You are on page 1of 33

Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

Contents lists available at SciVerse ScienceDirect

Neuroscience and Biobehavioral Reviews


journal homepage: www.elsevier.com/locate/neubiorev

Review

A behavioral neuroenergetics theory of ADHD


Peter R. Killeen a,∗ , Vivienne A. Russell b , Joseph A. Sergeant c
a
Department of Psychology, Arizona State University, Tempe, AZ 85287-1104, USA
b
Department of Human Biology, University of Cape Town, Faculty of Health Sciences, Observatory, 7925, South Africa
c
Vrije Universiteit, Department of Clinical Neuropsychology, Van der Boechorststraat 1, 1081 BT Amsterdam, The Netherlands

a r t i c l e i n f o a b s t r a c t

Article history: Energetic insufficiency in neurons due to inadequate lactate supply is implicated in several neuropatholo-
Received 2 February 2012 gies, including attention-deficit/hyperactivity disorder (ADHD). By formalizing the mechanism and
Received in revised form 2 February 2013 implications of such constraints on function, the behavioral Neuroenergetics Theory (NeT) predicts the
Accepted 18 February 2013
results of many neuropsychological tasks involving individuals with ADHD and kindred dysfunctions, and
entails many novel predictions. The associated diffusion model predicts that response times will follow a
Keywords:
mixture of Wald distributions from the attentive state, and ex-Wald distributions after attentional lapses.
ADHD
It is inferred from the model that ADHD participants can bring only 75–85% of the neurocognitive energy
Astrocyte
Attention
to bear on tasks, and allocate only about 85% of the cognitive resources of comparison groups. Parameters
Cadherin derived from the model in specific tasks predict performance in other tasks, and in clinical conditions
Diffusion model often associated with ADHD. The primary action of therapeutic stimulants is to increase norepinephrine
Energy in active regions of the brain. This activates glial adrenoceptors, increasing the release of lactate from
Lactate astrocytes to fuel depleted neurons. The theory is aligned with other approaches and integrated with
Memory more general theories of ADHD. Therapeutic implications are explored.
Fatigue © 2013 Elsevier Ltd. All rights reserved.
Neuropsychology
Norepinephrine
Vigilance

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 626
2. Exemplary phenomena to be explained . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 626
2.1. Defining characteristics and comorbidities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 627
2.2. Putative role of hypoenergetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 627
3. Neuroenergetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 628
3.1. The astrocyte-neuron lactate shuttle (ANLS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 628
3.2. A compartment model of neuroenergetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 629
3.2.1. Premises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 629
3.2.2. Neurophysiological bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 629
3.2.3. Mathematical bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 630
3.2.4. Averaged data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 630
3.3. Translating the basic model into response measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 631
3.4. Attention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 632
3.4.1. Inconspicuous stimuli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 633
3.4.2. Conspicuous stimuli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 633
4. Applications to data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 633
4.1. Sustained attention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 633
4.2. Event rate: inter-stimulus interval . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 635
4.3. The role of inhibition: the stop task . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 637

∗ Corresponding author. Tel.: +1 480 967 0560; fax: +1 480 965 8544.
E-mail address: Killeen@asu.edu (P.R. Killeen).

0149-7634/$ – see front matter © 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.neubiorev.2013.02.011
626 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

4.4. Working memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 637


4.5. Motor control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 639
4.6. Reinforcement frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 640
4.7. Delay discounting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 641
4.8. Spectral analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 641
4.9. Executive functioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 642
5. Embedding NeT in a causal framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643
5.1. Proximate models of ADHD: its definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643
5.1.1. Attention deficit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643
5.1.2. Hyperactivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643
5.1.3. Impulsivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 644
5.1.4. Comorbidity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 644
5.2. Ultimate models of ADHD: NeT and other theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 645
5.3. Ultimate material bases: genetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 646
5.4. The final causes of ADHD: what functions are served? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 646
6. The role of pharmacological agents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 646
7. Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 647
7.1. The neuroenergetics mass-action model (NEMA) and the behavioral neuroenergetics theory (NeT) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 647
7.2. Implications for therapy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 648
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 648
Appendix A. Attention drift and recapture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 649
Appendix B. Data analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 649
Appendix C. Sessional analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 650
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 650

Under Review, Neuroscience & Biobehavioral Reviews are differentially impacted—the right prefrontal cortex, basal gan-
glia, and vermis of the cerebellum—are smaller in individuals
1. Introduction with ADHD, and they also take up less glucose when activated
(Castellanos et al., 2001; Hart et al., 2012; Paloyelis et al., 2007;
the engine is intact, but there is a problem with the petrol supply Vaidya et al., 2005; Yu-Feng et al., 2007).
(Van der Meere, 2002) How does one translate this hypoenergetic hypothesis into
testable predictions? The strategy of this paper is to develop a
The human brain is distinguished from that of other species by cartoon compartment model of the neural energetics required
a cerebral cortex enlarged to support the development of language to support rapid neural firing—the neuroenergetics mass-action
and complex social behavior. Constituting only 2% of the body’s model (NEMA). Whereas the actual energetic process is very com-
weight, it utilizes 25% of total glucose production, for perception, plex (see Section 3), NEMA suffices to trace the time-course of
response generation, and the intrinsic neuronal processing that critical processes at a level that contacts the relevant behav-
informed responses require (Zhang and Raichle, 2010). Not only do ioral data, adumbrated in Section 2. Thereafter (Section 4) the
humans possess more neurons than other species, those neurons simplest of process models is adduced to map the behavior of
are hungrier, due to their expansive dendritic arbors and long- persons with ADHD and matched control groups to the neu-
range projecting axons throughout that large volume (Sherwood roenergetic processes—the behavioral neuroenergetics theory of
et al., 2006). The transport and refinement of their food stock, glu- ADHD (NeT). It is a behavioral theory because, whereas we exploit
cose, from blood vessels is mediated by glial cells, which, in the data collected under various cognitive rubrics such as inhibi-
human brain, are about as numerous as neurons (Azevedo et al., tion, inattention, working memory, and executive function, our
2009). Astrocytes ferry glucose from capillaries, store it as glycogen, theory neither invokes these, nor explains its data in terms of
and convert it to lactate, the primary fuel of rapidly firing neu- them. It is neurobehavioral. It combines a drift model of response
rons. Astroglia also assist the neuron in providing other nutrients, times from the attentive state with a Markov model of the
maintaining the composition of the extracellular fluid and clearing lapse and recovery of attention. In Section 5, NeT is related to
neurotransmitters from the synaptic cleft. other theories of ADHD, such as the cognitive-energetics the-
Todd and Botteron (2001) suggested that some forms of neu- ory (Sergeant, 2005), and provides the biological definition of
ropsychiatric disorder may be viewed as cortical energy-deficit energy pools, and quantitative predictions of task effects missing
syndromes secondary to hypofunctionality of catecholamine sys- from that theory. Section 6 reviews the role of pharmacological
tems that regulate astrocyte glucose and glycogen metabolism. agents. Section 7 summarizes these results, reviews predictions
This suggestion was examined in detail by Russell et al. (2006), of the theory, and draws implications for future research and
who explored its implications for attention-deficit hyperactivity treatment.
disorder (ADHD). They hypothesized that ADHD symptoms, par-
ticularly the marked intra-individual variation that characterizes
them, may arise as a result of impaired lactate production by astro- 2. Exemplary phenomena to be explained
cytes that is, insufficient to meet energy demands, resulting in a
supply of adenosine triphosphate (ATP) inadequate to maintain ion The diagnosis of ADHD is a complex and controversial adjudi-
gradients across neuronal membranes. This would impair timing cation, as ADHD constitutes not so much a taxon as an extreme
of motor responses, leading to slow and variable reaction times in range of scores in a multidimensional character state (Coghill
energy-demanding tasks like complex cognition and rapid exter- and Sonuga-Barke, 2012), parts of which range are also occu-
nally paced responding, such as finger tapping–especially when pied by other psychiatric disorders. It is therefore necessary to
regulated by the characteristically unmyelinated, energetically delimit the phenomenon and clarify the domain of the proposed
inefficient axons of dopaminergic neurons. Some areas of the brain theory.
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 627

2.1. Defining characteristics and comorbidities ADHD employ learning and memorial strategies that involve less
effort than controls (Egeland et al., 2010), are less adept at learning
Inspection of the Diagnostic and Statistical Manual (DSM- and memory (Andersen et al., 2012; Itami and Uno, 2002), and are
IV, American-Psychiatric-Association, 1994), and the forthcoming less able to sustain attention beyond the first few moments of a
DSM-5, reveals that attention deficits are a primary symptom in novel task (Sykes et al., 1973). Individuals with ADHD do not acti-
widely ranging disorders such as schizophrenia, depression and vate fronto-striatal regions in the same manner as TDCs in cognitive
anxiety. ADHD is associated with symptoms of inattention or tasks, but seem to rely on more diffuse networks of regions, includ-
hyperactivity/impulsivity, qualified by age and context. Some of ing more posterior and dorsolateral prefrontal regions (Durston
the categorizing symptoms of inattention are: careless mistakes in et al., 2003; cf. van Mourik et al., 2005). Individuals with ADHD
work; difficulty sustaining attention; difficulty in organizing and may have a deficiency in enzymes related to energy supply, which
failure to finish tasks; frequent distraction and loss of things; forget- prevents or delays development of fronto-striatal circuitry; or they
fulness. Some of the categorizing symptoms of hyperactivity are: may require more energy than normal individuals because they
fidgeting, squirming, running about; difficulty in quiet play; exces- have not developed energy-efficient neural networks, including
sive talking. Individuals must have a minimum number of the first fully myelinated axons (Fair et al., 2010; Nagel et al., 2011).
type or of the second, or a combination of both to be categorized or Impulsiveness and difficulty in carrying through plans for future
specified. Recent overviews of this disorder are available for both actions may be in part a consequence of impaired ability of starved
professionals (Stanford and Tannock, 2012) and the general public fast-spiking GABAergic interneurons to inhibit the upper motor
(Nigg, 2006). neurons that control the inappropriate responses (Koós and Tepper,
Only a minority of individuals diagnosed with ADHD have only 1999). This is especially the case for inhibitory circuits, because
ADHD; the majority (about 2/3) have other comorbid conditions. GABAergic interneurons are “special-needs” elements: cortical
One report from the Multimodal Treatment Study of Children with pyramidal neurons take up about 25% more sodium than the
ADHD (MTA: Jensen et al., 2001; Landau et al., 2012) found that 14% theoretical minimum required to generate action potentials, but
also had anxiety disorder, 30% also had oppositional-defiant disor- fast-spiking GABAergic interneurons, with their rapidly rising and
der or conduct disorder, and 25% had all three disorders. Depression falling action potentials, fail to completely inactivate sodium chan-
is common with ADHD, especially among girls (Angold et al., 1999). nels, thereby allowing twice as much sodium to enter the neuron
Kessler et al. (2006) found even greater comorbidities of ADHD (Carter and Bean, 2009). This inefficiency entails that substantially
with mood and anxiety disorders. There are also strong associations more ATP is necessary to re-establish their sodium gradients, which
with other conditions, such as learning disabilities (Willcutt et al., increases the vulnerability of inhibitory GABAergic interneurons
2010b) and impaired motor coordination (Lingam et al., 2010). to deficiencies in energy supply. Again, this may contribute to the
There are many possible reasons for this heavy burden of comorbid- impulsiveness that is characteristic of some categories of ADHD.
ity, thoughtfully discussed by Taylor (2010). Most observers agree Another common accompaniment of ADHD is clumsiness—poor
that ADHD is not simply a single disorder of executive function, but co-ordination of fine and gross body movements (for reviews see
rather a complex syndrome of impairments that overlap substan- Dyck et al., 2011; Gillberg and Kadesjö, 2003; Sergeant et al., 2006).
tially with other disorders (Polanczyk et al., 2007). The focus of this Poor performance in speeded reaction time tasks, slow and vari-
paper is primarily on the symptom of attention deficit, in particular able reaction times, and premature responses are widely reported
as manifest in neuropsychological tests; secondarily on hyperactiv- (Karalunas et al., 2012b; Van Meel et al., 2005). The organization of
ity; and only discursively on the related comorbidities. But there is a motor output is dependent on temporal synchronization of neural
common theme to many of these associated syndromes: comorbid- firing in cortico-striato-thalamo-cortical and cortico-cerebellar-
ity among ADHD, reading disabilities and mathematical disabilities thalamo-cortical circuits (Smith et al., 2003), and thus particularly
“is due to common genetic risk factors that lead to slow processing sensitive to impaired signaling, as seen in data reported by, for
speed” (Willcutt et al., 2010b, p. 533). NeT attributes the slow example, Noreika et al. (2013). Motivation cannot account for all
processing speed to insufficient neuronal energy supply, and the of the differences between groups (Van Meel et al., 2005). Deficient
consequent fatigue of rapidly firing neural networks. In turn, this action monitoring and error processing (also energy-demanding
leads to erratic performance and diffusion of attention. Our the- processes) are often associated with ADHD, leading to unrealistic
ory only pertains to this constellation of neurophysiology and its expectations by affected individuals, and increasing their difficulty
functional impact on behavior. It does not attempt to predict all of in learning from mistakes (Albrecht et al., 2008; Durston et al., 2003;
the symptoms used to categorize children with ADHD in the DSM. Sergeant and van der Meere, 1988). These are just some of the key
The differences among individuals may be due to the parts of the features that a comprehensive theory of ADHD must explain.
brain that are most affected by the neuroglial insufficiencies that we Many similar deficits are found in other developmental disor-
posit (de Zeeuw et al., 2012; Hart et al., 2013; Tafazoli et al., 2012), ders, and many symptoms that are significantly associated with
loci that may be a semi-random aspect of development, and to the classification are not found in every child with ADHD. The classifi-
functional heterogeneity of astrocytes (Theis and Giaume, 2012). cation “ADHD” is not a taxon (Coghill and Sonuga-Barke, 2012). As
The similarities among subtypes (Toplak et al., 2009) and related Dyck et al. (2011) noted, there is no evidence of “zones of rarity”
disorders (Banaschewski et al., 2005) may be due to a shared core between developmental disabilities: “with the exception of men-
deficit in energy resupply to neurons. tal retardation, the results imply there are no natural boundaries
between disorders or between disorders and normality”. Our the-
2.2. Putative role of hypoenergetics ory is not about an endophenotype for ADHD: it is a characterization
of a hypothesized brain disorder and its behavioral sequelae that
A number of strands of evidence are converging on energy insuf- is especially relevant to ADHD. It may characterize aspects of other
ficiency as a prominent factor in ADHD. Resting state MRI indicates developmental disorders as well.
clear differences between ADHD and typically developing con- Further evidence for the energy-deficiency hypothesis of ADHD
trols (TDC) (Castellanos et al., 2005; Uddin et al., 2008). The BOLD is provided by the recent finding that N-acetylaspartate (NAA) lev-
response measured by MRI indicates that the ability to summon els are altered in ADHD (Perlov et al., 2009; Yang et al., 2010).
glucose for oxidative metabolism, required for the restoration of ion NAA is a neuron-specific energy storage and transport form of
gradients in the fronto-parietal system, is impaired in persons with acetyl coenzyme A (Ariyannur et al., 2010). Clinical improvement
ADHD (Cortese et al., 2012; Zametkin et al., 1990). Individuals with brought about by treatment with methylphenidate is accompanied
628 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

by increased NAA in the anterior cingulate cortex of treatment- et al., 2012), including the ability to limit neuronal excitability, limit
naïve adults with ADHD (Kronenberg et al., 2008), and in various release probability, and silence synapses (Pannasch et al., 2011).
brain locations in children with ADHD (Wiguna et al., 2012).
The anterior cingulate cortex plays an essential role in error 3.1. The astrocyte-neuron lactate shuttle (ANLS)
processing—recognizing and learning from mistakes—which is one
of the deficits associated with ADHD (Bush et al., 1999; Durston A major part of the ATP used by neurons is for post-synaptic
et al., 2003; Itami and Uno, 2002; Van Meel et al., 2007). NAA restoration of ion gradients and transmission of action potentials
is important not only for energy metabolism, but also for lipid (Attwell and Laughlin, 2001; but see Harris et al., 2012). The neu-
synthesis, required for the formation of myelin sheaths, and as ron’s supply of energy for these processes involves a complex set of
a component of a molecular water pump in myelinated neurons reactions that starts with intra-neural resources, and quickly there-
(Baslow, 2002). The myelin sheath makes neurotransmission much after engages extracellular and glial-supplied lactate (Aubert et al.,
more energy efficient (Zhu et al., 2012), and its disturbance in ADHD 2005; Pellerin et al., 2007). Although neurons can produce ATP
(Nagel et al., 2011) indicates that axonal transmission of informa- by metabolizing glucose (Hyder et al., 2006), this is less efficient
tion is more energetically costly, and—key to our thesis—slower than utilizing lactate, and so they have net lactate consumption
(Harris and Attwell, 2012) and more variable (Walhovd and Fjell, (Jakoby et al., 2012; Pellerin et al., 2007). The astrocyte-neuron lac-
2007). NAA levels are considered a surrogate marker of neural tate shuttle (ANLS), controversial a decade ago, is now accepted
integrity in general, and are irregular in other psychiatric disor- as the primary source of energy to fuel sustained neuronal activity
ders, for example, multiple sclerosis (Rigotti et al., 2011), which (e.g., Brown and Ransom, 2007). The first 5–12 s of neural activity is
has behavioral symptoms in common with ADHD. We attempt to powered by oxidative metabolism within the neuron using stored
integrate these strands of evidence with a model of the neuroen- creatine phosphate and NAA. During this time, the neuron begins
ergetics of the individual neurons, and the consequent effects of to draw on the reservoir of extracellular lactate (Hu, 2002; Hu and
insufficient energy on the mass action of ensembles of neurons. Wilson, 1997; Mangia et al., 2003, 2006). Glutamate released by
the neuron during stimulation is co-transported with sodium ions
3. Neuroenergetics into the astrocytes (Fig. 1). Elevated levels of potassium resulting
from neural activity indirectly stimulate soluble adenylyl cyclase
In this section we describe how the release of glutamate by receptors on the astrocytes (Choi et al., 2012, p. 1094), causing
neurons stimulates astroglia, both directly and through secondary glycogen breakdown, enhancing glycolysis, and releasing lactate
release of norepinephrine from neural varicosities. That stimula- into the extracellular space, where it is available to neurons for use
tion increases the glial cells’ uptake and metabolism of glucose, to as an energy substrate (Bélanger et al., 2011). Depletion of ATP used
energize its own responsibilities in clearing glutamate, transfor- to re-establish the sodium gradient across the membrane further
ming it into glutamine, and transferring it back to the neuron. That stimulates glucose uptake by the astrocyte, which can be metabo-
stimulation also causes the glia to produce lactate, which is a cru- lized to lactate or used to produce ATP (Bélanger et al., 2011). The
cial fuel for the neuron. It is our thesis that this complementarity resulting ATP maintains glutamate transport into astrocytes and
between astrocyte and neuron fails in ADHD. It is not clear pre- supports its conversion to glutamine, which is shuttled to the neu-
cisely why this failure occurs, but we speculate about this below rons to restore their pools of neurotransmitters (Magistretti, 2009;
(Section 3.1). The remainder of this section outlines in detail how Pellerin et al., 2007). Surplus lactate enters the extracellular reser-
this energetic process is known to occur in the normal case. This voir, where its availability to the neuron favors lactate oxidation as
neurophysiological basis for attentional disorders such as ADHD the neuron’s primary source of energy to support sustained neural
constitutes our behavioral neuroenergetics theory, NeT. firing (Hyder et al., 2006; Wyss et al., 2011). Glial lactate is the major
Glia metabolize glucose to lactate which is released for use by contributor to neuronal oxidative phosphorylation of ADP to ATP
neurons to produce ATP, the energy currency of the cell. Neurons (Hertz et al., 2006). Glutamate also stimulates release of norepi-
metabolize glucose primarily in the pentose phosphate pathway nephrine from nearby varicosities in axons (Russell, 2001; Russell
to support protein synthesis and other housekeeping functions, and Wiggins, 2000), which, acting on ␤-adrenoceptors, further
including glutathione regeneration, which protects the neuron stimulates astrocytes to take up glucose and produce additional
from reactive oxygen species (Bélanger et al., 2011). Metabolism lactate (Gibbs et al., 2008b; Todd and Botteron, 2001). Such stim-
of glucose becomes rate-limiting when precise timing is required ulation has been shown to enhance memory (Gibbs et al., 2008a;
for the synchronization of neural networks firing at different fre- Hutchinson et al., 2007). The learning of new tasks differentially
quencies (e.g. theta and gamma) in different brain areas (Kann induces gliogenesis in the medial prefrontal cortex (Rapanelli et al.,
et al., 2011), which is associated with short-term learning (Fell and 2011). Fig. 1 sketches some of the key elements of this process.
Axmacher, 2011). In such cases, alternative intraneuronal energy Work of Caesar et al. (2008) draws a timeline of lactate pro-
stores, such as creatine phosphate and NAA, are utilized to produce duction, stimulated by activation of cerebellar climbing fibers, and
ATP (Ariyannur et al., 2010; Friedman and Roberts, 1994). These the ensuing glutamate-stimulated glycolysis, in astrocytes (Fig. 2).
stores also become depleted by the precisely timed patterns of After elicitation by glutamate, lactate production rises to exceed
firing of neural networks that are needed over extended periods baseline extracellular levels, with diffusion into local neurons
to maintain long-term potentiation (LTP) of synaptic strength replenishing their stores. At cessation of stimulation lactate slowly
(Richter and Klann, 2009). Astrocytic glycogen breakdown and lac- falls to the pre-stimulation levels. As shown, the half-life of both
tate release are essential for the maintenance of LTP elicited in vivo processes is several minutes. Cloutier et al. (2009) have constructed
(Newman et al., 2011; Suzuki et al., 2011), and for the long-term a relatively complete physiologically based pharmacokinetic model
memory that LTP enables (Costa-Mattioli et al., 2009; Mayford et al., of such processes involving 34 differential equations, with the ANLS
2012; Newman et al., 2011; Suzuki et al., 2011)—a kind of mem- playing a key role. We shall maintain emphasis on the ANLS as a
ory that is at disadvantage in ADHD (Rhodes et al., 2011). The key bottleneck in the process, but not engage those equations in
ATP generated within the astrocyte in response to the above pro- our model.
cess permits it to convert glutamate to glutamine for return to the Failure of this supply chain at any one of the reactions can
neuron (the glutamate-glutamine shuttle), for recycling into gluta- undermine functionality of the neuron, increasing variability in
mate (Magistretti, 2009). Astrocytes, we are learning, are involved responding as the neuron fatigues, and require that alternate
in neuronal computational processes in a “plethora” of ways (Min circuits be recruited to guide behavior. Therapeutic agents such
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 629

Fig. 1. The supply chain for adenosine triphosphate (ATP) production, that powers the neuron. Release of glutamate (yellow circles) stimulates glucose uptake (blue arrow)
and glycolysis in the astrocyte to produce lactate. The lactate diffuses into the extracellular space, to be absorbed by the neuron for ATP production, for restoration of ionic
gradients, and resequestration and encapsulation of neurotransmitters. Astrocytes also convert glutamate to glutamine, which is shuttled to the neurons to restore their
pools of neurotransmitters (yellow arrows). Glutamate, acting on AMPA receptors, stimulates norepinephrine release (red circles) from nearby noradrenergic varicosities.
These act on ␤-adrenoceptors, to further stimulate glucose uptake and glycogenolysis, causing astrocytes to produce more lactate to support sustained neural firing.

as amphetamine that block the monoamine transporters maintain 3.2. A compartment model of neuroenergetics
high extracellular concentrations of norepinephrine. Those raised
levels stimulate both cAMP and lactate production in astrocytes, Relatively complete models of energetic supply are available,
and can help compensate for any of several potential energetic but their many parameters overpower the available behavioral
bottlenecks (Pellerin and Magistretti, 2011; Sorg and Magistretti, data. A simpler phenomenological model may suffice to translate
1991). the fundamental energetic processes into behavioral predictions.
That is assayed here.

3.2.1. Premises
We present here the neurophysiological and mathematical
bases of our theory. NeT posits a particular brain dysfunction
(neural hypoenergetics) characteristic of individuals with ADHD.
Ancillary to this is a general model of response times based on
inferred neural processing speed and attentional processes, NEMA.
It bridges the neurophysiological facts, from the neuroenergetics of
individual neurons through the mass action of ensembles of neurons to
measures of behavior. Essentially, NeT calls upon NEMA as a sub-
routine to deliver predictions, and to interpret extant data from
studies of ADHD in terms relevant to the theory.

3.2.2. Neurophysiological bases


Action potentials, postsynaptic potentials, and the subse-
Fig. 2. Proportional change in extracellular concentration of lactate in rats’ cere-
quent resetting of ion gradients and clearance of glutamate, are
bellar cortex elicited by stimulation of the climbing fibers. Data from Caesar et al.
(2008); curve from Eq. (2), with s = d = 0.005/s. Both are rescaled to start at 0 and energy-demanding processes critical for high rates of informa-
asymptote at 1. tion transmission (Attwell and Gibb, 2005; Howarth et al., 2012;
630 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

Strelnikov, 2010). Both neurons and glia bear these costs. These are • Each astrocyte services scores of neurons, and connects to
the underlying facts on which we base the neuroenergetic part of thousands of others, whose excitatory and inhibitory interactions
our theory. The singular premise that one or more of these pro- may be treated as a random additive process, represented as a
cesses is disordered in ADHD constitutes our theory, NeT. random walk with drift, Eq. (5) (below).
• The ability to maintain attention to a task, like all responses,
• During stimulation, neuronal energy is initially provided by mito- covaries with the energy available for it. Response-time distribu-
chondrial oxidative processes, and later as a byproduct of glial tions are then mixtures of processes issuing from the inattentive
(astrocyte) processes (Mangia et al., 2003). and attentive states (Fig. 5).
• Glutamate released by neurons is taken up by the astrocytes,
powered by sodium influx. The ATP used to re-establish the Fig. 3 entails:
sodium gradient stimulates glycolysis, which restores the ATP
and also generates lactate (Hyder et al., 2006; Kasischke et al., dE
= −cfE + s(EM − E) c, f, s ≥ 0 (1)
2004). dt
• Additional energy may be recruited by astrocytes from their where E is the energy available to the neuron/glia process in the
glycogen stores (Benarroch, 2010). Such glycogenolysis is form of ATP, both endogenous to the neuron and derived from lac-
facilitated by noradrenergic stimulation of the astrocytes’ ␤- tate shuttled from the astrocyte. (There is an implicit constant of
adrenoceptors (Fillenz et al., 1999; Hertz et al., 2010). proportionality in both c and f that converts lactate to available
• Lactate is transported to the extracellular neuropil, where it is energy). EM is maximum energy—that in a fully rested neuron—and
taken up by neurons as their preferred energy source (Pellerin is set here to 1, so the basic model concerns proportional changes,
et al., 2007). and requires rescaling to speak to the units in which physiological
• The release of glutamate from presynaptic terminals is propor- variables are measured. The first addend has the energy decreasing
tional to: (a) the frequency of action potentials, and (b) the proportionate (−c) to the rate of stimulation (f) and the current
size of the readily releasable pool of vesicles, maintained by the energy level, as consistent with the first bullet of this section. The
astrocyte’s glutamate-glutamine shuttle. Both are impaired by second addend has the rate of repletion increasing proportionate
insufficient supplies of glucose or lactate (Magistretti, 2009). (s) to the neuron’s deficit, EM − E, consistent with the second bullet
point. Call the initial energy (at t = 0) E0 . Then the solution of Eq. (1)
3.2.3. Mathematical bases is:
The key premises in translating the above theory into a model
that enables quantitative predictions (NEMA) are: Et = E∞ + (E0 − E∞ )e−(s+cf )t (2)

The available energy changes exponentially from E0 to E∞ . For


• The depletion of energetic resources by neurons is a linear func-
the data analyzed in this paper we assume the subject starts in a
tion of both the rate of stimulation (Smith et al., 2002), and the
rested state, and thus set E0 to 1. The rate of approach to equilibrium
available energy (Leegsma-Vogt et al., 2004), implying (pseudo)
at E∞ depends on the sum of the rates of repletion and depletion,
first order kinetics as captured by Figs. 2 and 3 and Eqs. (1–2)
s + cf. As t → ∞, the available energy stabilizes at an asymptotic
(given below).
value E∞ that depends on the ratio of demand, fc, to supply s:
• The repletion of energy to the neuron involves classic
Michaelis–Menten processes (Aubert and Costalat, 2007; Pellerin s 1
E∞ = = (3)
et al., 2007) that may also be approximated by (pseudo) first order s + cf 1 + fc/s
kinetics, as in Figs. 2 and 3 and Eqs. (1–2).
• The simplicity of these relations despite the complexity of the If the rate of supply, s, is slow, or the cost of neural processing, c,
is high, Eq. (3) predicts a lower asymptotic energy. Eq. (3) is concave
energy supply is due to mass action of ensembles of neurons.
in s, so that variable rates of supply will also cause deficits below
“The apparent paradox of a range of energy-consuming processes
an invariant supply at the same mean rate. This is the prediction
being proportional to a single electrical activity [stimulation fre-
for ADHD.
quency] may be resolved if these processes are all coupled to
Because the supply from the astrocyte complements the
the averaged rate of the electrical activity” such as “the firing
demand from the neuron, this system describes the changes in
of an ensemble of pyramidal neurons” (Smith et al., 2002). Such
extracellular lactate shown in Fig. 2, where the origin from the pre-
“coherent functional networks” have been noted by Deco et al.
stimulation average concentration was set to 0, and the change
(2010).
• When such first order kinetics are averaged over individuals rescaled by dividing by the imputed asymptote 23 ␮M. Eq. (2)
draws the curve through those data. Upon glutamate signaling,
or epochs having different rate constants, the resulting average
lactate supply rises at the rate s + cf = (0.005 + 0.005)/s. When the
time-course is transformed into a power function, Eq. (4) (below).
stimulation stops, f goes to baseline (here 0), glutamate exocy-
tosis decreases to baseline, reducing the stimulus for glycolysis
and letting extracellular lactate decrease exponentially at the rate
of 0.005 ␮M/s. These are relatively slow processes, with half-lives
of several minutes. The above bullet points constitute the neuro-
physiological and mathematical bases of our theory.

3.2.4. Averaged data


Eq. (2) provides the core model of change in energy as a function
of time, but it applies only when there is no variance in the param-
Fig. 3. The neuroenergetics model. Major parts of the energy budget are postsy- eters. When exponential processes are averaged over data sets in
naptic processing and propagation of action potentials, both an increasing function which the rate constants vary among elements, as in ensemble fir-
of the frequency of stimulation. After several seconds of firing the neuron relies on
ing, the resulting curves morph into power functions, as derived for
lactate provided by the astrocyte in the ANLS. The hypothesis of this paper is that in
ADHD the cost of transmission of action potentials (c) may be increased, or the rate exponential constituents by Killeen (2001), and for a range of con-
of supply of lactate (s) may be compromised. stituent distributions by Murre and Chessa (2011). Eq. (2) becomes:
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 631

Et = E∞ + (E0 − E∞ )t − t≥1 (4) Table 1


Parameters for the neuroenergetics mass-action model (NEMA).
where the exponent  (rho) increases with s + cf. Parameter Module

Energetics
3.3. Translating the basic model into response measures Et Energy available in the form of lactate in the neuron/glial
system (v in Eq. (5))
In this section we present the key NEMA equations and illustrate Predicted by Eq. (2) or Eq. (4) as a proportion of the
maximum energy, EM
their use in data from archival studies of ADHD.
The speed of information processing is tightly coupled to the E∞ Equilibrium energy given by Eq. (3), or fit as free parameter
energy resources available: it has been estimated that one mole  Rate of approach to asymptote for group data, increasing
with s + cf
of glucose must be oxidized for each mole of neurotransmitter
recycled (Shulman et al., 2004). The result of these coordinated Attention
 Probability of a lapse of attention in any 1 s period
excitatory and inhibitory neurotransmissions may be treated as a
˛ Probability of a return of attention in any 1 s period
random walk toward a decision, a Brownian motion with “drift”  Attentional inertia: mean time required to refocus lapsed
toward an action. The speed of the progress, v, at any point in attention, or to shift set
time is proportional to the energy available for it, Et , as given by
Response time
Eq. (2) or Eq. (4). The times for completion of such a large series E Average available energy, determining speed of neural
of additive and subtractive (inhibitory) computations—the time computation
of first passage—progressing at the speed v is given by the Wald C Computational resources allocated for a response; Criterion
distribution: in drift model

Derived measures
C 2
f (t) = √ e−(C−vt) /2t t>0 (5) C/E Mean response time from attentive state
2t 3 C/E3 Response-time variance from attentive state
(CE)−1/2 Coefficient of variation from attentive state
The conditions necessary for a process to be characterized by [1 − (1 − ) ]
s
Expected additional reattentional latency s seconds into trial
Eq. (5), and for such inverse Gaussian distributions, when summed, [1 − (1 − ) ]
s 2
Expected additional reattentional variance s seconds into
to yield a distribution of the same form, are given by Folks and trial
Chhikara (1978). A clear and thorough review of the properties of
the distribution may be found in Chhikara and Folks (1989). Eq. (5) of response characteristics. NEMA’s parameters are summarized
gives the distributions of times for processes proceeding at the rate in Table 1. The mean of Eq. (5) is  = C/E, its standard deviation
of v to cross a criterion C for the first time. The value of C depends (C/E3 )1/2 , and its coefficient of variation (CV) is (CE)−1/2 . E may be

on the informational demands in all stages, and on the complete- estimated directly as /. It is a skewed distribution with a coef-
ness with which the individual processes that information. If more ficient of kurtosis equal to thrice the CV. These statistics are heavily
information is needed for one task than for another, the value of dependent on the speed of propagation: halving the velocity E
C will be larger for that task. In general C should be thought of as doubles the mean, and more than doubles the standard deviation.
a stopping criterion, giving a measure of the number of computa- This explains why variability is more diagnostic of slowed neural
tions involved in generating a response. More complex tasks will processing speed than are means. For the basic time-course data,
require more computations to succeed. Individuals must generally there are three key parameters in the model, , which is governed
make speed-accuracy tradeoffs, in the light of both task demands, by the sum of the rates of depletion and repletion, E∞ , the equilib-
and the resources available to them. Thus the value of C may also rium energy level of the glial/neuron system, governed by the ratio
be strategically biased. With experience, that bias may become the of the rates of repletion and depletion, and C, the response crite-
habitual default. rion. When only summary statistics rather than time-courses are
An additional parameter in the denominator of the exponent of available, the basic model reduces to two parameters: the average
Eq. (5), the variance in drift rate, is absorbed into C and v. If drift available energy E, and the criterion C. When additional data are
rate is more variable for individuals with ADHD, as expected from available concerning attentional processes, the model expands to
NeT, it would provide one explanation for the lower values of those include two additional parameters: the probability of attentional
parameters found for affected individuals, as documented below. lapses (), and the probability (˛; or latency, ) of return to atten-
C is a cousin of the criterion in signal detection theory (SDT). In tion, discussed below (Section 3.4.1 and Appendix A).
NEMA it is (proportional to) the number of computations before The use of a diffusion model such as Eq. (5) as a lingua franca
a positive response. In SDT it is the amount of evidence necessary between neural and behavioral processes is not novel. Ratcliff and
for a positive response. If that evidence accumulated as a biased McKoon (2008) review the rapid advances being made in inte-
random walk, the models converge, as the case in general drift grating such models of decision processes with neural data. That
models, which have separate criteria for the relative amount of evi- literature makes close contact with neurophysiological reports.
dence necessary for a positive or negative response. The difference Here, with focus on human subjects, the analysis must be hypo-
of those criteria gives a measure of bias, as in SDT. Drift models thetical. With the emergence of animal models of ADHD, better
typically let the speed of drift, v, depend on problem difficulty. In reification can be expected. With additional parameters this class
NEMA, drift rate is tied to the available energy, which determines of models can also accommodate error rates, as demonstrated in
processing speed, while the information processed determines C. the elegant applications of drift models to ADHD in the work by
Eq. (5) is a special case of the inverse Gaussian distribution Mulder et al. (2010) and Karalunas et al. (2012a). Although these
(Chhikara and Folks, 1989; Folks and Chhikara, 1978; Heathcote, models provide a more complete description of neuropsychological
2004). For parsimony we set the velocity of propagation of the data on short ISI experiments, they do not include the attentional
decision, v, proportional to the available energy in relevant neu- loop of NEMA as shown in Fig. 5, and therefore do not address inter-
ral ensembles, v = kEt , 0 ≤ Et ≤ 1. The units of v in this paper are stimulus-interval (ISI) effects. Huang-Pollock et al. (2012) provide
m s−1 . The constant k is set to 1, entailing that C must be under- a current and perceptive review of the utility of such drift models
stood as relative to this unknown constant that bridges from ATP for analyzing data from ADHD populations.
to the velocity of propagation of information. This is the basic mod- Amongst the few studies that report full response-time distri-
ule of NEMA, which translates neuroenergetics into predictions butions for ADHD subjects is that of Querne and Berquin (2009),
632 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

Fig. 5. The neuroenergetics model—NEMA. Trials start in the attentive state A. By


the end of 1 time period, attention will have lapsed with probability . Upon appear-
ance of a target stimulus, if the subject is in the A state a response is prepared and
emitted according to the Wald distribution (Eq. (5)). If the subject is in the ∼A state,
he must first re-attend, to perceive the stimulus and initiate the response. In exper-
iments with inconspicuous stimuli (vigilance or detection experiments), attention
drifts back to A at a rate of ˛ . In experiments with salient stimuli (performance
experiments, analyzed in this paper), conspicuous target stimuli capture attention,
driving ˛ larger. In that case the predicted response distribution is ex-Wald. The
latency to recapture attention adds the exponential process, with mean  = 1/˛ .
This additional latency is engaged with probability 1 − p(A), as given by Eq. (8). The
final distribution of responses is a mixture of the Wald and ex-Wald.

but second only to controls in the Hyperactive–Impulsive subtype.


The lowest values of E in the Inattentive and Combined groups, the
latter 68% of the value for TDC, are consistent with their having
the largest number of inattention symptoms. The higher energy of
the Hyperactive–Impulsive fuels the impulsivity, and along with
the lowered criterion speeds responses, including a disproportion-
ate number of errors of commission. Although this profile has face
validity with regard to the symptoms that are characteristic of these
groups, it is not a priori clear why the Hyperactive–Impulsive sub-
type should have the highest E of the ADHD subtypes, yet the lowest
criterion for action. Perhaps it is just that, given the constellation
of combinations of E and C that are possible for individuals with
ADHD, those with this combination—high E and low C—are indeed
hyperactive/impulsive.
Fig. 4. Relative frequency distributions of response times on trials with a cor-
Eq. (5) is closely related to other distributions. The ex-Gaussian,
rect response for four groups of children (Querne and Berquin, 2009). The curves a convolution of exponential and normal random variables, is
are Wald densities (Eq. (5)), with parameters displayed in the bottom panel. HI increasingly popular in describing response-time data from stud-
is Hyperactive–Impulsive; IA is Inattentive; and C is Combined types. As they are ies of ADHD. It mimics the Wald distribution (Gottlob, 2004;
arrayed in this figure, the criterion C decreased across groups, as did the available
Van Zandt and Ratcliff, 1995). The next section shows that an
energetic resources E, except for HI (Hyperactive Impulsive), which marshaled sub-
stantial energy but held the laxest criterion. NEMA accounts for 96% of the variance analogous distribution, a convolution of exponential and Wald
in the data. variates—the ex-Wald—provides the necessary final step in this
theory of response times.
which serves to illustrate the simplest deployment of the model
advocated here. Querne and Berquin matched four groups of 16 3.4. Attention
children for age and sex, and distinguished by membership in
three subtypes of ADHD (HI: Hyperactive–Impulsive; IA: Inatten- The above development carries the implicit assumption that
tive; C: Combined; distinguished by the numbers of hyperactive the subjects are always “on task”—that their behavior is under
and impulsive symptoms) or a control group. About every 3 s a the control of stimuli that are interesting to the experimenter or
visual target was presented which required discriminating higher teacher. This may be the case for brief experiments with keen
vs. lower positions on a screen. The resulting response-time distri- subjects, but it is obviously not in extended experiments on
butions are shown in Fig. 4. individuals with attention deficits. Subjects generally begin an
A marked difference in control and ADHD distributions is obvi- activity or experimental trial in an attentive state, but all may
ous; but how should this be interpreted? The parameters required experience lapses of attention thereafter, ranging from attentional
to fit the distributions tell the story: the criterion C (proportional blinks to attentional fugues. Attention then may drift back to
to the number of Computations made before a response) decreased the task (Section 3.4.1) or be captured by the appearance of
from Control through IA and Combined subtypes, to the HI sub- the target stimulus (Section 3.4.2). Such attentional fugues are
type. This is consistent with the Combined and HI subtypes having correlated with functional connectivity in the visual cortex, and
the greatest number of hyperactivity/impulsivity symptoms. The with response-time variability (Prado and Weissman, 2011).
energy available for the task, E (an indicator of the speed of This section explicates how NEMA deals with attentional drift
processing, equal to v in Eq. (5)), was greatest for the control group, and recapture. NEMA is the tool we deploy for highlighting the
less for the Inattentive subtype, still less for the Combined subtype, differences between special populations (Nesse and Stein, 2012).
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 633

NeT is a theory of the mechanisms of ADHD and its consequences. Table 2


Parameters for the curves in Figs. 6 and 7.
It predicts differences in parameters returned by NEMA, when
used to model the behavior of ADHD and TDC (Section 4). Parameter Börger et al. (1999)

Group
3.4.1. Inconspicuous stimuli
Control ADHD
Our model for this drift of attention from and then back to the
task, NEMA, starts with the simplest of Markov chains, diagrammed  0.23 0.09
in Fig. 5. Trial initiation brings the individual to attention, A, from E∞ 0.51 0.00
C 400 400
which he lapses with a probability  (lambda) into a non-attending  (s) 0.38 0.48
state, ∼A. If a target stimulus occurs while the subject is in the
A state, a perception/decision process is initiated, and a response
emitted according to the Wald distribution. If attention has wan- The probability of return, given stimulus onset, is a similar
dered, it will wander back to the task (State A) with probability ˛ exponential-integral process with a rate constant of ˛ rather than
(alpha). The long run probability of being in A is p∞ (A) = ˛/(˛ + ),  . These rate constants are primed because they are the continu-
and of being inattentive, its complement /(˛ + ). For  > 0, NEMA ous analogues of the probabilities shown in Fig. 5, denoted by ˛ and
predicts that the probability of being on task decreases approx- , the default parameters of NEMA. In the case that the probability
imately exponentially through a trial, from 1.0 toward p∞ (A). In of an attentional lapse approaches zero ( → 0), NEMA’s statistics
signal detection experiments, this process predicts the probability reduce to those of the basic Wald distribution, the first addends in
of a hit. The probability of maintaining focused attention during a Eqs. (6–7). The mixture is dominated by the Wald density because

trial of length t decreases with t approximately as e− t . Once in the the response, governed by the Wald process, is made on every trial
∼A state, the time to return to the attentive state is exponentially (every one that contributes data to these measures), and the addi-
distributed with mean of  (tau). (See Appendix A for a more precise tional lag due to inattention on only a portion of the trials. Adding
statement of this process). the inattention factor to the model of the data in Fig. 4 increased the
goodness of fit to over 97%, imputing larger values of  for the ADHD
3.4.2. Conspicuous stimuli groups. Because the fit of the simple Wald model was already excel-
The symmetry of Fig. 5 is broken by the ability of salient stimuli lent, however, little reliable information could be gleaned from that
to recapture attention. Vagrant attention that would only drift amplification.
back to the task with probability ˛ may be captured by the abrupt Vigilance is sustained hard work (Kaplan and Berman, 2010;
appearance of a target stimulus (Egeth and Yantis, 1997). This is Sarter et al., 2006; Warm et al., 2008). It is a core implication of our
called stimulus-driven attention. Different neural systems subserve theory that the resources to maintain attention decrease as energy
stimulus-driven and goal directed attention (Asplund et al., 2010; depletes through the course of a session (Eqs. (2 and 4)), with con-
Corbetta and Shulman, 2002; Sarter et al., 2005). The appearance comitant increases in the probability of lapses in attention (). This
of the target essentially jumps ˛ up to a large probability of is due to fatigue of neural circuits responsible for processing task-
transition back to A. Such on-demand attentional shifts are not relevant stimuli. When analyzing sessional data we assume that
immediate, but require up to ½ s in standard laboratory experi- the probability of maintaining attention (1 − ) is proportional to
ments (Shulman et al., 1979; Sperling and Reeves, 1980). We call the available energetic resources, 1 −  = kEt , so that  = 1 − kEt . In
this time to recover attention attentional inertia, and represent its most experiments analyzed here, only summary statistics are avail-
mean by . Attentional recapture adds an exponentially distributed able, so that estimates of E and  reflect the average operation of
attentional latency to a Wald-distributed perception-response those processes over the course of the session. Table 1 is a memo-
latency, resulting in an ex-Wald distribution. This distribution randum of these parameters. NEMA is now applied to a range of the
was well characterized by Schwarz (2001) and recently applied to available neuropsychological tasks employed in ADHD research, to
response latencies (Palmer et al., 2011). Responses issuing from help us evaluate the validity of NeT.
the attentive state are Wald distributed (Eq. (5)), while those from
the inattentive state add the attentional recapture process, trans-
4. Applications to data
forming it into an ex-Wald distribution. The expected response
times will be a Wald distribution from the attentive state with
Swanson et al. (Swanson et al., 2010; also see Willcutt et al.,
probability p(A) and an ex-Wald distribution from the inattentive
2008; Williams et al., 2010) summarized the meta-analyses of
state with probability p(∼A) = 1 − p(A), as shown in Fig. 5. The
neuropsychological tasks for ADHD reported by (Nigg, 2005) and
resulting mixture has a mean:
(Willcutt et al., 2005). Those tests with the largest effect sizes are
C analyzed here.
= + p(∼A) E>0 (6)
E
and standard deviation 4.1. Sustained attention

C The heart of NeT for analysis of special populations is its pre-
= + p(∼A) 2 (7)
E3 diction of differential rates of repletion of energetic stores, causing
Most experiments analyzed in this paper used conspicuous, increased probability of attentional lapses, a slowing of response
abrupt-onset stimuli, so that stimulus-driven attention preempts times and consequent increases in variance of those times. Such
goal-driven attention. To reduce parameter load and simplify the fatigue is, of course, a hallmark of the performance of all humans
attentional model, the probability of drift back to attention, ˛, is on sustained detection or performance tasks: all show decrements
therefore set to 0 until the presentation of the stimulus, and thence over the course of a session, attributed to attentional lapses as a
to a larger value close to 1. This permits a simple expression (see result of boredom or resource depletion (Grier et al., 2003; Helton
Appendix A) for the probability of being in the inattentive state as and Warm, 2008; Mackworth, 1956; Parasurman and Davies, 1982;
a function of time since trial onset t: Pattyn et al., 2008). An example of the greater difficulty had by indi-

viduals with ADHD is provided by Börger et al. (1999), who tested
P(∼A) = 1 − p(A) = 1 − e t (8) 21 ADHD boys without comorbid disorders, either medication naïve
634 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

Fig. 6. Top: The probability of not making off-task responses during a CPT for 21 boys
with ADHD and 16 control boys. In this and all other figures, triangles represent the
Fig. 7. Top: The probability of not making any off-task response during a CPT trial
ADHD group and controls are represented by circles. Bottom: Means and standard
for 17 boys with ADHD (triangles) and 15 control boys (circles). Bottom: Means and
deviations of response times on that task. Data from Börger et al. (1999); curves from
standard deviations of response times on that task. Data from Börger and van der
NEMA, with parameters in Table 2, accounting for a median 76% of the variance in
Meere (2000), are shown as symbols; predictions by NEMA, using the parameters
the data.
that drew Fig. 6, are drawn as lines. They account for a median of 45% of the variance
in the data.

or abstinent for 24 h, on a continuous performance task (CPT, see,


e.g., Riccio et al., 2002). Sixteen age-matched normal boys consti- Börger and van der Meere (2000) replicated their previous
tuted the control group. In the CPT either a target letter (Q; 20% experiment in one that reported irrelevant movements, such as
trials) or foil (O; 80% trials) was presented for 700 ms every 4 s. The looking away from the screen, using fixed (4 s) ITIs and variable ITIs.
investigators measured on task behavior as the probability that no Because some of the tasks were similar—although they employed
irrelevant response, such as bodily movements, occurred. The prob- different samples of subjects—it is possible to predict the course
ability of such full attention is shown in Fig. 6, along with the model of attention, and response-time measures, based on the parame-
performance. The model predictions are derived from Eqs. (4 and ters from Börger et al. (1999). Fig. 7 shows that NEMA was able to
8), with the probability of maintaining attention (i.e., not lapsing) predict the general lay of the data, although the control subjects in
from one second to the next set proportional to available energy, this experiment paid better attention than predicted, and latencies
1 −  = 1.5Et , and waning with it as a function of time through the were shorter than predicted. Such deviations may be due to proce-
session. Table 2 lists the key parameters for the Börger et al. (1999) dural differences, as the subjects were on the average 1 year older
study, and Appendix B provides modeling details. By the end of the than in the prior study.
session, the ADHD subtype manifested about 86% of the energy of The above results replicate in the laboratory the common obser-
the TDC. vation of off-task behavior in the classroom manifest by all children,
Off-task responses were made at relatively high levels that but in particular by those with ADHD (Rapport et al., 2009b; Sawyer
increased throughout the course of the session, making undivided et al., 2001). Similar data have recently been found in working
attention increasingly scarce. Given these observations of the actual memory tasks, with decreasing orientation on tasks as a function
behavior, wayward attention must be recognized as a viable expla- of load, with large differences between ADHDs and controls (Kofler
nation for increased latencies in these experimenter-paced tasks. et al., 2010). The present theory does not predict the temporal
Other research (Teicher et al., 2004) shows that methylphenidate patterning of inattention that Börger et al. found—their children
increases accuracy in this task with the probable mechanism being were able to shift their discursive behavior between trial onset and
a decrease in the entry into “distracted states”. Both the asymp- target presentation into parts of the interval that would interfere
totic energy levels (E∞ ), and the rate of approach to those levels (), minimally with detection rates. This is a good example of strategic
depend on the rate of energetic supply to the neurons. Therefore reallocation of attention, one that preserves energetic resources of
we predict that these parameters will be larger for controls than for the modules relevant to on task behavior to when they are most
ADHD. Table 2 shows that this was indeed the case, consistent with needed. The implications of such patterning are discussed in Sec-
this prediction. Controls also recovered from lapses more quickly tion 5.1.1.
( = 0.38 s vs. 0.48 s). Both groups made about the same nominal NeT predicts a divergence in the performance of ADHD and TDC,
number of neural computations (C). with the former performing worse over the course of the session.
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 635

This is a reliable finding: for instance, in a large study, Epstein et al.


(2003) found that the increased variability of RTs over the course
of the session was “highly associated with most ADHD symptoms”.
In a meta-analysis Huang-Pollock et al. (2012) found reliable evi-
dence for performance-over-time differences between groups. But
a number of papers report no interaction effect between ADHD
and controls for the deterioration in performance as a function of
time on task (e.g., Rapport et al., 2009b). Stins et al. (2005) found
that children with ADHD performed more slowly, less accurately,
more impulsively, and with less stability than controls. There was
an increase in errors for ADHD and controls as a function of time on
task, also as predicted, but those curves were parallel, with no hint
of faster deterioration for ADHD. There are several possible reasons
for this. One is that the task was essentially self-paced, so the partic-
ipants were not forced into a speed-accuracy tradeoff: individuals
with ADHD could just slow down a bit more over the session to
maintain their (already compromised) accuracy. Furthermore, this
test was embedded in a series of other tests. Divergence in rates
is predicted over the first few minutes of testing; after 10 min,
performances arising from Eq. (4) will have convergent slopes at
divergent levels: the theory does not predict a significant differ-
ence in slopes late in testing (see Fig. C1). In the Stins et al. study,
the CPT was embedded in a day of testing, and each task was prac-
ticed “as long as necessary” for the child to understand the tasks
well enough to perform accurately on it. Under those conditions we
predict no measurable interaction, and that was what was found.
Had the training occurred on a prior day, and groups tested on this
task alone the next day, NeT must predict that an interaction in
RT, and in RT variability, with group, would have been found, with
ADHD deteriorating at a faster rate than TDC.
Fig. 8. Top panel: Data from Hervey et al. (2006) comparing response times from
4.2. Event rate: inter-stimulus interval ADHD children (triangles) and controls (circles) on the Conners CPT. Curves are from
NEMA. With parameters shown in Table 3, they account for 98% of the variance in
the data. Bottom panel: Data from Epstein et al. (2006) comparing response times
It is assumed in Fig. 5 that the beginning of a trial calls the sub- as a function of medication status on the day of testing, using the same task and
ject to attention, which he leaves each second with probability . procedure. Triangles represent ADHD off medication, and circles ADHD on medica-
ADHD individuals lapse more quickly because the neural circuits tion. The curves are predictions from the model using the parameters from the top
panel, and account for 97% of the variance in the data.
processing the target location and the top–down template for the
awaited target stimulus fatigue more quickly, and at the same time
the circuits processing distractors are refreshed. If trials are quickly the average (267 ms) for ADHD than for controls (184 ms). These
paced, there is a good chance that the target stimulus will elicit a data provide little leverage on the values of E and C, which here
response directly from the initial attention state. If the trial dura- serve only to set the 0 s ISI intercepts. The near identical perfor-
tion, or ISI, or pre-warning period is long, there is a greater chance mance at the 1 s ISI is unusual for ADHD and TDC groups (but see
that attention will have wandered, and must return to the atten- Kuntsi et al., 2009); NEMA translates this similarity into similar val-
tive state before a response can be made. In the present studies the ues of E. This is inconsistent with the prediction of NeT, and other
appearance of the target forces the issue and recaptures attention. theories that predict RT data, that persons with ADHD will have
The recovery adds an average of s to the latencies and  2 to the slower and more variable RTs; and therefore counts as evidence
variance of the eventual response. against those theories. On the other hand, most studies find such
When data are averaged over a session information is lost about differences (Tamm et al., 2012), counting as evidence for those the-
changes in energy levels and increases in mean and variance of ories. The preponderance of evidence supports NeT’s predictions.
response times throughout. That decrease in degrees of freedom in A number of studies were published from the MTA using exactly
the data evokes a corresponding decrease in the degrees of free- the same procedures. This greatly reduces random-effects variance
dom of the model:  is retired, and energy is set at the estimated in comparing them (Killeen, 2007), and should make it possible
average value for the session, E. All of the effects of ISI are due to predict results of other studies using similar parameter values.
to the Greek parameters  (probability of attentional lapse) and Epstein et al. (2006) used the same CPT for 316 ADHD-C children
 (attentional inertia). Fig. 8 shows data from the large-scale MTA from four study sites, when on and off medication. (The Hervey et al.
assessment using the Conners’ CPT-II. In the top panel, Hervey et al. data were a subset of these). Results were separately analyzed for
(2006) report the results from of 65 ADHD children and adoles- 190 children who took stimulant medication on the day of the test,
cents who had previously taken themselves off medication, with and 126 who did not. Children were not randomly assigned, but
matched controls. Means and standard deviations of both groups rather self-selected off medication sometime before these trials.
increased with ISI, with the ADHD group lying above the controls in The model and parameters used in the top panel of Fig. 8 draw the
both cases. E and C set the origin of the curves at ISI = 0: C/E sets the curves through the Epstein data in the bottom panel of Fig. 8. In

means, and (C/E3 ) sets the standard deviations. Visual inspection both panels the ADHD off medication were selected in like man-
of Fig. 8 shows that these curves converge toward ISI 0, thus permit- ner from the same cohort—individuals having taken themselves
ting us to set E and C to similar values across groups. They increase off medication for an indeterminate length of time. It is therefore
therefrom due to attentional lapses. Those lapses increased with not surprising that these data and parameters are so similar. Some-

the ISI as 1 − p(A) = 1 − e− t , with recapture of attention slower on what more surprising is the equivalent performance of the matched
636 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

Table 3 is mixed with the Wald density, and is responsible for the differ-
Parameters for the curves in Figs. 8 and 9.
ences between the columns. Lambdas were somewhat different for
Hervey et al. (2006) Leth-Steensen et al. (2000) the two groups, leading to probabilities of inattention by the time
of the target stimulus of 0.31, 0.53, and 0.78 (controls) and 0.38,
Parameter Control ADHD Control ADHD
0.62, and 0.85 (ADHD) for the 2, 4, and 8 s stimuli. The major factor
E 0.46 0.46 0.24 0.13
in the difference between rows was the attentional inertia , the
C 147 144 138 96
 0.34 0.34 0.21 0.27 time required for recapturing attention and initiating the response
 (s) 0.18 0.27 0.20 0.43 process (Oosterlaan and Sergeant, 1998b), which was twice as long
for the ADHD sample. This greater attentional inertia for individ-
uals with ADHD is consistent with the greater difficulty they have
control group in the top panel, and the ADHD group on medication in cued task switching (Cepeda et al., 2000; Kramer et al., 2001).
in the bottom panel. The parameters  and  play a similar role, and are thus con-
Hervey et al. (2006) fitted the ex-Gaussian distribution to their founded (almost collinear) in NEMA. Increases in one, caused
data. The ex-Gaussian is a response-time (RT) model comprising perhaps by sampling error, can lead to decreases in the other.
an exponential distribution convolved with a normal distribution. Because of this, care must be taken in interpreting their precise
The authors made two claims, one of which we view as true: “The values. The probability of attention lapsing by 1 s is . The product
greater number of abnormally long RTs of children with ADHD  therefore gives the expected temporal cost of lapses of attention
[i.e., those governed by the exponential process] reflect atten- at that point in time, and may be a more stable index of between-
tional lapses on some but not all trials.” This is precisely the claim group differences than either of its components. Table 1 gives the
of our attentional model. Finding that the mean of the Gaussian general calculation for expected temporal cost of inattention.
process was smaller for ADHD, they went on to argue that this Experimental tactics to maintain attention on the task, such as
“contradicts previous interpretation that children with ADHD have jittering the time of onset of the target, may substantially reduce .
slower than normal responding”. In fact, ADHD children did not If such a tactic had been employed by Leth-Steensen et al. (2000)
respond more slowly than TDCs at the 1 s ITI (see Fig. 8); but typ- at the shortest ISI and had driven  to 0, it would have cut the
ically they do (Tamm et al., 2012). The mean of an ex-Gaussian exponential tail off the ex-Wald, and substantially reduced  in the
distribution is  + . Hervey inferred that, because  was not larger ex-Gaussians fit to such data. We predict in such a case decreases
for ADHD, ADHD do not respond more slowly than TDC (as atten- in their coefficients of variation (CV) from 0.24 to 0.18 for TDC, and
tional differences are partitioned out in ). But that logic is valid a larger one, from 0.38 to 0.28 for ADHD, as the expected cost of
only if responses issuing from the attentive state are normally dis- lapsing is greater for ADHD. Lee et al. (2012) performed such an
tributed, so that the mean of the normal part of the ex-Gaussian, , experiment using fixed and variable ISIs of 1.5 s, and found reduc-
constitutes a “true” measure of speed of responding, one uncon- tions in the ex-Gaussian  when the ISI was varied, as expected in
taminated by inattention. But in general “RT distributions are the above scenario. The correlations of the above CVs with those
decidedly nonnormal—they are almost always skewed to the right” reported by Lee et al. is r = 0.98.
(Wagenmakers and Brown, 2007). Tau in the ex-Gaussian does One may question why parameters such as  and  should not be
some of the work intrinsic to the Wald, and some of the work essentially invariant within groups across studies that used similar
of the attentional lapse measured by our . If one takes the ex stimuli and procedures, and whether their differences are problem-
out of the ex-Gaussian, it leaves a normal distribution that cannot atic for NeT. The simple reason for their difference is that the data
fit the vast majority of RT data. Inferences from the ex-Gaussian are different, and NEMA interprets those data differences in terms
parameters are not to some “true” distribution that characterizes of differences in these parameters. The data are different because
normal response times; they are to a chimera whose constituent in most cases the participants are different, with different histo-
parameters must be treated with caution. ries and different ages and different nuances of testing procedure.
The importance of the right tail of the response distribution as a Curves drawn with fixed representative parameters would reflect
discriminant of ADHD and control groups has been noted by many the trends in all of these studies (see, e.g., Fig. C1), and that is the
observers, in particular Leth-Steensen et al. (2000). They showed standard target for most theories. Predicting precise neuropsychol-
that the exponential parameter in the ex-Gaussian distribution that ogical parameters from epidemiological data is a challenge of the
they fit to their data permitted almost perfect discrimination of future for NeT, as it is for all theories of ADHD.
those groups. That parameter,  (tau), is similar to the exponential NeT has, nevertheless, predictive validity. Consider for exam-
time constant  in our ex-Wald model. An important difference is ple a study by Rucklidge and Tannock (2002) testing children with
that in NEMA,  is derived from a theory of attention and inter- ADHD, with reading disabilities (RD), with a combination of the
preted as attentional inertia, mixed into an already skewed Wald two, and TDCs, on a variety of tasks, including the stop-signal task.
distribution with probability p(∼A). The values of E computed from their “go” reaction times ranged
Leth-Steensen et al. (2000) reported full distributions of from 0.164 for TDC to 0.096 for combined ADHD + RD. Ten years
response times for different groups and different ISIs, setting a later Gooch et al. (2011) reported a similar study on children with
rich and important exercise for our model. These authors presented ADHD and with dyslexia; and another similar study the next year
four circles on a computer screen for 2, 4 or 8 s, in counterbalanced (Gooch et al., 2012). The correlation between values of E for Ruck-
blocks, to 17 ADHD and 18 non-ADHD aged-matched control chil- lidge and Tannock’s four groups and the latters’ were 0.929 and
dren. At the end of the fore period, one circle was colored in, and 0.952 respectively. The correlation between the energy levels in
remained on until the subject responded to a computer key (S, F, the two recent studies was 0.978. Values of E for ADHD were 0.73,
J, or L). Approximately 80 trials were conducted at each of the fore 0.84, and 0.87 of those for TDC in the three studies, respectively.
period durations. The resulting distributions of response times are The median correlation among studies for C was 0.918; lower, as
shown in Fig. 9. expected, because C depends on task difficulty and strategic speed-
The ADHD boys marshaled less energy for the task (only 55% that accuracy tradeoffs unique to each experiment. Correlations are not
of the controls) and had a lower criterion for making a response the same as point predictions, as they can be high when the range
(see Table 3). This was a long experiment, so that the low levels of variables is different between the studies, as was the case here.
of energy (E) are not surprising. The probability of lapsing, gov- Nonetheless, by the standards of psychology (Richard et al., 2003),
erned by  in Eq. (8), controls the proportion of the ex-Wald that and of public health (Fig. 1 in Huesmann, 2007), these results are
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 637

Fig. 9. The response-time distributions of 17 ADHD boys (bottom panels) and 18 age-matched control boys (top panels) on a continuous performance task (Leth-Steensen
et al., 2000). The parameter in each column gives the ISI. NEMA draws the curves using the parameters in Table 3, and accounts for 97% of the variance in the data.

encouraging. Of course, cognitive slowing is not the only difference they measured. However, all of the relevant stop-signal response
between these groups (Willcutt et al., 2010a). times were linear functions of the go signal response times. Both
One fact that NeT must predict is that the performance of indi- improved under methylphenidate by roughly the same amount, so
viduals with ADHD will be both slower and more variable than that that there was little differential effect on the stop response. Because
of controls, and increasingly so as a function of increasing ISI. How- all responses are slower and more variable in ADHD persons, and
ever, a recent meta-analysis concluded that there was no effect of all are improved by methylphenidate, there is no reason to single
event rate on RT variability (Metin et al., 2012). The authors did find out a subset, be it inhibitory or excitatory, for special note.
the interactions that NeT predicts for mean and variance between The SSP is an ideal test of the hypothesis that poorer inhibitory
group and event rate, but the latter did not achieve statistical sig- control is the core deficit in ADHD (Barkley et al., 2007). It is also
nificance (p ranged from 0.03 to 0.11 in the within-study analysis). a test of NeT, as our theory predicts greater latency and variabil-
The data analyzed were the differences in standard deviations nor- ity in task reengagement, which has been noted (Oosterlaan and
malized by the pooled standard deviations—the effect size metric Sergeant, 1998b), but posits no special role for inhibitory processes.
d. Because standard deviations were in both the numerator and Three meta-analyses of these experiments have come to the same
denominator of d, and they were expected to be larger under this conclusion (Alderson et al., 2007; Lijffijt et al., 2005; Oosterlaan
manipulation for ADHD, such pooling is inappropriate, as it will bias et al., 1998): whereas there are marked differences in response
effect sizes low. In a different meta-analysis of effect sizes, Toplak times and variability between ADHD persons and controls, there is
et al. (2008) avoided such pooling because “an underlying assump- no evidence that inhibitory responses to the stop signal are a special
tion of this statistic [d] is that the impact of treatment will not class: the proportional delay in the stop signal necessary to achieve
change the homogeneity of variance of the two sample means being 50% successful inhibitions is not generally significantly different
compared”, and “that was well-documented to not be the case”. between ADHD and controls. “Deficits in ADHD reflect slower and
Therefore, Toplak et al. (2008) normalized by the standard devi- more variable responding to visually presented stimuli and concur-
ation of the control group alone. We predict that, had Metin et al. rent processing of a second stimulus, rather than deficits of motor
adopted a similar conservative strategy, they would have found the behavioral inhibition” (Alderson et al., 2008, p. 989).
interactions predicted by NeT (as in fact they did), and furthermore The slower and more variable reaction times in children
that they would have achieved statistical significance. with ADHD have alternately been attributed to “slower cogni-
tive processing (Kalff et al., 2005), slower motor speed (Van Meel
et al., 2005), deficient cognitive energetic resources (Sergeant et al.,
4.3. The role of inhibition: the stop task
1999), and deficient attentional processes (Lijffijt et al., 2005)”
(Alderson et al., 2008, p. 996). NeT is consistent with all of these
Some of the major theories of ADHD identify poorer inhibitory
attributions. Inhibitory signaling in the brain does not directly
control as a prime deficit. Mounting evidence, however, sug-
recruit extrinsic energetic resources (Chatton et al., 2003), but is
gests that motor inhibition (as a factor distinct from other motor
regulated by operation of the glutamate-glutamine shuttle, as glu-
responses) is not a major problem in ADHD (Alderson et al., 2007).
tamate is needed for GABA synthesis. That shuttle is activated by
There are, of course, other types of inhibition, but the evidence
(non-inhibitory) glutamatergic neurons (Liang et al., 2006; Calvetti
for their role in ADHD is mixed at best (Nigg, 2001; Shallice et al.,
and Somersalo, 2012). To the extent that those are handicapped by
2002). In the laboratory, inhibitory motor control is often assessed
energetic dysfunction, to that extent GABAergic circuits would be
using the stop-signal paradigm (SSP) (Barkley, 1997a; Lijffijt et al.,
consequentially handicappped.
2005; Oosterlaan et al., 1998). In SSP, on some of the trials the sub-
ject must interrupt an already-initiated response. To do so requires
the processing of the stop signal itself, and the initiation of a new 4.4. Working memory
response to suppress the response, or to decelerate the hand. Like
all response latencies, SSRT will be greater for ADHD (Nigg, 2001; Attentional allocation (Engle, 2002; Kane et al., 2001) and lapses
Sergeant et al., 2002). Scheres et al. (2003) found that inhibitory (Kenemans et al., 2005; Kofler et al., 2010) play important roles in
control in children with ADHD improved under methylphenidate working memory capacity. It should therefore not be surprising
compared to placebo in two out of three types of inhibition that that the largest effect sizes in laboratory tests distinguishing ADHD
638 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

Table 4
Parameters for the curves in Fig. 10.

Visuospatial Phonological

Parameter Control ADHD Control ADHD

E 0.077 0.005 0.102 0.068


k (C = k·SS) 13.8 7.8 15.1 15.1

a large difference between groups, and that the visuospatial task


was more difficult than the phonological one. The second and third
panels break this summary performance down into accuracy for
each set size.
How does NeT speak to such data? The core of the theory as
implemented in NEMA is its predication of differences in speed
of neural computation, with C representing the criterial number
of computations in terms of E, the energy available to speed them
along. These two parameters characterize a random walk with drift,
resulting in a Wald distribution of task completion. Many processes
are operating in this complex task, but the over-writing of mem-
ory for old stimuli by the output process and by the presentation
of new stimuli is among the most important (e.g., Killeen, 2001; Qi
et al., 2010). It is clear that the requisite number of neural computa-
tions must increase with set size, and that the more quickly stimuli
can be processed, the fewer that will be overwritten by the next,
while awaiting their turn to be used. Assume for simplicity that
the necessary number of calculations for each correct response, C,
is proportional to the set size that must be manipulated for it. In
this experiment, C calculations need to be completed during the
time from one stimulus onset until the next, 1 s later. The propor-
tion of calculations that will complete during the 1 s trial is given
by the area under the Wald distribution from 0 to 1 s. Multiplying
the area under the Wald (probability of completion) by the number
presented per trial (number of attempts) to get the expected num-
ber of stimuli correct, the dependent variables in Fig. 10. E differs
between groups, and C is proportional to set size. For the parame-
Fig. 10. The performance of a dozen ADHD (triangles) and control (circles) boys on ters shown in Table 4, NEMA draws the lines through those three
two working memory tasks (Rapport et al., 2008). NEMA draws the lines through panels of data.
the data using the parameters in Table 4, accounting for a median of 94% of the data
Even with the simplifications, NEMA is able to recover the data
variance.
within experimental sampling error. Criteria increase with set size
(SS), as k·SS, for instance, as 45, 60, . . ., etc. in the phonological task.
from control groups involved working memory (Martinussen et al., The energy available for the tasks in the ADHD group is substantially
2005; Swanson et al., 2010). Working memory is that facility which less than that available in the control group for the visuospatial
permits us to hold a number of stimuli in mind while performing task. It is clear that when the key processes are treated as time-
mental operations on them, while ignoring irrelevant stimuli. Typ- limited computations—here limited by over-written and fading
ical tasks involve attending to a stream of stimuli, and operating on memory traces, and the paced onslaught of new stimuli—simple
the set to generate a sequence of responses (Cowan, 2008). Even assumptions about differential speeds of computations in ADHD
in studies where SSRTs of ADHD participants are as fast as those and control samples permit good characterization of the data.
of controls, there may be a significant deficit in the former’s work- Why should C increase proportionate to set size? This param-
ing memory (Clark et al., 2007; in this case the working memory eter typically blends both the demand characteristics of the task
of those with ADHD was nonetheless correlated with their SSRT). with the resources that the individual brings to bear. As set size
An experiment by Rapport et al. (2008) provides a clear example, increases, there is much more to process—each position must be
and dramatically displays differences between ADHD and control held in mind for each stimulus before any response can be made.
populations. In one visuospatial condition these authors presented If a single stimulus-location memory requires 15 computations, 6
a sequence of 3, 4, 5, or 6 circles (the set size, SS) in a 3 × 3 matrix will require 90. Within the confines of a 1-s trial, this means that the
on a computer screen, at the rate of 1/s. Thereupon subjects had to probability of a successful completion for each one—the area under
reproduce the positions of the stimuli on a keyboard matrix, in the the Wald from 0 to 1 s—will decrease as that distribution flattens
same order, except that the single (randomly chosen) stimulus of with increases in C. In the visuospatial task, where energy of the
color had to be entered last. In a phonological task, a sequence of ADHD group was severely compromised, there was also an appar-
3 . . . 6 symbols, all numbers except for one letter, were presented. ent decrease in the proportion of computations that they made.
The subjects had to say the numbers they had seen, but rearranged (There are not enough data here to be sure of this, as k could be held
into ascending order, with the letter being said last. constant at 13 for both groups and incur only a few percent decrease
Fig. 10 shows the results of this experiment for 12 ADHD boys in r2 .) Interestingly, the phonological task made less severe ener-
and 11 TDC, as presented in the authors’ Fig. 3. The top panel shows getic demands on both groups, and separate values of k provided no
the number of stimuli correctly reported per trial for each group, on improvement in fitting those data. The larger difference between
visual and phonological working memory tasks. This demonstrates ADHD and control groups for the visuospatial task is consistent with
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 639

Fig. 12. Variability of 11 ADHD (triangles) and 11 control (circles) boys on a syn-
Fig. 11. The parameters of ADHD (triangles) and control (circles) children on three chronized tapping task. Data from Rubia et al. (1999). The predictions from NEMA
working memory tasks of differing difficulty (Buzy et al., 2009). The values of these account for 57% of the variance in these data.
energetic parameters are inferred via Eq. (9) from the authors’ report of the param-
eters of the ex-Gaussian distribution of response times.
The criteria for emitting a response (C) dropped across both groups
as difficulty increased, most markedly for the ADHD group. This
the meta-analyses reported earlier (Martinussen et al., 2005), and is
should lead to more errors—as in fact it did: accuracy for the ADHD
a byproduct of the differences in demands of the task and resources
group was lower than for controls, and decreased linearly with load.
of the individuals.
Whereas we might expect the criterion to increase with task dif-
Although this implementation of NEMA is straightforward, it
ficulty, under the time-demands of fading memory traces these
is rare to find reports of experimental data that are sufficient to
subjects made a necessary speed-accuracy tradeoff by skimping
effect like analyses. Fortunately, there is a simple way to reinterpret
on processing in the most difficult conditions. The same situation
archival data. Increasingly researchers are reporting the parame-
might occur in the classroom: student-paced work can allow ade-
ters of ex-Gaussian densities fit to their RT data. The ex-Gaussian is
quate time for children to muster the resources necessary for a
a widely used descriptive distribution. It utilizes three parameters,
high criterion, whereas instructor paced work may entail the speed-
rather than the two of the Wald (in the simple model); or the three
accuracy tradeoffs manifest as reduced values of C seen in Fig. 11,
(of the sessional model, which must add  to track waning energy);
and characterized as impulsive behavior.
or the four (of the attentional model, which must add non-zero
Visuospatial working memory is one of the most effective labo-
values for  and ). It is possible, however, to approximate many
ratory discriminators between ADHD and TDC (Martinussen et al.,
data with the simple drift/Wald model, as was shown in Fig. 4. The
2005; but see Marzocchi et al., 2008; Nigg, 2005). Speed of visual
two parameters of that model are easily inferred from those of the
information processing has been shown to be a predictor of fatigue
ex-Gaussian by the method of matching moments (see Appendix B
in elementary and junior-high students (Mizuno et al., 2011). Both
for details):
reduced response speed and increased psychological fatigue may

+ be a common outcome of neural fatigue, caused by an inadequate
E= lactate supply to critical neural networks. Verbal working memory
2 + 2 (9)
is also affected in ADHD, but to a lesser extent than visuospatial
C = ( + )E (Martinussen et al., 2005). The processing component of working
All of the Greek parameters here belong to the ex-Gaussian. memory tasks share a common resource pool, with storage aspects
The utility of this recoding is seen with the data of Buzy et al. depending on domain-specific verbal and visual resources (Alloway
(2009), who studied 25 children with ADHD and 24 TDC on a visual et al., 2006, p. 1698). Why visuospatial storage should be more
serial addition task with different task difficulties. The task required affected than phonological/verbal storage in ADHD is uncertain,
the addition of two sequentially presented numbers, and judgment although ADHD persons also have relatively greater difficulty in
of whether that sum equaled a third number presented simulta- visually cued timing tasks than in auditorially cued timing tasks
neously with the second. The task difficulty was manipulated by (Toplak and Tannock, 2005).
using small (sum less than 5), medium (sum less than 10) and large In an extensive overview of dual-process theories of psycho-
(sum less than 19) digits. Fig. 11 shows the values of E and C imputed logical processes, Barrett et al. (2004, p. 554) observe that the
from Table 2 of Buzy et al. “ability to engage in controlled processing in attention-demanding
The energy available for allocation did not vary much across task circumstances, especially those that require the suppression or
difficulty (unfilled symbols), but was uniformly lower for ADHD inhibition of automatic processing, is related to individual differ-
than for control, as expected from NeT. On the average, the ADHD ences in [working memory capacity]”. Fig. 10 provides evidence
group commanded only 75% as much energy as did controls. We that differences in that capacity may be attributed to deficiencies
might have expected a greater covariation of energy with task dif- in speed of neural computation (E), and a compromised number
ficulty. Buzy et al. (2009) increased memory load over two 2–3 h of computations (C) that can be perfected within a temporal enve-
sessions. They did not randomize exposure because pilot experi- lope shaped by fading traces and ongoing stimulation. This close
ments had shown that many children first exposed to the high task association of processing speed and working memory is attested
difficulty would balk. That they didn’t balk when finally exposed to by recent analyses utilizing drift models (Karalunas and Huang-
that condition suggests that learning had occurred, which might Pollock, 2013).
have reduced the energetic demands of the moderate and high
difficulty tasks experienced later in testing. 4.5. Motor control
Given fixed resources, different for the two groups, increased
task difficulty entails reduced numbers of computations, as seen One of the areas of the brain affected in ADHD is the cerebellum
in the top of Fig. 11. This is the classic speed-accuracy tradeoff. (Berquin et al., 1998; Hart et al., 2012), which is implicated in simple
640 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

a standard timing model (Killeen, 2002; Killeen and Taylor, 2000).


This model predicts a linear relationship between the mean and
standard deviation, with the slope of the function being the Weber
fraction, w = [p(1 − p)]1/2 . The parameter p gives the probability that
transmission of information from one step in the accumulation
process to the next will occur without error. That probability is
p = 0.957 for controls and 0.919 for ADHDs. (These probabilities
can be related to the area under appropriate Wald distributions
for each of the composite steps, a nuance not indulged here.) Very
similar data were reported by Kerns et al. (2001), with comparable
compatibility with the timing model.
The decreased probability of transmitting timing information
from one stage to the next in persons with ADHD is consistent with
research that shows an important role for attention in the timing
Fig. 13. Variability of 101 ADHD + ODD teenagers (triangles) and 39 control (circles)
teenagers on a time reproduction task. Data from Barkley et al. (2001). The lines are process (Buhusi, 2003; Fortin, 2003). When attention is distracted
from a timing model that assumed that the probability of missing an internal time by concurrent stimulation or tasks, both the mean and variance of
signal was greater for ADHD (8%) than for controls (4%). They account for 99% of the temporal estimates or productions are affected, either due to the
variance in the data. failure to reset or increment the counts, or to fading control by
earlier count information due to competition in memory from con-
conditioning, motor control, and timing of stimuli and responses in current tasks (Buhusi and Meck, 2009). Thus, over two wide ranges
the deci-second range. Simple motor coordination tasks show reli- of timing tasks, models based on NeT and NEMA (Fig. 12) or at least
able differences between ADHDs and TDCs (e.g., Jacobi-Polishook consistent with them (Fig. 13) provide good (if cursory) descrip-
et al., 2009; Killeen et al., 2012; Kooistra et al., 2009; Meyer and tions of the data. Timing in ADHD persons is certainly handicapped
Sagvolden, 2006; Sergeant, 2005). In an early study of motor tim- by attentional, memorial and processing limitations. It is less clear
ing, Rubia et al. (1999) asked 11 ADHD and 11 control boys to tap that it is a handicapper, however—that is, that it plays a causal role
in synchrony with visual stimuli, which appeared at one of five in other deficits.
intervals. The abscissae of Fig. 12 displays the average produced
intervals, and the ordinate the standard deviations. NEMA predicts
a tight relationship between these variables, governed by the avail- 4.6. Reinforcement frequency
able energy and criterion. Given that the mean of the produced
interval is M = C/E, and variance is C/E3 , we can replace C with ME Rapidly paced trials clearly result in better performance for both
√ TDCs and ADHDs, shown in Figs. 7 and 8; this is a general find-
in the latter and predict that the standard deviation  = M/E. The
curves show these values, under the assumption that E for con- ing. According to NEMA, this effect is due to the recall of attention
trols is 0.24, and that for ADHDs is 63% of that value, 0.15. Some by trial initiation, and its subsequent drift (Fig. 5). Other events
of the deviation of the data from these trend lines may be due to that focus attention include the presence of the experimenter, the
sampling error in these small samples. A similar experiment (Rubia appearance of target stimuli, and the delivery of reinforcers. There
et al., 2003) and analysis returned comparable values of E of 0.28 are numerous demonstrations in the literature that more frequent
for TDC and 0.13 for ADHD. When the ADHDs were given placebo, presentation of stimuli (sometimes misleadingly called incentives)
single, or double doses of methylphenidate, E increased from 0.14 is more effective than infrequent presentation for both ADHD and
to 0.16 to 0.24, the last value close to that of controls in this and the TDC. Nevertheless, there are few demonstrations that the mecha-
prior study. nism is through reinforcement, or otherwise incentivizing the task.
Tiffin-Richards et al. (2004) conducted a variety of tapping tasks Reinforcers typically signal trial initiation, they should work as well
and found no differences between ADHD and control groups. Their as other salient stimuli. There is little evidence that they lead to bet-
task typically involved only 12 repetitions in the synchronization ter performance, or that they differentially increase motivation in
phase (as opposed to 60 in the study of Rubia et al. shown in Fig. 12), ADHD (Oosterlaan and Sergeant, 1998a). A major review of the lit-
which, as the authors noted, may have been inadequate to bring out erature (Luman et al., 2005) found the results to be mixed, with
systematic differences—differences that we predict should start out some evidence that frequent scheduling of reinforcement effected
minimal and increase with increasing time on task. A recent meta- a greater improvement in ADHD than TDC—but ceiling effects on
analysis (Noreika et al., 2013) demonstrated that individuals with TDC limited the interpretability of those results. When feedback is
ADHD had difficulty in a variety of tasks requiring timing of short, given on every trial, neither the frequency nor magnitude of rein-
intermediate and long intervals. forcement has any effect on speed of learning in ADHD children
Timing of longer intervals consists of a concatenation of elemen- (Luman et al., 2009). Thus, it is not clear that frequency of rein-
tal units akin to those involved in Fig. 12 (Cassenti, 2011), with the forcement needs to be considered to improve our theory’s account
Wald distribution governing the variance of the elements, and the of the data.
rules of their combination (Killeen and Weiss, 1987) determining The inability of individuals with ADHD to sustain attention is
their Weber functions (Fetterman and Killeen, 1990). An example often taken as prima facie evidence for their lack of motivation,
is given in Fig. 13, whose data were reported by Barkley et al. (2001) or lack of investment of effort. But appropriate contingencies of
for 100 teenagers with ADHD and ODD, compared to 39 matched reinforcement work at least as well with ADHD children as with
controls. Participants were given six intervals, ranging from 2 to controls (Drechsler et al., 2010; Luman et al., 2005). NeT claims
60 s, to estimate and reproduce. The authors reported the absolute that these tasks are harder for individuals with ADHD due to their
deviation from target time. In a Gaussian distribution, the aver- inadequate energy resources: it is not lack of motivation, nor lack
age absolute deviation from the mean (AD) is proportional to the of appreciation of reinforcement, which differentiates them from
√ their controls—it is their inability to maintain monotone behavior
standard deviation (AD = (2/)SD). Barkley et al. computed devia-
tions from the target duration, so his measures include any constant in the face of rapidly declining resources. This may lead to future
error in the participants’ responses. Nonetheless, the measures will, avoidance of situations that place such painful demands on them
to a first approximation, be comparable and may be analyzed with (Sonuga-Barke, 2005).
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 641

4.7. Delay discounting (exponential) decay of utility as a function of psychological delay


(Killeen, 2009), will accommodate delay-discounting data from
For a delayed reinforcer to make contact with a response, ADHD participants:
some memorial trace of that response must be present when rein-
ˇ
forcement occurs (Killeen, 2011). Where working memories are vt = v0 e−kt (10)
foreshortened by energetic insufficiency, we expect steeper delay
of reinforcement gradients (Johansen et al., 2007, 2009). Given a Here, v0 is the immediate value, k is the discount rate, and ˇ (beta) is
choice between small immediate and larger delayed outcomes, the psychophysical parameter for future time. Typically, ˇ is around
ADHD persons often prefer the immediate outcome somewhat 0.5 for young adults, less for children, and increases with maturity.
more than do controls (Luman et al., 2010; Marco et al., 2009; In the three studies we have analyzed (Crean et al., 2000; Hurst
Sonuga-Barke et al., 2008), which is evidence either of steeper gra- et al., 2011; Paloyelis et al., 2010a), Eq. (10) provided an excel-
dients or of a blunted effectiveness (i.e. faster decreasing marginal lent fit to the data, did not require different parameter values of
utility) of larger payoffs. But there are inconsistencies in this ˇ between ADHD and TDC, and showed substantially greater dis-
literature, suggesting that when real delays are involved, and counting, indexed by k, for ADHD. (Given the dependence of ˇ on IQ,
other factors such as age and intelligence are controlled for, the if that is not matched, some of the steeper discounting would be due
effects can be small (Banaschewski et al., 2012; Scheres et al., to lower IQ in the ADHD sample.) Given the generally greater liabil-
2006). ities of ADHD-Combined subtype seen in the present analyses, it is
The above research involves real delays to real not surprising that this is also the subtype most likely to show the
goods—experiential discounting. It plays out over a much smaller steepest discount functions (Paloyelis et al., 2010b; Scheres et al.,
range of delays and magnitudes of goods, and delivers discount 2008).
gradients that are steeper by orders of magnitude, than other The difficulty in maintaining fixed attention over extended
research included under the rubric delay discounting. When a periods that persons with ADHD experience would reasonably
person is given a hypothetical choice to buy or sell a good some make them averse to entering situations where such extended con-
time in the future, the value of that good is discounted as a sideration is necessary. Like Sonuga-Barke (2002, 2003), we suspect
function of the delay incurred. This is rational: when an outcome that learned aversions to the extended attentional requirements
is long delayed, we are denied the use of it—and of the money of deferred outcomes may be the mechanism of steeper discount-
we trade for it—until the delay has elapsed. Discount functions ing in the hypothetical tasks. It is perhaps therefore not surprising
are graphs of the present value of a delayed good, relative to that methylphenidate reduces experiential discount rates but not
the value of the same good delivered immediately, plotted as hypothetical discount rates (Shiels et al., 2009), which should
a function of delay. An individual whose discount function is have greater inertia due to learning history (Nevin and Grace,
steeper than the norm is often called impulsive (even though his 2001).
decisions may be deliberate, well-considered, and perhaps even
prudent). Various classes of individuals have been shown to be
more impulsive than the norm, or than their contrasting control 4.8. Spectral analyses
group. Young individuals discount more steeply than older ones,
poor more steeply than rich, addicts more steeply than sober, less The present analysis predicts slowing of response time through
intelligent more steeply than more intelligent, and ADHD more extended performance, and predicts that the slowing will be greater
steeply than TDC (Madden and Bickel, 2009; Paloyelis et al., 2009). for ADHD than for TDC. This is a common finding (e.g., Berlin
Evaluation of imaginary deferred outcomes, or forbearance during et al., 2003; Huang-Pollock et al., 2012; Johnson et al., 2007), more
the delay to real ones, would be difficult for individuals with ADHD evident in response times and event-related potentials (Heinrich
during the delay. Experience has inculcated the attractions of the et al., 2001) than other dependent variables. Spectral analyses
immediate—especially when “deferred” often means “denied” using fast Fourier transformation of a series of response times will
(Biederman et al., 2004). reflect that sessional difference, showing more energy (informa-
There are two importantly different ways to frame the behav- tion, power) at low frequencies (e.g., Gilden and Hancock, 2007),
ior of the subjects in these situations. Consider an automobile because those frequencies correspond to slow changes over the
whose delivery is delayed for 3 years. The individual may com- course of the session that we here identify as attentional fatigue.
pute the current dollar value of the automobile, and discount If the series of response times is detrended to remove such fatigue
that dollar value. Or he may compute the utility of that car effects and render the series stationary, NeT predicts that the differ-
now vs. 3 years hence, and base his decision on that evalua- ences in spectra between ADHD persons and controls will decrease
tion. Individuals tend to discount consumable goods at different or disappear. If the data are only approximately detrended, by
rates than money (Estle et al., 2007). If they converted to dol- analyzing residuals from a linear regression (e.g., Johnson et al.,
lar value and discounted that, however, this would not happen. 2007), the overall effect will be reduced, and may be moved from
Therefore people discount utility, not dollar value. But if util- lowest to slightly higher frequencies, depending on where the
ity is not a linear function of monetary value (and the law of deviation between the fatigue trend line and its linear approxi-
decreasing marginal utility (Bernoulli, 1738/1954) stipulates that mation is greatest. As an example, picture a straight line drawn
it is not), then goods of large magnitude must be discounted at a through the sessional trends of Fig. 6. The deviations between
lower rate in time than goods of small magnitude (Killeen, 2009). those fatigue curves and the straight line will be zero at two
This effect has been found, and christened the magnitude effect points, but loom large about 30% of the way into the session.
(Green et al., 1997). Experiments that vary both the delay and When these residuals are transformed to the frequency domain,
the magnitude of a delayed good permit us to concurrently estab- they will indicate unusual power corresponding to that period.
lish discount rates for increasing time and increasing magnitude, Absent a rationale for doing otherwise, however, NeT predicts
but there are few such that use ADHD subjects (Barkley et al., that with a properly detrended time series (e.g., by analyzing the
2001; Paloyelis et al., 2010a)—too few to draw firm conclusions residuals from a power function such as Eq. (4)), there will be
about their rate of magnitude discounting, especially so given the no substantial differences between ADHD and TDC in their power
genetic differences that add variance to these measures (Paloyelis spectra. Recent analyses validate this prediction (Karalunas et al.,
et al., 2010a). Therefore a simpler discounting model, a rational 2012b).
642 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

4.9. Executive functioning we can certainly manipulate them in an ordinal fashion within
the same type of task, as shown previously in Fig. 11. We may
Epstein et al. (2011) extended the above handful of tasks by ask, however, how well the following characterization will predict
studying a battery of neuropsychological tasks for ADHD (51 com- performance: ADHD will have less available energy, and whatever
bined subtype, 53 inattentive subtype), and control (47) children. computational resources TDC bring to bear on each of the tasks,
The tasks were: (a) The Attentional Network Test (ANT, Rueda ADHD will commit proportionally less. In particular, assume that
et al., 2004) that involved a difficult discrimination of the orien- the Inattentive subtype has only 84% of the energy of TDC, and com-
tation of stimuli on a screen in the context of distractors; (b) a mand 86% of the resources (C) that they do. And assume that the
stop-signal task (Verbruggen and Logan, 2008), with “go” responses Combined subtype is worse off, with 77% of the energy of TDC,
interrupted on 25% of the trials by a stop signal, presented at a lag and 83% of the resources. These generalizations predict the data,
sufficient to inhibit just half the responses; (c) an N-Back task, in as shown in the right panel of Fig. 14. This means that, as far as
which subjects had to press a key if the current stimulus was the may be inferred from response-time distributions, these tasks did
same as the prior one, which happened on 30% of the trials; (d) a not strongly interact with subtypes: that is, they did not contribute
Choice task, in which subjects had to press one key if a circle was unique information.
presented, and another if a square was presented, which happened It is obvious that the parameters depend on an interaction
with equal frequency; and (e) a Go/No-Go task, which required that of the task difficulty (Fig. 14), time-constraints on the resources
subjects hit the space key for target stimuli—any letter but X—and that may be allocated (whether those are dictated by task pac-
not respond when an X was presented, which happened on 10% of ing or memory decay) and characteristics of the individual subjects
the trials. (Figs. 14 and 15). We surmise that insufficiencies in energy—here
The authors presented stimuli paced at 1, 2 and 4 s, but did 84% and 77% for the ADHD subtypes—are characteristic traits,
not break out all performances by those lags. They reported mean, whereas resource allocation is also a function of task, setting,
standard deviation and coefficient of variation on all tasks, and training and motivation of the individuals. The former is more
the ex-Gaussian . Those summary statistics were converted to immediately affected by pharmacotherapy, the latter by accuracy-
the model parameters by Eq. (9), and are shown in Fig. 14 (left speed tradeoffs that may have become habitual, but which the
panel). Neural computations C were greatest for the ANT task, breathing space given by increased energy allows to shift toward
and least for the simple Go/No-Go task. That processing showed a accuracy. When errors of commission are made on such tasks,
very similar pattern across populations, with inattentive subtypes there is often a post-error slowing, suggesting the recalibration and
investing proportionally less than controls, and combined subtypes increase in resources dedicated to the task (see the review by Shiels
slightly less than inattentive. A similar pattern was seen for ener- and Hawk, 2010).
getic support (E), with the only nonmonotonicity being for the The agreement between data and theory shown in the right
choice task, which was less energetically demanding than N-back, panel of Fig. 14 used a substantial number of degrees of freedom
but to which no more or less was given in the way of computational from the data, being based on values of C and E inferred from the
resources. control group, and proportionate decreases in them for the ADHD
Unfortunately, we have no way of predicting a priori the groups. With those values in hand, however, it may be possible
demands that various tasks will place on resources, even though to project results in similar experiments. Klein et al. (2006) tested
55 children with ADHD, most Combined subtype, and matched
controls, on the stop-signal, Go/No-Go, and 1-back tests. The pre-
dictions for similar conditions and groups from the Epstein et al.
(2011) experiments (the ordinates in Fig. 14) are plotted as the
abscissae in Fig. 15. Against them are plotted the data obtained by
Klein et al. It is clear that the predictions fail, as the length of the
y-axis is 60% that of the x-axis. In the Klein study the trials were
much more rapidly paced, and testing was briefer, putting gen-
tler demands on energy. If these subjects, older by an average of
2 years, commanded 37% more energy, and required only 84% as

Fig. 14. Left panel: The resources expended on each of the tasks (C), and the energy
available (E) to reach those criteria. ANT = Attentional Network Test; S-Sig = Stop Sig-
nal; G/N-G = Go/No-Go. Right panel: Assuming proportional insufficiencies of energy Fig. 15. The predictions for the conditions of Epstein et al. (2011) shown in the
(84% for Inattentive, and 77% Combined type) and computation (86% for Inattentive, previous figure are applied to data of Klein et al. (2006) found in similar assessments.
83% for Combined type) in each of the groups recovers the data, accounting for over Triangles represent ADHD performance, circles controls. Despite the obvious strong
90% of its variance. correlation, the predictions account for none of the variance in the data, which they
Data from Epstein et al. (2011). systematically overestimate (note the different ranges of the ordinates).
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 643

cluster around two major dimensions: attention deficit, and


hyperactive–impulsive, with many children satisfying the criteria
for both (ADHD-Combined subtype). There is an important unitary
component to ADHD symptoms as well as those separable dimen-
sional (Normand et al., 2012), one that we hypothesize may have
something to do with a shared energetic insufficiency.

5.1.1. Attention deficit


Individuals with ADHD do not really have an attention
deficit—they have an attention surfeit: they attend too well to new
stimuli, quickly shifting attention to novel features of their environ-
ment (Landau et al., 2012). “Is often easily distracted” and “Often
has trouble keeping attention on tasks” are two of the DSM-IV crite-
Fig. 16. The resources expended on a stop-signal task (filled circles) and the aver- ria for categorization. It is concentration—sustained attention—that
age energy available (open circles) as a function of the dosage of methylphenidate is problematic in ADHD. Sustained behavior requires sustained fir-
(MPH). ing of functional groups of neurons. But after 10 s those neurons
Based on data from Scheres et al. (2003). require additional supplies of lactate, and it takes several minutes
for astrocytes to fully ramp up production to whatever level they
may. NeT entails a deficit in sustained monotone activity, be it
many computations, NEMA’s predictions would account for over
sensory, cognitive or motoric. Lapses of attention follow from the
77% of the variance in the data. Shifts in the key parameters are
cumulating energetic cost of sustaining attention, manifest in con-
not unexpected in light of those differences in age and pacing. NeT
trols, but more marked in ADHDs (Alberts and van der Meere, 1992).
therefore cannot predict a priori the exact values for E for ADHD
Lapses are preceded by reduced activity in the anterior cingulate
vs. TDC. Furthermore, in the effort to be parsimonious, we ask the
and right prefrontal regions (Weissman et al., 2006). During these
key parameters to absorb the effects of other potential modera-
lapses, the original neural complexes can begin to recover from the
tors (e.g., training trials, embedding in other tasks, length of trials,
prolonged demand. Individuals with ADHD can strategically time
presence or absence of experimenter, age, degree of severity of
their inattention. This can leave them better situated to exploit
symptoms, etc.), and this will increase their variability from one
attentional resources when most needed, as toward the end of a
study to the next. Until we can make all such predictions precisely,
long scheduled delay (Börger and van der Meere, 2000). On the trial
however, based on the conditions of the test and the characteristics
before an omission error on such tasks, response times lengthen.
of the subjects, or of the underlying neural information processing
After those omissions, they return close to baseline (Epstein et al.,
(Crumiller et al., 2011), this theory, and all that attempt similar
2010). The lengthening suggests fatigue and increased probability
predictions, may be considered incomplete. Thus, the difference in
of lapsed attention. The post omission quickening may be inter-
estimates for E and C between experiments such as those analyzed
preted as a restoration of energy, as a secondary benefit of that
here remains a target for future exploration.
lapse. Shifts of attention in self-paced tasks (including loosely mon-
Scheres et al. (2003) reported the mean and variance of response
itored classroom tasks) are more frequent for children with ADHD
times on a stop-signal task under various doses of methylphenidate.
(Kofler et al., 2008b) and may mitigate some of the fatigue effects
Fig. 16 shows the values of C and E computed from those statistics.
(van Mourik et al., 2007). (The variety of stimulation available in
It is clear that available energy increases with methylphenidate
electronic games may sustain attention due to the quickly shifting
dosage, allowing additional precision in computation while still
nature of their targets and sub-goals.) It is such shifts in utiliza-
empowering faster response times.
tion of alternate functional units that contribute substantially to
response variability (Prado et al., 2011). As children mature, they
5. Embedding NeT in a causal framework may condition alternative functional networks (Deco et al., 2010),
which can relieve the primary ones. Perhaps this is one of the rea-
Killeen et al. (2012) proposed an explanatory framework for sons for the decrease in the prevalence of ADHD in adulthood.
ADHD based on the four kinds of knowledge that comprise under-
standing: that of the causes, substrates, functions, and of the models 5.1.2. Hyperactivity
by which we define and interpret a thing. Each kind has both a prox- Whereas a deficit in sustained attention is a core prediction of
imate (immediate, molecular) core and an ultimate (longer-term, NeT, that theory does not predict hyperactivity. The best that we
molar) shell. The present paper has been concerned with the prox- can offer is a post hoc hypothesis. When attention to a primary
imate material causes of ADHD, the machinery that appears to be task becomes difficult due to energy depletion, shifts of attention
broken. It is important to situate this perspective within the other to other stimuli are inevitable. We predict a much higher rate of
kinds of knowledge: formal (definitions, models and theories), effi- endogenous task switching in ADHD than in TDC. Whereas this
cient (causes of symptoms and syndrome), and final (what sustains prediction may seem obvious (Kofler et al., 2008b), laboratory evi-
the behavior, and what sustains the syndrome). dence for it is scant, although higher levels of entropy (a measure of
diversity of behavior) have been demonstrated in an animal model
5.1. Proximate models of ADHD: its definition of ADHD (Johansen et al., 2007). Some of the diversity of behav-
ior called hyperactive may be attention/task-shifting, driven by the
The definition of ADHD has evolved over the years, along with quick dissipation of energetic resources for sustaining monotone
the criteria for categorization. ADHD shares many comorbidities, tasks. Some of it may be a poorly socialized manifestation of play
being more common in association with other psychiatric cat- behavior (Panksepp, 2007). It is strongly context-dependent (Dane
egories than by itself. Anxiety, depression, oppositional-defiant et al., 2000). As a reviewer noted, the data in Fig. 4 show that E
disorder, conduct disorder, and learning disabilities are not was not much lower for HI than for controls, and was larger than
uncommon correlates of ADHD. Clinical definition of ADHD has for the other categories. Hyperactivity–impulsivity is clearly asso-
evolved from that of Kramer and Pollnow (1930) to a list ciated with lower values of C, inordinate to the modest decrease in
of symptoms validated in careful clinical histories. Symptoms E that such individuals may suffer.
644 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

Rapport et al. (2009a) and associates (Alderson et al., 2012)


have noted that increased levels of activity in ADHD are correlated
with demands on working memory. Heightened motor activity may
increase sympathetic tone, providing the arousal necessary (Barry
et al., 2005) to a hypoactivated prefrontal cortex (Fan et al., 2012).
It is conceivable that HI is a subtype that is able to use such activ-
ity to stimulate NE mediated energetic resupply, much as some of
us will bounce a foot to maintain attention during a long collo-
quium, or manuscript such as this. But there is another possibility,
one that reverses this causal arrow. It has been established that
activity of the locus coeruleus norepinephrine arousal system is
necessary for optimal attentional performance (Arnsten and Rubia,
2012; Dietrich et al., 2012; Howells et al., 2012). To the extent that
activation of the noradrenergic system generates motor activity,
Rapport’s analysis is consistent with our theory, and may explain Fig. 17. The inferred values of C from RT data are plotted as a function of the individ-
some of the vigor of behavior called hyperactivity. The decrease in uals’ score on a personality inventory of impulsivity (data from Logan et al., 1997).
impulsivity scores under task demand (Raiker et al., 2012) sup- The predicted strong negative correlation between impulsivity and C is manifest.
ports the hypothesis that such tasks provoke the locus coeruleus to
release norepinephrine in the prefrontal cortex (Aston-Jones and
using a 9-point personality inventory. Fig. 17 displays the values of
Cohen, 2005; for a recent review of the role of the locus coeruleus
C inferred via Eq. (9) from their data for individuals scoring at each
in ADHD, see Imeraj et al., 2012). Hyperactivity may thus be a
level of the inventory. The correlation between C and impulsivity
byproduct, rather than instrumental component, of this syndrome.
was greater than the correlation between mean RT and impulsi-
Possibly, of course, causality flows in both directions (Kievit et al.,
vity. This study validates our interpretation of the criterion C as a
2011).
(inverse) measure of impulsivity.
5.1.3. Impulsivity
As inattention is a sin of omission, impulsivity is a sin of commis- 5.1.4. Comorbidity
sion. The many definitions of impulsivity (Evenden, 1999; Robbins In populations at large many different kinds of abilities (e.g.,
et al., 2012) are helpfully reviewed by Dalley et al. (2011). In the case selective attention, sustained attention, episodic memory retrieval,
of ADHD, the three that are diagnostic in the DSM-IV are “often problem solving, spatial working memory, etc.) are positively cor-
blurts out answers before questions have been completed; often related. This robust phenomenon has been called the “positive
has difficulty awaiting turn; often interrupts or intrudes on others”. manifold”. One explanation for it, stemming from early factor
In addition, many of the symptoms listed under hyperactivity (e.g., analysis and later structural equation modeling, is based on the
“often leaves seat in classroom”) could as easily be called impulsive. ability to statistically refer these abilities to a principal common
All of these symptoms may be thought of as a “failure of inhibi- factor. Spearman (1904) called the factor g, identifying it with
tion”, which motivated one of the major theories of ADHD (Barkley, general intelligence. It was a small step to reify the concept, and
1997b). But failures of inhibition have been harder to document in then to search for the causes of variation in g. There is another
the laboratory in paradigms such as the stop-signal task (Alderson interpretation of the manifold, however, due to a contemporary
et al., 2007; Winstanley et al., 2006). of Spearman. Instead of the manifold being caused by some sin-
The lower energetic resources of ADHD persons, and the (pos- gle “thing” such as a general computational ability (g) affecting
sibly compensatory) decreases in the caution (C) with which a all tasks, each task constituting the psychometric battery may
response is made, entail responding “before all of the data are in” sample resources from elements of a distributed neural system
(Sergeant and Scholten, 1985). Given the smaller working memory (Bartholomew et al., 2009). If there are strengths (or weakness)
capacity of the typical individual with ADHD, a response delayed in disparate parts of the brain that are shared by different tasks, the
is often forgotten. Prudence may dictate a response at the point task performances will be positively correlated. There need be no
at which more starts being lost than retained. Balancing the pros common latent factor that causes genius—or dysfunction—no cen-
and cons of an action, while holding it in abeyance, is often beyond tral executive or working memory module that is better or worse
the reach of a compromised working memory, just as is conjur- than average. All that is required is mutual exploitation of various
ing the long-term consequences of short-term action. If delayed unnamed parts of the brain that are better or worse than average
goods are discounted, then delayed consequences in general will (Marcus, 2006), parts that will vary between individuals and tasks
be discounted. This all may occur for the same reason: a decrease (Uttal, 2012). Such covariation is manifest also in the manifold of
in the criterion for responding, possibly as a compensation for the correlated symptoms within and across DSM categories (Cramer
deficit in energetic resources necessary to persist in carrying out et al., 2010). If distributed systems analysis of the manifold is cor-
the computation; possibly as an attempt to exploit memorial traces rect, it is unlikely that an endophenotype of ADHD can be found,
before they have faded beyond recall; eventually, possibly becom- and even less likely that it would have any single name.
ing established as a character trait. Vaughn et al. (2011) found some If neuroenergetic insufficiency is a cause of ADHD, there is no
evidence for the long-term persistence of impulsivity in ADHD chil- reason to expect its action to be restricted to just those parts of the
dren and controls at intervals of 2 and 3 years. Analysis of their CPT cortex that govern the behaviors associated with the DSM-5 crite-
data revealed increases in E and C for the controls; in contrast, the ria for ADHD. Increased variability of performance over intervals
values of E and C were smaller for ADHD children, and showed little of time is seen in many psychiatric disorders. The semi-random
to no improvement over time. failure of myelination or energy transport mechanisms in different
We have asserted that smaller values of C are the hallmark of regions of the brain (Cherkasova and Hechtman, 2009) may give
impulsive responding. What is the evidence for that assertion? In an rise to both the similarities and the differences between ADHD and
important article on impulsivity and inhibition, Logan et al. (1997) the many other psychopathologies that may be comorbid with it
measured the RT to the go signal in a stop-signal task. A novelty (Larson et al., 2011). Consider again the data collected by Rucklidge
of their study was an independent measurement of impulsivity and Tannock (2002) on children with ADHD, reading disabilities
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 645

consistent with the dynamic-developmental theory (DDT, Johansen


et al., 2005; Sagvolden et al., 2005) in that it predicts steeper delay
of reinforcement gradients, as a byproduct of the over-writing of a
more limited working memory by events that intervene between
stimulus and response—or between response and reinforcement.
The posited mechanisms are different, however: in DDT those are
dopamine-based (Johansen et al., 2002), and in NeT norepinephrine
based. As one of the proponents of DDT noted, “A challenge for
reinforcement theories of ADHD is to link the concepts of memory
and attention used in our analyses of behavior to the corresponding
concepts used in cognitive psychology” (Johansen et al., 2009). The
present theory takes a step in that direction.
Our theory is also consistent with the working memory the-
ory of ADHD (e.g., Kofler et al., 2008a, 2011; Sarver et al., 2012),
Fig. 18. Ordinates: The time to name 50 numbers, letters, colors and objects in an which posits deficiencies in working memory to be the core feature
experiment by Rucklidge and Tannock (2002). Abcissae: The predictions from NEMA, of ADHD, from which others devolve. NeT, however, views com-
based on the energy available to participants on a stop-signal task, and the number promised working memory as a sequela to energetic insufficiency,
of computations C necessary for TDC on the naming tasks.
which may directly affect other processes (e.g., motor coordination)
without mediation by working memory. The effects of ADHD per-
(RD), both, and TDC. In addition to the stop-signal RT task, from vade basic processes, as well as higher-order ones such as executive
which we inferred values of E and C mentioned above, they con- function and working memory (Mulder et al., 2010).
ducted four rapid naming tasks, for 50 numbers, letters, colors and Tripp and Wickens (2008) recently promulgated a dopamine
objects (such as pictures of books, chairs, etc.). The dependent vari- transfer deficit (DTD) theory of ADHD based on the impairments
able was total times to name all objects, in each series. We cannot in learning about predictors of reinforcement that are effected by
predict the values of C required for each of these lists of unique dopamine. Their theory has made successful predictions of the
items, but we can estimate them from the behavior of the TDC, effects of methylphenidate on delay discounting, and on the effects
as their total time divided by their values of E estimated from the of intermittent reinforcement in ADHD. Research has shown that
stop-signal task. Then, given the values of E estimated for the clini- stimulus–response coupling is impaired in ADHD (Luman et al.,
cal groups in the stop-signal task, we can predict how long it should 2009), as they would predict, as is learning (Mayes and Calhoun,
take them to complete the naming. Fig. 18 shows the predictions 2007). The data on intermittent reinforcement and ADHD are mixed
of the behavior of these special populations in novel experimental (Kollins et al., 1997). A comprehensive review of the issues rele-
paradigms. vant to evaluating DTD and other theories of ADHD is provided
In the earlier instantiation of NeT (Russell et al., 2006), the by Luman et al. (2010). Because DTD is based on impaired dopa-
overlap in symptoms and therapy with two other psychiatric minergic function, its premises are different than those of NeT.
classifications, phenylketonuria and narcolepsy, were reviewed. Indeed, many other theories of ADHD, predicated on the efficacy
Sleep disturbances are also noted in ADHD (Imeraj et al., 2012; of methylphenidate, premise a hypodopaminergic (or sometimes
Sergeant, 2005), and restful sleep is important for effective synaptic a hyperdopaminergic) condition as casual in ADHD (Gonon, 2009).
energy use (Harris et al., 2012). Modafinil is the pharmacothe- Energetic insufficiency is the core aspect of NeT, one mitigated by
rapy of choice in narcolepsy, and also has therapeutic value in norepinephrine, not dopamine.
the treatment of ADHD. In addition to its effect on the dopamine NeT has perhaps the greatest resonance with Sergeant’s
transporter (Zolkowska et al., 2009), modafinil acts on the norepi- cognitive-energetics theory (CEM, e.g., Sergeant et al., 1999,
nephrine transporter as well as the ␣2 - and ␤-adrenergic receptors 2003; Sergeant, 2000, 2005). This theory incorporates features of
(Minzenberg and Carter, 2007). The latter are associated with the Barkley’s inhibition model and resource demand imbalance model
stimulation of astroglia, eliciting the increased availability of lactate (Turgay et al., 2012) and Sonuga-Barke’s delay aversion model
to empower the propagation of neural signals. (Marco et al., 2009). It comprises three levels, with three energetic
In light of the additional emphasis that our theory places on pools corresponding to the encoding, central processing and motor
astrocytic adrenoceptors, it must predict that whenever those are organization stages, organized by the executive control system at
compromised, symptoms similar to those that are diagnostic of top, where liabilities in working memory and inhibition are to be
ADHD should result. Multiple sclerosis (MS) is a clear test case. found. At the intermediate level, there are no differential liabilities
Astrocytic ␤2 -adrenoceptors are lost in MS (DeKeyser et al., 2004). in encoding or central processing: liabilities are concentrated in
And, in fact, patients with MS have slowed information processing, motor organization or timing (Noreika et al., 2013). The cognitive
more variable response latencies, and impaired working memory energetic theory has been criticized because its energetic pools are
(Hoffmann et al., 2007). They may suffer from focused, divided and unobserved hypothetical constructs (Rapport et al., 2001). The cur-
sustained attention deficits: the severity of MS is “robustly cor- rent behavioral neuroenergetics theory of ADHD, NeT, moves them
related” with their processing speed (De Sonneville et al., 2002). to center-stage: there are energetic pools associated with all cere-
Patients show a substantially smaller increase in activation of bral actions, and these consist of the ATP immediately available to
various brain regions than healthy controls as task complexity the neurons, backed up by astrocytic glycogen, and by the astro-
increases, and “have reduced functional reserve for cognition rel- cytes’ ability to convert that into lactate to provide the long-term
evant to memory” (Cader et al., 2006). Modafinil relieves their energy required by sustained neuronal firing. The central construct
associated fatigue (Rammohan et al., 2002), and methylphenidate of NeT, the energy available to the neurons in the form of lactate
improves their attentional performance (Harel et al., 2009). that enables sustained attention, is represented by the variable E.
That letter could as well stand for the central energetic resource in
5.2. Ultimate models of ADHD: NeT and other theories cognitive energetics, which is both directly available to computa-
tional processes, and which, through attention switching, helps
The present behavioral neuroenergetics theory of ADHD, NeT, empower other cognitive processes such as stimulus encoding
has many themes in common with other treatments of ADHD. It is and response organization (Kaplan and Berman, 2010; Sergeant
646 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

et al., 2006). That construct plays a key role in another cognitive- those brought up to tread therein.” Despite its attendant risks, Plan
energetics theory of motivated cognition (Kruglanski et al., 2012) B works well enough to keep characteristics of ADHD an important
under the rubrics potential driving force, and energy supply. part of the character of Homo sapiens.

5.3. Ultimate material bases: genetics


6. The role of pharmacological agents
ADHD is highly heritable (Asherson and Gurling, 2012), jus-
tifying the search for genetic substrates. In a review of five All pharmacotherapies for ADHD facilitate catecholamine
genome-wide association studies, Franke et al. (2009) noted that, function, either by increasing extracellular concentrations
whereas few of the ‘classic’ candidate genes, such as DAT4 a of norepinephrine and dopamine (the transporter block-
dopamine transporter gene, showed a reliable association with ers: methylphenidate, amphetamine, atomoxetine, modafinil,
ADHD, the cell adhesion molecule cadherin CDH13 was the most desipramine, and monoamine oxidase inhibitors), or by acting
common associate among the studies (also see Banaschewski et al., directly on noradrenergic receptors (guanfacine, an ␣2A-
2010). CDH13 is associated with many of the symptoms of ADHD adrenoceptor agonist). Some of these are stimulants, some
(Mick et al., 2011), and with impaired working memory (Arias- not; what they have in common is their ability to stimulate the
Vásquez et al., 2011). CDH13 plays many roles in the brain (Rivero release of lactate from astrocytes. The prefrontal cortex is the
et al., 2012), among them a negative growth regulator for astrocytes prime target of these drugs, mediating top–down regulation
(Huang et al., 2003). It is highly expressed in brain regions that have of motor, cognitive, and emotional function through reciprocal
reduced volume in ADHD (Takeuchi et al., 2000). It is our hypothesis connections with sensory and limbic association cortices, basal
that one of the reasons for a failure of the ANLS may be impair- ganglia and cerebellum (Arnsten, 2011). At large behaviorally
ment of astrocyte growth, and of their junctions with neurons activating doses, psychostimulants produce large and widespread
(Togashi et al., 2009). Mice carrying a dominant negative form of the increases in extracellular levels of brain catecholamines (Berridge
astrocyte-neuron adhesion molecule SynCAM1 displayed ADHD- and Devilbiss, 2011). Low, cognition-enhancing doses exert
like behavior that was attenuated by d,l-amphetamine (Sandau regionally restricted actions, elevating extracellular catecholamine
et al., 2012). This provides further support for impaired astrocyte- levels and enhancing neuronal signal processing by activating ␣2-
neuron communication playing a role in ADHD. adrenoceptors and dopamine D1 receptors within the prefrontal
As the psychometric literature shows, there may be many cortex (Berridge and Devilbiss, 2011; Gamo et al., 2010).
different biological pathways to the manifestation of ADHD-like On top of these common effects, such drugs have unique
symptoms. Other causes of hypoenergetics leading to slower neural effects on different brain systems, and thus on function. Nor-
processing might include mitochondrial dysfunction, hyperpolar- epinephrine and dopamine have complementary effects on the
ization of neurons requiring additional energy to maintain their strength of network connections in the prefrontal cortex, and func-
regular firing, failure of tight junctions to maintain cell polarity, tion to co-ordinate cognitive state with physiological demands
and so on. Depending on the parts of the brain involved, these (Arnsten, 2011). The effects of the catecholamines are both
may be viewed as different presentations of ADHD, comorbidities of receptor and region specific: Guanfacine, an ␣2-adrenoceptor ago-
ADHD, or a different dysfunction altogether with some ADHD-like nist, improved prefrontal cortex function (i.e., performance in
symptoms. In such cases, treatments with stimulants may be less working memory, attention stop-signal tasks, Bari et al., 2011)
effective in increasing the ATP ultimately available to the neuron. but did not improve performance in a spatial memory task,
which was largely dependent on hippocampal function (Decamp
5.4. The final causes of ADHD: what functions are served? et al., 2011). Norepinephrine actually strengthens connectivity
in the lateral prefrontal cortex via ␣2A-adrenoceptor activation.
The last of the four causes to be explicated concerns function. That is because such activation inhibits cAMP formation, closing
What is the immediate function of the symptoms of ADHD, and the hyperpolarization-activated cyclic nucleotide-gated (HCN) K+
what is the ultimate function, if any, of the syndrome? In the case channels, which reduces shunting and allows the network to inter-
of symptoms, attention wanders because, perforce, it must. Dis- connect (Brennan and Arnsten, 2008). Dopamine shapes network
cursive behavior exploits undepleted elements of the CNS, towards inputs through activation of D1 receptors (Arnsten and Pliszka,
which attention is naturally drawn, away from the fatigued ele- 2011). Activation of ␤-adrenoceptors in the medial prefrontal cor-
ments. While off task, neurons involved in the target task have the tex enhances memory consolidation (Barsegyan et al., 2010) and
chance to replete, making a change of attention not only a relief, rodent performance in the stop-signal task, and alters neural cir-
but also a strategy. cuits that control impulsivity (Bari et al., 2011).
For the syndrome, the question of function is more unsettled. Underlying this diversity of action in diverse brain areas are
Many articles on ADHD open with a list of its costs to society and functions common across areas. Upon neuronal glutamate release,
the individual. Unexplained is why these, severe as they are, have astrocytes provide lactate as the preferred neuronal energy source
not eliminated the genotype. The marginally higher rate of procre- (Wyss et al., 2011). Norepinephrine activates ␤-adrenoceptors,
ation for young adults with ADHD (Weiss et al., 1985) can be only stimulating glycogenolysis in astrocytes, which then produce lac-
part of the answer. Another must be that there are environments tate from glucose and release it to provide the fuel necessary
in which ADHD traits have some advantage (Killeen et al., 2012; for rapid persistent neuron firing (Pellerin and Magistretti, 2011;
Landau et al., 2012; Williams and Taylor, 2006). Stressful environ- Sorg and Magistretti, 1991). The ␣2A-adrenoceptors do not stim-
ments, in utero or in development, can precipitate the emergence ulate glycogenolysis and lactate production, but they may have
of ADHD (Cohen, 2010). Fugacious attention (due to lowered E) and a long-term effect by increasing glycogen synthesis in astrocytes
impulsivity (lowered C) can lead adolescents out of those environ- (Hutchinson et al., 2011; Subbarao and Hertz, 1990), thus increas-
ments. Quick responding and breadth of attention have survival ing the stores of energy in astrocytes. An ample supply of glycogen
advantage in the novel environments they then encounter (Weiss offers some protection against hypoglycemic neural injury (Brown
and Hechtman, 1993). ADHD is Nature’s Plan B, the escape hatch and Ransom, 2007). Glycogen is depleted during the day and
from James’s (1890, p. 121) “enormous flywheel of society [that] restored during sleep, which may provide some explanation for
keeps us all within the bounds of ordinance, and . . . prevents the why the afternoon is a period of risk for ADHD, the irregular sleep
hardest and most repulsive walks of life from being deserted by patterns associated with it, and the potential of melatonin, as an
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 647

aid to sleep onset, to be of general use in the treatment of ADHD whose parameters are summarized in Table 1, instantiates the
(Imeraj et al., 2012). core predictions of NeT: all of the derived measures will be
As Biederman and Spencer (1999) had observed, evidence now greater for individuals with ADHD than for TDC. NEMA is then
strongly supports a key role for norepinephrine in ADHD: the “over- applied to the psychoneurological data that are the most discrim-
whelming majority” of the drugs that are effective in treating ADHD inating of ADHD. These reanalyses all pointed to a reduction of
have important effects on the noradrenergic system (del Campo 15–25% in the speed (E) and completeness (C) of neural infor-
et al., 2011), none of the drugs used to treat ADHD act exclusively on mation processing in ADHD. NEMA is a general model, applicable
the dopaminergic system, and some of the drugs (e.g., atomoxetine) to response times and variabilities for all populations. It may be
work primarily to block norepinephrine uptake. Methylphenidate extended to other dependent variables (e.g., Fig. 10). NeT makes
is about equally effective in blocking norepinephrine transporters general predictions for specific populations (here, ADHD), for the
as it is in blocking dopamine transporters (Han and Gu, 2006). time-course of slowing of information processing, and the underly-
Atomoxetine, which is equally effective as methylphenidate in ing neural processes responsible for them, and calls upon NEMA to
treating ADHD symptoms (Hanwella et al., 2011; Hazell et al., translate those into specific predictions relevant to experimental
2011), increases both dopamine and norepinephrine in the pre- data.
frontal cortex of the rat, but unlike methylphenidate, does not In their overview of the role of computational models in cogni-
increase dopamine levels in the striatum or nucleus accumbens tive neuroscience, Cowell et al. (2012) noted five critical properties
(Bymaster et al., 2002). Dopaminergic theories of ADHD often of a computational model that require explication. Whereas ours is
assert a reinforcement deficit that the action of methylphenidate a deterministic rather than connectionist model, those properties
at these sites meliorates. The present theory posits no dopamine- merit review here as well.
related reinforcement deficit, and is therefore unembarrassed
by the efficacy of atomoxetine despite its inactivity in these 1. Levels of biological organization. These may range from the
regions. organism embedded in its environment, down through systems
to the CNS, to synapses, to molecules. The levels of material
causes invoked by NeT are principally at the neuroglial level
7. Summary and conclusions (Fig. 1), with diffusion upward toward systems and downward
toward genetic mechanisms.
7.1. The neuroenergetics mass-action model (NEMA) and the 2. The problem space. If levels constitute the domain of a model,
behavioral neuroenergetics theory (NeT) the problem space constitutes its range, previewed in Section
2 of this paper. For NeT this is processing speed, manifest in
ADHD is clinically heterogeneous, and research into this dis- response-time distributions (Figs. 4 and 9), their means and vari-
order is diverse—not only in the methodologies used, but also ances (Figs. 8 and 14), attentional fugacity (Fig. 6), impaired
the range of results reported. There is a need to integrate the working memory capacity (Figs. 10 and 11), coordination and
diverse findings. Over 300 meta-analyses have been conducted timing (Figs. 12 and 13).
in an attempt to increase analytic power in specifying differ- 3. Biological plausibility. NeT is grounded in the neuroscientific
ences between ADHD and TDCs. This is an empirical, inductive data that indicate neuroenergetic insufficiencies as a basis of the
form of integration. Theoretical modeling—for example the dual phenomena in its problem space. If those are falsified, then NeT
pathway theory and its revision (Sonuga-Barke et al., 2010)—is must be rejected and NEMA orphaned.
a retroductive form of integration. Progress in understanding 4. Parsimony. In order to constrain the degrees of freedom in the
ADHD has become crucially dependent upon the quantitative model, thus stabilizing the parameters, parsimony has been a
predictions that theory affords; predictions that, when cor- ubiquitous concern. No attempt was made to predict errors
rect help to substantiate the theory, and when off the mark, of omission and commission, because the additional requisite
help to improve it. This paper integrates substantial domains parameter (a second criterion) would absorb all the degrees
within the field of ADHD research theoretically, and attempts of freedom in most such data without enabling new predic-
to further demonstrate the quantitative predictability of find- tions. Table 1 arranges the parameters, with only one application
ings between diverse domains based upon the mathematical (Fig. 6, tracing the evolution of behavior over time for two groups
model, NEMA, acting under the direction of our general theory, on three variables) requiring five parameters, and most four or
NeT. fewer parameters. Given that our goal has been accurate depic-
NeT is a theory of response slowing and attentional fugacity tion of the data, which we have generally achieved, we would
characteristic of ADHD. Its central thesis is that neurons may not be be surprised to find more parsimonious models able to give a
adequately resupplied with the energetic resources—lactate—that comparable account of existing data.
they require for prolonged, precisely timed firing. This could occur 5. Predictions. Sometimes trends suffice: “The key property of the
because of failures in any one of the components that are involved simulations is the qualitative . . . trends in the data that emerge”
in the astrocyte-neuron lactate shuttle (Fig. 1). To translate this (Cowell et al., 2012). We offer a set of such qualitative predictions
premise into predictions, we sketched a minimal dynamic model for future research:
of supply and demand (Fig. 3 and Eq. (1)), and generalized it to the • Rate of approach to asymptotic performance depends on the
case of ensembles of neurons (Eq. (4)). Rate of firing is closely cou- sum of rates of supply and demand (Eq. (3)); where supply is
pled to the energy available to reestablish gradients and reposition compromised, as in ANLS failure, that approach will be slower;
glutamate vesicles, so that energetic course was taken as predictive but where cost is increased, as in poor myelination or more
of the speed of information processing. Assuming in addition that intense processing demands, the approach will be faster.
a criterial number of computations (C) are necessary for a response • Interventions, such as the use of stimulants, which increase
leads to the inverse Gaussian distribution of response times called energetic supplies will increase the rate of approach to asymp-
the Wald distribution (Eq. (5)). totic performance.
It is difficult to persist in an activity, be it perceptual, cog- • Whenever astrocytic adrenoceptors are compromised, symp-
nitive, or motor, absent the requisite energy. Behavior wanders. toms similar to those that are diagnostic of ADHD should result.
Fig. 5 provides the apposite model of lapse and recollection of • All medications that are effective in ameliorating pha-
attention, and consequent response execution: NEMA. This model, sic symptoms of ADHD will do so by stimulating lactate
648 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

release from astrocytes in relevant brain systems. Such drugs reduce premature responding, and improve visuospatial attention
(both stimulants and non-stimulants) may act by increasing and working memory (Pattij et al., 2011).
extracellular levels of the catecholamines—particularly, nor- There is, however, a downside to tonic activation. Brains typi-
epinephrine, which stimulates energy repletion, and thus will cally buffer tonic changes in levels of neurotransmitters. The most
improve performance. widely used pharmacotherapies, therefore, are the psychostimu-
• Conversely, any treatment that stimulates astrocytes to take up lants methylphenidate and amphetamine, which block dopamine
glucose, convert it to lactate and release lactate into the extra- and norepinephrine transporters. Their action is more circum-
cellular space in response to signals of elevated extracellular scribed, both temporally and spatially, to neural activity: just the
glutamate or K+ levels will ameliorate ADHD symptoms. thing that needs support. The presence of glutamate stimulates nor-
• Dopamine insufficiency plays no role in ADHD. Norepi- epinephrine release from local axon varicosities in the prefrontal
nephrine is an effective therapeutic agent because of its ability cortex and hippocampus (Howells and Russell, 2008; Russell, 2001;
to stimulate astrocytes to release lactate. Russell and Wiggins, 2000). Noradrenergic regulation of lactate
• Because of the energy demands involved in neural synchro- release by astrocytes can thus be selectively enhanced in areas
nization and maintenance of LTP, both essential for learning, of increased neural activity (Fusar-Poli et al., 2012; Magistretti,
individuals with ADHD will be disadvantaged in complex tasks 2011). In contrast to tonic activation of adrenoceptors on both
involving learning, particularly at slow rates of stimulation, glia and neurons, drugs that block the norepinephrine trans-
where attention is also more likely to lapse. porter act locally to enhance the glutamate-stimulated increase
• Because of astrocytic involvement in memory formation and in extracellular norepinephrine, thereby increasing astrocyte pro-
consolidation, individuals with ADHD will be disadvantaged in duction of lactate and sustaining neural firing where and when
complex tasks involving memory. it is most needed. This reduces (although it does not eliminate)
• Because of their difficulty in maintaining sustained neural the down-regulation (Fleckenstein et al., 2009; Simchon et al.,
firing in functional systems, in situations requiring focused 2010), which leads to tolerance to pharmacotherapy (Swanson
attention the performance of individuals with ADHD will be et al., 1999; Vles et al., 2003; Yanofski, 2011). If our hypoth-
inferior to that of controls. esis is scrutinized, it will lead to a diversification of research
• Wandering of attention and discursive action will be especially that has, here-to-date, been dominated by the dopaminergic
pronounced for individuals with ADHD in monotone environ- hypothesis (Rastmanesh, 2010). If our hypothesis is sustained,
ments. it may lead to discovery of other sites of neuro-intervention at
• Group performances will diverge with time-through-trial points where the energy transfer may be compromised (such
(Fig. A1). as the orexin system: Sakurai et al., 2010; Tsujino and Sakurai,
• Group performances will diverge with time-on-task (Fig. C1). 2009).
• Spectra of latencies that have not been detrended will show Because one of the points of focus in this article has been on
greatest discrepancies at low frequencies corresponding to the the pharmacotherapy of ADHD, it is essential to note in closing the
first few minutes on task. limitations of this approach. ADHD often persists into adulthood
• Tonic catecholaminergic stimulation will be less effective than (Spencer et al., 2007), but the beneficial effects of pharmacothe-
phasic stimulation due to down-regulation. rapy seldom persist after it is discontinued, and interventions have
• Comorbid dysfunctions may co-occur because affected indi- little to no lasting benefit on behavioral trajectories (Molina et al.,
viduals suffer a similar neurocomputational slowing in 2009). Non-pharmacological interventions have little impact on
particular portions of the brain, apart from other factors that core symptoms (Sonuga-Barke et al., 2013). There are no phar-
distinguish the disorders (Fig. 18). macotherapies for ADHD: there are only pharmaco-prostheses.
• Quality of performance of ADHD will be improved in self-paced This is consistent with NeT, which posits no long-term structural
tasks, where, to a limited extent, their slower processing does changes due to the transient facilitation of the astrocyte-neuron
not incur an accuracy penalty. Accuracy scores will be less lactate shuttle, or to other treatment regimens. As prostheses,
discriminating of groups in such tasks. current medications give breathing space in which behavioral cop-
ing strategies may evolve or be trained (Chronis et al., 2006;
Sometimes quantitative predictions are also possible. Fabiano et al., 2009; Pelham and Fabiano, 2008). Halperin and
Healey (2011) have urged a new approach to ADHD, a program
• We have assayed such in the curves of Figs. 7, 8, 15 and 18. of directed play and exercise, which would not only have direct
• We predict that all the derived indices of performance in benefits on brain growth and social skills, but also stimulate sym-
Table 1 will be of greater magnitude for ADHD than for TDC. pathetic activity and norepinephrine release in the prefrontal

• We predict that E = M/SD, the ratio of the square root of the cortex (Barry et al., 2005; Rapport et al., 2009a), thereby nour-
mean RT to its standard deviation, will be of smaller magni- ishing depleted neurons. Indeed, the authors’ proposed regimens
tude for ADHD than for TDC. C = M3/2 /SD will be smallest for would benefit all children. They may cause a useful reframing
individuals presenting as Hyperactive–Impulsive, and more of our quest, from a treatment of ADHD as a psychiatric dis-
discriminating of groups on experimenter paced than on self- order, to prevention and treatment of symptoms that can put
paced tasks. every child at risk of failing to achieve their full potential in
life.
7.2. Implications for therapy
Acknowledgments
Noradrenergic neurons innervate functionally distinct corti-
cal areas, exerting widespread effects on energy metabolism This article would not have been written without the early
(Pellerin and Magistretti, 2011). This implies that drugs that support of the Norwegian Centre for Advanced Study, and
tonically activate ␣2- or ␤-adrenoceptors might have some the leadership there of Terje Sagvolden. We thank Rosemary
therapeutic potential in treating ADHD symptoms. Such drugs have Tannock for comments on the manuscript, and Paige Scalf for
been found to be beneficial in treating symptoms of inattention reminding us of the attention and organizational deficits found
and impulsivity in a rodent model (Arnsten and Pliszka, 2011): in MS patients. Two anonymous reviewers greatly sharpened the
Clenbuterol, a potent ␤2-adrenoceptor agonist, has been shown to arguments and text.
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 649

mean time to the first lapse is the reciprocal of this, 4.5 s. Treat-
ment of ˛ is exactly the same. The mean time to the recapture of
attention is  = 1/˛ .
Blatent stimuli that are apprehended even if the subject is not
focussing on the target call the subject back to attention (Fig. 5).
The shift back is not instantaneous, but requires an average of s.
The simple form of the model in Fig. 5 may be retained by having
target onset precipitate the return to attention, driving ˛ from 0 (or
any other assigned value) to a larger value ˛ close to 1.0. Then the
transition approximates an exponential process, with the probabil-
1
ity of reattending during the first second ˛ = 0 e−˛ t dt = 1 − e−˛ .
Inversely, the mean time to recapture is  = 1/˛ = −1/ln(1 − ˛) s.
Values for  range around 1/3 s, corresponding to ˛ = 3/s, and ˛
Fig. A1. The drift of attention off target (out of the A state), given by Eqs. (A1 and around 0.95.
A2).

Appendix B. Data analysis


Appendix A. Attention drift and recapture
Consider first the data in Fig. 4. The model assumes that the
The attentional module in Fig. 5 is the state diagram of a simple multiplicity of neural firings has the character of a random additive
Markov chain. The probability of residence in either of the states as walk toward C, whose time of first passage is given by the Wald
a function of time may be predicted from its transition matrix, Eq. distribution (Eq. (5)). The parameters of that equation, the best val-
(A1): ues of E and C, are found by an iterative search that maximizes the
  probability of the data given the model and its parameters. This
1−  

P=  (A1)
is similar to maximizing the sum of squares deviation between
˛ 1−˛ model and data, which was the cost function for analysis of all
non-distributional data, such as means and standard deviations.
The first row and column correspond to the A state, and the In the maximum likelihood case, binomial variability around the
second row and column to the ∼A state. The rows indicate the cur- predicted points gave the probability of the observed data given its
rent locus of attention, and the columns the probability that, after predicted locus, which was maximized using an iterative algorithm.
1 step, attention will be in the state given by the columns. Thus, if Consider next Fig. 6. The change in available energy is given by
starting in A the probability of staying there after 1 step is 1 − , Eq. (4), a power function with exponent  (rho) and right asymp-
and the probability of drifting to inattention, ∼A, is . If in the ∼A tote of E∞ . Evaluation of that function, with the parameters given
state, corresponding to the second row of the matrix, the probabil- in Table 2, gives the inferred change in energy (E) available to the
ity of returning to attention is ˛, and the probability of remaining neurocomputations at any point in time. For simple Wald distri-
inattentive (the second column) is 1 − ˛. butions, the mean and variance of those passages are C/E and C/E3 ,
To predict the probability of being in any of the states after s respectively. C remains invariant within group and task (after initial
steps, raise the matrix to the power s: learning of the task), but E changes as a function of energy deple-
p(As ) = P s (A2) tion. It is necessary to choose values of these three parameters: C,
and  and E∞ . From these latter two Et is derived for each of the
Fig. A1 demonstrates the loss of attention with time, in the case blocks of time into the session. That could immediately be done by
that  is 0.1, and where ˛ takes values indicated by the parameter. an iterative search that maximizes the probability of the data given
The process approaches an asymptote, whence the probability the model and its parameters. However, now we have a richer data
of finding the subject in the A state is given by: field, one that permits us to trace the course of attention as a func-
˛ tion of the inter-stimulus interval, permitting us to infer the effect
P∞ (A) = (A3) of changes in energy on the ability to remain focused on the task.
˛+
Here, and everywhere else in this paper, it is assumed for simplicity
For brief and inconspicuous stimuli, this is the long-range prob-
that the probability of being attentive from 1 s to the next (1 − ) is
ability of a successful detection. If the probability of drifting back
proportional to the available energy. Using the continuous (expo-
to attention is 0, as shown in the bottom curve in Fig. A1, then
nential) form: (1 −  ) = kEt . In the study analyzed here, k took the
the probability of maintaining attention is essentially a negative
value 1.46. The cost of fugacious attention, , the time required
exponential function of time:
to recall attention, must also be estimated. Instead of doing these

P(A) = (1 − ) ≈ e− t
t estimates piecemeal, the likelihood of the ensemble of data is max-
(A4)
imized to estimate the value of k (from which  is derived), and .
In making the transition from a probability () and its geometric Now the predicted densities are not Wald but a mixture of Wald and
evolution, to a rate ( ) and its exponential evolution, the numerical
1
a convolution of Wald and exponential, the ex-Wald, as depicted
value of the parameter must be adjusted:  = 0 e− t dt = 1 − e− . in Fig. 5. The mean and variance of the Wald distribution are C/E
For small values of , as called for if the probabilities were updated and C/E3 ; the recovery of attention adds to these (1 − p(A)) and
on a millisecond time base, they are essentially equal:  ≈  . The (1 − p(A)) 2 respectively. From Eq. (A4), the probability of atten-

time base has been set to 1 s, however. This is to keep these tempo- tion is p(A) = e− t . Heathcote (2004) provides S-Plus functions for
ral units in seconds, and to keep the parameter values in ranges that fitting the ex-Wald to data. Because  is not weighted by 1 − p(A)
are kinder to our intuitions. To convert from the rate constants (the in that package, expect estimates of it to grow larger as attention
primed symbols) to probabilities, compute the cumulative proba- wanes.
bility of a transition occurring at the rate  over the course of 1 s, Sometimes great pains are taken in the execution of an exper-
as in the above equation. Inversely,  = −ln(1 − ). A probability iment, but only cursory data, or, worse, inferential statistics, are
of  = 0.25 corresponds to a rate of approximately  = 0.22/s. The reported. (To compound this paucity of information, often positive
650 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

effects that do not achieve significance are taken as evidence


against an effect, misleading the readers. Lack of strong (e.g., statis-
tically significant) evidence against a null hypothesis should never
be interpreted as evidence for that hypothesis; cf. Gallistel, 2009).
It is still possible to derive information of use to the theory. When
a distribution of data is fairly well described by a model such as
the ex-Gaussian, another model may be used to fit the same data,
one step removed, by the method of matching moments. In its
present avatar this is less sophisticated than it sounds. The mean
of the ex-Gaussian distribution is  + . Its variance is  2 +  2 . We
may set these equal to the mean and variance given above for the
Wald distribution, and solve, yielding the relation of Eq. (9). Given
only the mean and variance of the data, we may still proceed, by
setting  = 0. This uses less information from the data, and thus
gives an inferior triangulation of the NEMA parameters, but is often
adequate to display trends. In both cases, the attentional module
is omitted: the probability of an attentional lapse is de facto set
to  = 0. This leaves the basic Wald to make up for some of the
instances of lapsed attention by deflating its estimates of E (to make
up for the missing exponential term in the ex-Wald). As this will
happen for ADHD and TDC alike, comparative measures retain some
validity.
Not predicted here are the proportions of misses and false posi-
tives in continuous detection tasks. Error distributions can shift
between these as a function of strategy, often reported as the
Fig. C1. Top: The decrease in energy available over the course of a session according
decision threshold. The standard approach to modeling such data to Eq. (4), with  = 0.25 (ADHD) and 0.3 (TDC), and asymptotic energy of 0.02 and
is the two-boundary generalization of the drift model presented 0.20, respectively. Bottom: predicted means and standard deviation of response
here (e.g., Ratcliff and McKoon, 2008; Ratcliff and Rouder, 1998). times, given by Eqs. (6 and 7), with  = 0, and C = 130 and 160 for ADHD and TDC.
Whereas such approaches have been tremendously successful, they
are more complicated (Brown and Heathcote, 2008; Ratcliff and first few minutes are often spent on task orientation, practice trials,
Tuerlinckx, 2002; Wagenmakers et al., 2007), and add degrees of or even other neuropsychological tests that deplete similar func-
freedom to the model that cannot be supported by most of the data tional units. Later in the session, although differences are maximal,
available in the literature on ADHD (but see, e.g., Huang-Pollock the rate of change in those differences becomes minimal.
et al., 2012; Karalunas and Huang-Pollock, 2013; Mulder et al.,
2010). In this paper simplicity was favored over those more sophis- References
ticated models, at the cost of a more general account of probabilities
and latencies of correct and error response times. Alberts, E., van der Meere, J., 1992. Observations of hyperactive behaviour during
An alternate analytic approach noted by a reviewer would be to vigilance. Journal of Child Psychology and Psychiatry, and Allied Disciplines 33,
1355–1364.
apply the model to data from a large sample, and show that NEMA Albrecht, B., Brandeis, D., Uebel, H., Heinrich, H., Mueller, U.C., Hasselhorn, M., Stein-
found no difference in E and C between individuals that were and hausen, H.C., Rothenberger, A., Banaschewski, T., 2008. Action monitoring in
were not ADHD, on tests that did not discriminate them; but suc- boys with attention-deficit/hyperactivity disorder, their nonaffected siblings,
and normal control subjects: evidence for an endophenotype. Biological Psychi-
cessfully predicted differences for tests that did. We are confident atry 64, 615–625.
that this would work, but only if other diagnostic categories were Alderson, R.M., Rapport, M.D., Kasper, L.J., Sarver, D.E., Kofler, M.J., 2012. Hyperactiv-
excluded, as they may share many of the symptoms of ADHD. ity in boys with attention deficit/hyperactivity disorder (ADHD): the association
between deficient behavioral inhibition, attentional processes, and objectively
measured activity. Child Neuropsychology: A Journal on Normal and Abnormal
Development in Childhood and Adolescence 18, 487–505.
Appendix C. Sessional analysis Alderson, R.M., Rapport, M.D., Kofler, M.J., 2007. Attention-deficit/hyperactivity
disorder and behavioral inhibition: a meta-analytic review of the stop-signal
paradigm. Journal of Abnormal Child Psychology 35, 745–758.
NeT predicts slower and more variable response times for ADHD Alderson, R.M., Rapport, M.D., Sarver, D.E., Kofler, M.J., 2008. ADHD and behavioral
than TDC, and that these differences will increase with time on task. inhibition: a re-examination of the stop-signal task. Journal of Abnormal Child
This is because functional areas of the brain are depleting their ATP, Psychology 36, 989–998.
Alloway, T.P., Gathercole, S.E., Pickering, S.J., 2006. Verbal and visuospatial short-
and lactate resupply is insufficient to generate all that is required. term and working memory in children: are they separable? Child Development
This happens to TDC also, but to a lesser extent. The effects should be 77, 1698–1716.
reduced by interlacing a variety of activities, and by self-paced tasks American-Psychiatric-Association, 1994. Diagnostic and Statistical Manual of Men-
tal Disorders. American Psychiatric Publishing, Inc, Washington, DC.
where lower computational speeds do not force accuracy trade- Andersen, P.N., Egeland, J., Øie, M., 2012. Learning and memory impairments in
offs. Yet, often only small to moderate effect sizes are found for children and adolescents with Attention-Deficit/Hyperactivity Disorder. Journal
increasing divergence as a function of time on task (Huang-Pollock of Learning Disabilities, http://dx.doi.org/10.1177/0022219412437040, Online
First.
et al., 2012). Why should this be the case? Fig. C1 depicts one of the Angold, A., Costello, E., Erkanli, A., Worthman, C., 1999. Pubertal changes in hormone
reasons given by NEMA. For representative parameters, available levels and depression in girls. Psychological Medicine 29, 1043–1053.
energy decreases as a convex function of time. Although the differ- Arias-Vásquez, A., Altink, M.E., Rommelse, N.N.J., Slaats-Willemse, D.I.E., Buschgens,
C.J.M., Fliers, E.A., Faraone, S.V., Sergeant, J.A., Oosterlaan, J., Franke, B.,
ence between energy levels is greatest toward the end of a session,
2011. CDH13 is associated with working memory performance in attention
the rate of divergence between them will be greatest toward the deficit/hyperactivity disorder. Genes, Brain and Behavior 10, 844–851.
beginning of the session. If, as typically the case, TDC have a slightly Ariyannur, P.S., Moffett, J.R., Manickam, P., Pattabiraman, N., Arun, P., Nitta, A.,
higher criterion for responding, there will be little or no difference Nabeshima, T., Madhavarao, C.N., Namboodiri, A., 2010. Methamphetamine-
induced neuronal protein NAT8L is the NAA biosynthetic enzyme: implications
in response rates at the start of the session (or possibly a crossover, for specialized acetyl coenzyme A metabolism in the CNS. Brain Research 1335,
with ADHD faster, depending on the parameters). However, those 1–13.
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 651

Arnsten, A.F.T., 2011. Catecholamine influences on dorsolateral prefrontal cortical Berridge, C., Devilbiss, D., 2011. Psychostimulants as cognitive enhancers: the pre-
networks. Biological Psychiatry 69, e89–e99. frontal cortex, catecholamines, and Attention-Deficit/Hyperactivity Disorder.
Arnsten, A.F.T., Pliszka, S.R., 2011. Catecholamine influences on prefrontal cortical Biological Psychiatry 69, e101–e111.
function: relevance to treatment of attention deficit/hyperactivity disorder and Biederman, J., Monuteaux, M.C., Doyle, A.E., Seidman, L.J., Wilens, T.E., Fer-
related disorders. Pharmacology, Biochemistry, and Behavior 99, 211–216. rero, F., Morgan, C.L., Faraone, S.V., 2004. Impact of executive function
Arnsten, A.F.T., Rubia, K., 2012. Neurobiological circuits regulating attention, cog- deficits and attention-deficit/hyperactivity disorder (ADHD) on academic
nitive control, motivation, and emotion: disruptions in neurodevelopmental outcomes in children. Journal of Consulting and Clinical Psychology 72,
psychiatric disorders. Journal of the American Academy of Child and Adolescent 757–766.
Psychiatry 51, 356–367. Biederman, J., Spencer, T., 1999. Attention-deficit/hyperactivity disorder (ADHD) as
Asherson, P., Gurling, H., 2012. Quantitative and molecular genetics of ADHD. In: a noradrenergic disorder. Biological Psychiatry 46, 1234–1242.
Behavioral Neuroscience of Attention Deficit Hyperactivity Disorder and its Börger, N., van der Meere, J., 2000. Motor control and state regulation in children
Treatment, pp. 239–272. with ADHD: a cardiac response study. Biological Psychology 51, 247–267.
Asplund, C.L., Todd, J.J., Snyder, A.P., Marois, R., 2010. A central role for the lat- Börger, N., van der Meere, J., Ronner, A., Alberts, E., Geuze, R., Bogte, H., 1999. Heart
eral prefrontal cortex in goal-directed and stimulus-driven attention. Nature rate variability and sustained attention in ADHD children. Journal of Abnormal
Neuroscience 13, 507–512. Child Psychology 27, 25–33.
Aston-Jones, G., Cohen, J.D., 2005. An integrative theory of locus coeruleus- Brennan, A.R., Arnsten, A.F.T., 2008. Neuronal mechanisms underlying attention
norepinephrine function: adaptive gain and optimal performance. Annual deficit hyperactivity disorder. Annals of the New York Academy of Sciences 1129,
Review of Neuroscience 28, 403–450. 236–245.
Attwell, D., Gibb, A., 2005. Neuroenergetics and the kinetic design of excitatory Brown, A.M., Ransom, B.R., 2007. Astrocyte glycogen and brain energy metabolism.
synapses. Nature Reviews. Neuroscience 6, 841–849. Glia 55, 1263–1271.
Attwell, D., Laughlin, S.B., 2001. An energy budget for signaling in the grey matter Brown, S.D., Heathcote, A., 2008. The simplest complete model of choice
of the brain. Journal of Cerebral Blood Flow and Metabolism 21, 1133–1145. response time: linear ballistic accumulation. Cognitive Psychology 57,
Aubert, A., Costalat, R., 2007. Compartmentalization of brain energy metabolism 153–178.
between glia and neurons: insights from mathematical modeling. Glia 55, Buhusi, C.V., 2003. Dopaminergic mechanisms of interval timing and attention. In:
1272–1279. Meck, W.H. (Ed.), Functional and Neural Mechanisms of Interval Timing. CRC
Aubert, A., Costalat, R., Magistretti, P.J., Pellerin, L., 2005. Brain lactate kinetics: mod- Press, Boca Ratan, FL, pp. 317–338.
eling evidence for neuronal lactate uptake upon activation. Proceedings of the Buhusi, C.V., Meck, W.H., 2009. Relative time sharing: new findings and an extension
National Academy of Sciences 102, 16448–16453. of the resource allocation model of temporal processing. Philosophical Transac-
Azevedo, F.A.C., Carvalho, L.R.B., Grinberg, L.T., Farfel, J.M., Ferretti, R.E.L., Leite, R.E.P., tions of the Royal Society B: Biological Sciences 364, 1875–1885.
2009. Equal numbers of neuronal and nonneuronal cells make the human brain Bush, G., Frazier, J.A., Rauch, S.L., Seidman, L.J., Whalen, P.J., Jenike, M.A., Rosen,
an isometrically scaled up primate brain. The Journal of Comparative Neurology B.R., Biederman, J., 1999. Anterior cingulate cortex dysfunction in attention-
513, 532–541. deficit/hyperactivity disorder revealed by fMRI and the Counting Stroop.
Banaschewski, T., Becker, K., Scherag, S., Franke, B., Coghill, D., 2010. Molecu- Biological Psychiatry 45, 1542–1552.
lar genetics of attention-deficit/hyperactivity disorder: an overview. European Buzy, W., Medoff, D., Schweitzer, J., 2009. Intra-individual variability among chil-
Child & Adolescent Psychiatry 19, 237–257. dren with ADHD on a working memory task: an Ex-Gaussian approach. Child
Banaschewski, T., Hollis, C., Oosterlaan, J., Roeyers, H., Rubia, K., Willcutt, E., Tay- Neuropsychology: A Journal on Normal and Abnormal Development in Child-
lor, E., 2005. Towards an understanding of unique and shared pathways in the hood and Adolescence (Neuropsychology, Development and Cognition: Section
psychopathophysiology of ADHD. Developmental Science 8, 132–140. C) 15, 441–459.
Banaschewski, T., Jennen-Steinmetz, C., Brandeis, D., Buitelaar, J.K., Kuntsi, J., Bymaster, F.P., Katner, J.S., Nelson, D.L., Hemrick-Luecke, S.K., Threlkeld, P.G.,
Poustka, L., Sergeant, J.A., Sonuga-Barke, E.J., Frazier-Wood, A.C., Albrecht, B., Heiligenstein, J.H., Morin, S.M., Gehlert, D.R., Perry, K.W., 2002. Atomoxetine
Chen, W., Uebel, H., Schlotz, W., van der Meere, J.J., Gill, M., Manor, I., Miranda, A., increases extracellular levels of norepinephrine and dopamine in prefrontal cor-
Mulas, F., Oades, R.D., Roeyers, H., Rothenberger, A., Steinhausen, H.-C., Faraone, tex of rat: a potential mechanism for efficacy in attention deficit/hyperactivity
S.V., Asherson, P., 2012. Neuropsychological correlates of emotional lability in disorder. Neuropsychopharmacology 27, 699–711.
children with ADHD. Journal of Child Psychology and Psychiatry 53, 1139–1148. Cader, S., Cifelli, A., Abu-Omar, Y., Palace, J., Matthews, P.M., 2006. Reduced brain
Bari, A., Mar, A.C., Theobald, D.E., Elands, S.A., Oganya, K.C.N.A., Eagle, D.M., Robbins, functional reserve and altered functional connectivity in patients with multiple
T.W., 2011. Prefrontal and monoaminergic contributions to stop-signal task per- sclerosis. Brain 129, 527–537.
formance in rats. The Journal of Neuroscience: the Official Journal of the Society Caesar, K., Hashemi, P., Douhou, A., Bonvento, G., Boutelle, M.G., Walls, A.B., Lau-
for Neuroscience 31, 9254–9263. ritzen, M., 2008. Glutamate receptor-dependent increments in lactate, glucose
Barkley, R.A., 1997a. Attention-deficit/hyperactivity disorder, self-regulation, and and oxygen metabolism evoked in rat cerebellum in vivo. The Journal of Physi-
time: toward a more comprehensive theory. Journal of Developmental and ology 586, 1337–1349.
Behavioral Pediatrics: JDBP 18, 271–279. Calvetti, D., Somersalo, E., 2012. Ménage å trois: the role of neurotransmitters in
Barkley, R.A., 1997b. Behavioral inhibition, sustained attention, and executive func- the energy metabolism of astrocytes, glutamatergic, and GABAergic neurons.
tions: constructing a unifying theory of ADHD. Psychological Bulletin 121, Journal of Cerebral Blood Flow and Metabolism 32, 1472–1483.
65–94. Carter, B.C., Bean, B.P., 2009. Sodium entry during action potentials of mammalian
Barkley, R.A., Edwards, G., Laneri, M., Fletcher, K., Metevia, L., 2001. Executive func- neurons: incomplete inactivation and reduced metabolic efficiency in fast-
tioning, temporal discounting, and sense of time in adolescents with attention spiking neurons. Neuron 64, 898–909.
deficit hyperactivity disorder (ADHD) and oppositional defiant disorder (ODD). Cassenti, D.N., 2011. The intrinsic link between motor behavior and temporal cog-
Journal of Abnormal Child Psychology 29, 541–556. nition. New Ideas in Psychology 29, 72–79.
Barkley, R.A., Murphy, K.R., Fischer, M., 2007. ADHD in Adults: What the Science Castellanos, F.X., Giedd, J.N., Berquin, P.C., Walter, J.M., Sharp, W., Tran, T., Vaituzis,
Says. The Guilford Press. A.C., Blumenthal, J.D., Nelson, J., Bastain, T.M., 2001. Quantitative brain mag-
Barrett, L.F., Tugade, M.M., Engle, R.W., 2004. Individual differences in working mem- netic resonance imaging in girls with attention-deficit/hyperactivity disorder.
ory capacity and dual-process theories of the mind. Psychological Bulletin 130, Archives of General Psychiatry 58, 289.
553–573. Castellanos, F.X., Sonuga-Barke, E.J., Scheres, A., Di Martino, A., Hyde, C., Walters,
Barry, R.J., Clarke, A.R., McCarthy, R., Selikowitz, M., Rushby, J.A., 2005. Arousal and J.R., 2005. Varieties of attention-deficit/hyperactivity disorder-related intra-
activation in a continuous performance task. Journal of Psychophysiology 19, individual variability. Biological Psychiatry 57, 1416–1423.
91–99. Cepeda, N.J., Cepeda, M.L., Kramer, A.F., 2000. Task switching and attention
Barsegyan, A., Mackenzie, S.M., Kurose, B.D., McGaugh, J.L., Roozendaal, B., 2010. Glu- deficit hyperactivity disorder. Journal of Abnormal Child Psychology 28,
cocorticoids in the prefrontal cortex enhance memory consolidation and impair 213–226.
working memory by a common neural mechanism. Proceedings of the National Chatton, J.-Y., Pellerin, L., Magistretti, P.J., 2003. GABA uptake into astrocytes is not
Academy of Sciences 107, 16655–16660. associated with significant metabolic cost: implications for brain imaging of
Bartholomew, D.J., Deary, I.J., Lawn, M., 2009. A new lease of life for Thomson’s bonds inhibitory transmission. Proceedings of the National Academy of Sciences 100,
model of intelligence. Psychological Review 116, 567–579. 12456–12461.
Baslow, M.H., 2002. Evidence supporting a role for N-acetyl-aspartate as a molecular Cherkasova, M.V., Hechtman, L., 2009. Neuroimaging in attention-deficit hyperactiv-
water pump in myelinated neurons in the central nervous system: an analytical ity disorder: beyond the frontostriatal circuitry. Canadian Journal of Psychtiatry
review. Neurochemistry International 40, 295–300. 54, 651–664.
Bélanger, M., Allaman, I., Magistretti, P.J., 2011. Brain energy metabolism: focus on Chhikara, R.S., Folks, L., 1989. The Inverse Gaussian Distribution: Theory, Methodol-
astrocyte-neuron metabolic cooperation. Cell Metabolism 14, 724–738. ogy, and Applications. CRC, Boca Raton, FL.
Benarroch, E.E., 2010. Glycogen metabolism. Neurology 74, 919. Choi, H.B., Gordon, G.R.J., Zhou, N., Tai, C., Rungta, R.L., Martinez, J., Milner, T.A., Ryu,
Berlin, L., Bohlin, G., Nyberg, L., Janols, L.O., 2003. Sustained performance and reg- J.K., McLarnon, J.G., Tresguerres, M., 2012. Metabolic communication between
ulation of effort in clinical and non-clinical hyperactive children. Child: Care, astrocytes and neurons via bicarbonate-responsive soluble adenylyl cyclase.
Health and Development 29, 257–267. Neuron 75, 1094–1104.
Bernoulli, D., 1738/1954. Specimen theoriae novae de mensura sortis. Commentarii Chronis, A.M., Jones, H.A., Raggi, V.L., 2006. Evidence-based psychosocial treatments
academiae scientarum imperialis Petropolitanae 5, pp. 175–192. for children and adolescents with attention-deficit/hyperactivity disorder. Clin-
Berquin, P., Castellanos, F., Giedd, J., Hamburger, S., Rapoport, J., 1998. Cerebellum in ical Psychology Review 26, 486–502.
attention Deficit/Hyperactivity Disorder: an MRI morphometric study. European Clark, L., Blackwell, A.D., Aron, A.R., Turner, D.C., Dowson, J., Robbins, T.W., Sahakian,
Psychiatry 13, 160s–161s. B.J., 2007. Association between response inhibition and working memory in
652 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

adult ADHD: a link to right frontal cortex pathology? Biological Psychiatry 61, Epstein, J.N., Langberg, J.M., Rosen, P.J., Graham, A., Narad, M.E., Antonini, T.N.,
1395–1401. Brinkman, W.B., Froehlich, T., Simon, J.O., Altaye, M., 2011. Evidence for higher
Cloutier, M., Bolger, F.B., Lowry, J.P., Wellstead, P., 2009. An integrative dynamic reaction time variability for children with adhd on a range of cognitive
model of brain energy metabolism using in vivo neurochemical measurements. tasks including reward and event rate manipulations. Neuropsychology 25,
Journal of Computational Neuroscience 27, 391–414. 427–441.
Coghill, D., Sonuga-Barke, E.J.S., 2012. Annual research in review: categories versus Estle, S.J., Green, L., Myerson, J., Holt, D.D., 2007. Discounting of monetary and
dimensions in the classification and conceptualisation of child and adolescent directly consumable rewards. Psychological Science: A Journal of the American
mental disorders: implications of recent empirical study. Journal of Child Psy- Psychological Society 18, 58–63.
chology and Psychiatry 53, 469–489. Evenden, J.L., 1999. Varieties of impulsivity. Psychopharmacology 146, 348–361.
Cohen, D., 2010. Probabilistic epigenesis: an alternative causal model for conduct Fabiano, G.A., Pelham Jr., W.E., Coles, E.K., Gnagy, E.M., Chronis-Tuscano, A.,
disorders in children and adolescents. Neuroscience and Biobehavioral Reviews O‘Connor, B.C., 2009. A meta-analysis of behavioral treatments for attention-
34, 119–129. deficit/hyperactivity disorder. Clinical Psychology Review 29, 129–140.
Corbetta, M., Shulman, G.L., 2002. Control of goal-directed and stimulus-driven Fair, D.A., Posner, J., Nagel, B.J., Bathula, D., Dias, T.G.C., Mills, K.L., Blythe, M.S.,
attention in the brain. Nature Reviews. Neuroscience 3, 201–215. Giwa, A., Schmitt, C.F., Nigg, J.T., 2010. Atypical default network connectivity
Cortese, S., Kelly, C., Chabernaud, C., Proal, E., Di Martino, A., Milham, M.P., Castel- in youth with attention-deficit/hyperactivity disorder. Biological Psychiatry 68,
lanos, F.X., 2012. Toward systems neuroscience of ADHD: a meta-analysis of 55 1084–1091.
fMRI studies. The American Journal of Psychiatry 169, 1038–1055. Fan, J., Xu, P., Van Dam, N.T., Eilam-Stock, T., Gu, X., Luo, Y., Hof, P.R., 2012. Sponta-
Costa-Mattioli, M., Sossin, W.S., Klann, E., Sonenberg, N., 2009. Translational control neous brain activity relates to autonomic arousal. The Journal of Neuroscience:
of long-lasting synaptic plasticity and memory. Neuron 61, 10–26. the Official Journal of the Society for Neuroscience 32, 11176–11186.
Cowan, N., 2008. What are the differences between long-term, short-term, and Fell, J., Axmacher, N., 2011. The role of phase synchronization in memory processes.
working memory? Progress in Brain Research 169, 323–338. Nature Reviews. Neuroscience 12, 105–118.
Cowell, R.A., Bussey, T.J., Saksida, L.M., 2012. Empiricists are from venus, model- Fetterman, J.G., Killeen, P.R., 1990. A componential analysis of pacemaker-counter
ers are from mars: reconciling experimental and computational approaches timing systems. Journal of Experimental Psychology. Human Perception and
in cognitive neuroscience. Neuroscience and Biobehavioral Reviews 36, Performance 16, 766–780.
2371–2379. Fillenz, M., Lowry, J.P., Boutelle, M.G., Fray, A.E., 1999. The role of astrocytes and
Cramer, A.O.J., Waldorp, L.J., van der Maas, H.L.J., Borsboom, D., 2010. Comorbidity: noradrenaline in neuronal glucose metabolism. Acta Physiologica Scandinavia
a network perspective. The Behavioral and Brain Sciences 33, 137–150. 167, 275–284.
Crean, J.P., de Wit, H., Richards, J.B., 2000. Reward discounting as a measure of impul- Fleckenstein, A.E., Volz, T.J., Hanson, G.R., 2009. Psychostimulant-induced alterations
sive behavior in a psychiatric outpatient population. Experimental and Clinical in vesicular monoamine transporter-2 function: neurotoxic and therapeutic
Psychopharmacology 8, 155–162. implications. Neuropharmacology 56, 133–138.
Crumiller, M., Knight, B., Yu, Y., Kaplan, E., 2011. Estimating the amount of infor- Folks, J., Chhikara, R., 1978. The inverse Gaussian distribution and its statistical
mation conveyed by a population of neurons. Frontiers in Neuroscience 5, 90, application – a review. Journal of the Royal Statistical Society. Series B (Method-
http://dx.doi.org/10.3389/fnins.2011.00090. ological) 40, 263–289.
Dalley, J.W., Everitt, B.J., Robbins, T.W., 2011. Impulsivity, compulsivity, and top- Fortin, C., 2003. Attentional time-sharing in interval timing. In: Meck, W.H. (Ed.),
down cognitive control. Neuron 69, 680–694. Functional and Neural Mechanisms of Interval Timing. CRC Press, Boca Ratan,
Dane, A.V., Schachar, R.J., Tannock, R., 2000. Does actigraphy differentiate ADHD FL, pp. 235–260.
subtypes in a clinical research setting? Journal of the American Academy of Franke, B., Neale, B.M., Faraone, S.V., 2009. Genome-wide association studies in
Child and Adolescent Psychiatry 39, 752–760. ADHD. Human Genetics 126, 13–50.
De Sonneville, L., Boringa, J., Reuling, I., Lazeron, R., Ader, H., Polman, C., 2002. Friedman, D.L., Roberts, R., 1994. Compartment of brain-type cretine kinase and
Information processing characteristics in subtypes of multiple sclerosis. Neu- ubiquitous mitochondrial cretine kinase in neurons: evidence for a cretine phos-
ropsychologia 40, 1751–1765. phate energy shuttle in adult rat brain. The Journal of Comparative Neurology
de Zeeuw, P., Weusten, J., van Dijk, S., van Belle, J., Durston, S., 2012. Deficits in cog- 343, 500–511.
nitive control, timing and reward sensitivity appear to be dissociable in ADHD. Fusar-Poli, P., Rubia, K., Rossi, G., Sartori, G., Balottin, U., 2012. Striatal dopamine
PLoS One 7, e51416. transporter alterations in ADHD: pathophysiology or adaptation to psychostim-
Decamp, E., Clark, K., Schneider, J.S., 2011. Effects of the alpha-2 adrenoceptor agonist ulants? a meta-analysis. The American Journal of Psychiatry 169, 264–272.
guanfacine on attention and working memory in aged non-human primates. The Gallistel, C., 2009. The importance of proving the null. Psychological Review 116,
European Journal of Neuroscience 34, 1018–1022. 439–453.
Deco, G., Jirsa, V.K., McIntosh, A.R., 2010. Emerging concepts for the dynamical orga- Gamo, N., Wang, M., Arnsten, A., 2010. Methylphenidate and atomoxetine improve
nization of resting-state activity in the brain. Nature Reviews. Neuroscience 12, prefrontal cortical function via noradrenergic ␣-2 and dopaminergic D1 receptor
43–56. stimulation. Journal of the American Academy of Child and Adolescent Psychi-
DeKeyser, J., Zeinstra, E., Wilczak, N., 2004. Astrocytic ␤2 -adrenergic receptors and atry 49, 1011–1023.
multiple sclerosis. Neurobiology of Disease 15, 331–339. Gibbs, M.E., Bowser, D.N., Hutchinson, D.S., Loiacono, R.E., Summers, R.J., 2008a.
del Campo, N., Chamberlain, S.R., Sahakian, B.J., Robbins, T.W., 2011. The roles Memory processing in the avian hippocampus involves interactions between
of dopamine and noradrenaline in the pathophysiology and treatment of ␤-adrenoceptors, glutamate receptors, and metabolism. Neuropsychopharma-
attention-deficit/hyperactivity disorder. Biological Psychiatry 69, e145–e157. cology 33, 2831–2846.
Dietrich, A., Althaus, M., Hartman, C.A., Buitelaar, J.K., Minderaa, R.B., van den Hoof- Gibbs, M.E., Hutchinson, D., Hertz, L., 2008b. Astrocytic involvement in learning and
dakker, B.J., Hoekstra, P.J., 2012. Baroreflex sensitivity during rest and executive memory consolidation. Neuroscience and Biobehavioral Reviews 32, 927–944.
functioning in attention-deficit/hyperactivity disorder. The TRAILS study. Bio- Gilden, D.L., Hancock, H., 2007. Response variability in attention-deficit disorders.
logical Psychology 90, 249–257. Psychological Science: A Journal of the American Psychological Society 18, 796.
Drechsler, R., Rizzo, P., Steinhausen, H.C., 2010. Decision making with uncertain Gillberg, C., Kadesjö, B., 2003. Why bother about clumsiness? The implications of
reinforcement in children with attention deficit/hyperactivity disorder (ADHD). having developmental coordination disorder (DCD). Neural Plasticity 10, 59–68.
Child Neuropsychology: A Journal on Normal and Abnormal Development in Gonon, F., 2009. The dopaminergic hypothesis of attention-deficit/hyperactivity dis-
Childhood and Adolescence 16, 145–161. order needs re-examining. Trends in Neurosciences 32, 2–8.
Durston, S., Tottenham, N.T., Thomas, K.M., Davidson, M.C., Eigsti, I.M., Yang, Y., Ulug, Gooch, D., Snowling, M., Hulme, C., 2011. Time perception, phonological skills and
A.M., Casey, B., 2003. Differential patterns of striatal activation in young children executive function in children with dyslexia and/or ADHD symptoms. Journal of
with and without ADHD. Biological Psychiatry 53, 871–878. Child Psychology and Psychiatry 52, 195–203.
Dyck, M.J., Piek, J.P., Patrick, J., 2011. The validity of psychiatric diagnoses: the case of Gooch, D., Snowling, M.J., Hulme, C., 2012. Reaction time variability in children
‘specific’ developmental disorders. Research in Developmental Disabilities 32, with ADHD symptoms and/or dyslexia. Developmental Neuropsychology 37,
2704–2713. 453–472.
Egeland, J., Johansen, S.N., Ueland, T., 2010. Do low-effort learning strategies mediate Gottlob, L.R., 2004. Location cuing and response time distributions in visual atten-
impaired memory in ADHD? Journal of Learning Disabilities 50, 347–354. tion. Perception & Psychophysics 66, 1293–1302.
Egeth, H.E., Yantis, S., 1997. Visual attention, control, representation, and time Green, L., Myerson, J., McFadden, E., 1997. Rate of temporal discounting decreases
course. Annual Review of Psychology 48, 269–297. with amount of reward. Memory & Cognition 25, 715–723.
Engle, R.W., 2002. Working memory capacity as executive attention. Current Direc- Grier, R.A., Warm, J.S., Dember, W.N., Matthews, G., Galinsky, T.L., Szalma, J.L.,
tions in Psychological Science 11, 19–23. Parasuraman, R., 2003. The vigilance decrement reflects limitations in effortful
Epstein, J.N., Erkanli, A., Conners, C.K., Klaric, J., Costello, J.E., Angold, A., 2003. Rela- attention, not mindlessness. Human Factors 45, 349–359.
tions between continuous performance test performance measures and ADHD Halperin, J.M., Healey, D.M., 2011. The influences of environmental enrichment,
behaviors. Journal of Abnormal Child Psychology 31, 543–554. cognitive enhancement, and physical exercise on brain development: can we
Epstein, J.N., Hwang, M.E., Antonini, T., Langberg, J.M., Altaye, M., Arnold, L.E., alter the developmental trajectory of ADHD? Neuroscience and Biobehavioral
2010. Examining predictors of reaction times in children with ADHD and nor- Reviews 35, 621–634.
mal controls. Journal of the International Neuropsychological Society: JINS 16, Han, D., Gu, H., 2006. Comparison of the monoamine transporters from human and
138–147. mouse in their sensitivities to psychostimulant drugs. BMC Pharmacology 6, 6.
Epstein, J.N., Keith Conners, C., Hervey, A.S., Tonev, S.T., Eugene Arnold, L., Abikoff, Hanwella, R., Senanayake, M., de Silva, V., 2011. Comparative efficacy and accept-
H.B., Elliott, G., Greenhill, L.L., Hechtman, L., Hoagwood, K., 2006. Assessing med- ability of methylphenidate and atomoxetine in treatment of attention deficit
ication effects in the MTA study using neuropsychological outcomes. Journal of hyperactivity disorder in children and adolescents: a meta-analysis. BMC Psy-
Child Psychology and Psychiatry, and Allied Disciplines 47, 446–456. chiatry 11, 176.
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 653

Harel, Y., Appleboim, N., Lavie, M., Achiron, A., 2009. Single dose of methylphenidate Jacobi-Polishook, T., Shorer, Z., Melzer, I., 2009. The effect of methylphenidate on
improves cognitive performance in multiple sclerosis patients with impaired postural stability under single and dual task conditions in children with atten-
attention process. Journal of the Neurological Sciences 276, 38–40. tion deficit hyperactivity disorder – a double blind randomized control trial.
Harris, J.J., Attwell, D., 2012. The energetics of CNS white matter. The Journal of Journal of the Neurological Sciences 280, 15–21.
Neuroscience: the Official Journal of the Society for Neuroscience 32, 356–371. Jakoby, P., Schmidt, E., Ruminot, I., Gutiérrez, R., Barros, L.F., Deitmer, J.W., 2012.
Harris, J.J., Jolivet, R., Attwell, D., 2012. Synaptic energy use and supply. Neuron 75, Higher transport and metabolism of glucose in astrocytes compared with neu-
762–777. rons: a multiphoton study of hippocampal and cerebellar tissue slices. Cerebral
Hart, H., Radua, J., Mataix-Cols, D., Rubia, K., 2012. Meta-analysis of fMRI studies Cortex, http://dx.doi.org/10.1093/cercor/bhs309, First Online.
of timing in attention deficit hyperactivity disorder (ADHD). Neuroscience and James, W., 1890. Principles of Psychology, New York.
Biobehavioral Reviews 36, 2248–2256. Jensen, P.S., Hinshaw, S.P., Kraemer, H.C., Lenora, N., Newcorn, J.H., Abikoff, H.B.,
Hart, H., Radua, J., Nakao, T., Mataix-Cols, D., Rubia, K., 2013. Meta-analysis of March, J.S., Arnold, L.E., Cantwell, D.P., Conners, C.K., 2001. ADHD comorbidity
functional magnetic resonance imaging studies of inhibition and attention findings from the MTA study: comparing comorbid subgroups. Journal of the
in attention-deficit/hyperactivity disorder: exploring task-specific, stimulant American Academy of Child and Adolescent Psychiatry 40, 147–158.
medication, and age effects. Journal of the American Medical Association: Psy- Johansen, E.B., Aase, H., Meyer, A., Sagvolden, T., 2002. Attention-
chiatry 70, 185–198. deficit/hyperactivity disorder (ADHD) behaviour explained by dysfunctioning
Hazell, P.L., Kohn, M.R., Dickson, R., Walton, R.J., Granger, R.E., van Wyk, G.W., 2011. reinforcement and extinction processes. Behavioural Brain Research 130,
Core ADHD symptom improvement with atomoxetine versus methylphenidate. 37–45.
Journal of Attention Disorders 15, 674–683. Johansen, E.B., Killeen, P.R., Russell, V.A., Tripp, G., Wickens, J.R., Tannock, R.,
Heathcote, A., 2004. Fitting Wald and ex-Wald distributions to response time data: Williams, J., Sagvolden, T., 2009. Origins of altered reinforcement effects in
an example using functions for the S-PLUS package. Behavior Research Methods, ADHD. Behavioral Brain Function 5, 7.
Instruments, & Computers 36, 678–694. Johansen, E.B., Killeen, P.R., Sagvolden, T., 2007. Behavioral variability, elimination of
Heinrich, H., Moll, G.H., Dickhaus, H., Kolev, V., Yordanova, J., Rothenberger, A., responses, and delay-of-reinforcement gradients in SHR and WKY rats. Behav-
2001. Time-on-task analysis using wavelet networks in an event-related poten- ioral Brain Function 3, 60.
tial study on attention-deficit hyperactivity disorder. Clinical Neurophysiology: Johansen, E.B., Sagvolden, T., Aase, H., Russell, V.A., 2005. The dynamic developmen-
Official Journal of the International Federation of Clinical Neurophysiology 112, tal theory of attention-deficit/hyperactivity disorder (ADHD): present status and
1280–1287. future perspectives. The Behavioral and Brain Sciences 28, 451–454.
Helton, W.S., Warm, J.S., 2008. Signal salience and the mindlessness theory of vigi- Johnson, K.A., Kelly, S.P., Bellgrove, M.A., Barry, E., Cox, M., Gill, M., Robertson, I.H.,
lance. Acta Psychologica 129, 18–25. 2007. Response variability in attention deficit hyperactivity disorder: evidence
Hertz, L., Lovatt, D., Goldman, S.A., Nedergaard, M., 2010. Adrenoceptors in brain: for neuropsychological heterogeneity. Neuropsychologia 45, 630–6388.
cellular gene expression and effects on astrocytic metabolism and [Ca2+ ]i . Neu- Kalff, A.C., De Sonneville, L.M.J., Hurks, P.P.M., Hendriksen, J.G.M., Kroes, M., Feron,
rochemistry International 57, 411–420. F.J.M., Steyaert, J., Van Zeben, T.M.C.B., Vles, J.S.H., Jolles, J., 2005. Speed, speed
Hertz, L., Peng, L., Dienel, G.A., 2006. Energy metabolism in astrocytes: high variability, and accuracy of information processing in 5 to 6-year-old children
rate of oxidative metabolism and spatiotemporal dependence on glycol- at risk of ADHD. Journal of the International Neuropsychological Society: JINS
ysis/glycogenolysis. Journal of Cerebral Blood Flow and Metabolism 27, 11, 173–183.
219–249. Kane, M.J., Bleckley, M.K., Conway, A.R.A., Engle, R.W., 2001. A controlled-attention
Hervey, A.S., Epstein, J.N., Curry, J.F., Tonev, S., Arnold, L.E., Conners, C.K., Hinshaw, view of working-memory capacity. Journal of Experimental Psychology. General
S.P., Swanson, J.M., Hechtman, L., 2006. Reaction time distribution analysis of 130, 169.
neuropsychological performance in an ADHD sample. Child Neuropsychology: Kann, O., Huchzermeyer, C., Kovács, R., Wirtz, S., Schuelke, M., 2011. Gamma oscil-
A Journal on Normal and Abnormal Development in Childhood and Adolescence lations in the hippocampus require high complex I gene expression and strong
12, 125–140. functional performance of mitochondria. Brain 134, 345–358.
Hoffmann, S., Tittgemeyer, M., Von Cramon, D.Y., 2007. Cognitive impairment in Kaplan, S., Berman, M.G., 2010. Directed attention as a common resource for exec-
multiple sclerosis. Current Opinion in Neurology 20, 275–280. utive functioning and self-regulation. Perspectives on Psychological Science 5,
Howarth, C., Gleeson, P., Attwell, D., 2012. Updated energy budgets for neural com- 43–57.
putation in the neocortex and cerebellum. Journal of Cerebral Blood Flow and Karalunas, S.L., Huang-Pollock, C.L., 2013. Integrating impairments in reaction time
Metabolism 32, 1222–1232. and executive function using a diffusion model framework. Journal of Abnormal
Howells, F.M., Russell, V.A., 2008. Glutamate-stimulated release of norepinephrine Child Psychology, http://dx.doi.org/10.1007/s10802-013-9715, Online First 1-.
in hippocampal slices of animal models of attention-deficit/hyperactivity disor- Karalunas, S.L., Huang-Pollock, C.L., Nigg, J.T., 2012a. Decomposing attention-
der (spontaneously hypertensive rat) and depression/anxiety-like behaviours deficit/hyperactivity disorder (ADHD)-related effects in response speed and
(Wistar-Kyoto rat). Brain Research 1200, 107–115. variability. Neuropsychology 26, 684.
Howells, F.M., Stein, D.J., Russell, V.A., 2012. Synergistic tonic and phasic activity Karalunas, S.L., Huang-Pollock, C.L., Nigg, J.T., 2012b. Is reaction time variability in
of the locus coeruleus norepinephrine (LC-NE) arousal system is required for ADHD mainly at low frequencies? Journal of Child Psychology and Psychiatry,
optimal attentional performance. Metabolic Brain Disease 27, 267–274. http://dx.doi.org/10.1111/jcpp.12028.
Hu, Y., 2002. A temporary local energy pool coupled to neuronal activity: fluctua- Kasischke, K.A., Vishwasrao, H.D., Fisher, P.J., Zipfel, W.R., Webb, W.W., 2004. Neural
tions of extracellular lactate levels in rat brain monitored with rapid-response activity triggers neuronal oxidative metabolism followed by astrocytic glycoly-
enzyme-based sensor. Journal of Neurochemistry 69, 1484–1490. sis. Science 305, 99–103.
Hu, Y., Wilson, G.S., 1997. A temporary local energy pool coupled to neuronal activity: Kenemans, J., Bekker, E., Lijffijt, M., Overtoom, C., Jonkman, L., Verbaten, M., 2005.
fluctuations of extracellular lactate levels in rat brain monitored with rapid- Attention deficit and impulsivity: selecting, shifting, and stopping. International
response enzyme-based sensor. Journal of Neurochemistry 69, 1484–1490. Journal of Psychophysiology: Official Journal of the International Organization
Huang, Z., Wu, Y.L., Hedrick, N., Gutmann, D.H., 2003. T-cadherin-mediated cell of Psychophysiology 58, 59–70.
growth regulation involves G2 phase arrest and requires p21CIP1/WAF1 expres- Kerns, K.A., McInerney, R.J., Wilde, N.J., 2001. Time reproduction, working memory,
sion. Molecular and Cellular Biology 23, 566–578. and behavioral inhibition in children with ADHD. Child Neuropsychology: A
Huang-Pollock, C.L., Karalunas, S.L., Tam, H., Moore, A.N., 2012. Evaluating vigilance Journal on Normal and Abnormal Development in Childhood and Adolescence
deficits in ADHD: a meta-analysis of CPT performance. Journal of Abnormal 7, 21–31.
Psychology 121, 360–371. Kessler, R.C., Adler, L., Barkley, R., Biederman, J., Conners, C.K., Demler, O., Faraone,
Huesmann, L.R., 2007. The impact of electronic media violence: scientific theory and S.V., Greenhill, L.L., Howes, M.J., Secnik, K., 2006. The prevalence and correlates of
research. The Journal of Adolescent Health: Official Publication of the Society for adult ADHD in the United States: results from the National Comorbidity Survey
Adolescent Medicine 41, S6–S13. Replication. The American Journal of Psychiatry 163, 716.
Hurst, R.M., Kepley, H.O., McCalla, M.K., Livermore, M.K., 2011. Internal consistency Kievit, R.A., Romeijn, J.W., Waldorp, L.J., Wicherts, J.M., Scholte, H.S., Borsboom,
and discriminant validity of a delay-discounting task with an adult self-reported D., 2011. Mind the gap: a psychometric approach to the reduction problem.
ADHD sample. Journal of Attention Disorders 15, 412–422. Psychological Inquiry 22, 67–87.
Hutchinson, D.S., Catus, S.L., Merlin, J., Summers, R.J., Gibbs, M.E., 2011. ␣2- Killeen, P.R., 2001. Writing and overwriting short-term memory. Psychonomic Bul-
Adrenoceptors activate noradrenaline-mediated glycogen turnover in chick letin & Review 8, 18–43.
astrocytes. Journal of Neurochemistry 117, 915–926. Killeen, P.R., 2002. Scalar counters. Learning and Motivation 33, 63–87.
Hutchinson, D.S., Summers, R.J., Gibbs, M.E., 2007. ␤2- and ␤3-Adrenoceptors acti- Killeen, P.R., 2007. Replication statistics. In: Osborne, J.W. (Ed.), Best Practices in
vate glucose uptake in chick astrocytes by distinct mechanisms: a mechanism Quantitative Methods. Sage, Thousand Oaks, CA, pp. 103–124.
for memory enhancement? Journal of Neurochemistry 103, 997–1008. Killeen, P.R., 2009. An additive-utility model of delay discounting. Psychological
Hyder, F., Patel, A.B., Gjedde, A., Rothman, D.L., Behar, K.L., Shulman, R.G., 2006. Review 116, 602–619.
Neuronal-glial glucose oxidation and glutamatergic-GABAergic function. Jour- Killeen, P.R., 2011. Models of trace decay, eligibility for reinforcement, and delay
nal of Cerebral Blood Flow and Metabolism 26, 865–877. of reinforcement gradients, from exponential to hyperboloid. Behavioural Pro-
Imeraj, L., Sonuga-Barke, E., Antrop, I., Roeyers, H., Wiersema, R., Bal, S., Deboutte, cesses 8, 57–63.
D., 2012. Altered circadian profiles in attention deficit/hyperactivity disorder: an Killeen, P.R., Tannock, R., Sagvolden, T., 2012. The four causes of ADHD: a framework.
integrative review and theoretical framework for future studies. Neuroscience In: Stanford, S.C., Tannock, R. (Eds.), Behavioral Neuroscience of Attention Deficit
and Biobehavioral Reviews 36, 1897–1919. Hyperactivity Disorder and its Treatment. Springer-Verlag, Berlin, pp. 391–425.
Itami, S., Uno, H., 2002. Orbitofrontal cortex dysfunction in attention-deficit hyper- Killeen, P.R., Taylor, T.J., 2000. How the propagation of error through stochastic
activity disorder revealed by reversal and extinction tasks. Neuroreport 13, counters affects time discrimination and other psychophysical judgments. Psy-
2453–2457. chological Review 107, 430–459.
654 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

Killeen, P.R., Weiss, N.A., 1987. Optimal timing and the Weber function. Psycholog- Mangia, S., Garreffa, G., Bianciardi, M., Giove, F., Di Salle, F., Maraviglia, B., 2003. The
ical Review 94, 455–468. aerobic brain: lactate decrease at the onset of neural activity. Neuroscience 118,
Klein, C., Wendling, K., Huettner, P., Ruder, H., Peper, M., 2006. Intra-subject 7–10.
variability in attention-deficit hyperactivity disorder. Biological Psychiatry 60, Mangia, S., Tkå, I., Gruetter, R., Van de Moortele, P.F., Maraviglia, B., Ugurbil, K., 2006.
1088–1097. Sustained neuronal activation raises oxidative metabolism to a new steady-state
Kofler, M.J., Rapport, M.D., Bolden, J., Altro, T.A., 2008a. Working memory as a core level: evidence from 1H NMR spectroscopy in the human visual cortex. Journal
deficit in ADHD: preliminary findings and implications. The ADHD Report 16, of Cerebral Blood Flow and Metabolism 27, 1055–1063.
8–14. Marco, R., Miranda, A., Schlotz, W., Melia, A., Mulligan, A., Müller, U., Andreou, P., But-
Kofler, M.J., Rapport, M.D., Bolden, J., Sarver, D.E., Raiker, J.S., 2010. ADHD and ler, L., Christiansen, H., Gabriels, I., 2009. Delay and reward choice in ADHD: An
working memory: the impact of central executive deficits and exceeding stor- experimental test of the role of delay aversion. Neuropsychology 23, 367–380.
age/rehearsal capacity on observed inattentive behavior. Journal of Abnormal Marcus, G.F., 2006. Cognitive architecture and descent with modification. Cognition
Child Psychology 38, 149–161. 101, 443–465.
Kofler, M.J., Rapport, M.D., Bolden, J., Sarver, D.E., Raiker, J.S., Alderson, R.M., 2011. Martinussen, R., Hayden, J., Hogg-Johnson, S., Tannock, R., 2005. A meta-analysis of
Working memory deficits and social problems in children with ADHD. Journal working memory impairments in children with attention-deficit/hyperactivity
of Abnormal Child Psychology 39, 805–817. disorder. Journal of the American Academy of Child and Adolescent Psychiatry
Kofler, M.J., Rapport, M.D., Matt Alderson, R., 2008b. Quantifying ADHD classroom 44, 377–384.
inattentiveness, its moderators, and variability: a meta-analytic review. Journal Marzocchi, G.M., Oosterlaan, J., Zuddas, A., Cavolina, P., Geurts, H., Redigolo, D., Vio, C.,
of Child Psychology and Psychiatry, and Allied Disciplines 49, 59–69. Sergeant, J.A., 2008. Contrasting deficits on executive functions between ADHD
Kollins, S.H., Lane, S.D., Shapiro, S.K., 1997. Experimental analysis of childhood and reading disabled children. Journal of Child Psychology and Psychiatry, and
psychopathology: a laboratory matching analysis of the behavior of children Allied Disciplines 49, 543–552.
diagnosed with attention-deficit hyperactivity disorder (ADHD). Psychological Mayes, S.D., Calhoun, S.L., 2007. Learning, attention, writing, and processing speed
Review 47, 25–44. in typical children and children with ADHD, autism, anxiety, depression, and
Kooistra, L., Ramage, B., Crawford, S., Cantell, M., Wormsbecker, S., Gibbard, B., oppositional-defiant disorder. Child Neuropsychology: A Journal on Normal and
Kaplan, B.J., 2009. Can attention deficit hyperactivity disorder and fetal alco- Abnormal Development in Childhood and Adolescence 13, 469–493.
hol spectrum disorder be differentiated by motor and balance deficits? Human Mayford, M., Siegelbaum, S.A., Kandel, E.R., 2012. Synapses and memory storage. In:
Movement Science 28, 529–542. Cold Spring Harbor Perspectives in Biology, p. 4.
Koós, T., Tepper, J.M., 1999. Inhibitory control of neostriatal projection neurons by Metin, B., Roeyers, H., Wiersema, J.R., van der Meere, J., Sonuga-Barke, E., 2012. A
GABAergic interneurons. Nature Neuroscience 2, 467–472. meta-analytic study of event rate effects on go/no-go performance in attention-
Kramer, A.F., Cepeda, N.J., Cepeda, M.L., 2001. Methylphenidate effects on task- deficit/hyperactivity disorder. Biological Psychiatry 72, 9900–9996.
switching performance in attention-deficit/hyperactivity disorder. Journal of Meyer, A., Sagvolden, T., 2006. Fine motor skills in South African children with
the American Academy of Child and Adolescent Psychiatry 40, 1277–1284. symptoms of ADHD: influence of subtype, gender, age, and hand dominance.
Kramer, F., Pollnow, H., 1930. Hyperkinetische Zustandsbilder im Kindesalter. Zen- Behavioral Brain Function 2, 33.
tralblatt für die gesamte Neurologie und Psychiatrie 57, 844–845. Mick, E., McGough, J., Loo, S., Doyle, A.E., Wozniak, J., Wilens, T.E., Smalley, S.,
Kronenberg, G., Ende, G., Alm, B., Deuschle, M., Heuser, I., Colla, M., 2008. Increased McCracken, J., Biederman, J., Faraone, S.V., 2011. Genome-wide association study
NAA and reduced choline levels in the anterior cingulum following chronic of the child behavior checklist dysregulation profile. Journal of the American
methylphenidate. European Archives of Psychiatry and Clinical Neuroscience Academy of Child & Adolescent Psychiatry 50, 807–817.
258, 446–450. Min, R., Santello, M., Nevian, T., 2012. The computational power of astrocyte medi-
Kruglanski, A.W., Bélanger, J.J., Chen, X., Köpetz, C., Pierro, A., Mannetti, L., 2012. The ated synaptic plasticity. Frontiers in Computational Neuroscience 6, 93.
energetics of motivated cognition. A force-field analysis. Psychological Review Minzenberg, M.J., Carter, C.S., 2007. Modafinil: a review of neurochemical actions
119, 1–20. and effects on cognition. Neuropsychopharmacology 33, 1477–1502.
Kuntsi, J., Wood, A.C., Van Der Meere, J., Asherson, P., 2009. Why cognitive per- Mizuno, K., Tanaka, M., Fukuda, S., Yamano, E., Shigihara, Y., Imai-Matsumura, K.,
formance in ADHD may not reveal true potential: findings from a large Watanabe, Y., 2011. Low visual information-processing speed and attention are
population-based sample. Journal of the International Neuropsychological Soci- predictors of fatigue in elementary and junior high school students. Behavioral
ety: JINS 15, 570. Brain Function 7, 20.
Landau, A.N., Elwan, D., Holtz, S., Prinzmetal, W., 2012. Voluntary and involuntary Molina, B.S.G., Hinshaw, S.P., Swanson, J.M., Arnold, L.E., Vitiello, B., Jensen, P.S.,
attention vary as a function of impulsivity. Psychonomic Bulletin & Review 19, Epstein, J.N., Hoza, B., Hechtman, L., Abikoff, H.B., 2009. The MTA at 8 years:
405–411. prospective follow-up of children treated for combined-type ADHD in a multi-
Larson, K., Russ, S.A., Kahn, R.S., Halfon, N., 2011. Patterns of comorbidity, function- site study. Journal of the American Academy of Child and Adolescent Psychiatry
ing, and service use for US children with ADHD, 2007. Pediatrics 127, 462–470. 48, 484–500.
Lee, R., Jacobson, L.A., Pritchard, A.E., Ryan, M.S., Yu, Q., Denckla, M.B., Mostofsky, Mulder, M.J., Bos, D., Weusten, J.M.H., van Belle, J., van Dijk, S.C., Simen, P., van Enge-
S., Mahone, E.M., 2012. Jitter reduces response-time variability in ADHD: an land, H., Durston, S., 2010. Basic impairments in regulating the speed-accuracy
Ex-Gaussian analysis. Journal of Attention Disorders, Online First. tradeoff predict symptoms of attention-deficit/hyperactivity disorder. Biological
Leegsma-Vogt, G., van der Werf, S., Venema, K., Korf, J., 2004. Modeling cerebral arte- Psychiatry 68, 1114–1119.
riovenous lactate kinetics after intravenous lactate infusion in the rat. Journal Murre, J.M.J., Chessa, A.G., 2011. Power laws from individual differences in learn-
of Cerebral Blood Flow and Metabolism 24, 1071–1080. ing and forgetting: mathematical analyses. Psychonomic Bulletin & Review 3,
Leth-Steensen, C., King-Elbaz, Z., Douglas, V.I., 2000. Mean response times, variabil- 592–597.
ity, and skew in the responding of ADHD children: a response time distributional Nagel, B.J., Bathula, D., Herting, M., Schmitt, C., Kroenke, C.D., Fair, D., Nigg,
approach. Acta Psychologica 104, 167–190. J.T., 2011. Altered white matter microstructure in children with attention-
Liang, S.L., Carlson, G.C., Coulter, D.A., 2006. Dynamic regulation of synaptic GABA deficit/hyperactivity disorder. Journal of the American Academy of Child and
release by the glutamate-glutamine cycle in hippocampal area CA1. The Jour- Adolescent Psychiatry 50, 283–292.
nal of Neuroscience: the Official Journal of the Society for Neuroscience 26, Nesse, R.M., Stein, D.J., 2012. Towards a genuinely medical model for psychiatric
8537–8548. nosology. BMC Medicine 10, 5.
Lijffijt, M., Kenemans, J.L., Verbaten, M.N., van Engeland, H., 2005. A meta-analytic Nevin, J.A., Grace, R.C., 2001. Behavioral momentum and the law of effect. The Behav-
review of stopping performance in attention-deficit/hyperactivity disorder: ioral and Brain Sciences 23, 73–90.
deficient inhibitory motor control? Journal of Abnormal Psychology 114, Newman, L.A., Korol, D.L., Gold, P.E., 2011. Lactate produced by glycogenolysis in
216–222. astrocytes regulates memory processing. PloS One 6, e28427.
Lingam, R., Golding, J., Jongmans, M.J., Hunt, L.P., Ellis, M., Emond, A., 2010. The asso- Nigg, J.T., 2001. Is ADHD a disinhibitory disorder? Psychological Bulletin 127,
ciation between developmental coordination disorder and other developmental 571–598.
traits. Pediatrics 126, e1109. Nigg, J.T., 2005. Neuropsychologic theory and findings in attention-
Logan, G.D., Schachar, R.J., Tannock, R., 1997. Impulsivity and inhibitory control. deficit/hyperactivity disorder: the state of the field and salient challenges for
Psychological Science: A Journal of the American Psychological Society 8, 60–64. the coming decade. Biological Psychiatry 57, 1424–1435.
Luman, M., Oosterlaan, J., Sergeant, J.A., 2005. The impact of reinforcement con- Nigg, J.T., 2006. What Causes ADHD? Understanding What Goes Wrong and Why.
tingencies on AD/HD: a review and theoretical appraisal. Clinical Psychology Guilford Press.
Review 25, 183–213. Noreika, V., Falter, C.M., Rubia, K., 2013. Timing deficits in Attention-
Luman, M., Tripp, G., Scheres, A., 2010. Identifying the neurobiology of altered rein- Deficit/Hyperactivity Disorder (ADHD): evidence from neurocognitive and
forcement sensitivity in ADHD: A review and research agenda. Neuroscience neuroimaging studies. Neuropsychologia 52, 235–266.
and Biobehavioral Reviews 34, 744–754. Normand, S., Flora, D.B., Toplak, M.E., Tannock, R., 2012. Evidence for a general ADHD
Luman, M., Van Meel, C.S., Oosterlaan, J., Sergeant, J.A., Geurts, H.M., 2009. Does factor from a longitudinal general school population study. Journal of Abnormal
reward frequency or magnitude drive reinforcement-learning in attention- Child Psychology 4, 555–567.
deficit/hyperactivity disorder? Psychiatry Research 168, 222–229. Oosterlaan, J., Logan, G.D., Sergeant, J.A., 1998. Response inhibition in AD/HD, CD,
Mackworth, N., 1956. Vigilance. Nature 178, 1375–1377. comorbid AD/HD + CD, anxious, and control children: a meta-analysis of stud-
Madden, G.J., Bickel, W.K., 2009. Impulsivity: Theory, Science, and Neuroscience of ies with the stop task. Journal of Child Psychology and Psychiatry, and Allied
Discounting. American Psychological Association, Washington, DC. Disciplines 39, 411–425.
Magistretti, P.J., 2009. Role of glutamate in neuron-glia metabolic coupling. The Oosterlaan, J., Sergeant, J.A., 1998a. Effects of reward and response cost on response
American Journal of Clinical Nutrition 90, 875S. inhibition in AD/HD, disruptive, anxious, and normal children. Journal of Abnor-
Magistretti, P.J., 2011. In: Russell, V.A. (Ed.), Just-in-Time Signalling, Capetown, SA. mal Child Psychology 26, 161–174.
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 655

Oosterlaan, J., Sergeant, J.A., 1998b. Response inhibition and response re- Rastmanesh, R., 2010. Drug policy and treatment bias due to the dopamine-deficit
engagement in attention-deficit/hyperactivity disorder, disruptive, anxious and theory of child attention-deficit hyperactivity disorder. Attention Deficit and
normal children. Behavioural Brain Research 94, 33–43. Hyperactivity Disorders 2, 149–157.
Palmer, E.M., Horowitz, T.S., Torralba, A., Wolfe, J.M., 2011. What are the shapes of Ratcliff, R., McKoon, G., 2008. The diffusion decision model. Theory and data for
response time distributions in visual search? Journal of Experimental Psychol- two-choice decision tasks. Neural Computation 20, 873–922.
ogy. Human Perception and Performance 37, 58–71. Ratcliff, R., Rouder, J.N., 1998. Modeling response times for two-choice decisions.
Paloyelis, Y., Asherson, P., Kuntsi, J., 2009. Are ADHD symptoms associated with Psychological Science: A Journal of the American Psychological Society 9,
delay aversion or choice impulsivity? A general population study. Journal of the 347–356.
American Academy of Child and Adolescent Psychiatry 48, 837–846. Ratcliff, R., Tuerlinckx, F., 2002. Estimating parameters of the diffusion model:
Paloyelis, Y., Asherson, P., Mehta, M.A., Faraone, S.V., Kuntsi, J., 2010a. DAT1 and approaches to dealing with contaminant reaction times and parameter vari-
COMT effects on delay discounting and trait impulsivity in male adolescents ability. Psychonomic Bulletin & Review 9, 438–481.
with attention deficit/hyperactivity disorder and healthy controls. Neuropsy- Rhodes, S.M., Park, J., Seth, S., Coghill, D.R., 2011. A comprehensive investigation of
chopharmacology 35, 2414–2426. memory impairment in attention deficit hyperactivity disorder and oppositional
Paloyelis, Y., Mehta, M.A., Kuntsi, J., Asherson, P., 2007. Functional MRI in ADHD: a defiant disorder. Journal of Child Psychology and Psychiatry 53, 128–137.
systematic literature review. Expert Review of Neurotherapeutics 7, 1337–1356. Riccio, C.A., Reynolds, C.R., Lowe, P., Moore, J.J., 2002. The continuous performance
Paloyelis, Y., Stahl, D.R., Mehta, M., 2010b. Are steeper discounting rates in attention- test: a window on the neural substrates for attention? Archives of Clinical
deficit/hyperactivity disorder specifically associated with hyperactivity- Neuropsychology: the Official Journal of the National Academy of Neuropsy-
impulsivity symptoms or is this a statistical artifact? Biological Psychiatry 68, chologists 17, 235–272.
e15–e16. Richard, F.D., Bond Jr., C.F., Stokes-Zoota, J.J., 2003. One hundred years of
Panksepp, J., 2007. Can PLAY diminish ADHD and facilitate the construction of the social psychology quantitatively described. Review of General Psychology 7,
social brain? Journal of the Canadian Academy of Child and Adolescent Psychi- 331–363.
atry 16, 57–66. Richter, J.D., Klann, E., 2009. Making synaptic plasticity and memory last: mecha-
Pannasch, U., Vargová, L., Reingruber, J., Ezan, P., Holcman, D., Giaume, C., Syková, nisms of translational regulation. Genes & Development 23, 1–11.
E., Rouach, N., 2011. Astroglial networks scale synaptic activity and plasticity. Rigotti, D.J., Gonen, O., Grossman, R.I., Babb, J.S., Falini, A., Benedetti, B., Filippi,
Proceedings of the National Academy of Sciences 108, 8467–8472. M., 2011. Global N-acetylaspartate declines even in benign multiple sclerosis.
Parasurman, R., Davies, D., 1982. The Psychology of Vigilance. Academic Press, Lon- American Journal of Neuroradiology 32, 204–209.
don. Rivero, O., Sich, S., Popp, S., Schmitt, A., Franke, B., Lesch, K.P., 2012.
Pattij, T., Schetters, D., Schoffelmeer, A.N.M., van Gaalen, M.M., 2011. On the Impact of the ADHD-susceptibility gene CDH13 on development and
improvement of inhibitory response control and visuospatial attention by indi- function of brain networks. European Neuropsychopharmacology:
rect and direct adrenoceptor agonists. Psychopharmacology 219, 327–340. the Journal of the European College of Neuropsychopharmacology,
Pattyn, N., Neyt, X., Henderickx, D., Soetens, E., 2008. Psychophysiological inves- http://dx.doi.org/10.1016/j.euroneuro.2012.06.00.
tigation of vigilance decrement: boredom or cognitive fatigue? Physiology & Robbins, T., Curran, H., de Wit, H., 2012. Special issue on impulsivity and compul-
Behavior 93, 369–378. sivity. Psychopharmacology 219, 251–252.
Pelham Jr., W.E., Fabiano, G.A., 2008. Evidence-based psychosocial treatments for Rubia, K., Noorloos, J., Smith, A., Gunning, B., Sergeant, J., 2003. Motor timing
attention-deficit/hyperactivity disorder. Journal of Clinical Child & Adolescent deficits in community and clinical boys with hyperactive behavior: the effect
Psychology 37, 184–214. of methylphenidate on motor timing. Journal of Abnormal Child Psychology 31,
Pellerin, L., Bouzier-Sore, A.K., Aubert, A., Serres, S., Merle, M., Costalat, R., Magistretti, 301–313.
P.J., 2007. Activity-dependent regulation of energy metabolism by astrocytes: an Rubia, K., Taylor, A., Taylor, E., Sergeant, J.A., 1999. Synchronization, anticipation, and
update. Glia 55, 1251–1262. consistency in motor timing of children with dimensionality defined attention
Pellerin, L., Magistretti, P.J., 2011. Sweet sixteen for ANLS. Journal of Cerebral Blood deficit hyperactive behaviour. Perceptual and Motor Skills 89, 1237–1258.
Flow and Metabolism 32, 1152–1166. Rucklidge, J.J., Tannock, R., 2002. Neuropsychological profiles of adolescents with
Perlov, E., Philipsen, A., Matthies, S., Drieling, T., Maier, S., Bubl, E., Hesslinger, B., ADHD: effects of reading difficulties and gender. Journal of Child Psychology
Buechert, M., Henning, J., Ebert, D., 2009. Spectroscopic findings in attention- and Psychiatry, and Allied Disciplines 43, 988–1003.
deficit/hyperactivity disorder: Review and meta-analysis. The World Journal of Rueda, M.R., Fan, J., McCandliss, B.D., Halparin, J.D., Gruber, D.B., Lercari, L.P., Posner,
Biological Psychiatry: the Official Journal of the World Federation of Societies of M.I., 2004. Development of attentional networks in childhood. Neuropsycholo-
Biological Psychiatry 10, 355–365. gia 42, 1029–1040.
Polanczyk, G., de Lima, M.É., Horta, B., Biederman, J., Rohde, L., 2007. The world- Russell, V.A., 2001. Increased AMPA receptor function in slices containing the pre-
wide prevalence of ADHD: a systematic review and metaregression analysis. frontal cortex of spontaneously hypertensive rats. Metabolic Brain Disease 16,
The American Journal of Psychiatry 164, 942–948. 143–149.
Prado, J., Carp, J., Weissman, D.H., 2011. Variations of response time in a selective Russell, V.A., Oades, R.D., Tannock, R., Killeen, P.R., Auerbach, J.G., Johansen, E.B.,
attention task are linked to variations of functional connectivity in the atten- Sagvolden, T., 2006. Response variability in Attention-Deficit/Hyperactivity Dis-
tional network. NeuroImage 54, 541–549. order: a neuronal and glial energetics hypothesis. Behavioral Brain Function 2,
Prado, J., Weissman, D.H., 2011. Spatial attention influences trial-by-trial relation- 30.
ships between response time and functional connectivity in the visual cortex. Russell, V.A., Wiggins, T.M., 2000. Increased glutamate-stimulated norepinephrine
NeuroImage 54, 465–473. release from prefrontal cortex slices of spontaneously hypertensive rats.
Qi, X., Katsuki, F., Meyer, T., Rawley, J.B., Zhou, X., Douglas, K.L., Constantinidis, Metabolic Brain Disease 15, 297–304.
C., 2010. Comparison of neural activity related to working memory in primate Sagvolden, T., Johansen, E.B., Aase, H., Russell, V.A., 2005. A dynamic developmen-
dorsolateral prefrontal and posterior parietal cortex. Frontiers in Systems Neu- tal theory of attention-deficit/hyperactivity disorder (ADHD) predominantly
roscience 4, 12. hyperactive/impulsive and combined subtypes. The Behavioral and Brain Sci-
Querne, L., Berquin, P., 2009. Distinct response time distributions in attention deficit ences 28, 397–419, discussion 419–468.
hyperactivity disorder subtypes. Journal of Attention Disorders 13, 66–77. Sakurai, T., Mieda, M., Tsujino, N., 2010. The orexin system: roles in sleep/wake
Raiker, J.S., Rapport, M.D., Kofler, M.J., Sarver, D.E., 2012. Objectively-measured regulation. Annals of the New York Academy of Sciences 1200, 149–161.
impulsivity and Attention-Deficit/Hyperactivity Disorder (ADHD): testing com- Sandau, U.S., Alderman, Z., Corfas, G., Ojeda, S.R., Raber, J., 2012. Astrocyte-specific
peting predictions from the working memory and behavioral inhibition models disruption of SynCAM1 signaling results in ADHD-like behavioral manifesta-
of ADHD. Journal of Abnormal Child Psychology 40, 699–713. tions. PloS One 7, e36424.
Rammohan, K., Rosenberg, J., Lynn, D., Blumenfeld, A., Pollak, C., Nagaraja, H., 2002. Sarter, M., Gehring, W.J., Kozak, R., 2006. More attention must be paid: the neurobi-
Efficacy and safety of modafinil (Provigil (r)) for the treatment of fatigue in mul- ology of attentional effort. Brain Research Reviews 51, 145–160.
tiple sclerosis: a two centre phase 2 study. Journal of Neurology, Neurosurgery, Sarter, M., Hasselmo, M.E., Bruno, J.P., Givens, B., 2005. Unraveling the attentional
and Psychiatry 72, 179–183. functions of cortical cholinergic inputs: interactions between signal-driven
Rapanelli, M., Frick, L.R., Zanutto, B.S., 2011. Learning an operant conditioning task and cognitive modulation of signal detection. Brain Research Reviews 48,
differentially induces gliogenesis in the medial prefrontal cortex and neurogen- 98–111.
esis in the hippocampus. PloS One 6, e14713. Sarver, D.E., Rapport, M.D., Kofler, M.J., Scanlan, S.W., Raiker, J.S., Altro, T.A., Bolden,
Rapport, M.D., Alderson, R.M., Kofler, M.J., Sarver, D.E., Bolden, J., Sims, V., 2008. J., 2012. Attention problems, phonological short-term memory, and visuospa-
Working memory deficits in boys with attention-deficit/hyperactivity disorder tial short-term memory: differential effects on near-and long-term scholastic
(ADHD): the contribution of central executive and subsystem processes. Journal achievement. Learning and Individual Differences 22, 8–19.
of Abnormal Child Psychology 36, 825–837. Sawyer, A.M., Taylor, E., Chadwick, O., 2001. The effect of off-task behaviors on the
Rapport, M.D., Bolden, J., Kofler, M.J., Sarver, D.E., Raiker, J.S., Alderson, R.M., 2009a. task performance of hyperkinetic children. Journal of Attention Disorders 5, 1.
Hyperactivity in boys with attention-deficit/hyperactivity disorder (ADHD): a Scheres, A., Dijkstra, M., Ainslie, E., Balkan, J., Reynolds, B., Sonuga-Barke, E., Castel-
ubiquitous core symptom or manifestation of working memory deficits. Journal lanos, F.X., 2006. Temporal and probabilistic discounting of rewards in children
of Abnormal Child Psychology 37, 521–534. and adolescents: effects of age and ADHD symptoms. Neuropsychologia 44,
Rapport, M.D., Chung, K.M., Shore, G., Isaacs, P., 2001. A conceptual model of child 2092–2103.
psychopathology: implications for understanding attention deficit hyperactiv- Scheres, A., Lee, A., Sumiya, M., 2008. Temporal reward discounting and ADHD:
ity disorder and treatment efficacy. Journal of Clinical Child Psychology 30, task and symptom specific effects. Journal of Neural Transmission 115,
48–58. 221–226.
Rapport, M.D., Kofler, M.J., Alderson, R.M., Timko, T.M., DuPaul, G.J., 2009b. Variability Scheres, A., Oosterlaan, J., Swanson, J., Morein-Zamir, S., Meiran, N., Schut, H.,
of attention processes in ADHD. Journal of Attention Disorders 12, 563–573. Vlasveld, L., Sergeant, J.A., 2003. The effect of methylphenidate on three forms of
656 P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657

response inhibition in boys with AD/HD. Journal of Abnormal Child Psychology Sperling, G., Reeves, A., 1980. Measuring the reaction time of a shift of visual atten-
31, 105–120. tion. In: Nickerson, R.S. (Ed.), Attention and Performance VIII. Erlbaum, Hillsdale,
Schwarz, W., 2001. The ex-Wald distribution as a descriptive model of response NJ, pp. 347–360.
times. Behavior Research Methods Computation 33, 457–469. Stanford, C., Tannock, R., 2012. Behavioral Neuroscience of Attention Deficit Hyper-
Sergeant, J., Oosterlaan, J., van der Meere, J., 1999. Information processing and activity Disorder and its Treatment. Springer.
energetic factors in attention-deficit/hyperactivity disorder. In: Handbook of Stins, J.F., Tollenaar, M.S., Slaats-Willemse, D.I., Buitelaar, J.K., Swaab-Barneveld, H.,
Disruptive Behavior Disorders, pp. 75–104. Verhulst, F.C., Polderman, T.C., Boomsma, D.I., 2005. Sustained attention and
Sergeant, J.A., 2000. The cognitive-energetic model: an empirical approach executive functioning performance in attention-deficit/hyperactivity disorder.
to attention-deficit hyperactivity disorder. Neuroscience and Biobehavioral Neuropsychological, Development, and Cognition. Section C, Child Neuropsy-
Reviews 24, 7–12. chology 11, 285–294.
Sergeant, J.A., 2005. Modeling attention-deficit/hyperactivity disorder: a crit- Strelnikov, K., 2010. Neuroimaging and neuroenergetics: brain activations as
ical appraisal of the cognitive-energetic model. Biological Psychiatry 57, information-driven reorganization of energy flows. Brain and Cognition 72,
1248–1255. 449–456.
Sergeant, J.A., Geurts, H., Huijbregts, S., Scheres, A., Oosterlaan, J., 2003. The top Subbarao, K.V., Hertz, L., 1990. Effect of adrenergic agonists on glycogenolysis in
and the bottom of ADHD: a neuropsychological perspective. Neuroscience and primary cultures of astrocytes. Brain Research 536, 220–226.
Biobehavioral Reviews 27, 583–592. Suzuki, A., Stern, S.A., Bozdagi, O., Huntley, G.W., Walker, R.H., Magistretti, P.J.,
Sergeant, J.A., Geurts, H., Oosterlaan, J., 2002. How specific is a deficit of execu- Alberini, C.M., 2011. Astrocyte-neuron lactate transport is required for long-
tive functioning for attention-deficit/hyperactivity disorder? Behavioural Brain term memory formation. Cell 144, 810–823.
Research 130, 3–28. Swanson, J., Baler, R.D., Volkow, N.D., 2010. Understanding the effects of stimu-
Sergeant, J.A., Piek, J.P., Oosterlaan, J., 2006. ADHD and DCD: a relationship in need lant medications on cognition in individuals with attention-deficit hyperactivity
of research. Human Movement Science 25, 76–89. disorder: a decade of progress. Neuropsychopharmcology 36, 207–226.
Sergeant, J.A., Scholten, G., 1985. On data limitations in hyperactivity. Journal of Swanson, J., Gupta, S., Guinta, D., Flynn, D., Agler, D., Lerner, M., Williams, L., Shoul-
Child Psychology and Psychiatry, and Allied Disciplines 26, 111–124. son, I., Wigal, S., 1999. Acute tolerance to methylphenidate in the treatment of
Sergeant, J.A., van der Meere, J., 1988. What happens after a hyperactive child com- attention deficit hyperactivity disorder in children. Clinical Pharmacology and
mits an error? Psychiatry Research 24, 157–164. Therapeutics 66, 295–305.
Shallice, T., Marzocchi, G.M., Coser, S., Del Savio, M., Meuter, R.F., Rumiati, R.I., 2002. Sykes, D.H., Douglas, V.I., Morgenstern, G., 1973. Sustained attention in hyperactive
Executive function profile of children with attention deficit hyperactivity disor- children. Journal of Child Psychology and Psychiatry, and Allied Disciplines 14,
der. Developmental Neuropsychology 21, 43–71. 213–220.
Sherwood, C.C., Stimpson, C.D., Raghanti, M.A., Wildman, D.E., Uddin, M., Gross- Tafazoli, S., O‘Neill, J., Bejjani, A., Ly, R., Salamon, N., McCracken, J.T., Alger, J.R.,
man, L.I., Goodman, M., Redmond, J.C., Bonar, C.J., Erwin, J.M., 2006. Evolution Levitt, J.G., 2012. 1 H MRSI of middle frontal gyrus in pediatric ADHD. Journal
of increased glia-neuron ratios in the human frontal cortex. Proceedings of the of Psychiatric Research 47, 505–512.
National Academy of Sciences 103, 13606–13611. Takeuchi, T., Misaki, A., Liang, S.B., Tachibana, A., Hayashi, N., Sonobe, H., Ohtsuki,
Shiels, K., Hawk Jr., L.W., 2010. Self-regulation in ADHD: the role of error processing. Y., 2000. Expression of T-Cadherin (CDH13, H-Cadherin) in human brain and
Clinical Psychology Review 30, 951. its characteristics as a negative growth regulator of epidermal growth gactor in
Shiels, K., Hawk Jr., L.W., Reynolds, B., Mazzullo, R.J., Rhodes, J.D., Pelham Jr., W.E., neuroblastoma cells. Journal of Neurochemistry 74, 1489–1497.
Waxmonsky, J.G., Gangloff, B.P., 2009. Effects of methylphenidate on discounting Tamm, L., Narad, M.E., Antonini, T.N., O‘Brien, K.M., Hawk, L.W., Epstein, J.N., 2012.
of delayed rewards in attention deficit/hyperactivity disorder. Experimental and Reaction time variability in ADHD: a review. Neurotherapeutics 9, 500–508.
Clinical Psychopharmacology 17, 291–301. Taylor, E., 2010. Comorbidity in neurodevelopmental disorders: the case of
Shulman, G.L., Remington, R.W., Mclean, J.P., 1979. Moving attention through visual attention-deficit-hyperactivity disorder. In: Bax, M., Gillberg, C. (Eds.), Comor-
space. Journal of Experimental Psychology. Human Perception and Performance bidities in Developmental Disorders. Mac Keith Press, London, pp. 60–74.
5, 522–526. Teicher, M.H., Lowen, S.B., Polcari, A., Foley, M., McGreenery, C.E., 2004. Novel
Shulman, R.G., Rothman, D.L., Behar, K.L., Hyder, F., 2004. Energetic basis of strategy for the analysis of CPT data provides new insight into the effects of
brain activity: implications for neuroimaging. Trends in Neurosciences 27, methylphenidate on attentional states in children with ADHD. Journal of Child
489–495. and Adolescent Psychopharmacology 14, 219–232.
Simchon, Y., Weizman, A., Rehavi, M., 2010. The effect of chronic methylphenidate Theis, M., Giaume, C., 2012. Connexin-based intercellular communication and astro-
administration on presynaptic dopaminergic parameters in a rat model for cyte heterogeneity. Brain Research 1487, 88–98.
ADHD. European Neuropsychopharmacology: the Journal of the European Col- Tiffin-Richards, M., Hasselhorn, M., Richards, M.L., Banaschewski, T., Rothen-
lege of Neuropsychopharmacology 20, 714–720. berger, A., 2004. Time reproduction in finger tapping tasks by children
Smith, A., Taylor, E., Lidzba, K., Rubia, K., 2003. A right hemispheric frontocerebellar with attention-deficit hyperactivity disorder and/or dyslexia. Dyslexia 10,
network for time discrimination of several hundreds of milliseconds. NeuroIm- 299–315.
age 20, 344–350. Todd, R.D., Botteron, K.N., 2001. Is attention-deficit/hyperactivity disorder an energy
Smith, A.J., Blumenfeld, H., Behar, K.L., Rothman, D.L., Shulman, R.G., Hyder, F., 2002. deficiency syndrome? Biological Psychiatry 50, 151–158.
Cerebral energetics and spiking frequency: the neurophysiological basis of fMRI. Togashi, H., Sakisaka, T., Takai, Y., 2009. Cell adhesion molecules in the central ner-
Proceedings of the National Academy of Sciences 99, 10765–10770. vous system. Cell Adhesion & Migration 3, 29–35.
Sonuga-Barke, E., Bitsakou, P., Thompson, M., 2010. Beyond the dual pathway Toplak, M.E., Connors, L., Shuster, J., Knezevic, B., Parks, S., 2008. Review of
model: evidence for the dissociation of timing, inhibitory, and delay-related cognitive, cognitive-behavioral, and neural-based interventions for attention-
impairments in attention-deficit/hyperactivity disorder. Journal of the American deficit/hyperactivity disorder (ADHD). Clinical Psychology Review 28,
Academy of Child and Adolescent Psychiatry 49, 345–355. 801–823.
Sonuga-Barke, E.J., 2002. Psychological heterogeneity in AD/HD – a dual path- Toplak, M.E., Pitch, A., Flora, D.B., Iwenofu, L., Ghelani, K., Jain, U., Tannock, R., 2009.
way model of behaviour and cognition. Behavioural Brain Research 130, The unity and diversity of inattention and hyperactivity/impulsivity in ADHD:
29–36. evidence for a general factor with separable dimensions. Journal of Abnormal
Sonuga-Barke, E.J., 2003. The dual pathway model of AD/HD: an elaboration of Child Psychology 37, 1137–1150.
neuro-developmental characteristics. Neuroscience and Biobehavioral Reviews Toplak, M.E., Tannock, R., 2005. Time perception: modality and duration effects
27, 593–604. in attention-deficit/hyperactivity disorder (ADHD). Journal of Abnormal Child
Sonuga-Barke, E.J., 2005. Causal models of attention-deficit/hyperactivity disorder: Psychology 33, 639–654.
from common simple deficits to multiple developmental pathways. Biological Tripp, G., Wickens, J., 2008. Research review: dopamine transfer deficit: a neurobi-
Psychiatry 57, 1231–1238. ological theory of altered reinforcement mechanisms in ADHD. Journal of Child
Sonuga-Barke, E.J.S., Brandeis, D., Cortese, S., Daley, D., Ferrin, M., Holtmann, M., Psychology and Psychiatry 49, 691–704.
Stevenson, J., Danckaerts, M., van der Oord, S., Döpfner, M., Dittmann, R.W., Tsujino, N., Sakurai, T., 2009. Orexin/hypocretin: a neuropeptide at the interface of
Siminoff, E., Zuddas, A., Banaschewski, T., Buitelaar, J., Coghill, D., Hollis, C., sleep, energy homeostasis, and reward system. Pharmacological Reviews 61,
Konofal, E., Lecendreux, M., Wong, I.C.K., Sergeant, J., 2013. Nonpharmacological 162–176.
interventions for ADHD: systematic review and meta-analyses of randomized Turgay, A., Goodman, D.W., Asherson, P., Lasser, R.A., Babcock, T.F., Pucci, M.L.,
controlled trials of dietary and psychological treatments. The American Journal Barkley, R., 2012. Lifespan persistence of ADHD: the life transition model and its
of Psychiatry 170, 275–289. application. The Journal of Clinical Psychiatry 73, 192–201.
Sonuga-Barke, E.J.S., Sergeant, J.A., Nigg, J.T., Willcutt, E., 2008. Executive dysfunc- Uddin, L.Q., Kelly, A., Biswal, B.B., Margulies, D.S., Shehzad, Z., Shaw, D., Ghaffari, M.,
tion and delay aversion in attention deficit hyperactivity disorder: nosologic Rotrosen, J., Adler, L.A., Castellanos, F.X., 2008. Network homogeneity reveals
and diagnostic implications. Child and Adolescent Psychiatric Clinics of North decreased integrity of default-mode network in ADHD. Journal of Neuroscience
America 17, 367–384. Methods 169, 249–254.
Sorg, O., Magistretti, P.J., 1991. Characterization of the glycogenolysis elicited by Uttal, W.R., 2012. Reliability in Cognitive Neuroscience: A Meta-Meta Analysis.
vasoactive intestinal peptide, noradrenaline and adenosine in primary cultures Vaidya, C.J., Bunge, S.A., Dudukovic, N.M., Zalecki, C.A., Elliott, G.R., Gabrieli, J.D.E.,
of mouse cerebral cortical astrocytes. Brain Research 563, 227–233. 2005. Altered neural substrates of cognitive control in childhood ADHD: evi-
Spearman, C., 1904. General Intelligence, objectively determined and measured. The dence from functional magnetic resonance imaging. The American Journal of
American Journal of Psychology 15, 201–292. Psychiatry 162, 1605–1613.
Spencer, T.J., Biederman, J., Mick, E., 2007. Attention-deficit/hyperactivity disorder: Van der Meere, J.J., 2002. The role of attention. In: Sandberg, S. (Ed.), Hyperactivity
diagnosis, lifespan, comorbidities, and neurobiology. Ambulatory Pediatrics 7, and Attention Disorders of Childhood. Cambriudge Univ Press, Cambridge, pp.
73–81. 162–213.
P.R. Killeen et al. / Neuroscience and Biobehavioral Reviews 37 (2013) 625–657 657

Van Meel, C.S., Heslenfeld, D.J., Oosterlaan, J., Sergeant, J.A., 2007. Adaptive con- Willcutt, E., Sonuga-Barke, E., Nigg, J., Sergeant, J., 2008. Recent developments in neu-
trol deficits in attention-deficit/hyperactivity disorder (ADHD). The role of error ropsychological models of childhood psychiatric disorders. In: Banaschewski, T.,
processing. Psychiatry Research 151, 211–220. Rohde, L.A. (Eds.), Biological Child Psychiatry. Recent Trends and Developments.
Van Meel, C.S., Oosterlaan, J., Heslenfeld, D.J., Sergeant, J.A., 2005. Motivational Adv Biol Psychiatry. Karger, Basel, pp. 195–226.
effects on motor timing in attention-deficit/hyperactivity disorder. Journal of Willcutt, E.G., Betjemann, R.S., McGrath, L.M., Chhabildas, N.A., Olson, R.K., DeFries,
the American Academy of Child and Adolescent Psychiatry 44, 451–460. J.C., Pennington, B.F., 2010a. Etiology and neuropsychology of comorbidity
van Mourik, R., Oosterlaan, J., Heslenfeld, D.J., Konig, C.E., Sergeant, J.A., 2007. When between RD and ADHD: the case for multiple-deficit models. Cortex 46,
distraction is not distracting: a behavioral and ERP study on distraction in ADHD. 1345–1361.
Clinical Neurophysiology: Official Journal of the International Federation of Clin- Willcutt, E.G., Doyle, A.E., Nigg, J.T., Faraone, S.V., Pennington, B.F., 2005. Validity
ical Neurophysiology 118, 1855–1865. of the executive function theory of attention-deficit/hyperactivity disorder: a
van Mourik, R., Oosterlaan, J., Sergeant, J.A., 2005. The Stroop revisited: a meta- meta-analytic review. Biological Psychiatry 57, 1336–1346.
analysis of interference control in AD/HD. Journal of Child Psychology and Willcutt, E.G., Pennington, B.F., Duncan, L., Smith, S.D., Keenan, J.M., Wadsworth, S.,
Psychiatry, and Allied Disciplines 46, 150–165. DeFries, J.C., Olson, R.K., 2010b. Understanding the complex etiologies of devel-
Van Zandt, T., Ratcliff, R., 1995. Statistical mimicking of reaction time data: single- opmental disorders: behavioral and molecular genetic approaches. Journal of
process models, parameter variability, and mixtures. Psychonomic Bulletin & Developmental and Behavioral Pediatrics: JDBP 31, 533–544.
Review 2, 20–54. Williams, J., Taylor, E., 2006. The evolution of hyperactivity, impulsivity and cogni-
Vaughn, A.J., Epstein, J.N., Rausch, J., Altaye, M., Langberg, J., Newcorn, J.H., Hinshaw, tive diversity. Journal of the Royal Society: Interface 3, 399–413.
S.P., Hechtman, L., Arnold, L.E., Swanson, J.M., 2011. Relation between outcomes Williams, L.M., Hermens, D.F., Thein, T., Clark, C.R., Cooper, N.J., Clarke,
on a continuous performance test and ADHD symptoms over time. Journal of S.D., Lamb, C., Gordon, E., Kohn, M.R., 2010. Using brain-based cognitive
Abnormal Child Psychology 39, 853–864. measures to support clinical decisions in ADHD. Pediatric Neurology 42,
Verbruggen, F., Logan, G.D., 2008. Response inhibition in the stop-signal paradigm. 118–126.
Trends in Cognitive Sciences 12, 418–424. Winstanley, C.A., Eagle, D.M., Robbins, T.W., 2006. Behavioral models of impulsivity
Vles, J., Feron, F., Hendriksen, J., Jolles, J., van Kroonenburgh, M., Weber, W., in relation to ADHD: translation between clinical and preclinical studies. Clinical
2003. Methylphenidate down-regulates the dopamine receptor and transporter Psychology Review 26, 379–395.
system in children with attention deficit hyperkinetic disorder (ADHD). Neuro- Wyss, M.T., Jolivet, R., Buck, A., Magistretti, P.J., Weber, B., 2011. In vivo evidence
pediatrics 34, 77–80. for lactate as a neuronal energy source. The Journal of Neuroscience: the Official
Wagenmakers, E.J., Brown, S., 2007. On the linear relation between the mean and the Journal of the Society for Neuroscience 31, 7477–7485.
standard deviation of a response time distribution. Psychological Review 114, Yang, P., Wu, M.T., Dung, S.S., Ko, C.W., 2010. Short-TE proton magnetic resonance
830–841. spectroscopy investigation in adolescents with attention-deficit hyperactivity
Wagenmakers, E.J., Van Der Maas, H.L.J., Grasman, R.P.P.P., 2007. An EZ-diffusion disorder. Psychiatry Research. Neuroimaging 181, 199–203.
model for response time and accuracy. Psychonomic Bulletin & Review 14, 3–22. Yanofski, J., 2011. The dopamine dilemma – Part II: could stimulants cause toler-
Walhovd, K.B., Fjell, A.M., 2007. White matter volume predicts reaction time insta- ance, dependence, and paradoxical decompensation? Innovations in Clinical
bility. Neuropsychologia 45, 2277–2284. Neuroscience 8, 47–53.
Warm, J.S., Parasuraman, R., Matthews, G., 2008. Vigilance requires hard mental Yu-Feng, Z., Yong, H., Chao-Zhe, Z., Qing-Jiu, C., Man-Qiu, S., Meng, L., Li-Xia, T., Tian-
work and is stressful. Human Factors 50, 433–441. Zi, J., Yu-Feng, W., 2007. Altered baseline brain activity in children with ADHD
Weiss, G., Hechtman, L., Milroy, T., Perlman, T., 1985. Psychiatric status of hyper- revealed by resting-state functional MRI. Brain & Development 29, 83–91.
actives as adults: a controlled prospective 15-year follow-up of 63 hyperactive Zametkin, A., Nordahl, T., Gross, M., King, C., Semple, W., Rumsey, J., Hamburger,
children. Journal of American Academy of Child & Adolescent Psychiatry 24, S., Cohen, R., 1990. Cerebral glucose metabolism in adults with hyperactivity of
211–220. childhood onset. The New England Journal of Medicine 323, 1361–1366.
Weiss, G., Hechtman, L.T., 1993. Hyperactive Children Grown Up: ADHD in Children, Zhang, D., Raichle, M.E., 2010. Disease and the brain’s dark energy. Nature Review.
Adolescents, and Adults. Guilford Press. Neurology 6, 15–28.
Weissman, D.H., Roberts, K.C., Visscher, K.M., Woldorff, M.G., 2006. The neural bases Zhu, X.H., Qiao, H., Du, F., Xiong, Q., Liu, X., Zhang, X., Ugurbil, K., Chen, W., 2012.
of momentary lapses in attention. Nature Neuroscience 9, 971–978. Quantitative imaging of energy expenditure in human brain. Neuroimage 60,
Wiguna, T., Guerrero, A.P.S., Wibisono, S., Sastroasmoro, S., 2012. Effect of 12- 2107–2117.
week administration of 20-mg long-acting methylphenidate on Glu/cr, NAA/Cr, Zolkowska, D., Jain, R., Rothman, R.B., Partilla, J.S., Roth, B.L., Setola, V., Prisinzano,
Cho/Cr, and mI/Cr ratios in the prefrontal cortices of school-age children in T.E., Baumann, M.H., 2009. Evidence for the involvement of dopamine trans-
indonesia: a study using 1 h magnetic resonance spectroscopy (MRS). Clinical porters in behavioral stimulant effects of modafinil. The Journal of Pharmacology
Neuropharmacology 35, 81–85. and Experimental Therapeutics 329, 738–746.

You might also like