Articles

https://doi.org/10.1038/s41593-019-0383-6

Causal contributions of parietal cortex to

perceptual decision-making during stimulus
categorization
Lin Zhong1,2,3, Yuan Zhang   1,2,3, Chunyu A. Duan1, Ji Deng1,2, Jingwei Pan1 and Ning-long Xu   1,2*

The posterior parietal cortex (PPC) has been implicated in perceptual decision-making and categorization, but whether its
activity plays a causal role remains controversial. Here we examined the population dynamics of PPC activity during an audi-
tory-guided decision task in mice. We found that silencing of PPC activity impaired several aspects of decision-making. First,
categorization of new, but not well-learned, stimuli was impaired. Second, re-categorization of previously experienced stimuli
based on newly learned categories was also impaired. Third, the bias on behavioral choices created by preceding trials signifi-
cantly increased. In vivo two-photon imaging of PPC activity during stimulus categorization revealed differential dynamics in
representations of new stimuli and learned categories, consistent with rapid incorporation of new sensory information during
categorization. At the circuit level, inactivation of PPC axonal projections to the auditory cortex also significantly reduced cat-
egorization performance. Thus, PPC circuits play a causal role in decision-making during stimulus categorization.

P
erceptual decision is a simple form of decision-making and has
been widely studied in a variety of animal species to under-
stand the neural basis of decision-making. The mammalian
posterior parietal cortex (PPC) is one of the brain areas that shows
strong decision-related activity in both primates1–6 and rodents7–11.
Computational theories have been developed based on PPC activity
to account for circuit mechanisms of decision-making2,12–14. However,
recent studies have reported somewhat inconsistent results regard-
ing the causal role of PPC, questioning its functional significance in
decision processes. For instance, inactivation of the PPC in macaques
had no major effect on choice behavior in a visual–motion percep-
tual decision task15, nor in a self-motion perceptual decision task16. In
rodents, silencing of PPC activity had no notable effects on behavioral
performance in either auditory11,17 or tactile decision tasks18. A com-
mon view regarding the lack of effect on choice behavior following
PPC inactivation is that decision-related activity in PPC may reflect
only a secondary process not directly involved in computations for
decision-making15,19. An alternative possibility, however, is that PPC
activity may be important in guiding decisions on unfamiliar or
unknown sensory stimuli based on previous knowledge, which can
be a crucial function in the real world, but this theory has not been
effectively evaluated in previously reported behavioral tasks using the
same set of well-trained stimuli for both training and testing.
To test this idea, we examined the functional significance of PPC
in auditory-guided perceptual decisions on new sensory stimuli not
previously used for training. Using pharmacological and optoge-
netic perturbations, we show that PPC activity causally contributes
to categorical decisions on newly introduced, but not well-experi-
enced, stimuli. Silencing of PPC activity also impaired the animals’
capability of reassigning previously learned stimuli into newly
learned categories, further supporting the function of the PPC in
the learning process of stimulus categorization. In addition, silenc-
ing of PPC activity increased the biasing effect created by preceding
trials, suggesting an additional contribution of the PPC to decision-
making by counterbalancing short-history bias. In vivo two-photon
imaging of PPC population activity during stimulus categoriza-
tion revealed both stable representations for learned categories and
evolving dynamics in response to new stimuli, consistent with a
learning process of categorizing new stimuli into existing associa-
tions10,20. Projection-specific chemogenetic perturbation of the PPC
to auditory cortex connections also significantly reduced catego-
rization performance, implying a circuit-level mechanism. These
findings reveal a causal role of PPC circuits in the dynamic learning
process of making perceptual decisions for stimulus categorization.
Results
Mice can make categorical decisions on auditory stimuli. We
trained head-fixed mice to perform a two-alternative, forced-choice
auditory discrimination task categorizing tone stimuli as either high
or low frequency using directional licking to read out the choices.
During the task, mice were required to lick the left or right lickport
following a low- or high-frequency tone, respectively, to receive a
water reward. We first trained mice to learn associations for two
training tones (8 and 32 kHz; Fig. 1a,b). After mice had reached
criterion performance (>85% correct), new tones at intermediate
frequencies (six tones separated by 0.2 octave) were introduced in
~30% of trials, randomly interleaved with training trials to test the
animals’ ability in stimulus categorization. Mice showed catego-
rization of the test stimuli in the first session with the introduc-
tion of new tones, as indicated by a steep slope in psychometric
function (Fig. 1c,d). The animals’ subjective category boundaries
(peak of slope function) were distributed in close proximity to an
experimentally defined category boundary (16 kHz; Fig. 1e; see also
Supplementary Fig. 1e), suggesting that their categorical decisions
were based on learned ‘categories’.
PPC activity is required for categorization of new auditory stimuli.
Next, we sought to determine whether PPC activity is important

1
Institute of Neuroscience, State Key Laboratory of Neuroscience, CAS Center for Excellence in Brain Science and Intelligence Technology, Shanghai
Institutes for Biological Sciences, Chinese Academy of Sciences, Shanghai, China. 2University of Chinese Academy of Sciences, Beijing, China.
3
These authors contributed equally to this work: Lin Zhong, Yuan Zhang. *e-mail: xunl@ion.ac.cn

Nature Neuroscience | VOL 22 | JUNE 2019 | 963–973 | www.nature.com/natureneuroscience 963

The Nature trademark is a registered trademark of Springer Nature Limited.
Articles
a Low High
Left choice Right choice
8 11.3 13 14.9 17.1 19.7 22.6
Training
Testing tones
tones
c Training trials d e
100 Testing trials 2
Reported high (%)
Slope
50
1
0 Psychometric curve
0
8 16 32
(0) (1) (2)
Frequency (kHz (octave))
Fig. 1 | Perceptual categorization of auditory stimulus in head-fixed mice. a, Schematic showing configuration of the behavioral experiments. b, Temporal
structure of behavioral trials. c, Psychometric function from a sample behavioral session with one mouse, when the test stimuli were first introduced.
Error bars, mean ± 95% confidence interval across trials. d, Slopes of the psychometric function shown in c, with the location of peak slope indicating the
categorization boundary. e, Histogram showing distribution of category boundaries (16.5 ± 0.3 kHz, mean ± s.e.m., n = 25 mice) in the first session with the
introduction of test stimuli.
in the decision process of categorizing new stimuli. We reversibly
silenced the PPC by bilateral micro-injection of muscimol, a selec-
tive gamma-aminobutyric acid (GABAA) receptor agonist, in the first
session (day 1) when the new test stimuli were introduced (Fig. 2a,c).
In contrast to previous reports in rodents11,17,19,21 and primates15,16, we
found that the categorical decisions made by the mice were mark-
edly impaired compared to control mice micro-injected with saline.
Impairment was measured as a significant reduction in the slope of
psychometric functions (saline, 1.46 ± 0.16; muscimol, 0.97 ± 0.09;
P = 0.019, Wilcoxon rank-sum test; Fig. 2d,f and Supplementary
Fig. 2), but not in the category boundaries (Fig. 2g).
Since muscimol inactivation does not have the temporal resolu-
tion to distinguish whether it was during training trials or testing
trials that PPC activity was required for stimulus categorization, we
employed optogenetics to inactivate the PPC in a more temporally
controlled fashion18. An adeno-associated virus (AAV) encoding
Cre-inducible channelrhodopsin 2 (ChR2)22 was injected bilater-
ally in the PPC of Vgat-Cre mice23 to express ChR2 in GABAergic-
inhibitory neurons. Glass optical windows were implanted above the
PPC of both hemispheres following virus injection. Collimated blue
laser was used to illuminate the glass window to achieve photoinhi-
bition of PPC (Fig. 2b,c, Supplementary Fig. 3 and Supplementary
Video 1)18. Another group of Vgat-Cre mice injected with AAV-
Flex-mCherry with windows implanted at the same locations was
used as the control group. To control for the potential distraction of
the blue light, two blue light-emitting diodes were placed in front
of both eyes of the mouse to deliver masking light with the same
temporal patterns as the photostimulation in all trials in both the
ChR2 and control groups. During task performance, photoinhibi-
tion was applied in all testing trials but only in 10% of training trials
for comparison, leaving the majority (90%) of training trials unaf-
fected. We found that this manipulation also significantly reduced
categorization performance for the new tone stimuli (control slope,
2.43 ± 0.46; photoinhibition slope, 1.17 ± 0.37; P = 0.02, Wilcoxon
rank-sum test; Fig. 2e,f and Supplementary Fig. 4), indicating that
PPC activity during testing trials, rather than during training tri-
964
The Nature trademark is a registered trademark of Springer Nature Limited.
NATure NeuroSCIenCe
b
1s
Stimulus
Delay
Response window
32 (kHz) Response time
Training
Water valve
tones
Defined
Maximum slope
boundary
Animal number
Category 5
boundary
8 16 32 8 16 32
(0) (1) (2) (0) (1) (2)
Frequency (kHz (octave)) Frequency (kHz (octave))
als, is required for categorization of new test stimuli. Moreover, we
found that when PPC photoinhibition was applied in 50% of the
testing trials (randomly interleaved), categorization performance
was not significantly affected (Supplementary Fig. 5), suggesting
that PPC activity in part of the testing trials on day 1 was sufficient
for categorizion of new tone stimuli, consistent with a learning-
dependent process.
It should be noted that neither muscimol silencing (Fig. 2h) nor
photoinhibition (Fig. 2i) influenced performance on the training
stimuli, nor was the miss rate affected (Fig. 2j,k and Supplementary
Fig. 6), suggesting that the established associations between learned
categories and motor responses are stable and do not require PPC
activity. It is thus likely that when the testing stimuli are well learned
and are incorporated into established associations, choices on test-
ing stimuli may no longer require PPC activity. This may explain the
previously reported lack of behavioral effects following PPC inacti-
vation. To test this possibility, we performed bilateral PPC inacti-
vation in a separate cohort of mice after the test stimuli had been
well experienced following continued training for two additional
sessions (Fig. 3a). We found that under this condition, inactivation
of PPC did not significantly influence categorization performance
(Fig. 3b–e and Supplementary Fig. 7), in agreement with the results
from previous studies in rodents and non-human primates assess-
ing the effect of PPC inactivation on choice behavior, where the
stimuli were probably also well experienced11,15–18. Taken together,
these data indicate that the PPC is causally involved in categoriza-
tion of new sensory stimuli, but not after the stimuli have already
been incorporated into existing categories.
PPC response properties for learned categories and new sen-
sory stimuli. The above results suggest that categorization of new
sensory stimuli is a dynamic process that may involve learning24–26
and that the PPC plays a critical role in this learning process. We
thus examined how PPC activity evolves with the learning process
of stimulus categorization when new sensory information is inte-
grated into learned categories. Using chronic in vivo two-photon
Nature Neuroscience | VOL 22 | JUNE 2019 | 963–973 | www.nature.com/natureneuroscience
NATure NeuroSCIenCe Articles
a b c Training stage Testing stage
C57/BL/6J Vgat-Cre
Col
r lim
ato ator
llim

8 kHz , 32 kHz
Co

Musci./saline 8 11.3 13 14.9 17.1 19.7 22.6 32 (kHz)

AAV-flex-ChR2
~10 days Day 1 Day 2
C57
C57
Opto Vgat, ChR2
Musci. 1 mm Light 1 mm Vgat, ChR2
Saline mask

d e f g
100
4 P = 0.022 1
P = 0.68 P = 0.73
Reported high (%)

Categorical boundary
Psychometric slope
3 P = 0.019
50

(octave)
2 0
Saline Mask
Musci. Mask + opto
0 1

2
Slope

0 –1

to

.
ci

to
0

as

ci
lin

as
lin
op
us

op
us
Sa

Sa

M
M

M
8 16 32 8 16 32

+
k

k
as

as
Frequency (kHz)

M
h P = 0.27 i P = 1.0 j P = 0.009 k
P = 0.10
100 100 100 100
of training trials (%)
of training trials (%)

Miss rate (%)

Miss rate (%)

Correct rate
Correct rate

50 50 50 50

0 0 0 0
.

o.

o
ci

to
lin

pt

ci
pt

as
lin
us

op
O

us
sa

-o

Sa

M
m

M
on

+
2
1

k
N
ay
ay

as
D
D

M
Fig. 2 | PPC activity is necessary for categorical decision-making on new sensory stimuli. a, Schematic and histology image showing the locations of
injected muscimol (musci.) or saline in the PPC. Red areas in the histology image are fluorescence from co-injected red beads used to mark the injection
sites. b, Schematic and histology image showing the locations of photostimulation and viral expression of ChR2-mCherry in GABAergic neurons in
bilateral PPC of Vgat-Cre mice. Red fluorescence represents mCherry expression. Opto, optogenics. c, Schematic showing experimental schedules for PPC
inactivation and control experiments on the first day following introducion of test stimuli. Symbols represent the four types of experiment as indicated
in a and b. Frequencies of tone stimulation at different stages are indicated. d, Psychometric functions (top) and corresponding slopes (bottom) from
sample muscimol injection and control sessions as indicated in a and c. Error bars, 95% confidence interval. See Supplementary Fig. 2 for individual cases.
e, As in d, for photoinhibition and control sessions as indicated in b and c. See Supplementary Fig. 4 for individual cases. f, Summary of group data for
peak psychometric slopes from experiments indicated in c. Muscimol, n = 13 mice, versus saline, n = 14 mice, P = 0.019, Wilcoxon rank-sum test, two-
sided; photoinhibition, n = 7 mice, versus mask only, n = 6 mice, P = 0.022, Wilcoxon rank-sum test, two-sided; errror bars, mean ± s.e.m. g, Summary
of category boundaries for experiments indicated in c (muscimol versus saline, P = 0.68, Wilcoxon rank-sum test, two-sided; photoinhibition with mask
versus mask only, P = 0.73, Wilcoxon rank-sum test, two-sided; error bars, mean ± s.e.m. h, Summary of performance (percentage correct) for training
stimuli, compared between muscimol injection and control, n = 13 mice; error bars, mean ± s.e.m.; paired t-test, two-sided. i, As in h for photoinhibition
versus control experiments, n = 7 mice; error bars, mean ± s.e.m.; paired t-test, two-sided. j,k, Miss rate for muscimol silencing experiments (j) and
photoinhibition experiments (k); n = 13 mice (muscimol), n = 14 mice (saline), n = 7 mice (mask + opto), n = 6 mice (mask); error bars, mean ± s.e.m.;
Wilcoxon rank-sum test, two-sided.

calcium imaging, we tracked PPC population activity across differ- categorizing new sensory stimuli. First, we found that the num-
ent sessions/days following the introduction of new test stimuli. The ber of neurons selective to training stimuli was almost unchanged
genetically encoded calcium indicator GCaMP6s27 was expressed across two consecutive days (21% on day 1 versus 20% on day 2;
unilaterally in the PPC using AAV vectors to indicate intracellular Supplementary Fig. 10e). Second, using receiver operator charac-
calcium dynamics, followed by implantation of a chronic imaging teristic (ROC) analysis, we computed the index of selectivity30 in
window28. Layer 2/3 neuron populations were repeatedly imaged each neuron for high versus low training frequencies for all neu-
during task performance starting from the session when the test rons (n = 757 neurons) over the trial period (Fig. 4c,d). We sorted
stimuli were first introduced (Fig. 4a and Supplementary Fig. 8). all neurons based on their selectivity index values on day 1 and used
Consistent with previous studies8,10,29,21, we found that PPC neurons the same order to sort values on day 2. We found that, across the
showed complex response selectivity and heterogeneous dynamics entire population and trial time period, neuronal selectivity for
(Fig. 4b, Supplementary Fig. 9 and Supplementary Table 1). high and low training frequencies was preserved as indicated by the
Previous studies reported strong decision-related activity in the resemblance of the selectivity index profiles between consecutive
PPC despite the absence of behavioral effects following PPC inac- days (Fig. 4d,e). Thus, neuronal selectivity for learned sensorimotor
tivation7,9,11,16,17,20. We first examined whether the representations associations was stable over the time course of learning to categorize
for well-trained stimuli would remain stable during the process of new stimuli. Third, we asked whether the population coding for the

Nature Neuroscience | VOL 22 | JUNE 2019 | 963–973 | www.nature.com/natureneuroscience 965

The Nature trademark is a registered trademark of Springer Nature Limited.
Articles
a when the same choices had been made20. Here we further exam-
Training stage
8 kHz, 32 kHz 8 11.3 13
~10 days Day 1
C57
Vgat, ChR2
b d
100
Psychometic slope
Reported high (%)
50
Day 1
Day 2
Day 3
0 (opto)
c e
Psychometic slope
3
Slope
0 1
8 16 32
Frequency (kHz)
Fig. 3 | PPC activity is not required for choice behavior at post-
categorization stage. a, Schematic showing experimental schedules for
PPC inactivation after 2 days of training with test stimuli. b,c, Psychometric
functions (b) and corresponding slopes (c) from three consecutive
sessions from an example mouse as indicated in a. Error bars, mean ± 95%
confidence interval across trials. d,e, Summary of grouped data for peak
psychometric slopes in experiments on PPC inactivation in the third session
using muscimol (d) or optogenetics (e); n = 6 mice (muscimol), n = 5 mice
(optogenetics); error bars, mean ± s.e.m., signed-rank test, two-sided. See
Supplementary Fig. 7 for further individual cases.
learned associations was also stable. We first examined whether the
animal’s choices on training trials in three consecutive days could
be decoded by the simultaneously imaged population activity only
from day 1, using a linear classifier31. Population activity in the
training trials on day 1 was used as a training set to predict the ani-
mals’ choices in the training trials on either day 1 or on days 2 and 3,
respectively (Fig. 4f). We found that decoding accuracy was mark-
edly higher than chance level starting from stimulus onset, and was
not significantly different between day 1 and days 2 and 3 (Fig. 4g),
indicating that population coding for the learned associations was
stable across the time course of learning. Stable population cod-
ing in training trials may reflect stable representations for learned
categories. We thus examined whether population activity in the
training trials could predict behavior in test trials over the learning
course. We trained the linear classifier with data from the training
trials on day 1 and examined decoding accuracy in predicting test
trials from three consecutive days. We found that prediction accu-
racy was also markedly higher than chance and was stable across
test days (Fig. 4h,i), suggesting that the representations for learned
categories in the PPC were stable over the learning process of cat-
egorizing new stimuli.
It has been proposed that PPC neurons incorporate new sen-
sory information into existing representations10,20 and that, during
learning, PPC populations distinguish novel and familiar cues even
966
The Nature trademark is a registered trademark of Springer Nature Limited.
NATure NeuroSCIenCe
Testing stage
ined whether PPC neurons would distinguish newly presented
14.9 17.1 19.7 22.6 32 (kHz) test stimuli versus training stimuli from the same choice category,
Day 2 Day 3 and whether such distinction would show dynamic changes over
the learning process of categorization. We found that some neu-
rons indeed showed differential responses to training and testing
stimuli even within the same category (Fig. 5a,b and Supplementary
Fig. 11). We computed the ROC-based index of selectivity distin-
P = 0.69 guishing training trials versus testing trials within the same category
P = 0.56 for individual neurons over the population, and we also examined
4
such selectivity changes across consecutive days. We found that
many neurons showed a distinct preference to test stimuli on day 1,
3
which was reduced over the following days (Fig. 5c,d). The propor-
2 tion of neurons showing significant preference to test stimuli were
reduced from 16% on day 1 to 8% on day 2. Such dynamic change
1
is consistent with the possibility that these neurons may have been
0
involved in the categoriziation of new sensory stimuli on day 1. In
neurons showing preferred responses to test stimuli on day 1, the
ROC values for distinguishing test and training stimuli of the same
category reduced over consecutive days (Fig. 5e). Taken together,
P = 0.31
these results show that over the learning process of categorization
4 P = 0.63
of new stimuli, PPC neurons showed both stable representation for
learned categories and evolving dynamics for new stimuli, with a
3
trend of converging representations for stimuli of the same category
2 over the time course of learning (Supplementary Fig. 12).
PPC activity counterbalances recent history bias. Decision-
making is often biased by recent behavioral history21,32–35, particu-
0
Day 1 Day 2 Day 3 larly when sensory evidence is weak34 or a delay is imposed between
stimulus and choice21,32. However, in a volatile environment, the
outcomes of events immediately preceding generally lack predic-
tive value for future stimuli, and decisions are guided rather by syn-
thesized previous knowledge from long-term past experiences and
current sensory evidence. Therefore, under certain circumstances
short-term history and long-term previous experience may have
opposing influences on decisions. PPC activity has been shown to
be related to both long-term sensory and behavioral history3,32 and
to short-term evidence accumulation8,15, and may therefore mediate
either short-term history bias or action against such bias. To distin-
guish these possibilities, we examined the effect of PPC activity on
recent history bias during categorization of new stimuli. We quanti-
fied the influence of the preceding trial on choices made by mice, by
comparing their performance on current test stimuli when the pre-
ceding training stimulus was of the same category as the current trial
versus when it was from a different category. An influence of pre-
ceding training trials would be expressed as decreased or increased
performance when the current test stimulus was of a different or
the same category as the preceding training stimulus, respectively.
We found that, under control conditions, preceding training trials
did not significantly bias performance in test trials on day 1 (Fig. 6a;
77.3 ± 4.6% for same category versus 74.3 ± 3.2% for a different
category, P = 0.32), nor after the test stimuli were well experienced
(Supplementary Fig. 13). However, when the PPC was inactivated
by either muscimol or optogenetics, we observed a significant
choice bias in test stimuli toward the category of preceding train-
ing stimuli (Fig. 6b,c; muscimol, same category 71.8 ± 3.4% versus
different category 63.5 ± 3.7%, P = 0.017; photoinhibition, same cat-
egory 70.3 ± 2.4% versus different category 55.9 ± 3.2%, P = 0.003).
Thus, whereas decisions were minimally influenced by preceding
events when the PPC was intact, its inactivation led to a significant
bias by short-term behavioral history (Fig. 6d). In contrast, when
newly introduced test stimuli became well experienced after contin-
ued training, PPC inactivation no longer resulted in a significantly
increased biasing effect (Supplementary Fig. 13e,f). These results
indicate that PPC activity additionally contributes to decisions on
new sensory stimuli by counterbalancing short-term history bias.
Nature Neuroscience | VOL 22 | JUNE 2019 | 963–973 | www.nature.com/natureneuroscience
NATure NeuroSCIenCe Articles
a c
16× NA
0.8 Depth: 0 µm Depth: 150 µm Day 1 Day 2

Training

Testing

1 mm 250 µm 100 µm
Testing
b 8 11.3 13 14.9 17.1 19.7 22.6 32 (kHz) 300

∆F/F (%)
Training
5

4
200 0

∆F/F (%)
Neurons

3
0
2

(∆F/F )
200%
0 6 0 6
1 Time from tone onset (s)
4s

d Day 1 Day 2 e
757
SI (0–1 s)

0.2
(sorted by day 1)
Neurons (n)

SI (day 2)
0

High
0.2 –0.2
r = 0.91
SI

Slope = 0.88
1 –0.2 –0.4
0 1 2 3 0 1 2 3 Low –0.4 –0.2 0 0.2
Time from tone onset (s) SI (day 1)

f Training trials g P = 0.91 h Testing trials i

P = 0.76
First
80 100 P = 0.31 80 First 100
lick
Prediction accuracy (%)

P = 0.69
Prediction accuracy (%)
Prediction accuracy (%)

Prediction accuracy (%)

lick

70 Day1 70
Day2
Day3 50 50
60 60

Shuffle
50 50
0 0
0 0.5 1 0 0.5 1
1

1
2

3
y

y
y

Time from tone onset (s) Time from tone onset (s)
Da

Da

Da

Da
Da

Da

Fig. 4 | PPC neurons show stable representation for learned categories. a, Schematic and images showing in vivo two-photon imaging from PPC
populations using GCaMP6s. The same imaging fields were repeatedly imaged over several days. b, Sample Ca2+ signal traces of PPC neurons showing
high signal/noise ratio with diverse selectivity and rich dynamics. c, Ca2+ signal of a sample PPC neuron imaged during behavioral sessions on consecutive
days. Top color raster of Ca2+ signals (% ∆F/F0) across trials. Blocks of trials presenting training and test stimuli of low- and high-frequency categories are
segregated. Trials within each block are aligned by stimulus onset time and sorted by time of first responding lick (vertical dashed lines, stimulus onset;
white dots, time of first responding lick). Bottom: traces of average Ca2+ signals from the blocks of trials in the top panel, with colors corresponding to tone
frequency. Shaded regions, mean ± s.e.m. d, Color raster plot showing index of selectivity (SI) between high- and low-frequency categories across trial time
for all neurons, as in across consecutive days. Neurons are sorted according to SI values on day 1. e, Scatter plot of mean SI values (n = 757 neurons) in the
1 s trial time window after stimulus onset, as in d. r, Pearson’s correlation coefficient. Slope, slope of the linear regression line (red). f, Decoding accuracy
across trial time for high and low training tones by population activity of simultaneously imaged neurons in each imaging field using a linear classifier.
The classifier was trained using neuronal activity in 80% of training trials on day 1 to decode stimulus categories on 3 consecutive days. Shaded regions,
mean ± s.e.m. g, Peak decoding accuracies in a 1 s time window following stimulus onset, as in f; n = 6 mice (day 1), n = 6 mice (day 2), n = 4 mice (day 3);
error bars, mean ± s.e.m., signed-rank test (day 1 versus day 2, two-sided), Wilcoxon rank-sum test (day 1 versus day 3, two-sided). h, Decoding accuracy of
a linear classifier discriminating high and low test trials. The classifier was trained using the population activity of training trials from day 1. Shaded regions,
mean ± s.e.m. i, Peak decoding accuracies in a 1 s time window following stimulus onset, as in h; n = 6 mice (day 1), n = 6 mice (day 2), n = 4 mice (day 3);
error bars, mean ± s.e.m., signed-rank test (day 1 versus day 2, two-sided), Wilcoxon rank-sum test (day 1 versus day 3, two-sided).

The contribution of the PPC to reduced short-term history categorization, or (2) a redundant expression of the same process.
bias could be either (1) an independent contribution to decision- To address this question, we simultaneously examined the con-
making in addition to its contribution to learning of stimulus tribution of the PPC to both categorization (Fig. 2) and reduced

Nature Neuroscience | VOL 22 | JUNE 2019 | 963–973 | www.nature.com/natureneuroscience 967

The Nature trademark is a registered trademark of Springer Nature Limited.
Articles NATure NeuroSCIenCe

a Day 1 Day 2 c Day 1 Day 2

Low
20
training

Sorted neurons
40

Category low
Low
testing
60
High
testing
80
200

∆F/F (%)
High 100
training 0 2 4 6 0 2 4 6

200 0
∆F/F (%)

20

Sorted neurons

Category high
40
0 6 0 6
Time from tone onset (s)
60
b
Day 1
1.0 80
Day 2
versus testing)
AUC (training

.75 Category low 100

0 2 4 6 0 2 4 6
Category high Time from tone onset (s)
.50
0 1 2 3 4 5 6 0.4 –0.4
Test SI Train
Time from tone onset (s)

d e P = 7.06 × 10–15
Day 1 Day 2
P = 3.27 × 10–15
200 NS 200
P < 0.05 0.60
150 150
AUC

P = 0.23
Cells

100 100 0.55

50 50

0.50
0 0
–0.5 0 0.5 –0.5 0 0.5
1

3
ay

ay

ay

Selectivity index Selectivity index

D

Fig. 5 | PPC representations show evolving dynamics over learning. a, Plot similar to that in Fig. 4c, showing Ca2+ signal of a sample PPC neuron with
dynamic changes in response in testing trials (vertical dashed lines, stimulus onset; white dots, time of first response lick). Shaded regions, mean ± s.e.m.
b, Area under ROC curves (AUC) distinguishing testing and training trials over trial period. c. Selectivity index (SI) between training and test stimuli over time
based on ROC analysis for all neurons in a sample imaging field. Positive SI values indicate preference for test stimuli and negative values for training stimuli.
d, Histogram of SI values (mean value in 1 s window after stimulus onset) for training versus test stimuli from the total of 757 neurons across 2 consecutive
days. Proportions of neurons showing significant preference (with AUC values above the 95% percentile of shuffled distribution) are color coded: green,
training stimuli; magenta, test stimuli. NS, not significant. e, Averaged AUC values over 3 consecutive days for neurons showing significant preference to test
stimuli on day 1, as in a and d; n = 57 data points (data for only 2 consecutive days were not included); paired t-test, two-sided; error bars, mean ± s.e.m.

recent history bias (Fig. 6a,c) by modeling the performance
of mice under control and silencing conditions using a series
of logistic regression models (generalized linear mixed model,
GLMM; Fig. 6e; also see Methods)32,35. Our models confirmed
the lack of recent history bias in control data, insofar as the
inclusion of previous trial choice as a regressor term did not
significantly improve the model prediction of the behavioral data
(Supplementary Fig. 14a). The models also confirmed the effect of
PPC inactivation on stimulus categorization, insofar as inclusion
of an interaction term between PPC silencing and sensory stimuli
significantly improved the model prediction on experimental data
showing reduced psychometric slopes (Supplementary Fig. 15).
Notably, when including an interaction term between previous
trial choice and the silencing condition as a regressor, the model
prediction was significantly better in comparison to the model
quantifying only the effect of silencing on psychometric slope
changes (Fig. 6f–h and Supplementary Fig. 14), indicating that the
effect of PPC silencing on short-term history bias is in addition to
its effect on categorization slopes. Thus, the PPC causally contrib-
utes to the decision process of categorizing new stimuli, by both
influencing psychometric slopes and counterbalancing short-term
history bias.

968 Nature Neuroscience | VOL 22 | JUNE 2019 | 963–973 | www.nature.com/natureneuroscience

The Nature trademark is a registered trademark of Springer Nature Limited.

NATure NeuroSCIenCe Articles
a b c No manipulation ChR2 + mask Saline
100 ChR2 + opto Muscimol
Opto-control ChR2 + opto
Saline Muscimol P = 0.005
Performance of Td (%)

40 * P = 0.019
30 **

∆History effect
50 NS NS

(Ts – TD, %)
20
NS
10
P = 0.32 P = 0.003
P = 0.14 P = 0.017 0
0
0 50 100 0 50 100 –10
Performance of Ts (%) Current tone is from the same category (Ts) or from a
different category (Td) compared to previous tone

d Current trial category e

Choice bias
+ Improvement Time
– Impairment t –1 t
High
Logistic()
1
(previous trial)

Sensory SL SH SL Ws
Frequency

Choice CL/R CL/R Wc p 0.5

Constant bias Wb0 0

Low High
s
Frequency
(current trial)

f Model 1: quantify only slope change g Model 2: quantify slope change and h Model comparison:
following silencing previous choice bias following silencing model 2 versus model 1

100 100 100 100 50 40 mice

Control data Silenced data Control data Silenced data (23,572 trials)
Reported high (%)
Reported high (%)

P = 3.14 × 10–72

Counts
50 50 50 50

0 0 0 0
0
8 16 32 8 16 32 8 16 32 8 16 32 –10 0 10
Frequency (kHz) Frequency (kHz) Frequency (kHz) Frequency (kHz) –log(likelihood ratio)

Red Pre-choice is left Model fitting

Blue Pre-choice is right Actual data pool across animals

Fig. 6 | PPC activity counterbalances short-term history bias on perceptual decisions. a, Comparison of behavioral performance on testing trials for
which the preceding training stimulus was of the same category (Ts) or a different category (Td) versus the current testing trial. The choices made were
not significantly biased by the previous trial under normal control conditions (saline, P = 0.14, n = 14 mice; control with light mask (opto-control), P = 0.32,
n = 6 mice; paired t-test, two-sided). b, Similar plot as in a, but for mice with PPC silenced on day 1. Choices made were significantly biased by previous
trials when the PPC was inactivated by either muscimol (P = 0.017, n = 13 mice) or photoinhibition (P = 0.003, n = 7 mice); paired t-test, two-sided. c,
Summary of changes in performance attributable to influence of preceding trial in a and b. Data from a separate group of control mice in the absence of
manipulation (n = 25 mice) were also included. P = 0.449 (no manipulation), P = 0.017 (muscimol), P = 0.143 (saline), P = 0.003 (ChR2 + opto), P = 0.319
(ChR2); one-sample t-test, two-sided; comparison between groups, two-sample t-test; error bars, mean ± s.e.m. d, Schematic showing the bias effect on
categorical choice by preceding trials when the PPC was silenced. e, Illustration of linear regression models (GLMM test) used to simultaneously quantify
the effects of PPC silencing on categorization slopes and short-term history bias (SL, stimuli of low-frequency category; SH, stimuli of high-frequency
category; CL/R, left or right choices; Ws, weight of sensory stimulus term; Wc, weight of choice term; Wb0, weight of constant bias term). f,g, GLMM fit to
data under control and silenced conditions. Lines, fit; dots, data; red, trials for which the previous choice was left; blue, trials for which the previous choice
was right. Model 1 (f) includes the regressor that quantifies slope change after PPC silencing. Model 2 (g) includes regressors that quantify both slope
change and previous change in choice bias after PPC silencing. h, Model comparison (likelihood ratio test) for model 1 versus model 2; one-sample t-test.

PPC activity is required for decisions based on newly learned cat- after these mice had learned to perform categorization of tones in
egories. A potential alterative explanation for the sharp slopes of the first frequency range, and whether the same tone stimuli in the
psychometric function we observed in mice is that these may reflect first frequency range could be ‘re-categorized’ based on the bound-
a bottom-up sensory property: for example, higher sensitivity in ary of the re-learned new category6. We designed tone stimuli from
frequency discrimination at the sensory level that coincides with the two overlapping frequency ranges (~1.75 octave for each range)
task-defined category boundary, rather than reflecting learned cate- with the midpoints of each range separated by ~1 octave as the
gories. To exclude this possibility we examined, in a separate cohort task-defined category boundaries (Fig. 7a; see also Methods). We
of mice, whether a different category boundary could be re-learned first trained mice to discriminate a pair of training tones in a lower

Nature Neuroscience | VOL 22 | JUNE 2019 | 963–973 | www.nature.com/natureneuroscience 969

The Nature trademark is a registered trademark of Springer Nature Limited.
Articles NATure NeuroSCIenCe

a Frequency (octave) b b1 b2
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75
100
b1

(Range 1)

Reported high (%)

Re-categorized
50

Trained
Category low
Training tone
Category high

0
(Range 2) 0 0.875 1.875 2.75

b2 Frequency (octave)

c d e f P = 0.54
Control Silenced 2 P = 0.017
2.75 100
P = 0.0087

∆ Category boundary
Category boundary

b2 b2

Correct rate (%)

1.875
(octave)
(octave)

1 50
b1 b1
0.875

0 0 0
1 2 1 2 Training Re-categorized
l

ed
e e e e
tro

ng ng ng ng tones tones
nc
on

Ra Ra Ra Ra
le
C

Si

Fig. 7 | PPC activity is required for stimulus re-categorization based on new decision boundaries. a, Schematic depicting auditory re-categorization
task. After learning the categorization task for the stimuli in the first frequency range (Range 1), mice were trained by two training stimuli in the second
frequency range followed by testing with tones of intermediate frequency in the second range (Range 2). Some of the stimuli in Range 1 are re-categorized
in Range 2. b, Sample psychometric functions showing re-categorization performance by a control mouse on the first day following introducion of test
stimuli in Range 2 (solid lines) and by a mouse with the PPC silenced in the first session following introducion of test stimuli in Range 2 (dashed lines; blue,
PPC silenced); error bars, mean ± 95% confidence interval across trials. c–d, Category boundaries in Range 1 and Range 2 for mice from control group
(c), n = 5 mice, and the group with PPC silenced on the first day of re-categorization (d), n = 6 mice; error bars, mean ± s.e.m. e, Difference in category
boundaries between Range 1 and Range 2 in control group (n = 5 mice) and PPC-silenced group (n = 6 mice) as in c and d, Wilcoxon rank-sum test, two-
sided. Error bars, mean ± s.e.m. f, Performance on training stimuli and re-categorized stimuli in control mice and PPC-silenced mice; Wilcoxon rank-sum
test, two-sided; error bars, mean ± s.e.m.

frequency range (4–13.5 kHz), associating the lower-frequency tone used the same photoinhibition approach as shown in Fig. 2b. After
(4 kHz) with left licking and the higher-frequency tone (13.5 kHz) mice had learned categorization tasks in the first frequency range
with right licking. After mice had reached a performance level and learned new categories using two train tones in the second
>85%, we introduced testing tones of intermediate frequency in the frequency range, we performed photoinhibition of the PPC while
first frequency range (six tones separated by ~0.25 octave) to obtain introducing test tones in the second frequency range. Compared
psychometric functions. Mice categorized the testing tones with to control mice, bilateral photoinhibition of the PPC significantly
sharp psychometric slopes and with subjective category boundar- reduced the extent of boundary shift during re-categorization
ies located near the task-defined boundary (midpoint of the first (Fig. 7d,e; changes in boundary, 1.26 ± 0.11 octave in control group,
range, b1; Fig. 7a,b). We then trained the same animals to learn 0.61 ± 0.12 octave in silencing group). The correct rate for re-cate-
associations for a different pair of training tones (8 and 26.9 kHz) gorization of test tones was also significantly reduced (78 ± 4 and
in the second frequency range, followed by the introduction of test- 53 ± 9% for control and silencing groups, respectively, P = 0.017;
ing tones from the second frequency range (six tones separated by Fig. 7f). Thus, inactivation of the PPC significantly impaired the
~0.25 octave; Fig. 7a). Since the two frequency ranges overlapped capability of mice in making categorical decisions based on newly
by ~0.75 octave, some of the testing tones had already been expe- learned criteria.
rienced and categorized by the animals during the first categoriza-
tion task and therefore they needed to re-categorize these stimuli PPC–auditory cortex circuits contribute to categorical auditory
according to the new category boundary in the second range. We decision-making. Neurons in the PPC comprise diverse popula-
found that the mice re-categorized these testing tones on the first tions interconnected with many brain regions36 and are involved in
day of testing in the second frequency range, with category bound- a myriad of cognitive functions1,2,4,8,37,38. It is likely that, for a given
aries located near the midpoint of the second frequency range (Fig. task, specific circuits interconnecting the PPC and modality-spe-
7a–c), indicating that they were capable of learning new categories cific brain regions are critically involved. We thus further examined
and making perceptual decisions to re-categorize sensory stimuli the circuit-level specificity of the causal role of PPC neurons in
based on learned criteria. our auditory categorization task by manipulating PPC-to-auditory
Next we sought to determine whether the PPC is involved in re- cortex projections using chemogenetics. We first determined the
categorization of tone stimuli based on newly learned categories. We projection patterns between the PPC and the auditory cortex by

970 Nature Neuroscience | VOL 22 | JUNE 2019 | 963–973 | www.nature.com/natureneuroscience

The Nature trademark is a registered trademark of Springer Nature Limited.

NATure NeuroSCIenCe Articles
a b c
PPC
PPC
PPC
Vis Bregma, –1.94 mm
1 mm

Dd
AU
Dp

Dd
AU

AU
AUDp

1 mm 1 mm
Bregma, –3.08 mm

d e f Model 1: without silencing

AAV-hM4D-mCherry/AAV-mCherry 3 P = 0.040 Model 2: with silencing
60

Psychometric slope
2 40 Mean = –4.94 ± 0.15

Counts
P = 5.42 × 10–99

CNO CNO 1 20

0 0
O –10 0 10

O
N

N
C

C
–log(likelihood ratio)
+

+
rry

4D
he

hM
C
m

Fig. 8 | PPC-to-auditory cortex projections are necessary for categorical decision-making on new sensory stimuli. a,b, Histology images from a sample
animal showing projection patterns between the PPC and auditory cortex. a, Red fluorescence indicates the site of injection of AAV-syn-mCherry in the
AUDp; green fluorescence in the AUDd indicates axons projecting from the PPC. Inset: site of injection of AAV-syn-GFP in the PPC. Vis, visual cortex.
b, Red fluorescence indicates axons projecting from the AUDp in the PPC. The images shown are from one mouse; the projection pattern was validated
using retrograde viral tracing from two mice. c, Schematic showing projection patterns between the PPC and auditory cortex. d, Schematic showing
chemogenetic silencing of the projections from the PPC to the auditory cortex. AAV-syn-hM4D-mCherry (experimental group) or AAV-syn-mCherry
(control group) were injected bilaterally in the PPC. CNO was micro injected bilaterally in the auditory cortex before the first session introducing test
tone stimuli. e, Summary of the effect of silencing of the PPC on auditory cortex projections on psychometric slopes. P < 0.05, n (control) = 7 mice, n
(silencing) = 8 mice; Wilcoxon rank-sum test, two-sided; error bars, mean ± s.e.m. f, Likelihood ratio test comparing regression models with or without
projection silencing as a regressor; n = 300 shuffles (data points), paired t-test, two-sided.

expressing green fluorescent proteins in the PPC (Fig. 8a) and red
fluorescent proteins (tdTomato) in the primary auditory cortex
(AUDp; Fig. 8b), using AAV vectors. We found that the PPC received
projections from the AUDp while sending projections to the dor-
sal portion of the auditory cortex (AUDd), indicating an inter-
connected circuit between the PPC and auditory cortex (Fig. 8c).
We next expressed hM4Di, a designer receptor exclusively activated
by designer drug, in the PPC of both hemispheres using an AAV
vector (AAV-Syn-hM4Di). On day 1 (test tones first introduced)
we injected clozapine-N-oxide (CNO) bilaterally into the AUDd to
silence projections from the PPC before the behavioral session (Fig.
8d). As control, a separate group of mice were injected with AAV-
Syn-mCherry in the PPC followed by injection with CNO in the
AUDd during task performance on day 1. We found that, compared
to the control group, silencing of the PPC to AUDd projections
impaired categorization performance with the new tone stimuli, as
indicated by a significant reduction in the slopes of psychometric
function (Fig. 8e). Quantification using a GLMM test also showed a
significantly enhanced prediction of data with the interaction term
of projection silencing and sensory stimuli compared to absence of
the interaction term (Fig. 8f). These results indicate that specific
projections from the PPC to the AUDd play a critical role in the
auditory categorical decisions made on new sensory stimuli, imply-
ing a circuit mechanism for information routing during the learning
process of categorization.
Discussion
When making decisions on unknown sensory stimuli, the brain
uses previous knowledge of sensorimotor associations to make cat-
egorical choices. Such integration of new sensory information with
previous knowledge to ensure efficient sensorimotor transforma-
tion is a dynamic process: new sensory stimuli are rapidly incor-
porated into existing categories, forming reliable associations with
motor actions. Our results support the concept that PPC circuits
play a critical role in this process. We show that silencing of the PPC
impaired categorical decisions on new sensory stimuli; however,
after the new sensory stimuli become well experienced, presumably
incorporated with existing categories, PPC activity was no longer
required for behavior regarding choice (Figs. 2, 3). Monitoring of
PPC population activity at single-cell resolution across this dynamic
process revealed both stable representation of existing categori-
cal choices (Fig. 4) and dynamic representations for new sensory
stimuli, with a trend of incorporating new sensory information into
learned categories (Fig. 5). The critical role of the PPC in making
categorical decisions is also manifest in its contribution to counter-
balancing short-term history bias (Fig. 6), and to the flexibility of
re-categorizing previously experienced sensory stimuli according to
newly learned categories (Fig. 7). At the circuit level, we identified
a role of the PPC essential to auditory cortex feedback projections,
implying circuit-level information routing during categorization of
new sensory stimuli (Fig. 8).

Nature Neuroscience | VOL 22 | JUNE 2019 | 963–973 | www.nature.com/natureneuroscience 971

The Nature trademark is a registered trademark of Springer Nature Limited.
Articles
The causal link between PPC activity and decision-making
behavior remains controversial. Some recent studies showed that
silencing of the PPC impaired visually guided8,11,29,19, but not audi-
tory-guided11,17,19, choice behaviors in rodents, which tends to sug-
gest that the PPC may have certain visual-specific functions but that
is not critical for general sensorimotor decisions11,19. In macaques,
silencing of the parietal cortex influenced neither visual motion15-
nor self-motion16-guided choice behavior, further questioning the
causal link between PPC activity and choice behavior. These contro-
versies call into question the functional significance of the PPC and
may lend support to the view that computations for perceptual deci-
sion-making are carried out in other brain circuits and that decision-
related activity in the PPC may reflect signals inherited from only
those circuits15,19. Our results, however, support the alternative idea
that the PPC plays a critical causal role in decision-making when the
subject is facing novel or unknown sensory stimuli, or when making
decisions on familiar stimuli according to new criteria.
The functional role of the PPC has also been studied in the con-
text of other cognitive functions, including working memory8,39,40
and attention41–43. A recent study showed that inactivation of the
PPC in rats during a delayed comparison auditory working mem-
ory task improved behavioral performance32. The seemingly oppo-
site behavioral effects in this study and in our study are probably
due to fundamental differences in task paradigms assessing distinct
cognitive functions. Performance on the working memory task was
dependent on both short-term sensory memory of the first stimulus
in each trial and a sensorimotor decision after the second stimulus.
In the previous study, only when the PPC was inactivated during
the delayed memory period was the task performance influenced32,
while such a memory component was not included in our task. It
was proposed that the observed interference with working memory
by previous sensory experience is attributable to biased choices
toward an average stimulus32. Because the PPC represents previous
experience, its inactivation resulted in less interference and thus
improved working memory performance. This is consistent with
the function of the PPC in stimulus categorization, since perceptual
classification of individual stimuli into a learned category (repre-
sented as either a prototype44 or average stimulus) may effectively
bias the perception or perceptual memory of individual stimuli
toward the ‘average stimulus’. In our categorization task, the PPC
represents a learned category and critically contributed to catego-
rization of new sensory stimuli, but may have somewhat biased the
short-term sensory memory toward learned categories in the work-
ing memory task, producing apparently opposite effects on task
performance. Therefore, the observed different behavioral effects
are not contradictory—rather, they reflect common PPC functions.
While the precise computational mechanism by which PPC neu-
rons implement categorization of new sensory stimuli may require
further investigation, our data, together with results from previous
studies10,20,32, suggest that PPC neurons represent previously learned
sensorimotor associations to guide decision-making on new sen-
sory stimuli. Several lines of evidence from our data support this
view. First, our in vivo imaging data show that the PPC represents
stable category/choice information over several days (Fig. 4 and
Supplementary Fig. 16), similar to results previously reported for
population representations20. Such stable representations may serve
as the basis for integration of new sensory stimuli. Second, our
manipulation experiments show that silencing of the PPC affected
choice performance only on new sensory stimuli (Figs. 2, 3 and
Supplementary Fig. 17), suggesting that the PPC is important in the
incorporation of new sensory information with existing representa-
tions whereas the execution of previously learned sensory–motor
associations is likely to be implemented by other circuits45. Third,
inactivation of the PPC significantly increased the biasing effect
from immediate preceding events, suggesting that the representa-
tions of previous category knowledge in the PPC counterbalance
972 Nature Neuroscience | VOL 22 | JUNE 2019 | 963–973 | www.nature.com/natureneuroscience
The Nature trademark is a registered trademark of Springer Nature Limited.
NATure NeuroSCIenCe
the uninformative influences from more recent events (Fig. 6).
Fourth, our observation of the impaired re-categorization of previ-
ously learned stimuli in the second frequency range following PPC
inactivation (Fig. 7) is consistent with the idea that when the guid-
ance from re-learned associations is compromised by PPC inacti-
vation, choices tend to depend more on associations established at
even earlier stages, that is, associations in the first frequency range.
Finally, the critical involvement of the PPC in auditory cortex feed-
back projections (Fig. 8) suggests a potential circuit mechanism that
may influence representations in sensory areas for stimulus catego-
rization. An important future direction would be to examine PPC-
dependent top-down influence on representations in the auditory
cortex during stimulus categorization.
Online content
Any methods, additional references, Nature Research reporting
summaries, source data, extended data, supplementary informa-
tion, acknowledgements, peer review information; details of author
contributions and competing interests; and statements of data and
code availability are available at https://doi.org/10.1038/s41593-
019-0383-6.
Received: 11 May 2018; Accepted: 13 March 2019;
Published online: 29 April 2019
References
1. Shadlen, M. N. & Newsome, W. T. Neural basis of a perceptual decision in
the parietal cortex (area LIP) of the rhesus monkey. J. Neurophysiol. 86,
1916–1936 (2001).
2. Gold, J. I. & Shadlen, M. N. The neural basis of decision making. Annu. Rev.
Neurosci. 30, 535–574 (2007).
3. Yang, T. & Shadlen, M. N. Probabilistic reasoning by neurons. Nature 447,
1075–1080 (2007).
4. Freedman, D. J. & Assad, J. A. A proposed common neural mechanism for
categorization and perceptual decisions. Nat. Neurosci. 14, 143–146 (2011).
5. Freedman, D. J. & Assad, J. A. Neuronal mechanisms of visual categorization:
an abstract view on decision making. Annu. Rev. Neurosci. 39,
129–147 (2016).
6. Freedman, D. J. & Assad, J. A. Experience-dependent representation of visual
categories in parietal cortex. Nature 443, 85–88 (2006).
7. Hanks, T. D. et al. Distinct relationships of parietal and prefrontal cortices to
evidence accumulation. Nature 520, 220–223 (2015).
8. Harvey, C. D., Coen, P. & Tank, D. W. Choice-specific sequences in parietal
cortex during a virtual-navigation decision task. Nature 484, 62–68 (2012).
9. Runyan, C. A., Piasini, E., Panzeri, S. & Harvey, C. D. Distinct timescales of
population coding across cortex. Nature 548, 92–96 (2017).
10. Morcos, A. S. & Harvey, C. D. History-dependent variability in population
dynamics during evidence accumulation in cortex. Nat. Neurosci. 19,
1672–1681 (2016).
11. Raposo, D., Kaufman, M. T. & Churchland, A. K. A category-free neural
population supports evolving demands during decision-making. Nat.
Neurosci. 17, 1784–1792 (2014).
12. Wang, X.-J. Decision making in recurrent neuronal circuits. Neuron 60,
215–234 (2008).
13. Wong, K.-F. & Wang, X.-J. A recurrent network mechanism of time
integration in perceptual decisions. J. Neurosci. 26, 1314–1328 (2006).
14. Wang, X.-J. Probabilistic decision making by slow reverberation in cortical
circuits. Neuron 36, 955–968 (2002).
15. Katz, L. N., Yates, J. L., Pillow, J. W. & Huk, A. C. Dissociated functional
significance of decision-related activity in the primate dorsal stream. Nature
535, 285–288 (2016).
16. Chen, A., Gu, Y., Liu, S., DeAngelis, G. C. & Angelaki, D. E. Evidence for a
causal contribution of macaque vestibular, but not intraparietal, cortex to
heading perception. J. Neurosci. 36, 3789–3798 (2016).
17. Erlich, J. C., Brunton, B. W., Duan, C. A., Hanks, T. D. & Brody, C. D.
Distinct effects of prefrontal and parietal cortex inactivations on an
accumulation of evidence task in the rat. eLife 4, e05457 (2015).
18. Guo, Z. V. et al. Flow of cortical activity underlying a tactile decision in mice.
Neuron 81, 179–194 (2014).
19. Licata, A. M. et al. Posterior parietal cortex guides visual decisions in rats.
J. Neurosci. 37, 4954–4966 (2017).
20. Driscoll, L. N., Pettit, N. L., Minderer, M., Chettih, S. N. & Harvey, C. D.
Dynamic reorganization of neuronal activity patterns in parietal cortex. Cell
170, 986–999.e16 (2017).
NATure NeuroSCIenCe
21. Hwang, E. J., Dahlen, J. E., Mukundan, M. & Komiyama, T. History-based
action selection bias in posterior parietal cortex. Nat. Commun. 8,
1242 (2017).
22. Zhang, F., Wang, L.-P., Boyden, E. S. & Deisseroth, K. Channelrhodopsin-2
and optical control of excitable cells. Nat. Methods 3, 785–792 (2006).
23. Chao, H.-T. et al. Dysfunction in GABA signalling mediates autism-like
stereotypies and Rett syndrome phenotypes. Nature 468, 263–269 (2010).
24. Ohl, F. W., Scheich, H. & Freeman, W. J. Change in pattern of ongoing
cortical activity with auditory category learning. Nature 412, 733–736 (2001).
25. Zentall, T. R., Galizio, M. & Critchfield, T. S. Categorization, concept
learning, and behavior analysis: an introduction. J. Exp. Anal. Behav. 78,
237–248 (2002).
26. Ashby, F. G. & Ell, S. W. The neurobiology of human category learning.
Trends Cogn. Sci. 5, 204–210 (2001).
27. Chen, T.-W. et al. Ultrasensitive fluorescent proteins for imaging neuronal
activity. Nature 499, 295–300 (2013).
28. Xu, N. et al. Nonlinear dendritic integration of sensory and motor input
during an active sensing task. Nature 492, 247–251 (2012).
29. Goard, M. J., Pho, G. N., Woodson, J. & Sur, M. Distinct roles of visual,
parietal, and frontal motor cortices in memory-guided sensorimotor
decisions. eLife 5, e13764 (2016).
30. Feierstein, C. E., Quirk, M. C., Uchida, N., Sosulski, D. L. & Mainen, Z. F.
Representation of spatial goals in rat orbitofrontal cortex. Neuron 51,
495–507 (2006).
31. Hung, C. P., Kreiman, G., Poggio, T. & DiCarlo, J. J. Fast readout of object
identity from macaque inferior temporal cortex. Science 310, 863–866 (2005).
32. Akrami, A., Kopec, C. D., Diamond, M. E. & Brody, C. D. Posterior parietal
cortex represents sensory history and mediates its effects on behaviour.
Nature 554, 368–372 (2018).
33. Bertelson, P. Serial choice reaction-time as a function of response versus
signal-and-response repetition. Nature 206, 217–218 (1965).
34. Gold, J. I., Law, C.-T., Connolly, P. & Bennur, S. The relative influences of
priors and sensory evidence on an oculomotor decision variable during
perceptual learning. J. Neurophysiol. 100, 2653–2668 (2008).
35. Busse, L. et al. The detection of visual contrast in the behaving mouse.
J. Neurosci. 31, 11351–11361 (2011).
36. Zingg, B. et al. Neural networks of the mouse neocortex. Cell 156,
1096–1111 (2014).
37. Andersen, R. A. & Buneo, C. A. Intentional maps in posterior parietal cortex.
Annu. Rev. Neurosci. 25, 189–220 (2002).
38. Bisley, J. W. & Goldberg, M. E. Attention, intention, and priority in the
parietal lobe. Annu. Rev. Neurosci. 33, 1–21 (2010).
39. Todd, J. J. & Marois, R. Capacity limit of visual short-term memory in human
posterior parietal cortex. Nature 428, 751–754 (2004).
40. Pesaran, B., Pezaris, J. S., Sahani, M., Mitra, P. P. & Andersen, R. A. Temporal
structure in neuronal activity during working memory in macaque parietal
cortex. Nat. Neurosci. 5, 805–811 (2002).
41. Buschman, T. J. & Miller, E. K. Top-down versus bottom-up control of
attention in the prefrontal and posterior parietal cortices. Science 315,
1860–1862 (2007).
Nature Neuroscience | VOL 22 | JUNE 2019 | 963–973 | www.nature.com/natureneuroscience
The Nature trademark is a registered trademark of Springer Nature Limited.
Articles
42. Corbetta, M., Kincade, J. M., Ollinger, J. M., McAvoy, M. P. & Shulman, G. L.
Voluntary orienting is dissociated from target detection in human posterior
parietal cortex. Nat. Neurosci. 3, 292–297 (2000).
43. Behrmann, M., Geng, J. J. & Shomstein, S. Parietal cortex and attention. Curr.
Opin. Neurobiol. 14, 212–217 (2004).
44. Rosch, E. Principles of categorization. in Cognition and Categorization (eds.
Rosch, E. & Lloyd, B.B.) 28–49 (L. Erlbaum Associates, 1978).
45. Znamenskiy, P. & Zador, A. M. Corticostriatal neurons in auditory cortex
drive decisions during auditory discrimination. Nature 497, 482–485 (2013).
Acknowledgements
We thank R. Egnor for advice on auditory behavioral apparatus, A. Kerlin for
communication on behavioral control software, N. Andersen for pilot optogenetic
experiments, T.T. Zhou for technical support, L. Cui, S. Tang and J. Xiao for helping
with behavior training, Y. Xin for discussions on data analysis, ION Gene-editing Core
Facility for providing the adeno-associated viruses and M.M. Poo for comments on the
manuscript. This work was supported by Key Research Program of Frontier Sciences,
CAS (grant No. QYZDB-SSW-SMC045), National Natural Science Foundation of China
(grant No. 31571081), the Strategic Priority Research Program of the Chinese Academy
of Sciences (grant No. XDB32010000), National Key R&D Program of China (grant Nos.
2017YFA0103900/2017YFA0103901), Shanghai Municipal Science and Technology Major
Project (grant No. 2018SHZDZX05) and the Youth Thousand Talents Plan (to N.L.X.).
C.A.D. is supported by the Simons Collaboration on the Global Brain Postdoctoral
Fellowship and the CPSF-CAS Joint Foundation for Excellent Postdoctoral Fellows.
Author contributions
L.Z. and N.L.X. conceived the project and designed the experiments. L.Z. performed all
the experiments and data analysis. Y.Z. performed the axon inactivation experiments.
L.Z. and C.A.D. conceived and performed the GLMM analysis. J.D. collected part of
the behavioral data. L.Z., J.P. and N.L.X. developed the hardware and software for the
behavior and imaging system. L.Z. and N.L.X. wrote the manuscript with contributions
from C.A.D and Y.Z.
Competing interests
The authors declare no competing interests.
Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/
s41593-019-0383-6.
Reprints and permissions information is available at www.nature.com/reprints.
Correspondence and requests for materials should be addressed to N.-l.X.
Journal peer review information Nature Neuroscience thanks Jonathan Whitlock and
the other anonymous reviewer(s) for their contribution to the peer review of this work.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
© The Author(s), under exclusive licence to Springer Nature America, Inc. 2019
973
Articles
Methods
Animal subjects and behavior. All experimental procedures were carried out in
compliance with those approved by the Animal Care and Use Committee of the
Institute of Neuroscience, Chinese Academy of Sciences. Data were acquired from
male C57BL/6J and VgaT-Cre mice (Jackson Laboratory, No. JAX 017535)27, aged
8–10 weeks at the start of the experiment. Mice used for behavioral tests were
housed in a room with a reverse light/dark cycle. Mice had no previous history of
any other experiments. On days when not tested, mice received 1 ml of free water;
on days of testing, behavioral sessions lasted for 1–2 h when they received water
mainly from task performance and with supplements to reach a total of 1 ml. Each
mouse’s weight was measured daily to ensure that weight loss did not exceed 20%
of pre-water restriction weight. Animals used in the control and testing groups
were randomly chosen from either the common pool of purchased animals or
in-house-bred litters.
After ~7 days’ recovery following surgery, mice were started with a water-
restriction procedure. Each mouse received 1 ml water per day and the body
weight was monitored. After ~7 days of water restriction, behavior training was
started28,46. Water consumption was calculated as total water = number of rewarded
trials × ~6 µl water in one delivery for each training session. If mice consumed
<0.5 ml water, additional supplementation was provided. Mice were allowed to
perform the task until sated.
Behavior was divided into training and testing stages. In the former, mice
learned to discriminate the tones of two simple frequencies separated by two
octaves (8 and 32 kHz). Each trial was started with a random delay of 500–
1,000 ms, followed by tone stimuli (duration, 300 ms). Mice were required to
respond by licking the left or right lickport within a response period of ~1–3 s to
indicate their choices. Correct answers were defined as licking the left lickport in
response to the lower-frequency tone (for example, 8 kHz) and licking the right
lickport in response to the higher-frequency tone (for example, 32 kHz). Correct
responses led to opening of the water valve to dispense a small amount of water
as reward (~6 µl). Error responses led to a time-out punishment of ~2–4 s, during
which time licking of the wrong side reinitiated the time-out period. When mice
made no response lick within the 3 s response window, the trial was defined as a
miss, leading to the initiation of an inter-trial interval. The level of correct response
reached >85% in 7–10 days (Supplementary Fig. 1c,d).
During the testing stage, eight different frequency tones (8,000, 11,300, 13,000,
14,900, 17,100, 19,700, 22,600 and 32,000; in octave difference related to 8,000 Hz:
0, 0.5, 0.7, 0.9, 1.1, 1.3, 1.5 and 2) were presented in pseudo-randomly interleaved
trials. Trials containing newly introduced frequencies were defined as testing trials,
constituting ~30% of total trials. Similar to most previous psychometric studies in
rodents and primates, testing trials were rewarded according to the experimentally
defined category boundary at the midpoint between two training tones.
The psychometric functions were fitted to choice data using a four-parameter
sigmoidal function47:
  s−μ  
f (s , μ, v , g , l ) = g + (1−g −l ) 1 + erf  
  v 2  
 
where s is the tone frequency in the octave, μ is the mean value of the distribution
representing the subject’s point of subjective equality (PSE), v is the variation in
distribution representing the subject’s discrimination sensitivity and g and l are the
guess rate and lapse rate, respectively, and erf is the error function.
In a separate group of mice, we also examined whether these individuals would
exhibit categorization without an externally imposed category boundary. Rather
than rewarding in the trials according to the experimentally defined categorization
boundary, we provided random rewards in the testing trials starting from the
day when the test tones were introduced and continued these for 5–7 days.
Mice showed clear categorical performance as indicated by the psychometric
functions (Supplementary Fig. 1e), suggesting that they tended to categorize newly
encountered stimuli according to previously trained associations.
Behavioral system. Experiments were conducted in custom-designed and
-fabricated double-walled, sound-attenuating boxes. Mice were head-fixed with
a pair of clamps and thumbscrews. Each mouse was contained within an acrylic
‘body tube’ (diameter, 25 mm), with the head extending out and the front paws
gripping the tube edge or a ledge after head fixation (Supplementary Video 1).
The holder and body tube in turn were attached to a custom-designed tube holder
fixed to a caddy which was mounted on the behavior box with table clamps after
head fixation. Water rewards were provided by two custom-made metal lickports
placed to the left and right of the mouse’s mouth. The lickports are connected to
a custom-desgined, capacitive-sensing circuit board that reliably detected contact
by the tongue during licking. The amount of water delivered was controlled by
the time during which a solenoid water valve was open, calibrated at least once
per week. Calibration was performed by estimating the duration of valve opening
necessary to deliver the desired volume per trial, by running 100 deliveries and
calculating water volume by weight. The fluid pressure was not adjusted within
each behavioral session.
The auditory discrimination behavior of mice was controlled by a custom-
developed, low-cost and high-precision real-time control system, the PX-Behavior
The Nature trademark is a registered trademark of Springer Nature Limited.
NATure NeuroSCIenCe
System (Supplementary Fig. 1a,b). This system includes a custom-designed tone-
generating module that generated sound waveforms of arbitrarily high frequency
at high fidelity, and a microcontroller (Arduino MEGA 2560) implementing
a real-time virtual-state machine framework for programming the behavioral
protocols, stimulus delivery and measurement of behavioral events. The Arduino
board communicated with the tone-generating module through a custom-designed
cache board without interrupting the flow of the behavioral protocols. The tone-
generating module sent the specified sound waveforms to an amplifier (ZB1PS,
Tucker-Davis Technologies) to drive a speaker (ES1, Tucker-Davis Technologies)
to produce acoustic stimuli. Cosine ramps (5 ms) were applied to the rising and
falling of all tones. Behavioral data were logged via a serial port using custom-
written software in python.
The sound system was calibrated using a free-field microphone (Type
4939, Brüel and Kjær) over 3–60 kHz that showed a smooth spectrum (±5 dB).
Measurements were performed with the behavioral box closed and the microphone
positioned at the location and orientation of the mouse’s ear in the presence of the
mouse-mounting system. The microphone was connected to a preamplifier (Type
2670, Brüel and Kjær) and signals were digitized with a National Instruments
acquisition card (NI 9201) at 500,000 samples s–1 for analysis with custom software
developed in LabVIEW (National Instruments).
Surgery and virus injection. During surgery, mice were anaesthetized
with isoflurane (~2%). For inactivation experiments, we targeted the
PPC bilaterally (2 mm caudal, 1.5 mm lateral to bregma, both hemispheres)
by micro-injection of muscimol in wild-type mice, or by viral solution
in VgaT-Cre mice. For photoinhibition experiments, craniotomies were made
over the PPC of VgaT-Cre mice following virus injection. Viral solution
containing either AAV2/8-CAG-FLEX-ChR2-mCherry (titer: ~1013 genomes ml–
1
) or AAV2/9-Syn-FLEX-ChrimsonR-tdTomato (titer: ~1013 genomes ml–1;
Shanghai Taitool Bioscience Co.) was injected slowly (~100 nl over ~20 min).
The injection system comprised a pulled-glass pipette (25–30 μm outside
diameter at the tip; Wiretrol II Capillary Microdispenser, Drummond Scientific)
back-filled with mineral oil. A fitted plunger was inserted into the pipette and
advanced to displace the contents using a hydraulic manipulator (MO-10,
Narashige). Retraction of the plunger was used to load the pipette with virus
solution. The injection pipette was advanced into the brain using a Sutter MP-
225 micromanipulator. After injection, craniotomies were covered with a circular
glass coverslip and sealed in place with dental cement (Jet Repair Acrylic, Lang
Dental Manufacturing).
For two-photon in vivo imaging, a circular craniotomy (diameter, 2–3 mm) was
made over the PPC of the left hemisphere, followed by injection of AAV2/9-hSyn-
GCaMP6s (titer: ~1013 genomes ml–1; Shanghai Taitool Bioscience Co.) at a depth
of ~250 μm. After injection, a double-layered imaging window was implanted at
the craniotomy as described previously31. The double-layered glass comprised
two 200 μm glass coverslips (diameter, ~3 mm) attached to a larger glass coverslip
(diameter, 5 mm) using ultraviolet-cured optical adhesive (Edmund Optics).
Following virus injection and window implantation, a titanium head-plate was
then attached to the skull with cyanoacrylate glue and dental cement to permit
head fixation during behavior. For pharmacological inactivation, the location of
the PPC of both hemispheres was marked for later injection. Mice were allowed at
least 7 days to recover before water restriction.
Pharmacological inactivation of PPC. We used the GABA agonist muscimol
hydrobromide, dissolved in saline, 5 mg ml–1 (Sigma-Aldrich) to reversibly silence
the PPC by stereotactic injection. We used the same nano-injection system used in
our virus injection to inject a small volume at slow speed (50 nl at 10 nl min–1) to
ensure accurate and localized injection volumes31, which minimized the spread of
muscimol to neighboring brain regions compared to conventional cannula-based
infusion systems. Diluted (1:20) red beads were added to identify the injection sites
for both muscimol and saline injections.
Photoinhibition. For optogenetic inactivation of the PPC, blue laser (473 nm)
was delivered to the PPC of both hemispheres through two collimators connected
to a splitting fiberoptic unit coupled to a diode-pumped solid-state laser (DPSS,
Shanghai Laser & Optics Century Co.). Photostimulation, composed of trains of
pulses at 40 Hz, pulse width 5 ms, was delivered from 0.1 s before the stimulus
onset and lasted for 1.4 s to cover the sampling and choice period. Laser power at
the cranial windows (beam size ~1 mm in diameter) was calibrated before the start
of each session (6.5–8.0 mW per mm2). To prevent disturbance to the mice by blue
light illumination, a ‘masking flash’ light was delivered with the same pattern of
photostimulation used throughout each trial using two blue light-emitting diodes
positioned near the eyes (Supplementary Video 1).
We investigated whether manipulation of PPC activity would directly influence
motor output, by plotting the reaction time and licking (Supplementary
Fig. 16). We observed no significant changes in reaction times when the PPC was
silenced by either muscimol or photoinhibition in comparison to control mice
(Supplementary Fig. 16a,b). Neither did we observe any significant changes in the
lick rate during the response period when the PPC was silenced (Supplementary
Fig. 16c,d).
Nature Neuroscience | www.nature.com/natureneuroscience
NATure NeuroSCIenCe
Chemogenetic inactivation of PPC projections. We used a designer receptor
exclusively activated by the designer drugs-based method to silence the PPC to
dorsal auditory cortex projections. AAV2/9-hSyn-DIO-hM4D(Gi)-mCherry
(1.72 × 1012 genomic copies ml–1) was injected bilaterally into the PPC and was
allowed for 4 weeks for development of expression. CNO was dissolved in saline
(0.9% NaCl solution) to give a stock solution of 20 mg ml–1 (stored at −20°C)
and was diluted to a working concentration of 1 µg µl–1 each day before
experiments. During the day of experimental testing, saline or CNO (1 µl per
hemisphere,1 µg µl–1) was micro-injected bilaterally into the AUDd ~30–40 min
before behavioral testing.
Two-photon in vivo calcium imaging. Imaging was performed using a
custom-built two-photon microscope (http://openwiki.janelia.org/wiki/display/
shareddesigns/MIMMS). GCaMP6s was excited using a Ti-Sapphire laser
(Chameleon Ultra II, Coherent) tuned to 920 nm. Images were acquired using
a 16 × 0.8 numerical aperture objective (Nikon); GCaMP6s fluorescence was
isolated using a bandpass filter (No. 525/50, Semrock) and detected using GaAsP
photomultiplier tubes (No. 10770PB-40, Hamamatsu). Horizontal scanning was
accomplished using a resonant galvanometer (Thorlabs; 16 kHz line rate, bi-
directional). The noise produced by the resonant scanner was attenuated to <30 dB
using an optical window sealing the output opening of the resonant scanner
module. The entire microscope was enclosed in a custom-designed, custom-
fabricated, double-walled sound attenuation box (internal noise level <30 dB with
the two-photon imaging system running). The average laser power for imaging
was ~70 mW, measured at the entrance pupil of the objective. The field of view
was 300 × 300 um (512 × 512 pixels), imaged at ~30 frames s–1. The system was
controlled using ScanImage (http://scanimage.org)48. The initiation of each trial
was synchronized between image acquisition and behavior by a trigger signal sent
from our PX-Behavior System to the ScanImage system. One imaging field was
acquired for each of six mice over either 2 days (two mice) or 3 days (four mice)
after introduction of the test tone stimuli. The imaging fields on days 2 and 3 were
aligned to that on day 1, based on the superficial blood vessels in zoom-out view
followed by careful alignment to single neurons at higher-level magnification.
Imaging data analysis. To correct for brain motion, all imaging frames from
each imaging/behavior session were aligned to a target image frame using a
cross-correlation-based registration algorithm (discrete Fourier transformation,
algorithm)49. The target image was obtained by mean projection of image frames
from a trial visually identified to contain still frames. To extract fluorescence
signals from individual neurons, regions of interest (ROIs) were drawn manually
based on neuronal shapes. Mean, maximum intensity and standard deviation
values of all frames in a session were used to determine the boundaries of the
neurons. The pixels in each ROI were averaged to estimate the time series of
fluorescence (F) of a single cell. Before calculating ∆F/F0, slow calcium fluorescence
change correction was performed by subtracting the eighth percentile of ROI
fluorescence within a 200 s time window for each frame. For each ROI, ∆F/F0 (%)
was calculated as (∆F/F0) × 100, where F0 is the index of the peak of the histogram
of F.
To quantify single-neuronal selectivity differentiating training stimuli and
newly introduced test stimuli, or the high and low categories, we used an ideal
observer decoding based on ROC analysis30,50. We defined relative selectivity
index (SI) as
SI = 2 × (AUC−0.5)
which is a measure with a range of −1 to 1. For selectivity of training versus testing,
selectivity index <0 means ‘preferring training stimuli’ and selectivity index >0
means ‘preferring test stimuli’, with 0 representing no selectivity. For stimulus
category selectivity, selectivity index <0 for high and >0 for low. Selectivity index
across the time course of the behavioral trial was computed from frame-by-frame
Ca2+ signals (Figs. 4d and 5b,c). To calculate the selectivity index in a specific
period—for example, ~0–1 s after tone onset—average ∆F/F0 values in this period
in each trial were used.
We used a linear classifier based on a support vector machine with linear kernels.
Simultaneously imaged calcium signals were arranged into a M × N matrix, where M
is the number of trials and N is the number of neurons. Each element in the matrix
is the ∆F/F0 of a particular neuron at a given image frame in a trial. Such matrices
were prepared for all imaging frames across time. To decode the stimulus categories
on day 1, we used data from equal numbers of low- and high-frequency training
trials on day 1 as the training set to train the linear kernel support vector machine
classifier, used ‘leave-one-out’ cross-validation to test the decoder and repeated
this process 1,000 times to obtain the average accuracy of decoding. To decode the
stimulus categories on days 2 and 3, we also used data from the training trials on
day 1 as the training set to predict the training trials on days 2 and 3 (Fig. 4f).
Linear regression models. To address the contribution of multiple variables
on animal behavior (auditory stimuli in octave, effect of silencing the PPC and
previous trial history), we fitted a series of GLMMs using the MATLAB function
Nature Neuroscience | www.nature.com/natureneuroscience
The Nature trademark is a registered trademark of Springer Nature Limited.
Articles
fitglme. First, to characterize the effect of previous trial history on control sessions
(Fig. 6f) we fitted a model where the probability of rightward choice was a logistic
function of pure tone stimuli (in octave) and animals’ choice on the previous trial.
We compared this model to a simpler model without the previous trial effect (see
Model comparison) and found that inclusion of the previous trial effect resulted in
a significantly worse fit to control behavior (P < 0.001).
Second, we quantified the effect of silencing bilateral PPC neurons on
psychometric performance by including an interaction term between auditory
stimuli and the silenced condition. The statistics of this interaction term reflect a
change in the slope of the psychometric curves (Supplementary Fig. 14). Addition
of this interaction term significantly improved the fit to animal behavior compared
to the simpler model of fitting auditory stimuli alone (P < 0.001).
Finally, we examined whether silencing of PPC neurons would also
change the effect of previous trial history, in addition to the silencing effect on
psychometric slopes. For this final model, choice was fit as a logistic function
of stimuli, the interaction between stimuli and the silencing condition and
the interaction of previous choice and the silencing condition (Fig. 6g and
Supplementary Fig. 15). The final model was a significantly better fit to behavior
than the model lacking the final interaction term (P < 0.001). For all GLMMs,
the rightward and leftward choices were defined as 1 and 0, respectively. For the
silenced condition, 0 and 1, respectively, denote whether the current trial was
from a control or silenced session. Random effects of psychometric biases were
included for individual animals.
Model comparison. To justify more complex models of behavior, we conducted
model comparisons using the cross-validated negative log likelihood measure. We
always compared a test model to one extra predictor (for example, the interaction
between stimuli and the silenced condition) against a null model lacking that extra
term. For each round of cross-validated model prediction, 70% of the trials were
randomly selected as the training set with the remaining 30% as the testing set. The
same dataset was used for both the test model and null model. The negative log
likelihood ratio (–logLR) for that round is the difference between the negative log
likelihoods of the two models (test – null). This procedure was repeated 300 times
and the distribution of –logLR was compared to 0. If the distribution of –logLR
was significantly less than 0, this meant that the test model outperformed the null
model in predicting animal behavior, thus justifying addition of the extra term.
Statistical analysis. All statistical analyses were performed in MATLAB. Datasets
were tested for normality, and appropriate statistical tests applied as described in
the text (for example, t-test for normally distributed data, Wilcoxon rank-sum test
for non-parametric data, GLMM model comparison and cross-validated negative
log likelihood test). Data distribution was assumed to be normal, but this was
not formally tested. Shaded regions surrounding line-plots indicate s.e.m. unless
otherwise stated. Unless indicated in Methods, no statistical methods were used
to predetermine sample sizes but our sample sizes are similar to those reported
in previous publications11,18,21,28. Data collection and analysis were not performed
blind to the conditions of the experiments, since all behavioral variables were
objectively measured by an automated hardware and software system that did not
require human intervention. See Reporting Summary for additional information.
Reporting Summary. Further information on research design is available in the
Nature Research Reporting Summary linked to this article.
Data availability
All data used to understand and assess the conclusions of this study are available in
the main text or supplementary materials. All the original behavioral, optogenetic,
imaging and histochemical data are archived in the Institute of Neuroscience,
Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, and are
available from the corresponding author upon reasonable request.
Code availability
All data acquisition and analysis code are archived in the Institute of Neuroscience,
Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences, and are
available from the corresponding author upon reasonable request.
References
46. Guo, Z. V. et al. Procedures for behavioral experiments in head-fixed mice.
PLoS One 9, e88678 (2014).
47. Wichmann, F. A. & Hill, N. J. The psychometric function: I. Fitting, sampling,
and goodness of fit. Percept. Psychophys. 63, 1293–1313 (2001).
48. Pologruto, T. A., Sabatini, B. L. & Svoboda, K. ScanImage: flexible software
for operating laser scanning microscopes. Biomed. Eng. OnLine 2, 13 (2003).
49. Guizar-Sicairos, M., Thurman, S. T. & Fienup, J. R. Efficient subpixel image
registration algorithms. Opt. Lett. 33, 156–158 (2008).
50. Green, D. M. & Swets, J. A. Signal Detection Theory and Psychophysics (John
Wiley and Sons, 1966).
 nature research | reporting summary
Corresponding author(s): Ning-long Xu
Last updated by author(s): Feb 26, 2019

Reporting Summary
Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency
in reporting. For further information on Nature Research policies, see Authors & Referees and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.
n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement
A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly
The statistical test(s) used AND whether they are one- or two-sided
Only common tests should be described solely by name; describe more complex techniques in the Methods section.

A description of all covariates tested

A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons
A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient)
AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals)

For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted
Give P values as exact values whenever suitable.

For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings
For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes
Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated
Our web collection on statistics for biologists contains articles on many of the points above.

Software and code

Policy information about availability of computer code
Data collection For behavior data collection, custom written program in Python (open source) and in Arduino based on Arduino IDE 1.5.6r2 (open
source). For calcium imaging data collection, ScanImage 4.2 (open source) in MATLAB (commercial, Mathworks). For auditory stimulation
calibration, custom written software developed in LabVIEW (National Instruments).

Data analysis Custom code written in MATLAB (Mathworks)

For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors/reviewers.
We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data
Policy information about availability of data
All manuscripts must include a data availability statement. This statement should provide the following information, where applicable:
- Accession codes, unique identifiers, or web links for publicly available datasets
- A list of figures that have associated raw data
- A description of any restrictions on data availability
October 2018

All data to understand and assess the conclusions of this study are available in the main text or supplementary materials. All the original behavioral, optogenetic,
imaging, histochemical data and analysis code are archived in the Institute of Neuroscience, Shanghai Institutes for Biological Sciences, Chinese Academy of
Sciences.

1
The Nature trademark is a registered trademark of Springer Nature Limited.
 nature research | reporting summary
Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.
Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design

All studies must disclose on these points even when the disclosure is negative.
Sample size Unless indicated in the methods, no statistical methods were used to pre-determine sample sizes, but our sample sizes are similar to those
reported in previous publications (Ref 11,18,22,29). All groups are with similar sample sizes. In this study, 6 mice were used for two-photon
imaging. In behavior manipulation, 34 mice were used for parmacological inactivation and saline control experiments, 24 mice were used for
optogenetics inacitvation and control experiments, 3 mice were used for electrophysiology, and 15 mice were used for chemogenetic
inactivation and control experiments.

Data exclusions No data were excluded from the analyses.

Replication Two different methods (pharmacological and optogenetics) were performed independently to verify the reproducibility of the experimental
findings and both reached the same conclusions.

Randomization Animals used in control and testing groups were randomly chosen from the common pool of purchased animals or in-house bred litters.
Stimuli presented in behavioral task were randomly interleaved trial-by-trials.

Blinding All behavioral variables were objectively measured by automated hardware and software system that do not require human intervention, and
were not blinded to investigators.

Reporting for specific materials, systems and methods

We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material,
system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.

Materials & experimental systems Methods

n/a Involved in the study n/a Involved in the study
Antibodies ChIP-seq
Eukaryotic cell lines Flow cytometry
Palaeontology MRI-based neuroimaging
Animals and other organisms
Human research participants
Clinical data

Animals and other organisms

Policy information about studies involving animals; ARRIVE guidelines recommended for reporting animal research
Laboratory animals Species: Mouse.
Strain: C57BL/6J and VGAT-Cre.
Sex: male.
Age: 8 weeks old from the begining of experiments

Wild animals The study did not involve the wild animals

Field-collected samples The study did not invovle the Field-collected samples from the filed
October 2018

Ethics oversight All experimental aspects were carried out in compliance with the procedures approved by the Animal Care and Use Committee
of the Institute of Neuroscience, Chinese Academy of Sciences.
Note that full information on the approval of the study protocol must also be provided in the manuscript.

2
The Nature trademark is a registered trademark of Springer Nature Limited.
Nature Research, brought to you courtesy of Springer Nature Limited (“Nature Research”)

Terms and Conditions

Nature Research supports a reasonable amount of sharing of content by authors, subscribers and authorised or authenticated users (“Users”), for
small-scale personal, non-commercial use provided that you respect and maintain all copyright, trade and service marks and other proprietary
notices. By accessing, viewing or using the nature content you agree to these terms of use (“Terms”). For these purposes, Nature Research
considers academic use (by researchers and students) to be non-commercial.

These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the terms, a site licence or a personal subscription (to the
extent of the conflict or ambiguity only). By sharing, or receiving the content from a shared source, Users agree to be bound by these Terms.

We collect and use personal data to provide access to the nature content. ResearchGate may also use these personal data internally within
ResearchGate and share it with Nature Research, in an anonymised way, for purposes of tracking, analysis and reporting. Nature Research will
not otherwise disclose your personal data unless we have your permission as detailed in the Privacy Policy.

Users and the recipients of the nature content may not:

1. use the nature content for the purpose of providing other users with access to content on a regular or large scale basis or as a means to
circumvent access control;
2. use the nature content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability,
or is otherwise unlawful;
3. falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by either Nature
Research or ResearchGate in writing;
4. use bots or other automated methods to access the nature content or redirect messages; or
5. override any security feature or exclusionary protocol.

These terms of use are reviewed regularly and may be amended at any time. We are not obligated to publish any information or content and may
remove it or features or functionality at our sole discretion, at any time with or without notice. We may revoke this licence to you at any time and
remove access to any copies of the shared content which have been saved.

Sharing of the nature content may not be done in order to create substitute for our own products or services or a systematic database of our
content. Furthermore, we do not allow the creation of a product or service that creates revenue, royalties, rent or income from our content or its
inclusion as part of a paid for service or for other commercial gain. Nature content cannot be used for inter-library loans and librarians may not
upload nature content on a large scale into their, or any other, institutional repository.

To the fullest extent permitted by law Nature Research makes no warranties, representations or guarantees to Users, either express or implied
with respect to the nature content and all parties disclaim and waive any implied warranties or warranties imposed by law, including
merchantability or fitness for any particular purpose.

Please note that these rights do not automatically extend to content, data or other material published by Nature Research that we license from
third parties.

If you intend to distribute our content to a wider audience on a regular basis or in any other manner not expressly permitted by these Terms
please contact us at

onlineservice@springernature.com
The Nature trademark is a registered trademark of Springer Nature Limited.