Received 13 January 2016; revised 25 July 2016; accepted 2 October 2016.

Date of publication 5 April 2017; date of current version 16 May 2017.

Digital Object Identifier 10.1109/LLS.2016.2644647

Multiscale Metabolic Modeling Approach for

Predicting Blood Alcohol Concentration
MASOOD KHAKSAR TOROGHI1 , WILLIAM R. CLUETT1 ,
AND RADHAKRISHNAN MAHADEVAN1,2
1 Department of Chemical Engineering and Applied Chemistry, University of Toronto, Toronto, ON M5S, Canada
2 Institute of Biomaterials and Biomedical Engineering, University of Toronto, Toronto, ON M5S, Canada

CORRESPONDING AUTHOR: R. MAHADEVAN (e-mail: krishna.mahadevan@utoronto.ca).

ABSTRACT Alcohol is one of the most widely consumed and abused substances, and is a major factor
in many alcohol-related diseases, incidents of impaired driving, and crimes. In this letter, we develop
a mechanistic model for alcohol metabolism in the human body based on the dynamic parsimonious
flux balance analysis technique. The developed whole body alcohol metabolic model contains two main
mechanisms for ethanol metabolism in the body, namely, oxidative and non-oxidative mechanisms. The
model is able to demonstrate the effect of variations in biochemical kinetics associated with the alcohol
dehydrogenase enzyme, gender differences, physiological properties of the human body such as age, weight,
and height, and the meal effect on the alcohol clearance from the body. Simulation results show that the model
predictions are consistent with in vivo studies. The results from this letter indicate that the proposed metabolic
modeling approach may open the door to new opportunities in the area of metabolic nutrition research and
personalized medicine since it accounts for physiological properties and biochemical information related to
the human body.

INDEX TERMS
Biomedical engineering, constraint-based modeling, medicine, precision medicine, system biology.

I. INTRODUCTION this variation has not been considered where the kinetic

A LCOHOL is a drug that slows down part of the

brain. Excessive alcohol use may lead to increased
risk of health problems and crimes [1], [2]. Globally, alco-
hol consumption leads to approximately 3.3 million deaths
each year [3]. These casualties have urged researchers
to develop several models to better understand alcohol
metabolism [4]. Alcohol metabolic modeling began in
the 1930s with Wedmark [5] and continues on today
in the form of physiologically based pharmacokinetic
models (PBPK) [6]. These models describe the distri-
bution, metabolism, and elimination of alcohol in the
body. However, the current state-of-the-art models fail to
account for several important factors that have a significant
impact on alcohol clearance and toxicity in a single modeling
framework. For instance, in the majority of the developed
metabolic models, alcohol metabolism has been modeled by
a single reaction in the liver [7], but the research has shown
that ethanol can be metabolized in the other human organs
using different mechanisms [8]–[10]. Moreover, studies have
shown that the variation in alcohol dehydrogenase (ADH)
enzyme activity leads to different elimination rates and
alcohol toxicity [11]. However, in the current PBPK models,
2332-7685
2017 IEEE. Translations and content mining are permitted for academic research only.
Personal use is also permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
VOLUME 2, NO. 4, DECEMBER 2016
parameters associated with the liver have been used [6]. In addi-
tion, experimental evidence has demonstrated that aging
has a significant impact on maximum alcohol concentration
in the body after consumption. In fact, decreased ADH
enzyme activity and decreased liver volume contribute to
having a higher level of alcohol concentration in older
adults [12]. Moreover, although the research has shown that
females with less ADH enzyme activity in the stomach have
a higher maximum alcohol concentration in their body than
males [13], gender difference has not been included in the
developed models as of yet [7]. One last parameter of interest
is to consider the effect of a meal on the blood alcohol
concentration (BAC) as studies have shown that the fed state
decreases the BAC [14].
The lack of a generic computational model that
includes these elements (i.e., different metabolic mecha-
nisms, aging, biochemical variation in enzyme activity, sex
differences, and the meal effect) in one modeling framework
and the growing research in the field of alcohol-related
diseases have motivated us to develop a new mathematical
model for alcohol metabolism. This model is built based on
our previous modeling framework described in [15].
59
 Toroghi et al.: Multiscale Metabolic Modeling Approach for Predicting Blood Alcohol Concentration

TABLE 1. Ethanol PBPK Model Parameters Here, Rs , Vm , and Km are the first-pass metabolism rate
and associated kinetic parameters, respectively; λ is a non-
linear function and is defined as follows:

1 A ≤ 25
λmale = −0.00102A2 + 0.067A − 0.0663 25 < A ≤ 55
A > 55

0.5
(3)

1 A ≤ 25
λfemale = −0.0003A2 + 0.0168A + 0.397 25 < A ≤ 55
A > 55.

0.38
(4)
Here, A is age in years. The model parameters for the
nonlinear function are obtained using spline interpolation
of experimental data presented in [12], indicating the effect
of age on ADH enzyme activity. To capture the oxida-
tive pathway in the human hepatocyte, maximization of
acetate production is considered as an objective function at
the cellular level for the pFBA problem described in [15].
In addition, an empirical model is adapted from [18] to
II. MODEL DEVELOPMENT estimate the liver volume as a function of physiological
The framework in [15] includes the whole body model properties of the human body (i.e., body weight, height, and
with 237 metabolites integrated with a human hepatocyte age). Here, the differential equations describing the alcohol
metabolic network. This framework uses a novel computa- concentration in the whole human body are presented [15].
tional approach for solving the problem using the dynamic Stomach first-pass metabolism
parsimonious flux balance analysis (pFBA) technique.
As known from the literature, alcohol can be metabolized V̇s = −Ks Vs − Rs . (5)
through two major pathways in the human body, namely the Gut compartment
oxidative and non-oxidative pathways. The oxidative mecha-
nism occurs in the liver and stomach, while the non-oxidative Qg Cg
C˙g = (Cart − ). (6)
mechanism takes place in the liver, heart, and pancreas. The Vg Kg
end products of the oxidative and non-oxidative mecha- Kidney compartment
nisms are acetate and fatty acid ethyl ester (FAEE), respec-
tively [9]. To include the non-oxidative pathway in the Qk Ck
C˙k = (Cart − ). (7)
model, kinetic parameters associated with FAEE formation in Vk Kk
the liver, pancreas, and heart are obtained from in vivo stud-
Brain compartment
ies [16], [17] and are used in the Michaelis—Menten kinetic
model as follows: Qb Cb
C˙b = (Cart − ). (8)
Vm,i Ci Vb Kb
Ri = . (1)
Km,i + Ci Heart compartment
Here, Ci represents the alcohol concentrations in the Qh Ch
C˙h = (Cart − ) − RH . (9)
liver, pancreas, and heart. Also, Vm,i and Km,i represent the Vh Kh
associated kinetic parameters.
Pancreas compartment
To incorporate the alcohol metabolism in the stomach, the
kinetic parameters associated with the ADH enzyme in the Qp Cp
stomach for males and females are obtained from an exper- C˙p = (Cart − ) − RP . (10)
Vp Kp
imental study (see Table 1) [12]. In this experiment, gas-
tric ADH activity was measured at high and low ethanol Liver compartment
concentration in 290 individuals of various ages. Based Ql Cl
on the experimental evidence, the activity of ADH in the Ċl = (Cl,input − ) − RL . (11)
Vl Kl
stomach decreases in older adults. In addition, the first-pass
metabolism in the gastric mucosa is lower for female subjects Adipose tissue
due to a decreased activity of the ADH enzyme. To capture Qa Ca
the effect of age and gender on the metabolic rate in the C˙a = (Cart − ). (12)
Va Ka
stomach, we fit a nonlinear function and introduce a modified
kinetic model as follows: Lung compartment
λVm Cs Qlu Clu
Rs = . (2) Ċlu = (Cven − ). (13)
Km + Cs Vlu Klu
60 VOLUME 2, NO. 4, DECEMBER 2016
Toroghi et al.: Multiscale Metabolic Modeling Approach for Predicting Blood Alcohol Concentration
TABLE 2. Details of the Test Subjects. Mean Values ± SD, N = 12 per Age
Group
Adrenal compartment
Qadr Cadr
Ċadr = (Cart − ).
Vadr Kadr
Blood compartment
Qlu Clu
Ċart = ( − Cart ) + Ks Vs .
Vlu Klu
In these equations, Vi , organ volume, and Qi , blood flow
rate to the organ, are physiological properties of the human
body, and Ki , called the tissue-partition coefficient, represents
pharmacokinetic properties of ethanol for the ith compart-
ment. Vs and Ks are the amount of alcohol in the stomach
and the stomach emptying rate, respectively. To incorporate
the variation in alcohol metabolism rate in the liver, dif-
ferent constraints are applied for the reaction related to
the ADH enzyme in hepatocyte metabolic network. The con-
straints are assigned based on the available information for
the ADH enzyme activity in different populations [19]. In the
next section, different simulations are conducted to illustrate
the performance of the developed model. The simulations
are compared with experimental data from different studies
to show the impact of variation in the liver ADH enzyme
activity, sex differences, aging, and the meal effect on alcohol
clearance from the body. In Table 1, ethanol PBPK model
parameters are presented.
III. SIMULATION RESULT
To demonstrate the ability of the model to estimate the BAC, a
set of experimental data from healthy Swedish volun-
teers published by Jones [20] is used. In this experi-
ment, healthy men within four age groups (20–29, 30–39,
40–49, and 50–59 years of age with 12 participants in each
group) drank 0.68g/kg(BW) ethanol as neat whisky after
fasting overnight. The drink was finished within 20 min
and the concentration levels of ethanol in their blood were
collected at 30–60 min intervals for 7 h. Table 2 gives
the details of age, weight, and height of the test subjects.
Using this experimental data, we identified the fitting param-
eter (Vm,ADH for the liver). In Fig. 1, the result of the fit for
each group is presented. As seen, the average BAC can be
described by the model. In addition, the simulation results
indicate that the time course of blood ethanol concentration
is similar in all four groups, with the curve peaking for
50–59-year-old men at the highest level and 20–29-year-old
men at the lowest level, which are consistent with experimen-
tal data.
As part of our model validation, to evaluate the ability of
the model to predict the alcohol concentration in a single indi-
vidual, we use a different set of experimental data published
VOLUME 2, NO. 4, DECEMBER 2016
(14)
FIGURE 1. Average BAC time courses in four age groups of healthy men.
(15) Vm,ADH = 4.5 mM min−1 . Mean absolute error (MAE) based on the data
points. (a) 0.41. (b) 0.36. (c) 0.68. (d) 0.68. Model −. Data ∗.
FIGURE 2. Concentration−time profiles of ethanol in two healthy Swedish
subjects after they drank a dose of 0.80 g/kg body weight on an empty
stomach. MAE based on the data points. (a) 0.93. (b) 1.28. Model −.
Data ∗.
in [21] associated with healthy Swedish subjects. Fig. 2 shows
that the model predictions have relatively good agreement
with data.
In the next simulation, the gender difference in BAC is
demonstrated. To do so, the data from an experimental
study on 17 Italian women 22–48-year-old and 14 Italian
men 26–53-year-old are used [13]. In this experiment, all
subjects were asked to drink 0.3g/kg(BW) ethanol in a fast-
ing state, and their BACs were collected for 300 min. The
average male BAC data are used to identify the fitting
parameter (Vm,ADH for the liver and this parameter is assumed
to be the same for female). Then, to test the model, the
simulation is performed for the women subjects. As seen
in Fig. 3(a), the proposed model is capable of showing the
difference in BAC due to gender.
In the last simulation, we aim to demonstrate the ability of
the model to capture the meal effect on BAC. To do so, a set
of experimental data of the average male BAC from [22] is
used to identify the fitting parameters (i.e., Vm,ADH for the
liver and alcohol stomach emptying rate in the fed state).
In this experiment, 10 healthy Swedish male subjects con-
sumed 0.3g/kg(BW) in two different conditions (i.e., empty
stomach and 1 h after eating a standardized breakfast).
In both conditions, blood samples were collected for 240 min
after the drink started. In Fig. 3(b), the performance of the
model to estimate the average blood concentrations in both
conditions is depicted.
61
 Toroghi et al.: Multiscale Metabolic Modeling Approach for Predicting Blood Alcohol Concentration
FIGURE 3. (a) Gender difference in alcohol concentration. Average
physiological properties. Men: BW = 63 kg, age = 39.5 years, and H = 164
cm. Women: BW = 63 kg, age = 35 years, H = 174 cm, and Vm,ADH = 5.5
mM min−1 . Male MAE = 0.19 and female MAE = 0.37. (b) Meal effect on
alcohol concentration. Average physiological properties: BW = 81.9 kg,
age = 26 years, H = 183 cm, Vm,ADH = 4.5 mM min−1 , fasted condition
MAE = 0.54, and fed condition MAE = 0.2.
FIGURE 4. Meal effect on BAC for two different subjects. Experimental
data are extracted from [22]. (a) Fasted MAE = 0.65 and fed MAE = 0.57.
b) Fasted MAE = 0.95 and fed MAE = 0.36.
To evaluate the ability of the model to describe the BAC in
the fed and fasted conditions for individuals participating in
this study (see [22]), the simulations are performed for two
subjects. As seen in Fig. 4, the model simulation for BAC has
good agreement with the experimental data in both the fed
and fasted states for both individuals.
IV. CONCLUSION
In this letter, we have developed an alcohol metabolic
model for humans based on the dynamic pFBA technique.
The proposed model captures the important elements for
alcohol metabolism in the human body. The simulation
results show that model predictions are consistent with
experimental data. The results from this letter indicate that
the proposed model has several potential applications. For
example, it could provide new opportunities in personalized
medicine to study some metabolic disorders associated with
alcohol metabolism, as the model accounts for biochemi-
cal information associated with each population as well as
the individual and physiological properties of the human
body. In addition, it could be used in forensic science to
predict BACs in an individual. Although the presented model
includes the most important enzymes for alcohol metabolism
in the human body, the impact of other enzymes such
as cytochrome CYP2E1, catalase, and phospholipase D for
oxidative and non-oxidative mechanisms could be added to
the model [23]–[25]. In addition, the effect of other factors
such as disease, stress, and sex hormones in BAC could be
added in future studies to enhance the model’s generality.
62
REFERENCES
[1] H. Fujii and N. Kawada, ‘‘Fibrogenesis in alcoholic liver disease,’’ World
J. Gastroentrol., vol. 20, pp. 8048–8054, Sep. 2014.
[2] R. S. O’Shea, S. Dasarathy, and A. J. McCullough, ‘‘Alcoholic liver
disease,’’ Amer. J. Gastroentrol., vol. 105, pp. 14–32, Sep. 2010.
[3] World Health Organization, ‘‘Global status report on alcohol and health,’’
WHO Library Cataloguing, vol. 978, pp. 1–61, Apr. 2014.
[4] A. W. Jones, ‘‘Evidence-based survey of the elimination rates of ethanol
from blood with applications in forensic casework,’’ Forensic Sci. Int.,
vol. 200, nos. 1–3, pp. 1–20, 2010.
[5] E. Widmark, ‘‘Principles and applications of mediocolegal alcohol deter-
mination,’’ J. Appl. Toxicol., vol. 2, p. 166, Apr. 1982.
[6] M. Plawecki, J.-J. Han, P. Doerschuk, V. Ramchandani, and
S. O’Connor, ‘‘Physiologically based pharmacokinetic (PBPK) models
for ethanol,’’ IEEE Trans. Biomed. Eng., vol. 55, no. 12, pp. 2691–2700,
Dec. 2008.
[7] D. M. Umulis, N. M. Gürmen, P. Singh, and H. S. Fogler, ‘‘A physio-
logically based model for ethanol and acetaldehyde metabolism in human
beings,’’ Alcohol, vol. 35, pp. 3–12, Apr. 2005.
[8] E. Czech and M. Hartleb, ‘‘Non-oxidative metabolism of ethanol and
its influence on the metabolic pathway of serotonin and transferrin,’’
Z Zagadnien Nauk Sadowych, vol. 52, pp. 37–51, Oct. 2002.
[9] S. Zakhari, ‘‘Overview: How is alcohol metabolized by the body?’’ Alcohol
Res. Health, vol. 29, pp. 245–254, Jun. 2006.
[10] C. M. Oneta et al., ‘‘First pass metabolism of ethanol is strik-
ingly influenced by the speed of gastric emptying,’’ Gut, vol. 43,
pp. 612–619, May 1998.
[11] A. Norberg, A. W. Jones, R. G. Hahn, and J. L. Gabrielsson, ‘‘Role of vari-
ability in explaining ethanol pharmacokinetics,’’ Clin. Pharmacokinetics,
vol. 42, no. 1, pp. 1–31, 2003.
[12] H. Seitz et al., ‘‘Human gastric alcohol dehydrogenase activity: Effect of
age, sex, and alcoholism,’’ Gut, vol. 34, pp. 1433–1437, Apr. 1993.
[13] M. Frezza et al., ‘‘Higher blood alcohol levels in women, the
role of decreased gasric alcohol dehygrogenase acitivy and first-pass
metabolism,’’ New England J. Med., vol. 322, pp. 95–99, Apr. 1990.
[14] A. Jones and K. Jönsson, ‘‘Food-induced lowering of blood-ethanol pro-
files and increased rate of elimination immediately after a meal,’’ J. Foren-
sic Sci., vol. 39, no. 4, pp. 1084–1093, 1994.
[15] M. K. Toroghi, W. R. Cluett, and R. Mahadevan, ‘‘A multi-scale model
of the whole human body based on dynamic parsimonious flux balance
analysis,’’ in Proc. 11th IFAC Symp. Dyn. Control Process Syst. Including
Biosyst., Trondheim, Norway, Jun. 2016, pp. 937–942.
[16] P. S. Bora, B. L. Guruge, D. D. Miller, B. R. Chaitman,
M. S. Ruyle, ‘‘Purification and characterization of human heart fatty
acid ethyl ester synthase/carboxylesterase,’’ J. Mol. Cellular Cardiol.,
vol. 28, pp. 2027–2032, Sep. 1996.
[17] D. Riley, E. Kyger, C. Spilburg, L. Lange, ‘‘Purification and characteri-
zation of human pancreatic fatty acid ethyl ester synthase,’’ Biochemistry,
vol. 29, pp. 3848–3852, Oct. 1990.
[18] T. N. Johnson, G. T. Tucker, M. S. Tanner, and A. Rostami-Hodjegan,
‘‘Changes in liver volume from birth to adulthood: A meta-analysis, ’’ Liver
Transplantation, vol. 12, pp. 1481–1493, Sep. 2005.
[19] D. W. Crabb, W. F. Bosron, T. K. Li, ‘‘Ethanol metabolism,’’ Pharmacol.
Therapeutics, vol. 34, pp. 59–73, Apr. 1987.
[20] A. Jones and A. Neri, ‘‘Age-related differences in blood ethanol parametere
and subjective feeling of intoxication in healthy men,’’ Alcohol Alcoholism,
vol. 20, pp. 52–54, Jun. 1985.
[21] A. W. Jones and K. A. Jonsson, ‘‘Between-subject and within-subject
variations in the pharmacokinetics of ethanol,’’ Brit. J. Clin. Pharmacol.,
vol. 37, no. 5, pp. 427–431, 1994.
[22] S. Kechagias, K. A. Jönsson, and A. W. Jones, ‘‘Impact of gastric emptying
on the pharmacokinetics of ethanol as influenced by cisapride,’’ Brit.
J. Clin. Pharmacol., vol. 48, no. 5, pp. 728–732, 1999.
[23] M. Jin, A. Ande, A. Kumar, and S. Kumar, ‘‘Regulation of cytochrome
P450 2e1 expression by ethanol: Role of oxidative stress-mediated
pkc/jnk/sp1 pathway,’’ Cell Death Disease, vol. 4, no. 3, p. e554, 2013.
[24] J. A. Handler and R. G. Thurman, ‘‘Redox interactions between catalase
and alcohol dehydrogenase pathways of ethanol metabolism in the per-
fused rat liver,’’ J. Biol. Chem., vol. 265, pp. 1510–1515, Sep. 1990.
[25] S. Aradottir, G. Asanovska, S. Gjerss, P. Hansson, and C. Alling, ‘‘Phos-
phatidylethanol (PEth) concentrations in blood are correlated to reported
alcohol intake in alcohol-dependent patients,’’ Alcohol Alcoholism, vol. 41,
pp. 431–437, Sep. 2006.
VOLUME 2, NO. 4, DECEMBER 2016