This action might not be possible to undo. Are you sure you want to continue?

# COMM-3248; No.

of Pages 14

ARTICLE IN PRESS

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e x x x ( 2 0 1 1 ) xxx–xxx

journal homepage: www.intl.elsevierhealth.com/journals/cmpb

Identiﬁcation of an integrated mathematical model of standard oral glucose tolerance test for characterization of insulin potentiation in health

Roberto Burattini ∗ , Micaela Morettini

Department of Information Engineering, Polytechnic University of Marche, Via Brecce Bianche, 60131 Ancona, Italy

a r t i c l e

Article history:

i n f o

a b s t r a c t

Two new formulations, respectively denominated INT M1 and INT M2, of an integrated mathematical model to describe the glycemic and insulinemic responses to a 75 g oral glucose tolerance test (OGTT) are proposed and compared. The INT M1 assumes a single compartment for the intestine and the derivative of a power exponential function for the gastric emptying rate, while, in the INT M2, a nonlinear three-compartment system model is adopted to produce a more realistic, multiphase gastric emptying rate. Both models were implemented in a Matlab-based, two-step procedure for estimation of seven adjustable coefﬁcients characterizing the gastric emptying rate and the incretin, insulin and glucose kinetics. Model behaviour was tested vs. mean plasma glucagon-like peptide 1 (GLP1), glucose-dependent insulinotropic polypeptide (GIP), glucose and insulin measurements from two different laboratories, where glycemic proﬁles observed during a 75 g OGTT were matched in healthy subjects (HC1- and HC2-group, respectively) by means of an isoglycemic intravenous glucose (I-IVG) infusion. Under the hypothesis of an additive effect of GLP-1 and GIP on insulin potentiation, our results demonstrated a substantial equivalence of the two models in matching the data. Model parameter estimates showed to be suitable markers of differences observed in the OGTT and matched I-IVG responses from the HC1-group compared to the HC2-group. Model implementation in our two-step parameter estimation procedure enhances the possibility of a prospective application for individualization of the incretin effect in a single subject, when his/her data are plugged in. © 2011 Elsevier Ireland Ltd. All rights reserved.

Received 10 December 2010 Received in revised form 14 May 2011 Accepted 4 July 2011 Keywords: Glucose-insulin system Gastric emptying Oral glucose absorption GIP GLP-1 Incretin effect Parameter estimation

1.

Introduction

Mathematical modelling in the assessment of insulin–glucose interactions has a longstanding tradition, and reported models show a wide degree of complexity depending on their purpose. Besides models of minimal complexity, some of which are referred to as “minimal models”, after Bergman et al. [1], increasing relevance is being assumed by integrated simu-

lation models of the glucose-insulin control system during an oral glucose tolerance test (OGTT) or meal glucose tolerance test (MGTT) to improve knowledge of diabetes pathophysiology and assess the efﬁcacy of hypoglycemic agents in clinical drug development [2–6]. In this context, a key issue concerns the modelling of glucose transit through the gastrointestinal tract, to describe glucose absorption [7–9] and the related “incretin effect” [8,10–19], mostly due to the contribution of glucagon-like peptide-1 (GLP-1) and glucose-dependent

Corresponding author. Tel.: +39 071 2204458; fax: +39 071 2204224. E-mail address: r.burattini@univpm.it (R. Burattini). 0169-2607/$ – see front matter © 2011 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.cmpb.2011.07.002

Please cite this article in press as: R. Burattini, M. Morettini, Identiﬁcation of an integrated mathematical model of standard oral glucose tolerance test for characterization of insulin potentiation in health, Comput. Methods Programs Biomed. (2011), doi:10.1016/j.cmpb.2011.07.002

∗

COMM-3248; No. of Pages 14

ARTICLE IN PRESS

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e x x x ( 2 0 1 1 ) xxx–xxx

2

insulinotropic polypeptide (GIP) released in response to nutrient ingestion, which stimulate oral glucose dependent insulin secretion. A promising simulation model of the oral glucose tolerance test, primarily intended to illustrate the importance of incretin within the normal ranges observed clinically in humans, has recently been proposed by Brubaker et al. [4]. Unfortunately, a simple empiric description of the rate of gastric emptying of ingested glucose into the gut, Gempt , constrains the model to the reproduction of glycemia and insulinemia responses to 50 and 100 g oral glucose loads, and does not allow an effective model testing versus experimental data measured during clinical OGTT protocols, which generally consist of the administration of a standard 75 g glucose dose. In its original formulation [4] the model was characterized by ﬁfteen parameters, ten of which were given numerical values known a priori from previously reported measurements, thus leaving ﬁve “adjustable” parameters. However, the lack of a validation against standard OGTT data limits the reliability of the values assumed for these “adjustable” parameters, thus affecting the predictive capabilities of the model. Based on these considerations, the aim of the present study was to improve the model formulation by incorporating a mathematical description of oral glucose absorption that allows model application to standard OGTT data, such that “adjustable” model parameters can be better assessed by ﬁtting to clinical data. Two competing integrated mathematical models of oral glucose tolerance test, denominated INT M1 and INT M2, respectively, were obtained by incorporating two alternative representations of glucose absorption. The former model incorporates a single compartment for the intestine and the derivative of a power exponential function for the gastric emptying rate [7]. In the latter, a nonlinear threecompartment system model is adopted to produce a more realistic multiphase gastric emptying rate [9]. A comparative analysis of INT M1 and INT M2 behaviour was performed in terms of their ability to reproduce the augmented glucosedependent plasma insulin concentration (generally referred to as incretin-induced insulin potentiation) observed after an OGTT, as it compares with the insulin response to an intravenous glucose infusion given in amount sufﬁcient to match the proﬁle of glucose concentration observed during OGTT (isoglycemic intravenous glucose, I-IVG, infusion) [15,18,19]. Especially, mean data of OGTT and I-IVG responses from two groups of metabolically healthy subjects (here denominated HC1- and HC2-group), respectively reported by Muscelli et al. [18] and Nauck et al. [15], were used to test our models capability to reproduce and interpret differences in essential aspects of insulin potentiation in health observed in the data sets from the two different laboratories.

2.

2.1.

Methods

Model formulation

The model previously proposed by Brubaker et al. [4] to simulate the glycemic and insulinemic responses to a 50 and a 100 g oral glucose load, with explicit incorporation of incretin action, was used as a basis to build-up an integrated model

upon an OGTT protocol consisting of oral administration of a standard 75 g oral glucose load (i.e. 1 g of glucose per kg body weight, BW, administered at time t = 0 min, and standardized to a 75 kg individual). To this aim the description of the oral glucose absorption was modiﬁed by incorporating either a one-compartment model [7], denoted as M1, or a three-compartment model [9], denoted as M2. The two versions of our integrated model, the ﬂow diagram of which is depicted in Fig. 1, were denoted as INT M1 when incorporating the M1, and INT M2 when incorporating the M2, respectively. The overall mathematical description is given in the Appendix A. Description, basal values and units for both the INT M1 and the INT M2 variables are given in Table 1. In the absence of measurements of basal hepatic glucose balance, HGBb , the value of 0.77 mmol min−1 was deduced from the literature by multiplying the BW of 75 kg times 10.25 × 10−3 mmol min−1 kg−1 determined as the mean of two hepatic glucose balance values per unit BW reported by Bell et al. [20] (and summarized at points 3 and 4 of Table A.1 in [21]) for humans with mean glycemia of 5.28 mmol L−1 and mean insulinemia of 10 mU L−1 , which are consistent with the mean fasting values of glycemia and insulinemia characterizing our HC1-group (5.4 mmol L−1 and 10.2 mU L−1 , respectively) and HC2-group (5.5 mmol L−1 and 7.6 mU L−1 , respectively). Besides the basal values of variables as given in Table 1, and considered that HGBb is treated as a ﬁxed parameter, the INT M1 formulation consists of ﬁfteen further independent parameters; namely, k3 , k4 , k5 , k6 , k7 , k8 , k9 , p, V, M, , kabs , f, ke and ˇ. Instead, the INT M2 is characterized by seventeen parameters, because in this model the ke and ˇ characteristic parameters of the monophase waveshape for the rate of gastric emptying (Gempt ; Eq. (A1)) are replaced by the c, b, kmin , and kmax coefﬁcients of the emptying function kempt (qsto ) of Eq. (A10). Quantiﬁcation of all these parameters was accomplished as follows. The k3 , k4 , k6 , k9 , p and V were assumed as ﬁxed and were given numerical values (Table 2) known a priori from observations reported in the literature, as discussed by Brubaker et al. [4]. The kabs , f, b and c parameters pertaining to glucose absorption were also assumed as ﬁxed (Table 2). In accordance with the validation studies on glucose-absorption models reported by Dalla Man et al. [9,22], the value of 0.22 min−1 was assumed for the rate constant, kabs , of intestinal glucose absorption, while the value of 0.90 was assumed for the fraction, f, of the ingested glucose dose, D, that is actually absorbed. Eventually, the b and c dimensionless coefﬁcients were given the values of 0.85 and 0.25, respectively, thus ﬁxing the percentage of the glucose dose, D, for which the emptying function kempt (Eq. (A10)) decreases (b) and subsequently increases (c) at (kmax − kmin )/2 [9]. The remaining independent model parameters, denoted as k5 , k7 , k8 , M and , which were assumed as “adjustable” in the original model of Brubaker et al. [4] and presumably manually adjusted, were considered as free parameters in addition to the ke and ˇ in the INT M1 model formulation, and in addition to kmin and kmax in the INT M2 model formulation. One novel aspect of the present work is that these free parameters were automatically estimated by a two step procedure as described in Section 2.3.

Please cite this article in press as: R. Burattini, M. Morettini, Identiﬁcation of an integrated mathematical model of standard oral glucose tolerance test for characterization of insulin potentiation in health, Comput. Methods Programs Biomed. (2011), doi:10.1016/j.cmpb.2011.07.002

COMM-3248; No. of Pages 14

ARTICLE IN PRESS

3

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e x x x ( 2 0 1 1 ) xxx–xxx

Fig. 1 – Flow diagram of our integrated model incorporating either one of two alternative formulations of glucose absorption, denoted as M1 and M2. The M1 representation assumes the derivative of a power exponential function for the gastric emptying rate (Gempt ) of the oral glucose dose (D), and a single compartment for the gut (qgut ). In the M2 conﬁguration, the oral glucose dose enters a two compartment model (qsto1 and qsto2 ) of the stomach, while a gastric emptying function (kempt ), dependent on the total amount of glucose in the stomach (qsto , equal to the sum of qsto1 and qsto2 ), plays a key role in determining the time course of both the rate of emptying of oral glucose load into the gut (Gempt ) and the glucose quantity in the gut compartment (qgut ). In both our integrated model formulations (denominated INT M1 and INT M2, respectively), the Gempt is transmuted by the k5 coefﬁcient into a signal that enters the compartment of plasma incretin (INC). Similarly, qgut is multiplied by the kabs rate constant of intestinal absorption and the f fraction of the intestinal absorption that actually appears into plasma, to predict the RaG rate of glucose absorption from the gut into the mesenteric circulation. This feeds the glucose compartment (G). Solid lines with arrows indicate directional ﬂows. Dashed lines with arrows indicate control actions of incretin (INC) and glucose (G) on post-hepatic insulin secretion, as well as control actions of insulin (I) on hepatic glucose balance (HGB) and glucose delivery to the peripheral tissues.

Please cite this article in press as: R. Burattini, M. Morettini, Identiﬁcation of an integrated mathematical model of standard oral glucose tolerance test for characterization of insulin potentiation in health, Comput. Methods Programs Biomed. (2011), doi:10.1016/j.cmpb.2011.07.002

COMM-3248; No. of Pages 14

ARTICLE IN PRESS

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e x x x ( 2 0 1 1 ) xxx–xxx

4

**Table 1 – Description, basal values and units for model variables. Constant
**

qsto1b qsto2b qgutb Gb Ib INCb HGBb Gemptb RaGb

Description

Amount of glucose in the compartment 1 of stomach. Amount of glucose in the compartment 2 of stomach. Glucose mass in the intestine. Plasma glucose concentration. Plasma insulin concentration. Plasma incretin concentration. Hepatic glucose balance. Rate of emptying of glucose load into the gut. Rate of appearance of glucose in the peripheral circulation.

Value

0 0 0 Fasting measurement Fasting measurement Fasting measurement 0.77 0 0

Units

mmol mmol mmol mmol L−1 mU L−1 ng L−1 mmol min−1 mmol min−1 mmol min−1

In accordance with Dalla Man et al. [9], to favour numerical identiﬁability of the INT M2 model, the constraint k21 = kmax was imposed. After INT M1 and INT M2 model parameter estimation was accomplished, the remaining dependent parameters ε1 , ε2 , RaINCb , , k1 , and k2 were computed by Eqs. (A11), (A12), (A15), (A17), (A20) and (A21), respectively.

2.2.

Model testing vs. reported experimental data

Our INT M1 and INT M2 model outputs were tested against data of plasma glucagon-like peptide-1 (GLP-1), glucosedependent insulinotropic polypeptide (GIP), glucose and insulin concentration, averaged over two groups of metabolically healthy subjects, here denominated HC1- and HC2group, respectively reported by Muscelli et al. [18] and Nauck et al. [15]. Essentially, the subjects involved in these previous studies underwent two different glucose-challenge protocols. On the ﬁrst occasion (OGTT protocol) the subject ingested 75 g of glucose, whereas on the second occasion (I-IVG protocol) an intravenous glucose infusion was given in amount sufﬁcient to match the proﬁle of glucose concentration observed during OGTT [15,18]. An augmented glucose-dependent insulin secretion (insulin potentiation), observed in these circumstances after OGTT, compared to the matched I-IVG infusion, is attributed to the inﬂuence of the so called “incretin effect” mostly due to the gut-derived incretin hormones, GLP-1 and

GIP, which enhance glucose-dependent insulin secretion by binding to speciﬁc receptors on the -cell [23–25]. These hormones are released during glucose or meal intake in proportion to nutrient transport across the intestinal epithelium [10], their effect seems to be additive, and they stimulate insulin secretion both at fasting and postprandial plasma glucose level [14]. On this basis, the mean levels of GLP-1 and GIP (expressed in pmol L−1 ) reported in [18] for the HC1-group, and in [15] for the HC2-group, were combined as follows to deﬁne an INC(t) signal for each group. First, the reported mean levels of GLP-1 and GIP, expressed in pmol L−1 , were converted into ng L−1 (conversion factors were: 1 pmol L−1 = 3.30 ng L−1 for the GLP-1, and 1 pmol L−1 = 4.98 ng L−1 for the GIP) and, then, for each group the INC(t) signal was determined as the sum of related GLP-1 and GIP. Eventually, this INC(t) signal was used together with the glycemia, G(t), and insulinemia, I(t), data, and the related Gb , Ib and INCb , fasting values, for model parameter estimation by the procedure described in Section 2.3.

2.3.

Parameter estimation procedure

As explained in Section 2.1, the INT M1 and the INT M2 are characterized by seven free parameters, i.e., k5 , k7 , k8 , M, , ˇ and ke , the former, and k5 , k7 , k8 , M, , kmin and kmax, the latter. To deﬁne an optimal value for each one of these parameters, a two-step parameter estimation procedure was set-up as follows.

**Table 2 – Description, values and units for ﬁxed INT M1 and INT M2 model parameters. Coefﬁcient
**

kabs f c b p k3 k4 k6 k9 V

Description

Rate constant of intestinal absorption. Fraction of the intestinal absorption which actually appears into plasma. Glucose dose percentage corresponding to the ﬁrst ﬂex point of the kempt (qsto ) function (Eq. (A10)). Glucose dose percentage corresponding to the second ﬂex point of the kempt (qsto ) function (Eq. (A10)). Ratio of non-insulin mediated to insulin mediated glucose uptake (NIMGU/IMGU; Eq. (A19)). Slope of renal glucose clearance. Intercept of renal glucose clearance. Measure of degradation/clearance of incretin. Measure of degradation/clearance of insulin. Volume of distribution.

Value

0.22 0.90 0.25 0.85 2 0.0718 0.717 0.1 0.1 15

Units

min−1

L min−1 mmol min−1 min−1 min−1 L

Source: Sources of the values given to p, k3 , k4 , k6 , k9 and V are found in Brubaker et al. [4]. Values of kabs , f, c and b are taken from Dalla Man et al. [9,22].

Please cite this article in press as: R. Burattini, M. Morettini, Identiﬁcation of an integrated mathematical model of standard oral glucose tolerance test for characterization of insulin potentiation in health, Comput. Methods Programs Biomed. (2011), doi:10.1016/j.cmpb.2011.07.002

COMM-3248; No. of Pages 14

ARTICLE IN PRESS

5

2.4. Insulin potentiation

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e x x x ( 2 0 1 1 ) xxx–xxx

The ﬁrst step was to estimate the k7 parameter by running either the INT M1 and the INT M2 to ﬁt mean plasma insulin concentrations from the I-IVG protocol. To this aim, Gempt (and, with it, the dynamic component of plasma incretin inﬂow in Eq. (A14)) was set to zero, while the incretin level was maintained at the basal value, INCb , and the glucose proﬁle matched to that measured during OGTT in the HC1- and HC2-group, respectively, was used as model input instead of the description by Eq. (A18). Under these conditions, glucose absorption is not involved, so that k7 (Eq. (A16)) is the only one free parameter that remains to be estimated from ﬁtting the model predicted I(t) to I-IVG insulin data. In a subsequent step, the k7 estimate was ﬁlled into the INT M1 and the INT M2, which were run to generate incretin, glycemia and insulinemia responses to a 75 g OGTT, to be simultaneously ﬁtted to the OGTT data (corresponding to each I-IVG), in order to estimate the six remaining unknown parameters (k5 , k8 , M, , ke and ˇ) of the INT M1, as well as the six remaining unknown parameters (k5 , k8 , M, , kmin , and kmax ) of the INT M2. All model formulations were implemented in Matlab–Simulink environment. Data ﬁt and parameter estimation were accomplished by means of a weighted least squares (WLS) procedure [26]. The following sum of square error (SSE) expression was used to ﬁt the model predicted I-IVG insulin output at the i-th instant, Im (ti ), to the corresponding insulin measurement, I(ti ):

In accordance with Nauck et al. [11], the quantity of -cell secretory response evoked by factors other than glucose itself (insulin potentiation) was deduced from the difference in integrated incremental responses (over basal) of insulinemia between OGTT and matched I-IVG protocols. This insulin potentiation, IP, can be quantiﬁed as the percentage of the OGTT response by the equation:

IP% =

AUCIOGTT − AUCII-IVG × 100 AUCIOGTT

(3)

where AUCIOGTT and AUCII-IVG represent the area under the curve of incremental insulin concentration over the OGTT and the matched I-IVG duration, respectively, according to the trapezoidal rule.

3.

Results

N

SSEI-IVG =

ì=1

Im (ti ) − I(ti )

Ii

2

(1)

The following SSE expression was subsequently used for simultaneous ﬁt of model predicted OGTT incretin, INCm (ti ), glycemia, Gm (ti ), and insulinemia, Im (ti ), outputs at the i-th instant, to the corresponding incretin, INC(ti ), glycemia, G(ti ), and insulinemia, I(ti ), measurements:

N

SSEOGTT =

ì=1

INCm (ti ) − INC(ti )

INCi 2

2

+

Gm (ti ) − G(ti )

Gi

2

+

Im (ti ) − I(ti )

Ii

(2)

The weights, INCi , Gi , Ii , were assumed equal to 5.5%, 1.5% and 4% of INC(ti ), G(ti ) and I(ti ), respectively, under the assumption of normal distribution with zero mean for the measurement errors [18,26]. The goodness of ﬁt was evaluated by calculation of the root-mean-square errors, RMSEI-IVG = SSEI-IVG /N and RMSEOGTT = SSEOGTT /3N Precision of all parameter estimates was expressed as percent coefﬁcient of variation: CV(pi )% = SDpi /pi × 100, where pi is the i-th component of the model parameters vector and SDpi is the standard deviation of pi , which is calculated as the square root of the diagonal terms of the inverse of the Fisher information matrix.

Mean fasting values of Gb , Ib , and INCb were 5.4 mmol L−1 , 10.2 mU L−1 , and 145 ng L−1 , respectively, for the HC1-group, and 5.5 mmol L−1 , 7.6 mU L−1 , and 90 ng L−1 , respectively, for the HC2-group. Fitting the INT M1 and INT M2 outputs (dash-dot line in Fig. 2C and F) to mean insulinemia data from OGTTmatched I-IVG protocols, reported for the HC1-group and HC2-group (open circles in Fig. 2C and F, respectively), yielded the RMSEI-IVG and the estimates of k7 given in Table 3 for the former model and in Table 4 for the latter. After feeding the INT M1 and the INT M2 with the k7 values obtained from the previous step, the INC(t), G(t) and I(t) responses to a 75 g OGTT were generated and ﬁtted to incretin, glycemia and insulinemia data from OGTT protocols corresponding to each I-IVG, thus estimating the k5 , k8 , M, , ke and ˇ values reported in Table 3 for the INT M1, and the k5 , k8 , M, , kmin and kmax values reported in Table 4 for the INT M2. Eventually, computation of ε1 (Eq. (A11)), ε2 (Eq. (A12)), RaINCb (Eq. (A15)), (Eq. (A17)), k1 (Eq. (A20)) and k2 (Eq. (A21)) yielded the values given in Table 5. Fig. 2 displays the quality of data ﬁt obtained from applying the INT M1 (dashed line) and the INT M2 (solid line) to mean OGTT data from the HC1-group (panels A–C), and the HC2-group (panels D–F). Values of RMSEOGTT are reported in Table 3 for the INT M1 and in Table 4 for the INT M2. Fig. 3 displays the time course of the Gempt and RaG proﬁles predicted by the INT M1 (dashed line) and the INT M2 (solid line) for the HC1-group (panels A and B, respectively) and the HC2-group (panels C and D, respectively). The grey area in panels B and D represents the range of variability of RaG , measured with the multiple tracer, tracer-to-tracee clamp technique during OGTT as reported by Dalla Man et al. [9]. Values of the insulin potentiation index, IP%, computed by Eq. (3) from OGTT and matched I-IVG insulinemia outputs produced by the INT M1 and INT M2 models are compared in Table 6 with the corresponding IP% values computed from mean insulinemia data of the HC1- and HC2-group.

Please cite this article in press as: R. Burattini, M. Morettini, Identiﬁcation of an integrated mathematical model of standard oral glucose tolerance test for characterization of insulin potentiation in health, Comput. Methods Programs Biomed. (2011), doi:10.1016/j.cmpb.2011.07.002

COMM-3248; No. of Pages 14

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e x x x ( 2 0 1 1 ) xxx–xxx

6

A

500

HC1

D

500 400 300 200 100

HC2

INC(t) (ng·L-1)

300 200 100 0 30 60 90 120 150 180 210 240

INC(t) (ng·L-1)

400

0

30

60

90

120 150 180 210 240

Time (min)

Time (min)

B

G(t) (mmol·L-1)

12 11 10 9 8 7 6 5 4 0 30 60 90 120 150 180 210 240

E

G(t) (mmol·L-1)

12 11 10 9 8 7 6 5 4 0 30 60 90 120 150 180 210 240

Time (min)

Time (min)

C

I(t) (mU·L-1)

120 100

F

120 100

I(t) (mU·L-1)

80 60 40 20 0 0 30 60 90 120 150 180 210 240

80 60 40 20 0 0 30 60 90 120 150 180 210 240

Time (min)

Time (min)

Fig. 2 – Mean data of plasma incretin (closed triangles in panels A and D), glucose (closed squares in panels B and E) and insulin (closed circles in panels C and F) concentrations in response to 75 g oral glucose challenge in the HC1-group (left hand column) and the HC2-group (right hand column) are matched by the proﬁles of INC(t), G(t), and I(t) outputs of the INT M1 (dashed lines) and the INT M2 (solid lines) models after free parameter optimization. Open circles in panels C and F are mean plasma insulin concentration data from the HC1- and HC2-group, respectively, measured after the glycemic proﬁles observed following glucose ingestion (OGTT) were matched by means of an isoglycemic intravenous glucose (I-IVG) infusion (open squares in panels B and E). The dash-dot line in panels C and F describes the best ﬁtting I-IVG insulin output provided by both the INT M1 and the INT M2 model.

4.

Discussion

This work yields an improved formulation of a mathematical model previously proposed by Brubaker et al. [4] to simulate the glycemia and insulinemia responses to 50 and 100 g oral glucose administration, by explicitly incorporating the incretin effect. The incretin response, INC(t), is thought to be a direct effect of the rate of emptying of ingested glucose into the gut, Gempt (t), through a zero-order transfer function, with k5 gain, between the Gempt and the dynamic component

of plasma incretin inﬂow (Eq. (A14)). An improved description of intestinal glucose absorption in response to 75 g OGTT, was accomplished here by replacing the previously proposed empirical description, restricted to 50 and 100 g glucose challenge [4], with either a one-compartment model (M1 in Fig. 1) and a three-compartment model (M2 in Fig. 1), thus giving rise to two integrated models denominated INT-M1 and INT-M2, respectively. Besides the k3 , k4 , k6 , k9 , p, and V parameters (Table 2), which were ﬁxed at numerical values known a priori from

COMM-3248; No. of Pages 14

ARTICLE IN PRESS

7

HC1

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e x x x ( 2 0 1 1 ) xxx–xxx

A

Gempt(t) (mmol·min-1)

6 5 4 3 2 1 0 0 30 60 90

C

Gempt(t) (mmol·min-1)

6 5 4 3 2 1 0

HC2

120

150

180

210

240

0

30

60

90

120

150

180

210

240

Time (min)

Time (min)

B

RaG(t) (mmol·min-1)

6 5 4 3 2 1 0 -1 0 30 60 90 120 150 180 210 240

D

RaG(t) (mmol·min-1)

6 5 4 3 2 1 0 -1 0 30 60 90 120 150 180 210 240

Time (min)

Time (min)

Fig. 3 – The proﬁles of the rate of emptying of oral glucose load into the gut, Gempt (t), and the rate of glucose absorption from the gut into the mesenteric circulation, RaG (t), for the HC1-group (Panel A and B, respectively) and the HC2-group (Panel C and D, respectively) as predicted by INT M1 (dashed line) and the INT M2 (solid line). The grey area in panels B and D represents the range of variability of RaG , measured with the multiple tracer, tracer-to-tracee clamp technique during OGTT as reported by Dalla Man et al. [9].

previously reported measurements, as discussed by Brubaker et al. [4], the basal hepatic glucose balance, HGBb , and the four more glucose absorption parameters, kabs , c, b and f, were ﬁxed at numerical values taken from the literature as described in Section 2.1. Consequently, seven free parameters (k5 , k7 , k8 , M, , ke and ˇ) characterize the INT M1 (Table 3), as well as seven are the free parameters (k5 , k7 , k8 , M, , kmin , and kmax ) that characterize the INT M2 (Table 4). Among these, the ﬁve parameters denominated k5 , k7 , k8 , M and are the same as those deﬁned “adjustable” by Brubaker et al. [4], and presumably manually adjusted in their study for qualitative representation of OGTT responses. The remaining two parameters, i.e. ke and ˇ for the INT M1 and kmin and kmax for the INT M2, are free parameters that allow individualization of the rate of ingested glucose absorption. Granted the suitability of all ﬁxed parameters, a novel aspect of the present study was the set-up of a two-step procedure that allows estimation of the free parameters of our INT M1 and INT M2 model formulations by ﬁtting to incretin, glycemia and insulinemia data taken from previously published studies [15,18], where glycemic proﬁles observed in two groups, (HC1 and HC2) of metabolically healthy subjects, during an OGTT, were used to quantify incretin-induced insulin potentiation by comparing the insulin response to OGTT with that obtained from an isoglycemic intravenous glucose (I-IVG) infusion.

According to our two-step estimation procedure, the k7 parameter was ﬁrst estimated from ﬁtting to I-IVG insulinemia data. To this aim the Gempt (and, with it, the dynamic component of plasma incretin inﬂow in Eq. (A14)) was set to zero and the plasma glucose compartment was fed with the glucose proﬁle matched to that measured during OGTT (isoglycemic proﬁle). Under these conditions the alternative formulations (M1 and M2) of glucose absorption do not come into play. This explains why both the INT M1 and the INT M2 produce the same k7 estimates and related RMSEI-IVG values, as given in Tables 3 and 4, respectively. The higher k7 estimate of 0.648 mU min−1 mmol−1.3 L−0.3 for the HC1-group, compared to the estimate of 0.158 mU min−1 mmol−1.3 L−0.3 for the HC2-group provides quantitative information on an enhanced insulin response to intravenous glucose infusion in the former group. This is consistent with the experimental ﬁnding of a higher value of the ratio (AUCII-IVG /AUCGI-IVG ) between the area under the curve of I-IVG suprabasal insulin to the area under the curve of the corresponding suprabasal proﬁle of glucose infused intravenously in the HC1-group (ratio of 14.3 mU mmol−1 from data of Fig. 2B and C) compared to that of the HC2-group (ratio of 5.7 mU mmol−1 from data of Fig. 2E and F). Once the estimated k7 values were ﬁlled into the INT M1 and the INT M2, these two models produced a comparable

8

COMM-3248; No. of Pages 14

Table 3 – Estimates of INT M1 model parameters in the presence of HGBb equal to 0.77 mmol min−1 . k7 (CV%) mU min−1 mmol−1.3 L−0.3

HC1 0.648 (2.2%) HC2 0.158 (2.3%)

**k5 (CV%) ng L−1 mmol−1
**

6.77 (5.3%) 8.77 (2.7%)

**k8 (CV%) mU min−1 ng−1
**

0.027 (4.7%) 0.017 (3.4%)

**(CV%) mmol mU−1
**

0 (–) 0.058 (4.9%)

**M (CV%) L2 mU−1 min−1
**

0.0024 (24%) 0.0034 (5.7%)

**ke (CV%) min−1
**

0.0095 (6.9%) 0.0155 (1.1%)

ˇ (CV%)

1.0 (2.6%) 1.52 (0.8%)

**RMSEI-IVG RMSEOGTT
**

2.96 4.16 2.21 c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e x x x ( 2 0 1 1 ) xxx–xxx 3.89

HC1, group of 11 subjects with normal glucose tolerance (NGT) from Muscelli et al. [18]; HC2, group of 10 metabolically healthy subjects from Nauck et al. [15]; k7 , coefﬁcient of appearance of insulin due to glucose, in the absence of incretin effect; k5 , gain of the zero-order transfer function between the rate of gastric emptying, Gempt , and the dynamic plasma incretin component; k8 , coefﬁcient of appearance of insulin due to incretin; , shaping factor for derivative control of insulin on glucose; M, coefﬁcient accounting for the effects of counter-regulatory factors on liver; ke and ˇ, fractional transfer coefﬁcient and dimensionless shape factor of gastric emptying, respectively; CV%, precision of all parameter estimates expressed as percent coefﬁcient of variation; RMSEI-IVG , root-mean-square error of I-IVG data ﬁt, RMSEOGTT , root-mean-square error of OGTT data ﬁt.

ARTICLE IN PRESS

Table 4 – Estimates of INT M2 model parameters in the presence of HGBb equal to 0.77 mmol min−1 . k7 (CV%) k5 (CV%) mU min−1 mmol−1.3 L−0.3 ng L−1 mmol−1

HC1 HC2 0.648 (2.2%) 0.158 (2.3%) 5.78 (3.9%) 8.82 (2.7%)

**k8 (CV%) mU min−1 ng−1
**

0.025 (4.9%) 0.017 (3.3%)

**(CV%) mmol mU−1
**

0.0028 (84%) 0.0605 (4.5%)

**M (CV%) L2 mU−1 min−1
**

0.0071 (5.9%) 0.0033 (5.1%)

**kmin (CV%) min−1
**

0.017 (2.6%) 0.027 (1.8%)

**kmax (CV%) min−1
**

0.038 (2.7%) 0.038 (1.4%)

**RMSEI-IVG RMSEOGTT
**

2.96 4.16 3.90 4.07

HC1, group of 11 subjects with normal glucose tolerance (NGT) from Muscelli et al. [18]; HC2, group of 10 metabolically healthy subjects from Nauck et al. [15]; k7 , coefﬁcient of appearance of insulin due to glucose, in the absence of incretin effect; k5 , gain of the zero-order transfer function between the rate of gastric emptying, Gempt , and the dynamic plasma incretin component; k8 , coefﬁcient of appearance of insulin due to incretin; , shaping factor for derivative control of insulin on glucose; M, coefﬁcient accounting for the effects of counter-regulatory factors on liver; kmin , minimum of the kempt (qsto ) function; kmax , maximum of kempt (qsto ) function; CV%, precision of all parameter estimates expressed as percent coefﬁcient of variation; RMSEI-IVG , root-mean-square error of I-IVG data ﬁt, RMSEOGTT , root-mean-square error of OGTT data ﬁt.

COMM-3248; No. of Pages 14

ARTICLE IN PRESS

9

HC1, group of 11 subjects with normal glucose tolerance (NGT) from Muscelli et al. [18]; HC2, group of 10 metabolically healthy subjects from Nauck et al. [15]; ε1 and ε2 , respectively, quantify the rate of decrease and subsequent increase of the gastric emptying function (Eqs. (A11) and (A12), respectively); RaINCb , basal rate of appearance of incretin in the peripheral circulation (Eq. (A15)); , effects of additional regulators of I(t) on insulin appearance (Eq. (A17)); k1 and k2 are coefﬁcients of linear glucose mediated and insulin mediated glucose uptake, respectively (Eqs. (A20) and (A21), respectively).

Table 5 – Computed values of INT M1 and INT M2 model parameters.

ﬁt to incretin (closed triangles in Fig. 2A and D), glycemia (closed squares in Fig. 2B and E) and insulinemia (closed circles in Fig. 2C and F), as judged from the dashed-line proﬁles for the INT M1 and the solid-line proﬁles for the INT M2, and the related RMSEOGTT reported in Tables 3 and 4, respectively. Due to their ability to approximate both the OGTT and the matched I-IVG insulin response, the two models yielded estimates of the IP% index of insulin potentiation (Eq. (3)) consistent with the corresponding values computed from measured data in the HC1- and HC2-group (Table 6). The augmented insulin secretion observed after oral glucose load, compared with intravenous glucose infusion, at similar plasma glucose concentration, is attributed to the inﬂuence of the incretin effect, most of which is due to the GLP-1 and GIP secretion [8,14,16,18,23–25]. As the contribution of intestinal hormones to oral glucose-stimulated insulin secretion is relevant, and being the major stimulus to the release of these hormones clearly related to the rate of ingested glucose delivery to the gut, Gempt (t), the INC(t) signal described in Eq. (A14) by the product k5 × Gempt (t) was derived from the GLP-1 and GIP responses to OGTT reported in [18] for the HC1-group and in [15] for the HC2group, respectively, by hypothesizing an additive effect as described in Section 2.2. With this assumption, the closedtriangle data of Fig. 2A and D were obtained. The lower k5 estimates provided by both the INT M1 (Table 3) and the INT M2 (Table 4) for the HC1-group, compared to those provided for the HC2-group, reﬂect the experimental observation of a lower area under the suprabasal curve of incretin in the HC1-group (25.6 × 103 ng L−1 over 180 min, Fig. 2A) compared to that of the HC2-group (40.0 × 103 ng L−1 over 180 min, Fig. 2D) in response to the 75 g oral glucose challenge. On this basis, the k5 free parameter appears a suitable marker of the amplitude of the incretin response to oral glucose administration. In spite of lower incretin response, the suprabasal insulinemia response of the HC1-group, quantiﬁed by the AUCIOGTT value of 9.03 × 103 mU L−1 over 180 min, is higher than that, 7.70 × 103 mU L−1 over 180 min, observed in the HC2-group (compare Fig. 2C and F). This implies the presence, in the HC1group, of a compensatory increase of the insulin response to the incretin, which is reﬂected in the increased k8 values estimated in the same group by the INT M1 (Table 3) and the INT M2 (Table 4), compared to those of the HC2-group. In the presence of this compensation, the enhanced insulin potentiation (higher IP% index; Table 6) in the HC2-group, compared to the HC1-group, is mainly explained by the observed reduction

k2 mmol mU−1 min−1

k1 min−1

HC2 HC1 HC2 mU L−1 min−1 HC2 RaINCb ng min−1 HC1 HC2 ε2 mmol−1 HC1 HC2 HC1

– 0.024

– 0.040

– 0.024

217 217

135 135

−8.66 −8.41

−2.22 −2.21

0.0064 0.0064

0.0062 0.0062

0.0017 0.0017

HC1

0.0022 0.0022

HC2

**Table 6 – Insulin potentiation. IP% INT M1
**

HC1 HC2 63.4 81.0

ε1 mmol−1

– 0.040

HC1

INT M2

62.9 81.1

EXP

63.0 78.1

HC1, group of 11 subjects with normal glucose tolerance (NGT) from Muscelli et al. [18]; HC2, group of 10 metabolically healthy subjects from Nauck et al. [15]; IP%, insulin potentiation computed by Eq. (3).

INT M1 INT M2

10

COMM-3248; No. of Pages 14

Table 7 – Estimates of INT M1 model parameters in the presence of HGBb equal to 0.92 mmol min−1 . k5 (CV%) k7 (CV%) mU min−1 mmol−1.3 L−0.3 ng L−1 mmol−1

HC1 HC2 0.648 (2.2%) 0.158 (2.3%) 6.86 (5.7%) 8.85 (1.3%)

**k8 (CV%) mU min−1 ng−1
**

0.026 (4.8%) 0.017 (1.2%)

**(CV%) mmol mU−1
**

0 (–) 0.0489 (5.0%)

**M (CV%) L2 mU−1 min−1
**

0 (–) 0.0017 (14%)

**ke (CV%) min−1
**

0.0093 (7.8%) 0.0156 (1.0%)

ˇ (CV%)

1.0 (2.9%) 1.47 (0.9%)

RMSEI-IVG

2.96 4.16

RMSEOGTT

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e x x x ( 2 0 1 1 ) xxx–xxx 2.09 4.0

See legend of Table 3 for meaning of symbols.

ARTICLE IN PRESS

Table 8 – Estimates of INT M2 model parameters in the presence of HGBb equal to 0.92 mmol min−1 . k7 (CV%) k5 (CV%) mU min−1 mmol−1.3 L−0.3 ng L−1 mmol−1

HC1 HC2 0.648 (2.2%) 0.158 (2.3%) 5.95 (3.8%) 8.93 (2.7%)

**k8 (CV%) mU min−1 ng−1
**

0.025 (4.8%) 0.016 (3.3%)

**(CV%) mmol mU−1
**

0.0032 (65%) 0.0566 (4.9%)

**M (CV%) L2 mU−1 min−1
**

0.0043 (9.1%) 0.0021 (8.0%)

**kmin (CV%) min−1
**

0.016 (2.4%) 0.026 (1.8%)

**kmax (CV%) min−1
**

0.037 (2.5%) 0.037 (1.4%)

**RMSEI-IVG RMSEOGTT
**

2.96 4.16 3.92 4.14

See legend of Table 4 for meaning of symbols.

COMM-3248; No. of Pages 14

ARTICLE IN PRESS

11

4.1. Monophase vs. multiphase waveshape of gastric emptying of glucose

In conceptual terms, our INT M1 model differs from the INT M2 in that the former incorporates the hypothesis of a monophasic waveshape for the rate of gastric emptying of ingested glucose into the gut, (Gempt ; Eq. (A1) and dashed line in Fig. 3A and C), characterized by the two identiﬁable parameters, ke and ˇ (Table 3). Instead, the INT M2 conﬁguration assumes a three-compartment model of glucose absorption, with a nonlinear gastric emptying function (kempt ; Eq. (A10)), that yields a multiphase Gempt (solid line in Fig. 3A and C). Unfortunately, the kempt is characterized by four independent parameters (b, c, kmin and kmax ) a couple of which needs to be ﬁxed to favour numerical identiﬁability of the INT M2 model. Our choice was, then, to ﬁx the b and c values (Table 2), which respectively locate the ﬂex points of kempt in correspondence of the percentage of the glucose dose, D, for which the emptying function kempt (Eq. (A10)) decreases and subsequently increases at (kmax − kmin )/2 [9]. The need of ﬁxing two of the four characteristic parameters of the kempt function yields a limitation in the INT M2 ability to match the data, such that no relevant beneﬁt is seen in the RMSEOGTT , compared to the INT M1 output.

of suprabasal insulin response to the OGTT-matched proﬁle of glucose infused intravenously, as marked by a lower k7 in the HC2-group. In spite of a similar behaviour of the INT M1 and INT-M2 models in accounting for variations of k5 , k7 and k8 estimates between the HC1- and the HC2-group of healthy subjects, a relevant diversity is seen in that, in the HC1-group, the INT M1 model estimated a zero value for the and a value of M 66% lower than that produced by the INT M2. No model independent information is available to judge the reliability of estimated values for , since the small derivative × dI/dt term was empirically introduced by Brubaker et al. [4] to sharpen the peak in glucose level following entry of glucose into the circulation, and was set equal to 0.06 when RaG > 0. This value is comparable with that estimated here for the HC2-group by both the INT M1 and the INT M2 (Tables 3 and 4). The zero-value estimated for by the INT M1 in the HC1 group (Table 3), as well as the corresponding very low estimate (with 84% estimation error) provided by the INT M2 (Table 4) might be ascribed to a vanishing role of the × dI/dt in the presence of a limited glycemia dynamics as observed in the HC1-group compared to the HC2-group (Fig. 2B with E). Concerning the M parameter, which modulates the bilinear glucose-insulin control on hepatic balance, it is likely that its estimate is affected by the value of HGBb (Eq. (A13)) assumed as ﬁxed to favour model identiﬁability. Because the value of 10.25 × 10−3 mmol min−1 kg−1 assumed here for basal hepatic glucose balance per unit BW (yielding HGBb equal to 0.77 mmol min−1 for a standardized 75 kg individual), differs from the value of 12.22 × 10−3 mmol min−1 kg−1 (yielding HGBb equal to 0.92 mmol min−1 ) previously assumed by Brubaker et al. [4], the effect of the latter HGBb value on the estimates of our free model parameters was tested. Results are given in Table 7 for the INT M1 and in Table 8 for the INT M2 model. Comparison with Tables 3 and 4, respectively, conﬁrms that the HGBb has a major effect on the M parameter estimation. In particular, the M values of Tables 7 and 8 are signiﬁcantly underestimated with respect to the corresponding values of Tables 3 and 4. The zerovalue of M produced by the INT M1 for the HC1 group (Table 7) implies a stationary unidirectional HGB ﬂux (Eq. (A13)), which is unlikely to occur under normal conditions [27–29]. For this reason the parameter estimates obtained from our HGBb assumption (Tables 3 and 4) are more physiologically consistent. The lack of quantitative information on the HGB dynamics in humans does not allow assessment of the reliability of absolute values of M when different from zero. In describing the glucose kinetics by Eq. (A18), a linear non-insulin mediated glucose uptake, k1 × G(t), was assumed here, rather than the k1 × G(t)1.3 power term originally assumed by Brubaker et al. [4]. Under our simplifying assumption, which is consistent with that characterizing minimal models of glucose kinetics [1], the estimated values for the k1 coefﬁcient (Table 5) increased in both our HC1 and HC2 groups by about 65%, compared to the k1 values computed in the presence of the k1 × G(t)1.3 power term. Practically no effect was observed in the RMSEOGTT of data ﬁt and in the estimates of free model parameters.

4.2.

Summary and conclusions

An improvement over a mathematical model previously proposed by Brubaker et al. [4] to simulate the glycemia and insulinemia responses to an OGTT, was accomplished here by replacing the previously proposed empirical description of glucose absorption, restricted to 50 and 100 g glucose challenge, with either the M1 and the M2 model conﬁgurations depicted in Fig. 1 and mathematically described in the Appendix A. This new arrangement, gave rise to our INT M1 and INT M2 integrated models, respectively, that allow a reliable description of glycemic and insulinemic responses to a clinically consistent 75 g OGTT. Under the hypothesis of an additive effect of GLP-1 and GIP on insulin potentiation, our results showed a substantial equivalence of the two alternative INT M1 and INT M2 models in reproducing insulin potentiation as observed by comparing mean OGTT responses to a 75 g oral glucose challenge with matched I-IVG responses, as reported in [18] and in [15] for two different groups of healthy subjects, here denominated HC1 and HC2, respectively. Especially, the implementation of both the INT M1 and INT M2 models in our two-step parameter estimation procedure allowed assessment of k5 , k7 and k8 parameters, which appear as suitable markers of the differences in the incretin, glucose and insulin responses between the two groups. Namely, k5 is a marker of the amplitude of the incretin response to oral glucose administration; k8 is a marker of the insulin response to the incretin, while k7 is a marker of the insulin response irrespective of incretin effect. A higher k8 value estimated by both the INT M1 and the INT M2 in the HC1-group, compared to the HC2, indicates the presence, in the former group of an increase of insulin response to incretin, which compensates for a lower incretin response to oral glucose load (lower k5 in the HC1). In the presence of this compensation, the enhanced insulin potentiation

COMM-3248; No. of Pages 14

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e x x x ( 2 0 1 1 ) xxx–xxx

12

(higher IP% index; Table 6) in the HC2-group, compared to the HC1-group, is mainly explained by the observed reduction of suprabasal insulin response to the OGTT-matched proﬁle of glucose infused intravenously, as marked by a lower k7 in the HC2-group. Owing to the substantial equivalence of INT M1 and INTM2 behaviour, the former model appears the best compromise between simplicity and predictive capability. Model implementation in our two-step parameter estimation procedure enhances the possibility of a prospective application for individualization of the incretin effect in a single subject, when his/her data are plugged in.

with the initial conditions being the basal values qsto1b , qsto2b and qgutb reported in Table 1. In these equations ı(t) is the impulse function, D (mmol) is the amount of ingested glucose, k21 (min−1 ) is the rate of grinding, Gempt (mmol min−1 ) is the rate of emptying of the oral glucose load into the gut, RaG (mmol min−1 ) is the rate of glucose absorption from the gut into the mesenteric circulation; kabs (min−1 ) is the rate constant of intestinal absorption, f is the fraction of the intestinal absorption which actually appears in the plasma, and kempt (min−1 ) is a gastric emptying function, which depends on the total amount of glucose in the stomach, qsto (mmol), as follows [9]: kempt (qsto ) = kmin + kmax − kmin × {tanh[ε1 × (qsto − b × D)] 2 (A10)

**Appendix A. Mathematical formulation of our 75 g OGTT integrated models
**

A.1. M1 model of oral glucose absorption

This model describes glucose absorption by the gut as shown in the M1 panel of Fig. 1. Under the assumption that, after ingestion of a glucose dose, D (75 g, equivalent to 417 mmol), the fraction of glucose in the duodenum follows a power exponential function increase, the rate of emptying of the oral glucose load into the gut, Gempt (mmol min−1 ), is described by the following equation [7,9]: Gempt (t) = D × ˇ × ke × tˇ−1 × e[−(ke t)

ˇ

ˇ

− tanh[ε2 × (qsto − c × D)] + 2}

As reported by Dalla Man et al. [9], this model is suitable to describe glucose absorption during an OGTT under the assumption of kempt = kmax at qsto = D (full stomach) and at qsto = 0 (empty stomach). This assumption yields the following expressions for the ε1 and ε2 parameters: ε1 = 5 2 × D × (1 − b) 5 2×D×c (A11)

]

(A1)

ε2 =

(A12)

where ke (min−1 ) is a fractional transfer coefﬁcient and ˇ a dimensionless shape factor. Denoting the amount of glucose in the gut as qgut (mmol), the rate constant of intestinal absorption as kabs (min−1 ), the fraction of the intestinal absorption which actually appears in plasma as f, and the rate of appearance of ingested glucose into plasma as RaG (mmol min−1 ), the following equations hold [7,9]: ˙ qgut (t) = −kabs × qgut (t) + Gempt (t) RaG (t) = f × kabs × qgut (t) (A2) (A3)

**A.3. Hepatic glucose balance
**

Hepatic glucose balance (HGB; mmol min−1 ) represents the net ﬂux of glucose G (mmol L−1 ) across the hepatic bed, thereby reﬂecting the sum of glucose production and glucose uptake from the mesenteric circulation (e.g., following the oral glucose load). In accordance with Brubaker et al. [4] the following equation is assumed to describe the HGB(t) dynamics in healthy conditions: HGB(t) = HGBb + M × [Gb − G(t)] × I(t) (A13)

**A.2. M2 model of oral glucose absorption
**

The representation of glucose absorption displayed in the M2 panel of Fig. 1 assumes two compartments (qsto1 and qsto2 , respectively) for the stomach, and a single compartment, qgut , for the intestine. Mathematical description is as follows [9]: ˙ qsto1 (t) = −k21 × qsto1 (t) + D × ı(t) ˙ qsto2 (t) = −kempt (qsto ) × qsto2 (t) + k21 × qsto1 (t) ˙ qgut (t) = −kabs × qgut (t) + kempt (qsto ) × qsto2 (t) qsto (t) = qsto1 (t) + qsto2 (t) Gempt (t) = kempt (qsto ) × qsto2 (t) RaG (t) = f × kabs × qgut (t) (A4) (A5) (A6) (A7) (A8) (A9)

In this equation, HGBb is basal hepatic glucose balance, Gb is basal glucose concentration, while G(t) and I(t) represent the time course of glucose and insulin concentrations. M is a modulating factor in the bilinear control term.

A.4. Incretin response

The following differential equation was assumed to describe the kinetics of the incretin [4]: RaINCb dINC(t) = + k5 × Gempt (t) − k6 × INC(t) dt V (A14)

According to this equation, three terms determine the rate of variation with time of circulating incretin concentrations (INC), based on the combined contribution of glucagonlike peptide-1 (GLP-1) and glucose-dependent insulinotropic polypeptide (GIP) [8,14,16,18,23–25]. The ﬁrst term is the basal rate of appearance of the incretin (RaINCb ; ng min−1 ),

COMM-3248; No. of Pages 14

ARTICLE IN PRESS

13

to insulin mediated glucose uptake, and assuming a value of 2 for basal conditions [31,32], the following relation holds: p= k1 × Gb =2 k2 × Ib (A19)

normalized to distribution volume (V); the second term is proportional, through the k5 constant (ng L−1 mmol−1 ), to the proﬁle of the rate of gastric emptying into the gut (Gempt ; mmol min−1 ); and, eventually the third term accounts for a negative contribution, proportional through the k6 (min−1 ) to the time course of the incretin (INC; ng L−1 ) due to the removal of bioactive GIP and GLP-1 from the circulation [30]. The quantity RaINCb = k6 × V × INCb (A15)

From Eq. (A18) written for steady state conditions (dG/dt = 0 and dI/dt = 0) and Eq. (A19), the values of the constants k1 (min−1 ) and k2 (mmol min−1 mU−1 ) are given by the following equations [4]: k1 = HGBb p × p+1 Gb × V 1 HGBb × p+1 Ib × V (A20)

is obtained from Eq. (A14) by setting the derivative equal to zero, which occurs when Gempt equals zero.

k2 =

(A21)

A.5. Insulin response

The following differential equation was assumed to describe insulin kinetics [4]: dI(t) 1.3 = k7 × G(t) + k8 × INC(t) − k9 × I(t) + dt As reported by Brubaker et al. [4], the Eq. (A18) works well under normoglycemic conditions. To account for urinary glucose loss during hyperglycemia (G > 10 mmol L−1 ), an additional term is added as shown in the following equation: RaG (t) HGB(t) dG(t) = + − k1 × G(t) − k2 × I(t) + dt V V × dI(t) k3 × G(t) − k4 − dt V (A22)

(A16)

Eq. (A16) is based on the assumption that four terms determine the rate of variation with time of circulating insulin. The ﬁrst nonlinear term, assumed equal to k7 × G(t)1.3 , accelerates the effect of plasma glucose on insulin release, with respect to a linear term; the second and the third terms are proportional, through the k8 (mU min−1 ng−1 ) and k9 (min−1 ) constants, respectively, to the circulating concentrations of the incretin and insulin; and eventually, the constant term, (mU L−1 min−1 ), which is expressed, in steady state, in terms of the known basal values of G and I, after setting the derivative term, dI/dt, equal to zero and solving for : = k9 × Ib − k8 × INCb − k7 × G1.3 b (A17)

references

**A.6. Plasma glucose kinetics
**

The following differential equation was assumed to describe the glucose kinetics [4]: RaG (t) HGB(t) dG(t) = + − k1 × G(t) − k2 × I(t) + dt V V dI(t) dt (A18)

×

According to this equation, the rate of change of glucose concentration during an OGTT depends on glucose absorption from the gut into the mesenteric circulation (RaG (t)/V); net absorption/production of glucose by the liver (HGB(t)/V); insulin mediated glucose uptake, k2 × I(t); and a small corrective term, × dI(t)/dt, which has the effect of sharpening the peak in glucose level following entry of glucose into the circulation. A linear, k1 × G(t), non-insulin mediated glucose uptake was assumed here, rather than the k1 × G(t)1.3 power term originally assumed by Brubaker et al. [4]. Under these assumptions, denoting as p the ratio of non-insulin mediated

[1] R.N. Bergman, Y.Z. Ider, C.R. Bowden, C. Cobelli, Quantitative estimation of insulin sensitivity , Am. J. Physiol. 236 (1979) E667–E677. [2] K. Thomaseth, A. Kautzky-Willer, B. Ludvik, R. Prager, G. Pacini, Integrated mathematical model to assess beta-cell activity during the oral glucose test , Am. J. Physiol. 270 (1996) E522–E531. [3] P.G. Fabietti, V. Canonico, M.O. Federici, M.M. Benedetti, E. Sarti, Control oriented model of insulin and glucose dynamics in type 1 diabetics , Med. Biol. Eng. Comput. 44 (2006) 69–78. [4] P.L. Brubaker, E.L. Ohayon, L.M. D’Alessandro, K.H. Norwich, A mathematical model of the oral glucose tolerance test illustrating the effects of the incretin , Ann. Biomed. Eng. 35 (2007) 1286–1300. [5] C. Dalla Man, R.A. Rizza, C. Cobelli, Meal simulation model of the glucose-insulin system , IEEE Trans. Biomed. Eng. 54 (2007) 1740–1749. [6] H.E. Silber, N. Frey, M.O. Karlsson, An integrated glucose-insulin model to describe oral glucose tolerance test data in healthy volunteers , J. Clin. Pharmacol. 50 (2010) 246–256. [7] J.D. Elashoff, T.J. Reedy, J.H. Meyer, Analysis of gastric emptying data , Gastroenterology 83 (1982) 1306–1312. [8] J. Schirra, M. Katschinski, C. Weidmann, T. Schäfer, U. Wank, R. Arnold, B. Göke, Gastric emptying and release of incretin hormones after glucose ingestion in humans , J. Clin. Invest. 97 (1996) 92–103. [9] C. Dalla Man, M. Camilleri, C. Cobelli, A system model of oral glucose absorption: validation on gold standard data , IEEE Trans. Biomed. Eng. 53 (2006) 2472–2478. [10] E. Ferrannini, E. Barrett, S. Bevilacqua, J. Dupre, R.A. DeFronzo, Sodium elevates the plasma glucose response to

COMM-3248; No. of Pages 14

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e x x x ( 2 0 1 1 ) xxx–xxx

14

[11]

[12]

[13]

[14]

[15]

[16]

[17] [18]

[19]

[20]

glucose ingestion in man , J. Clin. Endocrinol. Metab. 54 (1982) 455–458. M.A. Nauck, E. Homberger, E.G. Siegel, R.C. Allen, R.P. Eaton, R. Ebert, W. Creutzfeldt, Incretin effects of increasing glucose loads in man calculated from venous insulin and C-peptide responses , J. Clin. Endocrinol. Metab. 63 (1986) 492–498. E.T. Shapiro, H. Tillil, M.A. Miller, B.H. Frank, J.A. Galloway, A.H. Rubenstein, K.S. Polonsky, Insulin secretion and clearance. Comparison after oral and intravenous glucose , Diabetes 36 (1987) 1365–1371. L.T. Shuster, V.L. Go, R.A. Rizza, P.C. O’Brien, F.J. Service, Incretin effect due to increased secretion and decreased clearance of insulin in normal humans , Diabetes 37 (1988) 200–203. T. Vilsbøll, T. Krarup, S. Madsbad, J.J. Holst, Both GLP-1 and GIP are insulinotropic at basal and postprandial glucose levels and contribute nearly equally to the incretin effect of a meal in healthy subjects , Regul. Pept. 114 (2003) 115–121. M.A. Nauck, A. El-Ouaghlidi, B. Gabrys, K. Hücking, J.J. Holst, C.F. Deacon, B. Gallwitz, W.E. Schmidt, J.J. Meier, Secretion of incretin hormones (GIP and GLP-1) and incretin effect after oral glucose in ﬁrst-degree relatives of patients with type 2 diabetes , Regul. Pept. 122 (2004) 209–217. J.J. Meier, M.A. Nauck, D. Kranz, J.J. Holst, C.F. Deacon, D. Gaeckler, W.E. Schmidt, B. Gallwitz, Secretion, degradation, and elimination of glucagon-like peptide 1 and gastric inhibitory polypeptide in patients with chronic renal insufﬁciency and healthy control subjects , Diabetes 53 (2004) 654–662. W. Creutzfeldt, The [pre-] history of the incretin concept , Regul. Pept. 128 (2005) 87–91. E. Muscelli, A. Mari, A. Natali, B.D. Astiarraga, S. Camastra, S. Frascerra, J.J. Holst, E. Ferrannini, Impact of incretin hormones on -cell function in subjects with normal or impaired glucose tolerance , Am. J. Physiol. Endocrinol. Metab. 291 (2006) E1144–E1150. M. Campioni, G. Toffolo, L.T. Shuster, F.J. Service, R.A. Rizza, C. Cobelli, Incretin effect potentiates beta-cell responsivity to glucose as well as to its rate of change: OGTT and matched intravenous study , Am. J. Physiol. Endocrinol. Metab. 292 (2007) E54–E60. P.M. Bell, R.G. Firth, R.A. Rizza, Effects of hyperglicemia on glucose production and utilization in humans , Diabetes 35 (1986) 642–648.

[21] T. Arleth, S. Andreassen, M.O. Federici, M.M. Benedetti, A model of the endogenous glucose balance incorporating the characteristics of glucose transporters , Comput. Methods Programs Biomed. 62 (2000) 219–234. [22] C. Dalla Man, A. Caumo, R. Basu, R. Rizza, G. Toffolo, C. Cobelli, Minimal model estimation of glucose absorption and insulin sensitivity from oral test: validation with a tracer method , Am. J. Physiol. Endocrinol. Metab. 287 (2004) E637–E643. [23] J.J. Holst, Glucagonlike peptide 1: a newly discovered gastrointestinal hormone , Gastroenterology 107 (1994) 1848–1855. [24] P.E. MacDonald, W. El-Kholy, M.J. Riedel, A.M. Salapatek, P.E. Light, M.B. Wheeler, The multiple actions of GLP-1 on the process of glucose-stimulated insulin secretion , Diabetes 51 (2002) S434–S442. [25] P.E. MacDonald, A.M. Salapatek, M.B. Wheeler, Glucagon-like peptide-1 receptor activation antagonizes voltage-dependent repolarizing K(+) currents in beta-cells: a possible glucose-dependent insulinotropic mechanism , Diabetes 51 (2002) S443–S447. [26] E.R. Carson, C. Cobelli, L. Finkelstein, The Mathematical Modeling of Metabolic and Endocrine Systems , Wiley, New York, NY, 1983. [27] E. Ferrannini, O. Bjorkman, G.A. Reichard Jr., A. Pilo, M. Olsson, J. Wahren, R.A. DeFronzo, The disposal of an oral glucose load in healthy subjects. A quantitative study , Diabetes 34 (1985) 580–588. [28] E. Ferrannini, D.C. Simonson, L.D. Katz, G. Reichard Jr., S. Bevilacqua, E.J. Barrett, M. Olsson, R.A. DeFronzo, The disposal of an oral glucose load in patients with non-insulin-dependent diabetes , Metabolism 37 (1988) 79–85. [29] A.D. Cherrington, Control of glucose uptake and release by the liver in vivo , Diabetes 48 (1999) 1198–1214. [30] D.J. Drucker, Therapeutic potential of dipeptidyl peptidase IV inhibitors for the treatment of type 2 diabetes , Expert. Opin. Investig. Drugs 12 (2003) 87–100. [31] E. Ferrannini, J.D. Smith, C. Cobelli, G. Toffolo, A. Pilo, R.A. DeFronzo, Effect of insulin on the distribution and disposition of glucose in man , J. Clin. Invest. 76 (1985) 357–364. [32] A.D. Baron, G. Brechtel, P. Wallace, S.V. Edelman, Rate and tissues sites of non-insulin- and insulin-mediated glucose uptake in humans , Am. J. Physiol. 255 (1988) E769– E774.