Am J Physiol Endocrinol Metab 294: E261–E270, 2008.
First published November 14, 2007; doi:10.1152/ajpendo.00676.2007.
Comparison between surrogate indexes of insulin sensitivity and resistance
and hyperinsulinemic euglycemic clamp estimates in mice
Sihoon Lee,1* Ranganath Muniyappa,1* Xu Yan,2 Hui Chen,1 Lilly Q. Yue,2 Eun-Gyoung Hong,3
Jason K. Kim,3 and Michael J. Quon1
1
Diabetes Unit, National Center for Complementary and Alternative Medicine, National Institutes of Health, Bethesda,
Maryland; 2Division of Biostatistics, Center for Devices and Radiological Health, United States Food and Drug
Administration, Rockville, Maryland; 3Department of Cellular and Molecular Physiology, Pennsylvania State University
College of Medicine, Hershey, Pennsylvania
Submitted 22 October 2007; accepted in final form 9 November 2007
calibration model; quantitative insulin-sensitivity check index
(e.g., high-fat diet and high-sucrose diet; Refs. 8, 30). Using
these models to develop and evaluate novel therapeutic agents
for the treatment and/or prevention of insulin resistance and its
associated diseases is an important application. In particular,
mouse models of insulin resistance are very useful in preclinical evaluation during drug development. Therefore, the ability
to accurately and efficiently quantify insulin sensitivity and
resistance in large numbers of mice is of great interest. The
hyperinsulinemic euglycemic clamp (“glucose clamp”) technique is widely accepted as the reference standard for directly
determining metabolic insulin sensitivity in both humans and
animals. However, feasibility issues in applying the glucose
clamp procedure to humans preclude its use in epidemiological
studies and large clinical investigations. Therefore, a number
of simple surrogate indexes of insulin sensitivity and resistance
have been developed, validated, and effectively utilized (some
more than others; Refs. 3, 5, 17, 20, 21, 23). Some published
animal studies (13, 18, 22, 26, 27, 32) have also used these
surrogate indexes that were originally developed for humans.
However, human and mouse physiology differs and no direct
comparisons or validation of these surrogate indexes against
the reference standard glucose clamp has been performed in
mice. Thus, in the present study, we compare linear correlations between several simple surrogate indexes [quantitative
insulin-sensitivity check index (QUICKI), homeostasis model
assessment (HOMA), 1/HOMA, log(HOMA), and 1/fasting
insulin] and glucose clamp determinations in a cohort of mice
with a wide range of insulin sensitivity and resistance. In
addition, we perform calibration model analysis to evaluate the
predictive accuracy of these surrogate indexes relative to the
reference standard glucose clamp.
METHODS
to the pathophysiology of diabetes and is a hallmark of obesity, metabolic syndrome, and
many cardiovascular diseases (9, 28). To gain insight into
pathophysiological mechanisms underlying insulin resistance,
obesity, and diabetes, many mouse models have been extensively and productively investigated. Genetic manipulations of
mice have generated numerous models of insulin resistance,
diabetes, obesity, and cardiovascular complications (7, 14, 15).
Many mouse models have also been used to study acquired
insulin resistance from various environmental manipulations
Description of mouse models used. For the present study, we used
glucose clamp data from a total of 87 mice with normal, decreased, or
increased insulin sensitivity. Only data from mice where complete information was available to calculate both clamp-derived indexes of insulin
sensitivity and surrogate indexes of insulin sensitivity were used. Consequently, data from 23 out of an original cohort of 110 mice were not
included in this study for final analyses. C57BL/6 mice with normal
insulin sensitivity (n ⫽ 10) were used as controls for insulin-resistant
mice on a high-fat diet for 3 wk (n ⫽ 10), 6 wk (n ⫽ 6), or 3 mo (n ⫽
10) (24). Wild-type FVB mice with normal insulin sensitivity (n ⫽ 6)
* S. Lee and R. Muniyappa contributed equally to this work.
Address for reprint requests and other correspondence: M. J. Quon, Diabetes
Unit, NCCAM, NIH, 9 Memorial Drive, Building 9, Room 1N-105 MSC 0920,
Bethesda, MD 20892-0920 (e-mail: [email protected]).
The costs of publication of this article were defrayed in part by the payment
of page charges. The article must therefore be hereby marked “advertisement”
in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
INSULIN RESISTANCE CONTRIBUTES
http://www.ajpendo.org
0193-1849/08 $8.00 Copyright © 2008 the American Physiological Society
E261
Downloaded from http://ajpendo.physiology.org/ by 10.220.33.2 on June 15, 2017
Lee S, Muniyappa R, Yan X, Chen H, Yue LQ, Hong E-G, Kim
JK, Quon MJ. Comparison between surrogate indexes of insulin sensitivity and resistance and hyperinsulinemic euglycemic clamp estimates in
mice. Am J Physiol Endocrinol Metab 294: E261–E270, 2008. First
published November 14, 2007; doi:10.1152/ajpendo.00676.2007.—Insulin resistance contributes to the pathophysiology of diabetes, obesity,
and their cardiovascular complications. Mouse models of these human
diseases are useful for gaining insight into pathophysiological mechanisms. The reference standard for measuring insulin sensitivity in
both humans and animals is the euglycemic glucose clamp. Many
studies have compared surrogate indexes of insulin sensitivity and
resistance with glucose clamp estimates in humans. However, regulation of metabolic physiology in humans and rodents differs and
comparisons between surrogate indexes and the glucose clamp have
not been directly evaluated in rodents previously. Therefore, in the
present study, we compared glucose clamp-derived measures of insulin sensitivity (GIR and SIClamp) with surrogate indexes, including
quantitative insulin-sensitivity check index (QUICKI), homeostasis
model assessment (HOMA), 1/HOMA, log(HOMA), and 1/fasting
insulin, using data from 87 mice with a wide range of insulin
sensitivities. We evaluated simple linear correlations and performed
calibration model analyses to evaluate the predictive accuracy of each
surrogate. All surrogate indexes tested were modestly correlated with
both GIR and SIClamp. However, a stronger correlation between body
weight per se and both GIR and SIClamp was noted. Calibration
analyses of surrogate indexes adjusted for body weight demonstrated
improved predictive accuracy for GIR [e.g., R ⫽ 0.68, for QUICKI
and log(HOMA)]. We conclude that linear correlations of surrogate
indexes with clamp data and predictive accuracy of surrogate indexes
in mice are not as substantial as in humans. This may reflect intrinsic
differences between human and rodent physiology as well as increased technical difficulties in performing glucose clamps in mice.
E262
INSULIN SENSITIVITY INDEXES AND GLUCOSE CLAMP ESTIMATES
AJP-Endocrinol Metab • VOL
a robust, direct, and data-intensive protocol) was very small relative to
measurement error of simple surrogates determined from single fasting
measurements (e.g., QUICKI) and measurement error of body weight.
Therefore, to simplify the analysis, we neglected the measurement error
for GIR (or SIClamp) in our calibration model.
For each surrogate index, two types of predicted residuals were
considered. The first type of residual is the difference between
measured GIR (or SIClamp) (xi for the ith subject) and fitted GIR (or
^
SIClamp; ^xi ⫽ ␣^ ⫹ ␤^1Si ⫹ ␤^2Wi ⫹ ␤3Si 䡠 Wi). That is, the residual ei ⫽
^
xi ⫺ xi, where xi is derived from the calibration model with all subjects
included in the estimation of model parameters ␣, ␤1, ␤2, and ␤3. The
second type of residual considered is a cross-validation type predicted
residual e(i) ⫽ xi ⫺ ^x(i), where xi is still measured GIR (or SIClamp),
but x^(i) is predicted SIClamp from the calibration model that excludes the ith subject. The subscript (i) means “with the ith subject
deleted.” From these two types of residuals, criterion functions
were used to evaluate prediction accuracy: square root of the mean
squared error of prediction: RMSE ⫽ [¥ e2i /n⫺4]1/ 2 and leave-oneout cross-validation-type root mean squared error of prediction:
2
CVPE ⫽ [¥ e(i)
/n]1/ 2 . Smaller values of RMSE and CVPE indicate
better predictive power. RMSE is likely to underestimate prediction
errors while CVPE is more robust.
Statistical analysis. Pearson correlation coefficients (R) and respective P values were calculated to assess the statistical significance of
the model using a linear least-squares fit method to obtain the linear
regression. To compare predictive accuracy of QUICKI (adjusted for
weight) and other surrogates (adjusted for weight) in terms of CVPE
and RMSE, we performed hypothesis testing with the one-sided
alternative hypothesis that weight-adjusted QUICKI had a smaller
RMSE or CVPE than another weight-adjusted surrogate using a
Bootstrap percentile method with 60,000 replications performed for
each comparison (11). The bootstrap method is appropriate because
the RMSEs (or CVPEs) corresponding to QUICKI and other surrogates were derived from the same group of subjects and thus correlated.
The P values calculated from comparison of RMSE and CVPE were for
pair-wise comparisons. For example, when weight-adjusted QUICKI and
weight-adjusted HOMA were compared with respect to CVPE based on
87 mice, a bootstrap percentile method with 60,000 replications was used
to get a sample of 60,000 differences in CVPE [CVPE (HOMA) ⫺
CVPE (QUICKI)], and then a P value for one-sided superiority testing
was estimated as the proportion of the bootstrap replications less than
zero. One-sided hypothesis testing was used because multiple previous
studies in humans have demonstrated the superiority of QUICKI as a
surrogate index of insulin sensitivity from a bvariety of perspectives
supporting an a priori expectation. P ⬍ 0.05 was considered to
indicate statistical significance. The software used for statistical analysis and the random calibration model was SAS System V9.
RESULTS
Mice. Our original cohort consisted of 110 mice. However,
to do complete analyses for both GIR and SIClamp, 23 out of
these 110 mice were excluded because of incomplete data sets.
Metabolic characteristics of the 87 mice analyzed in our study
under fasting and steady-state glucose clamp conditions are
shown in Table 1. Note that appropriate control mice corresponding to each experimental group have been combined
together for a total of 32 control mice. Groups of mice with
decreased insulin sensitivity include MCK-hPTP1B transgenic
mice (36), MCK-hLAR transgenic mice (36), L-SACC1
transgenic mice (25), and mice fed with high-fat diet for 3 wk,
6 wk, or 3 mo, respectively (24). Groups of mice with increased insulin sensitivity include Par-1b null mice (16),
Ptpn6me-v/me-v mice (10), and CASA/Rk mice (33).
Determinations of insulin sensitivity and resistance. We
used both the steady-state GIR as well as SIClamp to obtain
294 • FEBRUARY 2008 •
www.ajpendo.org
Downloaded from http://ajpendo.physiology.org/ by 10.220.33.2 on June 15, 2017
were littermate controls for insulin-resistant transgenic mice overexpressing the protein-tyrosine phosphatases PTP1B or LAR in skeletal
muscle (MCK-hPTP1B, n ⫽ 7; MCK-hLAR, n ⫽ 7; Ref. 36).
Wild-type C57BL/6 ⫻ FVB mice with normal insulin sensitivity (n ⫽
4) were littermate controls for insulin-resistant transgenic mice overexpressing a dominant-negative mutant of CEACAM1 in liver (LSACC1, n ⫽ 4; Ref. 25). Wild-type C57BL/6 mice with normal
insulin sensitivity (n ⫽ 4) were littermate controls for Par-1b null
mice with increased insulin sensitivity (Par-1b KO, n ⫽ 3; Ref. 16).
Wild-type C57BL/6J mice with normal insulin sensitivity (n ⫽ 5)
were littermate controls for mice with functionally deficient SHP-1
protein that have increased insulin sensitivity (Ptpn6me-v/me-v, n ⫽ 6;
Ref. 10). Wild-type C57BL/6J mice with normal insulin sensitivity
(n ⫽ 3) were used as controls for CASA/Rk mice with increased
insulin sensitivity (n ⫽ 2; Ref. 33).
Hyperinsulinemic euglycemic clamp. Glucose clamp procedures in
conscious unanesthetized mice were performed as described previously
(10, 16, 24, 25, 36). In brief, mice were fasted overnight (⬃15 h) and a
2-h hyperinsulinemic euglycemic clamp procedure was conducted
with a primed-continuous infusion of insulin (15 pmol 䡠 kg⫺1 䡠 min⫺1;
humulin; Eli Lilly, Indianapolis, IN) and a variable infusion of 20%
glucose to maintain euglycemia. Basal and insulin-stimulated wholebody glucose turnover was estimated with a continuous infusion of
[3-3H]glucose (PerkinElmer Life and Analytical Sciences, Boston,
MA) for 2 h before (0.05 ␮Ci/min) and throughout the clamps (0.1
␮Ci/min), respectively. The glucose infusion rate (GIR) adjusted for
body weight was used as one measure of insulin sensitivity. In
addition, we calculated a clamp-derived index of insulin sensitivity as
SIClamp ⫽ M/(G ⫻ ⌬I) adjusted for body weight where M is the
steady-state glucose infusion rate (mg/min), G is the steady-state
blood glucose concentration (mg/dl), and ⌬I is the difference between
basal and steady-state plasma insulin concentrations (␮U/ml). Rates
of basal hepatic glucose production (HGP) and insulin-stimulated
whole-body glucose turnover were determined as the ratio of the
[3H]glucose infusion rate (disintegrations/min) to the specific activity
of plasma glucose (dpm/␮mol) at the end of the basal period and
during the final 30 min of the clamp procedure, respectively. HGP
during hyperinsulinemic clamp conditions was determined by subtracting the GIR from whole-body glucose turnover. All procedures
were approved by the appropriate Animal Care and Use Committees.
Surrogate indexes of insulin sensitivity and resistance. Surrogate
indexes were calculated from fasting blood glucose and plasma
insulin concentrations as follows: QUICKI ⫽ 1/[log(I0) ⫹ log(G0)],
where I0 is fasting insulin (␮U/ml) and G0 is fasting glucose (mg/dl;
Ref. 17); and HOMA ⫽ (G0 ⫻ I0)/22.5 (with glucose expressed as
mmol/l and insulin expressed as ␮U/ml; Ref. 21). Other surrogate
indexes examined were 1/HOMA, log(HOMA) and 1/fasting insulin.
Calibration model analysis of surrogate indexes adjusted for body
weight. Calibration is inverse regression (4). For the model y ⫽ f (x;
␪) ⫹ ε, x is the explanatory variable, y is the response variable, ␪ is
an unknown parameter, and ε is the random error. Using an estimated
^
model y ⫽ f (x; ␪) to predict a new y* for a given x* is regression.
Conversely, predicting a new x* for a given y* is calibration. If the
value of x is prespecified as part of an experimental design, this is
called classical or controlled calibration. Since QUICKI (or other
surrogates), GIR (or SIClamp), and weight are measured with error
from an experimental population, random calibration is the more
appropriate method to use. In random calibration, there is no difficulty
in specifying the conditional distribution of x given y, so that random
calibration is similar to regression in prediction. Here we fitted a
calibration model xi⫽ ␣ ⫹ ␤1Si ⫹ ␤2Wi ⫹ ␤3Si 䡠 Wi ⫹ εi, where xi is
GIR (or SIClamp), Si is the surrogate index, Wi is the body weight,
Si 䡠 Wi is the multiplication of surrogate index and body weight, and εi is
the random error for the ith subject. It was assumed that the random error
had Gaussian distribution with mean of 0 and a constant variance. Even
though GIR (or SIClamp) was measured with error, it was assumed for our
model that the measurement error of GIR (or SIClamp) (determined from
E263
INSULIN SENSITIVITY INDEXES AND GLUCOSE CLAMP ESTIMATES
Table 1. Metabolic characteristics of mice grouped according to insulin sensitivity relative to respective appropriate
control mice
Decreased Insulin Sensitivity
Increased Insulin Sensitivity
Groups
Control
hPTP1B-TG
hLAR-TG
L-SACC1
3-wk HFD
6-wk HFD
3-mo HFD
Par-1bKO
Ptpn-6me-v/me-v
CASA/Rk
n
Body weight, g
Fasting glucose, mg/dl
Fasting insulin, ␮U/ml
Clamp glucose, mg/dl
Clamp insulin, ␮U/ml
GIR, mg 䡠 kg⫺1 䡠 min⫺1
Basal HGP, mg 䡠 kg⫺1 䡠 min⫺1
Clamp HGP, mg 䡠 kg⫺1 䡠 min⫺1
WBGT, mg 䡠 kg⫺1 䡠 min⫺1
32
26⫾1
131⫾6
14⫾1
109⫾4
53⫾4
43⫾2
16⫾1
0⫾1
43⫾2
7
33⫾1
147⫾13
29⫾5
113⫾7
76⫾7
25⫾5
15⫾1
7⫾4
32⫾3
7
34⫾1
168⫾16
23⫾2
115⫾9
74⫾8
26⫾4
18⫾2
7⫾3
33⫾2
4
27⫾2
135⫾7
15⫾2
105⫾5
65⫾10
43⫾3
20⫾2
10⫾2
52⫾1
10
30⫾1
175⫾13
17⫾1
103⫾7
42⫾4
18⫾6
17⫾1
9⫾3
28⫾3
6
28⫾1
165⫾6
20⫾6
94⫾3
38⫾4
15⫾4
18⫾2
14⫾3
29⫾2
10
38⫾1
142⫾9
38⫾6
125⫾7
62⫾5
12⫾1
10⫾1
8⫾1
21⫾1
3
18⫾1
97⫾5
9⫾2
121⫾4
30⫾2
61⫾5
15⫾2
0⫾5
61⫾8
6
14⫾2
72⫾14
11⫾2
113⫾6
39⫾4
52⫾5
17⫾2
⫺4⫾3
49⫾2
2
26⫾1
150⫾6
15⫾2
110⫾7
32⫾4
47⫾3
18⫾1
1⫾1
47⫾3
direct estimates of insulin sensitivity adjusted for body weight
from glucose clamp data (Table 2). In addition, from fasting
insulin and glucose data, we calculated a variety of simple
surrogate indexes of insulin sensitivity and resistance, including QUICKI, HOMA, 1/HOMA, log(HOMA), and 1/fasting
insulin (Table 2). As expected, measures of insulin sensitivity
derived from glucose clamp data demonstrated lower mean
GIR and SIClamp for groups of mice with decreased insulin
sensitivity and higher mean GIR and SIClamp for groups of mice
with increased insulin sensitivity when compared with mean
values from the control group with normal insulin sensitivity.
Similarly, for surrogate indexes of insulin sensitivity (QUICKI,
1/HOMA, and 1/fasting insulin), lower mean values were
observed for groups of mice with decreased insulin sensitivity while higher mean values were observed for groups of
mice with increased insulin sensitivity when compared with
mean values from the control group with normal insulin
sensitivity. Finally, for surrogate indexes of insulin resistance [HOMA and log(HOMA)], higher mean values were
observed for groups of mice with decreased insulin sensitivity while lower mean values were observed for groups of
mice with increased insulin sensitivity when compared with
mean values from the control group with normal insulin
sensitivity. Thus, the rank order of insulin sensitivity and
resistance determined by all of the surrogate indexes tested
corresponded to the correct rank order of insulin sensitivity
determined by two direct measures obtained from glucose
clamp data. Subgroup analyses in specific groups of mice
were not considered a priori due to the small sample sizes of
subgroups.
Correlations between clamp-derived measures of insulin
sensitivity and surrogate indexes of insulin sensitivity and
resistance. To determine the relative merits of various surrogate indexes of insulin sensitivity and resistance, we first
analyzed linear correlations between all of the surrogate indexes and GIR (Fig. 1) or SIClamp (Fig. 2). Modest linear
correlations were observed for all comparisons. 1/fasting insulin demonstrated the weakest correlation with GIR or SIClamp
(R ⬇ 0.30; P ⬍ 0.008). All of the other surrogate indexes
[QUICKI, HOMA, 1/HOMA, and log(HOMA)] had stronger
linear correlations with GIR or SIClamp (R ⬇ 0.45; P ⬍
0.0001). In the original cohort (n ⫽ 110), linear correlations
between GIR and other surrogate indexes [QUICKI, HOMA,
1/HOMA, and log(HOMA)] were also comparable (R ⬇ 0.40;
P ⬍ 0.0001). However, based on the similar correlation coefficients, we could not distinguish any particular advantage
among these remaining surrogate indexes. Moreover, the linear
correlations between surrogate indexes and clamp-derived estimates of insulin sensitivity we observed in mice were not as
substantial as those previously reported in humans (5, 6, 17,
19). In addition, the linear correlations between surrogate
indexes and whole-body glucose turnover were generally
weaker than correlations with GIR or SIClamp (data not shown).
Finally, we did not observe any substantial linear correlations
between any of the surrogate indexes tested and insulinmediated suppression of HGP during the glucose clamp (data
not shown).
Correlations between GIR or SIClamp and body weight in
mice. One potential explanation for disparity between mice and
humans with respect to linear correlations of direct and surro-
Table 2. Glucose clamp derived measures of insulin sensitivity (GIR and SIClamp) as well as surrogate indexes of insulin
sensitivity and resistance in mice grouped according to insulin sensitivity
Groups
n
Control mice
32
Decreased insulin sensitivity 44
Increased insulin sensitivity 11
GIR, mg 䡠 kg⫺1 䡠 min⫺1
SIClamp,
(10⫺4 䡠 dl 䡠 kg⫺1 䡠 min⫺1)/(␮U/ml)
43.5⫾2.0
21.1⫾2.3
51.1⫾3.0
131⫾12
68⫾10
195⫾18
QUICKI
HOMA
1/HOMA
Log(HOMA)
1/Fasting
Insulin, ml/␮U
0.315⫾0.004 4.51⫾0.44 0.30⫾0.03 0.590⫾0.043 0.092⫾0.010
0.287⫾0.002 8.58⫾0.61 0.14⫾0.01 0.892⫾0.029 0.053⫾0.004
0.345⫾0.008 2.36⫾0.46 0.54⫾0.07 0.310⫾0.069 0.113⫾0.014
Data shown are mean ⫾ SE from glucose clamp studies at baseline (fasting) or under steady-state clamp conditions. GIR, glucose infusion rate; QUICKI,
quantitative insulin sensitivity check index; HOMA, homeostasis model assessment.
AJP-Endocrinol Metab • VOL
294 • FEBRUARY 2008 •
www.ajpendo.org
Downloaded from http://ajpendo.physiology.org/ by 10.220.33.2 on June 15, 2017
Data shown are mean ⫾ SE from glucose clamp studies at baseline (fasting) or under steady-state clamp conditions. TG, transgenic mice; KO, knockout mice;
GIR, glucose infusion rate; HGP, hepatic glucose production; WBGT, whole body glucose turnover; HFD, high fat diet.
E264
INSULIN SENSITIVITY INDEXES AND GLUCOSE CLAMP ESTIMATES
Downloaded from http://ajpendo.physiology.org/ by 10.220.33.2 on June 15, 2017
Fig. 1. Linear correlation between glucose infusion rate (GIR) and various surrogate indexes of insulin sensitivity and resistance derived from glucose clamp studies; n ⫽ 87 with
32 control mice (F), 44 mice from decreased
insulin sensitivity groups (E), and 11 mice from
increased insulin sensitivity groups (‚). Solid
line represents the linear least squares fit between
GIR and each surrogate indexes of insulin sensitivity and resistance for all mice. Correlation
coefficients (R) and respective P values are
shown in A–E. A: results derived from Quantitative insulin-sensitivity check index (QUICKI).
B: results derived from homeostasis model assessment (HOMA). C: results derived from
1/HOMA. D: results derived from log(HOMA).
E: results derived from 1/fasting insulin.
gate indexes of insulin sensitivity may be that the ratio of body
surface area to mass is quite different. In addition, unlike
humans, rodents do not have a true physiological fasting state
and their body weight may change significantly upon fasting.
Therefore, we examined linear correlations between GIR or
SIClamp and body weight per se in our mice (Fig. 3). Body
weight of our mice had a substantially stronger correlation with
clamp-derived measures of insulin sensitivity than any of the
surrogate indexes tested (R ⫽ ⫺0.65; P ⬍ 0.0001 for GIR).
Therefore, it may be useful in mice to adjust surrogate indexes
of insulin sensitivity and resistance according to body weight.
Calibration model analysis of surrogate indexes adjusted for
body weight. We tested the absolute accuracy of various
surrogate indexes adjusted for body weight using calibration
model analysis. The calibration model generates predictions of
AJP-Endocrinol Metab • VOL
GIR (Fig. 4) or SIClamp (Fig. 5) based on data from each
surrogate index. If a surrogate index adjusted for body weight
perfectly predicts GIR or SIClamp, a plot of the predicted vs.
observed values of GIR or SIClamp for a given surrogate index
would generate a straight line with a slope of 1 and a yintercept of 0. To generate the plots in Figs. 4 and 5, we used
the leave-one-out cross-validation approach to the calibration
model (as described in MATERIALS AND METHODS). When comparing error functions CVPE or RMSE for each of the surrogate indexes, the error was much smaller for predicting GIR
than for predicting SIClamp (Table 3). However, when evaluating the absolute accuracy of all of the surrogate indexes
adjusted for body weight that were tested, no significant
differences were noted among any of the indexes when compared with QUICKI (evaluated by either CVPE or RMSE and
294 • FEBRUARY 2008 •
www.ajpendo.org
INSULIN SENSITIVITY INDEXES AND GLUCOSE CLAMP ESTIMATES
E265
either with GIR or SIClamp; Table 3). Thus in mice, adjusting
any of the common surrogate indexes of insulin sensitivity and
resistance for body weight results in reasonable predictive
accuracy for determining GIR.
DISCUSSION
Insulin resistance plays a significant role in the pathogenesis
of diabetes and is strongly associated with obesity and cardiovascular diseases (9, 28). Mouse models of these human
diseases with acquired or genetic forms of insulin resistance
and sensitivity are useful for gaining insight into pathophysiological mechanisms, for evaluating therapeutic interventions,
and for screening novel therapeutic compounds and approaches. The hyperinsulinemic euglycemic clamp technique
is considered the reference standard for determining insulin
sensitivity and resistance in both humans and rodents. HowAJP-Endocrinol Metab • VOL
ever, the glucose clamp is laborious, time consuming, and
technically demanding to perform (more so in rodents than in
humans). In humans, a number of surrogate indexes for insulin
sensitivity and resistance have been developed that correlate
reasonably well with glucose clamp results (2, 3, 5, 17, 20, 21,
23). However, the present study is the first to evaluate comparisons between surrogate indexes of insulin sensitivity and
resistance and glucose clamp estimates in mice. It is important
to explicitly evaluate these comparisons in mice because validation of simple, accurate, and reliable surrogate indexes of insulin
sensitivity and resistance in mice will be valuable for drug development and screening purposes as well as for other studies where
feasibility issues prevent the use of the glucose clamp.
Comparisons between glucose clamp estimates and surrogate indexes of insulin sensitivity and resistance. In the present
study, we used data from a total of 87 mice that exhibited a
294 • FEBRUARY 2008 •
www.ajpendo.org
Downloaded from http://ajpendo.physiology.org/ by 10.220.33.2 on June 15, 2017
Fig. 2. Linear correlation between clamp-derived index of insulin sensitivity (SIClamp) and
various surrogate indexes of insulin sensitivity
and resistance derived from glucose clamp
studies; n ⫽ 87 with 32 control mice (F), 44
mice from decreased insulin sensitivity groups
(E), and 11 mice from increased insulin sensitivity groups (‚). Solid line represents the linear
least squares fit between SIClamp and each surrogate indexes of insulin sensitivity and resistance for all mice. Correlation coefficients (R)
and respective P values are shown in A–E.
A: results derived from QUICKI. B: results derived from HOMA. C: results derived from
1/HOMA. D: results derived from log(HOMA).
E: results derived from 1/fasting insulin.
E266
INSULIN SENSITIVITY INDEXES AND GLUCOSE CLAMP ESTIMATES
wide range of insulin sensitivity and resistance resulting from
either genetic or environmental manipulations. We only
used mice where complete data was available to obtain GIR,
SIClamp, and fasting surrogate indexes. Consequently, due to
the small sample sizes in some of the specific groups of mice,
we have not performed any subgroup analyses. Metabolic
phenotyping and comparison of glucose metabolism between
the specific models of insulin resistance and sensitivity and
their respective wild-type littermates have been previously
published (10, 16, 24, 25, 33, 36). In this study however, we
first grouped mice according to normal insulin sensitivity
(control mice), decreased insulin sensitivity, and increased
insulin sensitivity as determined by GIR. Importantly,
SIClamp, as well as all the simple surrogate indexes evaluated, correctly identified the same rank order of insulin
sensitivity and resistance. Thus, all of the surrogate indexes
tested in mice worked equally well for determining large
differences in insulin sensitivity and resistance when compared with two direct measures (GIR and SIClamp). This is
similar to findings in humans (5, 6, 17, 19).
AJP-Endocrinol Metab • VOL
294 • FEBRUARY 2008 •
www.ajpendo.org
Downloaded from http://ajpendo.physiology.org/ by 10.220.33.2 on June 15, 2017
Fig. 3. Linear correlation between body weight and GIR (A) or SIClamp (B);
n ⫽ 87 with 32 control mice (F), 44 mice from decreased insulin sensitivity
groups (E), and 11 mice from increased insulin sensitivity groups (‚). Solid
line represents the linear least squares fit for all mice. Correlation coefficients
(R) and respective P values are shown in A and B.
To more rigorously evaluate the surrogate indexes we also
evaluated linear correlations between surrogates and either
GIR or SIClamp. All of the surrogate indexes evaluated
[QUICKI, HOMA, 1/HOMA, log(HOMA), and 1/fasting insulin] had modest linear correlations with both GIR and
SIClamp. However, the correlation between GIR or SIClamp and
1/fasting insulin (R ⬇ 0.30) was not as high as with the other
surrogates (R ⬇ 0.45). Thus, this analysis revealed modest
linear correlations between direct measures of insulin sensitivity and surrogate indexes in mice but no clear advantage among
QUICKI, HOMA, 1/HOMA, or log(HOMA). It is possible that
using larger numbers of mice may improve correlations between surrogates and glucose clamp-derived measures of insulin sensitivity. However, this seems unlikely as linear correlations between GIR and surrogates for 110 mice were similar
to those we observed for the set of 87 mice with complete data.
By contrast, in humans, both QUICKI and log(HOMA) have
much more substantial and stronger linear correlations with
glucose clamp estimates in multiple independent studies (R ⬇
0.70 – 0.90; Refs. 5, 6, 17, 19, 29, 34, 35).
There are several possible explanations for these differences
observed between mice and humans. First, all of the surrogate
indexes rely on the assumption that fasting glucose and insulin
levels represent a basal steady-state condition. However, unlike
humans, rodents have no true physiological fasting state. Mice
feed primarily during the dark hours and prolonged fasting (15
h) increases peripheral insulin sensitivity and decreases lean
body mass, hepatic glycogen content, plasma glucose, and
insulin levels (1, 12). Consequently, changes in peripheral
insulin sensitivity, fasting insulin, and glucose concentrations
during fasting in mice occur in a time-dependent fashion. That
is, nonphysiological fasting conditions imposed on mice may
not determine steady-state conditions. Second, the ratio of
body surface area to mass in mice is quite different than in
humans. Thus, mice may have substantial changes in body
weight upon fasting (1). Indeed, we found a more substantial
linear correlation between GIR (or SIClamp) and body weight
per se than with any of the surrogate indexes tested. This
suggests that in mice it may be useful to adjust surrogate
indexes for body weight. Third, it is likely that the glucose
clamp technique in mice may not be as accurate and reliable as
in humans (1). Mice have very small blood volumes (⬃2 ml)
so that limited sampling ability makes it difficult to achieve and
rigorously verify steady-state clamp conditions (especially
with respect to insulin levels). For example, in humans, the
clamp steady-state values for GIR, blood glucose, and plasma
insulin typically vary by ⬍5% over 60 min. In mice, because
of small blood volumes, it is not possible to measure multiple
insulin concentrations during the clamp to verify steady-state
insulin levels. In addition, the glucose clamp procedure is
stressful in mice (even if done under anesthetized conditions).
Inherent limitations and variability associated with measurement of low levels of plasma insulin during the fasting state
may also contribute to the modest correlations observed in the
present study. Another consideration is that many monogenic
forms of insulin resistance in mice preferentially cause
skeletal muscle insulin resistance without substantially affecting hepatic insulin sensitivity. The GIR from the clamp
predominantly reflects skeletal muscle insulin sensitivity,
while the surrogate indexes based on fasting glucose and
insulin levels primarily reflect hepatic insulin sensitivity. In
INSULIN SENSITIVITY INDEXES AND GLUCOSE CLAMP ESTIMATES
E267
humans, under most conditions, hepatic and muscle insulin
sensitivity and resistance are proportional. However, in mice, if
skeletal muscle and hepatic insulin sensitivity and resistance
are uncoupled, this may contribute to weaker correlations
between surrogate indexes and glucose clamp estimates. In
rats, correlations between surrogate measures of insulin sensitivity (QUICKI, HOMA, and fasting insulin) and clamp-derived estimates of glucose disposal are also modest (31). It is
possible that limitations that apply to mice may also affect
similar studies in rats.
Calibration model analysis of surrogate indexes adjusted for
body weight. To evaluate the absolute accuracy of various
surrogate indexes of insulin sensitivity and resistance, we used
calibration model analysis to assess the predictive accuracy of
each surrogate index adjusted for body weight. For this analysis, we assumed that error in GIR or SIClamp is small when
compared with error in surrogate indexes. With the use of
AJP-Endocrinol Metab • VOL
calibration model analysis, weight-adjusted QUICKI (or other
weight-adjusted surrogate indexes) was reasonable at predicting GIR. In contrast, in humans, QUICKI and log(HOMA)
were both excellent at predicting SIClamp, with a much smaller
CVPE and RMSE compared with this study in mice (5). In
addition, both error functions CVPE and RMSE were consistently smaller for surrogates to predict GIR than for surrogates
to predict SIClamp. This suggests that SIClamp in mice may have
a larger error than GIR for reasons related to small blood
volumes and difficulties in obtaining sufficient insulin determinations in mice as described above. In humans, QUICKI and
log(HOMA) have the best linear correlation with and strongest
predictive accuracy for glucose clamp estimates of insulin
sensitivity and resistance (5). However, in our calibration
model analysis of mice, there were no significant differences in
predictive accuracy among any of the surrogates tested. Moreover, predictive accuracy of surrogate indexes in mice even
294 • FEBRUARY 2008 •
www.ajpendo.org
Downloaded from http://ajpendo.physiology.org/ by 10.220.33.2 on June 15, 2017
Fig. 4. Comparison between observed GIR during
glucose clamp and predicted GIR derived from the
calibration model of various surrogate indexes of
insulin sensitivity and resistance adjusted for body
weight (leave-one-out cross validation analysis) as
described in MATERIALS AND METHODS; n ⫽ 87 with
32 control mice (F), 44 mice from decreased insulin
sensitivity groups (E), and 11 mice from increased
insulin sensitivity groups (‚). The dashed line indicates ideal predictive accuracy with a slope of 1 and
intercept of 0. A: results derived from QUICKI
adjusted for body weight. B: results derived from
HOMA adjusted for body weight. C: results derived
from 1/HOMA adjusted for body weight. D: results
derived from log(HOMA) adjusted for body weight.
E: results derived from 1/fasting insulin adjusted for
body weight.
E268
INSULIN SENSITIVITY INDEXES AND GLUCOSE CLAMP ESTIMATES
after adjusting for body weight as well as linear correlations is
not substantial as in humans. This may reflect intrinsic differences between human and rodent physiology (e.g., using monogenic mouse models of polygenic human disease) as well as
increased technical difficulties in performing glucose clamps in
mice with small blood volumes.
In summary, in mice, many common simple surrogate indexes of insulin sensitivity and resistance are modestly corre-
Table 3. CVPE and RMSE calculated from calibration analysis of various surrogate indexes of insulin sensitivity adjusted for weight
GIR
CVPE
QUICKI
HOMA
1/HOMA
Log(HOMA)
1/fasting insulin
13.94
14.75
13.68
14.07
14.31
SIClamp
P
RMSE
0.36
0.68
0.32
0.28
13.54
13.83
13.43
13.60
13.92
P
CVPE
0.22
0.79
0.20
0.30
70.02
71.40
69.59
70.26
69.32
P
RMSE
P
0.39
0.54
0.42
0.60
67.73
68.91
67.23
68.00
68.32
0.20
0.71
0.23
0.71
These error estimates are derived from comparing observed versus predicted GIR or SIClamp from glucose clamp studies as described in MATERIALS AND METHODS
(n ⫽ 87). P values correspond to statistical comparisons between CVPE or RMSE for QUICKI adjusted for weight versus alternative surrogate indexes adjusted for
weight. QCVPE, leave-one-out cross-validation-type root mean squared error of prediction; RMSE, square root of the mean squared error of prediction.
AJP-Endocrinol Metab • VOL
294 • FEBRUARY 2008 •
www.ajpendo.org
Downloaded from http://ajpendo.physiology.org/ by 10.220.33.2 on June 15, 2017
Fig. 5. Comparison between observed SIClamp derived
from glucose clamp data and predicted SIClamp derived
from the calibration model of various surrogate indexes of insulin sensitivity and resistance adjusted for
body weight (leave-one-out cross validation analysis)
as described in MATERIALS AND METHODS; n ⫽ 87 with
32 control mice (F), 44 mice from decreased insulin
sensitivity groups (E), and 11 mice from increased
insulin sensitivity groups (‚). Dashed line indicates
ideal predictive accuracy with a slope of 1 and intercept of 0. A: results derived from QUICKI adjusted for
body weight. B: results derived from HOMA adjusted
for body weight. C: results derived from 1/HOMA
adjusted for body weight. D: results derived from
log(HOMA) adjusted for body weight. E: results derived from 1/fasting insulin adjusted for body weight.
INSULIN SENSITIVITY INDEXES AND GLUCOSE CLAMP ESTIMATES
11.
12.
13.
14.
15.
16.
17.
ACKNOWLEDGMENTS
18.
We thank J. L. Breslow (The Rockefeller University, New York, NY) for
kindly providing the CASA/Rk transgenic mice. Glucose clamp data were
obtained from previously published and unpublished studies performed at the
Yale Mouse Metabolic Phenotyping Center and the Penn State Mouse Metabolic Phenotyping Center.
19.
GRANTS
This work was supported, in part, by the Intramural Research Program,
National Center for Complementary and Alternative Medicine, National Institutes of Health and by a Mentor-based Post-Doctoral Fellowship Award from
the American Diabetes Association (M. J. Quon). These studies were funded
by the National Institute of Diabetes and Digestive and Kidney Diseases Grant
U24-DK-59635 and American Diabetes Association Grants 1-04-RA-47 and
7-07-RA-80 (to J. K. Kim).
20.
21.
22.
REFERENCES
1. Ayala JE, Bracy DP, McGuinness OP, Wasserman DH. Considerations
in the design of hyperinsulinemic-euglycemic clamps in the conscious
mouse. Diabetes 55: 390 –397, 2006.
2. Bergman RN, Hope ID, Yang YJ, Watanabe RM, Meador MA, Youn
JH, Ader M. Assessment of insulin sensitivity in vivo: a critical review.
Diabetes Metab Rev 5: 411– 429, 1989.
3. Bonora E, Targher G, Alberiche M, Bonadonna RC, Saggiani F,
Zenere MB, Monauni T, Muggeo M. Homeostasis model assessment
closely mirrors the glucose clamp technique in the assessment of insulin
sensitivity: studies in subjects with various degrees of glucose tolerance
and insulin sensitivity. Diabetes Care 23: 57– 63, 2000.
4. Brown PJ. Measurement, Regression and Calibration. Oxford, UK:
Clarendon, 1993.
5. Chen H, Sullivan G, Quon MJ. Assessing the predictive accuracy of
QUICKI as a surrogate index for insulin sensitivity using a calibration
model. Diabetes 54: 1914 –1925, 2005.
6. Chen H, Sullivan G, Yue LQ, Katz A, Quon MJ. QUICKI is a useful
index of insulin sensitivity in subjects with hypertension. Am J Physiol
Endocrinol Metab 284: E804 –E812, 2003.
7. Clee SM, Attie AD. The genetic landscape of type 2 diabetes in mice.
Endocr Rev 28: 48 – 83, 2007.
8. Collins S, Martin TL, Surwit RS, Robidoux J. Genetic vulnerability to
diet-induced obesity in the C57BL/6J mouse: physiological and molecular
characteristics. Physiol Behav 81: 243–248, 2004.
9. DeFronzo RA. Pathogenesis of type 2 diabetes mellitus. Med Clin North
Am 88: 787– 835, ix, 2004.
10. Dubois MJ, Bergeron S, Kim HJ, Dombrowski L, Perreault M,
Fournes B, Faure R, Olivier M, Beauchemin N, Shulman GI, SimiAJP-Endocrinol Metab • VOL
23.
24.
25.
26.
27.
28.
29.
novitch KA, Kim JK, Marette A. The SHP-1 protein tyrosine phosphatase negatively modulates glucose homeostasis. Nat Med 12: 549 –556,
2006.
Efron B, Tibshirani R. An Introduction to the Bootstrap. Boca Raton, FL:
Chapman & Hall/CRC, 1998.
Heijboer AC, Donga E, Voshol PJ, Dang ZC, Havekes LM, Romijn
JA, Corssmit EP. Sixteen hours of fasting differentially affects hepatic
and muscle insulin sensitivity in mice. J Lipid Res 46: 582–588, 2005.
Herbach N, Rathkolb B, Kemter E, Pichl L, Klaften M, de Angelis
MH, Halban PA, Wolf E, Aigner B, Wanke R. Dominant-negative
effects of a novel mutated Ins2 allele causes early-onset diabetes and
severe beta-cell loss in Munich Ins2C95S mutant mice. Diabetes 56:
1268 –1276, 2007.
Hribal ML, Oriente F, Accili D. Mouse models of insulin resistance.
Am J Physiol Endocrinol Metab 282: E977–E981, 2002.
Hsueh W, Abel ED, Breslow JL, Maeda N, Davis RC, Fisher EA,
Dansky H, McClain DA, McIndoe R, Wassef MK, Rabadan-Diehl C,
Goldberg IJ. Recipes for creating animal models of diabetic cardiovascular disease. Circ Res 100: 1415–1427, 2007.
Hurov JB, Huang M, White LS, Lennerz J, Choi CS, Cho YR, Kim
HJ, Prior JL, Piwnica-Worms D, Cantley LC, Kim JK, Shulman GI,
Piwnica-Worms H. Loss of the Par-1b/MARK2 polarity kinase leads to
increased metabolic rate, decreased adiposity, and insulin hypersensitivity
in vivo. Proc Natl Acad Sci USA 104: 5680 –5685, 2007.
Katz A, Nambi SS, Mather K, Baron AD, Follmann DA, Sullivan G,
Quon MJ. Quantitative insulin sensitivity check index: a simple, accurate
method for assessing insulin sensitivity in humans. J Clin Endocrinol
Metab 85: 2402–2410, 2000.
Maeda N, Shimomura I, Kishida K, Nishizawa H, Matsuda M, Nagaretani H, Furuyama N, Kondo H, Takahashi M, Arita Y, Komuro R,
Ouchi N, Kihara S, Tochino Y, Okutomi K, Horie M, Takeda S,
Aoyama T, Funahashi T, Matsuzawa Y. Diet-induced insulin resistance
in mice lacking adiponectin/ACRP30. Nat Med 8: 731–737, 2002.
Mather KJ, Hunt AE, Steinberg HO, Paradisi G, Hook G, Katz A,
Quon MJ, Baron AD. Repeatability characteristics of simple indices of
insulin resistance: implications for research applications. J Clin Endocrinol Metab 86: 5457–5464, 2001.
Matsuda M, DeFronzo RA. Insulin sensitivity indices obtained from oral
glucose tolerance testing: comparison with the euglycemic insulin clamp.
Diabetes Care 22: 1462–1470, 1999.
Matthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF,
Turner RC. Homeostasis model assessment: insulin resistance and betacell function from fasting plasma glucose and insulin concentrations in
man. Diabetologia 28: 412– 419, 1985.
Moloney F, Toomey S, Noone E, Nugent A, Allan B, Loscher CE,
Roche HM. Antidiabetic effects of cis-9, trans-11-conjugated linoleic acid
may be mediated via anti-inflammatory effects in white adipose tissue.
Diabetes 56: 574 –582, 2007.
Pacini G, Mari A. Methods for clinical assessment of insulin sensitivity
and beta-cell function. Best Pract Res Clin Endocrinol Metab 17: 305–
322, 2003.
Park SY, Cho YR, Kim HJ, Higashimori T, Danton C, Lee MK, Dey
A, Rothermel B, Kim YB, Kalinowski A, Russell KS, Kim JK.
Unraveling the temporal pattern of diet-induced insulin resistance in
individual organs and cardiac dysfunction in C57BL/6 mice. Diabetes 54:
3530 –3540, 2005.
Park SY, Cho YR, Kim HJ, Hong EG, Higashimori T, Lee SJ,
Goldberg IJ, Shulman GI, Najjar SM, Kim JK. Mechanism of glucose
intolerance in mice with dominant negative mutation of CEACAM1. Am J
Physiol Endocrinol Metab 291: E517–E524, 2006.
Potenza MA, Marasciulo FL, Tarquinio M, Quon MJ, Montagnani M.
Treatment of spontaneously hypertensive rats with rosiglitazone and/or
enalapril restores balance between vasodilator and vasoconstrictor actions
of insulin with simultaneous improvement in hypertension and insulin
resistance. Diabetes 55: 3594 –3603, 2006.
Potenza MA, Marasciulo FL, Tarquinio M, Tiravanti E, Colantuono
G, Federici A, Kim JA, Quon MJ, Montagnani M. EGCG, a green tea
polyphenol, improves endothelial function and insulin sensitivity, reduces
blood pressure, and protects against myocardial I/R injury in SHR. Am J
Physiol Endocrinol Metab 292: E1378 –E1387, 2007.
Reaven G, Abbasi F, McLaughlin T. Obesity, insulin resistance, and
cardiovascular disease. Recent Prog Horm Res 59: 207–223, 2004.
Skrha J, Haas T, Sindelka G, Prazny M, Widimsky J, Cibula D,
Svacina S. Comparison of the insulin action parameters from hyperinsu-
294 • FEBRUARY 2008 •
www.ajpendo.org
Downloaded from http://ajpendo.physiology.org/ by 10.220.33.2 on June 15, 2017
lated with GIR from glucose clamp studies. The predictive
accuracy of these surrogates is improved by adjusting the
surrogates for body weight (an important determinant of insulin sensitivity in mice). However, unlike in humans where
QUICKI and log(HOMA) are superior to other surrogates (5),
we could not establish the superiority of a particular surrogate
index in mice. This may be due to inherent technical difficulties in performing clamps in mice (that are not present in
human studies) or differences between mouse and human
physiology that render the surrogate indexes less appropriate.
Further studies are required to distinguish among these possibilities. We conclude that many of the surrogate indexes
commonly used in humans may be used in mice to roughly
determine insulin sensitivity and resistance. Thus, for drug
screening and evaluation, studies where insulin sensitivity is a
secondary outcome, or for conditions where feasibility issues
preclude the use of glucose clamps, it may be advantageous to
use surrogate indexes of insulin sensitivity and resistance
adjusted for body weight. However, for studies where feasibility is not an issue and determination of insulin sensitivity is of
primary interest, using a direct measure derived from glucose
clamp studies is preferable.
E269
E270
INSULIN SENSITIVITY INDEXES AND GLUCOSE CLAMP ESTIMATES
AJP-Endocrinol Metab • VOL
33. Vitarius JA, Sehayek E, Breslow JL. Identification of quantitative trait
loci affecting body composition in a mouse intercross. Proc Natl Acad Sci
USA 103: 19860 –19865, 2006.
34. Yokoyama H, Emoto M, Fujiwara S, Motoyama K, Morioka T,
Komatsu M, Tahara H, Koyama H, Shoji T, Inaba M, Nishizawa Y.
Quantitative insulin sensitivity check index and the reciprocal index of
homeostasis model assessment are useful indexes of insulin resistance in
type 2 diabetic patients with wide range of fasting plasma glucose. J Clin
Endocrinol Metab 89: 1481–1484, 2004.
35. Yokoyama H, Emoto M, Fujiwara S, Motoyama K, Morioka T,
Komatsu M, Tahara H, Shoji T, Okuno Y, Nishizawa Y. Quantitative
insulin sensitivity check index and the reciprocal index of homeostasis
model assessment in normal range weight and moderately obese type 2
diabetic patients. Diabetes Care 26: 2426 –2432, 2003.
36. Zabolotny JM, Haj FG, Kim YB, Kim HJ, Shulman GI, Kim JK, Neel
BG, Kahn BB. Transgenic overexpression of protein-tyrosine phosphatase 1B in muscle causes insulin resistance, but overexpression with
leukocyte antigen-related phosphatase does not additively impair insulin
action. J Biol Chem 279: 24844 –24851, 2004.
294 • FEBRUARY 2008 •
www.ajpendo.org
Downloaded from http://ajpendo.physiology.org/ by 10.220.33.2 on June 15, 2017
linemic clamps with homeostasis model assessment and QUICKI indexes
in subjects with different endocrine disorders. J Clin Endocrinol Metab
89: 135–141, 2004.
30. Sumiyoshi M, Sakanaka M, Kimura Y. Chronic intake of high-fat and
high-sucrose diets differentially affects glucose intolerance in mice. J Nutr
136: 582–587, 2006.
31. Tran TT, Gupta N, Goh T, Naigamwalla D, Chia MC, Koohestani
N, Mehrotra S, McKeown-Eyssen G, Giacca A, Bruce WR. Direct
measure of insulin sensitivity with the hyperinsulinemic-euglycemic
clamp and surrogate measures of insulin sensitivity with the oral
glucose tolerance test: correlations with aberrant crypt foci promotion
in rats. Cancer Epidemiol Biomarkers Prev 12: 47–56, 2003.
32. Verreth W, De Keyzer D, Pelat M, Verhamme P, Ganame J,
Bielicki JK, Mertens A, Quarck R, Benhabiles N, Marguerie G,
Mackness B, Mackness M, Ninio E, Herregods MC, Balligand JL,
Holvoet P. Weight-loss-associated induction of peroxisome proliferator-activated receptor-alpha and peroxisome proliferator-activated receptor-gamma correlate with reduced atherosclerosis and improved
cardiovascular function in obese insulin-resistant mice. Circulation
110: 3259 –3269, 2004.