Stimulated Insulin Secretion Predicts Changes in Body Composition Following Weight Loss in Adults with High BMI

Abstract

Background

The aim of obesity treatment is to promote loss of fat relative to lean mass. However, body composition changes with calorie restriction differ among individuals.

Objectives

The goal of this study was to test the hypothesis that insulin secretion predicts body composition changes among young and middle-age adults with high BMI (in kg/m²) following major weight loss.

Methods

Exploratory analyses were conducted with pre-randomization data from 2 large feeding trials: the Framingham, Boston, Bloomington, Birmingham, and Baylor study (FB4; n = 82, 43.9% women, BMI ≥27) and the Framingham State Food Study [(FS)2; n = 161, 69.6% women, BMI ≥25]. Participants in the 2 trials consumed calorie-restricted moderate-carbohydrate or very-low-carbohydrate diets to produce 12–18% weight loss in ∼14 wk or 10–14% in ∼10 wk, respectively. We determined insulin concentration 30 min after a 75-g oral glucose load (insulin-30) as a measure of insulin secretion and HOMA-IR as a measure of insulin resistance at baseline. Body composition was determined by DXA at baseline and post–weight loss. Associations were analyzed using general linear models with adjustment for covariates.

Results

In FB4, higher insulin-30 was associated with a smaller decrease in fat mass (0.441 kg per 100 μIU/mL increment in baseline insulin-30; P = 0.005; –1.20-kg mean difference between the first compared with the fifth group of insulin-30) and a larger decrease in lean mass (–0.465 kg per 100 μIU/mL; P = 0.004; 1.27-kg difference). Participants with higher insulin-30 lost a smaller proportion of weight loss as fat (–3.37% per 100 μIU/mL; P = 0.003; 9.20% difference). Greater HOMA-IR was also significantly associated with adverse body composition changes. Results from (FS)2 were qualitatively similar but of a smaller magnitude.

Conclusions

Baseline insulin dynamics predict substantial individual differences in body composition following weight loss. These findings may inform understanding of the pathophysiological basis for weight regain and the design of more effective obesity treatment. Registered at http://www.clinicaltrials.gov clinicaltrials.gov as NCT03394664 and NCT02068885.

See corresponding editorial on https://doi.org/10.1093/jn/nxac001 page 649.

Introduction

Excess body weight is associated with increased risk of all-cause mortality (1, 2). Current recommendations for obesity management focus on strategies to reduce body weight for chronic disease risk reduction (3), with less attention given to body composition. Measures of body weight and BMI (in kg/m²) do not discriminate between fat mass and lean mass, where these 2 compartments can vary widely between individuals. Excess loss of lean mass has been shown to result in various adverse effects, including reduced energy expenditure, and may predispose to weight regain (4., 5., 6., 7.).

Energy-restricted diets for weight loss lead to reductions in both fat mass and lean mass (8, 9), and changes in these 2 compartments with weight loss interventions vary between individuals. This heterogeneity may relate to many factors, including genetics, sex, age, initial body weight or composition, physical activity levels, dietary composition, rate of weight loss, metabolic profile or hormonal response, and gut microbiome (10). Individual differences in insulin dynamics may also influence changes in body composition in response to weight loss. As an anabolic hormone, insulin promotes metabolic deposition of substrates, including lipids, and suppresses lipolysis and fat oxidation (11). Increased insulin secretion is associated with higher body weight and adiposity (12, 13), as is insulin administration for intensive diabetes management (14, 15). In contrast, pharmacological agents that suppress insulin secretion facilitate loss of weight and fat mass (16).

Insulin dynamics at baseline (BSL; pretreatment) have been shown to modify the effect of macronutrient composition on changes in body weight (17., 18., 19., 20.), but few studies have assessed changes in body composition. Measures of insulin dynamics of special interest include insulin secretion, measured 30 min following a 75-g oral glucose load (insulin-30) (21, 22), and insulin resistance, calculated by HOMA-IR using fasting glucose and insulin concentrations (23). In 2 dietary trials, insulin-30 modified the response to diet, with a tendency for less weight loss among individuals with high insulin-30 on a high– compared with low–glycemic load diet (17, 18). In a controlled feeding trial, higher insulin-30 was associated with adverse body composition changes following weight loss and lower energy expenditure during weight-loss maintenance (24). Interventional data also suggest an interaction between insulin resistance and macronutrient composition in response to weight loss (19, 20).

The primary aim of this exploratory study was to examine associations between insulin-30 and changes in body composition during the weight-loss run-in phase of 2 metabolic feeding studies. We also examined associations involving HOMA-IR.

Methods

Study design

Data from the pre-randomization weight-loss run-in phases from the Framingham, Boston, Bloomington, Birmingham, and Baylor study (FB4) and the Framingham State Food Study [(FS)2] (25, 26) were analyzed separately.

FB4

Participants consumed a hypocaloric very-low-carbohydrate (VLC) diet (60% of estimated energy needs; 7.5% of energy from carbohydrate, 67.5% from fat, and 25% from protein) to promote weight loss of 12–18% relative to BSL body weight in ∼14 wk. Study foods were fully prepared and delivered to the homes of participants using a commercial food delivery service (Metabolic Meals). Participants had the option of adding their own nonstarchy vegetables to delivered meals.

(FS)2

A similar weight-loss run-in phase was utilized in this trial. Methodologic details of the full study and the primary outcome involving total energy expenditure were previously presented (25., 26., 27.). Participants consumed a hypocaloric moderate-carbohydrate diet (60% of estimated energy needs; 45% of energy from carbohydrate, 30% from fat, and 25% from protein) to promote weight loss of 10–14% relative to BSL body weight in ∼10 wk. All meals and snacks were prepared by Sodexo, the food service contractor at Framingham State University (FSU), and consumed either on the FSU campus or as takeout based on a defined schedule.

Participants

For FB4, participants aged 18–50 y with a BMI ≥27 underwent a multistep screening and enrollment process to determine eligibility. For (FS)2, participants aged 18–65 y with a BMI ≥25 underwent a similar multistep screening and enrollment process (25). Key inclusion criteria in both trials included body weight ≤350 lb (159 kg; to avoid exceeding upper weight limits for some assessment equipment), willingness to eat and drink only study foods and beverages, and no major food allergies or aversions. Key exclusionary criteria included recent and substantial weight change, chronic use of mediation that could affect study outcomes, abnormal screening laboratory tests, and physician diagnosis of a major medical illness or eating disorder. Participants who completed the weight-loss run-in phase and achieved the minimum weight loss target in FB4 and (FS)2 were included in this exploratory study.

Outcomes

In both trials, DXA (Horizon A; Hologic) was performed by licensed radiologic technologists. Whole-body (fat mass and pure lean mass) and regional (trunk fat mass and android:gynoid ratio) body compositions were assessed at BSL and the end of the weight-loss run-in phase [post–weight loss (PWL)]. Whole-body pure lean mass refers to lean soft tissue mass excluding bone mass, henceforth referred to as lean mass. Whole-body and regional percentages were calculated as the proportion of fat mass, lean mass, or trunk fat mass relative to total body weight (e.g., percentage fat mass was calculated by dividing fat mass by body weight and then multiplying by 100). The android:gynoid ratio was calculated by dividing android fat by gynoid fat. Android fat refers to the proportion of android fat mass relative to android region mass, and gynoid fat refers to the proportion of gynoid fat mass relative to gynoid region mass. Total body weight was calculated from DXA as the sum of 3 components: total fat mass, lean mass, and bone mineral content. Changes (absolute or percentage) in outcome variables were calculated as the difference at PWL – BSL. Proportion of weight loss as fat (expressed as percentage) was calculated by dividing change in fat mass by change in body weight and then multiplying by 100. At BSL, we conducted a 2-h oral–glucose–tolerance test (OGTT) using a standard 75-g dose of dextrose (Trutol; ThermoFisher Scientific). Plasma blood samples were collected for glucose and insulin analysis at –10, –5, 0, 10, 20, 30, 60, 90, and 120 min relative to the start time of dextrose consumption. For this report, we used insulin-30 as a measure of insulin secretion, consistent with previous studies (18, 25, 28). We calculated HOMA-IR using BSL fasting glucose and insulin concentrations (average of –10, –5, and 0 min) as a measure of insulin resistance (23). All participants with BSL insulin-30 and both DXA measurements were included in the main analyses. Among these participants, secondary analyses included those with BSL HOMA-IR.

Ethics

All study protocols for FB4 and (FS)2 were approved by the Institutional Review Board at Boston Children's Hospital. Written informed consent was obtained prior to BSL assessments.

Statistical analyses

Statistical analyses were conducted separately for FB4 and (FS)2 using SAS version 9.4 (SAS Institute). Statistical significance was defined as P < 0.05 in a 2-sided test.

Baseline participant characteristics were summarized with descriptive statistics (median and first and third quartiles for continuous variables, and frequency and percentage for categorical variables). To assess BSL cross-sectional associations between insulin-30 or HOMA-IR and body composition measures, we used general linear models (GLMs) with adjustment for age (calculated as age at BSL DXA visit) and sex.

Associations between insulin-30 and changes in body composition from BSL to PWL were assessed using GLMs. In each model, we adjusted for the following covariates: age, sex, race, ethnicity, percentage weight loss at PWL, and BSL measure of the dependent variable (e.g., when change in fat mass was the dependent variable, BSL fat mass was included as a covariate). To characterize body composition changes further, insulin-30 was divided with quintiles into 5 groups (with the first group representing 20% of participants with lowest insulin-30 and the fifth group representing 20% of participants with highest insulin-30) to provide an estimate of the effect size within the observed range in each study. Mean changes among participants in the first group compared with the fifth group were estimated and compared using estimated contrasts from GLMs adjusting for the aforementioned covariates. We also assessed associations between HOMA-IR and changes in body composition applying the same model and covariates used for the insulin-30 analyses.

We assessed model assumptions by inspecting residuals from models using QQ plots and skewness statistics (29) for normality and plots of residuals by predicted values for homoscedasticity (30, 31). Because skewness was less than ±1.0 in all models, and variance was relatively constant across predictor values, we deemed assumptions to be satisfied for robust analyses (32). We assessed multicollinearity using type 2 tolerance, where a threshold of 0.1 was used, and BSL body weight was removed after causing a violation in initial models. We also performed 3 sets of sensitivity analyses and compared the results. First, we log-transformed non-normally distributed insulin predictor variables (insulin-30 and HOMA-IR) due to potential impacts of influence on the regression line. In a second sensitivity analysis, we removed data points with absolute values of studentized residuals ≥3 as potential outliers (33). Third, acknowledging possible limitations of using ratios (e.g., android:gynoid ratio) and percentages (e.g., fat mass divided by total body weight), we performed sensitivity analyses using regression methods to adjust for the denominators of ratios or proportions (e.g., gynoid fat or total body weight) as covariates (34). We did so by using variables in the numerator of the original ratio or percentage as outcomes (e.g., android fat change or fat mass change in kilograms), and variables in the denominator as covariates (e.g., gynoid fat change for android:gynoid ratio and total body weight change for percentage changes in body composition), along with BSL measures of the numerator and denominator and the same covariates as the original model except for ratios or percentages. As additional exploratory analyses, we assessed the association of OGTT insulin at other time points [0 (fasting), 10, 20, 60, 90, and 120 min] on changes in outcomes using the same type of GLM as described previously.

Results

Baseline characteristics for participants in FB4 and (FS)2 are presented in Table 1. In FB4, 82 participants completed the weight-loss run-in phase (Supplemental Figure 1). At BSL, median body weight was 98.9 kg, with 38.0 kg (38.8% of body weight) as fat mass and 58.1 kg (58.5%) as lean mass (Table 1). In (FS)2, 161 participants completed the run-in weight-loss phase (Supplemental Figure 2). At BSL, median body weight was 88.4 kg, with 35.1 kg (41.6% of body weight) as fat mass and 49.5 kg (56.0%) as lean mass (Table 1). Median (quartile 1, quartile 3) values for insulin-30 were 109 μIU/mL (73.7, 182 μIU/mL) for FB4 and 114 μIU/mL (77.2, 171 μIU/mL) for (FS)2. Values for HOMA-IR in FB4 (n = 69) and (FS)2 (n = 161) were 3.23 (2.02, 5.19) and 3.04 (2.09, 4.59), respectively.

Table 1.

Baseline characteristics of FB4 and (FS)2 participants

Characteristic1	FB4 (n = 82)	(FS)2 (n = 161)
Sex, n (%)
Men	46 (56.1)	49 (30.4)
Women	36 (43.9)	112 (69.6)
Race or ethnic group, n (%)
Race
White	63 (76.8)	126 (78.3)
Black	8 (9.8)	16 (9.9)
Asian	1 (1.2)	5 (3.1)
Multiple or other	10 (12.2)	14 (8.7)
Ethnic group
Hispanic	23 (28.0)	25 (15.5)
Non-Hispanic	59 (72.0)	136 (84.5)
Age at baseline DXA, y	34.8 (25.8, 42.4)	35.2 (23.6, 50.1)
Insulin-30, μIU/mL	109 (73.7,182)	114 (77.2,171)
HOMA-IR2	3.23 (2.02, 5.19)	3.04 (2.09, 4.59)
Body weight, kg	98.9 (88.2,110)	88.4 (77.8,103)
Body composition, kg
Fat mass	38.0 (32.8, 45.8)	35.1 (30.5, 42.8)
Trunk fat mass	19.2 (15.2, 21.3)	16.7 (13.9, 19.8)
Lean mass	58.1 (49.0, 67.1)	49.5 (42.0, 60.5)
Body composition (relative to total body weight), %
Fat mass	38.8 (34.5, 46.0)	41.6 (37.1, 45.2)
Trunk fat mass	19.0 (16.0, 21.2)	19.1 (16.9, 21.5)
Lean mass	58.5 (51.7, 62.7)	56.0 (52.0, 60.4)
Android:gynoid ratio	1.03 (0.950, 1.12)	1.01 (0.934, 1.10)

¹Values are medians (Q1, Q3), unless otherwise stated. FB4, Framingham, Boston, Bloomington, Birmingham, and Baylor study; (FS)2, Framingham State Food Study; insulin-30, insulin concentration at 30 min following a 75-g oral glucose load.

²HOMA-IR available for n = 69 in FB4.

In cross-sectional analyses at BSL (Table 2), higher insulin-30 was associated with greater body weight (4.33 kg per 100 μIU/mL increment in insulin-30, P < 0.001), fat mass (1.10% per 100 μIU/mL, P = 0.004), and less lean mass (–0.957% per 100 μIU/mL, P = 0.008), after adjustment for age and sex in (FS)2. Insulin-30 was also positively associated with central adiposity (trunk fat mass and android:gynoid ratio). Qualitatively similar but generally weaker cross-sectional associations were observed in FB4. Cross-sectional associations between HOMA-IR and body composition measurements were directionally similar to those observed for insulin-30 in both studies (Table 2).

Table 2.

Baseline associations of body composition measurements with insulin-30 and HOMA-IR in FB4 and (FS)2, adjusted for age and sex¹

	FB4		(FS)2
Baseline body composition measurements	β estimate ± SE2	P value	β estimate ± SE2	P value
Insulin-30
Body weight, kg	2.21 ± 1.55	0.16	4.33 ± 1.16	<0.001
Body composition (relative to total body weight), %
Fat mass	0.515 ± 0.511	0.32	1.10 ± 0.376	0.004
Trunk fat mass	0.711 ± 0.307	0.023	1.05 ± 0.238	<0.001
Lean mass	–0.406 ± 0.481	0.40	–0.957 ± 0.354	0.008
Android:gynoid ratio	0.032 ± 0.010	0.002	0.030 ± 0.010	0.003
HOMA-IR
Body weight, kg	2.04 ± 0.709	0.005	1.32 ± 0.288	<0.001
Body composition (relative to total body weight), %
Fat mass	0.601 ± 0.216	0.007	0.191 ± 0.097	0.050
Trunk fat mass	0.588 ± 0.127	<0.001	0.264 ± 0.060	<0.001
Lean mass	–0.528 ± 0.204	0.012	–0.151 ± 0.091	0.10
Android:gynoid ratio	0.021 ± 0.004	<0.001	0.012 ± 0.002	<0.001

¹Insulin-30 available for n = 82 for FB4 and n = 161 for (FS)2. HOMA-IR available for n = 69 for FB4 and n = 161 for (FS)2. β, unstandardized regression coefficient; FB4, Framingham, Boston, Bloomington, Birmingham, and Baylor study; (FS)2, Framingham State Food Study; insulin-30, insulin concentration at 30 min following a 75-g oral glucose load.

²β estimate ± SE expressed per 100 μIU/mL increment in insulin-30 or per unit increment in HOMA-IR.

After the weight-loss run-in phases, the mean ± SD weight loss was –14.8 ± 2.24% in FB4 and –10.5 ± 1.96% in (FS)2. Percentage changes in body weight were not associated with insulin-30 (P = 0.32 and P = 0.77, respectively), adjusting for age, sex, race, ethnicity, and BSL body weight. In prospective FB4 analyses of body composition (Figure 1, Table 3), higher insulin-30 was associated with a lesser decrease in fat mass (0.441 kg per 100 μIU/mL increment in insulin-30, P = 0.005; mean difference between first and fifth group of insulin-30: –1.20 kg) and a greater reduction in lean mass (–0.465 kg per 100 μIU/mL, P = 0.004; first compared with fifth difference: 1.27 kg). For participants with higher insulin-30, a smaller proportion of weight loss was fat (–3.37% per 100 μIU/mL, P = 0.003; first compared with fifth difference: 9.20%), related in part to a smaller decrease in central adiposity (trunk fat mass). As shown in Table 3, higher HOMA-IR was associated with a smaller decrease in fat mass (0.303 kg per unit increase in HOMA-IR, P < 0.001; first compared with fifth difference: –2.08 kg), a greater reduction in lean mass (–0.278 kg, P < 0.001; first compared with fifth difference: 1.91 kg), and a smaller proportion of weight loss as fat (–2.14%, P < 0.001; first compared with fifth difference: 14.7%). Associations between insulin-30 or HOMA-IR and body composition variables were qualitatively similar in direction but of lesser magnitude in (FS)2 (Table 3). For the other OGTT time points, associations with changes in body composition were qualitatively similar in direction for FB4 and (FS)2 (Supplemental Tables 1 and 2).

Figure 1.

Associations between baseline insulin-30 and changes in body fat mass, lean mass, and proportion weight loss as fat in FB4 (A–C) and (FS)2 (D–F). General linear models for associations (β and P values) are adjusted for the following covariates: age, sex, race, ethnicity, percentage weight loss at post–weight loss, and outcome variable at baseline. β, unstandardized regression coefficient; FB4, Framingham, Boston, Bloomington, Birmingham, and Baylor study; (FS)2, Framingham State Food Study; insulin-30, insulin concentration at 30 min following a 75-g oral glucose load.

Table 3.

Baseline insulin-30 and HOMA-IR¹ as predictors of change in outcome variables following weight loss in FB4 and (FS)2²

	FB4			(FS)2
Change in outcome variables between baseline and post–weight loss periods	β estimate ± SE3	P value4	Mean effect of first vs. fifth group5	β estimate ± SE3	P value4	Mean effect of first vs. fifth group5
Insulin-30
Body composition, kg
Fat mass	0.441 ± 0.153	0.005	–1.20 ± 0.417	0.194 ± 0.086	0.025	–0.491 ± 0.217
Trunk fat mass	0.277 ± 0.106	0.011	–0.755 ± 0.289	0.119 ± 0.080	0.14	–0.302 ± 0.202
Lean mass	–0.465 ± 0.158	0.004	1.27 ± 0.431	–0.126 ± 0.086	0.15	0.319 ± 0.219
Body composition (relative to total body weight), %
Fat mass	0.589 ± 0.174	0.001	–1.61 ± 0.475	0.260 ± 0.102	0.012	–0.661 ± 0.259
Trunk fat mass	0.302 ± 0.124	0.018	–0.823 ± 0.339	0.149 ± 0.092	0.11	–0.378 ± 0.232
Lean mass	–0.596 ± 0.174	0.001	1.63 ± 0.473	–0.248 ± 0.102	0.016	0.629 ± 0.258
Proportion weight loss as fat, %	–3.37 ± 1.08	0.003	9.20 ± 2.93	–2.00 ± 0.933	0.034	5.06 ± 2.37
Android:gynoid ratio	0.014 ± 0.007	0.047	–0.039 ± 0.019	0.017 ± 0.005	0.001	–0.044 ± 0.013
HOMA-IR
Body composition, kg
Fat mass	0.303 ± 0.071	<0.001	–2.08 ± 0.486	0.060 ± 0.021	0.005	–0.458 ± 0.161
Trunk fat mass	0.179 ± 0.050	0.001	–1.23 ± 0.344	0.050 ± 0.020	0.012	–0.385 ± 0.151
Lean mass	–0.278 ± 0.071	<0.001	1.91 ± 0.484	–0.056 ± 0.022	0.011	0.431 ± 0.167
Body composition (relative to total body weight), %
Fat mass	0.365 ± 0.078	<0.001	–2.51 ± 0.539	0.077 ± 0.025	0.002	–0.585 ± 0.188
Trunk fat mass	0.198 ± 0.060	0.002	–1.36 ± 0.414	0.051 ± 0.022	0.025	–0.388 ± 0.171
Lean mass	–0.370 ± 0.078	<0.001	2.54 ± 0.534	–0.073 ± 0.025	0.003	0.561 ± 0.187
Proportion weight loss as fat, %	–2.14 ± 0.483	<0.001	14.7 ± 3.32	–0.700 ± 0.223	0.002	5.36 ± 1.71
Android:gynoid ratio	0.006 ± 0.003	0.06	–0.041 ± 0.022	0.004 ± 0.001	0.007	–0.028 ± 0.010

¹Insulin-30 available for n = 82 in FB4 and n = 161 in (FS)2. HOMA-IR available for n = 69 in FB4 and n = 161 in (FS)2. β, unstandardized regression coefficient; FB4, Framingham, Boston, Bloomington, Birmingham, and Baylor study; (FS)2, Framingham State Food Study; insulin-30, insulin concentration at 30 min following a 75-g oral glucose load.

²Mean ± SD weight loss of –14.8 ± 2.24% in FB4 (n = 82) and –10.5 ± 1.96% in (FS)2 (n = 161).

³β estimate ± SE expressed per 100 μIU/mL increment in insulin-30 or per unit increment in HOMA-IR.

⁴P values represent associations between insulin-30 or HOMA-IR and change in outcome variable adjusted for the following covariates: age, sex, race, ethnicity, percentage weight loss at post–weight loss, and baseline measure of outcome variable.

⁵Insulin-30 was divided with quintiles into 5 groups, where the first group represents 20% of participants with lowest insulin-30 and the fifth group represents 20% of participants with highest insulin-30. Mean insulin-30 was 56.0 μIU/mL in the first group and 329 μIU/mL in the fifth group in FB4. Mean insulin-30 was 46.9 μIU/mL in the first group and 301 μIU/mL in the fifth group in (FS)2. Mean HOMA-IR was 1.42 in the first group and 8.29 in the fifth group in FB4. Mean HOMA-IR was 1.52 in the first group and 9.17 in the fifth group in (FS)2. Estimated mean ± SE for outcomes is reported for each corresponding concentration of insulin-30 and HOMA-IR.

For sensitivity analyses, results with the log-transformed insulin-30 and HOMA-IR remained at the same level of nominal significance (P < 0.05) as reported in Table 3. In FB4, change in android:gynoid ratio became significant after the log transformation of HOMA-IR (P = 0.036), whereas results for changes in android:gynoid ratio and percentage trunk fat mass became nonsignificant (P = 0.07 and P = 0.09, respectively) after the log transformation of insulin-30. When we removed studentized residuals with absolute values ≥3, insulin-30 became significantly associated with change in trunk fat mass (P = 0.009) and change in percentage trunk fat mass (P = 0.010) in (FS)2. There were no other changes in significance level in (FS)2 and FB4. For sensitivity analyses using covariates to adjust for the denominators of ratios or proportions as described previously, nominal significance levels (P < 0.05) remained the same as those for the primary analyses on change in kilograms. In FS(2), insulin-30 became significantly associated with change in lean mass (P = 0.037), whereas in FB4, insulin-30 became nonsignificant for changes in android:gynoid ratio (P = 0.46).

Discussion

In 2 controlled feeding trials, BSL insulin secretion predicted adverse changes in body composition independent of total weight loss. Participants with a higher insulin secretion lost less weight as fat mass, lost more weight as lean mass, and had greater central adiposity compared with those with lower insulin secretion. Similarly, adverse body composition changes were predicted by higher insulin resistance at BSL. These results suggest the BSL phenotypical differences in insulin dynamics may influence metabolic substrate partitioning and fat oxidation during energy restriction, with implications to the long-term success of weight-loss maintenance and chronic health.

Excessive reduction in lean mass during weight loss tends to decrease energy expenditure, requiring an increasingly severe energy restriction to sustain ongoing weight loss and a lower maintenance energy intake to prevent weight regain. Furthermore, loss of lean mass may elicit compensatory mechanisms, including increased hunger, that simultaneously promote gain in body fat (4., 5., 6., 7., 35, 36). With regard to chronic health, high body weight as defined by BMI is associated with increased risk of death from cardiovascular disease (37). However, BMI does not distinguish between fat and lean mass. In a pooled analysis of 7 prospective cohorts with median follow-up of 14 y, higher fat mass was associated with increased risk for mortality, whereas higher fat-free mass was associated with protection (38). These body composition exposures were independent of each other and stronger predictors than BMI. With regard to regional body fat distribution, greater central adiposity, as observed among our participants with high insulin secretion and high insulin resistance, is associated with increased cardiometabolic risk (39).

Our results are broadly consistent with those of previous studies relating individual differences in insulin dynamics to changes in body weight or body composition with weight-loss diets (17., 18., 19., 20., 24). These findings have specific relevance to the carbohydrate–insulin model (CIM) of obesity (11). The CIM predicts that 1) high insulin secretion characterizes a phenotype that responds poorly to energy restriction, related to adverse changes in energy expenditure and body composition; and 2) a low–glycemic load diet would attenuate the metabolic disadvantages of this phenotype. These predictions receive support from animal research (40), a 6-y observational study (41), a Mendelian randomization analysis (13), and several (17, 18, 27) but not all (42) clinical trials with a duration of 5–18 mo. Here, we found support for the first but not the second prediction. The associations between insulin secretion and adverse changes in body composition did not appear attenuated with the VLC diet of FB4 compared with the moderate-carbohydrate diet in (FS)2 (this comparison could be confounded by differences in cohort characteristics and study design features). Although we cannot ascertain the reason for these discrepant findings, study duration may be a critical factor.

For individuals habituated to a conventional, high-carbohydrate diet, imposition of energy restriction with a weight-loss diet represents a challenge to energy homeostasis, requiring upregulation of lipolysis and fat oxidation. This challenge is acutely greater with a ketogenic VLC diet, in which the body must also shift from a dependency on carbohydrate to ketones for much of the energy needs of the brain (43, 44). However, even with complete fasting, serum ketones rise slowly, reaching steady state after ≥3 wk (45). During this initial, adaptive period—either fasting or with consumption of a ketogenic diet—the body must catabolize protein from lean mass for gluconeogenic substrates, as demonstrated by a negative nitrogen balance for the first month on an energy-restricted ketogenic diet (46). Individuals with high insulin secretion, perhaps related to the suppressive effects of insulin on lipolysis, may have an exaggerated initial response to this metabolic stress. In the current study, the ∼3.5-mo duration of the run-in phase, with a rapid rate of weight loss, may not have been sufficient for adaptation to, and compensation for, the initial loss of lean body mass elicited by this phenotype.

Key strengths of this study include the same weight-loss run-in diet for all participants within each trial, facilitating unconfounded examination of interindividual differences; controlled feeding protocols to maximize adherence; substantial weight loss, improving the ability to detect meaningful differences in body composition; and measurement of body composition by DXA (considered state-of-the-art methodology). Several methodological and interpretative issues warrant consideration. First, the observational nature of the study precludes causal determination, especially with the cross-sectional analyses, and other factors associated with insulin dynamics may have mediated the observed findings. Furthermore, given the exploratory nature of the statistical analyses, P values are presented as unadjusted for multiple comparisons, which may result in type 1 error rates larger than the specified 0.05. Nevertheless, the consistency in outcomes within and between the 2 independent trials provides some reassurance regarding the possibility of false discovery. Second, the study may have been too short to examine long-term effects, limiting generalizability. However, the qualitative consistency between the cross-sectional and prospective associations suggests a chronic influence of insulin dynamics on body composition, exacerbated by energy restriction. Third, the measures of insulin dynamics may have both experimental and biological imprecision to varying degrees, limiting direct comparisons between them. Specifically, blood for insulin-30 is obtained during rapid changes in insulin concentration following oral glucose, such that minor variations in timing could affect this measure, in contrast to HOMA-IR, which is obtained during steady-state conditions. Conversely, HOMA-IR reflects systemic insulin resistance in the fasting state; it does not capture potentially important postprandial effects or differences between tissues, such as adipose compared with muscle.

In conclusion, individual differences in BSL insulin dynamics predicted changes in body composition independent of weight loss. High insulin secretion and insulin resistance may characterize phenotypes that respond adversely to calorie restriction, potentially predisposing to weight regain and elevated metabolic risk. Future studies are needed to confirm these exploratory findings with other populations, varying rates of weight loss, differing dietary composition, and employing longer intervention periods. Results of this line of investigation may inform understanding of the biological mechanisms influencing substrate partitioning and fat storage and also the design of more effective approaches to treat obesity, consistent with the aims of precision medicine.