figure a

Introduction

Bariatric/metabolic surgery has been shown to improve glycaemic control and often induces remission of type 2 diabetes in obese individuals [1,2,3]. Previous studies have consistently demonstrated improvements in insulin sensitivity proportional to the amount of weight loss following bariatric surgery [4,5,6]. Given the importance of declining islet-cell function in the pathogenesis of type 2 diabetes [7] and the prominent effects of Roux-en-Y gastric bypass (RYGB) on the secretion of gut hormones involved in islet-cell secretory responses, we designed a study to better understand the effect of RYGB on islet-cell function in response to both oral (mixed meal [MM]) and intravenous nutrient stimulation. Study times after surgery included an early point at which weight loss was still ongoing and a later visit when weight was at or near its nadir [8].

We hypothesised that, because of altered nutrient flow through the gastrointestinal tract after RYGB, changes in key gut hormones known to influence islet-cell insulin secretion (such as glucagon-like peptide 1 [GLP-1] and possibly gastric inhibitory polypeptide [GIP]) would occur early and be maintained long-term. In addition, we hypothesised that in individuals with and without type 2 diabetes, insulin sensitivity would improve with postsurgical weight loss in both groups and that the proinsulin-to-insulin ratio would decrease, reflecting improved islet-cell insulin-processing efficiency. On the other hand, we hypothesised that the postsurgical stimulated-islet-cell secretory responses would differ. In non-diabetic individuals in whom hyperinsulinaemia compensates for insulin resistance allowing maintenance of normal glucose levels at baseline, insulin secretion would be reduced after surgery, reflecting decreased demand accompanying improved insulin sensitivity. In contrast, in individuals with type 2 diabetes, in whom defective islet function manifests in baseline hyperglycaemia, post-RYGB stimulated insulin secretory capacity would recover (increase). If the latter is shown to be true, these data would not only provide important insights into mechanisms of diabetes remission and prevention after RYGB, but would also place this procedure among a handful of interventions shown to reverse the decline in beta cell function that accompanies onset and progression of type 2 diabetes [9].

Methods

The Longitudinal Assessment of Bariatric Surgery (LABS) study is a prospective, longitudinal cohort study with three phases: LABS-1, LABS-2 and LABS-3 [10]. LABS-1 is complete and reported 30 day postsurgical adverse outcomes in nearly 5000 participants [11]. LABS-2 is a cohort of 2458 participants and has a primary goal of evaluating long-term efficacy of bariatric surgery [8]. The LABS-3 Diabetes substudy comprised a subcohort from the LABS-2 study recruited specifically to better understand the physiological mechanisms of glucose metabolism improvement following RYGB.

Participants and study visits

Candidates were approached for inclusion if they were scheduled for RYGB at a participating LABS-2 site at the University of Pittsburgh, University of Washington or Oregon Health & Science University (ESM Fig. 1). Individuals with diabetes were included if they were documented to have a fasting plasma glucose ≥7.0 mmol/l and ≤10 mmol/l or HbA1c ≥48 mmol/mol (6.5%) and ≤69 mmol/mol (8.5%) and were treated with lifestyle interventions alone or metformin and/or sulfonylurea. Exclusion criteria included type 1 diabetes, creatinine >150 μmol/l, treatment with insulin or other non-metformin/non-sulfonylurea diabetes medications, treatment with weight-loss medications and individuals unlikely or unwilling to comply with follow-up visits. Participants on metformin, without a pre-treatment glucose or HbA1c measurement but meeting criteria for diabetes, were assumed to be on this therapy for diabetes prevention, weight loss or polycystic ovarian syndrome, and were excluded. Control participants (those without diabetes) were included if their fasting plasma glucose was <7.0 mmol/l and/or their HbA1c was <48 mmol/mol (6.5%) and they were not taking any diabetes medications. They were matched to participants with diabetes based on age, sex and pre-surgical BMI. All studies were approved by the investigational review boards at each site and consent was obtained from each participant before enrolment.

In the week before each research study visit, participants met with a registered dietitian to receive instruction regarding a standardised diet consisting of ~30% total energy from fat, ~10–15% from protein and ~55–60% from carbohydrates. During this time, participants with diabetes also stopped their diabetes medications and self-monitored their capillary blood glucose levels at home. All other usual medications for comorbid conditions (i.e. hypertension, gastroesophageal reflux and hyperlipidaemia) were continued, as was treatment for obstructive sleep apnoea.

At the end of this pre-study week, participants presented to the clinical research centres at their respective institutions following an overnight fast on each of 2 days, scheduled not more than a week apart. On one day, fasting blood samples were drawn for lipids and proinsulin levels after which participants were given an MM (BOOST Plus, Nestlé Health Science, Epalinges, Switzerland; 1506 kJ (360 kcal), 45 g carbohydrate, 14 g fat, 14 g protein). Blood samples were obtained for glucose, insulin, C-peptide, glucagon, GLP-1 and GIP 15 min before and 0, 10, 20, 30, 60, 90, 120, 150, 180 and 240 min after meal consumption. On the other day, they underwent an insulin-modified frequently-sampled IVGTT (FSIVGTT) [12].

Following these baseline studies, participants underwent identical procedures as described above at 6 and 24 months after RYGB.

Biochemical analysis

Determination of glucose was performed by the hexokinase method using Roche reagents (Roche Diagnostics, Indianapolis, IN, USA). Levels of insulin and C-peptide were measured by a two-site enzymometric assay using Tosoh reagents on a Tosoh 2000 autoanalyzer (Tosoh Corp., Tokyo, Japan). Proinsulin levels were determined by radioimmunoassay using a Millipore kit (MilliporeSigma, Burlington, MA, USA). All lipid analyses were performed at the Northwest Lipid Metabolism and Diabetes Research laboratory, University of Washington, Seattle, WA, USA.

Total GLP-1, total GIP, total adiponectin and glucagon were measured by the Oregon Health & Science University Oregon Clinical and Translational Research Laboratory, Portland, OR, USA. To determine total GLP-1 and total GIP, blood was collected into EDTA vacutainers (prepared with 500 KU aprotinin and 10 μl dipeptidyl-peptidase-4 inhibitor per ml of whole blood) on ice. To determine adiponectin, blood was collected into EDTA vacutainers (prepared with 500 KU aprotinin/ml whole blood) on ice. These three analytes were measured using single-plex immunoassays from Meso Scale Discovery/Sector Imager (Gaithersburg, MD, USA). Glucagon was measured in blood collected in heparin vacutainers on ice, prepared with 500 KU aprotinin/ml whole blood, using an RIA from EMD Millipore (Billerica, MA, USA). The specificity of the glucagon assay was established by testing cross-reactivity with several gut hormones: oxyntomodulin (0.02%), 7-37 GLP-1 (none), 7-36 GLP-1 (<0.01%) and 1-37 GLP-1 (none).

Definition of diabetes and diabetes remission

Diabetes remission included both partial and complete remission, according to the American Diabetes Association Consensus Group recommendation [13], as HbA1c <48 mmol/mol (6.5%) (or fasting glucose ≤6.9 mmol/l if HbA1c was not available) and the absence of concurrent pharmacological therapy for diabetes.

Modelling for insulin secretory response and insulin sensitivity

Insulin sensitivity (SI), beta cell responsivity to glucose (Φ; measured as Φ total, Φ basal, Φ dynamic and Φ static) and hepatic insulin extraction (HE basal and HE total) were determined during the MM as previously described [14]. Disposition index (DI) during the MM was derived from the product of SI and Φ total. Variables of sensitivity to insulin and secretion response during the FSIVGTT, including SI, glucose effectiveness (SG) and acute insulin response to glucose (AIRglu), were modelled as previously described [15]. The DI during the FSIVGTT was derived from the product of SI and AIRglu.

Statistical analysis

Forty participants with type 2 diabetes and 20 without diabetes (60 in total) were estimated to provide an effective sample size for detecting modest-to-large effects on insulin sensitivity and stimulated-islet-cell secretion for within-person analyses and large-to-very-large effect sizes for between-group analyses based on prior studies [16, 17]. For continuous measures, participants in the diabetes vs no-diabetes groups were compared based on a Wilcoxon Rank Sum test to assess statistical significance of differences, unless normally distributed, in which case paired t tests were performed. For categorical measures, the frequency and percentage of each category was reported, and a Pearson χ2 test was used for statistical significance. For tables with small cell frequencies, Fisher’s exact test was used. Measures of insulin secretion and sensitivity from the FSIVGTT and MM over time were modelled using generalised linear models which included a heterogeneous autoregressive working correlation matrix to account for the correlation between successive measures. The model assumed a γ distribution for the measures to account for the highly skewed nature of the data. The mean measures at 6 months and 24 months were compared with baseline measures using a Wald t test. Missing values were treated as random occurrences. The generalised linear models used here utilised likelihood-based inference under which estimates are unbiased when data are missing completely at random.

Results

Sixty-two participants (40 with type 2 diabetes and 22 without diabetes) were enrolled and did not differ by group with respect to percentages of women and race, or by baseline age, BMI, percentage body fat or presence of obstructive sleep apnoea (Table 1). In participants with diabetes in whom the information was obtained (n = 24), the median duration of diabetes was 3.0 years. At baseline, HbA1c, fasting glucose, proinsulin and proinsulin:insulin ratios were all higher in the participants with diabetes than in the non-diabetic participants, but fasting insulin, C-peptide and total adiponectin levels were no different (Table 2).

Table 1  Baseline characteristics of participants with and without type 2 diabetes
Full size table
Table 2  Fasting metabolic variables prior to and after RYGB in participants with and without type 2 diabetes
Full size table

Median weight loss was similar in both groups at 6 months (24% vs 26% in participants with vs without diabetes, p = 0.27) and 24 months (29% vs 32% in participants with vs without diabetes, p = 0.57) (Fig. 1). Accompanying this weight loss, fasting glucose and insulin levels were lower than baseline at each follow-up visit in both groups (Fig. 2 and Table 2). Compared with baseline, two years after surgery HbA1c levels significantly decreased to 31 mmol/mol (5.0%) in non-diabetic participants and 33 mmol/mol (5.2%) in participants with type 2 diabetes (Table 2), remaining marginally higher in the diabetes group (p = 0.017). Compared with baseline, at both follow-up visits the time to peak glucose and insulin levels during the MM shifted to an earlier time point (30 vs 60–90 min) in each group (p < 0.001, Fig. 2). On the other hand, the peak postprandial glucose level increased compared with baseline in the non-diabetic participants (p < 0.001) but not in those with diabetes (Fig. 2 and Table 3). Similarly, peak postprandial insulin concentrations at each follow-up time point were higher than at baseline in both groups (Fig. 2 and Table 3). Despite these high early postprandial levels for both glucose and insulin following a meal, AUC levels had decreased by the visit at 6 months and remained lower than baseline at 24 months (Table 3).

Fig. 1
figure 1

 Median (bold black line) and individual (grey lines) BMI at baseline and at 6 and 24 months after RYGB in 40 participants with type 2 diabetes (a) and 22 participants without diabetes (b). Median total weight loss after 6 months was 24% in participants with type 2 diabetes and 26% in participants without diabetes (p = 0.27), and after 24 months was 29% in participants with type 2 diabetes and 32% in participants without diabetes (p = 0.57)

Full size image
Fig. 2
figure 2

 Mean ± SEM for glucose (a, b) and insulin (c, d) levels during the MM in participants with (a, c) and without (b, d) type 2 diabetes at baseline (circles) and at 6 months (squares) and 24 months (triangles) after RYGB

Full size image
Table 3  Baseline and postsurgical levels of glucose, insulin and gut hormones secreted following an MM prior to and after RYGB in participants with and without type 2 diabetes
Full size table

Fasting levels of GIP were not different and fasting GLP-1 levels were lower after surgery compared with baseline in both groups (although the difference in GLP-1 did not reach statistical significance in the non-diabetes group) (Table 3). However, the postprandial increases in levels of these hormones following an MM mirrored those of glucose and insulin in both groups (Fig. 3). Peak GIP shifted 30 min earlier and AUC GIP was lower at 6 and 24 months compared with baseline (Fig. 3 and Table 3). Peak GLP-1 levels increased seven- to eightfold, and AUC levels were significantly increased in both groups at 6 and 24 months after surgery compared with baseline (Fig. 3 and Table 3). Also in both groups, fasting glucagon levels decreased after surgery but peak levels following the MM increased and the increase in AUC approached statistical significance at both 6 months and 24 months after surgery (Fig. 3 and Table 3).

Fig. 3
figure 3

 Mean ± SEM for GLP-1 (a, b), GIP (c, d) and glucagon (e, f) during the MM in participants with (a, c, e) and without (b, d, f) type 2 diabetes at baseline (circles) and at 6 months (squares) and 24 months (triangles) after RYGB

Full size image

Glucose metabolism variables following an MM

Using a standardised MM to derive meal-related variables of insulin secretion and sensitivity [14], participants with and without type 2 diabetes exhibited reduced basal (fasting) beta cell responsivity (Φb) and slight increases in basal HE (Table 4) (p < 0.001 for all comparisons) at 6 and 24 months after RYGB. During the follow-up visit at 24 months, despite an overall 32% reduction in the amount of secreted insulin per unit increase of glucose during the dynamic phase of the meal, or dynamic beta cell responsivity (Φd), Φd remained significantly higher in the non-diabetic participants than in those with diabetes, in whom Φd did not significantly change after surgery (Table 4). Static beta cell responsivity (Φs, the amount of insulin secreted in response to glucose during the static phase compared with basal state) and total Φ (the overall beta cell responsivity derived from Φd and Φs) were higher in the participants without diabetes than in those with type 2 diabetes at each study visit, but did not significantly change in either group as a result of the surgery and accompanying weight loss (Table 4). Conversely, the total HE during the meal was greater in participants with diabetes than in those without, both at baseline and during follow-up, but did not significantly change in either group following surgery. Insulin sensitivity derived during the meal test (SI-MM) improved 6 months after surgery in both groups and continued to increase in participants with diabetes thereafter. However, the increase was no longer significantly different from baseline in participants without diabetes after 24 months, despite continued weight loss (Table 4). Despite this increase in SI-MM, insulin sensitivity remained lower in participants with diabetes than in those without at all three study visits. The meal-derived disposition index (DIMM,) representing insulin secretion response to a given level of insulin sensitivity (calculated as the product of total Φ and SI-MM), was lower in participants with type 2 diabetes than in those without diabetes, at each visit. In the diabetes group, DIMM was higher 6 and 24 months after surgery compared with baseline, whereas in the non-diabetes group it remained high and unchanged from baseline throughout (Table 4 and Fig. 4).

Table 4  Variables of insulin sensitivity and secretion derived from the MM test prior to and after RYGB in participants with and without type 2 diabetes
Full size table
Fig. 4
figure 4

 Change in DI after RYGB for participants with (white circles) and without (black squares) type 2 diabetes at baseline (BL) and at 6 months (6 mo) and 24 months (24 mo) after RYGB. Mean ± SEM for AIRglu vs SI derived from an FSIVGTT (a) and insulin secretion (Φ total) vs SI derived from an MM (b). Background solid and dotted lines represent percentiles derived from normal populations as previously described [15, 47]

Full size image

Glucose metabolism variables derived from FSIVGTTs

Derived measures of insulin secretion and sensitivity during the FSIVGTT (Table 5) showed that the acute insulin response to intravenously administered glucose, AIRglu, significantly increased in the participants with type 2 diabetes and decreased in the non-diabetic participants. Insulin sensitivity (SI-FSIVGTT) was greater at baseline in participants without diabetes than in those with diabetes, and both groups experienced significant improvements in insulin sensitivity such that by 24 months they were no longer significantly different from one another. Despite divergent changes in AIRglu, the FSIVGTT-derived disposition index (DIFSIVGTT) increased roughly ninefold in participants with diabetes and 2.5-fold in non-diabetic participants by 24 months after surgery compared with baseline (Table 5 and Fig. 4). Glucose effectiveness (SG) did not differ between the groups and was not affected by surgery or weight loss.

Table 5  Variables of insulin sensitivity and secretion derived from the insulin-modified FSIVGTT prior to and after RYGB in participants with and without type 2 diabetes
Full size table

Diabetes remission and characteristics of participants with diabetes that persisted or recurred after surgery

None of the non-diabetic participants developed diabetes during the 24 months of follow-up. Of the 40 participants with diabetes enrolled at baseline, 34 had data to determine diabetes status at 24 months. Of these, three were classified as having diabetes at 24 months (91% remission rate). Two of these three participants were taking diabetes medication at 6 months, whereas the third was not taking any diabetes medications at either of those time points.

Discussion

The current study was undertaken to compare 2 year changes in insulin sensitivity and secretory responses, and levels of gastrointestinal hormones, measured during oral (MM) and intravenous (FSIVGTT) challenges in obese participants with and without diabetes following RYGB.

Although previously reported [18,19,20,21], in the largest studied groups to date after RYGB we found that fasting glucose and insulin levels declined, HbA1c levels improved in both participants with and without diabetes to become nearly identical and not clinically meaningfully different, and a fundamental change in post-meal glucose appearance and disappearance was readily apparent. Through creation of a gastrojejunostomy that bypasses the pyloric valve, gastric emptying is accelerated [21,22,23] and peak postprandial glucose levels are shifted to an earlier time point after meal ingestion (30 min) in individuals with and without diabetes. Another notable finding was that, after surgery, peak postprandial glucose levels remained equal to baseline levels in the participants with type 2 diabetes and were actually higher than baseline in participants without diabetes. Additionally, postprandial glucose levels rapidly declined following these peaks in both groups, resulting in lower AUC glucose and HbA1c levels during follow-up (i.e. improved glucose tolerance and glycaemic control, respectively). Such marked glucose fluctuations (including early postprandial hyperglycaemia), referred to as glycaemic variability or ‘dysglycaemia’, have been suggested to play a role in diabetes complications independently of HbA1c levels [24]. However, several cohort studies that have included large numbers of post-RYGB participants have reported reduced microvascular and macrovascular diabetes-related complications and mortality compared with non-surgical control groups [25, 26], suggesting that despite this postprandial hyperglycaemia, the overall reductions in fasting and postprandial glucose exposure following RYGB are beneficial.

Plasma insulin responses, as well as those of GIP and GLP-1, were augmented in a pattern that mirrored the accelerated postprandial glucose appearance and disappearance following surgery. The seven- to eightfold increases in peak GLP-1 levels and GLP-1 AUC after surgery seen in both groups are thought to contribute to improved insulin secretion [27,28,29,30] and remained robust throughout the 24 months of follow-up. Counterintuitively, we found that despite large postsurgical weight loss and increases in peak insulin, glucose and GLP-1 levels [31, 32], the post-meal glucagon AUC did not change, and peak concentrations increased. This paradoxical meal-related glucagon response has previously been reported after RYGB in smaller studies [33,34,35,36] and in this report we confirm that this is not due to cross-reactivity of our assay with either GLP-1 or oxyntomodulin. Interestingly, our data are consistent with a recently demonstrated dual role for increased glucagon and GLP-1 levels in facilitating long-term weight-loss maintenance after RYGB through appetite regulation [37].

Insulin sensitivity measured during the FSIVGTT increased postoperatively in participants with and without diabetes. In agreement with our hypothesis and extending findings of a previous smaller study [38], the acute insulin response to intravenously administered glucose decreased in the non-diabetic participants but increased in those with type 2 diabetes up to 2 years after RYGB. Despite opposing stimulated insulin responses, RYGB restored diminished pancreatic insulin secretory capacity (as demonstrated by increases in DIFSIVGTT) in both groups, with the non-diabetic group showing improvement from the fifth to the 25th percentile of normative population values. A larger relative increase in DIFSIVGTT occurred in the diabetes group, but the pre-surgery secretory defect was so profound that even with an approximately ninefold increase, the DIFSIVGTT still remained below the fifth percentile of the normal values 24 months after surgery. This meant that, despite equal weight loss, similarly impressive gains in insulin sensitivity by 24 months and less ‘glucotoxicity’ [39], the insulin secretory capacity of participants with type 2 diabetes did not recover fully enough after 2 years to even match the pre-surgical values of non-diabetic participants. Likewise, proinsulin levels and insulin-processing efficiency improved in both groups after surgery but still showed persistent defects (higher levels of proinsulin and proinsulin:insulin ratio) in the type 2 diabetes group compared with the non-diabetic group at each visit.

Of note, 2 years after surgery, fasting plasma glucose or HbA1c levels reverted to ‘normal’ in over 90% of the cohort with diabetes at baseline, becoming virtually indistinguishable from the non-diabetic control group. As already discussed, however, this does not mean that the groups became metabolically equivalent. Indeed, recent reports have shown that a persistent impairment in insulin secretion is a major factor contributing to the failure to achieve or sustain diabetes remission after surgically induced weight loss [40, 41]. These reports and our data indicate that simple clinical indicators used to determine normal glucose tolerance or ‘non-diabetic’ status in patients who have undergone weight loss surgeries fail to adequately describe the metabolic heterogeneity within this group or adequately assign risk for subsequent long-term deterioration in glucose control.

Previous studies have shown improvements of insulin secretion in response to both intravenous and oral (glucose and MM) challenges within 3 months of RYGB and biliopancreatic diversion [42, 43]. Longer-term, however, we find that the changes in insulin sensitivity and insulin secretion derived from the responses to an MM differed from the responses to an FSIVGTT. Basal insulin responsiveness (Φb) derived from the MM improved in both groups. However, despite a robust increase in postprandial GLP-1 levels that persisted for 2 years after RYGB, which should correspond with improved beta cell glucose sensitivity [6, 44,45,46], it is somewhat surprising that the corresponding measures of meal-related dynamic insulin responsiveness (Φd) did not change in the type 2 diabetes group and decreased in the non-diabetic control group, and that total insulin responsiveness (total Φ) did not change in either group. One potential explanation for this discrepancy between oral and intravenous stimulated insulin secretory responses is that Φ values are normalised to glucose levels during modelling of the MM data, whereas calculations of AIRglu from the FSIVGTT are not. Another explanation is that the Φ and SI values derived during the MM were already close to population norms (see Fig. 4) in participants without diabetes, so did not have much room for improvement. It is also possible that improvements in SI-MM after surgery in the participants with type 2 diabetes group resulted in a reduced demand on islet cells, in which case maintenance of insulin responsiveness represents an improvement in the relative islet insulin secretory capacity, as reflected by a significant increase in DIMM in this group.

In summary, we demonstrate that for up to 2 years after RYGB, obese adults experience improvements in SI as well as improved insulin secretory responses to intravenously administered glucose as measured by DI with weight loss, despite opposite directionality of AIRglu responses in those with type 2 diabetes (increased) vs without diabetes (decreased). The gastric bypass procedure is therefore among a handful of interventions shown to reverse the decline in beta cell function that accompanies onset and progression of type 2 diabetes. For participants with type 2 diabetes, however, the baseline beta cell secretory defect was so profound that even with a marked postsurgical improvement, mean DI remained below the fifth percentile of a normative population 2 years after surgery. Combining our findings with recent reports that islet-cell recovery is an important determinant of diabetes remission status following bariatric surgery [40, 41, 46], consideration should be given to recommending RYGB as early as possible after the diagnosis of diabetes (the median diabetes duration in our study was only 3 years) or, more optimally, in obese individuals at risk for diabetes in whom we show that restoration of beta cell secretory capacity to variables within population norms is more likely. Our data also highlight a paradoxical increase in peak postprandial glucagon levels that, rather than negatively affecting postprandial glucose metabolism, supports an alternative, beneficial role for glucagon (along with markedly higher meal-stimulated GLP-1 levels) in appetite regulation and weight-loss maintenance after RYGB. Finally, studying the durability of beta cell recovery during longer-term follow-up will be vital for a deeper understanding of the mechanisms involved in type 2 diabetes improvement, remission and recurrence following RYGB.