Metabolite patterns differ in PBH.

Unsupervised hierarchical clustering of the 189 metabolites detected at all 3 time points and with significant FDR less than 0.05 (F test) is presented in the heatmap in Figure 1A, with a full list of cluster components provided in Supplemental Table 2 and full statistical analysis in Supplemental Table 3. As expected, ingestion of the mixed meal in nonsurgical controls resulted in decreased abundance of multiple lipid species and increased abundance of AAs at 30 minutes, with return to baseline at 120 minutes.

Open in a separate window

Figure 1

Metabolite abundance in individuals with PBH, Asx, and Ow/Ob in the fasting state and at 30 and 120 minutes after mixed meal ingestion.

(A) Heatmap of z-scored log₂-transformed metabolite abundance for the top 190 most altered metabolites between groups, with unsupervised hierarchical clustering of rows. Cluster annotations include lipid species (clusters 1–6), amino acids (AA; cluster 7), tryptophan metabolites (cluster 8), carnitines (CAR; cluster 9), and tricarboxylic acid (TCA) cycle metabolites (cluster 10). (B–D) Pathway enrichment analysis for PBH versus Asx individuals in the fasting state and 30 and 120 minutes after meal ingestion, respectively. Downregulated pathways are colored blue, while upregulated pathways are colored red. The –log₁₀ P value for enrichment is represented on the x axis; the vertical line indicates nominal P less than 0.05.

Exploratory analysis of these 189 metabolites revealed that 158 metabolites were altered in Asx individuals versus nonsurgical controls, reflecting the impact of surgery (FDR < 0.25). Sixty-two metabolites were altered in PBH versus Asx post-surgical individuals, and 178 metabolites were differentially abundant in both comparisons. Interestingly, 19 of these 178 had opposite directionality between PBH and Asx, including adenine, oxalate, and hydroxyphenylpyruvate (Supplemental Figure 2, A–C). These findings support the concept that PBH is not simply an “extreme” example of post-surgical metabolic perturbations.

Pathway analysis was performed using the Fry function of the rotation gene set test (ROAST) in the limma R package. Multiple pathways were enriched (FDR < 0.25) in the comparison of PBH and Asx post-surgical groups (Figure 1, B–D). In the fasting state, pathways downregulated in PBH included spermidine/spermine biosynthesis, tryptophan and methionine metabolism, phospholipid synthesis, and glycolysis (Figure 1B). After meals, both spermidine/spermine and methionine metabolism pathways remained downregulated at 30 minutes (Figure 1C). By 2 hours, enriched pathways were upregulated in PBH, including starch/sucrose and pyruvate metabolism, gluconeogenesis, and citric acid cycle pathways (Figure 1D). Additional comparisons are presented in Supplemental Figure 3 (PBH vs. Ow/Ob) and Supplemental Figure 4 (Asx vs. Ow/Ob).

Correlation of metabolites with glucose.

As expected, both AAs and lipids were highly correlated with glucose levels. The top metabolites that correlated with glucose in the fasting state (FDR < 0.05) (Supplemental Table 4) were AAs (Phe, Tyr, Met, Leu, Trp, His, Ala, and Asn), with similar patterns at 30 minutes. Additional metabolites correlating with glucose in the fasting state included NMMA (N-methylmalonamic acid), carnitine, UDP (uridine-5′-diphosphate), and oxalate. At 30 minutes, metabolites correlating with glucose included AAs (Leu, Met, Ile, Phe, Tyr, Gln, Pro, Arg, Trp, Gly, Ser, His, Val, Lys, Thr), citrulline, choline, NMMA, and ornithine, and the triglycerides C56:1, C48:0, and C50:0 (FDR < 0.05). There were no metabolites significantly correlating (FDR < 0.05) with glucose at 120 minutes (Supplemental Table 4).

Given that hypoglycemia typically develops 2–3 hours after meals, we asked whether metabolites present in the fasting state correlated with post-meal glucose. While relationships were not as robust as for the matched-time analyses, fasting levels of 17 metabolites were positively correlated with glucose at 120 minutes, including lactate, AAs (Gln, Pro, Trp, Tyr, betaine), carnitine, and multiple lipids (lysophosphatidylcholine [LPC] 18:0, 18:2, 20:3, and 22:6; lysophosphatidylethanolamine [LPE] 18:1 and 22:6) (FDR < 0.25) (Supplemental Figure 5).

Plasma levels of TCA cycle intermediates are increased in PBH.

Hypoglycemia ultimately may reflect interaction between excessive insulin secretion, inadequate counterregulatory hormone levels and/or action, reduced glycogen stores, and inadequate substrate or activity within gluconeogenic pathways. As noted above, multiple AAs were reduced in PBH, potentially contributing to reduced formation of intermediates such as pyruvate (alanine), acetyl-CoA (tryptophan, tyrosine, leucine, isoleucine, and phenylalanine), and succinyl-CoA (methionine, valine, and threonine). We interrogated metabolite patterns in central glucose metabolic pathways, projecting differences in PBH onto a pathway map at each time point (Figure 2).

Open in a separate window

Figure 2

Model demonstrating the metabolites of the glycolysis, gluconeogenesis, and TCA cycle pathways during mixed meal tolerance test.

Individuals with PBH and Asx are compared in the fasting state (A) and at 30 minutes (B) and 120 minutes (C) after mixed meal ingestion. Metabolites are colored blue if decreased (nominal P < 0.05), red if increased (nominal P < 0.05), and white if unchanged in PBH versus Asx.

In the fasting state (Figure 2A), PBH was characterized not only by decreased levels of glucose (20% lower, P = 0.001, FDR = 0.04) but also by decreases in gluconeogenesis intermediates such as glucose-6-phosphate (40% reduction, P = 0.02, FDR = 0.18) and pyruvate (80% lower, P < 0.0001, FDR = 0.003). Conversely, TCA cycle intermediates such as aconitate, isocitrate, and succinate were increased by 30% in PBH versus Asx in the fasting state (P < 0.05 for all, FDR < 0.25). Moreover, the ketone β-hydroxybutyrate was increased by 80%, as noted above (fold change = 1.8, P = 0.04, FDR < 0.25).

At 30 minutes after meals, only glucose and pyruvate remained lower (20% and 50% lower, P < 0.01, FDR < 0.1), and isocitrate remained higher (30% higher, P = 0.009, FDR = 0.14), in PBH versus Asx (Figure 2B). At 2 hours, 3-phosphoglycerate was 2-fold higher (P = 0.002, FDR = 0.002), while several TCA intermediates were again higher in PBH, including 30%–40% increases in aconitate, isocitrate, and malate (all P < 0.01, FDR < 0.01) (Figure 2C).

Lipid, ketone, and bile acid metabolism is perturbed in PBH.

Multiple lipid species were decreased in both surgical groups at 30 minutes, consistent with postprandial rises in insulin secretion. Additional lipid species differentially abundant in PBH versus Asx included LPE 20:4 and C22:6, LPC C22:6, and triacylglycerol (TAG) C50:5, C54:5, and C56:7, all of which were decreased by 30%–50% in PBH versus Asx (P < 0.05, FDR < 0.25) (Supplemental Figure 6, A–F). By contrast, C56:1 TAG was markedly increased in PBH versus Asx at 30 and 120 minutes (29- and 59-fold higher, P = 0.01, FDR < 0.25 for both; Supplemental Figure 6G). In agreement, abundance of C56:1 TAG was inversely correlated with glucose levels at both 30 (r = –0.67, P = 0.01) and 120 minutes (r = –0.65, P = 0.02) in PBH, but not in the fasting state.

Ketones, produced in the liver from 2-carbon products of β-oxidation, can be utilized by some tissues as an alternative fuel when glucose availability is limited, as with prolonged fasting. Interestingly, the ketone β-hydroxybutyrate was increased in the fasting state by 80% in PBH versus Asx (P = 0.04, FDR < 0.25) and more than 3-fold in comparison with Ow/Ob (P < 0.001, FDR = 0.01). These differences remained at 120 minutes (Supplemental Figure 6H).

Bile acids have been proposed as mediators of glycemic improvement after RYGB (4), potentially via effects to increase both incretin hormones and FGF-19 (4, 10). Fasting levels of the conjugated bile acid taurocholate were 2.5-fold higher in PBH versus Asx (P = 0.03, FDR = 0.22) (Supplemental Figure 6I). However, there were no significant differences in fasting or postprandial levels of other measured bile acids in PBH versus asymptomatic post-surgical participants (4, 9, 11).

Biomarkers of insulin resistance and T2D are decreased in PBH.

Previous studies have identified metabolite biomarkers of insulin resistance and risk for T2D (12, 13). These include AAs (branched-chain AAs, Phe, and Tyr), and triglyceride species characterized by lower carbon number and double bond content (14). Indeed, post-surgical individuals had reductions in these biomarker AAs and lipids, consistent with improved metabolism after RYGB, with further reductions in PBH. Moreover, α-hydroxybutyrate, an organic acid derived from α-ketobutyrate (produced by AA catabolism) and associated with insulin resistance, was 30% lower in PBH at both fasting (P = 0.03, FDR = 0.21) and 30 minutes after meals (P = 0.03, FDR = 0.22), as compared with Asx (Supplemental Figure 7A). Likewise, 2-aminoadipate, a product of lysine degradation associated with insulin resistance (13, 15), was also 6-fold lower in PBH versus Asx (P = 0.001, FDR = 0.04) in the fasting state (Supplemental Figure 7B).

Levels of B-complex vitamins are reduced in PBH.

B-complex vitamins play important roles as cofactors for key reactions in cellular metabolism. Several water-soluble B vitamins, including niacinamide (B₃) and pantothenate (B₅), were decreased 2-fold in the fasting state in PBH versus Asx (P < 0.05 for all, FDR < 0.25). Pantothenate was also 2-fold lower at 30 and 120 minutes (P < 0.01 for both, FDR < 0.05) (Supplemental Figure 7, C and D). Moreover, fasting glucose was positively correlated with levels of both vitamins B₃ (fasting: r = 0.48, P = 0.02, FDR 0.11) and B₅ (fasting: r = 0.52, P = 0.009, FDR = 0.07; 30 minutes: r = 0.49, P = 0.017, FDR = 0.14).

AA levels are reduced in PBH in the fasting state.

In the fasted state, 10 AAs were significantly reduced in PBH as compared with both asymptomatic and nonsurgical groups. Leucine, isoleucine, valine, alanine, tryptophan, phenylalanine, threonine, tyrosine, methionine, and asparagine were reduced by 17%–29% (P < 0.05, FDR < 0.25) in PBH versus Asx and Ow/Ob groups (Figure 3, A–J). After meal ingestion, AA levels increased markedly in both surgical groups, but only leucine, tryptophan, and phenylalanine remained lower in PBH as compared with Asx. AA levels for each participant are presented in Supplemental Figure 8. Notably, the reductions in AAs in PBH included both essential and non-essential AAs, and both gluconeogenic (e.g., Ala) and ketogenic (e.g., Leu) AAs. Arginine abundance was increased in both PBH and Asx compared with Ow/Ob (Figure 3K), while glutamine did not differ between groups (Figure 3L). The sum of all AAs was strongly positively correlated with glucose in the fasting state (r = 0.78, P = 0.001) and at 30 minutes (r = 0.59, P = 0.033); these correlations did not remain at 120 minutes (r = 0.07, P = 0.82) (Figure 3, M–O). The top-ranking specific AAs correlated with glucose levels both in the fasting state and at 30 minutes are provided in Supplemental Figure 9.

Open in a separate window

Figure 3

Abundance of 12 plasma AAs in PBH, Asx, and Ow/Ob individuals in the fasting state and at 30 and 120 minutes after mixed meal ingestion.

(A–L) Abundance of leucine (A), isoleucine (B), valine (C), alanine (D), tryptophan (E), phenylalanine (F), threonine (G), tyrosine (H), methionine (I), asparagine (J), arginine (K), and glutamine (L). (M–O) Correlation between glucose levels and sum of AA abundance in the fasting state (red, M) and at 30 minutes (red, N) and 120 minutes (red, O) after meal. Data are mean ± SD. *^,#,$P < 0.05; **^,##P < 0.01; ^###P < 0.001. Two-way ANOVA with Tukey’s multiple-comparison test was performed.

Serotonin levels are uniquely increased in the postprandial state in individuals with PBH.

Both pathway and individual metabolite analysis revealed differential abundance of tryptophan pathway intermediates in PBH, including tryptophan itself and its downstream metabolites kynurenic acid, xanthurenate, and serotonin (Figure 4). For example, tryptophan levels were 40% lower in the fasting state in PBH versus Asx (P = 0.003, FDR = 0.06), and 20%–30% lower at 30 and 120 minutes (P = 0.028 and 0.019, respectively, FDR < 0.25) (Figure 4A). Similarly, kynurenic acid was reduced by 70% at all 3 time points (P < 0.01 for all, FDR < 0.25), while xanthurenate was 8-fold lower in the fasting state (P = 0.003, FDR = 0.06) and 6-fold lower at 2 hours (P = 0.008, FDR < 0.01) (Figure 4, B and C).

Open in a separate window

Figure 4

Abundance of tryptophan, tryptophan-derived metabolites, and serotonin in PBH, Asx, and Ow/Ob individuals in the fasting state and after mixed meal ingestion.

Center: Schematic of tryptophan metabolism and downstream metabolites. (A–E) Abundance of tryptophan (A), kynurenic acid (B), xanthurenate (C), serotonin (D), and serotonin by ELISA (E). (F–H) Serotonin ELISA of individual participants in PBH (red), Asx (green), and Ow/Ob (blue) groups. Data are mean ± SD. *^,#P < 0.05; **^,##P < 0.01; ^###P < 0.001;****P < 0.0001 for PBH vs. Asx; ^####P < 0.0001 for PBH vs. Ow/Ob. Two-way ANOVA with Tukey’s multiple-comparison test was performed. Blue color of circles in the schematic indicates metabolites with decreased abundance, while white indicates unchanged abundance.

Serotonin levels were 10-fold lower in the fasting state in PBH, as compared with Asx (P = 0.019, FDR = 0.16) (Figure 4D). Interestingly, serotonin patterns in response to meal ingestion were distinct in PBH, increasing 3.2- and 4.7-fold at 30 and 120 minutes, respectively; by comparison, serotonin levels decreased after meal ingestion in Asx and nonsurgical groups.

To confirm alterations in serotonin dynamics in PBH, we analyzed serotonin using ELISA on samples collected in the fasting state and at 30 and 120 minutes after a meal originating from a newly recruited cohort described in Supplemental Table 5. Consistent with the metabolomics data, serotonin levels in individuals with PBH increased 1.9-fold after a meal (P < 0.001), whereas Asx and Ow/Ob patients had a 1.4- to 1.6-fold (P < 0.001 for both) decrease in serotonin levels over the same time course (Figure 4E). Serotonin levels for each participant are presented in Figure 4, F–H, demonstrating the distinct responses in all individuals with PBH.

Correlation analysis between metabolites and glucose levels during mixed meal tolerance test showed that tryptophan was correlated with glucose in the fasting state (r = 0.67, P < 0.001) and at 30 (r = 0.65, P < 0.001) and 120 minutes (r = 0.47, P = 0.023) (Supplemental Table 4). While there was no correlation between serotonin (measured by metabolomics assay) and glucose (Supplemental Table 4), plasma serotonin, measured by ELISA, was inversely correlated with glucose at 120 minutes after the meal in PBH (r = –0.78, P = 0.007) (Supplemental Figure 10).

To exclude the possibility that high plasma insulin or low glucose levels could contribute to observed increases in serotonin in the postprandial state, we analyzed serotonin levels in response to hyperinsulinemic-hypoglycemic clamps in PBH, Asx, and Ow/Ob participants (Supplemental Figure 16A). Consistent with the design of the clamp, plasma glucose levels at 100 minutes were similarly reduced in all groups (PBH: 51 ± 7 mg/dL; Asx: 51 ± 7 mg/dL; Ow/Ob: 53 ± 5 mg/dL; Supplemental Figure 16B). However, serotonin levels did not differ between groups, either in the fasting state (PBH: 103 ± 57 ng/mL; Asx: 152 ± 73 ng/mL; Ow/Ob: 174 ± 81 ng/mL) or during experimental hypoglycemia (PBH: 142 ± 72 ng/mL; Asx: 119 ± 89 ng/mL; Ow/Ob: 187 ± 53 ng/mL) (Supplemental Figure 16C). Similarly, serotonin levels did not change after insulin injection in mice (2 U/kg, i.p.) (Supplemental Figure 16, D and E) as compared with equal-volume saline at either 15 minutes (insulin: 430 ± 166 ng/mL; vs. saline: 344 ± 56 ng/mL) or 30 minutes (insulin: 344 ± 87 ng/mL; vs. saline: 351 ± 175 ng/mL) (Supplemental Figure 16F). Together, these results in both humans and mice suggest that hyperinsulinemia or hypoglycemia per se is not likely to mediate increases in postprandial serotonin in PBH.

Serotonin induces hypoglycemia in mice.

Given that serotonin levels were uniquely increased in the postprandial state in individuals with PBH (4.7-fold at 120 minutes), and not influenced by experimental hypoglycemia or hyperinsulinemia, we sought to determine whether serotonin could modulate glucose levels in vivo in healthy C57BL/6J mice. We injected wild-type mice with serotonin (20 mg/kg, i.p.) or an equal volume of saline (Figure 5A). In serotonin-injected mice, serotonin levels increased rapidly to a peak of 539 ng/mL at 30 minutes (Figure 5B). To determine the impact of serotonin on glucose metabolism, we measured blood glucose at baseline and at 15, 30, and 60 minutes. Glucose levels were significantly reduced in serotonin-injected male mice at 15 (–39%, P < 0.001), 30 (–43%, P < 0.002), and 60 minutes (–44%, P < 0.018) (Figure 5C), with reductions of similar magnitude in females (–29% at 15 minutes, P < 0.001; –48% at 30 minutes, P < 0.001; and –48% at 60 minutes, P = 0.005) (Supplemental Figure 11). We next assessed whether serotonin-induced reduction in glucose was mediated by an insulin-dependent mechanism; plasma insulin increased 5- and 2.5-fold at 30 and 60 minutes after serotonin injection versus saline (P = 0.004 and P = 0.002) (Figure 5D). Interestingly, serotonin administration also significantly increased plasma GLP-1 levels 1.6- to 2.1-fold at 15, 30, and 60 minutes after injection (P = 0.011, P < 0.001, and P < 0.001, respectively) (Figure 5E).

Open in a separate window

Figure 5

Serotonin reduces glucose and increases insulin and GLP-1 secretion.

(A) Experimental protocol showing exogenous administration of serotonin (20 mg/kg) or saline in C57BL/6J mice. (B) Serotonin levels achieved by serotonin injection. (C) Glucose levels after serotonin or saline injection. (D and E) Insulin levels (D) and GLP-1 levels (E) after serotonin or saline injection. In all panels, **P < 0.01, ***P < 0.001, ****P < 0.0001 by 2-way ANOVA with Tukey’s multiple-comparison test.

To mimic the postprandial glucose dynamics and exaggerated insulin secretion characteristic of PBH in mice, we developed an insulin-augmented mixed meal tolerance test. After a 4-hour fast, mice were gavaged with a liquid mixed meal (200 μL Ensure Compact, Abbott Laboratories) and injected with insulin (2 U/kg, i.p.; Humulin R, Lilly), before injection with serotonin or saline (Figure 6A). Consistent with prior experiments, serotonin (20 mg/kg) also reduced glucose by 38% versus saline in this setting, achieving levels below fasting level (serotonin, 80 mg/dL, vs. saline, 128 mg/dL, at 60 minutes; P = 0.021; Figure 6B).

Open in a separate window

Figure 6

Serotonin induces hypoglycemia during meal tolerance test; hypoglycemia is blocked by cyprophetadine and more specifically by ketanserin.

Reductions in glucose are blocked by the serotonin antagonist cyproheptadine and the specific serotonin receptor 2 antagonist ketanserin. (A) Scheme showing administration of cyproheptadine (50 mg/kg) or saline and, after 30 minutes, oral gavage with 200 μL of Ensure, insulin injection (2 U/kg), and administration of serotonin (20 mg/kg) or saline in C57BL/6J mice. MMTT, mixed meal tolerance test. (B) Glucose levels during experiment. (C) Scheme showing administration of ketanserin (5 mg/kg) and exogenous administration of serotonin (20 mg/kg) or saline in C57BL/6J mice. (D) Glucose levels after serotonin or saline injection. (E and F) Insulin levels (E) and GLP-1 levels (F) after serotonin or saline injection. In all panels, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001 by 2-way ANOVA with Tukey’s multiple-comparison test.

Given that serotonin increased GLP-1 levels, we asked whether GLP-1 receptor blockade would attenuate serotonin affects. We first confirmed that the GLP-1 receptor antagonist avexitide (30 mmol/kg) blocked semaglutide-induced increases in insulin secretion (Supplemental Figure 13B) and, in parallel, attenuated the hypoglycemic effects of semaglutide (Supplemental Figure 13C) and reduced GLP-1 levels (Supplemental Figure 13D). Next, we tested whether avexitide could block serotonin-induced hypoglycemia (Supplemental Figure 14A). Treatment of mice with avexitide rapidly increased glucose levels in comparison with saline-injected mice (P = 0.012 at 15 minutes). However, avexitide was unable to block serotonin-induced hypoglycemia (Supplemental Figure 14B). Likewise, avexitide did not affect serotonin-induced increases in insulin and GLP-1 (Supplemental Figure 14, C and D). Thus, these data suggest that GLP-1 receptor–mediated signaling is not likely to be a primary mechanism mediating serotonin-induced hypoglycemia.

Participants.

The Joslin Diabetes Center Institutional Review Board approved the study. Written informed consent was obtained from all participants. A detailed description of the previous clinical study that generated residual samples for analysis of the present study is provided in ref. 5. In brief, 3 study groups were analyzed (Supplemental Table 1): (a) PBH: 13 participants with history of RYGB and neuroglycopenia recruited from the hypoglycemia clinic at Joslin Diabetes Center; (b) asymptomatic post-RYGB group (Asx): 10 participants who underwent uncomplicated RYGB 2–4 years previously with no history of hypoglycemia; and (c) overweight/obese (Ow/Ob): 8 individuals without history of gastrointestinal surgery (control group). Blood samples were obtained after an overnight fast and at 30 and 120 minutes after consumption of a liquid mixed meal (Ensure, 9 g protein, 40 g carbohydrate, 6 g fat, 240 mL; Abbott Laboratories).

A new cohort (Supplemental Table 5) was recruited from 2020 to 2023 and included 15 participants with PBH, 15 Asx, and 10 Ow/Ob. Samples from this cohort were used to validate serotonin levels by ELISA during mixed meal tolerance test and during hyperinsulinemic-hypoglycemic clamps.

Metabolomics analysis.

Plasma collected at time 0, 30, and 120 minutes after mixed meal (5, 9) was used for untargeted semiquantitative metabolomic analysis using flow injection–tandem mass spectrometry, with detection of 189 metabolites.

Hyperinsulinemic-hypoglycemic clamp.

A primed, continuous infusion of insulin (80 mU/m²/min) was used together with a 20% glucose infusion, titrated every 5 minutes to gradually lower the glucose from fasting levels to a target of 50 mg/dL at a rate of 1 mg/dL/min (65); the target glucose was maintained for 30 minutes to ensure a consistent stimulus for counterregulatory responses.

Animal studies.

All animal studies were approved by the Institutional Animal Care and Use Committee of Joslin Diabetes Center. C57BL/6J male and female mice, age 6 weeks, were purchased from The Jackson Laboratory. Mice were maintained on a 12-hour light/12-hour dark cycle with ad libitum access to water and chow diet.

The impact of serotonin was assessed after a 4-hour fast; mice were injected with 20 mg/kg of serotonin (serotonin hydrochloride, Sigma-Aldrich) or an equal volume of saline intraperitoneally. Blood was collected at baseline (before injection) and at 15, 30, and 60 minutes after injection from the tail vein or via cardiac puncture at the time of euthanasia in EDTA-coated microtubes containing DPP-IV inhibitor (Millipore); blood glucose was measured using the Prodigy AutoCode meter.

To mimic the postprandial glucose dynamics and high insulin secretion characteristic of PBH, the effect of serotonin on glucose metabolism in response to an insulin-augmented mixed meal tolerance test was assessed. After a 4-hour fast, mice were gavaged with a liquid mixed meal (200 μL Ensure Compact, Abbott Laboratories) and i.p. injection of insulin (2 U/kg; Humulin R, Lilly). Serotonin (20 mg/kg) or an equal volume of saline was injected i.p., immediately after insulin injection. For some experiments, mice were treated with the nonspecific serotonin antagonist cyproheptadine (50 mg/kg; MilliporeSigma) i.p. or an equal volume of saline, 30 minutes before mixed meal. For all experiments, blood glucose was measured in the fasting state, at the time of gavage (time 0), and at 15, 30, and 60 minutes thereafter.

In additional studies, mice were pretreated with the GLP-1 receptor antagonist avexitide (30 mmol/kg; Eiger Biopharmaceuticals), the serotonin receptor 3 antagonist ondansetron (3 mg/kg; Cayman Chemical), and the serotonin receptor 2 antagonist ketanserin (5 mg/kg; MilliporeSigma).

To determine whether insulin could impact serotonin levels in mice, mice fasted for 4 hours were injected i.p. with insulin (2 U/kg; Humulin R, Lilly) or saline. Blood was collected via tail vein at time 0 and 15 and 30 minutes after injection.

ELISA and metabolite assays.

For human samples, plasma collected at baseline (fasting) and at 30 and 120 minutes after the mixed meal or plasma collected at baseline (fasting) and at 40 and 100 minutes after initiation of the hyperinsulinemic-hypoglycemic clamp was assayed by serotonin ELISA (ADI-900-175, Enzo) following the manufacturer’s instructions.

For mouse samples, plasma collected at fasting (time 0) and subsequently at 15, 30, and 60 minutes after injections was assayed by Ultra-Sensitive Mouse Insulin ELISA (90082, Crystal Chem), serotonin ELISA (ADI-900-175, Enzo), and mouse GLP-1 ELISA (81508, Crystal Chem). All sampled blood was collected via tail vein or cardiac puncture in EDTA-coated microtubes containing DPP-IV inhibitor (Millipore). All assays were performed according to manufacturers’ instructions.

Statistics.

All metabolite data were log-transformed, and those with more than 75% missing values were filtered out. The remaining missing values were imputed with half of the minimum values of the corresponding metabolites. Each sample’s abundances were scaled so that all samples had the same median abundance. Differential abundance was assessed by linear regression modeling with limma, which applies an empirical Bayesian approach to moderate each metabolite’s variance (66). Two-group comparisons were done by moderated t tests and multiple-group comparisons by moderated F tests (both 2-tailed, unpaired). P values were corrected using the Benjamini-Hochberg false discovery rate (FDR) (67). Metabolite clusters in a hierarchical dendrogram were detected using a variable height cutoff approach (68). The volcano plot was made using ggplot2 (69) and heatmap by pheatmap (70). Metabolite pathway sets were downloaded from the Small Molecule Pathway Database (SMPDB) (71), and metabolite set enrichment was tested using the limma ROAST method (72). All bioinformatics analysis was done in the R software, and graphs were generated using R and GraphPad Prism 9 (GraphPad Software Inc.).

For non-metabolic data, data were analyzed using either 1- or 2-way ANOVA, as specified in the figure legends. The graphical abstract was created with BioRender (biorender.com).

Study approval.

The study was approved by the Committee on Human Subjects, Joslin Diabetes Center (Boston, Massachusetts, USA). Written informed consent was obtained from all participants. Animal procedures were performed using protocols approved by the Institutional Animal Care and Use Committee of Joslin Diabetes Center.

We thank the participants in the human studies; and the Clinical Research Center nursing staff for the excellent support. This work was supported by National Institutes of Health (NIH) grant R01-DK121995 (to MEP), NIH grant P30-DK036836 (Diabetes Research Center grant, Joslin Diabetes Center), and Fundação de Amparo à Pesquisa do Estado de São Paulo grant 2018/22111-2 (to RFB).