Genetic Evidence of Causal Relation Between Intestinal Glucose Absorption and Early Postprandial Glucose Response: A Mendelian Randomization Study

Type 2 diabetes (T2D) is classically characterized by the combination of defects in insulin secretion and action, resulting in both fasting and postprandial hyperglycemia (1). Postprandial hyperglycemia is an early and key feature of T2D dysglycemia (2,3). A longitudinal clinical 14-year study associated 2-h postload blood glucose with T2D risk and all-cause mortality (4). In addition, 1-h postload blood glucose has been shown to be an independent predictor of progression to prediabetes in youth with obesity (5).

Early blood glucose response after carbohydrates intake is influenced by multiple physiological processes, including preprandial glycemia, gastric emptying, carbohydrate digestion, intestinal glucose absorption, incretin secretion, splanchnic glucose uptake, β-cell function, and insulin sensitivity (6). Of these, intestinal glucose absorption has received increasing attention with the advent of therapies targeting sodium-glucose cotransporters (7) and cross-sectional studies relating plasma glucose levels and diabetes status with intestinal glucose absorption profiles (8).

Intestinal glucose absorption involves primarily SGLT1, a rate-limiting transport protein, responsible for the active cotransport of glucose and sodium from the intestinal lumen in enterocytes (9). Complete loss-of-function mutations in SGLT1 result in severe intestinal glucose-galactose malabsorption (10). In contrast, intestinal expression of SGLT1 is increased in T2D (11), and positively correlated with early (1-h) postload glucose excursion in normoglycemic individuals (12). Double-tracer studies, estimating intestinal glucose absorption by the rate of systemic appearance of oral labeled glucose, showed a close association between intestinal glucose absorption and 1-h postload blood glucose levels (5). We previously reached similar conclusion by using d-xylose to estimate intestinal glucose absorption when modulating intestinal glucose-sodium cotransport, first by diverting ingested carbohydrates from bile-derived sodium with Roux-en-Y gastric bypass (13), or second, by reducing the apical expression of SGLT1 in enterocytes with acute metformin oral administration (14). Finally, the administration of sotaglifozine, a dual inhibitor of SGLT1 and SGLT2, induced a reduction in intestinal glucose absorption that paralleled a decrease in the 30-min postprandial plasma glucose level in healthy individuals (15). Altogether, available evidence indicates a reciprocal relation between SGLT1-mediated intestinal glucose absorption and early postprandial glucose excursion. However, the causal association between these two variables remains to be demonstrated.

The aim of the current study was to leverage the Mendelian randomization approach to explore the causal effect of intestinal SGLT1 expression on early postload glucose.

For that purpose, we used a loss-of-function SGLT1 haplotype (Asn51Ser, Ala411Thr, and His615Gln) as the genetic instrument, intestinal SGLT1 expression as the exposure, and early glucose response after an oral glucose load as the outcome.

The present retrospective study analyzed prospectively collected data from the Atlas Biologique de l’Obésité Sévère (ABOS) cohort (NCT01129297) (16) (Fig. 1).

All ABOS participants who underwent bariatric surgery between January 2006 and December 2017 were enrolled in the current study. All subjects provided written informed consent prior to inclusion. Ethical approval for the study was granted by the Comité de Protection des Personnes Nord Ouest VI (CP 06/49) (Lille, France). We provided the patients with basic educational materials and a research tutorial to help encourage familiarity with research concepts and terminology.

Clinical characteristics and biological data of the patients were prospectively collected prior to surgery. A 75-g oral glucose tolerance test (OGTT) was performed after an overnight fast with plasma glucose measurements at fasting, and 30 and 120 min after glucose ingestion as previously described (17). The 30-min postload glucose response (Δ30 glucose) was calculated as the difference between 30-min postload glycemia and fasting glycemia. Diabetes status was defined based on the previous history of diabetes, use of antidiabetic medication and/or fasting blood glycemia >7 mmol/L (126 mg/dL) or 2-h blood glucose >11.1 mmol/L (≥200 mg/dL) during OGTT, and/or HbA_1c ≥6.5% (48 mmol/mol) (18). Matsuda index, reflecting peripheral insulin sensitivity, was calculated using the following formula: Matsuda index = 500,000/√([C-peptide_0-min × glucose_0-min × 333] × [C-peptide_120-min × glucose_120-min × 333]) (19). Β-cell function was estimated using the acute insulin response or Δ30 insulin (insulin_30-min – insulin_0-min), and the insulinogenic index = (insulin_30-min – insulin_0-min)/(glucose_30-min – glucose_0-min) (20).

Genotyping was available for 1,342 participants. The analysis was conducted at the SNO&SEQ Technology Platform, Molecular Medicine in Uppsala, Sweden, using the Illumina Infinium assay (21,22). Results were analyzed using the software GenomeStudio 2.0.3.

In this study, we identified three specific nonsynonymous, protein-altering variants in the SLC5A1 gene, which encodes SGLT1: rs17683011 A>G (Asn51Ser), rs33954001 C>G (His615Gln), and rs17683430 G>A (Ala411Thr). These variants have been previously recognized by Seidelmann et al. (23) as functionally damaging missense variants offering protection against diet-induced hyperglycemia across various populations. The combination of Asn51Ser, Ala411Thr, and His615Gln mutations was considered as representative of the loss-of-function haplotype described by Seidelmann et al. (23). The term “SGLT1 loss-of-function haplotype” will be used throughout the rest of the study to refer to the combined presence of these three mutations.

Transcriptomic data were available in a subset of 508 ABOS participants who underwent Roux-en-Y gastric bypass between December 2010 and May 2017. Jejunal biopsies were collected 70 cm distal to the duodenum at the time of intestinal division to create the alimentary limb. Jejunum biopsies were immediately frozen in liquid nitrogen, then stored at −80°C until processing. RNA extractions were performed with the RNeasy mini kit (Qiagen) following the manufacturer’s instructions. RNA quality was controlled on an Agilent 2100 BioAnalyzer. The RNA integrity number of the 508 samples being >7, GeneChip Human Transcriptome Array 2.0 (Affymetrix) analysis was performed to determine the whole-genome transcriptomic profile as described (24). Two samples did not successfully pass Affymetrix QC standards. Raw data were normalized with APTtool version 2.11.6 (Affymetrix Power Tools; www.thermofisher.com/fr/fr/home/life-science/microarray-analysis/microarray-analysis-partners-programs/affymetrix-developers-network/affymetrix-power-tools.html) (apt-probeset-summarize function with options: gc correction, scale intensity, and rma at probeset level). Log2-normalized expressions were averaged per Gene Symbol (annotation provided by Affymetrix: NetAffx Annotation release 36, July 2016).

For differential expression analysis, carriers of the SGLT1 loss-of-function haplotype were matched with noncarriers using the R matchIt package (nearest method). Sex, age, and BMI were used for the propensity score calculation. Differences of gene expression between the two groups of patients were analyzed using the Linear Models for Microarray Data method and the Galaxy-based tool for Interactive ANalysis of Transcriptomic data tool (25).

The Wald ratio method was used to estimate the association of genetically proxied intestinal SGLT1 mRNA expression with the study’s primary outcome (Δ30 glucose), whereby the exposure-outcome estimate is derived from the variant-outcome association divided by the variant-exposure association. MR estimates were scaled to 1 SD. Analysis using multiple instruments for genetically predicted Δ30 glucose was performed using the inverse variance–weighted method, which provides a weighted average of variant estimates analogous to a fixed-effect meta-analysis (26).

The instrumental variables were validated across the three required assumptions for MR (27): First, we tested the association of the SLC5A1 variants with the exposure. Second, the variants should not share any common cause with the outcome. This assumption is not empirically verifiable, although a prior study of these SLC5A1 variants showed that they were independent of smoking, alcohol, or total energy intake (23). Third, variants should not affect the outcome except through the risk factor. Use of a missense variant with plausible biology reduces the risk of this bias.

We obtained instrumental-exposure association by linear regression calculation in R (version 4); each model was adjusted for age, sex, BMI, Δ30 insulin, and Matsuda index.

Variance explained (r²) and F statistic were calculated with the ivreg package in R (28). An F statistic > 10 was considered suggestive of adequate instrument strength (29). The study was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization guidelines (30).

Student t test, two-way ANOVA, and P value adjustments were performed using R software. Calculation of 95% CI for proportions was performed by the Wilson method (31). For categorical variables, χ² tests were performed. The Spearman correlation test was used for the correlations between clinical features and SGLT1 mRNA levels, and graphs were made using the ggpubr package. OGTT graphs were obtained using GraphPad prism (version 9). The P values were considered significant when <0.05. The P values of the Linear Models for Microarray Data transcriptomic analysis were adjusted by the Benjamini-Hochberg method (false discovery rate [FDR]). A gene was considered differentially expressed when the log2 fold change was ≥0.263 and the FDR P value was <0.05.

The study protocol and methods have been published (NCT01129297), as well as the ABOS cohort profile, and are unrestrictedly available. Clinical data sets generated during and/or analyzed during the current study are subject to national data protection laws and restrictions imposed by the ethics committee to ensure the data privacy of the study participants. They are not publicly available; however, they can be applied for through an individual project agreement with the ABOS scientific committee of the University Hospital of Lille, Lille, France.

The flowchart of the study participants is detailed in Fig. 1. Among 1,547 ABOS participants, 1,488 (96.2%) underwent OGTT, including 995 (64.3%) with 30-min glucose results available. The mean age of participants was 41.6 ± 11.7 years, 425 (27.5%) of the participants were men, and the mean BMI was 46.1 ± 9.0 kg/m². Among the participants, 565 (36.5%) had T2D, 537 (34.7%) were glucose intolerant, and 366 (23.6%) were normoglycemic (Supplementary Table 1).

We analyzed DNA samples from 1,342 participants (86.7%). Among them, we identified rs17683011 (Asn51Ser), rs17683430 (Ala411Thr), and rs33954001 (His615Gln) mutations in 172 participants (12.8%). These three variants, carried by the same patients, were in high linkage disequilibrium, comprising a risk haplotype with a minor allele frequency of 6.9% (Supplementary Table 2).

The clinico-biological traits of carriers and noncarriers of the SGLT1 loss-of-function haplotype are detailed in Table 1. Baseline characteristics, including sex, age, BMI, and overall glucose metabolism variables (fasting glycemia and insulinemia, HbA_1c, and Matsuda and insulinogenic indexes), did not differ significantly between carriers and noncarriers. Moreover, we did not observe significant differences in diabetes prevalence or treatment between carriers and noncarriers (Table 1).

We further studied the association between the SGLT1 loss-of-function haplotype and postload glucose response in 778 participants who underwent an OGTT with all three time points available, namely, at fasting and 30 and 120 min after glucose load, in the absence of insulin treatment. In univariate analysis, a tendency of a postload glucose response was idenfitied in haplotype carriers (Fig. 2A). The mean Δ30 glucose was 0.41 mmol/L lower in the 102 (13.1%) carriers of the SGLT1 loss-of-function haplotype than in the 676 (86.9%) noncarriers (Table 1 and Fig. 2B) (P < 0.05). There was no significant difference in insulin response between the two subgroups. In multivariable analysis, following adjustment to other confounding variables including age, sex, BMI, Δ30 insulin, and Matsuda index, the SGLT1 loss-of-function haplotype was associated with a significant reduction of Δ30 glucose (−0.314 mmol/L; 95% CI −0.503; −0.126) (P < 0.001) (Fig. 2C). In sensitivity analyses, we also found a significant association between SGLT1 loss-of-function haplotype and 30-min glycemia (Supplementary Fig. 1A), but not with 120-min glycemia (Supplementary Fig. 2A), nor Δ120 glucose (Supplementary Fig. 3A).

Intestinal transcriptome and SGLT1 loss-of-function haplotype status were both available for 374 ABOS participants: 55 carriers (12.5%; 95% CI 9.7; 15.9) and 386 noncarriers (87.5%; 95% CI 84.1; 90.1). Carriers of SGLT1 loss-of-function haplotype expressed lower intestinal levels of SGLT1 mRNA (P < 0.001) when compared with noncarriers (Fig. 3A). In multivariable analysis, the SGLT1 loss-of-function haplotype was the only variable significantly associated with intestinal SGLT1 mRNA expression (Fig. 3B). The SGLT1 loss-of-function haplotype was associated with a decrease of intestinal SGLT1 mRNA expression of −0.991 log2 (95% CI −1.28; −0.702; P < 0.001). None of the confounding variables, including age, sex, and BMI, was significantly associated with SGLT1 mRNA expression.

To achieve an in-depth understanding, an additional analysis focusing on the entire intestinal transcriptomic signature was undertaken. We explored the overall intestinal transcriptomic signature in a propensity score study comparing 55 carriers of the SGLT1 loss-of-function haplotype with 55 noncarrier individuals matched by age, sex, and BMI (Supplementary Table 3). The only gene that was differentially expressed between the two groups was SGLT1: the intestinal SGLT1 mRNA level was significantly lower in carriers of the loss-of-function SGLT1 haplotype (log2 fold change −0.287, FDR adjusted P value < 0.001) (Fig. 3C).

We analyzed the association between intestinal SGLT1 mRNA expression and Δ30 glucose level in 449 patients for whom both the intestinal transcriptome and complete OGTT data were available. As shown in Fig. 4A, the Δ30 glucose level was positively correlated with the level of intestinal expression of SGLT1 mRNA (r = 0.094, P = 0.047). Multivariable analysis (Fig. 4B) showed that the SGLT1 mRNA level was significantly associated with the Δ30 glucose level, each decrease of 1 SD of SGLT1 mRNA expression (0.234 log2) resulting in a reduction of −0.10 mmol/L of Δ30 glucose level (95% CI −0.189; −0.01; P = 0.037). In sensitivity analyses, we found a similar and significant association between SGLT1 expression (Supplementary Fig. 1B) and 30-min glycemia, but not with 120-min glycemia (Supplementary Fig. 2B) or Δ120 glucose (Supplementary Fig. 3B).

For the MR analysis, the intestinal SGLT1 expression level was used as the exposure, the SGLT1 loss-of-function haplotype as the genetic instrument, and Δ30 glucose as the outcome. Genetically predicted intestinal SGLT1 mRNA expression (instrumented using the SGLT1 loss-of-function haplotype with a mean F statistic of 35.838, r² = −0.07) was significantly and positively related to the Δ30 glucose level (Fig. 5). A reduction of genetically predicted SGLT1 mRNA expression by 1 SD (0.234 log2) was associated with a reduction in the Δ30 glucose level of −0.353 mmol/L (95% CI −0.701; −0.01; P = 0.046). Using the Durbin-Wu Hausman test, we found no significant difference between the effect of 1 SD of observed SGLT1 mRNA expression on Δ30 glycemia and the effect of its MR estimate (P = 0.1).

The current study provides genetic evidence that variance in intestinal expression of SGLT1 has a causal impact on early (30-min) postload glucose response, independent of any defect in insulin secretion and insulin action.

To the best of our knowledge, this study is the first to use Mendelian randomization to explore the relationships between intestinal glucose absorption and metabolic traits. In observational epidemiological studies, various risk factors have been associated with the alteration of postprandial glucose response. Likewise, our study confirmed the association between intestinal expression of SGLT1 and early postprandial glucose response already reported by Fiorentino et al. (12). However, the causal nature of this association, and thus the suitability of SGLT1 as an effective intervention target, remained unclear. Here, we leveraged the Mendelian randomization method to infer the causal contribution of SGLT1-mediated intestinal glucose transport to early postprandial glucose response.

The selected genetic instrument was a loss-of-function haplotype that combines two prevalent nonsynonymous SGLT1 mutations (rs17683011/Asn51Ser and rs33954001/His615Gln), in high linkage desequilibirum, with an allele frequency of 6.9%. Each of them has been previously associated with a partial reduction (30%) of active sodium-glucose transport in transfected oocytes as compared with native SGLT1 (32). In humans, the same missense mutations have been also associated with a 25% reduction of the incidence of glucose intolerance (23).

Importantly, the proposed loss-of-function SGLT1 haplotype complied with the three core assumptions underlying causal inference with an instrumental variable.

First the studied haplotype was strongly associated with a reduction of intestinal expression of SGLT1 mRNA, the study’s exposure (F statistics > 10). This indicates that this loss-of-function SGLT1 haplotype that combines two cis SNPs significantly impacts the transcription of SGLT1 in the intestine, likely resulting in a decreased expression of a functional transporter, consequently limiting intestinal glucose uptake (relevance assumption).

Second, this haplotype was not associated with potential confounders of the exposure-outcome relationship (independence assumption). Indeed, we found no significant difference between carriers and noncarriers of the SGLT1 haplotype in terms of clinico-biological characteristics and diabetes prevalence at the time of their enrollment in the ABOS cohort, except for Δ30 glucose (Table 1). The lower postload glucose response observed in the haplotype carriers was independent of age, sex, and BMI, as well as insulin secretion and insulin sensitivity (Fig. 2).

Third, the SGLT1 loss-of-function haplotype was not differentially associated with the expression of any other intestinal transcripts, including those coding for other glucose transporters, including SGLT2, GLUT1, or GLUT2 (Fig. 4). This suggests that the studied haplotype was merely associated with the outcome through the study exposure (SGLT1 intestinal expression), and mitigates the risk of horizontal pleiotropy (exclusion restriction assumption).

Overall, the estimated reduction of the Δ30 glucose response associated with a 1-SD reduction of intestinal expression of SGLT1 was consistent in the observational study and in the Mendelian randomization instrumented by the SGLT1 haplotype (0.097 mmol/L and 0.353 mmol/L, respectively; P = 0.10, Durbin-Wu Hausman test).

Our study has certain limitations. First is its noninterventional, cross-sectional design. Second, this is a retrospective study including mostly White individuals with obesity who have undergone bariatric surgery, with a potential selection bias that may have influenced the observed rates of SGLT1 mutations that might not be indicative for the general population. Of note, the allele frequency of the SGLT1 loss-of-function haplotype in our cohort (6.9%) was similar to the frequency reported by Seidelmann et al. (23) in European-American participants of the Atherosclerosis Risk in Communities Study cohort (6.7%) but higher than in African American subjects (1.5%).

Furthermore, we did not directly measure intestinal glucose absorption with the gold standard technique using labeled glucose tracers. As the exposure, we rather used the measured intestinal expression of SGLT1, and the protein level was not determined. However, similar patterns between the intestinal expression of SGLT1 mRNA and SGLT1 protein have been previously described (12). This suggests that the combined genetic variations of this haplotype either directly or indirectly affect the transcription or stability of the SGLT1 mRNA, leading to a modified expression and potentially altered function of the protein. Likewise, a positive relation between SGLT1 mRNA levels and 1-h postload glucose has been previously reported elsewhere (33). Additionally, we did not investigate gastric emptying, which can impact 30-min postload glucose (34). However, a recent study also showed an association between 30-min postload glucose and intestinal glucose absorption, independently from gastric emptying (6). We did not study the potential effect in SGLT1 haplotype carriers on GLP-1 and PYY secretion. As less glucose is absorbed, more glucose is allowed to reach the distal intestine and promote the secretion of incretins, which coul d also account for better-observed glucose control (35). Finally, the reduction in key baseline glycemic traits observed in carriers of the loss-of-function SGLT1 haplotype carriers did not reach statistical significance in our study, in contrast to the results reported in a larger cohort by Seidelmann et al. (23). This lack of significance is likely attributable to limited statistical power. Of note, very strong evidence also links loss-of-function SGLT1 haplotypes with a better glucose metabolism in the general population. In the T2D portal database (https://t2d.hugeamp.org/), rs17683011 (Asn51Ser) is associated with a lower HbA_1c level (P = 1.18e-21, β −0.0171, in 1,092,580 individuals) and a decreased prevalence of T2D (P = 2.97e-8, odds ratio = 0.9597 in 1,458,670 individuals).

In summary, based on comprehensive data available from a large single cohort, this Mendelian randomization study provides genetic evidence that differences in intestinal expression of SGLT1 causally influence 30-min postload glucose response. This could have important clinical implications: first, to better characterize patients at risk for impaired glucose tolerance and T2D, and, second, to incorporate intestinal glucose absorption for applying precision medicine for preventing and treating T2D. This hypothesis is further supported by the clinical benefits of the administration of a specific SGLT1 antagonist (35), which may represent a particularly interesting therapeutic option in patients displaying higher levels of intestinal glucose absorption. Our results also support further research to develop alternative strategies aiming at modulating the intestinal sodium-glucose cotransport, through the reduction of ingested sodium (36), the distal intestinal diversion of sodium-rich bile and other digestive fluids (13), or the reduction of SGLT1 trafficking in enterocytes (14).

Funding. This work was supported by the Programme d’Investissement d’Avenir (ANR-16-RHUS-0006_PreciNASH; European Genomic Institute for Diabetes, ANR-10-LABX-0046), Lille University (WILL-CHAlRES-23-001), Agence Nationale de la Recherche (EQU202303016330 PATTOU), EU Horizon 2020 research and innovation program (Innovative Medicines Initiative 2, project Stratification of Obesity Phenotypes to Optimize Future Therapy 875534), and Fondation Francophone pour la Recherche sur le Diabète (FFRD 2015).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. V.R., S.L., and F.P. contributed to the study concept and design. S.P., V.R., P.L., B.S., S.L., and F.P. contributed to drafting the manuscript. S.P. and P.B. contributed to performing the statistical analysis. All authors contributed to the analysis and interpretation of the data, critically revised the manuscript for important intellectual content, and approved the final version for submission. All authors had full access to all the data in the study and took responsibility for the integrity of the data and the accuracy of the data analysis. F.P. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.