Longitudinal clinical and proteomic diabetes signatures in women with a history of gestational diabetes

Abstract

We characterized the longitudinal serum protein signatures of women 6 and 10 years after having gestational diabetes mellitus (GDM) to identify factors associated with the development of type 2 diabetes mellitus (T2D) and prediabetes in this at-risk post-GDM population, aiming to discover potential biomarkers for early diagnosis and prevention of T2D. Our study identified 75 T2D-associated serum proteins and 23 prediabetes-associated proteins, some of which were validated in an independent T2D cohort. Machine learning (ML) performed on the longitudinal proteomics highlighted protein signatures associated with progression to post-GDM diabetes. We also proposed prognostic biomarker candidates that were differentially regulated in healthy participants at 6 years postpartum who later progressed to having T2D. Our longitudinal study revealed T2D risk factors for post-GDM populations who are relatively young and healthy, providing insights for clinical decisions and early lifestyle interventions.

Keywords: Diabetes; Metabolism; Proteomics.

Figure 1. Comparisons of proteomes in T2D and prediabetes versus healthy participants.

(A) Overview of clinical cohort and the serum proteomic workflow. Proteomic profiling was performed in serum samples collected at 6-year and 10-year postpregnancy follow-ups from participants diagnosed with GDM. Serum samples were prepared and analyzed by LC-MS/MS using data-independent acquisition (DIA). The total number of participants per subgroup is shown. (B and C) Volcano plots comparing the serum proteomes of T2D versus healthy participants at 6-year (B) and 10-year (C) follow-ups post-GDM. The log₂ fold-change in protein abundance is displayed on the x axis, and the –log₁₀ 2-tailed t test P value is displayed on the y axis. Color coding is based on P values (P < 0.05: in yellow), with directionality of difference in protein abundance (purple: significantly increased in T2D; blue: significantly decreased in T2D). Highlighted proteins are further discussed in the text. (D) Dysregulated proteins in T2D with the most pronounced magnitude of dysregulation (log₂ fold change > 0.5) or observed at both 6- and 10-year visits, as shown in B and C. Proteins significantly dysregulated in T2D compared with healthy controls were marked with a star. (E and F) Volcano plots of prediabetic versus healthy participants proteome at 6-year (E) and 10-year (F) post-GDM visits. Notations are as in B and C. (G) Log₂ fold change protein abundance in prediabetes compared with healthy participants at 6 and 10 years after GDM. Proteins found to be T2D associated in B or C are labeled purple and prediabetic-specific proteins are labeled black. Proteins significantly dysregulated in prediabetes compared with healthy controls were marked with a star.

Figure 2. Proteins associated with diabetes severity or clinical traits.

(A) Intensities of the selected proteins (1-way ANOVA, P_adj. <0.01) across subpopulations. Log₂ intensity of participants in each subpopulation were displayed as a box-and-whiskers plot showing median and interquartile range (IQR). Adjusted P value by Brown-Forsythe and Welch 1-way ANOVA tests and Dunnett’s T3 multiple comparisons test were labeled if P < 0.05. (B) Associations of post-GDM T2D markers from Figure 1, B and C, with clinical characteristics. Selected proteins with absolute Spearman correlation ρ > 0.4 with at least 1 of the clinical characteristics were displayed. Heatmaps show correlation coefficients between protein levels in serum and clinical characteristics of all participants at 10 years follow-up visits. Row clustering was based on log₂ intensity of the protein. (C) Examples of the correlations from B; abundance of PON3 protein with HDL and insulin resistance of all participants at 10-year time point.

Figure 3. Biomarker candidates of diabetes progression in longitudinal analysis.

(A) Left: Overview of clinical cohorts for clinical and serum proteomic longitudinal analyses. Among the healthy participants at the 6-year visit, subpopulations who developed either prediabetes (n = 19) or T2D (n = 9) were labeled as the diabetes-progressors, while nonprogressors stayed healthy at the 10-year visit (n = 41). Right: Progression to diabetes between 6-year and 10-year follow-up visits was associated with increase in BMI and HOMA-IR. The plots show within-participant changes in BMI (top) and HOMA-IR (bottom) with Wilcoxon P values of progressors versus nonprogressors. (B) Protein changes at 10 years versus 6 years within a participant across each subcohorts are represented with individual lines. PON3 (top) and PLTP (bottom) changed most in T2D-progressors between 6 and 10 years. Data represent mean ± SD of each year. *P_adj. < 0.05, **P_adj. < 0.01. (C) Protein change between 6 and 10 years within a participant across each subcohort is displayed in colors corresponding to the mean of the log₂ fold change (10 years/6 years) within a participant. Among the T2D or prediabetes-specific proteins from Figure 1, proteins mostly changed in T2D-progressor or prediabetes-progressor are shown (Supplemental Data File 5). A star is displayed in each cell if abundance of a protein was significantly changed at 10 years compared with 6 years by paired 2-tailed t test. (D) ROC curve and corresponding AUC statistics using random forest model, using a 2-fold stratified cross-validation and repeated process over 5,000 within-class shuffling to differentiate T2D-progressors and prediabetes-progressors from nonprogressors. (E) The 10 most discriminating features of T2D-progressors versus nonprogressors for model training.

Figure 4. Differential expression of nonprogressor versus T2D-progressor at 6-year visit, when subcohorts were healthy.

(A) Subcohorts overview for prognostic signature discovery. (B) Response to OGTT in healthy participants at 6 years showed early signs of their 10-year outcome. Data are shown as mean ± SD levels over time after the oral glucose challenge based on subgroup. Due to limited sample size, this analysis was performed on all participants who had 6-year OGTT data and known outcome at 10 years, and it was not restricted to participants who contributed serum for proteomics. N healthy at both 6 and 10 years (“H”, nonprogressor) = 51, N healthy at 6 years who progressed to prediabetes at 10 years (“P”, prediabetes-progressor) = 24, N healthy at 6 years who progressed to T2D at 10 years (“T”, T2D-progressors) = 11. Significant differences between progressors and nonprogressors at specific time points after OGTT (P < 0.05) is denoted with an asterisk. (C) Volcano plot displaying differential protein expression in healthy to T2D-progressors (n = 9) versus participants staying healthy at 6-year visit (n = 41). The log₂ fold change in protein level is displayed on the x axis, and the –log₁₀ Welch’s 2-tailed t test P value is displayed on the y axis. Color coding is based on P values (P < 0.05: in yellow), with directionality of difference in protein abundance (purple: significantly increased; blue: decreased in T2D-progressors). Significantly different proteins are highlighted and further discussed in the text. (D) IGFBP2 and SHBG protein change at 10 years versus 6 years; Left: Healthy nonprogressors at both time points. Right: Healthy to T2D-progressors. Log₂ intensity of participants in each subpopulation were displayed as a box-and-whisker plot showing median and IQR (Tukey method). (E) IGFBP2 and SHBG protein abundance with corresponding insulin resistance of all participants at the 6-year time point. Spearman correlation coefficient and P value were displayed. (F) ROC curve and corresponding AUC statistics to differentiate T2D-prognostic features. (G) The 10 most discriminating features as T2D-prognostic marker candidates.

Figure 5. Comparison and validation of PONCH findings with external studies.

(A) Correlation of T2D-associated proteins in PONCH study and corresponding proteins from validation cohort (Benjamini-Hochberg FDR < 30%). Spearman correlation coefficient and P value displayed tight correlation between the proteins from 2 independent studies. Data point of proteins downregulated in T2D external cohort (Benjamini-Hochberg FDR < 15%) were in blue and labeled in bold. Characteristics of the cohorts are presented in Table 2. (B) Strong correlation between the proteins from PONCH study versus 2 independent studies. Log₂ fold-changes (T2D/Ctrl) for T2D-associated proteins identified in the PONCH 10-year cohort and reported by Liu et al. (left) (40), and Diamanti et al. (right) (41). Left: Forty-nine proteins common across the 2 studies with the same directional change in T2D, highlighting those that increase with T2D (purple) and those that decrease (blue). Right: Comparison with Diamanti et al. Twenty-four proteins decreased/increased in a mixed sex subcohort of Diamanti et al. as observed in the PONCH cohort at 10 years. Data point of significantly changed proteins in Liu et al. were labeled in gray (P < 0.05). Signature proteins are highlighted in Table 3; CRP and ADIPOQ were labeled with their names. Spearman correlation coefficient and P value displayed. Characteristics of the cohorts are presented in Table 2.

Figure 6. Summary of experimental cohort and corresponding analyses.