Pregnancy induces pancreatic insulin secretion in women with long-standing type 1 diabetes

Study design and eligibility

Pregnant women with L-T1D (n=15) were recruited to a prospective cohort study at Uppsala and Örebro University Hospitals, Sweden. Pregnant patients were eligible for the study if they had been diagnosed with T1D for >10 years at conception and were <12 weeks of gestation. Research visits were performed on three occasions (described as mean value (range)): (1) gestation week 11.6 (10–13), (2) gestation week 34.4 (33–36) and (3) 6.2 (5–8) weeks post partum. Two women had a consecutive pregnancy shortly after the first delivery and were followed once again according to the study protocol. Therefore, they were considered as two new biological samples (total sample size, n=17). All women except two were breast feeding at the postpartum visit. Samples from a single time-point acquired from NP women with L-T1D (n=30) were used as a control for selected variables.

Clinical chemistry and oral glucose tolerance test

All visits were performed in the morning (08:00–10:00 hours) after an overnight fast. Peripheral venous blood was collected for analyses at Clinical Chemistry Laboratory, Uppsala University Hospital, and Department of Laboratory Medicine, Örebro University Hospital according to clinical routine. Triglycerides were not analyzed for patients included at Örebro University Hospital, rendering 11 samples for statistical analysis. EDTA-plasma was stored at −70°C prior to analysis of C-peptide and circulating proteins.

At each visit, an oral glucose tolerance test (OGTT) was performed. The OGTT comprised ingestion of a 50 g glucose solution (250 mL), instead of the prevalent 75 g glucose load, and blood was collected at T₀=before ingestion, T₁=10 min, T₂=30 min, T₃=60 min, T₄=90 min, and T₅=120 min. Long-acting insulin was administered as per routine but no rapid-acting insulin was administered prior to the OGTT.

Enzyme-linked immunosorbent assay

An ultrasensitive ELISA (Mercodia; Sweden, product no. 10-1141-01) was used to measure fasting and stimulated C-peptide. The detection limit for the assay was 2.5 pmol/L (0.0076 µg/L). All samples from both pregnant and NP women were stored at −70°C for <12 months prior to C-peptide measurements.

Proximity extension assay

Protein analysis was performed in undiluted plasma by multiplex proximity extension assay (PEA) at Olink Proteomics, Uppsala, Sweden. Two validated 92-plex panels, Olink CELL REGULATION and Olink IMMUNE RESPONSE (online supplemental tables 1 and 2), were used to measure analytes associated with biological processes such as lymphocyte activation, cytokine-mediated signaling, apoptosis, and differentiation.

Data analysis and statistics

The ratio between triglycerides and high-density lipoprotein-cholesterol was used to calculate insulin resistance at each visit, where a cut-off value of 1.65 was applied.19 Patients were considered positive for autoantibodies if anti-GAD was >5 IU/mL and anti-IA2 >7 kU/L. Area under the curve (AUC) for C-peptide values during OGTTs was calculated by the trapezoid rule. Statistical analyses were conducted with GraphPad Prism V.9 and IBM SPSS Statistics V.27. Sharpio-Wilk and Kolmogorov-Smirnov tests were applied to determine the data distribution of clinical variables. C-peptide data were transformed to 10-logarithmic values when a Gaussian distribution was not followed, and normality tests were again performed to evaluate transformed data. Two-tailed unpaired Welch’s t-test or two-tailed unpaired Mann-Whitney U test was used to compare data between pregnant participants and NP participants. A linear mixed effects model, followed by the Benjamini-Hochberg method for false discovery rate, were applied to compare longitudinal changes in clinical variables. For non-parametric longitudinal data, Friedman test and uncorrected Dunn’s post hoc test were used. The χ² test was applied to compare frequencies of categorical variables, such as C-peptide detectability and autoantibody positivity. A significance level of α=0.05 was defined for all tests.

For analysis of protein levels, data processing was performed in Olink NPX Manager. By using predetermined correction factors for each analyte, adjusted Cq-values were inverted and transformed into a relative arbitrary unit called Normalized Protein eXpression (NPX). NPX is expressed on a logarithmic scale (log2), meaning that a difference of +1 NPX corresponds to a concentration doubling. Statistical analyses were conducted with RStudio V.4.0.0. Principal component analysis (PCA) was applied to identify outliers and clusters. Two-sided unpaired Welch’s t-test was used to compare protein levels between NP individuals and both trimesters, separately. A linear mixed effects model was applied to assess longitudinal changes in protein levels between trimesters and the postpartum period. A maximum likelihood approach was chosen for parameter estimation, where visits and individuals constituted fixed and random effects, respectively. An unstructured covariance matrix was assumed. The Benjamini-Hochberg method was applied for correction of multiple testing in the model, and a false discovery rate <0.05 was considered significant. A network analysis was performed on the online platform STRING V.11.0 (STRING: functional protein association networks (string-db.org)). A full network type was chosen to identify both functional and physical protein interactions, the interaction score was set at a high confidence level (0.700), and edges illustrate the strength of data support (confidence). A network was considered to have significantly more interactions than expected at a protein–protein interaction (PPI) enrichment p value <0.05. UniProtKB (UniProt) and The Human Protein Atlas V.19.3 (The Human Protein Atlas) were implemented to obtain in-depth information on protein function and tissue expression. Explorations were made using UniProt numbers specified in Olink panels.

Demographic data for study subjects

The pregnant participants were in average 3.2 years older than the NP women (table 1).

The presence of GAD and IA2 autoantibodies in peripheral blood was similar between pregnant and NP women and did not change along the study. However, the number of IA2-positive women increased by one at each study visit. Total cholesterol, LDL, and triglyceride levels increased considerably to attain a peak at late gestational ages, which was ensued by a reduction after parturition. Nevertheless, total cholesterol and LDL remained more abundant 2 months post partum compared with NP women and during the first trimester. Due to decreased plasma triglyceride levels, the degree of insulin resistance was lower in pregnant women at first trimester compared with NP individuals with L-T1D. In parallel with elevated daily insulin doses, the insulin resistance increased during the third trimester. Interestingly, insulin doses were overall lower for women 2 months after parturition than in the NP group despite no differences in insulin resistance. The glycemic control improved during pregnancy and persisted after delivery in comparison with NP women with L-T1D. Glycated hemoglobin (HbA1c) was even lower during the third trimester than at early gestational ages. PRL was only measured in serum samples obtained from pregnant women. PRL levels increased during pregnancy and peaked at third trimester prior to a decline after parturition, although PRL remained elevated compared with the first trimester.

β-Cell function in women with L-T1D was recovered during gestation, but decreased after parturition

To assess the effect of pregnancy on β-cell function, fasting and stimulated C-peptide were measured. First, it was evaluated whether C-peptide becomes measurable during pregnancy in comparison with NP individuals. The frequency of women with L-T1D having detectable fasting C-peptide was consistently higher during gestation and after parturition (first trimester: 64.7%, third trimester: 76.5%, and 2 months post partum: 64.7%) than for NP women (26.7%, figure 1A). Despite an apparent variance within the NP group, the absolute concentration of fasting C-peptide was also higher in pregnant women than in NP individuals with L-T1D (figure 1B, table 1). This difference between the groups remained 2 months after parturition.

At a group level, fasting C-peptide levels did not change between gestational ages and the postpartum period (figure 1C). However, at the individual level most women with L-T1D had a rise in fasting C-peptide during pregnancy followed by a reduction or stabilization after parturition. Few participants had a nadir of fasting C-peptide that was recovered in the postpartum period. Both AUC and peak C-peptide after a glucose load increased during pregnancy (figure 1D,E). Moreover, the difference between the highest C-peptide concentrations and fasting values was enhanced between first and third trimesters (figure 1F). The glucose-induced insulin secretion declined after parturition, although the percental response during OGTT did not seem to be reduced (figure 1G). All C-peptide and glucose data from the OGTT is presented in online supplemental table 4.

The circulating protein levels of pregnant women with L-T1D is dynamic, but few functional clusters were related to the maternal immune system and potential modulators of β-cell function.

Plasma proteins associated with cellular and immunological processes were measured to investigate how they varies during gestation in women with T1D. Seventy-seven of 85 samples for Olink CELL REGULATION and 78 of 85 samples for Olink IMMUNE RESPONSE passed the quality control (online supplemental figure 1), rendering 16 samples from the first trimester, 15 samples from the third trimester, and 15 samples from the postpartum period for analysis. All 184 analytes were included in the analysis, as they were detected in ≥25% of the samples. First, a two-dimensional projection of the multivariate dataset and samples was computed by unsupervised PCA to identify outliers and potential clusters. Neither outliers nor clusters were found among the samples, meaning that the study groups had similar protein concentrations at a global level (online supplemental figure 2).

To investigate whether the protein composition diverges during pregnancy from a ‘T1D-background’, plasma samples from each trimester were compared with the NP individuals. Only 16 analytes were differently abundant between women during early gestation and NP participants with L-T1D, whereas the discrepancy between later gestational ages and the control cohort was enhanced (online supplemental table 3). A supervised network analysis was then performed to investigate functional associations between diverging analytes (n=49). A network comprising 11 edges between 16 nodes was identified, which could be divided into five clusters (PPI value=0.0008, figure 2A). Only 35% of the network was associated with maternal immune processes as visualized by functional enrichments. The largest cluster comprised several nodes that had a single interaction with the pleiotropic cytokine interleukin (IL)-6. Levels of IL-6 were higher in women with L-T1D during the third trimester than in the NP group (figure 2B). The chemokine CCL11 (eotaxin) was continuously less abundant in the pregnant women than in the NP individuals. Fibroblast growth factor (FGF)-21 was functionally connected to glucagon (GCG) signaling, and plasma levels of this factor were lower at first trimester compared with NP women with L-T1D. IL-6 was directly associated with amphiregulin (AREG), which is a protein that mediates PG production in placental tissue.20 21 AREG was also higher during the whole gestational period than in the NP group. Cytoskeleton-associated protein (CKAP)4 was associated with neutrophil-mediated immunity and was more abundant during pregnancy than in the NP participants. CKAP4 was connected to members of the C-type lectin (CLEC) domain family. CLEC4C was less abundant, while CLEC4D and CLEC7A were more abundant during the third trimester than in the NP women. A cluster classified as endocrine modification was identified, which comprised prokineticin (PROK)-1 and GCG (figure 2A and C). PROK-1 was more abundant throughout gestation in comparison with the NP participants, while plasma levels of GCG were only higher during the third trimester. A cluster related to antigen endocytosis was identified, where CLEC4A and lymphocyte antigen (LY)75 were reduced in plasma from women with L-T1D at third trimester (figure 2A and D). Two minor clusters related to diabetic complications (figure 2E) and neutrophil-associated hydrolases (figure 2F) were also found.

A linear mixed model was also applied to assess variations in the protein composition of individuals with L-T1D during pregnancy and after parturition. Temporal differences were identified for 79 analytes, where the largest contrast was observed between the third trimester and 2 months post partum (online supplemental table 3). The supervised network analysis showed that only 19 proteins formed 15 interactions at high confidence level, but this network was non-random (PPI value=0.0007, figure 3A). Only 35% of all nodes were related to the immune system, as visualized by functional enrichments within the network. IL-6 was again connected to CCL11, FGF-21, AREG, and CKAP4, but to some extent with different functional enrichments. IL-6 increased from first to third trimester, as mirrored by the comparison with the NP group, and remained stable after parturition (figure 3B). The nadir of CCL11 during gestation disappeared 2 months after parturition and became similar to levels found in the NP women. FGF-21 was associated with growth factor activity and cell death inhibition in this network (figure 3A), which increased in plasma until the third trimester and then declined during the postpartum period (figure 3B). AREG and CKAP4 peaked at third trimester prior to a decline after parturition. In contrast to the previous network analysis, IL-6 was connected to the T cell co-receptor CD28 and indirectly to integrin proteins. Soluble CD28 was continuously reduced during pregnancy, but it increased again after parturition to higher levels than at early gestational ages. PROK-1 and GCG still formed a functional interaction within this network, although parathyroid hormone-related peptide receptor was included (figure 3A). PROK-1 levels were constantly high in pregnant women with L-T1D prior to a nadir 2 months post partum, while GCG peaked at third trimester prior to a decrease after parturition (figure 3C). CLEC4A, LY75, CD83, and lysosomal associated membrane protein (LAMP)3 were classified into one cluster, as they are functionally related to antigen-presenting cells (APC). In addition, these four proteins had similar longitudinal changes in women with L-T1D: suppressed during pregnancy before a rebound after childbirth (figure 3D). Three enzymes catalyzing protein glycosylation, indirectly regulating leukocyte migration and extravasation, were predominant in plasma samples from the third trimester (figure 3A and E).