Placental RNA sequencing implicates IGFBP1 in insulin sensitivity during pregnancy and in gestational diabetes

Reduced insulin sensitivity (or greater insulin resistance) is a hallmark of normal physiology in late pregnancy and also underlies gestational diabetes mellitus (GDM) pathophysiology. We conducted transcriptomic profiling of 434 human placentas and identified a strong positive association between insulin-like growth factor binding protein 1 gene (IGFBP1) expression in the placenta and insulin sensitivity at ~ 26 weeks’ gestation. Circulating IGFBP1 protein levels rose over the course of pregnancy and declined postpartum, which together with high placental gene expression levels, suggests a placental source. Higher circulating IGFBP1 levels were strongly associated with greater insulin sensitivity (lesser insulin resistance) at ~ 26 weeks’ gestation in the same cohort and two additional pregnancy cohorts. In addition, low circulating IGFBP1 levels in early pregnancy predicted subsequent GDM diagnosis in two cohorts. These results implicate IGFBP1 in the glycemic physiology of pregnancy and suggest a role for placental IGFBP1 deficiency in GDM pathogenesis.


Introduction
Gestational diabetes mellitus (GDM) affects 1 in 7 pregnancies worldwide 1 and is associated with maternal and offspring adverse health outcomes during pregnancy, at delivery, and over the life course 2 .
Prior research has established that a defect in insulin sensitivity (i.e., greater insulin resistance) contributes to GDM 3,4 .In addition, we and others [5][6][7][8][9][10] have previously shown that among individuals with GDM, those with the lowest insulin sensitivity (insulin resistant GDM) have the greatest risk of hyperglycemia-associated pregnancy complications, suggesting that reduced insulin sensitivity is a key contributor not only to GDM itself, but also to the negative health outcomes that accompany it.
The placenta is the major driver of marked changes in insulin physiology during pregnancy, including the drastic decline in insulin sensitivity, which occurs even in those without GDM.This has been attributed to hormonal factors released by the placenta that lead to insulin resistance 11 .Yet, the speci c placental circulating factors that mediate the profound change in insulin sensitivity during pregnancy are still unknown, and the classically-implicated pregnancy hormones (e.g., human chorionic gonadotropin, human placental lactogen, and placental growth hormone) are poorly correlated with insulin sensitivity in pregnancy 12 .Better understanding of the placental factors driving the pregnancy-related decline in insulin sensitivity could lead to novel therapeutic approaches to hyperglycemia, early identi cation of those at risk of developing GDM, and recognition of those most likely to have GDM-related pregnancy complications.Despite the hallmark reduction in insulin sensitivity in all pregnancies, most pregnant individuals do not develop GDM.This phenomenon suggests that additional factors, yet unknown, may contribute to the maintenance of euglycemia in pregnancy.Indeed, a variable improvement in insulin sensitivity has been reported in early pregnancy in several studies 13,14 .A systematic search for placental factors that are related to insulin sensitivity in pregnancy may also uncover those that improve it.
The overarching goal of this study was to discover novel placental factors implicated in physiologic changes in insulin sensitivity during pregnancy and that contribute to GDM pathophysiology.We conducted a placental genome-wide transcriptomic study using RNA sequencing (RNA-seq) to identify genes whose expression in the placenta was associated with insulin sensitivity in pregnancy.We identi ed IGFBP1 as the most strongly associated placental transcript.We then measured circulating IGFBP1 protein in plasma samples collected from three different pregnancy cohorts, at multiple timepoints during and after gestation.Using these data, we investigated associations between circulating IGFBP1 levels and insulin sensitivity, other pregnancy-related metabolic traits, birth anthropometric measurements, and risk of GDM.

Results
Participants included in placental genome-wide RNA sequencing analyses We conducted a genome-wide RNA-seq study using placental samples collected from 434 participants in the Genetic of Glucose regulation in Gestation and Growth (Gen3G) prospective pregnancy cohort 15 (Table 1).At study entry (median: 9 weeks' gestation), participants' mean ± SD age was 28.7 ± 4.4 years, and median [IQR] body mass index (BMI) was 23.8 [21.4-27.9]kg/m 2 .We excluded individuals with diabetes present prior to pregnancy.Participants underwent a fasting 75g oral glucose tolerance (75g-OGTT) in the late second trimester (median: 26 weeks' gestation), during which we collected plasma samples and measured glucose and insulin levels at multiple time-points to estimate insulin sensitivity (using the Matsuda index, which has been previously validated against euglycemic clamps in pregnancy 16 ).At delivery (median [IQR] = 39.6 [38.7-40.3]weeks), we collected samples from the maternal-facing side of the placenta using standardized protocols for collection and storage for future analyses by RNA-seq (see Methods).

Differential placenta RNA expression in relation to insulin sensitivity in pregnancy
After processing and quality control of the placental RNA-seq dataset, we investigated differential expression of 15,202 genes in relation to insulin sensitivity (Matsuda index, log2 transformed) in late second trimester.We identi ed 14 genes whose placental RNA expression levels were associated with insulin sensitivity (P-values < 1.0x10 − 3 ; Supp Table 1) after accounting for technical variability (37 Surrogate variables (SV)), precision variables (gestational age at delivery, fetal sex), and potential confounders (gravidity, maternal age, and BMI) using multivariate linear regression models.We observed the strongest association between insulin sensitivity and insulin-like growth factor binding protein 1 gene (IGFBP1; β = 0.43; P = 2.5x10 − 5 ), where higher placental expression levels were associated with greater insulin sensitivity (Fig. 1).IGFBP1 is a binding protein that is primarily produced by the liver outside of pregnancy and is highly expressed by the placenta 17 (https://www.proteinatlas.org/ENSG00000146678-IGFBP1), as replicated in our RNA-seq data.IGFBP1 has been implicated in the modulation of the biological activity of IGF-1 and IGF-2, which are key regulators of growth and metabolism in post-natal and fetal life 18 .We did not observe strong associations between the Matsuda index and other genes in IGF-related pathways, or genes encoding classic pregnancy-speci c placental hormones, or genes encoding in ammatory proteins secreted by the placenta that have been previously associated with insulin sensitivity in pregnancy 12 (Supp.Table 2).In SPRING participants who remained normoglycemic throughout pregnancy (n = 65), we observed that median plasma levels of IGFBP1 rose between the 1st trimester (66,610 pg/mL) and 24 to 32 weeks' gestation (79,379 pg/mL), then declined dramatically post-partum (16,588 pg/mL; paired t-tests P < 0.001 for differences between plasma levels across pregnancy and postpartum; Supp. Figure 1).This pattern, combined with high placental expression levels, suggests a placental origin of high circulating IGFBP1 levels during pregnancy.
In a subset of Gen3G participants (n = 27) in whom we assayed serial IGFBP1 levels during the 75g-OGTT (Supp.Figure 2), we observed that circulating IGFPB1 levels were stable over the rst hour of the OGTT (median levels: fasting = 87,008 pg/mL; 1h-post load = 91,485 pg/mL; paired t-test P-value = 0.13), but declined 2h-post glucose load (median = 60,920 pg/mL; paired t-test P-value = 0.0007 compared to fasting).The change in plasma insulin levels from baseline to 1h (delta insulin 0-60 minutes) appeared to be inversely associated with the IGFBP1 levels at 1h (r= -0.39; P = 0.047) and at 2h (r= -0.31; P = 0.11) during the OGTT.This is consistent with the known negative feedback regulation of IGFBP1 expression by insulin, albeit only shown in hepatocytes 20 .
Circulating IGFBP1 is correlated with insulin sensitivity in three independent pregnancy cohorts Higher plasma IGFBP1 levels were associated with greater insulin sensitivity in all three pregnancy cohorts examined (Table 2).The strong positive correlations (Pearson r = 0.5 to 0.6; P < 0.001) between plasma IGFBP1 levels and insulin sensitivity were consistent across different periods of pregnancy, as well as in the postpartum period (SPRING).Adjusting for maternal age and gestational age at the time of blood sampling did not in uence correlations.The strength of association was modestly attenuated by further adjustment for maternal BMI, but it remained highly statistically signi cant (r = 0.34 to 0.48, P < 0.001; Table 2).
In Gen3G, we further assessed Pearson correlations between plasma IGFBP1 (in the rst and second trimester) and various maternal metabolic traits and neonatal anthropometric measures (Supp.Table 5).
Higher maternal BMI was associated with lower plasma IGFBP1 in the rst trimester (r= -0.27) and in the late second trimester (r= -0.54; both P < 0.001).Plasma IGFBP1 in the late second trimester was negatively correlated with glucose (r= -0.28 to -0.30) and insulin levels (r= -0.40) during the OGTT (all P < 0.001).Lower IGFBP1 levels at both time-points were also associated with higher birth weight z-scores (standardized for gestational age and sex) at delivery (r= -0.15 and r= -0.21 for IGFBP1 at the rst and second trimester visits, respectively; both P < 0.001).Adjusting for maternal BMI or for maternal glucose reduced the strength of associations, but the correlations remained statistically signi cant (e.g.second trimester IGFBP1 partial correlations with birth weight Z-score adjusted for maternal BMI r= -0.12; P < 0.001; or adjusted for maternal glucose (glucose area under the curve (AUC) during the OGTT) r= -0.17; P < 0.001).

Early pregnancy circulating IGFBP1 independently predicts GDM
We tested whether plasma IGFBP1 measured in early pregnancy (median 9 weeks' gestation) predicts GDM (diagnosed with International Association of the Diabetes and Pregnancy Study Groups (IADPSG) criteria applied to a 75g-OGTT at a median of 26 weeks' gestation) in Gen3G participants (n = 837), independent of known clinical risk factors.Overall, 70 participants (8.4%) developed GDM (Supp.Table 3).Early pregnancy IGFBP1 levels alone predicted risk of incident GDM with a modest receiver operating characteristic (ROC) AUC value of 0.64.A model including only clinical variables (maternal age, gravidity, family history of diabetes, maternal BMI, gestational week at blood sampling) without IGFBP1 levels yielded ROC AUC of 0.66 (Fig. 2).A model with the same clinical variables but also incorporating early pregnancy IGFBP1 levels improved predictive ability (Fig. 2, ROC AUC = 0.72 compared to 0.66; P = 0.008).Using a logistic regression model, one SD increase in plasma IGFBP1 levels in early pregnancy was associated with a greater than 50% reduction in the risk for GDM in Gen3G (OR = 0.44; IQR = 0.30-0.64;P < 0.001; adjusted for maternal age, gravidity, gestational age at plasma IGFBP1 measurements, and maternal BMI; see Table 3).
We replicated the predictive association between early pregnancy IGFBP1 levels and GDM incidence in a nested case-control study in the MOMS cohort (n = 55 GDM cases, diagnosed based on Carpenter-Coustan criteria at a median of 29 weeks' gestation; matched 1:2 with non-cases): the OR was 0.40 (95% CI: 0.24-0.67;P < 0.001, adjusted for maternal age and BMI) per SD increase in plasma IGFBP1 (measured at a median of 17 weeks' gestation).In the SPRING cohort, we combined all GDM cases (n = 44, diagnosed in early pregnancy or at 24-32 weeks' gestation based on IADPSG criteria) and observed an OR of 0.75 (95% CI: 0.46-1.25;P = 0.28; adjusted for maternal age, BMI, and gestational age at blood samples) for each SD increase in plasma IGFBP1 measured in the rst trimester (median = 13 weeks' gestation).

Circulating IGFBP1 variations during pregnancy by GDM physiologic subtype
We have previously documented pathophysiologic heterogeneity underlying GDM, demonstrating that GDM cases with a predominant insulin sensitivity defect (insulin-resistant GDM) are more likely to experience GDM-related complications at birth 10 , whereas those with a predominant insulin secretion defect (insulin-de cient GDM) had a similar risk to those with normal glucose tolerance (NGT).Given the strong association between plasma IGFPB1 and insulin sensitivity in pregnancy, we investigated the longitudinal changes in plasma IGFBP1 across pregnancy in different physiologic subtypes of GDM and in participants with NGT in Gen3G (Fig. 3).All GDM subtypes had lower mean plasma IGFBP1 levels in early pregnancy compared to the NGT group.However, the insulin-resistant GDM group had a blunted increase in IGFBP1 levels between the rst and second trimester; in contrast, in those with insulinde cient GDM, IGFBP1 levels reached similar levels to those in the NGT group during the second trimester (Fig. 3).The group who had GDM with both insulin resistance and insulin de ciency (mixed defect GDM) showed an IGFBP1 trajectory that was intermediate between the other GDM subtypes.
In Gen3G, we also found that low IGFBP1 levels in rst trimester were associated with subsequent diagnosis of both insulin-resistant GDM and insulin-de cient GDM with ORs ~ 0.4 (in fully adjusted models, including maternal BMI) similar to prediction models where the outcome was all GDM (see Model 3, Table 3).However, IGFBP1 levels in the second trimester were only associated with insulin-resistant GDM (OR = 0.28 [0.16-0.47]per SD increase in IGFBP1 levels; P < 0.001); there was no signi cant association between second trimester IGFBP1 plasma levels and insulin-de cient GDM (Table 3).

Discussion
In this study, using genome-wide RNA-seq of placental tissue, we identify IGFBP1 as a key placental transcript associated with insulin sensitivity in human pregnancy.Our ndings implicate IGFBP1 de ciency in GDM pathophysiology.We show that circulating IGFBP1 levels rise during pregnancy and are much higher in pregnancy than in the non-pregnant state, supporting the contribution of placental IGFBP1 to elevated circulating IGFBP1 in pregnancy.In three independent pregnancy cohorts, we demonstrate a strong and consistent correlation between higher circulating IGFBP1 and greater insulin sensitivity (lesser insulin resistance), uncovering a potential compensatory mechanism in euglycemic pregnancy.Moreover, we show that low plasma IGFBP1 levels in the rst trimester of pregnancy predict the later diagnosis of GDM, independent of maternal clinical risk factors (including BMI).Finally, we note that the normal pregnancy rise in IGFBP1 levels is attenuated in insulin-resistant GDM, suggesting that a defect in placental IGFBP1 release may contribute speci cally to this GDM physiologic subtype.
In placental tissues, IGFPB1 expression has previously been detected in decidual cells and in fetal placental macrophages or Hofbauer cells 21 , however, there is limited knowledge of IGFBP1 regulation and actions in pregnancy.In in-vitro experiments in decidualized human endometrial stromal cells, IGFBP1 was regulated by cAMP, progesterone and relaxin 22,23 -the latter two being critical hormones for establishment and maintenance of a pregnancy 24 .Outside of pregnancy, IGFBP1 is almost exclusively expressed by the liver 17 and its production is regulated by insulin which inhibits its gene transcription in hepatocytes 25 .Our observations that plasma IGFBP1 levels decline after a plasma insulin rise in response to an oral glucose load introduce the possibility that insulin may downregulate the production and/or release of IGFBP1 from the placenta, similarly to the downregulation observed in hepatocytes 25 .It is also possible that other insulin sensitivity endocrine factor such as adiponectin regulates IGFBP1 expression in placental cells 26 .
Functional studies suggest that IGFBP1 binds IGF-1 and IGF-2 with equal a nity and can either inhibit or enhance IGF actions, depending on the context 20 .In post-natal life, IGF-1 is the main active growth factor and is essential for normal growth during childhood and adolescence; while during fetal development, both IGF-1 and IGF-2 are key regulators of fetal growth 20 .Outside of pregnancy, IGF-1 enhances insulin sensitivity by suppressing hepatic glucose production 27,28 and promoting glucose uptake in peripheral tissues 29,30 .IGF2 is a highly expressed imprinted gene which is a key regulator of fetal growth in mammals 31 .In a recent study, pregnant mice with an IGF2 deletion speci c to placental endocrine cells did not develop the normal insulin resistance of pregnancy and gave birth to fetuses that were growth restricted and hypoglycemic 32 .In general, IGFs have higher a nity for IGFBPs than for cellular IGFreceptors, and thus, IGFBPs often act as inhibitors of biological activity 22 .However, they may also function as a circulating pool of IGFs by prolonging their half-lives and creating IGF reservoirs 18,20 .In addition, IGFBP1 has putative IGF-independent effects, and may activate PI3K/AKT signaling pathways involved in post-receptor insulin signaling directly 33 .In line with this, in vivo injection of an active IGFBP1 peptide improved insulin sensitivity in a diet-induced obesity mouse model 34 .These multiple mechanisms of action might explain some of the inconsistencies from previous animal studies attempting to establish the effects of IGFBP1 on glucose regulation [35][36][37] .
None of these prior studies provide insights on the speci c role that IGFBP1 may have in the context of pregnancy, when there are high circulating levels of IGFs, which are suspected to in uence glucose metabolism 20,32 .We speculate that placental release of IGFBP1 may regulate insulin sensitivity in pregnancy -via direct and/or indirect effects -physiologically contributing to homeostatic mechanisms to balance maternal and fetal nutrient needs.An alternative explanation is that low levels of IGFBP1 in GDM are a consequence of hyperinsulinemia with another upstream cause, but this would not be in line with the rise of circulating IGFBP1 throughout pregnancy (which is characterized by hyperinsulinemia that increases as well).In the context of GDM pathophysiology, based on our ndings in individuals with insulin-resistant GDM, we speculate that the placenta may be unable to produce increasing amounts of IGFBP1 as pregnancy progresses; this de ciency in circulating IGFBP1 could contribute to excessive insulin resistance, and thus to maternal hyperglycemia detected in the late second trimester in this GDM subtype.In individuals with insulin-de cient GDM, IGFBP1 levels were low in the rst trimester but were at similar concentration to levels in those without GDM at the end of the second trimester, suggesting that other pathophysiologic factors contribute to hyperglycemia in this GDM subtype.Given the differences in IGFBP1 in different GDM subtypes and increasing recognition in the eld that GDM is a heterogeneous condition 38 , our nding of persistently lower IGFBP1 levels in the second trimester of pregnancies affected by insulin-resistant GDM may have implications for GDM precision medicine 39,40 .Our ndings suggest that in insulin-resistant GDM, the placenta does not increase IGFBP1 production su ciently; if this association is demonstrated to be causal, this opens the door to a novel therapeutic target for this GDM subtype.Beyond GDM, the association between lower circulating IGFBP1 levels and higher birth weight is in line with similar observations in prior report 41 and suggests a potential explanation for the greater risk of large-for-gestational-age birthweight that we previously observed in insulin-resistant GDM 10 .
Accurately predicting GDM incidence based on early pregnancy markers could allow development and implementation of interventions to prevent GDM and its complications.However, most predictive models that rely on established clinical risk factors perform poorly 42,43 , and thus, there has been a search for reliable and replicable biomarkers.We found that low levels of circulating IGFBP1 in early pregnancy predict later diagnosis of GDM in a large population-based cohort (Gen3G), with external replication and consistent effect sizes in a separate cohort (MOMS).The effect size was more modest and not statistically signi cant in a cohort study of participants who all had GDM risk factors (SPRING); these inclusion criteria may have diminished the predictive ability of circulating IGFBP1 in this population.
Previous studies have been inconsistent with regard to circulating IGFBP1 as a predictive biomarker for GDM, with only one study reporting on IGFPB1 levels measured before 20 weeks of gestation 44 .Our ROC analyses showed that circulating IGFBP1 levels in early pregnancy have a predictive ability beyond that of established GDM risk factors (including maternal BMI and family history of diabetes), however, the moderate ROC AUC value in a model that included IGFBP1 levels along with these clinical factors suggests that additional biomarkers would be necessary for clinical utility.

Strengths and limitations
Our investigation has several strengths.We included a large number of placentas in our expression pro ling, used transcriptome-wide RNA-seq, and leveraged an agnostic approach to implicate genes and their products in insulin sensitivity during pregnancy.Furthermore, we examined not only placental expression of IGFBP1, but also circulating IGFBP1 levels in three pregnancy cohorts.Our analyses included measurement of circulating IGFBP1 levels over a longitudinal timeframe that spanned both pregnancy and postpartum.In addition, we used an OGTT-based measure of insulin sensitivity that has been validated against euglycemic clamps in pregnancy.Our study also had some limitations.Although we had a large overall sample size, the number of GDM cases was somewhat modest, and the sample size for each GDM physiologic subtype was even more limited.Although we were able to tie placental expression and circulating IGFBP1 levels to detailed physiologic phenotyping, our study was observational and thus cannot establish mechanisms or causality for the associations we observed.

Conclusions
In this study which utilized placental gene expression pro ling, we implicated IGFBP1 in insulin sensitivity during pregnancy.IGFBP1 is highly expressed by the placenta and maternal IGFBP1 levels are markedly elevated during gestation, increasing across pregnancy and dropping substantially postpartum.Both At V2, we performed similar anthropometric measurements and questionnaires as at V1. V2 occurred at the time of the fasting 75g-OGTT, which was standard clinical practice for screening and diagnosis of GDM at the CHUS.We collected additional blood samples at the fasting, 1h, and 2h time-points of the 75g-OGTT to measure insulin at each time-point, in addition to glucose.We measured glucose levels via the hexokinase method (Roche Diagnostics; CHUS biochemistry laboratory) as soon as samples were collected.We measured insulin levels via multiplexed particle-based ow cytometric assays (Human Milliplex MAP kits; EMD Millipore) from the previously frozen plasma samples (stored at -80℃ until measurement).We estimated insulin sensitivity using the Matsuda Index 45 (using glucose and insulin values during the OGTT), as previously validated against euglycemic clamps performed in pregnancy 16 .
At delivery, we collected newborn age and sex at birth using medical records, in addition to details from the end of pregnancy and delivery complications.Trained study staff collected placentas within 30minutes of delivery using a standardized protocol.In brief, 1-cm 3 placental tissue sample was collected from the maternal facing side, including decidual tissue (within a 5cm radius of the corresponding location of cord insertion on the other side).Each collected sample was immediately put in RNA-Later for at least 24 hours at 4°C before storage at -80°C until RNA extraction.
RNA Extraction, Sequencing and Quality Control extracted RNA (average = 19.7 ± 7.1 µg) and checked the quality of each sample using an Aligent Bioanalyzer to determine the RNA Integrity Number (RIN; average RIN = 6.7 ± 0.8).We shipped samples (3 µg) with an RIN value ≥ 5 to the Broad Institute (Cambridge, MA, USA) for sequencing.In a second round of sample quality control at the Broad Institute (Caliper Life Sciences LabChip GX system), the RNA Quality Score (RQS) for each sample ranged from 3.3 to 7.8 (average RQS = 5.9).We submitted all samples with an RQS value of 4 or higher for RNA sequencing (N = 466).We completed library preparation with 250ng of each sample, using an automated variant of the Illumina TruSeq™ Stranded mRNA Sample Preparation Kit (Illumina, cat #RS-122-2103).We performed Flowcell cluster ampli cation and sequencing according to the manufacturer's protocols using the Illumina HiSeq 4000, to generate 101-bp paired end reads, average of 113M total reads (range 33M to 378M) per sample.
Following quanti cation, we applied additional quality control (QC) steps.Of the 466 samples sequenced, we excluded those with > 1% of outlier genes (> 3 times the inter quartile range (IQR) above Q3 or > 3 IQR below Q1), leaving 459 samples for our nal analytical data set.Among these, we had complete data on the phenotype of interest (Matsuda index) and covariates for 434 samples.Prior to differential gene expression analysis, we removed genes with low abundance, keeping only those genes with at least a sensitivity 54 (https://www.rdm.ox.ac.uk/about/our-clinical-facilities-and-mrcunits/DTU/software/homa/download).
Bioassays for circulating IGFBP1 (all three cohorts) We measured circulating IGFBP1 in plasma samples from all 3 cohorts in the same laboratory using a commercially available ELISA that measures free IGFBP1 (Catalog # DGB100, R&D systems, MN).The precision for the assays were: intra-assay CVs 5.6% and inter-assay CVs of 9.5%.We measured IGFBP1 levels in a blinded fashion, and we followed protocol for measurement per manufacturer's instructions.

Statistical analyses
For characteristics participants in all three cohorts, we reported normally distributed continuous variables as mean ± SD, non-normally distributed continuous variables as median and interquartile range (IQR), and categorical variables as percentages.We used a log2-transformation for Matsuda index (to approach a normal distribution) in the differential placental RNA expression analyses.

Placental differential expression analyses (using RNA-Seq data
We adjusted models for maternal age, gravidity, maternal BMI at the rst trimester visit, sex of offspring, and gestational age at delivery, in addition to computed surrogate variables (SVs) to account for unmeasured sources of variability, including batch effects and cell types.We used the EstDimRMTfunction from the R package isva 55 to estimate the number of SVs to include given the residuals from the regression of Matsuda and biological covariates from the normalized counts, which resulted in 37 SVs computed by the R package SmartSVA 56 recommended for our processed RNAseq dataset.We used Limma 57 to identify differentially expressed genes with log 2 Matsuda as a continuous independent variable.We reported genes that had differential expression in relation to Matsuda with Pvalues < 1.0 x 10 − 3 .

Circulating IGFBP1 correlation analyses
We used a box-cox transformation for plasma IGFBP1 levels in Gen3G (from MASS package 58 in R) since it was the best way to approximate a normal distribution.We conducted analyses in SPRING and MOMS cohorts using plasma IGFBP1 levels without transformation, given distributions that were relatively normal.We used Pearson correlations between circulating IGFBP1 levels and Mastuda index (logtransformed) in all three cohorts; we used partial correlations to assess the associations while taking into account maternal age, gestational age at blood draw, and maternal BMI.In Gen3G, we also used Pearson correlations to assess associations between plasma IGFBP1 (box-cox transformed) and maternal metabolic markers, as well as newborn anthropometry (transformed if needed).

Circulating IGFBP1 and risk of GDM analyses
We conducted logistic regression analyses with the levels of circulating IGFBP1 as the independent variable and GDM as the dependent variable in Gen3G and SPRING; in MOMS, due to the matched casecontrol design of GDM cases to controls, we used conditional logistic regression.In Gen3G and SPRING, we used international criteria (IADPSG) 59 to ascertain GDM, while in MOMS, we used the Carpenter-Coustan criteria 60 .In Gen3G, additionally sub-classi ed GDM by the insulin physiology defect driving hyperglycemia (insulin-resistant GDM, insulin-de cient GDM or mixed defect GDM, as previously described 10 ).We rst built unadjusted logistic regression analyses (Model 1).We adjusted for maternal characteristics (maternal age, gravidity, gestational age at plasma samples) in Model 2, and additionally adjusted for maternal BMI in Model 3. We calculated pro led log-likelihood con dence intervals along with likelihood ratio test p-values (using MASS 58 and glmglrt (https://CRAN.Rproject.org/package=glmglrt)packages 47 in R).In SPRING and MOMS cohorts, we employed similar modeling strategies using maximum likelihood dichotomous logistic models.
We conducted GDM predictive analyses using Receiver Operating Characteristic (ROC) curves in Gen3G to compare the predictive ability of rst trimester (V1) plasma IGFBP1 levels in addition to commonly measured GDM clinical risk factors (maternal age, gravidity, family history of diabetes, gestational age at V1, and maternal BMI at V1).We compared the ROC AUC values using all the clinical factors with and without rst trimester (V1) plasma IGFBP1 levels (after Box-Cox transformation).We compared the ROC AUC values from nested models using DeLong's test using the "roc.test"function from the pROC package in R 61 .We considered differences between AUC values to be statistically signi cant if P < 0.05.
In Gen3G, we performed analyses using R version 4.

Supplementary Files
This is a list of supplementary les associated with this preprint.Click to download.HivertSupplementaryMaterialIGFBP110.17

Table 1
Characteristics of Gen3G participants included in the placental RNA sequencing analyses (N = 434) BMI = body mass index; Matsuda calculated at the 2nd trimester visit; *by International Association of the Diabetes in Pregnancy Study Groups (IADPSG) criteria

Table 2
Footnote: plasma IGFBP1 levels transformed using box-cox for optimal normal distribution, then translated in z-score for the logistic regression analyses.OR and [95% CI] are per 1 SD increase of plasma IGFBP1 Footnote: plasma IGFBP1 levels transformed using box-cox for optimal normal distribution, then translated in z-score for the logistic regression analyses.OR and [95% CI] are per 1 SD increase of plasma IGFBP1 Model 2: adjusted for maternal age, gravidity, gestational age at plasma samples (V1 and V2 respectively) Model 3: Model 2 covariates + maternal BMI measured at V1 GDM diagnosis was made at a median of 26 weeks' gestation