Changes in the composition and function of the gut microbiome accompany type 1 diabetes in pregnancy


 Background The gut microbiome changes in response to a range of external influences, life events and disease. Pregnancy is a natural life event involving major physiological adaptation but studies of the microbiome across pregnancy are scarce and findings inconsistent. Pregnancy in which the mother has type 1 diabetes (T1D) with disturbed glucose-insulin homeostasis increases maternal and fetal risks and knowledge of the gut microbiome in this context is lacking. To understand how T1D impacts the gut microbiome in pregnancy we used a combination of whole metagenomic sequencing (WMS) and 16S rRNA sequencing to define the composition and function of the gut bacterial microbiome across 145 pregnancies, 94 in women with T1D. Women were participants in the Environmental Determinants of Islet Autoimmunity (ENDIA) study, in which the child has a first-degree relative with T1D.Results Pregnant women with and without T1D exhibited taxonomic and functional changes in gut microbial composition across pregnancy. Profiles in women with T1D were distinct, with a decrease and increase, respectively, of bacteria with anti- and pro-inflammatory properties, and a decrease in bacteria that synthesize essential vitamins and metabolites. These changes were accentuated in trimester 3, often in opposite directions to women without T1D.Conclusions The gut microbiome changes in composition and function across pregnancy, but distinctly in women with T1D. The decrease in bacteria that produce anti-inflammatory mediators and essential B-group vitamins suggests that the gut microbiome may contribute to maternal and fetal complications in the T1D pregnancy, potentially opening new avenues for therapeutic intervention.


Background
The gut microbiome provides essential metabolites, vitamins, co-factors and hormones, protects against pathogenic microorganisms and has a key role in the development of the immune and other systems [1,2]. Adaptations or shifts in the composition of the gut microbiome occur throughout ageing and in association with environmental conditions, life events and disease states [2][3][4]. In pregnancy, women undergo signi cant physiological changes, but only recently has the gut microbiome been studied in this context [5,6]. Koren et al [5] sampled the gut microbiome in the rst and third trimesters and found that the taxonomic composition in the rst trimester was similar to that of non-pregnant women, but in the third trimester observed an increase in Actinobacteria and Proteobacteria phyla along with an overall decrease in bacterial richness (alpha diversity). Germ-free mice inoculated with third trimester feces had greater weight gain, insulin resistance and gut in ammation than mice inoculated with rst trimester feces, which they interpreted as an adaptive proin ammatory response to defend the fetus from pathogens and provide it with nutrients [5,7]. On the other hand, DiGiulio et al [6] collected weekly fecal samples across pregnancy but found no signi cant temporal differences in diversity or composition of the gut microbiome. These contrary ndings along with a dearth of studies warrant further investigation of the gut microbiome across pregnancy. Type 1 diabetes (T1D) is an autoimmune disease in which insulin-producing β cells in the islets of the pancreas are destroyed by T lymphocytes leading to insulin de ciency [8]. T1D increases pregnancy risks for mother and fetus [9] but the role of the gut microbiome in this context is unclear. Alterations in the bacterial gut microbiome have been reported in T1D, mainly in children at high risk and at diagnosis (reviewed in [10], [11][12][13][14][15][16]). They encompass a decrease in the Firmicutes/Bacteroidetes phylum ratio [11,12], a decrease in richness (alpha diversity) [11][12][13] an increase in in ammation-associated bacteria [12,13] and a decrease in bacteria with anti-in ammatory properties, in particular lactate-and butyrateproducing and mucin-degrading species [12][13][14][15][16]. Functionally, this dysbiosis is re ected by a decrease in the abundance of genes that encode related metabolic pathways and enzymes, e.g. butyryl-coenzyme A (CoA)-CoA transferase [14] and butyryl-CoA dehydrogenase [15]. While these changes are not necessarily speci c for T1D they nevertheless may have clinical consequences. Gut butyrate is a key determinant of gut health and regulator of gene expression and homeostatic immunity [17][18][19]. It is the major energy source for the colonic mucosa, induces synthesis of mucin and promotes gut epithelial integrity, preventing 'gut leakiness' and translocation of toxins and dietary antigens into the blood. Such translocation may lead to systemic in ammation and predispose to immune diseases such as T1D. Indeed, dietary butyrate induced regulatory T cell production and decreased the incidence of spontaneous diabetes in the non-obese diabetic (NOD) mouse model of T1D [19]. Increased gut permeability has been described in established T1D [20] and recently by ourselves in association with gut microbiome dysbiosis in children with islet autoimmunity who progressed to T1D [21]. In the T1D pregnancy, gut microbiome dysbiosis could have functional consequences for the mother and the fetus. Consensus about the gut microbiome in pregnancy, even in the absence of T1D, is lacking. To address this knowledge gap, we used a combination of whole metagenomic sequencing (WMS) and 16S rRNA sequencing to analyse the gut microbiome across the three trimesters of pregnancy of women with and without T1D who are participating in the Australia-wide Environmental Determinants of Islet Autoimmunity study (ENDIA).

Study population
Fecal samples were collected from women enrolled in the ENDIA study across ve Australian States between February 2013 and October 2017 [22]. ENDIA is a prospective, pregnancy-birth cohort study that follows 1500 Australian children who have a rst-degree relative with T1D. Fifteen women (16 pregnancies) without T1D (non-T1D) and 16 with T1D had each provided three fecal samples across pregnancy (total of 96 samples) ( Table 1). These were analysed by shotgun WMS ( Figure S1A). A second sample set comprising a total of 354 fecal samples collected longitudinally across 145 pregnancies from 51 non-T1D women and 94 T1D women, including the 96 samples analysed by WMS, was analysed by 16S rRNA gene sequencing ( Figure S1B). Table 1 summarizes and compares characteristics of the non-T1D and T1D pregnancies.
Whole metagenomic sequencing and 16S sequencing For the WMS dataset, 43,809,936 ± 9,156,972 (mean ± SD) paired end reads per sample were obtained using an Illumina NovaSeq 6000. Raw reads (SRA accession: PRJNA604850) were pre-processed using KneadData bioBakery tool [23] to eliminate human DNA sequences and lter sequences with poor quality which on average removed 6% of the reads. Reads of 41,013,121 ± 8,739,281 (mean ± SD) were obtained after quality control and read lter steps (Excel le E0). For the 16S rRNA dataset, 42,104 ± 17,296 (mean ± SD) paired end reads per sample were obtained with an Illumina MiSeq. QIIME2 was used to demultiplex and quality-lter the raw sequences (SRA accession: PRJNA604850), generating 24,777 ± 10,749 (mean ± SD) reads per sample (Excel le E0).
Taxonomic diversity and composition of the gut microbiome in non-T1D and T1D women during pregnancy Metagenomic sequences of the women who provided three fecal samples across trimesters were analysed using MetaPhLan2 implemented within the HUMAnN2 pipeline. Overall, 298 bacterial species were identi ed with an average of 97 ± 11 (mean ± SD) species per sample.
The top 25 most abundant species accounted for more than 50% of the gut microbiome composition of each subject in any given trimester ( Figure 1). The Firmicutes/Bacteroidetes phylum ratio across trimesters was unchanged in non-T1D women but decreased in women with T1D ( Figure S2).
Alpha diversity (observed richness or number of species) per sample was calculated (Excel le E1) and generalized estimating equations (GEE) were applied to test for differences between women without and with T1D, between trimesters and to determine if there was an interaction between T1D status and trimester. No signi cant interactions or differences in richness were found ( Figure S3, Excel le E1).
For beta diversity analysis, Bray-Curtis coe cients were calculated between sample pairs, ordinated and plotted by principal coordinate analysis (PCoA; Figure 2). To test for differences in beta diversity, a repeated-measure aware permutational analysis of variance (RMA-PERMANOVA) of the Bray-Curtis coe cients was performed on normalized data. This revealed a signi cant interaction between T1D status and time at all taxonomic levels. Therefore, differences between T1D and non-T1D women were assessed within trimester. No signi cant differences were detected in trimester 1. However, differences were signi cant at the strain level (P=0.02) in trimester 2 and at the strain (P=0.002), species (P=0.001), genus (P=0.015), family (P=0.029) order (P=0.034) and phylum (P=0.032) taxonomic levels in trimester 3 (Excel le E2).
Differences in beta diversity re ect differences in taxonomic composition. To identify differences in taxa between non-T1D and T1D women in pregnancy, differential abundance was analysed in limma; the Benjamini-Hochberg method was used to control the false discovery rate. Taxa were considered differentially abundant if the adjusted P-value expressed as the False Discovery Rate (FDR) was < 0.05; however, borderline signi cant differences, i.e. FDR from 0.05 to 0.1, were also reported. Note that only taxa for which the prevalence (i.e., proportion of samples with those taxa) was above 50% in at least one group and with a log2 fold-change (logFC) greater than 0.5 or less than -0.5 were considered. T1D status and trimester were combined into a single factor with six levels and comparisons of interests were de ned as contrasts. Across all trimesters, the species Bacteroides eggerthii (FDR=0.033), Bacteroides clarus (FDR=0.005), Alistipes sp. AP11 (FDR=0.009), Escherichia coli (FDR=0.002), an unclassi ed Escherichia bacterium (FDR=0.042), Roseburia hominis (FDR=0.08) and Bacteroides uniformis (FDR=0.088), as well as the family Enterobacteriaceae (FDR=0.09) and the order Enterobacteriales (FDR=0.03), were increased in women with T1D ( Figure 3; Excel le E3). On the other hand, the species Bacteroides massiliensis (FDR=0.043) and the family Prevotellaceae (FDR=0.05) were decreased in T1D ( Figure 3; Excel le E3).
Most differences between non-T1D and T1D women were in trimester 3, in which species In order to expand on these observations, we sequenced DNA of the V4 region of the 16S rRNA marker gene in a larger dataset comprising 354 samples from 51 non-T1D and 94 T1D women, including the original 96 samples. Sequences were processed with QIIME2 to obtain a feature per sample table and further analysis was performed in R (see Methods). Features were agglomerated based on a phylogenetic tree into an overall total of 349 operational taxonomic units (OTUs) with a mean ± SD of 82 ± 24 (mean ± SD) OTUs per sample (Excel le E0). Fourteen of the 25 most abundant species detected by WMS were also within the 25 most abundant OTUs in the larger 16S rRNA dataset. Moreover, one OTU had 100% sequence identity with both Bacteroides vulgatus and B. dorei and therefore these two species could not be distinguished by 16S rRNA gene sequencing.
Consistent with the metagenomic analysis, alpha diversity did not differ by T1D status in the larger 16S rRNA dataset in any trimester (Excel le E1). Also, in accord with the WMS analysis, an interaction between T1D status and time was signi cant for beta diversity at the genus and family taxonomic levels and borderline signi cant at the OTU taxonomic level (Excel le E2). Therefore, differences in beta diversity were also assessed within trimester. The gut microbiome composition differed signi cantly between non-T1D and T1D women within trimester 2 and 3 at the OTU, genus and family taxonomic levels, but not within trimester 1 (Excel le E2). At the order and phylum taxonomic levels, T1D status and the interaction between T1D status and time were not signi cant.
Only one OTU classi ed to the species Bacteroides caccae was differentially abundant in trimester 3 (FDR=0.02) ( Figure 4A; Excel le E3). Generally, the prevalence of taxa was lower with 16S rRNA sequencing than WMS. To con rm the differentially abundant taxa detected between non-T1D and T1D women by WMS, relative abundances by 16S rRNA sequencing were agglomerated at family, order and phylum taxonomic levels. Most of the taxa identi ed in the 16S rRNA dataset exhibited the same trend in abundance between non-T1D and T1D as observed with WMS ( Figures 4B-F).
Effect of time and other factors on the gut microbiome during pregnancy WMS revealed no signi cant differences in alpha diversity in non-T1D or T1D women due either to time analysed as days into gestation or by trimester, i.e., as continuous or categorical variables, respectively (Excel le E1). Due to a signi cant interaction between T1D status and time, differences in beta diversity between trimesters were assessed in non-T1D and T1D women separately (Excel Table E2). No signi cant differences in beta diversity due to trimester were observed at any taxonomic level in non-T1D or T1D women (Excel Table E2). However, signi cant differences were present by time analysed as days into gestation at the strain and genus levels (and borderline differences at the species and family levels) in T1D samples and at the strain, species, genus and family levels in non-T1D samples (Excel le E2).
No signi cant associations were found between beta diversity and other factors included in the models, viz. age at conception, body mass index (BMI), parity and human leukocyte antigen (HLA) class II genotype, for either metagenomic or 16S rRNA data with the exception of parity which was signi cantly different only in trimester 1 at the order taxonomic level, for the 16S rRNA data (Excel le E2). As expected, non-T1D and T1D women differed in serum 1,5-AG, a marker of short-term glycemic control [24] ( Table 1), but in T1D women serum 1,5-anhydroglucitrol (1,5-AG) was not related to beta diversity (Excel le E2). The frequency of pre-eclampsia was higher in the larger group of 145 women (Table 1) but was not related to beta diversity (Excel le E2). The frequency of elective Caesarean section was higher in T1D women ( Table 1) but was not related to beta diversity (Excel le E2). Non-T1D and T1D women also differed in total carbohydrate but not ber intake (Table 1). In the larger group of 145 pregnancies, total carbohydrate intake was lower in those with T1D (P=0.016), most likely re ecting dietary advice. Based on 16S data, interactions were found between carbohydrate and ber intake and time; therefore, beta diversity was assessed per trimester. Carbohydrate intake was related (borderline P-values greater than 0.05 but less than 0.1) to beta diversity at family, order and phylum levels only in trimester 2. Fiber intake was signi cantly related to beta diversity at genus and family levels only (borderline signi cance for order and phylum taxonomic levels) in trimester 3 (Excel le E2). No relationships between either carbohydrate or ber intake and beta diversity were found with WMS data.

Functional annotation of gut microbiome taxa
For functional pro ling, metagenomic sequences were processed with HUMAnN2. Sequences were annotated, gene abundances calculated and regrouped into KO (Kegg Orthology) and MetaCyc reaction functional categories, and complete metabolic pathways were quanti ed obtaining a total of three functional pro les. A total of 5,480 KO, 3,014 metaCyc reactions and 420 complete pathways, were obtained. No signi cant differences in richness were detected between T1D and non-T1D women or across pregnancy in any of the three functional pro les (Excel le E1). On the other hand, in the analysis of beta diversity, the interaction between T1D status and time was signi cant. Therefore, differences between non-T1D and T1D women were assessed in each trimester and were only signi cant in trimester 3 for pathways, metaCyc reactions and KO functions (Figures 6; Excel le E2). Differential abundance analysis revealed signi cant differences between non-T1D and T1D in speci c pathways and enzymes only within trimesters 1 and 3 ( Figure 7; Excel les E5-7). A comprehensive list of differentially abundant pathways, KOs and MetaCyc reactions is presented in Supplementary Excel les E5-10). Of interest, a pathway (CMP-3-deoxy-D-manno-octulosonate biosynthesis I) and two enzymes (K02852 and K01791) involved in the synthesis of bacterial lipopolysaccharides (LPSs) were enriched in T1D women (Figure 7; Excel les E5 and E7), with contributions from Alistipes shahii, B. thetaiotaomicron, B. ovatus, B. vulgatus, B. cellulosilyticus, Parabacteroides distasonis, Bacteroides uniformis among others ( Figure S4). In addition, a KO and a MetaCyc reaction that were increased in T1D women in trimester 3 were involved in bio lm formation (K04334) and the synthesis of the antibiotic mannopeptimycin (K00815), respectively, both being contributed solely by E. coli (Figures 7 and S5; Excel le E5). Table 2 summarizes these and other key pathways and enzymes differentially abundant in T1D women.
Functional differences were also present between trimesters (Excel les E8-E10). Most involved a decrease in abundance from trimester 1 to 3. Thus, in T1D women only the abundance of beta-Nacetylhexosaminidase (K01207) involved in the degradation of mucin decreased signi cantly from trimester 1 to 3, associated with a decrease in A. muciniphila, R. hominis, B. longum and B. adolescentis (Figures 7 and S12; Excel le E8). Enzymes involved in the synthesis of acetate (K01512), butyrate (K00248 and K00209), acetyl-CoA (K13788; a precursor of acetate and butyrate) and vitamins B1 (K00878) and B5 (K00997) also decreased from trimester 1 to 3 in T1D women only, while pyruvate ferredoxin oxidoreductase (K00171), involved in the conversion of pyruvate into acetyl-CoA, increased from trimester 1 to 3 in non-T1D women only (Figures 7 and S12-S14; Excel le E8).

Discussion
The gut microbiome in pregnancy in healthy women was previously analysed in two studies that reported con icting ndings [5,6], but has not been studied in pregnancy in the context of T1D. We characterized the taxonomic composition and annotated functions of the gut microbiome across pregnancy in women without and with T1D. Although we found no signi cant differences in alpha or beta diversity between trimesters, or in alpha diversity between non-T1D and T1D women, we observed changes in the relative abundance of speci c taxa across pregnancy that were often opposite in directions in non-T1D and T1D women, particularly in trimester 3. Generally, pregnant women with T1D exhibited a more pro-in ammatory and catabolic gut microbiome pro le. The taxonomic differences between T1D and non-T1D women were reinforced by functional annotation of WMS data, revealing differential abundance in enzymes and pathways as pregnancy progressed. These differences could not be attributed to demographic or other factors including diet. It is important to emphasize that our ndings are based on DNA analysis and might not necessarily re ect changes at the RNA or protein level. Nevertheless, the functional consequences of the dysbiosis we observed in women with T1D could contribute to the known increase in maternal and fetal complications in T1D. This has implications for improving the management of women with T1D in pregnancy.
As a discovery dataset, we rst employed WMS on 96 fecal samples collected from 15 non-T1D and 16 T1D women in each trimester of pregnancy. To substantiate our observations, we then performed nextgeneration amplicon sequencing of the 16S rRNA gene V4 region on a larger sample set from 51 non-T1D and 94 T1D women. Koren et al [5] and DiGiulio et al [6] analysed fecal samples from healthy pregnancies by pyrosequencing the 16S rRNA gene V1-V2 and V3-V5 regions, respectively, but arrived at different conclusions. Koren et al [5] compared single samples from trimesters 1 and 3 from 91 pregnancies and reported a decrease in alpha diversity and 'remodelling of the gut microbiome' by the third trimester, speci cally a decrease in the abundance of taxa in the genus Faecalibacterium that generate anti-in ammatory butyrate [17] and an increase in taxa in the phylum Proteobacteria recognised to be pro-in ammatory [25]. DiGiulio et al [6], on the other hand, by weekly sampling from 49 women found no signi cant changes in diversity or composition across pregnancy. Similar to DiGiulio et al [6] we observed no differences in alpha or beta diversity, but detected changes in the relative abundance of speci c taxa across pregnancy with progression to a more pro-in ammatory microbiome particularly in women with T1D. In non-T1D women, similar to Koren et al [5], we observed an increase in the phylum Actinobacteria, in particular the order Bi dobacteriales, from trimester 1 to 3. However, in contrast, we observed a decrease in the abundance of taxa in the Proteobacteria.
The ratio of Firmicutes/Bacteroidetes phyla, regarded as an index of gut health in some publications [12], declined between the rst and third trimesters in T1D but not non-T1D women. In examining differentially abundant taxa we observed three main patterns: 1) taxa that were differentially abundant between non-T1D and T1D women across all trimesters, 2) taxa that decreased in T1D women only as pregnancy progressed, and 3) taxa that were similar in abundance in non-T1D and T1D women in trimester 1 but decreased or increased to be differentially abundant in trimester 3. In the rst category, in which taxa were differentially abundant between non-T1D and T1D women across all trimesters, the order Enterobacteriales, the family Enterobacteriaceae and six species including Escherichia coli were increased in T1D compared to non-T1D women across all trimesters. Enrichment in Enterobacteriaceae, in particular E. coli, is associated with intestinal in ammation [26,27]. Moreover, an increase in these facultative anaerobic bacteria may displace obligate anaerobic bacteria that produce SCFAs, further accentuating in ammation. In keeping with this, T1D women had a decrease in the abundance of the family Prevotellaceae, which produces succinate and the SCFAs propionate and acetate known to be associated with improved glucose metabolism [18, 28, 29]. In T1D women, E. coli contributed to an increased abundance of the major curlin subunit (K04334) involved in bio lm formation and of a tyrosine aminotransferase (EC. 2.6.1.107) involved in the biosynthesis of mannopeptimycin, a peptide antibiotic active on gram-positive bacteria. Bacterial bio lms confer increased tolerance to antibiotics and host immune responses [30] and may provide E. coli with a protective advantage over other more sensitive bacteria that compete for the same resources in the gut. Moreover, mannopeptimycin may provide E. coli with a competitive advantage to displace gram-positive butyrate-producing bacteria [17] such as B. adolescentis and R. bromii, which we observed to be decreased in T1D.
In the second category, taxa decreased in T1D women only or in contrast to non-T1D women demonstrated no increase as pregnancy progressed. We observed a decrease in Roseburia intestinalis (and the order Clostridiales) and Akkermansia muciniphila (and the order Verrucomicrobiales) from trimester 1 to trimester 3. De ciency of these taxa is associated with gut in ammation and impaired gut barrier function. R. intestinalis produces butyrate [31,32]. The abundance of A. muciniphila correlates with richness of the gut microbiome and protection from type 1 diabetes in the NOD mouse model [33] and inversely with markers of dysmetabolism [34]. Early life treatment with vancomycin propagated A. muciniphila and reduced diabetes incidence in the NOD mouse [33]. The bene ts of A. muciniphila relate to its ability to promote host production of anti-in ammatory lipids and mucus, and glucagon-like peptides which regulate glucose homeostasis [35]. Furthermore, by also degrading mucins, A. muciniphila produces oligosaccharides, and acetate and propionate, which together then stimulate mucus production and enhance epithelial integrity [36]. Beta-N-acetylhexosaminidase (K01207), which degrades mucin [37], was associated mostly with A. muciniphila, and decreased in women with T1D from trimester 1 to 3. Women with T1D also had a decrease in the abundance of pathways involved in the synthesis of the BCAAs, leucine, isoleucine and valine, contributed mainly by F. prausnitzii, A. muciniphila, B. adolescentis and some unidenti ed bacteria. BCAAs have a wide range of bene cial anabolic functions including promotion of glucose utilization, protein synthesis, intestinal epithelial integrity and mucin production, milk protein production and immune function [38]. Their de ciency could therefore be detrimental to both mother and fetus in the T1D pregnancy and ENDIA will provides the opportunity to investigate this.
In the third category (taxa that became differentially abundant by trimester 3) we observed that the family Bacteroidaceae and ve species, four of which belonged to the genus Bacteroides (B. caccae, B. uniformis, B. salyersiae and B. faecis) and Parabacteroides distasonis, increased in T1D and decreased in non-T1D women. Bacteroidaceae and its genus Bacteroidess [11,12], and Parabacteroides distasonis [39], were reported to be signi cantly more abundant in children with islet autoimmunity compared to healthy controls. The CMP-3-deoxy-D-manno-octulosonate biosynthesis I complete pathway and enzymes UDP-N-acetylglucosamine 2-epimerase (K01791) and P-N-acetyl-D-mannosamino uronate:lipid I N-acetyl-D-mannosamino uronosyltransferase (K02852) involved in LPS biosynthesis, contributed mostly by bacteria from the genus Bacteroides (including B. caccae) and by Parabacteroides distasonis, increased in T1D and decreased in non-T1D women throughout pregnancy, becoming differentially abundant by trimester 3. Immunostimulatory LPS could contribute to the pro-in ammatory gut microbiome, again potentially increasing risk of complications for mother and fetus in the T1D pregnancy [40]. Within this third category, we also observed that the species Bi dobacterium adolescentis and Ruminococcus bromii, and the order Bi dobacteriales, decreased in T1D by the third trimester. In addition, the phylum Actinobacteria (spp. B. adolescentis) and the order Clostridiales ( spp. Ruminococcus bromii), decreased only in women with T1D. Bi dobacterium adolescentis degrades resistant starches into lactate, acetate and malto-oligosaccharides, which then act as substrates to 'cross-feed' butyrateproducing bacteria [41]. Ruminococcus bromii is a 'keystone' degrader of resistant starches to butyrate [42]. Our functional analysis revealed that the enzyme cyclomaltodextrinase (K01208), involved in the degradation of starch into maltodextrin, contributed mainly by Bi dobacterium adolescentis, decreased in T1D women as pregnancy progressed and was signi cantly less abundant compared to non-T1D women by trimester 3. Enzymes involved in the conversion of pyruvate to acetyl-CoA (pyruvate ferredoxin oxidoreductase [K00171] and phosphate acetyltransferase [K13788]) and acetate (acylphosphatase [K01512]) decreased during pregnancy only in T1D. For the synthesis of butyrate, acetyl-CoA is converted into butyryl-CoA ultimately by butyryl-CoA dehydrogenase (K00248) [42] and trans-2-enoyl-CoA reductase (K00209) [43], both of which also decreased in T1D women by trimester 3. Thus, different bacterial species with diverse biochemical pathways implicated in the production of acetate and butyrate were decreased in T1D compared to non-T1D women, especially towards the end of pregnancy.
We also detected a decrease in T1D women over pregnancy of gut microbiome functions associated with the metabolism of the B-group vitamins B1 (thiamine), B5 (pantothenate), B6 (pyridoxine), B7 (biotin) and B9 (folate). Mammals cannot synthesize B-group vitamins and must acquire them from the diet or gut microorganisms [44]. All B-group vitamins contribute to regulation of immunity-in ammation and their de ciency may be associated with in ammatory disorders [45]. In addition, within the tricarboxylic acid (TCA) or Krebs cycle, which fuels ATP production via acetyl-CoA, B1 is a cofactor for pyruvate dehydrogenase to catalyse formation of acetyl CoA, B5 is a precursor of coenzyme A (CoA) and B7 is a cofactor for acetyl-CoA [44,45]. Succinate, a precursor of propionate and butyrate, is produced by the TCA cycle [46]. Thus, the lower abundance of B1, B5 and B7 could also contribute to the relative de ciency of SCFAs in T1D women. In addition to modulating immune responses, B9 (folate) is essential for the growth and development of the fetus; the association of folate de ciency in early pregnancy with neural tube defects is well-established [47]. However, pregnancy in women with T1D carries an increased risk for a range of fetal malformations [48] and is also associated with an increased risk of pre-eclampsia in which folate de ciency has been implicated [49,50]. B6 de ciency has been associated with in ammatory markers in population-based studies [51] and is reported to be common in T1D [52,53]. Of interest therefore, we found that the key enzymes in B6 synthesis, pyridoxine kinase (K00868) and pyridoxal 5'-phosphate synthase subunits pdxT and pdxS (K08681 and K06215), were decreased in T1D women across pregnancy. The majority of both non-T1D and T1D women reported taking multi B-group vitamins from early pregnancy (Table 1) and there was no difference in the circulating concentrations of vitamins B6 or B9 between non-T1D and T1D women (Table 1). However, the relative de ciency of these vitamin-synthesizing bacteria in T1D women might not only compound other changes in the gut microbiome but underscores the importance of dietary supplementation in this group of women.
Our ndings show that the composition of the gut microbiome not only changes across pregnancy but in a distinct way in women with T1D. By the third trimester, T1D women exhibit a more pro-in ammatory and catabolic gut microbiome pro le, re ected by a de ciency of SCFA-producing bacteria, as well a decrease in B-group vitamin-synthesizing bacteria. These changes could impair epithelial barrier function and contribute to systemic in ammation, a risk marker for pre-eclampsia, which is more common in T1D [54], as well as accentuating the insulin resistance of later pregnancy. Furthermore, a pro-in ammatory gut microbiome in the mother may impact the infant postnatally. In an elegant study in mice, Aguero et al [55] found that transient exposure to an auxotrophic E. coli mutant in the intestine of germ-free mothers in pregnancy accentuated innate immune development in the intestine of their germ-free offspring. This effect was mediated by the transfer, in part via maternal antibodies, of a range of E. coli products across the placenta and in the mother's serum and milk. Thus, with a single gut bacterium, the mother conditioned the immune system of her offspring, before their exposure to the external environment. This was proposed to be an adaptive mechanism to cope with the known higher risk of infection in neonates [55]. The question arises therefore could T1D mothers who display an increased abundance of E. coli and other LPS-producing bacteria induce stronger conditioning of innate immunity in their offspring?
Moreover, if T1D is triggered by an infectious agent [56] this mechanism could account for the lower transmission of T1D from maternal compared to paternal probands [57].

Conclusions
The gut microbiome changes across normal pregnancy. However, in women with T1D these changes are distinct from other pregnancies. They include an increase in pro-in ammatory bacteria, a decrease in antiin ammatory bacteria and a decrease in bacteria that synthesize essential vitamins and metabolites.
This gut microbiome dysbiosis could contribute to the known increased risks for mother and fetus in the T1D pregnancy and activate innate immunity in the fetus to modify its immune reactivity postnatally. In T1D women in pregnancy the relative de ciency of gut bacteria that synthesize a range of antiin ammatory compounds and B-group vitamins has implications for management of the T1D pregnancy. It remains to be determined if intervention by safe means to promote a less proin ammatory gut microbiome, e.g. with dietary ber and other pre-biotics, could decrease the higher risk of maternal and fetal complications in the T1D pregnancy.

Participants, study design and sample collection
This study involved 145 pregnancies in women participating in the ENDIA pregnancy-birth cohort study [22], 51 in healthy women with no history of gestational diabetes and 94 in women with established T1D on daily insulin treatment. The main criterion for participation in ENDIA was an unborn child with a rstdegree relative with T1D. Table 1 provides summary statistics for participant pregnancies, on 139 women with 145 pregnancies (six sibling pairs) and 147 babies (two sets of twins). Therefore, six women were included twice in the study population (each with two different pregnancies). The unit of observation is the pregnancy, and therefore observations from the same mother but different pregnancies have been included as separate observations, as many characteristics might change between pregnancies.
Women provided written informed consent and were enrolled into the study between 2013-2016 at one of eight clinical sites. Up to three study visits occurred during pregnancy, ideally one in each trimester. The study was approved by a Human Research Ethics Committee (HREC) at each clinical site, with the Women's and Children's Health Network HREC in Adelaide acting as the lead HREC under the Australian National Mutual Acceptance Scheme (reference number HREC/16/WCHN/066). ENDIA is registered on the Australia New Zealand Clinical Trials Registry (ACTRN1261300794707).
Maternal and paternal demographics, medical history, past-pregnancy history, pre-pregnancy weight, assisted conception status, and plurality of pregnancy were recorded at the rst opportunity. HLA DR typing was performed on DNA in saliva collected with OG-500 Oragene DNA tubes (DNA Genotek, Ontario, Canada) by TaqMan-based PCR-typing and imputation from three single nucleotide polymorphisms (rs3104413, rs2187668 and rs9275495), as described previously [59].
For continuous responses, where appropriate the summary tables present the mean and standard deviation derived from tting a linear mixed model. The model t for each continuous response adjusts for the fact that the observations from mothers with more than one pregnancy are not fully independent but may be correlated. For some response variables, the assumption of normally distributed residuals was not met. In these analyses, the response variable was transformed using a square root or log transformation, as appropriate. For transformed responses, the back transformed means and approximated standard deviations are presented. A Wald's test is used to determine whether the groups are signi cantly different.
For categorical responses, summary tables show numbers and percentages. The percentage was calculated using the total number of pregnancies or samples as the denominator. To determine whether the distribution of observations between groups for categorical data were similar or not, a generalized linear mixed model was tted, with a random effect for mother. Such models adjust for correlated mother observations. To determine whether groups were signi cantly different, the change in deviance of the nal model (i.e. a likelihood ratio test), which includes and excludes the treatment term, was examined. A then transferred into a sterile 70 mL collection jar. Participants were instructed to store the sample in the refrigerator prior to transport to the laboratory in an insulated bag within 24 h. Samples were divided into aliquots with a sterile spatula in a Biosafety Level 2 cabinet, then stored at -80°C. A total of 354 fecal samples were collected from the 145 pregnancies with either two or three samples collected longitudinally in each pregnancy ( Figure S1). Beta diversity (diversity between microbial communities) was determined with phyloseq (function distance, method="bray"). This function calculates Bray-Curtis coe cients, which measure the distance between communities based on the taxa/functions that they contain and their abundances. Differences in beta diversity were assessed by PERMANOVA using Bray-Curtis dissimilarities with the Adonis function from the vegan [77] R package. For tests that included multiple samples across trimesters from the same participant (i.e. longitudinal analysis) a modi ed version of Adonis, which performs a RMA-PERMANOVA test [27] was employed. This statistical model included T1D status and time with their interaction adjusted as in the alpha diversity model. In addition, interactions between time and other factors were also tested as described in the results section. When an interaction was either signi cant (i.e. FDR<0.05) or borderline signi cant (i.e. FDR greater than 0.05 and less than 0.1) statistical analysis was performed within trimester (i.e. when testing for differences between non-T1D and T1D women) or by separating data from non-T1D women and T1D women (i.e. when testing for differences in time).
Differential abundance of taxa and functions was analysed with the R package limma [78]. Library sizes were normalized using the trimmed mean of log expression ratios (TMM) method [79]. Counts were transformed to log2-counts per million (CPM) with associated precision weights using voom [80]. In the models, a consensus correlation was estimated with the limma duplicateCorrelation function in which "women IDs" are blocks and the consensus correlation is incorporated to account for multiple measurements while estimating statistics using linear models with the lmFit function and empirical Bayes moderated t statistics. For the taxonomic analysis, a correction was applied in limma to account for high sparsity of data (i.e. the large number of zeroes) that leads to underestimation of the "genewise" variances [81]. Differential abundance analyses of gene categories and metabolic pathways from HUMAnN2 was performed similarly to that of taxa but without the sparsity correction in limma. Since we have samples for all possible combinations of T1D status and trimester, this is a factorial design.
Therefore, in order to build our model, factors T1D status and trimester were combined into a single factor with six levels and the comparisons of interests were de ned as contrasts. P-values were adjusted with the Benjamini and Hochberg method to control the FDR. FDR ≤ 0.05 were considered signi cant. FDR greater than 0.05 but less than 0.1 were considered borderline. Taxa or functions signi cantly different with an abundance logFC greater than 0.5 or less than -0.5 and present in at least 50% of the samples in either of the groups being compared were regarded as biologically signi cant. participants provided written informed consent and were free to withdraw from the study at any time.

Consent for publication
Not applicable Availability of data and materials The demultiplexed raw datasets supporting the conclusions of this study can be accessed in the NCBI SRA https://www.ncbi.nlm.nih.gov/sra (project number PRJNA604850). All the commands used to run QIIME2, the python commands used to run HUMAnN2 and the R code used to perform statistical analyses are available at GitHub (https://github.com/PapenfussLab/RothSchulze_pregnancy-gutmicrobiome-T1D) as R markdown coding and knitr html les along with the necessary R objects which contain taxonomic and functional pro les with metadata.

Competing interests
The authors declare they have no competing interests. LCH designed and supervised the study, analysed data and wrote the manuscript with AJR-S.
All authors reviewed the manuscript.        Differentially abundant species (A-L and P-R), families (M, N and S) and orders (O and T) (mean ± SEM) detected by WMS between non-T1D (blue) and T1D (red) women within a speci c trimester or across trimesters. * Denotes a signi cant difference between non-T1D and T1D; + denotes borderline (FDR > 0.05 and< 0.1) signi cance. * or + in the top right corner denotes a difference across the trimesters; * or + between points denotes a difference between groups in that trimester.

Figure 4
Abundance of taxa detected by WMS between non-T1D (blue) and T1D (red) women across trimesters (mean ± SEM). *denotes a signi cant difference in the trimester.

Figure 5
Differentially abundant taxa detected by WMS (mean ± SEM) between trimesters in non-T1D (blue) and T1D (red) women. The * between trimesters denotes a signi cant difference between those trimesters while the * after the trend line denotes signi cant difference between trimesters 1 and 3. The * color denotes if differences between trimesters are within non-T1D (blue) or T1D (red) women.  represent Category 1 functions that were consistently enriched in T1D relative to non-T1D women across all trimesters. (D to Q) represent Category 2 functions that became less abundant in T1D relative to non-T1D women as pregnancy progressed. (R to T) represent Category 3 functions that were decreased from trimester 1 to 3 in T1D women. * Denotes a signi cant difference in the trimester; + denotes borderline (FDR > 0.05 and < 0.1) signi cance. A black * or + in trimester 3 denotes differences between T1D and non-T1D in that trimester; a colored * denotes signi cant differences between trimester 1 and 3 within non-T1D (blue) or T1D (red) women.

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