Gestational diabetes, human milk oligosaccharide concentrations, and their links to infant weight gain and the gut microbiome in a United States observational cohort

Abstract

Background

Gestational diabetes mellitus (GDM) increases offspring obesity risk, but whether this occurs via changes in human milk composition, including alterations in human milk oligosaccharides (HMOs), is unknown.

Objectives

This study aimed to identify differences in HMO concentrations in mothers with and without GDM and test whether GDM-associated HMOs are associated with infant growth, body composition, and fecal microbiome characteristics over the first 6-mo of life.

Methods

Human milk was collected at 1-mo postpartum from 337 females (49 with GDM) who fed their infants breastmilk exclusively. HMOs were quantified by high-performance liquid chromatography and multivariate regression models were used to test differences in HMO concentrations by GDM status (false discovery rate adjustment for multiple testing set at q < 0.05). HMOs associated with GDM were then tested for associations with infant growth, body composition, and 1 and 6-mo infant fecal microbial abundances measured by metagenomic whole-genome sequencing.

Results

Participants with GDM had ∼1 SD higher milk 6′sialyllactose (6′SL) {[β (95% confidence interval): 0.58 (0.20, 0.96)] and lacto-N-fucopentaose III (LNFP III) III [95% CI: 0.55 (0.16, 0.94)]} compared with those without GDM and 6′SL concentration was also positively associated with weight and length gain. Although infants of mothers with GDM had lower 1-mo fecal α-diversity and altered abundances of 6 of 56 microbial species detected compared with those without GDM, microbial features were not associated with the concentration of either 6′SL or LNFP III and evidence for mediation of GDM-growth and GDM-microbiome by HMOs was not found.

Conclusions

Mothers with a GDM diagnosis had higher milk concentrations of LNFP III and 6′SL, and 6′SL was in turn associated with increased infant growth rate, but neither HMO was associated with differential infant gut microbial abundances. The results suggest that the link between 6′SL and faster infant growth, if causal, occurs via mechanisms independent of the infant gut microbiome.

This study was registered at clinicaltrials.gov as NCT03301753.

Introduction

The prevalence of gestational diabetes mellitus (GDM) in the United States has progressively increased and is currently 8.3% [1,2], contributing to a variety of adverse perinatal and chronic disease outcomes. Although GDM typically resolves after delivery, it can have a lasting imprint on both parent [3] and offspring [3,4]. The mechanisms responsible for these latent disease outcomes in offspring of mothers with GDM are not completely understood, but findings from the literature on lactational programming of chronic disease [5] suggest that differences in human milk composition may play a mediating role linking maternal metabolic health to growth, development, and future disease risk in the offspring.

Human milk oligosaccharides (HMOs) are complex carbohydrates accounting for the third most abundant solid component in human milk after lactose and lipids [6]. HMOs act as prebiotics and antimicrobials, modulate the immune system, and have been linked to infant growth and development [7]. Infants fed human milk have different gut microbiomes (e.g., higher Bifidobacteria abundance) than infants fed standard formulas [8], which largely lack HMOs. Evidence in the preterm population suggests that HMO-enriched formulas may reduce the risk of necrotizing enterocolitis [9] and HMO profiles have been shown to be associated with infant gut microbiome characteristics [[10], [11], [12], [13], [14]]. Although this certainly suggests HMO concentrations are important in shaping the infant gut microbiome, it remains to be confirmed the degree to which common clinical conditions like GDM alter human milk HMO levels and the impact of these alterations on infant growth and health.

Evidence supporting the association of metabolic dysfunction with HMO variation includes both altered activity of glycosyltransferases involved in HMO biosynthesis [15,16] and altered serum concentrations of HMOs [17] in pregnancies affected by GDM, and associations of human milk HMO concentrations with pregnancy glucose dysregulation in lactating mothers without GDM [[18], [19], [20]]. Very few studies have tested differences in HMO profiles by GDM status [[20], [21], [22]]. The existing studies are generally small and have not jointly assessed whether GDM-associated differences in HMO profiles meaningfully impact infant growth and gut microbiome development.

To address this gap, we used data from the Mothers and Infants LinKed for Healthy Growth (MILk) observational cohort study of mother–infant dyads to test 3 aims (Figure 1). The first aim was to test differences in mature milk HMO composition between participants with and without a GDM diagnosis, adjusting for potential confounders and technical covariates. The second and third aims were to extend prior studies by assessing the relationship of GDM, and GDM-associated HMOs, with infant growth rate and body composition and with infant fecal microbiome characteristics, respectively. We hypothesized that HMO profiles would differ in participants with and without GDM, and that the concentration of GDM-associated HMOs would mediate relationships of GDM with infant growth and gut microbiome characteristics.

FIGURE 1.

Overview of the study aims. Aim 1: compare human milk HMO concentrations by gestational diabetes status. Aim 2a: test relationship of GDM status with infant growth and body composition. Aim 2b: test relationship of GDM-associated HMO concentrations with infant growth and body composition. Aim 3a: test relationship of GDM with infant gut microbial abundances. Aim 3b: test relationship of GDM-associated HMO concentrations with infant gut microbial abundances. GDM, gestational diabetes mellitus; HMO, human milk oligosaccharide.

Methods

Study design and participants

The MILk Study is an ongoing longitudinal cohort study of human milk composition and infant outcomes. Participants were enrolled between 15 October, 2014 and 23 September, 2024 at 2 United States study centers: the University of Minnesota in Minneapolis, MN and the Oklahoma University Health Sciences Center (OUHSC) in Oklahoma City, OK, as described previously [23]. Participants at the University of Minnesota center were recruited and enrolled during pregnancy from patient populations at 2 sites: through collaboration with researchers at a regional health insurance and health care provider system (HealthPartners) and at the MHealth-Fairview Health system affiliated with the University of Minnesota. The participants at the Oklahoma center were recruited through OUHSC obstetric clinics, through mass mailings to the OUHSC community, and through area events for families with children. The eligibility criteria were as follows: females aged 21–45 y old; pregnant; prepregnancy BMI of 18.5–45.0 kg/m²; giving birth to a baby weighing 2500–4500 g between 37 and 42 wk of gestation; and reporting an intention to only feed their singleton infants breastmilk for ≤3 mo. Exclusion criteria included drinking alcohol (≥1 drinks/wk) or ever smoking during pregnancy or lactation; having a history of or currently presenting with diabetes mellitus (type 1 or 2); having a presumed or known congenital metabolic or endocrine disease (other than GDM), or a congenital illness likely to interfere with the conduct of the study; and not speaking English. After delivery, participants were excluded if they had a preterm birth, a low-birth-weight infant, or were not exclusively feeding breastmilk to their infants at their 1-mo study visit. Exclusive breastmilk feeding was defined as the provision of only human milk, vitamins, and a cumulative total of <8 ounces of formula to the infant since birth, as well as no infant formula or water in the 2 wk before the 1-mo study visit. Dyads were scheduled for study visits at 1, 3, and 6 mo postpartum which occurred between 22 December, 2014 and 21 March, 2025.

The selection of samples for HMO analysis was as follows: of the 465 dyads that had completed the 1-mo study visit. We selected all those who had a GDM diagnosis (n = 53) and a random selection of 300 who did not have a GDM diagnosis. The random selection was accomplished using PROC SURVEYSELECT in SAS (SAS Institute, version 9.4). Subsequently, it was determined that 4 with GDM and 12 without GDM no longer had sufficient 1-mo milk for HMO analysis, leaving a total sample of 337 maternal–infant dyads (49 with GDM and 288 without GDM) in the present analysis. All participants provided written informed consent, and the institutional review boards at the University of Minnesota and Health Partners Institute approved all study protocols (STUDY00009021).

Definition and treatment of GDM

GDM was defined based on diagnostic codes within participant electronic health records at each recruitment site. Criteria for GDM diagnosis varied across sites; the OUHSC center and UM-Fairview site used the Carpenter-Coustan criteria: at 24–28 wk of gestation, a 50 g 1-h oral glucose challenge test (OGCT) was followed by a 100 g 3-h oral glucose tolerance test if OGCT >130 g/dL [24]. The HealthPartners site used a 50 g 1-h OGCT followed by a 75 g 2-h oral glucose tolerance test if OGCT >134 g/dL, and first-trimester screening for gestational diabetes was conducted for those at increased risk [25]. Participants with GDM were treated via insulin (50%), dietary modification (44%), oral medication (4%), or both diet and medication (2%).

Covariates

Prepregnancy BMI (kg/m²) was estimated using the first available height and weight in the pregnancy medical record (no later than 8 wk of gestation). Gestational weight gain (GWG) was calculated by delivery weight minus the prepregnancy weight. Information on maternal educational attainment, household income group, diet quality, and infant feeding mode (exclusively breastmilk, some breastmilk, or no breastmilk) at 3 and 6 mo postpartum) was collected through self-reported questionnaires at study visits. Although all dyads were exclusively breastmilk feeding (via breast and/or via bottle) at 1 mo postpartum when the milk sample was collected for HMO quantitation, a small percentage of participants began to use formula for feeding their infant by the 3 and 6-mo time points, which is captured by a 3-level infant feeding variable at the 3 and 6-mo time points: exclusive breastmilk feeding, mixed breastmilk and formula feeding, or formula feeding. Maternal diet quality was assessed using the 2015 Healthy Eating Index total score calculated from past-month food frequency intakes self-reported at 1-mo postpartum, where a higher score indicates greater adherence to United States Dietary Guidelines [26]. Race was self-reported by the parent using the following categories to choose from: White, Black/African/African American, Asian/Asian American, American Indian/Native American, Hawaiian/Pacific Islander, another race not listed, >1 race, and prefer not to say. Ethnicity (language) was self-reported as Hispanic. Non-Hispanic or Prefer not to say. Information abstracted from medical records included maternal age, mode of delivery (Cesarean section, assisted vaginal birth, or spontaneous vaginal birth), maternal group B Streptococcus status, and gestational age at delivery (in wk), whereas infant exact age at study visit, study center, and HMO measurement batch (1 or 2) were recorded by study staff.

Human milk collection

The details of milk collection and the assay procedure have been reported elsewhere [23,27]. Briefly, human milk was obtained at 1 mo (±5 d) postpartum when all infants were reported by their mothers to be consuming nothing but human milk. Participants provided a complete single breast expression sample at the study center from the right breast using a hospital-grade electric breast pump (Medela Symphony; Medela AG.), ∼2 h after they had breastfed their infant at the study center to control for time since last feed, and typically between 10:00 and 12:00, to control for diurnal variation. Human milk was gently mixed, aliquoted into 2 mL microcentrifuge tubes, and stored within 20 min of collection at −80°C until analysis to minimize degradation of milk bioactives.

HMO analysis

Human milk aliquots were shipped on dry ice to the University of California San Diego for analysis. HMO concentrations were measured by HPLC (Dionex Ultimate 3000, Dionex, now Thermo Fisher) on an amide-80 column (15 cm length, 2 mm inner diameter, 3 μm particle size; Tosoh Bioscience) with fluorescent detection after adding an internal standard and labeling with the fluorescent tag 2-aminobzamide as previously described [28]. Concentration and relative abundance of 19 of the most abundant and commonly observed HMOs with available reference standards were quantified: 2′Fucosyllactose (2′FL), 3-fucosyllactose (3FL), 3′-Sialyllactose (3′SL), 6′-Siallactose (6′SL), difucosyllactose (DFLac), difucosyllacto-N-hexaose (DFLNH), difucosyllacto-N-tetrose (DFLNT), disialyllacto-N-hexaose (DSLNH), disialyllacto-N-tetraose (DSLNT), fucodisialyllacto-N-hexaose (FDSLNH), fucosyllacto-N-hexaose (FLNH), LNFP I, lacto-N-fucopentaose II (LNFP II), lacto-N-fucopentaose III (LNFP III), LNH, lacto-N-neotetraose (LNnT), lacto-N-tetrose (LNT), sialyl-lacto-N-tetraose b (LSTb), and sialyl-lacto-N-tetarose c (LSTc). To determine secretor status, we used a 2′FL concentration threshold of 600 nmol/mL. For 2′FL concentrations between 300 and 600 nmol/mL, we also considered DFLac with a 10 nmol/mL threshold. HMO-bound fucose and HMO-bound sialic acid were calculated on a molar basis. The assays are performed in 96-well plate format, and 1 well is always reserved for the same pooled milk sample as a reference to track plate-to-plate variation. Technical replicate coefficient of variation (CV)% ranges from 2.1% to 4.8% depending on the HMO. Highly abundant HMOs like 2′FL and LNT have lower CV% values (2%–3%), whereas low abundance HMOs like LNFP III have higher CV% values (4%–5%). Samples are diluted before HPLC injection so that all HMOs are within the detection range, and the lower detection limit is <100 pmol after dilution.

Infant anthropometrics and body composition

At 1, 3, and 6 mo of age, infant anthropometrics (weight and length) were obtained as described previously [29]. The WHO growth charts were used to calculate sex- and age-standardized z-scores [30] [weight-for-age (WAZ); length-for-age (LAZ); and weight-for-length (WLZ)]. Infant body composition [fat mass (FM) fat-free mass (FFM); and percent body fat] was assessed using air displacement plethysmography [31] (Pea Pod, Cosmed Ltd) at 1 and 3 mo and using dual energy X-ray absorptiometry (Lunar iDXAv11-30.062 scanner; analysis via enCore 2007 software, GE) at 6 mo, when infants are too large to fit within the Pea Pod [32].

Infant fecal microbiome

Infant fecal collection, metagenomic DNA extraction, and microbial taxon abundance estimation have been previously described [[33], [34], [35]]. Stool samples were either collected from diapers during a study visit or at home by a parent. For samples collected at study visits, samples were immediately frozen at −80°C. Samples collected at home (within 3 wk of scheduled visit) were stored in 2 mL cryovials with 600 μL RNALater (Ambion/Invitrogen), mailed at ambient temperature, and stored at −80°C on arrival to the lab at the University of Minnesota. DNA extraction was performed with PowerSoil kit (QIAGEN). Sequencing libraries were constructed using the Illumina Nextera XT kit (Illumina) and sequenced on an Illumina NovaSeq system (Illumina) using the S4 flow cell with 2 × 150 bp paired-end V4 chemistry kit at the University of Minnesota Genomics Center.

Microbial taxon abundances were generated by first processing metagenomic fastq files with Shi7 version 1.0.1: sequences were trimmed, filtered by quality scores, and stitched per the learned parameters in Shi7 [36]. Processed sequences were aligned to a reference genome database generated from GTDB r95 (https://gtdb.ecogenomic.org/stats/r95) using BURST version 1.00 (see https://zenodo.org/records/3779009). Taxonomy tables were generated using the “embalmulate” function with “Ggtrim” activated [37]

Statistical analysis

R Studio [38] SAS Enterprise Guide 8.3 (SAS Institute), and other specific R packages (see below) were used for all statistical analyses. Crude differences in maternal and infant characteristics by GDM status were tested using Pearson’s chi-square, Fisher’s exact, and Welch t-tests. Missing data were handled using multiple imputation (PROC MI procedure in SAS) followed by SAS PROC MIANALYZE to obtain estimates and SEs. Associations between GDM and HMO composition were analyzed using separate multiple linear regression models for each of the 19 individual HMOs and the 2 summary HMO traits (HMO-bound fucose and HMO-bound sialic acid). Given the strong dependence of multiple HMO concentrations on secretor status, maternal secretor phenotype was included as a main effect and as an interaction term with GDM status in all analyses. For these and subsequent multivariate regression models, both potential confounders and covariates of the outcome were included as covariates to obtain direct effect estimates of the exposure-outcome relationships. Potential confounders included those occurring before and known to be associated with GDM (maternal prepregnancy BMI, age, maternal education, household income, and parity), factors potentially associated with HMO concentration (maternal diet quality, infant sex, and age at milk collection) and technical covariates (HMO batch number and study center). Models testing the relationships of GDM with infant age- and sex-specific WAZ, WLZ, and LAZ scores included the same confounders, covariates, and technical factors as above, as well as infant gestational age at birth. Infant body composition models were additionally adjusted for infant sex and exact age at study visit. The same covariates were included in models testing the relationship of 1 mo HMOs with infant growth and body composition measures, except in this case potential confounders also included GDM, GWG, and delivery mode and covariates included infant feeding mode at 3 and 6 mo as they are not on the causal pathway of HMOs to infant outcomes. If there were HMOs that were associated with both GDM and with infant growth or body composition, we conducted an exploratory mediation analysis SAS PROC CAUSALMED [39], with the identified GDM-associated HMOs as mediators of the GDM–infant growth relationships.

Regarding the associations of GDM-associated HMO and the infant fecal microbiome, α-diversity metrics were calculated from the species-level count matrix using the “Vegan” R package [40]. The Shannon index was calculated with the “diversity” function, and rarefied species richness was calculated with the “rarefy” function with a subsample size of the minimum of species-assigned read counts for a single sample (177,729 counts). This rarefaction approach eliminates differences in read depth between samples. Infant fecal metagenomic data were summarized as taxon abundances and filtered to include only species-level taxa with relative abundance >0.001 in ≤10% of 1-mo or 6-mo samples, leaving 56 species. A centered log-ratio (CLR) transformation was performed on the relative abundances of each sample. We used a linear mixed model to test for associations between GDM and each species abundance or species richness. To leverage our longitudinal sampling of the infant gut microbiome, we included species-level abundances from both time points in linear mixed models including data from 1- to 6-mo time points. We modeled the species CLR-transformed relative abundance or diversity metric as the response variable; included fixed effects of GDM status, potential confounders listed above as well as GWG and delivery mode, and covariates potentially associated with the outcome (infant age at sample collection, maternal group B Streptococcus status, fecal sample collection site (study center or home), and infant feeding status at 6 mo (exclusively breastmilk or any formula feeding). The dyad unique identifier was included as a random effect. Linear mixed models were performed with the “lmerTest” package in R [41]. We used a Benjamini–Hochberg false discovery rate to control for the 56 individual species tests. To test for associations with GDM status and α-diversity or species abundance at a single time point (1 or 6 mo), we used a simple linear regression including covariates of the variables used as fixed effects in the mixed linear model above, excluding the 1 or 6 mo binary variable and adding infant age in days at time of study visit. Finally, to test the relationship of GDM-associated HMO concentrations with GDM-associated infant fecal taxon abundances or species richness, we used a similar linear mixed model as above, with CLR-transformed taxon abundance as the response variable, but with the addition of HMO batch as a fixed effect. For all analyses, Benjamini–Hochberg false discovery rate was used to correct for multiple testing.

Results

Maternal and infant characteristics

Overall, there were 337 mother–infant dyads in the study sample, of which 49 had received a diagnosis of GDM. Characteristics of the study sample, by GDM status, are presented in Table 1. Gestational age at delivery was significantly lower in the GDM group as compared with non-GDM. Participants with GDM were more likely to be older at delivery, have higher BMI, and to deliver via cesarean section. Infants whose mothers had a GDM diagnosis had slightly lower WAZ at birth. Distribution of secretor phenotype percentage, diet quality score, GWG, parity (birth order of infant), infant feeding at 3 and 6 mo, and the proportion of male compared with female infants were similar between those with and without GDM. Descriptive statistics on infant anthropometrics and body composition at 1, 3, and 6 mo are provided in Supplemental Table 1. Infants exposed to GDM had higher WAZ at 1 mo, WLZ at 3 and 6 mo, and FFM at 1 mo, and a greater decrement in LAZ from birth to 6 mo compared with those unexposed to GDM.

TABLE 1.

Characteristics of the study sample, by maternal GDM status.

	Non-GDM (n = 288)		GDM (n = 49)
	n	%	n	%
Secretor phenotype (n, %)
Secretor	214	74.3	40	81.6
Non-secretor	74	25.7	9	18.4
Age at delivery, y [mean ± SD (n missing)]	31.1 ± 4.0 (2)		34.1 ± 4.4 (0)
Race (n, %)
American Indian/Native American/Hawaiian Native	4	1.4	1	2.0
Asian	6	2.1	9	18.4
Black/African American	11	3.9	2	4.1
White	247	86.7	34	69.4
Other race not listed	8	2.8	1	2.0
>1 race	6	2.1	1	2.0
Prefer not to say	6	2.1	1	2.0
Ethnicity (n, %)
Hispanic or Latino	6	2.1	2	4.1
Not Hispanic or Latino	277	96.2	47	95.9
Prefer not to say	3	1.0	3	6.1
Diet quality score [mean ± SD (n missing)]	65.9 ± 8.6 (7)		64.2 ± 9.8 (3)
Prepregnancy BMI (mean ± SD (n missing))	26.4 ± 5.6 (1)		29.2 ± 6.6 (0)
Gestational weight gain (n, %) 1
Insufficient	71	24.7	18	36.7
Appropriate	87	30.2	13	26.5
Excessive	121	42.0	13	26.5
Unknown/missing	9	3.1	5	10.2
Mode of delivery (n, %)
Spontaneous vaginal delivery	212	73.6	25	51.0
Assisted vaginal delivery	7	2.4	3	6.1
Cesarean section delivery	62	21.5	21	42.9
Unknown/missing	7	2.4	0	0
Birth order of infant (n, %)
1	115	39.9	18	36.7
2	107	37.2	22	44.9
≥3	60	20.8	7	14.3
Unknown/missing	6	2.1	2	4.1
Gestational age at delivery, wk [mean ± SD (n missing)]	39.7 ± 1.3 (4)		38.7 ± 1.1 (0)
Infant feeding at 3 mo (n, %)
Exclusively breast milk	264	91.7	37	75.5
Mixed feeding	16	5.6	4	8.2
Exclusively formula	3	1.0	0	0.0
Missing	5	1.7	8	16.3
Infant feeding at 6 mo (n, %)
Exclusively breast milk	218	75.7	29	59.2
Mixed feeding	39	13.5	6	12.2
Exclusively formula	19	6.6	2	4.1
Missing	12	4.2	12	24.5
Infant sex (n, %)
Male	150	52.1	24	49.0
Female	138	47.9	25	51.0
Infant age at 1 mo visit (mean ± SD)	32.3 ± 6.4		32.3 ± 6.8
Infant birth anthropometrics2 [mean ± SD (n missing)]
Weight-for-age z-score	0.43 ± 0.86 (5)		0.03 ± 0.88 (0)
Length-for-age z-score	1.16 ± 1.2 (11)		0.89 ± 1.1 (3)
Weight-for-length z-score	−0.74 ± 1.40 (12)		−0.96 ± 1.43 (3)
Laboratory batch (n, %)
1	58	20.1	31	63.3
2	230	79.9	18	36.7
Study site (n, %)
Oklahoma	91	31.6	2	4.1
Minnesota: Health Partners	182	63.2	16	32.6
Minnesota: University of Minnesota	15	5.2	31	63.3

Abbreviations: GDM, gestational diabetes mellitus.

¹Institute of Medicine. Weight gain during pregnancy: reexamining the guidelines. Washington, DC: National Academies Press, 2009.

²World Health Organization. WHO Child Growth Standards: Length/Height-for-Age, Weight-for-Age, Weight-for-Length, Weight-for-Height and Body Mass Index-for-Age: Methods and Development. Geneva, Switzerland: World Health Organization, 2006.

HMO concentrations by GDM status

Supplemental Table 2 presents the mean concentration and abundance (as a percent of total HMO) for the 19 individual HMO and the 4 HMO groups and traits (HMO-bound fucose, HMO-bound sialic acid, sum of HMO, and HMO diversity and evenness) measured in this analysis. The mean total concentration of HMOs was ∼10.3 g/L (or ∼ 1% by volume) which is what is typically observed in human milk, and the top 5 most abundant HMOs were 2′FL, 3FL, LNT, LNFP II, and LNFP I. Although the mean concentrations of most HMOs were very similar to those of other studies, mean LNFP III concentration (0.018 g/L) was lower than many other published studies of 1 mo milk (∼0.2–0.4 g/L) [42]. Table 2 presents the regression estimates for the association of GDM status with each of the 19 HMOs and HMO groups, ordered by nominal P value. There were 4 individual HMOs and 1 HMO group that were nominally higher in those with GDM as compared with those without GDM: 6′SL, LNFP III, DSLNT, LSTb and total HMO-bound sialic acid. After multiple testing correction, milk 6′SL {β [95% confidence interval (CI): 0.58 (0.20, 0.96)] and LNFP III [95% CI 0.55 (0.16, 0.94)]} remained higher in those with GDM as compared with those without GDM. These differences were relatively large in magnitude, equivalent to 1 SD higher in those with GDM. However, the CIs were wide, meaning there is significant overlap in values between the 2 groups. For these 2 HMOs and 5 other HMOs (DFLac, LSTb, LSTc, DSLNT, and HMO-bound sialic acid), there were significant GDM × secretor status interactions, and in all of them, the associations of GDM with HMO concentration were stronger in nonsecretors than secretors (data not shown). Figure 2 displays concentrations of 6′SL and LNFP III (μg/mL) by secretor status and GDM status.

TABLE 2.

Multivariate linear regression models testing differences in HMO concentrations in human milk obtained at 1 mo from patients diagnosed with GDM, as compared with those without a GDM diagnosis, n = 337.

HMO	Log mean ± SD (μg/mL)	Unadjusted estimate1 β (95% CI)	Adjusted estimate2 β (95% CI)	q value3
LNFP III	2.95 ± 0.52	0.11 (−0.05, 0.27)	0.58 (0.20, 0.96)	0.026
6'SL	6.06 ± 0.6	0.34 (0.16, 0.52)	0.55 (0.16, 0.94)	0.040
DSLNT	5.30 ± 0.48	−0.04 (−0.19, 0.11)	0.45 (0.10, 0.80)	0.103
Sia	7.76 ± 0.39	0.08 (−0.04, 0.20)	0.30 (0.04, 0.56)	0.165
LSTb	4.68 ± 0.49	0.05 (−0.10, 0.20)	0.39 (0.03, 0.75)	0.271
Fuc	9.36 ± 0.37	−0.09 (−0.20, 0.02)	−0.20 (−0.44, 0.03)	0.509
DSLNH	5.72 ± 0.52	−0.04 (−0.20, 0.11)	0.32 (−0.06, 0.70)	0.800
3FL	6.44 ± 0.89	−0.35 (−0.62, −0.08)	−0.41 (−0.93, 0.11)	0.948
3'SL	4.60 ± 1.06	0.52 (0.20, 0.83)	−0.43 (−1.17, 0.30)	0.991
DFLac	4.48 ± 1.84	−0.35 (−0.91, 0.21)	−0.67 (−1.34, 0.01)	1.000
FDSLNH	5.31 ± 1.12	−0.60 (−0.94, −0.26)	−0.42 (−1.01, 0.16)	1.000
Diversity	4.93 ± 1.41	−0.23 (−0.66, 0.19)	−0.58 (−1.61, 0.45)	1.000
LNT	6.80 ± 0.58	−0.14 (−0.32, 0.03)	0.25 (−0.15, 0.64)	1.000
FLNH	6.08 ± 0.47	0.00 (−0.15, 0.14)	0.19 (−0.15, 0.52)	1.000
DFLNT	6.22 ± 1.45	−0.48 (−0.91, −0.04)	−0.50 (−1.57, 0.56)	1.000
2'FL	7.33 ± 1.29	0.17 (−0.22, 0.56)	0.19 (−0.22, 0.60)	1.000
LNFP II	6.56 ± 0.74	−0.20 (−0.43, 0.02)	−0.17 (−0.62, 0.28)	1.000
SUM	9.20 ± 0.26	−0.04 (−0.12, 0.04)	0.08 (−0.12, 0.27)	1.000
DFLNH	5.22 ± 0.63	−0.11 (−0.30, 0.08)	−0.14 (−0.61, 0.33)	1.000
LNnT	4.82 ± 0.71	−0.13 (−0.35, 0.08)	−0.14 (−0.62, 0.34)	1.000
LSTc	5.04 ± 0.59	−0.10 (−0.28, 0.08)	0.09 (−0.33, 0.50)	1.000
LNH	4.85 ± 0.62	−0.18 (−0.37, 0.01)	−0.13 (−0.57, 0.32)	1.000
LNFP I	6.43 ± 0.95	0.17 (−0.12, 0.45)	0.01 (−0.52, 0.53)	1.000

Abbreviations: 2′FL, 2′ fucosyllactose; 3′SL/6′SL, 3′/6, sialyllactose; CI, confidence interval; DFLac, difucosyllactose; DFLNH, difucosyllacto-N-hexaose; DSLNH, disialyllacto-N-hexaose; DSLNT, disialyllacto-N-tetraose; FDSLNH, fucodisialyllacto-N-hexaose; Fuc, Fucose-bound; GDM, gestational diabetes mellitus; HMO, human milk oligosaccharide; LNFP, lacto-N-fucopentaose; LNH, lacto-N- hexose; LNnT, lacto-N-neotetraose; LSTb/ LSTc, sialyllacto-N-tetraose b/c; Sia, sialic acid bound HMO.

¹The unadjusted linear regression model included GDM, maternal secretor phenotype status, and secretor status × GDM interaction.

²The adjusted model additionally included prepregnancy BMI, age at delivery, maternal education, maternal dietary quality index, household income, infant birth order, infant exact age at the 1 mo visit, HMO batch number, and study center. Effect estimates represent the difference in the natural logarithm of the HMO concentrations between GDM groups, with positive coefficients indicating HMO concentrations that were higher in those with GDM, and negative coefficients indicating HMO concentrations were lower in those with GDM, as compared with those without GDM.

³q values are adjusted P values using the Holm–Bonferroni false discovery rate method to control for multiple comparisons.

FIGURE 2.

Concentrations of the HMOs LNFP III and 6′SL in 1 mo mature milk, by secretor status and GDM status. Violin plots show secretor stratified concentrations (μg/mL) of LNFP III and 6′SL by GDM status. Bars show median and interquartile ranges. GDM, gestational diabetes mellitus; HMO, human milk oligosaccharide; LNFP III, lacto-N-fucopentaose III; SL, sialyllactose.

GDM, GDM-associated HMOs, and infant growth and body composition

In multivariate adjusted models, offspring affected by GDM had lower WLZ at 3 and 6 mo and higher LAZ at 3 mo (Supplemental Table 3). Infant body composition variables did not differ appreciably by GDM status (Supplemental Table 4), and there were no statistically significant associations observed between LNFP III and any infant anthropometric or body composition measures (P > 0.05 for all models; data not shown). 6′SL concentration was not associated with variation in either anthropometrics or body composition at 1 mo, but there were positive associations of 6′SL concentration with infant anthropometrics at later ages (WAZ at 3 and 6 mo, LAZ at 3 mo, WLZ at 6 mo) and with the changes in WAZ and LAZ from birth to 6 mo. For example, for each SD greater log 6′SL concentration, infant WLZ at 6 mo was ∼0.2 z-score greater (Figure 3, Supplemental Table 5). However, the variance around these estimates was wide. To illustrate the observed relationship of 6′SL to infant weight gain, we plotted the model-predicted values of WAZ change from birth to 6 mo against log 6′SL (Figure 4). From the lower to the higher end of the observed milk log-6′SL values, the difference in infant WAZ change was appreciable. Among infants consuming milk with the lowest 6′SL concentrations, WAZ from birth to 6 mo of age decreased by and average of ∼1.0 z-score, whereas among infants consuming milk with the highest 6′SL concentrations, WAZ changed negligibly (mean change of ∼0.0 Z) over the first 6 mo. In terms of infant body composition, 6′SL concentration was positively associated with FFM at 3 mo, and both FM and FFM at 6 mo, and with the change in FFM from 1 to 3 mo (Figure 3, Supplemental Table 5). Again, for each 1SD greater log 6′SL, 6 mo infant FM and FFM were higher by ∼0.2 SD, with wide variance. There were only 2 infant outcomes (LAZ 3 mo and WLZ 6 mo) that were associated with both GDM status and 6′SL. Mediation analysis did not detect significant mediation of these GDM-growth relationships by 6′SL (data not shown), and the results did not differ in models that did or did not adjust for maternal secretor status.

FIGURE 3.

Multivariate adjusted standardized linear regression effect estimates of the relationship of infant anthropometrics and body composition at 1, 3, and 6 mo, and their longitudinal changes, to the concentration of the human milk oligosaccharide (HMO) 6′SL assessed at 1 mo. Bars show the multivariate adjusted linear regression estimate (± SE) for infant anthropometric z-scores (WHO 2006 references), standardized body composition measures (air displacement plethysmography at 1 and 3 mo and dual energy X-ray absorptiometry at 6 mo), and their longitudinal changes during early infancy against natural logarithm-transformed (log) human milk 6′SL concentrations. Anthropometric z-score changes are the difference between the 0 and 6 mo z-scores, and body composition changes are the difference between the 1 and 3 mo standardized measures. Anthropometric z-score models adjusted for maternal age, prepregnancy BMI, gestational weight gain, delivery mode, education category, income category, delivery mode, gestational diabetes status, diet quality score, infant gestational age at birth, infant feeding mode at 3 and 6 mo postpartum (all infants exclusively breastfed at 1 mo postpartum), study center, and HMO batch. Body composition models included the same covariates and in addition included infant sex and infant exact age at each study visit. BF, body fat; FFM, fat-free mass; FM, fat mass; HMO, human milk oligosaccharide; LAZ, length-for-age z-score; SL, sialyllactose; WAZ, weight-for-age z-score; WLZ, weight-for-length z-score.

FIGURE 4.

Scatter plot showing the relationship of human milk 6′SL concentration measured at 1 mo to predicted values of infant WAZ change between birth and 6 mo. The figure shows predicted values from multivariate linear regression models of the change in WAZ from birth to 6 mo regressed on log10 6′SL concentration (μg/mL). Best fit linear regression line shown in solid black line. 95% confidence limits for the best fit line shown in blue shading. 95% individual value prediction limits shown in blue-dashed lines. Covariates in the regression model included maternal age, prepregnancy BMI, gestational weight gain, delivery mode, education category, income category, delivery mode, gestational diabetes status, diet quality score, infant gestational age at birth, infant feeding mode at 3 and 6 mo postpartum (all infants exclusively breastfed at 1 mo postpartum), study center, and HMO batch. HMO, human milk oligosaccharide; SL, sialyllactose; WAZ, weight-for-age z-score; WLZ, weight-for-length z-score.

GDM, GDM-associated HMOs, and infant fecal microbiome characteristics

We next assessed the relationship between GDM status and the infant gut microbiome. GDM was associated with lower within-sample species richness when both 1- and 6-mo samples were modeled together in a linear mixed model and for 1-mo samples only, but not for 6-mo samples only. Infant microbiome Shannon index, another measure of within-sample diversity that measures both richness and evenness, was also lower (at 1 mo only) among infants born to individuals with GDM. We then tested for associations between GDM status and individual microbial species abundances in the infant gut. We found that 6 of 56 species were associated with GDM status (q value < 5%, Table 3). The strongest association was with the species Klebsiella variicola which was more abundant in infants whose mothers had GDM as was Klebsiella A oxytoca. Two Clostridium species (C. tertium and C. P. perfringens) were less abundant in infants whose mothers had GDM (GDM effect estimate = −0.70, P value = 2.9 × 10⁻⁴, q value = 0.016). Bacteroides fragilis and Bacteroides ovatus were also less abundant in infants whose mothers had GDM (Table 3). Hypothesizing that GDM-associated HMO concentrations could play a role in shaping these microbial characteristics, we then tested the correlations between the concentration of each of the nominally GDM-associated individual HMOs (6′SL, LNFP III, DSLNT, and DFLac) and the species richness for each of the 6 infant fecal microbial species that were associated with GDM (above). None of these associations were statistically different from zero after multiple testing correction (Supplemental Table 6), and therefore no formal test of mediation was applied.

TABLE 3.

Associations between maternal gestational diabetes (GDM) status and species abundances in the infant fecal microbiome at 1 and 6 mo,¹ by adjusted p (q) value.

Taxon	Estimate (β)	SE	P value	q value2
Klebsiella variicola	0.744	0.213	0.001	0.020
Clostridium tertium	−0.767	0.224	0.001	0.020
Bacteroides fragilis A	−0.813	0.244	0.001	0.020
Clostridium P perfringens	−0.618	0.199	0.002	0.032
Klebsiella A oxytoca	0.691	0.240	0.005	0.049
Bacteroides ovatus	−0.647	0.228	0.005	0.049
Escherichia sp004211955	0.597	0.231	0.011	0.087
Escherichia sp002965065	0.606	0.241	0.013	0.091
Lacticaseibacillus rhamnosus	0.541	0.224	0.016	0.095
Bacteroides thetaiotaomicron	−0.574	0.238	0.017	0.095
Bacteroides fragilis	−0.571	0.243	0.020	0.102
Bacteroides xylanisolvens	−0.548	0.246	0.028	0.129
Bacteroides sp003463205	−0.494	0.242	0.043	0.171
Klebsiella pneumoniae	0.443	0.218	0.044	0.171
Parabacteroides distasonis	−0.511	0.253	0.046	0.171
Klebsiella A michiganensis	0.445	0.227	0.053	0.184
Escherichia coli	0.467	0.256	0.070	0.226
Escherichia sp005843885	0.436	0.241	0.073	0.226
Parabacteroides distasonis A	−0.437	0.252	0.086	0.233
Bifidobacterium kashiwanohense	0.364	0.214	0.092	0.233
Streptococcus sp000187445	0.383	0.226	0.092	0.233
Phocaeicola dorei	−0.427	0.254	0.094	0.233
Bifidobacterium infantis	0.340	0.203	0.096	0.233
Bifidobacterium kashiwanohense A	0.352	0.214	0.102	0.237
Bifidobacterium bifidum	0.353	0.239	0.142	0.302
Escherichia coli D	0.353	0.242	0.148	0.302
Escherichia dysenteriae	0.341	0.235	0.150	0.302
Escherichia flexneri	0.339	0.235	0.151	0.302
Enterococcus faecalis	−0.305	0.224	0.176	0.340
Escherichia fergusonii	0.319	0.243	0.190	0.348
Bifidobacterium catenulatum	0.295	0.227	0.196	0.348
Escherichia coli C	0.314	0.243	0.199	0.348
Bacteroides caecimuris	−0.312	0.249	0.211	0.358
Bacteroides sp900557355	−0.271	0.234	0.248	0.408
Bifidobacterium reuteri	0.235	0.212	0.269	0.430
Clostridium paraputrificum	0.246	0.238	0.302	0.470
Bifidobacterium pseudocatenulatum	0.235	0.233	0.315	0.478
Veillonella atypica	−0.151	0.211	0.476	0.695
Veillonella parvula A	−0.155	0.221	0.484	0.695
Veillonella parvula	−0.143	0.222	0.520	0.729
Veillonella dispar A	−0.134	0.218	0.540	0.732
Bifidobacterium breve	0.133	0.221	0.549	0.732
Bacteroides uniformis	−0.145	0.260	0.579	0.754
Eggerthella lenta	0.113	0.213	0.598	0.760
Phocaeicola sartorii	−0.122	0.268	0.649	0.807
Erysipelatoclostridium ramosum	−0.080	0.220	0.718	0.865
Clostridium neonatale	0.080	0.227	0.726	0.865
Phocaeicola sp900554435	0.060	0.249	0.811	0.935
Bifidobacterium longum	0.052	0.225	0.818	0.935
Longicatena innocuum	−0.050	0.246	0.841	0.935
Veillonella sp900556785	−0.040	0.224	0.858	0.935
43−108 sp001915545	−0.043	0.259	0.868	0.935
Bacteroides finegoldii	−0.025	0.262	0.924	0.976
Phocaeicola massiliensis	0.008	0.260	0.976	0.985
Ruminococcus B gnavus	−0.004	0.219	0.985	0.985
Phocaeicola vulgatus	−0.005	0.262	0.985	0.985

Abbreviation: GDM, gestational diabetes mellitus.

¹Associations between gestational diabetes status and infant fecal microbial species abundances at 1 and 6 mo (estimated via shotgun sequencing) were assessed using a linear mixed-effect regression model. Fixed effects estimates were obtained for GDM status (shown above, β) as well as for covariates maternal prepregnancy BMI, gestational weight gain, delivery mode, birth order, maternal group B Streptococcus status, exclusive breastfeeding status at 6 mo, infant age at time of sample, and study center. A dyad unique identifier was included as a random effect to account for the repeated measures of fecal microbial species abundance at 1 and 6 mo.

²q values are adjusted P values using the Benjamini–Hochberg false discovery rate method to control for multiple comparisons.

Discussion

GDM and human milk HMO concentrations

To our knowledge, this is the first study in a United States population to test associations between HMO concentrations and GDM status, infant growth, and infant fecal microbiome characteristics. Compared with individuals without GDM, GDM was associated with higher levels of the neutral fucosylated HMO LNFP III, and the sialylated HMO 6′SL. The differences were relatively large in magnitude (∼1 SD higher in GDM than in those without GDM) and were stronger in nonsecretors than in secretors. Prior studies on the topic were all based on patient populations in China. Li et al. [20] found that 6′SL and 8 other HMOs (2′FL, 3′SL, LNT, LNFP I, LSTb, LSTc, 3′SLNFP II, 6′SLNFP VI, and DFLNHa) were higher in the milk from GDM-exposed participants at day 42 of lactation [20]. The greater number of identified HMO differences by GDM status in the study by Li et al. [20] may be due to both greater precision (higher numbers of GDM-affected participants) or residual confounding. In a study of n = 15 secretor-positive participants, Dou et al. [21] found higher levels of LNnT, LNFP II, 3′SL, and DSLNT in milk from those with GDM, whereas Ma et al. [22] examined HMO composition across lactation stage, finding that milk LNFP and DFLNH sharply decreased over the course of lactation in those with GDM as compared with those without GDM. There is a strong theoretical basis for the association of GDM with human milk HMO abundance given that it alters levels of the glycosyltransferases that govern their biosynthesis [16]; nonetheless, there is as yet little replication among published associations between glucose dysregulation phenotypes and specific milk HMOs. This could stem from the wide variety of methods used to quantify HMOs [[20], [21], [22]], baseline cohort characteristics like secretor status prevalence, GDM treatment, and other factors. Approximately half of the GDM cases in the present study were treated with medication (primarily insulin) and half with diet and weight control. Given the potent effect of insulin and glucose on milk synthesis [16] and suggested influences of maternal diet on HMO levels [43], it is likely that differences in the severity and mode of treatment of GDM across studies may have contributed to differences in findings.

GDM-associated HMOs and infant growth and body composition

We are among the first studies to test the relationship of HMO concentrations with serial infant body composition and the first to include offspring affected by GDM. One of the stronger findings in this study was the positive association of 1-mo human milk 6′SL with infant body size (WAZ and LAZ) from 0 to 6 mo and in both the fat-free and fat components of body weight, regardless of secretor or GDM status. This replicates findings in a nondiabetic United States cohort in which milk 6′SL and LNFP III were positively associated with FM from 2 to 6 mo [44] and by Cheema et al. [45] who also found a positive association between 6′SL and infant weight. The strongest evidence for milk 6′SL as a mediator of infant development is in the realm of myelination and cognition [46,47]. Research investigating HMOs and infant growth is relatively scant; a recent systematic review [48] identified only 13 papers to date testing associations of individual HMOs or HMO features with infant growth, and the results were largely inconsistent. Two of these (Tonon et al. [7], n = 78, Lagstrom et al. [49], n = 802) found negative associations between the HMO 6′SL and infant weight, opposite in direction to our study findings. As Brockway et al. [48] noted, the variability in methodology (e.g., lactation stage, exclusivity of human milk feeding, analytic techniques, and body composition assessment) continues to hinder clear conclusions about the overall impact of individual HMOs on infant growth. In addition, despite the consistent 6-SL-infant growth associations, we found no evidence that 6′SL mediates relationships of GDM to infant size and tissue accrual. Our findings suggest that 6′SL may positively impact early physical development; however, there was wide variance in the effects, meaning that 6′SL is clearly but one of many other factors influencing growth rate in the population.

GDM and fecal microbiome

GDM diagnosis was associated with decreased species richness and differences in the abundance of multiple bacterial taxa in the infant gut microbiome. Species members within Clostridium and Bacteroides were less abundant in those diagnosed with GDM, whereas members of Klebsiella were more abundant in those with GDM, as compared with those without GDM. Previous studies have found varying patterns of associations between GDM and the infant gut microbiome. Song et al. [50] similarly observed reduced alpha diversity and lower levels of the genus Clostridium in the group with GDM, but also saw higher Bifidobacterium and Ralstonia in the GDM group which we did not observe. Caution should be applied in comparing results across studies given the many methodological differences, particularly that of the method of assaying the microbiome (16S sequencing compared with the shotgun metagenomic sequencing employed here) [[50], [51], [52], [53]]. Contrary to our hypothesis, we did not find evidence that human milk HMOs associated with GDM mediated the relationship between GDM and the infant gut microbiome. We can conclude that although in our study GDM status was associated with differences in both human milk HMO profiles and the infant gut microbiome, these occur independently of one another. Furthermore, the results suggest that the link between 6′SL and faster infant growth, if causal, occurs via mechanisms independent of the infant gut microbiome.

Limitations and strengths

There were limitations of the present study. As an observational cohort study, causality of the reported associations cannot be inferred. HMOs were measured at only a single time point, and represent the most abundant HMO species with chemically pure standards for absolute quantification, which is only a partial picture of the variation in ∼150 known HMOs. We did not estimate infant HMO daily intake which may have biased the results of the infant growth and infant gut microbiome analyses toward the null hypothesis. The mothers in the study had all received consistent prenatal care, were predominantly White and non-Hispanic, had relatively high educational status, and were exclusively feeding their infants breastmilk, and so results are not representative of all maternal–infant dyads in the United States. There were also several important strengths of the present study. Infant feeding mode, stage of lactation, effects of time of day and time since last feed, method of milk expression and handling were all rigorously controlled by the study design. Rigorous confounder and multiple testing adjustments were applied. Infant body composition assessment provided insight into differential tissue accretion that is possible with anthropometry alone, and metagenomic sequencing provided species-level information on the infant gut microbiome impossible with 16S sequencing.

In conclusion, in a United States cohort of lactating mothers and their exclusively breastmilk-fed term infants, GDM was associated with differences in the infant gut microbiome and in the concentrations of 2 HMOs, 1 of which (6′SL) was also positively associated with both infant weight and length gains, and fat and FFM accrual. However, contrary to our hypotheses, we did not find statistical evidence that HMO levels mediated the relationship between GDM and subsequent infant growth and body composition or gut microbiome development. Overall, the role of human milk HMOs in the intergenerational programming of metabolic disease remains unclear, with few studies on the topic and little replication of results among them likely stemming from differences in study population, laboratory methods for quantitating HMOs, and statistical approaches. Given the importance of identifying the bioactive elements in human milk that together enhance infant growth and health, support for a large multicohort milk composition consortium is needed to facilitate coordinated, collaborative, and methodologically harmonized human milk research for the United States population.

Author contributions

The authors’ responsibilities were as follows – EWD, CAG, LB: developed the study concept and design; AJF, NY, EWD, LB, KP, SP, DAF, AK, SH, KJ: were responsible for data acquisition and curation; AJF, EMN, KEJ, AK: performed the statistical analyses; AJF, NY: performed the HMO quantifications; KEJ, CAG: performed the microbiome analyses; EMN: performed the infant growth and body composition analyses; EWD, LB, CAG, AJF, KP, SP, DAF, SH, TTG, KEJ: interpreted the results of the study; EWD: oversaw the project and acquired the necessary funding; AJF: wrote the first draft of the manuscript; EWD: revised the first draft of the manuscript; and all authors: contributed to critically revising and approving the final version.

Data availability

Data described in the manuscript, code book, and analytic code will be made available on request pending review by the study investigators.

Declaration of generative AI and AI-assisted technologies in the writing process

Neither generative AI or other AI-assisted technologies were used in the writing process for this manuscript.

Funding

The Mothers and Infants LinKed for Healthy Growth Study is supported by NIH/National Institute of Child Health and Development (NICHD) grant R01HD080444 to EWD, DAF, and KP and NIH/NICHD grant R01HD109830 to EWD, CAG, LB, and EMI. ENN was supported by NIH/NICHD grant K99HD108276. KEJ was supported by MinnCResT postdoctoral training grant T32 DE007288 and NIH/NICHD National Research Service Award (NRSA) postdoctoral fellowship F32 HD105364. TTG is supported by a University of Minnesota Presidential Postdoctoral Fellowship. LB is UC San Diego Chair of Collaborative Human Milk Research endowed by the Family Larsson-Rosenquist Foundation, Switzerland.

Conflict of interest

KP reports a relationship with HealthPartners Institute that includes: funding grants. All other authors report no conflicts of interest.

Appendix A. Supplementary data

The following is the Supplementary data to this article: