Abstract

   Infants born to mothers with obesity have increased risk for later
   development of metabolic dysfunction-associated steatotic liver disease
   (MASLD); however early hepatic changes that occur in these infants
   remain unclear. We integrated metabolomic analysis of umbilical cord
   plasma and transcriptomic analysis of cord plasma-exposed hepatocytes
   to examine differences in the intrauterine environment of pregnancies
   with and without maternal obesity. We identified significantly higher
   abundance of fatty acids, bioactive lipids, and upregulation of
   glutathione metabolism in cord plasma from pregnancies with obesity
   compared to normal weight pregnancies. Hepatocytes exposed to cord
   plasma from pregnancies with obesity exhibited distinct transcriptional
   changes that favored cellular injury and inflammation, and impaired
   hepatocyte development compared to hepatocytes exposed to cord plasma
   of normal weight pregnancies. In integrated analysis, metabolite-gene
   relationships were distinct between pregnancies with and without
   obesity, and the abundance of lipids were positively correlated with
   the expression of KDM4D, DTX3, and NOTCH4 and negatively correlated
   with the expression of MT_TS1. Our findings provide novel insights into
   transcriptional changes induced in hepatocytes by circulating factors
   in the intrauterine environment of pregnancies with obesity. Excess
   intrauterine lipids may contribute to hepatic injury, inflammation,
   impaired mitochondrial function, and impaired hepatocyte development in
   infants of pregnancies with obesity.

   Keywords: Maternal obesity, Cord plasma, Metabolomics, Transcriptomics

   Subject terms: Non-alcoholic fatty liver disease, Paediatric research

Introduction

   Non-alcoholic fatty liver disease (MASLD) is currently the most common
   liver disease worldwide^[36]1. The disease spectrum of MASLD ranges
   from simple hepatic steatosis to steatosis with liver inflammation,
   injury, and fibrosis, also known as metabolic dysfunction-associated
   steatohepatitis (MASH). In the U.S., MASLD is the leading cause of
   cirrhosis, liver failure, and hepatocellular carcinoma, and the second
   leading cause of liver transplantation^[37]2–[38]4. Patients with MASLD
   are also at increased risk for comorbidities including obesity, type II
   diabetes, cardiovascular disease, obstructive sleep apnea, and
   polycystic ovary syndrome^[39]1,[40]5. Perhaps more concerning is the
   rise in pediatric MASLD. MASLD is now the most common liver disease in
   children and adolescents^[41]6, and the prevalence of pediatric
   MASLD/MASH has increased over the past 3 decades irrespective of sex,
   age, and sociodemographic index^[42]7. Together, these factors
   emphasize the significant public health burden of MASLD and the need to
   understand how early life factors contribute to MASLD.

   Evidence indicate that offspring of mothers with obesity have increased
   risk for later development of MASLD^[43]8–[44]10. Children born to
   mothers with greater pre-pregnancy body mass index (BMI) had higher
   radiographic findings of hepatic lipid content in infancy and
   adolescence^[45]11–[46]13. These data are supported by rodent and
   non-human primate studies demonstrating the association between
   maternal obesity and fetal offspring hepatic steatosis, lipogenesis,
   inflammation, and oxidative stress that prime offspring to later
   development of steatohepatitis^[47]14–[48]16. With more than one-third
   of all women in the U.S. entering pregnancy with obesity^[49]17, a
   significant number of fetuses may be at increased risk for MASLD.
   However, early hepatocellular changes that occur in infants born to
   mothers with obesity remain poorly understood, largely due to
   difficulties obtaining liver samples in fetal and infant populations.

   In this study, we aimed to use umbilical cord blood plasma,
   metabolomics, and a transcriptomic hepatocyte reporter cell assay to
   examine how the intrauterine environment may impact fetuses of mothers
   with obesity compared to mothers with normal weight. We hypothesized
   that circulating factors in cord plasma of obese pregnancies would
   adversely impact hepatocyte gene expression and hepatocyte development.

Results

Study cohorts

   A total of 15 matched cord plasma samples from pregnancies with obesity
   and 15 samples from normal weight pregnancies met study criteria. The
   overall mean maternal age was 28.7 ± 3.3 years, infant gestational age
   was 39.4 ± 0.6 weeks, and infant birth weight was 3.7 ± 0.3 kg.
   Characteristics of maternal-infant dyads for each group are shown in
   Table [50]1. Groups were similar in maternal age, ethnicity, gravidity,
   parity, pregnancy weight gain, and infant gestational age, growth
   parameters at birth, and sex between groups. The majority of cord
   plasma samples were from vaginal deliveries.

Table 1.

   Clinical characteristics of maternal-infant dyads.
   Normal weight pregnancies (n = 15) Pregnancies with obesity (n = 15)
   P-value
   Maternal
   Age (years) 29.9 ± 2.4 27.5 ± 3.8 0.05^a
   Pre-pregnancy BMI (kg/m^2) 22.7 ± 1.6 37.3 ± 4.7  < 0.001^a*
   Net pregnancy weight gain (kg) 11.8 ± 1.7 10.6 ± 6.5 0.48^a
   Race 0.17
   White 13 (87%) 10 (66.7%)
   Black 1 (6.7%) 5 (33.3%)
   Asian 1 (6.7%) 0 (0%)
   Ethnicity N/A
   Non-Hispanic 15 (100%) 15 (100%)
   Gravidity  > 0.99^c
   1 5 (33.3%) 4 (26.7%)
   2 5 (33.3%) 5 (33.3%)
   3 3 (20%) 3 (20%)
   4 +  2 (13.3%) 3 (20%)
   Parity 0.80^c
   0 5 (33.3%) 4 (26.7%)
   1 6 (40%) 8 (53.3%)
   2 4 (26.7%) 3 (20%)
   SSRI medication 0.60^a
   Yes 1 (6.7%) 3 (20%)
   Mode of Delivery 0.67
   Vaginal delivery 12 (80%) 12 (80%)
   Infant
   Gestational age (weeks) 39.4 ± 0.7 39.3 ± 0.6 0.52^a
   Birth weight (kg) 3.63 ± 0.26 3.73 ± 0.38 0.41^a
   Birth length (cm) 51.6 ± 1.7 52.4 ± 2.7 0.30^a
   Sex  > 0.99^b
   Male 9 (60%) 9 (60%)
   [51]Open in a new tab

   Data are presented as a mean ± standard deviation or n (%). Percentages
   may not add up to 100 due to rounding.^aTwo-sided t-test; ^bChi-square
   test; ^cFisher’s Exact test. *P-value < 0.05. BMI, body mass index; kg,
   kilograms; m, meters; SSRI, selective serotonin reuptake inhibitor; cm,
   centimeters.

Higher fatty acids, bioactive lipids, and glutathione metabolism in cord
plasma from pregnancies with obesity

   To investigate whether there were metabolite differences in the
   intrauterine environment of pregnancies with and without obesity,
   untargeted metabolomic profiling was performed on cord plasma
   specimens. A total of 13,236 metabolites were detected, of which 10,092
   metabolites were considered reliable based on quality of the data. To
   focus the analysis, we examined the most substantial metabolite
   differences in cord plasma, and differentially abundant metabolites
   (DM) were defined as those with a |log2FC|≥ 0.5 (at least a 40%
   difference in abundance) and p-value < 0.05. We found a total of 183 DM
   in cord plasma samples from pregnancies with obesity compared to normal
   weight pregnancies. These DM showed clustering based on maternal
   obesity status (Fig. [52]1A). The majority of DM (83%) were higher in
   abundance in pregnancies with obesity (Fig. [53]1B). There were no
   differences in DM based on infant sex.

Fig. 1.

   [54]Fig. 1
   [55]Open in a new tab

   Metabolite differences in cord plasma cluster based on maternal obesity
   status. (A) Principal component analysis (PCA) plot showing the
   variance in 183 significant differentially abundant metabolites (DM)
   among cord plasma samples from pregnancies with obesity (n = 15) and
   normal weight pregnancies (n = 15). The variance captured within each
   coordinate is shown in parenthesis. Smaller symbols represent
   individual samples, larger symbols represent the centroid (average of
   each group), and the ellipses represent the 95^th percentile of each
   group. Samples from pregnancies with female (F) infants are indicated.
   (B) Heatmap showing differences in the abundance of DM among cord
   plasma samples from pregnancies with obesity and normal weight
   pregnancies. Individual cord samples are shown in columns, and
   individual metabolites are shown in the rows.

   Among DM in cord plasma from pregnancies with obesity compared to
   normal weight pregnancies, 82 (44.8%) could be identified based on a
   known or putative formula (Supplementary Table S1). Figure [56]2A shows
   the distribution of DM based on main class. Fatty acids were the
   predominant class of metabolites that were higher in pregnancies with
   obesity, followed by amino acids and peptides. Select DM by pregnancy
   group are shown in Fig. [57]2B. Very long chain fatty acids ((VLCFA),
   such as tricosanoic, pentacosanoic, heptacosatrienoic,
   tricosatetraenoic, octacosatrienoic acid) and very long chain hydroxy
   fatty acids (hydroxyheptacosanoic, hydroxyheptacosatrienoic, and
   hydroxynonacosatetraenoic acid) were higher in cord plasma from
   pregnancies with obesity compared to normal weight pregnancies.
   Pregnancies with obesity had elevated levels of bioactive signaling
   lipids lysophosphatidylcholine (LPC 22:5) and lysophosphatidic acid
   (LPA 18:3). Additionally, metabolites involved in protection against
   oxidative stress (glutathione and its synthesis bi-product
   2-hydroxybutyric acid) were higher in the cord plasma from pregnancies
   with obesity. Deoxycholate (DCA), a secondary bile acid made by
   intestinal bacteria, was also elevated in pregnancies with obesity. The
   majority of amino acids were branched chain amino acid leucine or
   isoleucine adducts to 2-hydroxybutyric acid. These findings show that
   there are higher fatty acids, bioactive lipid species, and metabolites
   that mitigate reactive oxygen species in the cord plasma from
   pregnancies with obesity compared to normal weight pregnancies.

Fig. 2.

   [58]Fig. 2
   [59]Open in a new tab

   Higher fatty acids, bioactive lipids, and glutathione in cord plasma
   from pregnancies with obesity. (A) Bar graph showing the proportion of
   identified differentially abundant metabolites (DM) in each main class
   of metabolites. The black bars denote higher relative abundance, and
   white bars denote lower relative abundance in pregnancies with obesity
   compared to normal weight pregnancies. Classes of metabolites that were
   each < 1.5% of identified DM are combined as other. Percentages may not
   add up to 100 due to rounding. (B) Dot plots showing abundances for
   select differentially abundant metabolites in the cord plasma from
   pregnancies with obesity and normal weight pregnancies. Means and
   standard deviations are shown. P-value < 0.05 for all metabolite
   comparisons between pregnancy groups. (C) Pathways associated with the
   DM between pregnancies with obesity compared to normal weight
   pregnancies as determined by MetaboAnalyst. The y-axis represents
   -log10 of the p-value and the x-axis represents the pathway impact
   score, or the importance of the pathway. Matched pathways are
   represented as circles. The color and size of each circle corresponds
   to the P-value and pathway impact value, respectively (darker color
   represents smaller P-value and larger circle represents higher impact
   score). Glutathione metabolism impact score = 0.26, P-value = 0.02.

   To examine biological processes associated with identified DM between
   pregnancies with and without obesity, we performed pathway enrichment
   analysis using MetaboAnalyst. We found that pathways involving
   glutathione, glycerophospholipid, porphyrin and chlorophyll,
   proprionate, and tyrosine metabolism and aminoacyl-tRNA biosynthesis
   were significantly enriched by DM in the cord plasma from pregnancies
   with obesity compared to normal weight pregnancies (p-value < 0.05).
   Notably, glutathione metabolism had the highest impact score
   (Fig. [60]2C). Given the role of glutathione as an important
   antioxidant, these findings suggest that infants exposed to maternal
   obesity may have increased intrauterine oxidative stress.

Differences in cord plasma-induced gene expression in hepatocyte reporter
cells

   To investigate whether HepaRG hepatocyte reporter cells respond
   differently to circulating plasma mediators in pregnancies complicated
   by maternal obesity, we compared gene transcripts from RNA-sequencing
   (RNA-seq) of hepatocytes exposed to cord plasma from pregnancies with
   obesity and those exposed to cord plasma from normal weight
   pregnancies. Differentially expressed genes (DEG) were defined as those
   with |Log2FC|> 0.5 (at least a 40% difference in expression) and
   p-value < 0.05. Cord plasma exposure induced a total of 442 DEG in
   hepatocytes that showed distinct clustering based on maternal obesity
   status (Fig. [61]3A). There were 228 (52%) upregulated and 214 (48%)
   downregulated DEG associated with pregnancies with obesity compared to
   normal weight pregnancies (Fig. [62]3B and Supplementary Tables S2 and
   S3). Hepatocytes exposed to cord plasma from pregnancies with obesity
   had higher expression of genes encoding proteins involved in epigenetic
   histone demethylation and promotion of hepatic inflammation and
   fibrogenesis (KDM4D (lysine demethylase 4D)), remodeling of the
   extracellular matrix (MMP21, MMP1, MMP23B, COL10A1), activation of
   immune response (IGHV7-81, IL2RB, RAET1G), and activation of Notch
   signaling that direct progenitor cell specification away from
   hepatocyte differentiation and toward biliary cell fate (NOTCH4 and
   DTX3 (deltex E3 ubiquitin ligase 3, a regulator of Notch signaling))
   compared to normal weight pregnancies. In addition, exposure to cord
   plasma from pregnancies with obesity was associated with lower
   expression of genes involved in hepatocyte differentiation and
   maturation (WNT10A and OSM (oncostatin M)), and mitochondrial function
   (MT-TS1 (mitochondrially encoded tRNA Serine 1) and CKMT1A (ubiquitous
   mitochondrial creatine kinase)). These findings suggest that genes
   involved in epigenetic modification, inflammation, remodeling of the
   extracellular matrix, and inhibition of hepatic cell fate are
   increased, while those involved in promoting hepatocyte differentiation
   and maturation and mitochondrial function are inhibited by circulating
   mediators in the cord plasma from pregnancies with obesity.

Fig. 3.

   [63]Fig. 3
   [64]Open in a new tab

   Differentially expressed genes in cord plasma-exposed hepatocytes. (A)
   Principal component analysis (PCA) plot showing the variance in 442
   significant differentially expressed genes (DEG) in hepatocyte reporter
   cells exposed to cord plasma from pregnancies with obesity and normal
   weight pregnancies. The variance captured within each coordinate is
   shown in parenthesis. Smaller symbols represent individual samples,
   larger symbols represent the centroid (average of each group), and the
   ellipses represent the 95^th percentile of each group. (B) Volcano plot
   showing the distribution of upregulated (red) and downregulated (green)
   DEG induced in hepatocytes exposed to cord plasma from pregnancies with
   obesity compared to normal weight pregnancies. (C) Gene Set Enrichment
   Analysis of DEG showing enrichment of gene sets involved in DNA repair,
   inflammation, and cell cycle regulation in hepatocytes exposed to cord
   plasma from pregnancies with obesity compared to normal weight
   pregnancies.

   We performed unbiased, genome-wide analysis using all of the RNA-seq
   data to examine potential pathways impacted by maternal obesity.
   Directional enrichment of gene sets in hepatocytes exposed to cord
   plasma from pregnancies with and without obesity were determined using
   GSEA. The most enriched gene sets in hepatocytes exposed to cord plasma
   from pregnancies with obesity compared to normal weight pregnancies
   broadly involved DNA repair, inflammation, and cell cycle
   (Fig. [65]3C). These findings indicate that transcriptomic changes
   related to increased DNA repair, inflammation, and regulation of
   cellular growth and proliferation occur in hepatocytes in response to
   cord plasma from obese pregnancies.

Correlation between lipids and gene alterations in cord plasma-exposed
hepatocytes

   To integrate and highlight global correlations across the DM and DEG
   datasets, we first examined the relationship using canonical
   correlation analysis (CCA). CCA is a multivariate statistical method
   used to determine the relationships between two sets of variables
   measured on the same subjects^[66]18. CCA correlation coefficients
   range from −1 to 1 and measure the strength of the association, with
   higher absolute values indicating a stronger relationship between the
   two sets. We found that DM and DEG were highly correlated, CCA
   correlation coefficient = 0.97, and that the first pair of canonical
   variates for metabolites and gene expression showed distinct separation
   of obese and normal weight pregnancies (Fig. [67]4A). These findings
   indicate that DM in cord plasma were closely correlated with DEG in
   cord plasma-exposed hepatocytes and that the correlation pattern
   discriminated pregnancies with obesity from normal weight pregnancies.

Fig. 4.

   [68]Fig. 4
   [69]Open in a new tab

   Relationship between DM in cord plasma and DEG in cord plasma-exposed
   hepatocytes. (A) Scatter plot of the first pair of canonical variates
   for gene expression (X[1]) and metabolites (Y[1]) in canonical
   correlation analysis (CCA). (B) Arrow network plot showing distinct
   clustering of metabolite-gene relationships in pregnancies with obesity
   (orange) and normal weight pregnancies (blue) using genes and
   metabolites identified by sparse generalized canonical correlation
   analysis (SGCCA). Each arrow represents one pregnancy with one end of
   the arrow representing gene expression and the other end representing
   metabolite abundance. Length of arrows represent the strength of the
   relationship with shorter arrows showing stronger correlations between
   metabolites and genes. (C) Heatmap showing significantly correlated
   metabolite-gene pairs between 31 identified DM (columns) in cord plasma
   and DEG encoding proteins, microRNA, and long non-coding RNA in cord
   plasma-exposed hepatocytes (rows) identified by SGCCA. DM and DEG are
   ordered by their loadings on the first SGCCA component from highest to
   lowest (left to right for DM and top to bottom for DEG). Heatmap colors
   represent the Spearman’s ρ of each significant metabolite-gene pair
   correlation (P-value < 0.05).

   To identify core metabolites and genes driving this relationship and to
   determine the strength of their association, we then used sparse
   generalized canonical correlation analysis (SGCCA) and pairwise
   Spearman rank correlation analysis. There were 72 DM and 170 DEG with
   SGCCA |canonical coefficient|> 0. Metabolite-gene relationships
   segregated and clustered based on maternal obesity status
   (Fig. [70]4B). Lipid species, predominantly LPC 22:5, VLCFA, MG 22:1,
   MG 24:2, and very long chain hydroxy fatty acids; and the protein
   coding genes SMCO1, KDM4D, MT_TS1, DTX3, and NOTCH4 were among the main
   contributors to the correlation between metabolites and genes.
   (Supplementary Figure S4). Significantly correlated metabolite-gene
   pairs are shown in Fig. [71]4C. Notably, the abundance LPC 22:5 and
   many VLCFA and their hydroxy derivatives were positively correlated
   with the expression of KDM4D, DTX3, and NOTCH4, and negatively
   correlated with the expression of MT_TS1. These findings indicate that
   the relationship between metabolites in cord plasma and gene expression
   in cord plasma-exposed hepatocytes are distinct in pregnancies with and
   without obesity, and that the abundance of lipids and gene changes that
   may promote hepatic inflammation, activate Notch signaling, and impair
   mitochondrial function are among the top drivers of this relationship.

Discussion

   In this study, we used umbilical cord plasma, metabolomics, and a
   transcriptomic hepatocyte reporter cell assay to investigate exposures
   that may alter hepatocytes and predispose infants of mothers with
   obesity to later MASLD. Our metabolomic analysis of cord plasma showed
   that the intrauterine environment of pregnancies with obesity is
   characterized by significantly higher abundance of fatty acids,
   bioactive lipids, glutathione and its synthesis bi-product
   2-hydroxybutyric acid, and upregulation of glutathione metabolism
   compared to normal weight pregnancies. Hepatocytes exposed to cord
   plasma from pregnancies with obesity exhibited distinct transcriptional
   changes that favor liver injury and inflammation, and impaired
   hepatocyte development compared to hepatocytes exposed to cord plasma
   from normal weight pregnancies. In integrated analyses, we found that
   DM in cord plasma were highly correlated with DEG in cord
   plasma-exposed hepatocytes based on maternal obesity status and that
   the abundance of lipids, particularly LPC 22:5 and VLCFA and their
   hydroxy derivatives, and expression of KDM4D, MT-TS1, DTX3, and NOTCH4,
   were among the top drivers of the distinctive metabolite-gene
   relationship between pregnancies with and without obesity.

   Umbilical cord plasma is in contact with all fetal tissues and is used
   to assess fetal status and the intrauterine fetal
   environment^[72]19–[73]21. Similar to prior studies, we found
   significantly higher abundance of fatty acids in cord plasma from
   pregnancies with obesity compared to normal weight pregnancies.
   Maternal pre-pregnancy BMI has been positively associated with the
   abundance of maternal fatty acids and phospholipids during
   gestation^[74]22 and cord blood fatty acids^[75]23,[76]24. Rodents with
   high-fat diet induced obesity during pregnancy have higher placental
   expression of fatty acid transporters and increased placental transfer
   of maternally derived dietary lipids to the fetus^[77]25,[78]26.
   Interestingly, we also found significantly higher abundance of LPC 22:5
   and LPA 18:3 in cord plasma from pregnancies with obesity. LPCs and
   LPAs are potent signaling molecules that activate inflammatory
   pathways, and both have been directly correlated with steatosis and
   fibrosis in patients with MASLD^[79]27,[80]28. Moreover, laboratory
   studies have shown that LPCs induce lipoapotosis, disrupt mitochondrial
   integrity, enhance cytochrome C release, and increase oxidative stress
   in hepatocytes^[81]29,[82]30. This suggest that fetuses of pregnancies
   with obesity may be exposed to excess fatty acids and potentially
   hepatotoxic lipid species during development.

   These lipid findings are interesting in the context of the higher
   abundance of glutathione and 2-hydroxybutyric acid, and upregulation of
   glutathione metabolism in the cord plasma from pregnancies with obesity
   compared to normal weight pregnancies. Fatty acids are primarily broken
   down in the liver via mitochondrial and peroxisomal β-oxidation,
   increasing reactive oxygen species in the process. However, we
   previously found that the human fetal liver may be uniquely vulnerable
   to lipotoxicity and oxidative injury due to lower expression of genes
   involved in fatty acid metabolism and peroxisome pathways and
   antioxidant enzymes compared to adult liver^[83]31. Glutathione is a
   potent antioxidant that minimizes lipid peroxidation of cellular
   membranes by excess reactive oxygen species, and oxidative stress can
   dramatically increase the rate of hepatic glutathione synthesis^[84]32.
   During the synthesis of glutathione, supplies of L-cysteine for
   glutathione synthesis become limiting and homocysteine is diverted into
   the transsulfuration pathway forming cystathione. When cystathione is
   cleaved to cysteine for incorporation into glutathione,
   2-hydroxybutyrate is released as a byproduct. Taken together, the
   intrauterine exposure to excess fatty acids and bioactive lipids such
   as LPC that disrupt mitochondrial integrity in pregnancies with obesity
   may challenge the fetal liver, resulting in increased oxidative stress.
   Laboratory studies of murine and non-human primates have demonstrated
   increased liver fat, hepatic inflammation, oxidative stress, and
   mitochondrial dysfunction in offspring of obese mothers fed a high-fat
   diet,(Reviewed in^[85]9,[86]33) The results of our metabolomic analysis
   of cord plasma add to the existing literature and suggest that
   lipotoxicity and increased oxidative stress also occur in infants born
   to mothers with obesity.

   In animal models using maternal high-fat diet, fetuses and offspring
   with excess intrauterine lipid exposure exhibit hepatic steatosis,
   inflammation, oxidative stress, apoptosis, and dysregulation of genes
   that prime adult offspring to steatohepatitis^[87]9,[88]14. However,
   early hepatic changes in infants of mothers with obesity remain poorly
   understood due to the lack of liver samples from living fetal/infant
   populations. To overcome limited access to affected organs,
   transcriptomic reporter cell assays using patient plasma or serum to
   induce gene expression in a reporter cell population have been used as
   non-invasive tools to predict and understand a variety of
   diseases^[89]34–[90]37. We have shown the utility of using patient sera
   to induce disease-specific transcriptomic profiles in hepatocyte
   reporter cells in pediatric MASLD^[91]38, and employed the same
   strategy in this study to examine the transcriptional response of
   cultured hepatocytes to circulating factors in the cord plasma for
   insights into early hepatocellular changes in infants of pregnancies
   with and without obesity.

   In our transcriptomic analysis of cord plasma-exposed hepatocytes, we
   demonstrated that cord plasma from pregnancies with obesity induced
   transcriptional changes in key genes and pathways involved in liver
   injury and inflammation. Specifically, the enrichment of DNA repair and
   inflammatory gene sets and increased expression of KDM4D, MMP21, MMP1,
   MMP23B, and COL10A1 in hepatocytes exposed to cord plasma from
   pregnancies with obesity are consistent with the role of these genes in
   hepatic inflammation, regeneration, and fibrosis. KDM4D encodes an
   epigenetic histone demethylase that responds to DNA damage sites to
   promote DNA repair^[92]39. Using animal models and hepatic stellate
   cells, Dong et al. found that KDM4D also promotes hepatic inflammation
   and fibrogenesis by facilitating TLR4 transcription through
   demethylation of H3K3 and activating TLR4/NF-κB signaling
   pathways^[93]40. Matrix metalloproteinases and collagen proteins
   remodel the extracellular matrix in response to hepatocyte
   injury^[94]41. These findings suggest that circulating factors in the
   cord plasma from pregnancies with obesity induce injury and/or
   inflammation in hepatocytes.

   We also found that hepatocytes exposed to cord plasma from pregnancies
   with obesity had lower expression of important genes and pathways
   involved in regulation of cell cycle, cell fate differentiation, and
   hepatocyte maturation when compared to normal weight pregnancies. Wnt
   and Notch signaling are known to be critical regulators of
   hepatobiliary development and cell fate determination in
   embryology^[95]42,[96]43. Notch signaling is critical for
   differentiating bipotential hepatoblasts and adult liver progenitor
   cells to the biliary, instead of hepatocyte lineage and reprogramming
   mature hepatocytes to biliary cells^[97]44. During development, Wnt
   signaling induces blastopore formation, which promotes gastrulation and
   eventually gives rise to the mesoderm and definitive endoderm from
   which the liver is formed^[98]45. Wnt signaling also promotes the
   differentiation of liver progenitor cells into hepatocytes^[99]46. In
   addition, oncostatin M is a cytokine that promotes hepatocyte
   differentiation and maturation^[100]47. Taken together, our findings of
   increased NOTCH4 and DTX3 expression and decreased WNT10A and OSM
   expression in cord plasma-exposed hepatocytes suggest that circulating
   factors in cord plasma from pregnancies with obesity may impair
   processes that are critical in hepatocyte development compared to
   normal weight pregnancies.

   It should be noted that while hepatocyte Notch signaling is quiescent
   after development, aberrant hepatocyte Notch signaling is involved in
   the development of MASLD and MASH. Hepatic Notch activity has been
   positively correlated with insulin resistance, MASLD disease severity,
   and liver transaminase levels in adults with MASLD^[101]48. In mouse
   models of MASH, inhibition of Notch signaling improves liver
   metabolism, ameliorates steatosis, and attenuates MASH-associated liver
   fibrosis^[102]49–[103]51. Recently, Yu et al. showed that activation of
   inter-hepatocyte Jagged1/Notch signaling by hepatocyte TLR4 increased
   Sox9-dependent osteopontin expression and secretion from hepatocytes,
   which activated hepatic stellate cells and drove liver
   fibrosis^[104]52. Whether increased Notch signaling during development
   similarly impact insulin resistance and inflammation in the developing
   liver or relates to reactivation of Notch signaling and MASLD later in
   life is unknown.

   Through integrated analysis of cord plasma metabolomic data and cord
   plasma-exposed hepatocyte transcriptome data, our study found
   distinctive metabolite-gene relationships between pregnancies with and
   without obesity that were driven, in part, by the abundance of lipids,
   particularly LPC 22:5 and VLCFA and their hydroxy derivatives, and
   expression of KDM4D, MT-TS1, DTX3, and NOTCH4. In our correlation of
   metabolite-gene pairs, we found that LPC 22:5 and many VLCFA and their
   hydroxy derivatives were positively correlated to the expression of
   KDM4D, DTX3, and NOTCH4, and negatively correlated with the expression
   of MT_TS1. As noted previously, LPCs induce lipoapotosis, disrupt
   mitochondrial integrity, and increases oxidative stress in hepatocytes.
   VLCFA are essential components of cellular processes including lipid
   metabolism, membrane composition and fluidity, and cell cycle control
   in cell proliferation and differentiation. Elevated VLCFA are seen in
   individuals with metabolic disease including obesity and type II
   diabetes^[105]53. By serving as precursors for lipid signaling
   molecules including ceramides and sphingolipids, elevated VLCFAs can
   alter signal transduction pathways^[106]54. These findings suggest that
   the relationship between metabolites in cord plasma and gene expression
   in cord plasma-exposed hepatocytes are distinct based on maternal
   obesity status, and that lipids of pregnancies with obesity may induce
   gene changes that may promote injury and inflammation, activate Notch
   signaling, and impair mitochondrial function in the fetal liver.

   While epidemiologic and laboratory data suggest that intrauterine
   exposure to maternal overnutrition and obesity predispose offspring to
   MASLD, early changes in humans are lacking due to difficulties
   obtaining liver samples in fetal populations. Here, we used metabolomic
   analysis of cord plasma and transcriptomic analysis of hepatocyte cells
   exposed to cord plasma to examine how circulating factors in the
   intrauterine milieu may influence developing hepatocytes in pregnancies
   with and without obesity. This allowed us to identify distinctive
   metabolite-gene relationships in pregnancies with obesity compared to
   normal weight pregnancies. Our findings suggest that circulating lipids
   may play a role in inducing changes in gene expression in hepatocytes
   that could alter fetal liver development and could be relevant to the
   pathogenesis of MASLD.

   Our study has limitations. First the small sample size may have limited
   our ability to detect differences in other metabolites and our findings
   need to be validated in larger cohorts. Second, because of our focus on
   maternal obesity, we did not include pregnancies of mothers who are
   overweight or underweight, and future studies should include
   pregnancies across the maternal weight spectrum. We also used very
   strict criteria in our cohort selection to limit the confounding
   effects of other maternal diagnoses and pregnancy related
   complications. While this allowed us to examine the impact of
   circulating factors associated with an obese intrauterine environment,
   it limited our ability to examine how other maternal factors, such as
   maternal gestational diabetes mellitus (GDM), interact with maternal
   obesity to influence offspring risk for MASLD and limits the
   generalizability of our findings to other patient groups. Third,
   reporter cells used in this study were adult-derived HepaRG cells and
   may not capture unique changes that occur in developing hepatocytes.
   We, and others, have shown that the fetal liver has decreased
   expression of genes involved in several biological processes including
   xenobiotic, bile acid, and fatty acid metabolism, peroxisome, and
   oxidative phosphorylation pathways compared to adult
   liver^[107]31,[108]55,[109]56; and it is possible that these
   differences may not have been captured in our study. Additionally,
   because cells maintained in culture for prolonged periods of time are
   subject to change and die over time, we were unable to fully mimic the
   prolonged exposure that occurs during pregnancy or any adaptations that
   may take place over time. Next, although bioinformatic tools, such as
   Metaboanalyst, are often used to link metabolomic data to metabolomic
   pathways and facilitate the integration of multiple -omic data, they
   have limitations. Pathway analysis of metabolomic data may be
   misleading due to lower coherence of metabolite levels in linear
   biosynthetic pathways, rapid transport of metabolites to and from their
   sites of synthesis, frequent circulation and exchange of metabolites
   between organs, and tissues, changes in a given metabolite may be due
   to multiple pathways, and bioinformatic limitations and unique
   analytical constraints of metabolites^[110]57. Lastly, eighty percent
   of cord plasma samples from each of our cohorts were from vaginal
   deliveries and may not be entirely representative of Cesarean section
   deliveries as labor has been associated with increased
   cytokines^[111]58 and prolonged second stage of labor has been
   associated with lower cord blood pH and increased nucleated
   cells^[112]59,[113]60. Future in vivo and in vitro studies are also
   needed to further examine mechanisms through which lipids and/or other
   circulating factors alter hepatocyte gene expression in offspring
   exposed to maternal obesity.

   In conclusion, this study provides novel insights into transcriptional
   changes induced in hepatocytes by circulating factors in the
   intrauterine environment of pregnancies with obesity. We identified
   significantly higher fatty acids, bioactive lipids, and upregulation of
   glutathione metabolism in cord plasma from pregnancies with obesity. We
   also found that cord plasma from pregnancies with obesity induced
   transcriptional changes in hepatocytes that favor injury and
   inflammation and impair hepatocyte development compared to cord plasma
   of normal weight pregnancies. We speculate that excess intrauterine
   lipid exposure may contribute to hepatic injury, inflammation, impaired
   mitochondrial function, and impaired hepatocyte development in infants
   of mothers with obesity. Further work should examine how these early
   hepatocyte changes may contribute to later development of MASLD.

Methods

Study design

   This was a cohort study utilizing umbilical cord plasma from
   pregnancies complicated by maternal obesity and pregnancies of mothers
   with normal weight. MCW Tissue Bank records and maternal medical charts
   were reviewed to identify cord plasma samples from healthy full-term
   pregnancies with obesity (pre-pregnancy BMI ≥ 30 kg/m^2) and with
   normal weight (pre-pregnancy BMI 18.5—25 kg/m^2), matched for maternal
   age group and mode of delivery. Exclusion criteria were pregnancies
   with maternal diagnosis of diabetes (gestational and pre-gestational),
   chronic hypertension, hypo/hyperthyroidism, in-vitro fertilization,
   polycystic ovary syndrome, anemia, prescription drug use (except
   selective serotonin inhibitor medications and as needed antiemetics),
   drug/tobacco/alcohol use, multiple gestation, placental abnormalities,
   prolonged rupture of membranes, congenital infection, chromosomal and
   congenital anomalies. Cord plasma samples used in this study were
   prospectively collected by the MCW Tissue Bank between January 2016 to
   December 2018.

Metabolomic profiling

   Untargeted metabolomic profiling was performed with liquid
   chromatography–mass spectroscopy (LC–MS) at the Michigan Regional
   Comprehensive Metabolomics Core (Ann Arbor, MI, USA) as previously
   described^[114]61,[115]62. Briefly, metabolites were extracted from
   cord plasma samples using extraction solvent including internal
   standards, with subsequent supernatant drying and reconstitution in
   water:methanol solution. Samples were analyzed on an Agilent 1290
   Infinity II/6545 qTOF mass spectrometer system with the JetStream
   electrospray ionization source (Agilent Technologies, Inc., Santa
   Clara, CA USA). Each sample was analyzed in both positive and negative
   ion mode electrospray ionization. Chromatographic peaks, representative
   of metabolite features, were detected using commercial software and a
   hybrid targeted/non-targeted approach for metabolomics.
   Semi-quantitative data for known compounds was obtained by manual
   integration using Profinder v8.00 (Agilent Technologies, Santa Clara,
   CA). Non-targeted data analysis was performed using Agilent’s
   MassHunter Find by Molecular Feature workflow (v7.0) with recursion
   using Agilent’s Mass Profiler Pro (v8.0). A combined feature set was
   generated by merging untargeted metabolite features and named
   metabolites into a single metabolite feature list. The combined
   metabolite feature set underwent data reduction using Binner^[116]61.
   Peak intensities below the detection were imputed using the K-nearest
   neighbor algorithm with K = 5 using R package. The coefficient of
   variance (CV) was calculated for each metabolite within each batch
   using the pooled samples. To optimize reliable data, metabolites with a
   CV of greater than 25% and those with less than 50% detection across
   samples were excluded from subsequent analyses. Metabolites were
   characterized and identified by matching their MS/MS spectra to
   authentic standards, in-house library of standards, and online
   databases such as Human Metabolome Database, Madison Metabolomics
   Consortium Database, Kyoto Encyclopedia of Genes and Genomes, and
   Lipidmaps^[117]61,[118]62. Of note, these analytical methods are not
   capable of determining positions of double bonds. Certain compounds
   were identified as fragments, adducts, or homodimers of the original
   compound with high confidence and denoted as adducts of the original
   compound. Different structural isomers of the same compound were noted.
   While all metabolites were considered in the statistical analysis, the
   interpretation of results focuses on compounds that are known or could
   be assigned a putative formula. Pathway analysis of differentially
   abundant metabolites was performed using MetaboAnalyst software
   ([119]https://www.metaboanalyst.ca/).

Hepatocyte cell culture and cord plasma exposure

   Reporter hepatocytes were exposed to cord plasma similar to our prior
   study^[120]38. Briefly, cryopreserved terminally differentiated HepaRG
   cells (BioPredic International, Saint Grégoire, France) were cultured
   per manufacturer guidelines. Cells were resuspended with HepaRG Thaw,
   Plate, and General Purpose medium (Invitrogen; Waltham, MA) and plated
   on to a single 48 well polypropylene tissue culture plate at
   1.06 × 10^6 cells/mL. Cells were then cultured at 37ºC in 5% CO2 and
   fed with HepaRG Maintenance/Metabolism medium (Invitrogen; Waltham, MA)
   on days 1, 4, and 7. After preliminary studies were conducted to
   determine the volume and duration of cord plasma exposure that
   optimized transcriptional response without increased cell death,
   cultured terminally differentiated HepaRG cells were exposed to media
   supplemented with 30% cord plasma (one cord plasma from one unique
   pregnancy per well of the same plate) and incubated at 37ºC in 5% CO2.
   After 24 h, cord plasma-exposed cells were harvested with Trizol (Life
   Technologies; Carlsbad, CA) and stored at −80ºC until use.

RNA preparation

   Total mRNA was isolated from cord plasma-exposed cells using the RNeasy
   Mini Kit with on-column DNase treatment (Qiagen; Germantown, MD) per
   manufacturer instructions and quantified by Qubit and quantitative
   real-time PCR (qRT-PCR). RNA integrity number (RIN) was determined
   using the Agilent Technologies 2100 Bioanalyzer (Santa Clara, CA) and
   all samples with RIN > 9 underwent RNA-sequencing (RNA-seq).

Library construction, quality control, and RNA-sequencing

   Sequencing libraries of RNA from cord plasma-exposed HepaRG cells were
   prepared by Novogene Corporation Inc. (Sacramento, CA) using the low
   input pipeline and sequenced on an NovaSeq 6000 (PE150) to generate
   6 Gb of raw data per sample.

Computational analysis

   Quantification analysis of RNA-seq data was performed by Novogene
   Corporation Inc. (Sacramento, CA). Briefly, FastQC software (version
   0.11) was used to validate raw sequence data quality. Reads were
   aligned to the Human reference genome (GRCh38) using Hisat2 (version
   2.0.5). FeatureCounts (version 1.5.0-p3)^[121]63 was used to count the
   read numbers mapped to each gene, and fragments per kilobase of
   transcript sequence per million base pairs sequenced was calculated
   based on the length of the gene and read count mapped to the gene.
   Differential expression analysis of RNA-seq data was done using DESeq2
   R package (version 1.20.0)^[122]64 and the Benjamini–Hochberg approach
   to estimate false discovery rate and adjust p-values for multiple
   testing. All gene expression data were used as an input for gene
   set/pathway analysis with Gene Set Enrichment Analysis software
   (version 4.0.3) (GSEA,
   [123]http://software.broadinstitute.org/gsea/index.jsp)^[124]65, using
   the Molecular Signatures Database hallmark gene set collection
   generated by computational methodology to include genes from
   well-defined biological processes^[125]66.

Integrated analysis of cord plasma metabolites and hepatocyte reporter gene
expression

   Dimension reduction approaches were used to integrate differentially
   abundant metabolites (DM) and differentially expressed genes (DEG)
   datasets. Canonical correlation analysis (CCA) was performed using all
   DM and DEG. Sparse generalized canonical correlation analysis (SGCCA)
   of all DM and DEG were used to identify strongly associated
   metabolite-gene pairs and univariate pairwise Spearman correlation
   analysis was used to determine the strength of association between DM
   and DEG pairs with SGCCA |canonical coefficient|> 0.

Statistics

   Data were presented as mean and standard deviation (SD) or n (%).
   Continuous variables were compared by two-sample t-test while
   categorical variables were compared by Chi-square test or Fisher’s
   exact test. Metabolite data were log2 transformed after imputation and
   compared using a two-sample t-test between pregnancies with obesity and
   normal weight pregnancies. Log2 fold change (FC) were then calculated.
   Unsupervised principal component analysis (PCA) of the significant
   metabolites was performed. Hierarchical clustering of the significant
   metabolites was also performed using average linkage clustering with
   Euclidian distance for patients and correlation distance for
   metabolites to generate a heatmap. Two-sided tests were used and,
   unless otherwise stated, p-value < 0.05 was considered significant.
   Statistical analyses were performed using SAS 9.4 (SAS Institute, Cary,
   NC, USA), R software^[126]67, and CIMminer
   ([127]https://discover.nci.nih.gov/cimminer/home.do).

Acknowledgements