Introduction
Cholestasis is a decrease in bile formation or flow, causing bile retention, liver toxicity, and conjugated hyperbilirubinemia. Premature infants, defined as infants born before 37 weeks of gestation, have numerous risk factors for cholestasis, including immaturity of metabolic pathways, small-for-gestational-age status, severe illness, and other factors limiting enteral nutrition, thus exposing them to more parenteral nutrition (PN), which is a known hepatotoxin.1 Parenteral nutrition-associated cholestasis (PNAC) is defined as a conjugated bilirubin level ≥ 2 mg/dL with PN exposure for 14 or more days without another underlying etiology. PNAC is a very common entity in the neonatal intensive care unit (NICU): in one cohort of infants less than 30 weeks gestational age, the incidence of cholestasis was 10.7% within the first 28 days of life, with greater PN exposure correlating to a significantly increased risk.2 Left untreated, alongside ongoing PN exposure, PNAC can result in significant hepatic fibrosis and is an indication for liver transplant.1
The human microbiome is a set of microorganisms that inhabit and interact with the human body, contributing to both health and disease.3 The preterm infant gut microbiome has been connected to conditions such as necrotizing enterocolitis (NEC) and late-onset sepsis, both of which are also associated with an increased risk of cholestasis.1,4,5 Research has associated both PN use and cholestasis with microbiome changes in term and near-term cohorts.6–8 To date, there has been little research into the gut microbiome and PNAC in preterm infants. By further clarifying microbiome patterns associated with PNAC, opportunities for both diagnostic and therapeutic interventions may emerge for this vulnerable patient population.
Bile acids, a major component of bile, are synthesized in the liver through cholesterol metabolism. Released into the gut, they serve as detergents to facilitate nutrient absorption and are recirculated back to the liver through the portal system in a phenomenon known as enterohepatic circulation. Interruptions in enterohepatic circulation and alterations in bile acid homeostasis are implicated in both nutrient malabsorption and liver disease pathogenesis.9 Bile acid metabolism is dependent on and regulated by the gut microbiome.10 There is minimal published research comparing the bile acid profiles of preterm infants to those of term infants and adults. Based on a small study of nine healthy infants from 32–36 weeks of gestation, preterm infants were found to have a relatively small bile acid pool in comparison to term infants.11 The infant bile acid profile transitions to a more “adult-like” composition during the second year of life, on a similar timeframe to the transition to a mature gut microbiome.12 There is no published research to date investigating the role of the gut microbiome in bile acid metabolism in preterm infants on PN.
Given the complex and interrelated nature of cholestasis, bile acid metabolism, PN, and the microbiome, there is a significant opportunity for research into these pathologies. The purpose of this study was to examine the fecal microbiome and bile acid content of premature infants on PN to determine if there is a relationship with the development of PNAC. We hypothesize that in preterm infants on PN and varying amounts of enteral nutrition, features of the early gut microbiome and differences in fecal bile acid content are associated with the development of PNAC.
Methods
Study design
Study subjects’ clinical data and biospecimens were examined retrospectively from the Microbiome in Neonatal Development (MIND) clinical cohort of preterm infants at the University of Chicago Comer Children’s Hospital, which was established to determine the relationship between the preterm infant microbiome and neurodevelopment. The inclusion criteria of the MIND clinical cohort were preterm infants (born at less than 37 weeks gestational age) without any genetic syndromes or severe congenital anomalies that were deemed viable after birth by the attending physician. Thus, this clinical cohort had the characteristics necessary to address the study research questions without any inherent restraints, as the enrollment criteria were inclusive, the clinical data collection had a broad mandate, and the biospecimen acquisition was microbiome-focused. An additional inclusion criterion for the specific goals of this study was applied to select a subset of patients who received PN, for the purpose of comparing preterm infant patients who did and did not develop cholestasis while on PN, and cholestasis could be accurately diagnosed, as serial bilirubin measurements are taken from infants while on PN per hospital protocol. The Duchossois Family Institute (DFI) was utilized for preterm infant fecal sample analysis; the DFI was established at the University of Chicago for the specific purpose of investigating the microbiome to improve human health, and therefore houses facilities containing state-of-the-art methods for analyzing the microbiome. The methods employed in this study include 16S rRNA gene sequencing and deep shotgun metagenomics sequencing, which are the gold standard in the microbiome research field for understanding microbiome composition and functional capacity, respectively, and a comprehensive bile acid panel, which included the 49 most common and abundant primary, secondary, and glyco-/tauro-conjugated subclasses of bile acids. Together, these measurements allowed for a detailed evaluation of whether and how the fecal microbiome and bile acid content are associated with the development of cholestasis in preterm infants while on PN.
Cohort identification
Participants were drawn from the MIND study, which is approved by the University of Chicago Institutional Review Board (IRB16-1431) and is conducted in accordance with both the Declarations of Helsinki and Istanbul. After receiving written informed consent from a parent, patients were enrolled in the NICU at the University of Chicago Comer Children’s Hospital between November 2010 and July 2018. Inclusion criteria were infants with a gestational age of less than 37 weeks. Infants with genetic or severe congenital anomalies (including major congenital heart disease, kidney, lung, or brain malformations) and infants deemed not to be viable by the attending physician were excluded. For the present study, all infants in the MIND cohort with exposure to PN for at least four weeks, with stool samples and bilirubin measurements obtained during the time period of PN exposure, were included.
Clinical data and sample collection
Infant diapers were collected weekly by nursing staff, and samples were stored at −80°C in order to enable assessment of the fecal microbiome. As per hospital protocol, bilirubin levels were measured at least once weekly for infants receiving PN. Additional potentially contributing clinical variables were gathered from patient charts by extracting information from the electronic medical record system at the University of Chicago Comer Children’s Hospital. Data collected included gestational age at birth, sex, birth weight, information on nutrition regimens (percent feeds human milk, average received enteral nutrition in mL per kg body weight, average received PN in mL per kg body weight, and average received lipid in mL per kg body weight), total number of days of antibiotics received, length of NICU stay, and corrected gestational age at discharge. Chart review also captured additional diagnoses, such as NEC and sepsis, and additional therapies utilized, such as Ursodiol and SMOFlipid (Fresenius Kabi; Bad Homburg, Germany). For purposes of this study, NEC was defined by modified Bell’s criteria, with disease classified if meeting at least Bell’s stage 2.13 Sepsis consisted of both those with early (<72 h) and late (>72 h) blood culture positivity.
Illumina 16S rRNA gene sequencing and data processing
Genomic DNA extraction and Illumina 16S rRNA gene sequencing of patient fecal samples were conducted by the Environmental Sample Preparation and Sequencing Facility at Argonne National Laboratory (Lemont, IL, USA).14,15 The raw data were subsequently processed using the DADA2 pipeline16 implemented into QIIME2 version 2019.717 and classified to the genus level using the IDTAXA method18 included in R package DECIPHER version 2.14.0 and the Genome Taxonomy Database version 8.19 An additional classification step for determining species-like groups (i.e., the highest taxonomic resolution possible by 16S rRNA gene sequencing alone) was employed using the online National Center for Biotechnology Information Nucleotide Basic Local Alignment Search Tool (4,5)20 with a ≥97% identity threshold, and very low abundant species-like groups of <0.1% mean abundance across all samples were culled. Low quality samples that had a sequence read depth of <1,000 were also removed prior to analysis. For the calculation of α-diversity, the R package iNEXT version 2.0.20 was utilized to obtain the richness, Simpson, and Shannon diversity metrics from the genus-level data.21 In terms of β-diversity analysis, the taxonomic levels of phylum, family, and genus were individually studied. For the determination of overall β-diversity, community typing analysis was conducted using the R package DirichletMultinomial version 1.36.0 with 1 to n-1 community types considered and the optimum solution selected by the minimum Laplace value.22 For examining individual taxa, the data were center-log ratio transformed via the R package ALDEx2 version 1.18.0 to account for compositionality prior to statistical analysis.23
Fecal bile acid analysis
Bile acid analysis was performed on available samples of infants who were on PN at the time of stool sample acquisition (n = 15) through the University of Chicago DFI. In brief, bile acids were extracted with organic solvent, dried down, and resuspended for direct analysis. The bile acids had previously been validated by the DFI through retention time and fragmentation comparison to standards and available databases. Using negative mode liquid chromatography–electrospray ionization–quadrupole time-of-flight MS ((-)LC-ESI-QTOF-MS, Agilent, 6546), 49 bile acids from the primary, secondary, and glyco-/tauro-conjugated subclasses were analyzed.24 In addition to retention time validation, the standard intact and fragment masses are routinely detected with differences < 5 ppm compared to calculated values. Metabolite-normalized peak abundances were recorded in the supplementary materials (Supplementary Table 1; qualitative). Quantitative values were also determined for 17 of the most common and abundant bile acids in human fecal samples; they are recorded in the supplementary materials (Supplementary Table 1; quantitative).
Shotgun metagenomics sequencing and data processing
Shotgun metagenomics sequencing was additionally conducted at the DFI on the same set of stool samples sent for bile acid analysis (n = 15), which were collected while the infants were on PN. The DFI completed genomic DNA extraction (QIAamp PowerFecal Pro DNA Kit, Qiagen 51804), library generation (QIAseq FX Library Kit, Qiagen 180479), and shotgun metagenomics sequencing on the Illumina NovaSeq 6000 platform using the 2 × 150 paired-end read cassette. Data obtained from the DFI were processed using the bioBakery whole metagenome shotgun workflow version 3.0.0.,25 which includes the KneadData tool for quality control and removal of host reads, and the HUMAnN tool for generation of microbial gene abundance profiles stratified by contributing microbial taxa utilizing the Kyoto Encyclopedia of Genes and Genomes (hereinafter referred to as KEGG) database.26 The compositional data produced from this workflow were center-log ratio transformed, with zeroes imputed by the nonparametric multiplicative simple method through the R package zCompositions version 1.4.0.1.
Statistical analysis
All analyses were conducted in R statistical software version 4.2.1, and plots were generated using the R packages ggplot2 version 3.4.4 and patchwork version 4.2.3. The outcome (dependent variable, y) of this study was binary: patients with cholestasis versus no cholestasis. The microbiome metrics of interest included α-diversity, overall β-diversity, individual abundances of taxa, and KEGG Orthology abundances. This study aimed to determine the associations between these microbiome metrics and the amounts of bile acids (as individual independent variables, x) with the outcome, cholestasis. To accomplish this task, mixed effect binomial regression models were built after variable standardization using R packages lme4 version 1.1.34 and standardize version 0.2.2, with adjustment for postmenstrual age (completed weeks), enteral feeding amount (mL/kg body weight), enteral feeding amount × microbiome metric/bile acid interaction, and patient (as a random effect) via the following formula: y ∼ x + postmenstrual age + enteral feeding amount + enteral feeding amount*x + (1/patient). Model fit was evaluated using the R package DHARMA version 0.4.6, along with calculation of the variance inflation factor through the R package car version 3.1.2 to assess collinearity (maximum < 5 considered not collinear). After fitting, p-values were determined from the Wald chi-square test implemented in the R package car and were subsequently adjusted by the Benjamini-Hochberg method. If the microbiome metric/bile acid p-value was <0.05 and adjustment indicated a false-positive rate of <1%, the microbiome metric/bile acid was next examined for substantial presence among the studied cohort (microbiome metric/bile acid needed to be present in ≥ three patients in at least one of the study groups) before being considered significant. Results and assumption tests of all models (significant or non-significant) are included in the supplementary materials (Supplementary Table 2). The associations between the significant microbiome metrics/bile acids (as the outcome, y) and the number of days of antibiotics or diagnosis of NEC (as individual independent variables, x) were also examined with adjustment for postmenstrual age (completed weeks) and patient (as a random effect) using linear mixed effect models via the same strategy as above: y ∼ x + postmenstrual age + (1/patient). If the number of days of antibiotics and/or diagnosis of NEC were found to be significantly associated with the microbiome metrics/bile acids, the number of days of antibiotics and/or diagnosis of NEC was then added to the above binomial mixed effect models as additional independent variables. Only if the microbiome metrics/bile acids remained significant after this additional adjustment were they considered to be associated with cholestasis. These regression results for evaluation of the number of days of antibiotics and the diagnosis of NEC as potential confounding variables are included in the supplementary materials (Supplementary Table 3).
Results
Cohort definition
Twenty-two infants met inclusion criteria; eleven patients had PNAC, and eleven were classified as non-cholestatic. Cohort clinical characteristics are summarized in Table 1. The cohorts were well-matched in terms of gestational age at birth, with a mean of 25.8 weeks, ranging from 24 to 30 weeks. Mode of delivery, sex, and birth weight were also similar between groups. There were no significant differences in the percentage of human milk feeds nor in the diagnosis of sepsis, which had low prevalence in this cohort (one case). There were significant differences in average total enteral feeds per day, with the non-cholestatic cohort taking more on average than the cholestatic. In turn, the cholestatic group was exposed to significantly more PN and lipid per day over their length of stay compared to the non-cholestatic cohort. The cholestatic group also received significantly more days of antibiotics than the non-cholestatic group, specifically penicillins, gentamicin, vancomycin, and metronidazole. There were significantly more diagnoses of NEC in the cholestatic cohort, as well as significantly higher utilization of SMOFlipid (Fresenius Kabi; Bad Homburg, Germany) and ursodiol, each of which is used to treat PNAC. The cholestatic cohort had a significantly increased length of NICU stay in days, as well as a greater corrected gestational age at discharge in completed weeks.
Table 1Cohort clinical characteristics
| Clinical characteristic | Cholestatic (n = 11) | Non-Cholestatic (n = 11) | p-value |
|---|
| Gestational age at birth, completed weeks | 25.3 ± 1.6 | 26.4 ± 1.7 | 0.11 |
| Mode of delivery, vaginal | 18.2% (2) | 18.2% (2) | 1.0 |
| Sex, male | 54.5% (6) | 63.6% (7) | 0.66 |
| Birth weight, kg | 0.76 ± 0.12 | 0.80 ± 0.25 | 0.63 |
| % Feeds human milk | 26.8 ± 27.8 | 40.9 ± 34.3 | 0.27 |
| Average total enteral feeds per length of NICU stay (mL/kg bodyweight) | 57.0 ± 28.5 | 85.2 ± 23.2 | 0.014 |
| Average PN received per length of NICU stay (mL/kg bodyweight) | 34.8 ± 9.9 | 24.1 ± 10.9 | 0.020 |
| Average lipid received per length of NICU stay (g/kg bodyweight) | 0.81 ± 0.16 | 0.58 ± 0.24 | 0.010 |
| Days of antibiotics | 36.1 ± 21.4 | 16.5 ± 12.6 | 0.010 |
| Days of penicillins | 24.4 ± 20.8 | 6.5 ± 3.1 | 0.00032 |
| Days of gentamicin | 14.9 ± 7.6 | 9.3 ± 5.3 | 0.042 |
| Days of vancomycin | 7.9 ± 5.8 | 3.5 ± 2.9 | 0.023 |
| Days of cephalosporins | 10.8 ± 10.8 | 3.3 ± 6.9 | 0.052 |
| Days of erythromycin | 1.8 ± 6.0 | 6.0 ± 11.3 | 0.25 |
| Days of antifungals | 2.1 ± 4.3 | 1.7 ± 4.1 | 0.83 |
| Days of clindamycin | 0.8 ± 1.8 | 0.3 ± 0.9 | 0.35 |
| Days of metronidazole | 6.6 ± 9.8 | 0.3 ± 0.9 | 0.020 |
| Sepsis | 9.1% (1) | 0.0% (0) | 0.23 |
| NEC | 27.3% (3) | 0.0% (0) | 0.031 |
| Use of SMOFlipid | 27.3% (3) | 0.0% (0) | 0.031 |
| Use of ursodiol | 36.4% (4) | 0.0% (0) | 0.011 |
| Length of NICU stay, days | 142.0 ± 46.0 | 95.5 ± 19.2 | 0.0038 |
| Corrected gestational age at discharge, weeks | 45.1 ± 5.6 | 39.6 ± 2.6 | 0.0054 |
Differences in α diversity but not overall β diversity were observed between cohorts
In order to interrogate the role of the fecal microbiome in PNAC, we first analyzed overall microbiome diversity. Microbiome richness was significantly higher in the non-cholestatic cohort (n = 11 patients, n = 29 samples) compared to the cholestatic cohort (n = 11 patients, n = 34 samples) when controlling for postmenstrual age and amount of enteral feeds (Fig. 1A; p < 2E-16). There were no differences in overall microbiome β diversity between cohorts (Fig. 1B; p = 1.0).
Taxa-specific analysis showed differences between cholestatic and non-cholestatic cohorts
After identifying a reduction in microbiome richness in cholestatic patients, we aimed to determine which taxa were depleted in these patients. There was a noted expansion of the phylum Proteobacteria both in cholestatic infants when accounting for postmenstrual age (Fig. 2A; direct coefficient 0.79; p < 2E-16), with a 23% mean difference from 28–30 weeks postmenstrual age, as well as in cholestatic infants on lower amounts of enteral nutrition (Fig. 2B; interaction coefficient −0.72; p < 2E-16), with a 20% mean difference from 0–50 mL/kg enteral feeds. Bacteroidota displayed a protective pattern: regardless of postmenstrual age, a significantly greater abundance of Bacteroidota was associated with lack of cholestasis (Fig. 2C; direct coefficient −33.53; p < 2E-16), with a 4% overall mean difference. With regard to enteral feed amount, greater Bacteroidota abundance was similarly associated with lack of cholestasis, and this was more pronounced in infants on greater amounts of enteral nutrition (Fig. 2D; interaction coefficient −32.80; p < 2E-16), with a 3% mean difference at >50 mL/kg enteral feeds. There was a notable increase in the incidence of cholestasis in infants with a greater abundance of Fusobacteriota, particularly at low enteral nutrition (Fig. 2E; interaction coefficient −3.56; p < 2E-16), with a 2% mean difference from 0–50 mL/kg enteral feeds. No Fusobacteriota were detected in the non-cholestatic group.
At the family level, significant patterns emerged with Bifidobacteriaceae, Bacteroidaceae, Dialisteraceae, Ruminococcaceae, Akkermansiaceae, and Erysipelatoclostridiaceae (p < 2E-16). At the genus level, Akkermansia were found to be present in high amounts in both non-cholestatic and cholestatic infants on low enteral nutrition; however, as feeds increased, a significantly greater abundance of Akkermansia was associated with the non-cholestatic state (Fig. 2F; interaction coefficient −28.30; p < 2E-16), with a 0.004% mean difference at >50 mL/kg enteral feeds, and no Akkermansia were detected in the cholestatic group.
Overall bile acid content was greater in non-cholestatic infants, with a significant increase in deoxycholic acid (DCA) specifically
Bile acid analysis was subsequently conducted in order to examine any potential differences in fecal bile acid content between cohorts. A smaller number of infants (n = 8 cholestatic, n = 7 non-cholestatic) had samples collected while on PN available for bile acid analysis. Among those analyzed, it was noted that non-cholestatic infants had significantly greater total bile acid content in their stool than cholestatic infants (Fig. 3A; direct coefficient −7.86; p = 0.0073). Cholestatic infants had significantly greater amounts of conjugated cholic acid than chenodeoxycholic acid (Fig. 3B; direct coefficient = 1.56; p = 0.034). A significant pattern also emerged with DCA; although the overall prevalence of DCA was low among both cholestatic and non-cholestatic infants, cholestatic patients had significantly lower levels of DCA than non-cholestatic infants when controlling for postmenstrual age and amount of enteral feeds (Fig. 3C; direct coefficient −171.58; p = 0.035).
Metagenomics analysis revealed functional differences in chemotaxis, carbohydrate metabolism, stress response, and zinc transporters
In order to evaluate specific functional differences between the microbiomes of cholestatic and non-cholestatic infants, metagenomics analysis was completed on the same fifteen samples as bile acid analysis (n = 8 cholestatic, n = 7 non-cholestatic). The Proteobacteria present in preterm infants without cholestasis had a feed-independent increase in chemotaxis-related proteins, amino acid transporters, stress response proteins, and microcin C permease. Further, these Proteobacteria had a feed-dependent increase in arginine, thiamine, glutathione, and sugar and sugar alcohol transporters, in addition to increased capabilities in polysaccharide degradation such as pectinase and monosaccharide utilization such as L-fuculose and D-allose kinases. Together, these findings suggest that the Proteobacteria are under more stress in the gut environment of non-cholestatic compared to cholestatic preterm infants from both nutrient competition and microbiota-produced antimicrobials including microcin C.27 These Proteobacteria are thus required to maintain a diversity of nutrient sensing, searching, uptake, and consumption proteins, in addition to stress response mechanisms. In contrast, preterm infants with cholestasis were found to have Proteobacteria and Streptococcus with a feed-dependent increase in zinc transporters. Full metagenomic analysis results are reported in Supplementary Table 4.
Discussion
PN is both a known hepatotoxin and an essential component of care for premature infants who are unable to tolerate adequate enteral nutrition for growth. Although not every infant on PN develops PNAC, those who do may go on to develop significant liver damage with ongoing PN exposure. This study is the first to compare the gut microbiome and fecal bile acid content in premature infants with PNAC to non-cholestatic peers. We hypothesized that in premature infants on PN and varying amounts of enteral feeds, features of the early gut microbiome, as well as differences in fecal bile acid content, are associated with the development of PNAC. In this cohort of 22 preterm infants exposed to PN for four or more weeks, there is significantly greater microbiome richness in non-cholestatic infants compared to cholestatic; however, no significant differences were observed in overall β-diversity. At the phylum level, cholestatic infants were found to have significantly more Proteobacteria and Fusobacteriota than non-cholestatic peers, a pattern that was more pronounced at low enteral nutrition. Regardless of enteral feed amount, non-cholestatic infants had a significantly greater abundance of Bacteroidota than cholestatic infants. The genus Akkermansia was found to be abundant in both cholestatic and non-cholestatic infants on low enteral feeding volumes; however, as enteral feed amount increased, it was noted that non-cholestatic infants had a significantly increased abundance of Akkermansia compared to cholestatic infants. A sub-analysis of fecal bile acid content in fifteen infants showed overall low concentrations of DCA among all patients, with significantly lower levels in cholestatic patients when controlling for postmenstrual age and enteral feed amount. Metagenomic analysis showed an increase of Proteobacteria requiring an augmented stress response in non-cholestatic infants and an increase in zinc transporter genes in Proteobacteria and Streptococcus in cholestatic infants.
The phylum Proteobacteria, which includes genera such as Escherichia, Klebsiella, and Enterobacter, has been shown by Dahlgren et al. to be expanded in premature infants on PN with cholestasis, but not reaching statistical significance due to cohort size.6 Mokha et al. in 2019 compared the fecal microbiomes of premature infants on PN who did not become cholestatic to those who did. Interestingly, in that small study, Proteobacteria were found to be elevated in the control group, with longitudinal samples from the pre-cholestasis, cholestasis, and post-cholestasis infants showing expansion of the phylum over time.28 Although this result differs from our cohort, we did not have enough representative samples from the pre-cholestasis time point to perform an analysis. Other studies in premature infants have shown a predominance of Proteobacteria in those who would go on to develop NEC, as well as those who display poor tolerance to enteral feeds compared to peers.29,30 These findings, in conjunction with our results, which show higher Proteobacteria in cholestatic premature infants compared to non-cholestatic, particularly at low enteral nutrition, suggest that Proteobacteria can be associated with health outcomes. In adults, Proteobacteria have been associated with relative enteral starvation, which, when coupled with inflammation, contributes to their proliferation.31 The findings for phylum Fusobacteriota were similar in that cholestatic infants had higher concentrations at lower enteral nutrition. Although there is some adult literature to suggest specific species of Fusobacterium contribute to disease states (F. nucleatum is associated with colorectal cancer, inflammatory bowel disease, and appendicitis), this finding in premature infants with cholestasis is novel.32
Bacteroidota are common gastrointestinal tract colonizers, known to be passed on to infants through vaginal birth.33 In the present study, regardless of enteral feed amount, Bacteroidota were found in higher concentrations in non-cholestatic than in cholestatic infants. This echoes the findings of Dahlgren et al., who found lower Bacteroidota in premature infants on PN, as well as Guo et al., who reported lower Bacteroidota in cholestatic term infants.6,7Bacteroidota are involved in bile acid metabolism at multiple levels, including bile salt hydrolase (BSH) activity.10,34 Although direct research into BSH activity in preterm infants is limited, Lynch et al. found BSH activity to be profoundly decreased in preterm infants who developed cholestasis, with correlation to a relative depletion of C. perfringens compared to controls.35 In the present cohort, there were no significant findings relative to C. perfringens nor with signaling for BSH pathways in metagenomics analysis, which may reflect lower availability of fecal material for analysis. Regardless, our findings suggest that premature infants without cholestasis are able to more efficiently metabolize bile acids due to the abundance of Bacteroidota in comparison to their cholestatic peers.
At the genus level, Akkermansia were noted to be more abundant in both cholestatic and non-cholestatic infants on low enteral feeding volumes. Akkermansia is a mucin-degrading genus that has a competitive advantage during states of low enteral nutrition due to its ability to use mucus as its sole carbon and nitrogen source. The presence of Akkermansia has been considered a biomarker of a healthy, protective, and anti-inflammatory mucous layer: low Akkermansia abundance has been correlated with disease states such as Crohn’s disease, ulcerative colitis, and appendicitis in human subjects.36 This is because, while Akkermansia paradoxically degrades mucin, it simultaneously increases both the number of goblet cells and the total mucus secretion by goblet cells through cross-talk with the host intestinal epithelium, with the net result being a thicker mucosal layer and stronger intestinal barrier.37 Although there has been no direct research implicating Akkermansia in cholestatic liver disease, one study utilizing a mouse model for alcoholic liver disease demonstrated that therapeutic A. muciniphila administration ameliorated hepatic injury.38 The mechanism proposed by the authors was that the thicker mucus layer, and thus improved intestinal barrier induced by A. muciniphila, reduced the systemic endotoxin load (i.e., serum lipopolysaccharide concentrations), which ameliorated liver injury through decreased inflammation. In our cohort, as enteral feeding increased, Akkermansia were found in higher abundance in non-cholestatic patients, which is in line with prior research, reflecting its contribution to a protective mucous layer. Studies in both humans and mice also suggest that the presence of DCA is correlated with increased growth of Akkermansia spp. (specifically A. muciniphila).39,40 Our results, which show both high Akkermansia and high DCA in non-cholestatic infants, align with these conclusions and are a novel finding in preterm infants.
Bile acid analysis was performed on a subset of samples and showed overall low bile acid levels but significantly decreased abundance of DCA in cholestatic compared to non-cholestatic infants. DCA is a secondary bile acid produced by 7α-dehydroxylation. Its presence in the stool is reflective of BSH activity, which can be induced by Bacteroidota.10,41 DCA is also a known farnesoid X receptor (FXR) agonist. Disrupted signaling of FXR, which has an important role in overall bile acid homeostasis, has been proposed as a mechanism for PNAC.42,43 Findings in the present study suggest that non-cholestatic infants have a more balanced bile acid profile, which may be attributed to a higher abundance of Bacteroidota in their microbiome.
Metagenomics analysis showed a pattern of Proteobacteria in the microbiome of non-cholestatic infants with increases in stress management, protein misfold correction, and chemotaxis capabilities, as well as amino acid and carbohydrate metabolism. This, along with the noted increase in Proteobacteria in the cholestatic cohort, suggests that competition from a microbiome with more richness in non-cholestatic infants may place selective pressure on the survival of Proteobacteria with a greater ability for adaptation to stressful conditions, in which bacterial mobility and metabolism are emphasized. This may further reflect that non-cholestatic infants possess a more “resilient” microbiome that is better able to prevent uncontrolled proliferation of the phylum.
Preterm infants with cholestasis were noted to have feed-dependent increases of both Proteobacteria and Streptococcus with zinc transporters. Zinc metabolism has been a topic of research in premature infants, as they are at risk for deficiency given that the majority of accretion occurs during the third trimester of pregnancy.44 Abnormalities in zinc metabolism are described in chronic liver disease, given that bile is an excretory route for zinc.45 While one study of extrahepatic biliary atresia patients shows low hepatic zinc, which is in line with other chronic liver diseases, there have also been some reports of increased hepatic zinc content in a group of consanguineous older children with cholestatic liver disease, suggesting a potential genetic mechanism.45,46 Existing literature on zinc and the human microbiome demonstrates an association between zinc deficiency and reduced gut microbiota biodiversity.47 Our study is the first to demonstrate a relationship between zinc metabolism and the microbiome in preterm infants with cholestasis, which suggests a potential selective pressure on the gut microbiome either to promote increased need for zinc absorption in a cholestatic state or as a reflection of a selective advantage for microbes that are able to metabolize zinc given the presence of excess.
This study is the first of its kind to directly explore the relationship between the susceptibility of certain preterm infants to PNAC and the fecal microbiome, as well as fecal bile acid content. Limitations include the small cohort size, with twenty-two infants meeting criteria for study inclusion. Patients in the study ranged in gestational age from 24 to 30 weeks. Although the analysis controlled for gestational age, the authors additionally note that patients may have had inherent differences in baseline liver maturity and fecal microbiome content based on gestational age alone.
Analysis was also limited by the amount of fecal matter available for analysis, particularly for bile acid sampling, which is a common problem in research from preterm infants on PN who are prone to infrequent defecation due to low enteral feed volumes and ongoing critical illness. Although the results reached statistical significance, group differences for relatively rare bile acids in this patient population should be interpreted with caution. In particular, given the limited data on extremely preterm and term infant fecal bile acid content, validation with larger cohorts is recommended. Additionally, the presence of bile acids and/or microbes in the stool may not be reflective of metabolic utility, as they are eliminated in the stool. Regardless, the analysis of our small cohort showing statistically significant results demonstrates the large effect size of these specific associations. Additionally, replicable findings with different populations are promising and suggest a more global effect. Nonetheless, future studies in this important patient population will benefit from enrollment of larger and more diverse cohorts.
Although our patients were well-matched in terms of gestational age at birth, mode of delivery, sex, birth weight, and diagnosis of sepsis, infants with cholestasis had higher rates of antibiotic exposure and more diagnoses of NEC than non-cholestatic peers. NEC can contribute to cholestasis directly by impacting portal blood supply as well as through bacterial translocation into portal blood.5 Antibiotic administration has also been implicated in cholestasis through alterations in bile acid metabolism via disruption in gut microbiota, although these associations are stronger for enteral antibiotics than parenteral.48 Of the antibiotics with statistically significant differences—penicillins, gentamicin, vancomycin, and metronidazole—there is no significant literature published to date attributing their parenteral administration to neonatal cholestasis. Regardless, the significant differences in antibiotic exposure between groups may limit the external applicability of these findings.