Graphical abstract

   graphic file with name fx1.jpg
   [37]Open in a new tab

Highlights

     * •
       Metabolic differentiation is already initiated in the first week of
       life
     * •
       Early metabolic alterations involve neuroactive and neurotoxic
       compounds
     * •
       Several recently discovered bile acid amino acid conjugates are
       detected
     * •
       Changed abundance of heme metabolites, bile acids, and steroid
       sulfates is observed
     __________________________________________________________________

   Pristner et al. use untargeted and targeted metabolomics to investigate
   the plasma and fecal metabolome of extremely premature infants with
   brain injury and the correlation with gut microbiome and immunological
   data. The results show an early differentiation in neuroactive
   metabolites and bile acids in infants diagnosed with brain injury later
   in life.

Introduction

   Extremely premature infants are born before 28 weeks of gestation and
   represent a highly vulnerable patient cohort. While the overall
   survival rate of premature infants has improved in the last decades, a
   significant number of infants still suffer from severe morbidities and
   life-long neurodevelopmental impairment.[38]^1 Extremely premature
   infants are born during a critical window of human fetal development,
   leaving them prone to early physical, microbial, and environmental
   exposures. Within this third-trimester window, neurophysiological
   maturation and development appears to be highly influenced by postnatal
   factors, especially by the developing gut microbiome. We recently
   described gut-immune-brain interactions in a clinical cohort of 60
   extremely premature infants and found that Klebsiella overgrowth is
   associated with pro-inflammatory immunological imprinting and neonatal
   brain injury.[39]^2 Therefore, early gut-immune-mediated and
   host-microbe signaling might have implications for early
   development.[40]^3

   The gut-immune-brain axis represents a complex bidirectional
   communication system composed of various neural pathways with
   neuropeptides, cytokines, and hormones as signaling molecules. An
   important communication pathway between gut microbiota and the central
   nervous system (CNS) is the enteric nervous system.[41]^4 Recent
   advances in understanding the gut-immune-brain axis have shown that
   microbes play a significant role in signal transduction, which, in
   return, can modulate the physiology of the host.[42]^5^,[43]^6 With
   host-affecting bacterial metabolites such as bile acid metabolites,
   short-chain fatty acids, amino acid metabolites, serotonin, and
   γ-aminobutyric acid (GABA),[44]^7 gut bacteria have been associated
   with functions and diseases of the CNS.[45]^8^,[46]^9 The synthesis of
   phenylalanine, tryptophan (TRP), and tyrosine via the shikimate pathway
   and subsequent catabolism results in formation of neurotransmitters
   L-3,4-dihydroxyphenylalanine, dopamine, epinephrine, and
   norepinephrine.[47]^8 TRP-derived indole and its derivatives
   indoleacetic acid, indolepropionic acid, and tryptamine induce
   neuroactivity and potentially limit TRP availability for the
   neurotransmitter synthesis.[48]^10

   Recently, it has been shown that the communication between the bile
   system and gut microbiota is bidirectional in nature, potentially
   involving neural modulation. In humans, de novo synthesized primary
   bile acids are cholic acid (CA) and chenodeoxycholic acid (CDCA). Upon
   performing their role in lipid digestion,[49]^11 most bile acids are
   recycled back to the liver via the enterohepatic cycle. Various members
   of the gut microbiota express bile salt hydrolase enzymes that
   deconjugate bile acids from taurine and glycine, which reduces the
   efficiency of their reabsorption in the enterohepatic
   cycle.[50]^12^,[51]^13 Deconjugation and dehydroxylation, oxidation,
   and epimerization of bile acids by gut microbiota are the main
   mechanisms for the formation of secondary bile acids such as
   deoxycholic acid (DCA), lithocholic acid (LCA), hyodeoxycholic acid
   (HDCA), and ursodeoxycholic acid (UDCA).[52]^14 These acids can be
   absorbed into the colon by passive diffusion and transported to the
   liver, where a repeated conjugation with glycine and taurine can occur,
   followed by secretion into the intestinal tract.[53]^15 A particularly
   important role of bile acids is their relation to inflammatory
   responses. While hydrophobic bile acids may act as pro-inflammatory
   signals, hydrophilic bile acids were shown to have the opposite effect.
   Moreover, deficits in luminal levels of bile acids have been correlated
   with gut bacterial overgrowth, intestinal inflammation, and tissue
   damage.[54]^16^,[55]^17 Recent evidence suggested that bile acids might
   directly affect the brain.[56]^18^,[57]^19

   The investigation of the diverse group of chemical compounds affecting
   the brain-gut axis poses a methodological challenge and requires a
   comprehensive analytical approach. High-resolution mass spectrometry
   (HRMS)-based untargeted metabolomics is an emerging and powerful
   approach that enables the global analysis of endogenous and exogenous
   metabolites, thereby revealing information about progression, causes,
   and potential biomarkers of diseases.[58]^20^,[59]^21 Metabolomics
   investigations of extremely premature infants have thus far been
   limited to only a few publications investigating only a single sample
   matrix or using a less comprehensive
   approach.[60]^22^,[61]^23^,[62]^24^,[63]^25

   Based on the previously found association of the overgrowth of
   Klebsiella with pro-inflammatory immunological imprinting and neonatal
   brain injury,[64]^2 an investigation of the blood and fecal metabolome
   was performed as the next step for generating insights into the
   possible molecular mechanisms underlying brain injury and for
   deciphering potential links to the gut microbiome. As a conclusive
   diagnosis of brain injury by magnetic resonance imaging (MRI) can only
   be done at a later developmental stage (term-equivalent age), there is
   also a need for early metabolic indicators for the later occurrence of
   brain injury. A deeper knowledge about underlying mechanisms and early
   metabolic predictors might pave the way for the development of an early
   intervention strategy to stop the formation of brain injury or at least
   reduce the degree of neurological damage. Here, we investigated the
   fecal and plasma metabolome in samples from a cohort[65]^2 of extremely
   premature infants collected at several time points by untargeted liquid
   chromatography (LC)-HRMS and targeted LC-tandem mass spectrometry
   (LC-MS/MS) based metabolomics. The resulting data were integrated with
   16S rRNA gene amplicon gut microbiome profiles as well as patient
   cytokine, growth factor, and T cell profiles.

Results

   From a cohort of 51 extremely premature infants hospitalized in the
   same hospital, 120 plasma samples at five time points (day 3, day 7,
   day 28, correction week 32, and term-equivalent age) and 75 fecal
   samples at three time points (day 7, day 28, and term-equivalent age)
   were analyzed. The cohort consisted of two patient groups, one with
   age-adequate findings or mild brain injuries diagnosed by cranial MRI
   at term-equivalent age, named the control group (CTR, n = 36), and one
   group with pathological and severe brain injuries, named the
   pathological group (PAT, n = 15).

   A visual summary of the patient cohort, sample composition, and
   experiments is illustrated in [66]Figure 1A. Additional information
   about the patient cohort, sampling procedures, and details of previous
   T cell, cytokine, growth factor, and 16S rRNA gene amplicon analysis,
   and determination of brain damage, is described in detail in the
   [67]supplemental information and our previous publication.[68]^2
   [69]Table 1 presents an overview of the subcohort used for this study,
   including the statistical significance of associations between
   variables and experimental groups. Retinopathy of prematurity (ROP) was
   observed in all infants of the pathological group and exhibited a
   significant relationship with the experimental groups ([70]Table 1). A
   significant association between mode of delivery, with fewer C-sections
   in the pathological group, was also observed but was not statistically
   significant after multiple testing correction in the subgroup of
   infants. An in-depth discussion of these findings can be found in our
   previous publication.[71]^2

Figure 1.

   [72]Figure 1
   [73]Open in a new tab

   Study design and metabolomic differences over time

   (A) Visualization of the study design.

   (B) PCA of untargeted metabolomics dataset and respective distances to
   the centroid for (B) feces and (C) plasma.

Table 1.

   Overview of the cohort of extremely premature infants
   Subject cohort demographics
     __________________________________________________________________

   Variable CTR PAT Fisher’s exact test/Student’s t test p value
   Bonferroni corrected p value
   No. of infants 36 15 – –
   Birth date (quarter) – – 0.336 1
   Birth date (month) – – 0.294 1
   Sex assigned at birth (% female) 56 47 0.759 1
   BPD (%) 25 47 0.187 1
   ROP (%) 42 100 <0.001 0.002
   PDA (%) 31 40 0.532 1
   NEC (%) 8 0 0.545 1
   Maternal antibiotics (%) 50 60 0.536 1
   C-section (%) 86 46 0.011 0.264
   Connatal infection (%) 36 53 0.734 1
   Antibiotic intervention (%) 72 87 0.470
   pPROM (%) 56 47 0.75 1
   Gestational age (weeks) 25.6 ± 1.2 24.9 ± 1.4 0.137 1
   Birth weight (g) 774.4 ± 138.6 740.3 ± 168.8 0.497 1
   Birth length (cm) 33.1 ± 2.2 32.8 ± 2.3 0.635 1
   Birth head circumference (cm) 23.2 ± 1.5 23 ± 1.9 0.621 1
   Weight slope 0.016 ± 0.001 0.015 ± 0.002 0.225 1
   Apgar score at 1 min 7.3 ± 0.8 7.1 ± 1.5 0.761 1
   Apgar score at 5 min 8.5 ± 0.6 8.3 ± 1.3 0.436 1
   Apgar score at 10 min 8.9 ± 0.4 8.9 ± 0.4 0.868 1
   Hospitalization time (days) 88.2 ± 23.3 96.6 ± 29.5 0.605 1
   Antibiotic intervention (days) 8.7 ± 5.9 9.9 ± 8.7 0.662 1
   Age at discharge (weeks) 38.2 ± 3.1 38.9 ± 3.2 0.780 1
   Discharge head circumference (cm) 32.7 ± 2.5 33.1 ± 2.8 0.979 1
   Discharge length (cm) 46.2 ± 3.9 46.3 ± 3.7 0.716 1
   Discharge weight (kg) 2.9 ± 0.8 2.8 ± 0.5 0.420 1
   O[2 max] at birth (%) 66.7 ± 21.1 57.3 ± 22.4 0.180 1
   [74]Open in a new tab

   CTR, control group; PAT, pathological group; pPROM, preterm premature
   rupture of membranes; BPD, bronchopulmonary dysplasia; ROP, retinopathy
   of prematurity; PDA, persistent ductus arteriosus; NEC, necrotizing
   enterocolitis. Data are presented as mean ± SD or percentage.

Development of the plasma and fecal metabolome of extremely premature infants
over time

   For characterization of the gut and plasma metabolome, each sample was
   analyzed by LC-HRMS in an untargeted manner using two complementary
   separation methods, reversed-phase and hydrophilic interaction
   chromatography, in positive and negative polarity mode to ensure the
   coverage of a broad spectrum of compounds and chemical classes. For a
   general overview of the differences between the groups over time, a
   principal component analysis (PCA) was performed on the combined data
   of the untargeted LC-HRMS experiments.

   This was followed by a permutational multivariate analysis of variance
   (PERMANOVA), which found that time point was a significant factor
   (plasma, p < 0.001; feces, p < 0.001; variation explained by
   experimental group: plasma = 21%, feces = 19%), while disease diagnosis
   was not significant. The experimental group explained in plasma <1% and
   in feces <1% of PERMANOVA variation, while patient ID explained 30% in
   plasma and 55% in feces. In addition, the dispersion of samples at the
   different sampling time points in plasma (average Euclidean distance to
   centroid: day 3 = 157.0, day 7 = 164.5, day 28 = 151.8, week 32 =
   133.6, term = 122.6) ([75]Figure 1C) and feces (average Euclidean
   distance to centroid: day 7 = 160.1, day 28 = 132.9, term = 146.1)
   ([76]Figure 1B) was analyzed. The difference in dispersion based on
   time point was statistically significant by ANOVA in both plasma and
   feces (plasma, p < 0.001; feces, p = 0.017).

Metabolic changes from a pathway perspective

   To put the untargeted LC-HRMS experiments into biological context, we
   used a combination of mummichog,[77]^26 a tool for pathway enrichment
   based on putative compound annotation, and a gene set enrichment
   analysis-related approach ([78]Figures 2A–2C). In [79]Figure 2 the
   altered metabolic pathways in plasma ([80]Figure 2A), feces
   ([81]Figure 2B), and the intersection between plasma and feces
   ([82]Figure 2C) can be seen. Pathway enrichment analysis is restricted
   to the predefined human genome-scale metabolic model, and its
   application for the analysis of microbial pathways is limited.
   Therefore, we also created activity networks for the fecal samples
   using the Python package of mummichog ([83]Figures 2D–2F).

Figure 2.

   [84]Figure 2
   [85]Open in a new tab

   Observed metabolic changes at the pathway level

   (A–C) Enriched metabolic pathways and their combined p values for
   (A) plasma, (B) feces, and (C) their intersection.

   (D) Activity networks of feces created by mummichog.

Plasma

   The metabolism of several amino acids was significantly modulated in
   plasma at day 3, day 7, and day 28, including the metabolism of
   glutamate, an important neurotransmitter, at day 3 and day 28 (day 3:
   p = 0.003; day 28: p = 0.007). Also, pathways involved in carbohydrate
   metabolism were significantly altered at day 7 and to a lesser extent
   at day 28.

Feces

   At all three time points, a significant or near-significant enrichment
   was found in biopterin metabolism (day 7: p = 0.004; day 28: p = 0.055;
   term: p = 0.010). Various pathways involved in carbohydrate metabolism
   were significantly modulated both at day 7 and day 28 in the feces. The
   pathway of the urea cycle was also significantly affected at both day
   28 and term-equivalent age (day 28: p = 0.023; term: p = 0.001) in
   feces. Tyrosine metabolism, connected to the synthesis of the
   neurotransmitter dopamine, was significantly altered at day 28 and
   term-equivalent age in feces (day 28: p = 0.034; term: p = 0.005).
   Biopterin metabolism was affected significantly in the feces at each
   sampling time point. In the activity networks of the feces
   ([86]Figure 2D), amino acids together with pyruvate and their related
   metabolites were the most prominent and well-connected metabolites at
   all three time points.

Metabolic profiles in plasma are already differentiated 3 days after delivery

   Plasma samples were investigated by untargeted metabolomics followed by
   evaluation in a purely untargeted fashion combined with a targeted
   suspect screening for compounds of interest. A positive fold change
   indicates an increase in the pathological group ([87]Figure 3).

Figure 3.

   [88]Figure 3
   [89]Open in a new tab

   Selected metabolites altered between groups at different time points in
   plasma and feces

   (A–G) Box plots of selected metabolites. Plasma: day 3: n[CTR] = 14,
   n[PAT] = 10; day 7: n[CTR] = 16, n[PAT] = 11; day 28: n[CTR] = 14,
   n[PAT] = 7; week 32: n[CTR] = 15, n[PAT] = 8; term age: n[CTR] = 16,
   n[PAT] = 9. Feces: day 7: n[CTR] = 4, n[PAT] = 8; day 28: n[CTR] = 31,
   n[PAT] = 11. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

   (H) Selected overview of differentiated metabolites and pathways
   between groups over all time points. Arrows pointing up indicate an
   increase in the pathological group.

Day 3

   The untargeted evaluation of the metabolomics data revealed 112
   significantly altered features (identification confidence level 2
   [ICL-2]: 1; ICL-3: 17; ICL-4: 19; unidentified: 75) with a log[2] fold
   change (FC) greater than |2|. The metabolites with the highest
   discriminatory power between the experimental groups and highest
   annotation confidence consisted of the bile acid conjugate
   tauro-1-hydroxycholic acid (ICL = 3, score = 0.88, p = 0.006,
   FC-log[2] = 3.1), the steroid sulfate
   11,20-dioxo-5b-pregnane-3-α,17,21-triol-3-sulfate (ICL = 3, score =
   0.79, p = 0.026, FC-log[2] = −2.2), hydroxyphenyllactic acid (ICL = 3,
   score = 0.99, p = 0.049, FC-log[2] = 3.6), and 4-hydroxycinnamic acid
   (ICL = 3, score = 0.99, p = 0.024, FC-log[2] = 4.1) and its glycoside
   1-O-p-coumaroyl-β-D-glucose (ICL = 3, score = 0.79, p = 0.003,
   FC-log[2] = 3.1). During the suspect screening, 26 metabolites (ICL-1:
   15; ICL-3: 11) were significantly changed, including, among others, the
   amino acids alanine (ICL = 1, p = 0.001, FC-log[2] = 0.6), glutamine
   (ICL = 1, p = 0.002, FC-log[2] = 0.5), asparagine (ICL = 1, p = 0.005,
   FC-log[2] = 0.9), serine (ICL = 1, p = 0.007, FC-log[2] = 0.7),
   homoserine (ICL = 1, p = 0.042, FC-log[2] = 0.6), citrulline (ICL = 1,
   p = 0.042, FC-log[2] = 0.8), and sarcosine (ICL = 1, p = 0.048,
   FC-log[2] = 1.3), glyoxylic acid (ICL = 1, p = 0.048, FC-log[2] = 0.7),
   the TRP metabolite tryptamine (ICL = 1, p = 0.012, FC-log[2] = 2.7),
   the metabolites of galactose metabolism galactitol (ICL = 1, p = 0.006,
   FC-log[2] = 1.4) and galactonic acid (ICL = 3, score = 0.99, p = 0.007,
   FC-log[2] = −1.1), and the sulfated compounds 19-nortestosterone
   sulfate (ICL = 3, score = 0.75, p = 0.002, FC-log[2] = −1.2), dopamine
   3-O-sulfate (ICL = 3, score = 0.78, p = 0.002, FC-log[2] = −1.3),
   3-β,16α-dihydroxyandrostenone sulfate (ICL = 3, score = 0.85, p =
   0.026, FC-log[2] = −1.1), and 3,5-tetrahydroaldosterone sulfate (ICL =
   3, score = 0.76, p = 0.007, FC-log[2] = −1.0).

Day 7

   Overall, 12 significantly altered features (ICL-4: 2; unidentified: 10)
   with a log[2] FC greater than |2| were observed during the untargeted
   evaluation, and four metabolites (ICL-1: 1; ICL-2: 1; ICL-3: 2) were
   significantly changed in the scope of the targeted evaluation. The
   significant metabolites consisted of lanthionine (ICL = 3, score =
   0.82, p = 0.039, FC-log[2] = −1.1) and the TRP metabolites tryptamine
   (ICL = 1, p = 0.048, FC-log[2] = −1.4), indoleacrylic acid (ICL = 3,
   score = 0.83, p = 0.044, FC-log[2] = 1.1), and indolelactic acid (ICL =
   3, score = 0.87, p = 0.044, FC-log[2] = 1.0).

Day 28

   The significantly changed features with a log[2] FC greater than
   |2| consisted of 39 features (ICL-2: 4; ICL-3: 8; ICL-4: 6;
   unidentified: 21) for the untargeted evaluation. The metabolite group
   with the highest distinction between the experimental groups and
   highest annotation confidence was N-alkylated ethanolamines, including
   N-decyl-ethanolamine (ICL = 3, score = 0.79, p < 0.001, FC-log[2] =
   2.5), N-caprylyl-diethanolamine (ICL = 3, score = 0.81, p = 0.001,
   FC-log[2] = 5.2), N-dodecanoyl-ethanolamine (ICL = 3, score = 0.75, p =
   0.005, FC-log[2] = 4.8), N-decanoyl-ethanolamine (ICL = 3, score =
   0.76, p = 0.009, FC-log[2] = 3.0), and N-octyl-ethanolamine (ICL = 3,
   score = 0.86, p = 0.002, FC-log[2] = 3.1). During the targeted
   evaluation, only N-acetylcitrulline (ICL = 3, score = 0.75, p = 0.024,
   FC-log[2] = −1.6) was regarded as significant. In addition, elevated
   levels of p-cresol sulfate were observed in the pathological group
   (ICL = 1, p = 0.067, FC-log[2] = 1.1), although not deemed
   statistically significant based on p value.

Week 32

   The untargeted evaluation of the metabolomics data for plasma at week
   32 revealed 72 significantly altered features (ICL-2: 1; ICL-3: 9;
   ICL-4: 10; unidentified: 52) with a log[2] FC greater than |2|. The
   metabolites with the highest discriminatory power between the
   experimental groups and highest annotation confidence consisted of the
   bile acids glycochenodeoxycholyl-7-sulfate (ICL = 3, score = 0.79, p =
   0.007, FC-log[2] = −8.0), 7-oxoglycodeoxycholate (ICL = 3, score =
   0.83, p = 0.004, FC-log[2] = −2.9), and 7-oxotaurodeoxycholate (ICL =
   3, score = 0.93, p = 0.004, FC-log[2] = −2.5). For the suspect
   screening, six metabolites (ICL-1: 3; ICL-3: 3) were significantly
   altered between groups, consisting of spermidine (ICL = 1, p = 0.047,
   FC-log[2] = 1.7), kynurenic acid (ICL = 1, p = 0.032, FC-log[2] =
   −1.0), isocitrate (ICL = 1, p = 0.013, FC-log[2] = 0.7), and the
   sulfated compounds dopamine 3-O-sulfate (ICL = 3, score = 0.78, p =
   0.013, FC-log[2] = −1.8), estriol 3-sulfate (ICL = 3, score = 0.80, p =
   0.046, FC-log[2] = −0.9), and cortisol 21-sulfate (ICL = 3, score =
   0.96, p = 0.021, FC-log[2] = −0.6).

Term-equivalent age

   At term-equivalent age, 170 features (ICL-3: 22; ICL-4: 29;
   unidentified: 119) with a log[2] FC greater than |2| were significantly
   altered during the untargeted evaluation. The most significantly
   altered metabolites with high annotation confidence were
   3-carboxy-4-methyl-5-propyl-2-furanpropionic acid (ICL = 3, score =
   0.99, p = 0.008, FC-log[2] = −4.6) and 1-methyluric acid (ICL = 3,
   score = 0.99, p = 0.034, FC-log[2] = −2.4). In the scope of the suspect
   screening, 15 metabolites (ICL-1: 7; ICL-3: 8) were significantly
   changed. The most significant altered metabolites were mevalonic acid
   (ICL = 1, p = 0.032, FC-log[2] = −1.5), indole (ICL = 1, p = 0.022,
   FC-log[2] = 1.2), the fatty acids undecanedioic acid (ICL = 3, score =
   0.99, p = 0.001, FC-log[2] = −1.1), 1,11-undecanedicarboxylic acid
   (ICL = 3, score = 0.99, p = 0.005, FC-log[2] = −1.2), and α-aminoadipic
   acid (ICL = 1, p = 0.027, FC-log[2] = −1.0), the steroid sulfates
   cortisol 21-sulfate (ICL = 3, score = 0.96, p = 0.007, FC-log[2] =
   −1.1) and estriol 3-sulfate (ICL = 3, score = 0.80, p = 0.023,
   FC-log[2] = −1.5), and several metabolites of the heme degradation
   pathway consisting of biliverdin IXβ (ICL = 1, p = 0.027, FC-log[2] =
   −1.4), bilirubin IXα (ICL = 2.a, score = 0.94, p = 0.038, FC-log[2] =
   −1.6), and 5-oxo-δ-bilirubin (ICL = 3, score = 0.89, p = 0.013,
   FC-log[2] = −1.4). Furthermore, two metabolites of the heme degradation
   pathway, biliverdin IXα (ICL = 2.a, score = 0.92, p = 0.060,
   FC-log[2] = −1.0) and bilirubin IXβ (ICL = 1, p = 0.052, FC-log[2] =
   −1.5), were decreased in the pathological group, although not deemed
   significant based on their p value.

Neuroactive metabolites are altered in feces of preterm infants with brain
injury

   The fecal samples were investigated by untargeted metabolomics followed
   by an evaluation in a purely untargeted fashion combined with a
   targeted suspect screening for compounds of interest. A positive fold
   change indicates an increase in the pathological group ([90]Figure 3).

Day 7

   The untargeted evaluation of the metabolomics data revealed 520
   significantly altered features (ICL-1: 2; ICL-2: 6; ICL-3: 91; ICL-4:
   145; unidentified: 275) with a log[2] FC greater than |2|. The most
   distinct metabolites between experimental groups with high annotation
   confidence were the bile acids α-phocaecholic acid (ICL = 3, score =
   0.88, p = 0.004, FC-log[2] = 4.0), 1,3,7,12-tetrahydroxycholan-24-oic
   acid (ICL = 3, score = 0.90, p = 0.016, FC-log[2] = 2.7), and
   3,12-dioxochol-4-en-24-oic acid (ICL = 3, score = 0.75, p = 0.014,
   FC-log[2] = 2.4), the neurosteroid epipregnanolone sulfate (ICL = 3,
   score = 0.99, p = 0.004, FC-log[2] = 2.7), the tryptophan metabolite
   indolelactic acid (ICL = 2.a, score = 0.87, p = 0.048, FC-log[2] =
   4.9), 2-hydroxy-3-methylbutyric acid (ICL = 3, score = 0.92, p = 0.016,
   FC-log[2] = 2.3), 2-hydroxy-3-(sulfooxy)benzoic acid (ICL = 3, score =
   0.75, p = 0.017, FC-log[2] = 6.8), glucuronic acid (ICL = 3, score =
   0.95, p = 0.004, FC-log[2] = 3.0), and the dipeptides valylproline
   (ICL = 3, score = 0.99, p = 0.008, FC-log[2] = −2.3) and valylvaline
   (ICL = 3, score = 0.83, p = 0.016, FC-log[2] = −2.4), whereas
   epipregnanolone sulfate and indolelactic acid were also part of the
   suspect screening. In the scope of the suspect screening, five
   metabolites (ICL-1: 2; ICL-2: 1; ICL-3: 2) were significantly altered
   between groups, including uracil (ICL = 1, p = 0.041, FC-log[2] = 3.4),
   methionine sulfone (ICL = 1, p = 0.004, FC-log[2] = −4.6), and the
   steroid sulfate corticosterone sulfate (ICL = 3, score = 0.86,
   FC-log[2] = 1.4, p = 0.048). Another tryptophan metabolite, indoxyl
   sulfate (ICL = 3, score = 0.83), was found on day 7 in the majority of
   samples from the pathological group but was not observed in the control
   group. In addition, the levels of the neurotransmitter histamine (ICL =
   3, score = 0.96, p = 0.368, FC-log[2] = 6.9), the neuroactive
   tryptophan metabolite tryptamine (ICL = 1, p = 0.269, FC-log[2] = 3.3),
   and the steroid sulfates 16α-hydroxydehydroepiandrosterone 3-sulfate
   (ICL = 3, score = 0.76, p = 0.073, FC-log[2] = 1.2) and pregnenolone
   sulfate (ICL = 3, score = 0.81, p = 0.109, FC-log[2] = 1.8) were all
   elevated in the pathological group, although not significant based on
   their p values.

Day 28

   Overall, 51 significantly altered features (ICL-2: 1; ICL-3: 15; ICL-4:
   7; unidentified: 28) with a log[2] FC greater than |2| were observed
   during the untargeted evaluation. The most significantly altered
   metabolites with high annotation confidence were the bile acid
   7β-hydroxy-3,12-dioxo-5β-cholanic acid (ICL = 3, score = 0.93, p =
   0.016, FC-log[2] = −2.7), the tryptophan metabolite 5-hydroxyindole
   (ICL = 3, score = 0.94, p = 0.005, FC-log[2] = −2.7), the dopamine
   metabolite 5-hydroxyindoleacetic acid (ICL = 3, score = 0.90, p =
   0.035, FC-log[2] = −2.5), glucuronolactone (ICL = 3, score = 0.97, p =
   0.001, FC-log[2] = −5.7), and the dipeptide prolylhydroxyproline (ICL =
   3, score = 0.86, p = 0.020, FC-log[2] = −3.3). During the suspect
   screening, only dopamine (ICL = 1, p = 0.042, FC-log[2] = −0.8) and
   adipic acid (ICL = 3, score = 0.76, p = 0.004, FC-log[2] = 1.7) were
   deemed significantly changed.

Term-equivalent age

   Overall, nine features (ICL-3: 6; ICL-4: 2; unidentified: 1) with a
   log[2] FC greater than |2| were significantly altered during untargeted
   evaluation, and no significantly changed metabolites were found during
   the suspect screening. The most distinct metabolites between
   experimental groups were nicotinic acid (ICL = 3, score = 0.96, p =
   0.010, FC-log[2] = −2.1) and valyl-prolyl-valine (ICL = 3, score =
   0.79, p = 0.003, FC-log[2] = 2.2).

Correlations between gut microbiome and neuroactive compounds observed in
feces

   The results of the 16S rRNA gene amplicon sequencing of the fecal
   samples, and the cytokine, growth factor, and T cell profiles obtained
   from this cohort of extremely premature infants in our previous
   study[91]^2 were correlated with the data from the targeted and
   untargeted metabolomics experiments ([92]Figure 4) over all sampling
   time points and samples available for this study.

Figure 4.

   [93]Figure 4
   [94]Open in a new tab

   Multi-omics correlation matrix

   Correlation matrix of 16S rRNA microbiome, T cell, cytokine, and growth
   factor data with metabolomics data over all time points of (A) plasma
   (T cell: n = 27; cytokines and growth factors: n = 23; 16S rRNA: n =
   78) and (B) feces (T cell: n = 34; cytokines and growth factors: n =
   27; 16S rRNA: n = 71). ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

Plasma

   In plasma ([95]Figure 4A), tryptamine was significantly positively
   correlated with pro-inflammatory factors interleukin-22 (IL-22), tumor
   necrosis factor α (TNF-α), and CXCL12, which play a role in
   neurogenesis and neuroinflammation.[96]^27 Other TRP metabolites,
   namely indolepropionic acid (ICL = 1), indoleacrylic acid, and
   indolelactic acid, were negatively correlated with Enterococcus and
   Streptococcus and positively with the growth factors brain-derived
   neurotrophic factor (BDNF) and platelet-derived growth factor BB
   (PDGF-BB) as well as CXCL3, a chemokine associated with neuroprotective
   and anti-inflammatory activities,[97]^28 while for indoleacetic acid
   and indol-3-carbaldehyde (ICL = 1) the exact opposite trend was
   visible. Bilirubin, biliverdin, oxo-δ-bilirubin, and the compounds
   p-cresol sulfate, kynurenic acid, and galactonic acid were all
   negatively correlated with the pro-inflammatory factors γδ-T cells,
   Vδ2^+ γδ-T cells, TNF-α, IL-6, IL-1β, and IL-8, while at the same time
   they were positively correlated with the anti-inflammatory factors
   regulatory T cells and CXCL3. Enterococcus, Streptococcus, and γδ-T
   cells were positively correlated with 3-hydroxyoctanoic acid (ICL = 3,
   score = 0.78) as were the growth factors BDNF and PDGF-BB. For the
   compounds phenylalanine, tyrosine, serine, adenosine, galactitol,
   methionine sulfone, and 4-hydroxybenzoic acid (ICL = 3, score = 0.79),
   a negative correlation with naive CD4^+ T cells and effector memory
   CD4^+ T cells was observable. Klebsiella exhibited negative
   correlations with galactitol and GABA. Further correlations are
   visualized in [98]Figure 4A.

Feces

   We observed significant negative correlations between Klebsiella and
   hydroxyproline (ICL = 1), phenylalanine (ICL = 1), and 5-hydroxyindole
   ([99]Figure 4B). Significant negative correlations were also observed
   with the neurotransmitter GABA (ICL = 1) and N-acetyl-L-aspartic acid
   (ICL = 1), the precursor of the neuronal dipeptide
   N-acetylaspartylglutamate. Staphylococcus exhibited strong positive
   correlations with trimethylaminoxide (ICL = 1), glutamine, asparagine,
   and the sialic acid N-acetylneuraminic acid (ICL = 3, score = 0.84),
   but was negatively correlated with the neuroactive TRP metabolite
   indole. Tryptamine, also a neuroactive TRP metabolite, was positively
   correlated with Enterococcus, as were carnitine (ICL = 1) and
   N-acetyl-L-aspartic acid (ICL = 1). There was a strong negative
   correlation between Enterococcus and the abundance of the amino acid
   tyrosine (ICL = 1) in the feces. Bifidobacterium showed significant
   positive correlations with the TRP metabolites indolelactic acid and
   indoleacetic acid (ICL = 1), GABA, and the nucleobases cytosine (ICL =
   1), uracil, and adenine (ICL = 1). Negative correlations were found for
   the taurine conjugate CDCA (ICL = 1). The strongest correlations
   between metabolites and the 16S rRNA gene amplicon data were observed
   in Escherichia-Shigella, consisting of strong negative correlations
   with the amino acids asparagine, arginine, glutamate, and glutamine.
   Escherichia-Shigella also exhibited significant positive correlations
   with GABA and the TRP metabolites indole-3-carbaldehyde (ICL = 1) and
   indole. The sialic acids and intestinal mucus components
   deamino-neuraminic acid (ICL = 3, score = 0.75), N-acetylneuraminic
   acid, and N-acetylgalactosamine 6-sulfate (ICL = 3, score = 0.76) were
   all significantly negatively correlated with IL-17a and
   Escherichia-Shigella while being positively correlated with IL-22.
   Further correlations are depicted in [100]Figure 4B.

Bile acid profiles of plasma and feces in extremely premature infants

   We batch-synthesized more than 120 reference standards including the
   majority of thus far described bacterial-derived conjugates of bile
   acids and amino acids,[101]^29 consisting of the primary bile acids CA
   and CDCA, the secondary bile acids UDCA, DCA, HDCA, and LCA, 19 amino
   acids, and taurine. A targeted LC-MS/MS method for the free bile acids
   and their respective conjugates was developed, allowing the
   quantification of >120 analytes while ensuring the chromatographic
   separation of the isomeric bile acids CDCA, HDCA, DCA, and UDCA and
   their respective conjugates.

Plasma

   Six of the recently described bile acid conjugates were detected
   ([102]Figure 5D) in the plasma samples, interestingly including
   ornithine-LCA, which was not detected a single time in its free form or
   conjugated form in the fecal samples. Furthermore, taurine conjugates
   of the secondary bile acids HDCA and DCA were already observed in most
   samples at day 3. In addition, lower levels of taurine-CA (Tau-CA) (p =
   0.036, FC-log[2] = −1.5) and glycine-CA (Gly-CA) (p = 0.110,
   FC-log[2] = −1.1) in the pathological group were detected at day 3.

Figure 5.

   [103]Figure 5
   [104]Open in a new tab

   Profiles of bile acids and their conjugates in feces and plasma

   (A and C) Presence of bile acids (%) in experimental groups at
   different time points.

   (B and D) Bile acid profiles of individual samples.

   (E) Box plots of bile acids altered between experimental groups
   (plasma: n[CTR] = 14, n[PAT] = 10; feces: n[CTR] = 3, n[PAT] = 8). ∗p <
   0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

   (F) Chromatographic peaks of the detected analytes.

Feces

   In the fecal samples of extremely premature infants, we identified 40
   of the recently discovered bile acid conjugates of five bile acids
   ([105]Figure 5B), with the exception of lithocholic acid conjugates. A
   clear trend between the abundance of the primary bile acids CDCA (p =
   0.033, FC-log[2] = 3.3), CA (p = 0.110, FC-log[2] = 1.6), and their
   taurine conjugates (Tau-CDCA: p = 0.067, FC-log[2] = −3.0; Tau-CA: p =
   0.110, FC-log[2] = −3.0) between the experimental groups was observed
   ([106]Figure 5E), although only one was regarded as significant based
   on their p values.

Discussion

Time-dependent development is the dominating factor in overall differences in
plasma and fecal metabolome

   Generally, the metabolome of the pathological and control groups
   exhibited no group-specific clustering in the PCA. However, there was
   clear clustering between sampling time points, as seen in plasma and
   feces ([107]Figures 1B and 1C). This finding was supported by the
   PERMANOVA, which found that time point was a significant factor while
   group affiliation was not deemed to be significant. This indicates that
   metabolic changes specific for the differentiation between the
   pathological and control groups might be only based on a subset of
   metabolites, and in general individual factors dominate the differences
   in the metabolome between patients. The plasma metabolome in the early
   time points, especially on day 7, displayed a less tight clustering and
   seemed to exhibit more spread compared to term-equivalent age
   ([108]Figure 1C), indicating a higher susceptibility to variation
   during earlier periods of infant development. At term-equivalent age a
   tight clustering between samples was seen, indicating general
   stabilization and convergence of the plasma metabolome over time. The
   opposite trend was observed in the fecal samples ([109]Figure 1B),
   where a slightly tighter clustering was shown at day 28 accompanied by
   a trend to spread out at term-equivalent age, indicating a
   diversification of the gut metabolome over time and possibly also the
   gut microbiome.[110]^30

Metabolic differentiation between experimental groups is detectable in the
earliest sample time point in both plasma and feces

   Despite the findings above, significant changes in several metabolic
   pathways between the pathological and control groups were already
   observed at the earliest time points ([111]Figures 2A and 2B). Pathways
   related to the metabolism of several amino acids dominated in plasma
   from day 3 to week 32 while being not affected at the last time point.
   Carbohydrate metabolism was most prominent in feces at day 7, while at
   term-equivalent age several pathways related to fatty acids and amino
   acid metabolism were the driving force. At day 28 (feces), a mix of
   altered carbohydrate and amino acid metabolism was observed, closing
   the gap between the previous and following time point. In general, a
   development over time is apparent between different sampling time
   points in both plasma and feces, where the affected pathways during
   term-equivalent age are most distinct from the other time points.
   Interestingly, time point day 28 in feces and plasma exhibited the most
   significantly changed pathways, in both cases including sialic acid
   metabolism. This pathway includes N-acetylneuraminic acid, which
   participates in the brain as an integral part of ganglioside structure
   in synaptogenesis.[112]^31 Furthermore, in feces, at all three time
   points a significant or near-significant enrichment was found in
   biopterin metabolism. Being a cofactor for aromatic amino acid
   hydroxylases, biopterin is involved in the synthesis of
   neurotransmitters including dopamine and serotonin.[113]^32 A
   difference in the presence and abundance of several metabolites in both
   plasma and feces was already observed between experimental groups in
   the first week of life, many of them associated with neuroactivity,
   neuroprotection, neurotoxicity, and immunomodulatory activity. In
   [114]Figures 3A–3G, a portion of these differentiated metabolites are
   highlighted. Several amino acids were significantly increased in plasma
   samples from the pathological group at day 3, mainly alanine, aspartic
   acid, serine, glycine, proline, lysine, homoserine, citrulline, and
   glutamine, which is the principal excitatory neurotransmitter in the
   brain.[115]^33 One interesting compound elevated in the pathological
   group at day 3 was galactitol, a reduction product of galactose.
   Elevated levels are associated with the presence of
   galactosemia[116]^34 and other genetic diseases marked by the
   deficiency in galactitol metabolism. In addition, it has been
   implicated for its harmful neurotoxic properties at elevated
   concentrations and is connected to brain damage[117]^35^,[118]^36 and
   retinopathy.[119]^34^,[120]^37 Sorbitol, a reduction product of
   glucose, was associated with peripheral neuropathy[121]^38 as well as
   retinopathy,[122]^37 like galactitol. Mannitol, a commonly used
   additive in medication, elevates blood plasma osmolality, thereby
   reducing cerebral edema, elevated intracranial pressure, and
   cerebrospinal fluid volume and pressure.[123]^39 The potential of these
   two compounds to cause retinopathy is especially interesting, since all
   infants in the pathological groups developed ROP and furthermore was
   significantly associated with the pathological group. Comparing the
   galactitol levels at day 3 between infants with and without ROP
   regardless of experimental group revealed a significant association
   between increased galactitol levels at day 3 and the development of ROP
   (Wilcoxon rank-sum test: p = 0.013). Although a reference standard for
   galactitol was acquired (which confirms the identity at an ICL 1,
   [124]Figure S2B),[125]^40 it cannot be excluded that it is its
   stereoisomer mannitol/sorbitol or a mixture of them, since their
   retention time and MS/MS spectra might be identical. Interestingly, at
   the same time a decrease in the alternative galactose metabolite
   galactonic acid was observed ([126]Figure 3B), indicating altered
   galactose metabolism between experimental groups shortly after birth.
   Another elevated neurotoxic compound was p-cresol sulfate at day 28
   (although not deemed significant, p = 0.067), the sulfate conjugate of
   the bacteria-derived compound p-cresol.[127]^41 It is associated with
   neurological impairments in relation to kidney disease, disease
   progression in Parkinson’s disease,[128]^42^,[129]^43 and
   neurotoxicity.[130]^44 A decrease of lanthionine, a biomarker for H[2]S
   production, was observed in plasma at day 7, indicating that there
   might be a difference in hydrogen sulfide-related metabolism in the
   first week after birth. A strong increased FC-log[2] of histamine was
   also observed at day 7 in the feces, although not regarded as
   significant. Histamine is an important neurotransmitter, especially in
   relation to immune cells, that influences various brain functions such
   as control of pituitary hormone secretion and cognitive
   functions[131]^45 and is known to be secreted by the human gut
   microbiome. Increased levels of lactate and ratio of lactate/pyruvate
   at birth have been used as a predictor of occurrence of neonatal
   encephalopathy in term and preterm infants in another study.[132]^46 We
   found a similar distribution of the lactate/pyruvate ratio
   ([133]Figure S6) for brain injury in the pathological group, although
   it was not statistically significant.

Altered tryptophan metabolites and their activity with the CNS

   Several compounds related to TRP metabolism with relevant biological
   activity, some likely of bacterial origin, were found to be
   significantly altered between experimental groups ([134]Figure 3A).
   Tryptamine was increased in plasma at day 3 in the pathological group
   while being decreased at day 7. In feces, increased levels were found
   at day 7, although not deemed significant. Tryptamine is a bacterial
   neuroactive TRP metabolite generated by multiple bacterial taxa[135]^10
   and has been associated with neurotoxicity, neurodegeneration, and the
   onset of autism spectrum disorder (ASD) and
   schizophrenia.[136]^8^,[137]^47^,[138]^48^,[139]^49^,[140]^50 It
   modulates intestinal homeostasis by acting on the enteric nervous
   system and can induce the release of serotonin and ion secretion by
   intestinal epithelial cells.[141]^51 Indolelactic acid, a metabolite
   produced by Bifidobacterium,[142]^52 was significantly increased in the
   pathological group at day 7 in both plasma and feces. This TRP
   metabolite was found to impact immune function in early
   life[143]^53^,[144]^54 and to be elevated in children with ASD.[145]^55
   Moreover, it also has been suggested to exhibit neuroprotective
   properties in infants,[146]^53 and its levels were found to strongly
   correlate with inflammation responses in humans.[147]^56 Another TRP
   metabolite affected in plasma and decreased at day 7 was indoleacrylic
   acid, which promotes intestinal epithelial barrier function and
   mitigation of inflammatory responses, thereby having an important
   anti-inflammatory function in the intestine.[148]^57 5-Hydroxyindole
   was found to be significantly decreased in the pathological group in
   feces at day 28 and is known to stimulate gastrointestinal
   motility[149]^58 and affect the α-7-nicotinic
   receptor,[150]^59^,[151]^60^,[152]^61 which has a potential role in the
   reduction of inflammatory and neurotoxicity in sepsis, myocardial
   infarction, stroke, and Alzheimer’s
   disease.[153]^62^,[154]^63^,[155]^64 Indole was also increased in
   plasma of the pathological group at term-equivalent age and has been
   found at high levels in children with ASD.[156]^8 Indoxyl sulfate, an
   indole metabolite, was increased in the pathological group at day 7 in
   the feces. It has been associated with ASD in children[157]^55 and acts
   as an agonist of the aryl-hydrogen receptor (AhR),[158]^65 a receptor
   in the gut-brain axis[159]^66 that is involved in inflammatory and
   neoplastic diseases.[160]^67 Kynurenic acid, a metabolite of the
   kynurenine pathway of endogenous TRP metabolism, was significantly
   decreased in plasma at week 32 ([161]Figure 3G). It is known to
   interact with AhR, glutamate, and N-methyl-D-aspartic acid (NMDA)
   receptors and has influence on important neurophysiological and
   neuropathological processes including
   neuroprotection.[162]^68^,[163]^69 Furthermore, kynurenic acid is also
   involved in the immune system and inflammation, exhibiting
   anti-inflammatory and immunosuppressive functions.[164]^70

Steroid sulfates and their role in the gut-brain axis and neurodevelopment

   At day 3 and week 32 in plasma, dopamine 3-O-sulfate was significantly
   decreased in the pathological group. In human blood circulation
   endogenous dopamine exists predominantly in the sulfated form, and
   sulfonation is an important regulation process, since the sulfated form
   does not bind to dopamine receptors.[165]^71 Furthermore, dopamine
   itself and its metabolite 5-hydroxyindoleacetic acid were significantly
   decreased in the feces of the pathological group at day 28. Both in
   feces and plasma, the abundance of several steroid sulfates was altered
   between experimental groups at different time points ([166]Figures 3D
   and 3E). Steroids play a role in neuromodulation and are contributors
   to gut-brain-axis signaling. Furthermore, they are influenced by the
   gut microbiome, presumably through modulating GABA[A] and NMDA
   receptors.[167]^72 Steroid sulfation and desulfation greatly affect the
   biological activity, regulating the steroid bioactivity and, in many
   cases, unconjugated and sulfated steroids acting in opposite ways on
   the same receptors. Steroid sulfates also have an important role in the
   regulation of steroid metabolism and transport.[168]^73 Sulfation also
   increases the polarity, thereby reducing their passage through the
   blood-brain barrier (BBB). Furthermore, steroid sulfonation by gut
   bacteria has been associated with alteration of immune cell trafficking
   in the host.[169]^74 A higher abundance of neurosteroids, including
   epipregnenolone sulfate and pregnenolone sulfate (not deemed
   significant), were found in the feces of the pathological group at day
   7. Epipregnenolone sulfate and pregnenolone sulfate are neurosteroids
   and agonists of TRPM3[170]^75 and negative allosteric modulators of
   GABA[A][171]^76 and NMDA receptors,[172]^77 both being relevant parts
   of the gut-brain axis.[173]^78^,[174]^79 Furthermore, hypofunction of
   NMDA receptors was associated with impairment of synaptic
   plasticity.[175]^80 In plasma a decrease in several steroid sulfates
   was observed at day 3 and then again at week 32 and term-equivalent
   age, including estriol 3-sulfate. Estriol can only be produced from
   fetal precursors[176]^81 and plays a particularly prominent role in the
   context of pregnancy and fetal development. It also exhibited
   neuroprotective properties in multiple sclerosis by reversing myelin
   breakdown in the brain. A clear interpretation of altered steroid
   sulfate levels is difficult, since there are several potential causes
   such as a change in sulfonation or desulfonation, increased excretion,
   and augmented uptake into tissue. Mevalonic acid, which was decreased
   in plasma at term-equivalent age ([177]Figure 3G), is essential for
   inducing trained innate immunity and is an integral part of cholesterol
   synthesis.[178]^82 Besides being the precursor of the synthesis of
   other steroids, cholesterol is also an important constituent of
   neuronal myelin and is crucial for brain development during the
   embryonic and postnatal stages.[179]^83

Brain- and immune-system-affecting plasma metabolites differentiate at later
time points

   During enrichment analysis ([180]Figure 2B), the metabolic pathway
   porphyrin metabolism, including heme degradation, was significantly
   altered between experimental groups at term-equivalent age in plasma.
   Abundance of the heme degradation metabolites biliverdin, bilirubin,
   and 5-oxo-δ-bilirubin were significantly decreased in the pathological
   group at term-equivalent age ([181]Figure 3F). Biliverdin and
   5-oxo-δ-bilirubin are products of heme catabolism, and biliverdin is
   further metabolized to bilirubin, which has been associated with
   neurological dysfunction and neurotoxicity.[182]^84^,[183]^85 Increased
   levels of bilirubin lead to jaundice, and in infants the underdeveloped
   BBB of neonates allows bilirubin to diffuse across and accumulate in
   the brain, causing permanent brain damage. Besides these negative
   effects, bilirubin and biliverdin have been linked to numerous
   anti-inflammatory and immunosuppressive actions and regulation of the
   innate immune system.[184]^86^,[185]^87 Furthermore, they are known
   activate the AhR.[186]^88 The decrease of bilirubin and biliverdin
   might be caused by an altered heme abundance or heme degradation.
   Another possibility is increased excretion, although no difference in
   their abundance or of their metabolites was observed in the feces. An
   increased permeation through an underdeveloped BBB may also occur,
   potentially contributing to brain injury owing to the neurotoxic
   effects of these compounds. We observed several significant
   correlations between the metabolomics data and the 16S rRNA gene
   amplicon, cytokine, growth factor, and T cell data. Plasma levels of
   heme degradation metabolites were decreased in the pathological group
   at term-equivalent age, and positive correlations with
   anti-inflammatory and neuroprotective factors were observed while
   correlations with pro-inflammatory factors were negative
   ([187]Figure 4A). The same trend was observed for galactonic acid and
   indoleacrylic acid. A significant negative correlation between
   galactitol in plasma and Klebsiella was observed. This aligns well with
   increased levels of the neurotoxic galactitol in plasma at day 3 in the
   pathological group and the decreased levels of Klebsiella in the first
   week of life.[188]^30 Klebsiella overgrowth at day 28 was previously
   associated with brain injury in extremely premature infants and could
   be a potential predictor of pathological neurophysiological
   development. The same trend was found in the feces, where there was a
   negative correlation between Klebsiella and the TRP metabolite
   5-hydroxyindole, both significantly altered at day 28. Several steroid
   sulfates, found to be differently abundant between experimental groups
   in the feces, were significantly negatively correlated with vascular
   endothelial growth factor A (VEGF-A), which was associated with
   pathological brain development in premature infants.[189]^2

Differentiation of bile acids and potential differences in metabolic activity
of the gut microbiome

   Recently, more than 145 as yet undescribed bacterial-derived bile acid
   amino acid conjugates[190]^29 were discovered that have been connected
   to disease phenotypes in Crohn’s disease, ulcerative colitis, and
   disease severity in inflammatory bowel disease.[191]^89 Initiated by
   this recent discovery, we batch-synthesized more than 120 bile acid
   conjugates consisting of CA, CDCA, DCA, HDCA, UDCA, and LCA coupled to
   19 amino acids and taurine, followed by the development of a targeted
   LC-MS/MS method for detection and quantitation of those bile acids and
   their conjugates. Utilizing this developed assay, we detected 40 of the
   recently discovered bile acid conjugates in the fecal samples
   ([192]Figures 4A and 4B), consisting mainly of primary bile acid amino
   acid conjugates of CA and CDCA, although secondary bile acid amino acid
   conjugates of DCA, HDCA and UDCA were also detected. In the fecal
   sample at day 7 ([193]Figure 5E), a significantly higher abundance of
   CDCA was observed for the pathological group, while for the
   taurine-conjugated form of CDCA an opposite trend was observed. In the
   plasma, no significant changes for CDCA or Tau-CDCA at were found at
   day 7. Chenodeoxycholic acid is conjugated to taurine in the liver, and
   the resulting Tau-CDCA is excreted via the bile into the intestinal
   tract where it can be reabsorbed into the bloodstream
   (i.e., enterohepatic circulation). It can also be deconjugated back to
   CDCA and metabolized to secondary bile acids or other bile acid
   metabolites by the gut microbiome. This increase in CDCA coupled with a
   decrease in Tau-CDCA in feces might be an indicator of a general or a
   bile acid specific elevation of bacterial metabolism in the intestinal
   tract. CDCA has been shown to decrease trans-epithelial electrical
   resistance, increase intestinal permeability, and elevate levels of
   IL-8, IL-6, TNF-α, and VEGF release.[194]^90 In contrast to the fecal
   samples, only six of the bile acid amino acid conjugates were sparingly
   detected in plasma samples ([195]Figures 5C and 5D). It is not clear
   whether this was caused by the absence of those compounds or by the
   expected lower concentration in the blood compared to feces, rendering
   them to be under the limit of detection. The taurine and glycine
   conjugates of the gut-microbiome-derived secondary bile acids HDCA,
   DCA, and, partially, UDCA were already detected in plasma samples at
   day 3. This observation might indicate that 3 days after delivery there
   is already a metabolically active gut microbiome established, producing
   several bile acid transformation products which are reabsorbed into the
   bloodstream with the potential of affecting the host. The findings in
   our previous publication[196]^2 showed the presence of a gut microbiome
   in fecal samples as early as day 3, exhibiting already significant
   differences between the experimental groups. However, it is also known
   that primary and secondary bile acids in their free form can be
   transferred through breast milk from the mother to the
   infant,[197]^91^,[198]^92 and to a much lesser degree glycine and
   taurine conjugates.[199]^93 Therefore, it cannot be excluded that these
   secondary bile acids might originate partially or altogether from the
   consumed breast milk. At day 3, lower plasma levels of the primary bile
   acid conjugates Tau-CA and Gly-CA were observed in the pathological
   group ([200]Figure 5E). The reported changes in bile acids at the
   earliest sampling time points in both plasma and feces clearly
   strengthen the observations from the untargeted metabolomics
   experiments, observing the same early onset of differentiation in
   metabolites between experimental groups. In general, there was no clear
   distinction in the occurrence of the amino acid conjugates between
   experimental groups, with their presence seeming to be a highly
   individual factor ([201]Figure 5A).

Limitations of the study

   Due to the nature and circumstances of extremely premature infants, it
   is exceptionally difficult to obtain samples in greater number from the
   same cohort or same patient. This results in typically smaller sample
   size but also a limited amount of possible sampling time points with
   greater time gaps in between. Furthermore, this makes it difficult to
   obtain a fully longitudinal sample set. Also, the sample amount
   available is limited, making the sample preparation and subsequent
   measurements more challenging. Another problematic factor is the
   potential for different onset times in the pathological processes
   behind brain injury among patients. This potentially delayed onset
   combined with the greater time gaps between sampling time points poses
   a major problem for a quantitative time-course investigation,
   especially regarding statistical analysis, since not every patient
   might be in the exact same step of pathological progression at the
   sampling time point. At the early time points this can lead to missing
   values of potential key metabolites, and combined with the small sample
   size this increases the difficulty of analysis and reduces the
   statistical power needed for deeming changes as significant. In
   addition, for one sampling time point (feces day 7) only a very low
   number of samples in the control group was available; therefore, the
   results regarding this group must be interpreted with caution.
   Furthermore, the annotation of some significantly altered metabolites
   are based on in silico tools, so a false-positive identification cannot
   be excluded. In general, it is challenging to differentiate between
   potential causes and symptoms of pathological neurological development
   due to metabolic and bacterial changes. An early neurological injury
   might cause differences in various areas such as feeding, antibiotic
   use, gastrointestinal motility, or development of the BBB, causing
   metabolic or microbial changes later on. The observed correlations
   between 16S rRNA gene amplicon data and metabolites warrant future
   studies employing metagenomic analyses to investigate the metabolic
   potential of the observed bacterial strains and draw a stronger
   conclusion regarding the relation between the observed metabolites and
   the microbiome. Despite these difficulties, we were able to collect the
   most comprehensive sample set to date over several time points of
   extremely premature infants and obtained comprehensive and highly
   relevant insights by employing highly optimized sample preparation
   procedures and MS workflows to deal with the very low sample amounts
   available.

Conclusion and outlook

   The occurrence of extremely premature births is rising globally,
   leading to an ever-growing highly vulnerable patient group. Although
   survival rates of extremely premature infants have been increasing as a
   result of improved neonatal intensive care, a considerable number of
   infants suffer from severe morbidity and life-long neurodevelopmental
   impairment. In this work we determined the metabolic profile of plasma
   and fecal samples from different sampling time points in two groups of
   patients based on MRI findings at term-equivalent age. Despite the
   limited sample amount and the difficult nature of the biological
   background of this patient group, we observed relevant molecular
   markers, indicating an early onset of differentiation between the group
   with brain abnormalities and the control group without any
   abnormalities. This distinction is characterized by a difference
   regarding the abundance and presence of primary bile acids/conjugates
   and numerous other compounds with relevant bioactivities, some of them
   derived from the gut microbiome. This includes neuroactivity,
   neurotoxicity, neuroprotection, immunomodulation, and inflammatory and
   anti-inflammatory actions. Furthermore, a broad array of significantly
   altered molecules is known to interact with AhR, GABA, and NMDA
   receptors, which are part of the gut-brain axis and are involved in the
   immune system, CNS, inflammation, neurodevelopment, and synaptic
   plasticity. Several compounds correlated to pro- and anti-inflammatory
   factors were significantly altered between the pathological group and
   the control group. These compounds might be beneficial in early
   diagnosis of pathological development. Furthermore, microbially derived
   secondary bile acids and bile acid amino acid conjugates were already
   present in the plasma 3 days after delivery, suggesting the
   establishment of an already metabolic active gut microbiome and the
   absorption of its metabolites into the bloodstream. This would indicate
   that the potential early influence of the gut microbiome on its host is
   not limited to the gut-brain axis but already takes place via direct
   uptake of metabolites. The findings in this study help to better
   understand the cause and development of brain impairments in extremely
   premature infants and their connection to the gut microbiome and lay
   the foundation for mechanistic studies.

STAR★Methods

Key resources table

   REAGENT or RESOURCE SOURCE IDENTIFIER
   Biological samples
     __________________________________________________________________

   Peripheral blood samples from extremely premature infants General
   Hospital of Vienna N/A
   Fecal samples from extremely premature infants General Hospital of
   Vienna N/A
     __________________________________________________________________

   Deposited data
     __________________________________________________________________

   Raw metabolomics data
   [202]https://www.ebi.ac.uk/metabolights/editor/MTBLS7560 Metabolights:
   MTBLS7560
   Supplementary spectral data [203]https://phaidra.univie.ac.at/o:2044701
   Phaidra: 2044701
   16S rRNA gene amplicon sequencing, T-cells, cytokines, growth factor
   data [204]https://doi.org/10.1016/j.chom.2021.08.004 NCBI: PRJNA715072
     __________________________________________________________________

   Chemicals, peptides, and recombinant proteins
     __________________________________________________________________

   Water LC-MS grade VWR Cat # 83645.320P
   Acetonitrile LC-MS grade Honeywell Riedel de-Haën Cat # 14261
   Methanol LC-MS grade Honeywell Riedel de-Haën Cat # 14262
   Formic acid Promochem Cat # FLK.14265.0050.GF
   In-house chemical reference standard library METHOD DETAILS/Chemicals
   and reagents [205]Table S1
   SRM 1950[206]^107 NIST, Gaithersburg, US
   [207]https://doi.org/10.1021/ac402503m
     __________________________________________________________________

   Software and algorithms
     __________________________________________________________________

   Python (3.8) Python Software Foundation [208]https://www.python.org
   R (4.1) The R Project for Statistical Computing
   [209]https://www.R-project.org/
   XCMS (3.14) Smith et al.[210]^94
   [211]https://bioconductor.org/packages/release/bioc/html/xcms.html
   Proteo Wizard (3.0.18) Chambers[212]^95
   [213]https://proteowizard.sourceforge.io/
   CAMERA (1.48) Kuhl et al.[214]^96
   [215]https://www.bioconductor.org/packages/release/bioc/html/CAMERA.htm
   l
   NeatMS (0.9) Gloaguen et al.[216]^97
   [217]https://github.com/bihealth/NeatMS
   SIRIUS 4 (4.9) Dührkop et al.[218]^98
   [219]https://bio.informatik.uni-jena.de/software/sirius-4-0-1/
   Skyline (20.2) Adams et al.[220]^99
   [221]https://skyline.ms/project/home/software/Skyline/begin.view
   qvalue (2.28) Storey et al.[222]^100
   [223]https://www.bioconductor.org/packages/release/bioc/html/qvalue.htm
   l
   GNPS Aron et al.[224]^101
   [225]https://gnps.ucsd.edu/ProteoSAFe/static/gnps-splash.jsp
   MetaboAnalyst (5.0) Pang et al.[226]^102
   [227]https://www.metaboanalyst.ca/
   mummichog (version 2) Li et al.[228]^26
   [229]https://github.com/shuzhao-li/mummichog
   MixOmics (6.20) Rohart et al.[230]^103
   [231]https://bioconductor.org/packages/release/bioc/html/mixOmics.html
   Cytoscape (3.9) Shannon et al.[232]^103 [233]https://cytoscape.org/
   vegan (2.3) Oksanen et al.[234]^104
   [235]https://cran.r-project.org/web/packages/vegan/index.html
   Hmisc (4.7) Harrell et al.[236]^105
   [237]https://cran.r-project.org/web/packages/Hmisc/index.html
   Nontargeted Analysis Study Reporting Tool Peter et al.[238]^106
   [239]https://doi.org/10.1021/acs.analchem.1c02621
   [240]Open in a new tab

Resource availability

Lead contact

   Further information and requests for resources and reagents should be
   directed to the lead contact, Benedikt Warth
   (benedikt.warth@univie.ac.at).

Materials availability

   This study did not generate new unique reagents.

Data and code availability

   The raw LC-HRMS and LC-MS/MS data are publicly provided via the
   Metabolights repository ([241]https://www.ebi.ac.uk/metabolights) under
   the identifier MTBLS7560. The 16S rRNA gene amplicon sequencing,
   T-cells, cytokines, growth factor data can be found at NCBI under the
   identifier PRJNA715072 (https://doi.org/10.1016/ j.chom.2021.08.004).
   Additional supplementary spectral data can be found under
   [242]https://phaidra.univie.ac.at/o:2044701. No custom code is reported
   in this paper. Any additional information required to reanalyze the
   data reported in this paper is available from the [243]lead contact
   upon request.

Experimental model and study participant details

Cohort of extremely premature infants and sample collection

   This study was approved by the ethics committee of the Medical
   University of Vienna (Ethics number 1348/2017). In a single cohort of
   sixty extremely premature infants, neurophysiological development was
   monitored during hospitalization and brain injuries were identified by
   a combination of ultrasound imaging at multiple time points and
   magnetic resonance imaging (MRI) at term-equivalent age. The infants
   were enrolled between September 2017 and June 2019 at the Medical
   University of Vienna. All parents gave written-informed consent before
   study inclusion. Out of the cohort of 60 infants, 53 survived and 38 of
   the remaining infants displayed age-adequate MRIs or only mild brain
   injuries while 15 infants were diagnosed with severe pathological brain
   injuries. Infants diagnosed with age-adequate or only mild brain
   injuries were assigned to the control group (CTR) while the
   pathological group (PAT) consisted of infants with severe pathological
   brain injuries. Among the observed injuries were periventricular
   hemorrhagic infarctions, intraventricular hemorrhages, cerebellar
   hemorrhages, and periventricular leukomalacia. A reduction of brain
   volume and an enlargement of the subarachnoid spaces was accompanying
   many of those injuries. Out of those surviving infants, samples from 51
   infants were available and used for this study ([244]Figure 1A). The
   distribution of sex assigned at birth within and between experimental
   groups was close to 50% (CTR: 56% female, PAT: 47% female) and showed
   no statistical significance ([245]Table 1). Every infant received the
   probiotic preparation Infloran (Bifidobacterium bifidum and
   Lactobacillus acidophilus) from the first day of life until the
   corrected age of 34 weeks of gestational age. The enteral feeding
   regimen consisted of the infant’s own pasteurized mother’s milk or
   pasteurized human donor milk. After achieving 100 mL kg^−1 enteral
   feeding, the milk was fortified with 4% bovine milk fortifier until
   term-equivalent age. Characteristics of the study cohort used in this
   study can be found in [246]Table 1 and the SI (SI_Cohort_Information).
   Over the course of hospitalization and when possible, fecal samples
   were collected, and blood was drawn at multiple time points with a
   focus on day 3, day 7, day 28, correction week 32 and term-equivalent
   age. The fecal samples from all extremely premature infants were
   collected from diapers via collection tubes and stored at −80°C until
   further analysis. For minimizing the stress for patients, blood samples
   were only taken combined with clinical routine sampling. The blood
   samples were transferred into Lithium Heparin S-Monovettes and the
   plasma was separated and stored at −80°C until further analysis.
   Additional information about the cohort and MRI-results and
   neurophysiological assessment can be found in previous work.[247]^2

Method details

Chemicals and reagents

   Acetonitrile (ACN), methanol (MeOH) and water were LC-MS grade and
   purchased from VWR Chemicals (water) or Honeywell Riedel de-Haën (ACN,
   MeOH). Formic acid (FA) used as an additive for eluents was purchased
   from Promochem. Chemical reference standards used during LC-MS
   measurements were acquired from Sigma Aldrich, Carbosynth or Toronto
   Research Chemicals. The bile acid conjugate reference standards were
   synthesized as described in the supporting information
   ([248]Figure S1A), and the used reagents were purchased from Sigma
   Aldrich. Isotopically labeled compounds used as internal standards,
   were purchased from Cambridge Isotope Laboratories.

Batch synthesis of bile acid conjugates

   A solution of a single bile acid (40 μL, 2.5 mM, 1 eq) was combined
   with freshly prepared solutions of triphenylphosphine (40 μL, 10 mM, 4
   eq) and dipyridyldisulfide (40 μL, 10 mM, 4 eq), vortexed and incubated
   for 1 min. A amino acid stock solution (5 μL, 2 mM, 0.25 eq),
   containing 10 amino acids, was added and the reaction mixture was
   incubated on a thermo-shaker (TS-100, Biosan) at 60°C for 30 min,
   followed by the addition of 300 μL of ACN. Taurine conjugates were
   synthesized in the same manner. This procedure was repeated for all
   bile acids until all desired compounds were acquired. From each mixture
   150 μL were taken, combined, evaporated and reconstituted in 150 μL ACN
   to generate a concentrated standard.

Sample preparation of plasma samples

   Plasma samples obtained from premature infants with pathological damage
   or age-adequate damage were used. The samples were stored at −80°C
   prior to extraction. During the whole sample preparation, the samples
   were kept on ice. Homogenization by vortexing was followed by the
   addition of the extraction solvent (ACN/MeOH, 50:50, v/v) to the
   homogenized sample, resulting in a 1:4 volume ratio of sample and
   extraction solvent. After further vortexing and sonication in an ice
   bath for 15 min, the sample solutions were stored overnight at −20 C°.
   Afterward the samples were centrifuged at 18,000 x g and 4°C for 10 min
   and the supernatant was transferred to another tube. For targeted
   LC-MS/MS analysis of bile acid conjugates, aliquots were taken from the
   solution and stored at −80°C until further use. The remaining solution
   was evaporated to dryness at 4°C using a SpeedVac concentrator
   (Labconco). The residues were reconstituted (ACN/water, 50:50, v/v) to
   one-fifth of the volume prior to evaporation, vortexed and sonicated in
   an ice bath for 15 min. After centrifugation at 18,000 x g and 4°C for
   10 min, the supernatants were transferred to LC vials and stored
   at −80°C until analysis. An internal standard (^2H[4]-ursudeoxycholic
   acid) was added to the aliquots taken for the targeted LC-MS/MS
   measurements to a final concentration of 2 μM, followed by a transfer
   of the samples to LC vials and storage at −80°C until analysis.

Sample preparation of fecal samples

   Stool samples obtained from premature infants with pathological damage
   or age adequate damage were used. The samples were stored at −80°C
   prior to the extraction. During the whole sample preparation the
   samples were kept on ice. The samples were manually homogenized using a
   spatula and dried at 4°C using a SpeedVac concentrator (Labconco) for
   60 h followed by the addition of MeOH/ACN/H[2]O (40:40:20; v/v:v) at a
   ratio of 50 μL for each mg of dried sample. After vortexing and
   sonication on ice for 10 min, the samples were homogenized in a bead
   shaker (4 m/s, MP FastPrep-24 5G) for 15 s and stored overnight at
   −20°C. Afterward, the samples were centrifuged at 18000 g and 4°C for
   10 min and transferred to another tube. The solution was diluted 1:1
   with ACN/H[2]O (20:80; v/v) containing internal standards
   (^13C[6]-bisphenol A, ^2H[4]-genistein). After centrifugation at 18,000
   x g and 4°C for 10 min, samples were sonicated in an ice bath for
   10 min to dissolve potential precipitate and transferred to HPLC vials
   and stored at −80°C until analysis.

   The fecal samples for the targeted LC-MS/MS analysis of bile acid
   conjugates were dried as described above, followed by the addition of
   25 μL H[2]O for each mg of dried sample and the homogenization in a
   bead shaker FastPrep-24 5G (4 m/s) for 10 s. Subsequently the samples
   were centrifuged at 18,000 x g and 4°C for 10 min and half the volume
   of the supernatant was transferred to another tube and stored for
   additional biochemical assays. The next step was the addition of
   MeOH/ACN (1:1; v/v) to a final solution of MeOH/ACN/H[2]O (40:40:20;
   v/v:v), following the extraction with a bead shaker (4 m/s, MP
   FastPrep-24 5G) for 10 s. After sonication in an ice bath for 10 min,
   the sample solutions were stored overnight at −20 C°. Afterward the
   samples were centrifuged at 18,000 x g and 4°C for 10 min and
   transferred to another tube. The remaining solution was evaporated to
   dryness at 4°C in a SpeedVac (Labconco) and the residues were
   reconstituted with (ACN/H[2]O, 50:50, v/v), containing an internal
   standard (^2H[4]-ursudeoxycholic acid, 2 μM), to the same volume before
   evaporation, vortexed and sonicated in an ice bath for 15 min. After
   centrifugation at 18,000 x g and 4°C for 10 min, the samples were
   transferred to HPLC vials and stored at −80°C until analysis.

Untargeted metabolomics experiments

   For the analysis a Vanquish UHPLC system coupled to a QExactive HF
   quadrupole-Orbitrap (Thermo) mass spectrometer via an ESI interface was
   used. Two columns, a HILIC column (SeQuant ZIC-pHILIC; 5 μm,
   150 × 2.1 mm) and an RP column (Waters Acquity HSS T3; 1.8 μm,
   100 × 2.1 mm) were chosen for complementary and broad coverage of
   analytes. In both cases a precolumn made of the respective material was
   used. An injection volume of 3 μL, a flow rate of 0.3 mL/min and a
   column compartment temperature of 40°C was selected for both columns.
   The HILIC chromatography utilized H[2]O and ACN with 10 mM NH[4]HCO[3]
   (90:10, v/v, pH 9.2, eluent A) and ACN (eluent B) for the following
   gradient: 0–6 min linear decrease from 75% B to 45%B, 6–7 min 45% B to
   30% B, 7–10 min constant at 30% B, 10 min until 10.1 min increase from
   30% B to 75% B and maintained constant at 75% B for 5 min. The RP
   chromatography utilized H[2]O with 0.1% formic acid (eluent A) and MeOH
   with 0.1% formic acid (eluent B) for the following gradient: 0–0.5 min
   stay at 5% B, 0.5–11.5 min linear increase to 95% B, 11.5–15.5 hold at
   95% B,15.5–15.6 decrease from 95% B to 5% B, 15.6–17 min stay at 5% B.
   Measurements were conducted in either full scan with polarity switching
   mode or in positive and negative mode with data dependent MS^2
   collection. The settings of the ESI interface were: sheath gas, 48 au;
   auxiliary gas, 15 au; sweep gas flow, 2 au; capillary voltage, 3.5 kV
   (positive), 2.8 kV (negative); capillary temperature, 300°C; auxiliary
   gas heater, 350°C. The scan range was from m/z 62 to 900. For full scan
   measurements the resolution was set to 60,000 with an automatic gain
   control (AGC) of 1x10^6 and a maximum injection time of 200 ms. For
   generating MS^2 spectra, data dependent acquisition in combination with
   an inclusion list with features of interest, based on the untargeted
   and targeted evaluation of the MS1 data, was used. For the
   data-dependent MS^2 acquisition a resolution of 60 000 with an AGC
   target of 1x10^6 and a maximum injection time of 100 ms were chosen for
   the full scan. The settings for the MS^2 collection were as following:
   resolution 30,000, AGC target 5x10^5, maximum injection time 120 ms,
   loop count 10, isolation window 1.0 m/z, stepped collision energy 30–50
   eV; minimum AGC 8x10^3.

Targeted LC-MS/MS for bile acids and their conjugates

   A targeted LC-MS/MS for the investigation of bile acids and their
   conjugates was developed employing commercially purchased or self
   synthesized reference standards ([249]Figure S1). For the targeted
   LC-MS analysis a Dionex Ultimate 3000 UHPLC (Thermo) system coupled to
   a TSQ Vantage triple quadrupole mass spectrometer (Thermo) via an ESI
   interface was used. An Atlantis T3 column (Waters, 3 μm,
   3 mm × 150 mm), including a precolumn made of the same material, at a
   flow rate of 0.6 mL/min was used and maintained at 40°C. The injection
   volume was 5 μL. Chromatography used H[2]O with 0.1% formic acid
   (eluent A) and ACN/MeOH (1:1, v/v, eluent B) with 2% H[2]O and 0.1%
   formic acid for the following gradient: 0–1 min at 5% B; rise to 50% B
   until 1.5 min; increase to 95% B until 10 min; 10–13 min at 95% B;
   13–13.1 min from 95% B to 75% B and maintain until 19 min before
   re-equilibrating at 5% B until 21 min. Negative ESI mode was used with
   the following ion source parameters: Capillary temperature of 300 (°C),
   vaporizer temperature of 300 (°C), sheath gas pressure of 40 (Arb), aux
   valve flow of 10 (Arb), declustering potential of 14 (V), collision gas
   pressure of 1.5 (mTorr) and a spray voltage of 3000 (V). The cycle time
   was set to 0.6 (s), while Q1 and Q3 peak width was kept at 0.7 (FWHM).

QC & QA measurements for LC-MS methods

   For the untargeted metabolomics measurements multiple measures to
   assure high quality data were taken. During the sample preparation
   three process blanks were processed together with the samples to
   investigate potential contaminations during the procedure and measured
   together with solvent blanks at the beginning and end of each sequence.
   For system suitability a mixture of reference standards (2 μM, 159
   compounds, [250]Table S1) was measured at the beginning and the end of
   the sequence, and for monitoring potential carry over between samples,
   solvent blank injections (ACN/H[2]O, v/v, 1:1) were measured. A pooled
   QC prepared from all samples was injected at the beginning, the end and
   after every five samples and analyzed throughout the sequence and were
   used together with internal standards (^13C[6]-bisphenol A,
   ^2H[4]-genisteine, 2 μM) for monitoring potential analytical deviations
   during the measurement. An SRM 1950 sample (metabolites in frozen
   plasma, NIST, Gaithersburg, USA) was extracted according to
   protocol[251]^107 and measured at the beginning and the end of a
   sequence.

   During the targeted LC-MS/MS measurements process blanks and solvent
   blanks were measured at the beginning and end of each sequence. A
   pooled QC, spiked with the reference standards (n = 126, SI:
   SI_MRMS_Targeted_LC_MSMS_Bile_acids), was injected at the beginning,
   the end and after every 20 samples, and was used together with an
   internal standard (^2H[4]-ursudeoxycholic acid, 2 μM) for monitoring
   potential analytical deviations during the measurement. The injection
   order of the samples during LC-HRMS and LC-MS/MS experiments was
   randomized to avoid any experimental group specific bias. The NTA Study
   Reporting Tool (SRT)[252]^106 was used as guidance for the construction
   of the materials and methods section to ensure high reporting quality.

Quantification and statistical analysis

Statistical analysis of the cohort characteristics

   Fisher’s exact test and Student’s t test followed by multiple testing
   correction with Bonferroni’s method, all performed via R (4.1), were
   used to analyze the relationship between cohort characteristics and
   experimental groups ([253]Table 1).

Data processing and integration of LC-MS acquired data

   Thermo Fisher LC-HRMS(/MS) raw data files from the untargeted
   metabolomics measurements were converted into mzXML files, centroided
   and separated into two files according to polarity using Proteo Wizard
   (3.0.18).[254]^95 All experimental sampling time points from their
   respective biological matrix were processed together. Data
   pre-processing was done using the R (4.1) package XCMS[255]^94 (3.14),
   starting with the import of the mzXML files. The next step was peak
   detection using the centWave algorithm (parameters: ppm=5,
   peakwidth=5/15; snthresh=10, prefilter=3/5000) followed by the removing
   of peaks with a too wide peak width. The retention time between samples
   were aligned employing the obiwarp algorithm (parameters: binSize=0.6,
   distFun="cor_opt") with a previous alignment of pooled QC samples
   followed by the alignment of the samples. The detected peaks were
   grouped to features with a minimum threshold of 40% in at least one
   experimental group (parameters: bw=1, minFraction=0.4, minSamples=5).
   Features missing a peak in certain samples were integrated in the
   respective retention time width of the feature. The resulting feature
   list was further processed by the R package CAMERA (1.48)[256]^96
   (parameters: perfwhm=0.6, ppm=5), creating adduct and isotopologue
   annotations. Features annotated as isotopologues and features with an
   adduct annotation different from [M+H]^+ or [M-H]^- were removed from
   the feature list. In addition, features with a signal in the process
   blank or solvent blank higher than 20% of the median signal in the
   pooled QC were excluded. All features eluting in the void volume were
   also removed from the feature list (HILIC: 1 min, RP: 0.65 min). The
   Python-based package NeatMS (0.9),[257]^97 based on neuronal
   networking, was used to classify the peak quality. Features mainly
   consisting of the peak label “Noise”/“Low Quality” were filtered out.
   For each sampling time point a subset of the feature list was created
   and further processed. Features with more than 20% missing values were
   removed and the missing values were imputed by using half of the
   minimum value. Fold changes (log[2]) between groups were determined for
   each time point, and unknown features found in the untargeted data
   within a threshold of fold change log[2] < |2| were considered for
   further MS2 experiments, annotation and following analysis. Feature
   annotation was carried out by in silico fragmentation employing SIRIUS
   4[258]^98 (parameter: 5 ppm; possible Ionizations [M-H]^-, [M+Cl]^-,
   [M+H]^+, [M+Na]^+; DB: Bio database), based on the received score and
   manual reviewing, and was expanded by comparing experimental MS^2
   spectra with spectral libraries using GNPS[259]^101 (parameters: Score
   Threshold: 0.7, Precursor Ion Mass Tolerance: 0.05, Min Matched Peaks:
   5) whenever possible. Features annotated by SIRIUS employing
   fragmentation based molecular formula prediction were assigned an
   identification confidence level (ICL) of 4, features annotated by in
   silico fragmentation an ICL of 3, features with an unambiguous match in
   a spectral library with an ICL of 2.a, while compounds with an identity
   confirmed by reference standards, were assigned an ICL of 1.[260]^40 In
   addition to the untargeted evaluation, a targeted suspect screening of
   compounds ([261]Table S2) with available reference standard and/or
   compounds of high interest based on previous measurements was done
   using the software Skyline (20.2).[262]^99 Fold changes (log[2])
   between groups were determined and corresponding p values were
   calculated using Wilcoxon Rank-sum test. The raw files obtained from
   the LC-MS/MS experiments targeting bile acids and their conjugates were
   also processed and quantified by Skyline (20.2). Fold changes (log[2])
   between groups were determined and corresponding p values were
   calculated using Wilcoxon Rank-sum test. FDR was controlled using the
   ‘qvalue’ (2.28)[263]^100^,[264]^108 R package with a cutoff of <0.2.
   For a putative Gene Set Enrichment Analysis (GSEA) and
   over-representation analysis (ORA, mummichog[265]^26) the ‘MS Peaks to
   Pathways’ tool from the web application MetaboAnalyst (5.0)[266]^102
   was used (parameters: p value cutoff: 0.1, mummichog+GSEA, min. 5
   entries, MFN DB). For this purpose, the datasets of the HILIC and the
   RP chromatography and positive and negative polarity mode were combined
   to one dataset for each time point. The activity network was created
   using the Python (3.8) package mummichog (version 2)[267]^26 and was
   visualized using Cytoscape (3.9).[268]^109 Multivariate statistical
   analysis (PCA) was done using the R package ‘MixOmics’ (6.20)[269]^103
   on the auto-scaled and combined data of both positive and negative
   polarity of RP and HILIC chromatography. Correlation analysis was done
   using the R package ‘Hmisc’ (4.7)[270]^105 based on Spearman’s
   correlation and for the PERMANOVA and dispersion analysis the R package
   ‘vegan’ (2.3)[271]^104 was used.

Acknowledgments