Ethics
Written informed consent was obtained before specimen collection from all enrolled mothers after a detailed consultation. All aspects of recruitment as well as collection, handling, processing and storing of samples and data were approved by the Luxembourg ethics board, the Comité national d’éthique de recherche, under reference number 201110/06 and by the Luxembourg National Commission for Data Protection under reference number A005335/R000058.
Clinical metadata
All study participants were enrolled and gave birth at the Centre Hospitalier de Luxembourg (CHL). Exclusion criteria for mother–neonate pairs included the administration of antibiotics to neonates immediately postpartum, birth prior to 34 weeks of gestation, and maternal gestational diabetes. Clinical metadata for all the analysed time points (days 1, 3 and 5 postpartum) are listed in Supplementary Data 1. Recorded metadata include information on the delivery mode, classification of caesarean section as elective or emergent, birth weight, gestational age, identification of the neonate as small for gestational age (SGA status) where relevant, gender, body length, weight and feeding regime. If a neonate received formula milk at any collection time point, the neonate was considered having received combined feeding for the remainder of the study, as even short-term formula feeding has been shown to cause profound and long-lasting shifts in the gastrointestinal microbiome composition53. Enrolled pairs of mothers and neonates (n = 16 pairs) included one twin birth (C115 (CSD) and C116 (CSD + SGA)).
Sample collection
Neonatal faecal samples were collected during the first 24 h as well as at days 3 and 5 after birth. Samples and data were collected at the CHL until day 3 after birth; subsequent samples were collected at home by trained study nurses. From the 33 neonates that were recruited into the study, the gut microbiome of 15 (Supplementary Data 1) had previously been characterized using a combination of 16S rRNA gene amplicon sequencing and quantitative real-time PCR7. For a subset of neonates, the mother was sampled additionally. Maternal samples (vaginal swabs and faeces) were collected less than 24 h before delivery. Samples were collected into sterile plastic tubes, immediately flash-frozen in liquid nitrogen and stored at −80 °C until further processing. Neonatal blood was collected by capillary or venous sampling, and plasma was isolated and stored at −80 °C from 31 healthy neonates (13 VD, 13 CSD, five CSD + SGA) at day 3 (28 samples) or day 5 (three samples) after birth, including 15 of the 16 neonates for whom metagenomic data were analysed (six VD, four CSD, five CSD + SGA), two neonates for whom no metagenomic but 16S rRNA gene amplicon sequencing data were available (two CSD) and 14 neonates (seven VD, seven CSD) that were sampled under the same conditions7 (Supplementary Data 1). Clinical data were stored on secure servers at the Luxembourg Centre for Systems Biomedicine (LCSB), and biological samples were stored until further processing at the Integrated BioBank of Luxembourg (IBBL), which is NF S96-900:2011 certified.
Sample processing and extraction of nucleic acids
Genomic DNA was isolated from vaginal swabs with the PowerSoil DNA isolation kit (MO BIO Laboratories; Antwerp, Belgium) with an additional step to increase extraction yield involving the incubation of the samples in PowerSoil tubes with solution C1 at 65 °C for 10 min prior to homogenization for 5 min at 20 Hz in an Oscillating Mill MM 400 (Retsch, Haan, Germany). DNA was subsequently extracted following the manufacturer’s instructions.
Faecal samples and cell-culture pellets were processed with the Powerlyzer PowerSoil DNA isolation kit (MO BIO Laboratories), optimized for low-yield samples. Bead solution (500 µl), C1 (60 µl), UltraPure™ Phenol:Chloroform:Isoamyl Alcohol (25:24:1, v/v; Invitrogen, Aalst, Belgium; 200 µl) and 50 mg neonatal stool or 150 mg maternal stool were added to a dry glass bead tube, incubated at 65 °C for 10 min, and homogenized by milling for 45 s at 4 m s−1 in a FastPrep-24 5 G (MP Biomedicals, Illkirch-Graffenstaden, France). Samples were centrifuged for 1 min at 12,000 g. Solutions C2 (250 µl) and C3 (100 µl) were added to the supernatant and incubated at 4 °C for 5 min, centrifuged for 1 min at 12,000 g, then 700 µl of solution C4 and 600 µl of 100% ethanol were added to the supernatant and mixed. 650 µl were loaded onto a Spin Filter and centrifuged at 10,000 × g for 1 min. This step was repeated until all lysate had passed through the filter. For the higher input-mass maternal faecal samples, the same isolation procedure was followed except that the filters were washed with a mix of 300 µl solution C4 and 370 µl 100% ethanol, with centrifugation at 10,000 × g for 1 min. This latter step was omitted for the low input neonatal samples. All filters were washed with 650 µl 100% ethanol, then 500 µl solution C5. After drying, 60 µl solution C6 was added to the centre of the filter and incubated at room temperature for 5 min. DNA was eluted by centrifugation at 10,000 × g for 30 s. RNase A (100 µg ml−1, 2 µl) was added and incubated at 37 °C for ≥ 30 min. Then, one-tenth volume 3 M sodium acetate (pH 6.8) and two volumes isopropanol were added to precipitate the DNA on ice prior to centrifugation. The pellet was washed with 150 µl 70% ethanol, before the dried DNA was dissolved in 50 µl (neonatal faecal samples) or 100 µl (maternal faecal samples) RNase-free water. To obtain an artefact control sample, DNA was extracted from 800,000 trypsinized Caco-2 cells/ml. Caco-2 cells were grown in Dulbecco’s Modified Eagle’s Medium (Thermo Fisher Scientific, Ghent, Belgium) containing 20% v/v foetal bovine serum and 1% penicillin–streptomycin (Invitrogen) to prevent microbial growth. DNA was extracted with the low-biomass protocol described above, subsequently titrated and samples with 480, 240, 120, 60 and 30 ng total mass were sequenced. DNA integrity and quantity were determined for extracted samples of all origins on 1% agarose gels and in a Qubit 2.0 fluorometer (Thermo Fisher Scientific). Extracted DNA was stored at −80 °C until further use.
DNA sequencing
All DNA samples (along with 8 controls) underwent standard amplicon sequencing of the V4 region of 16S rRNA genes using primers 515F- 5′-GTGBCAGCMGCCGCGGTAA-3′ and 805R- 5′-GACTACHVGGGTATCTAATCC-3′ at the Center for Analytical Research and Technology–Groupe Interdisciplinaire de Génoprotéomique Appliquée (CART-GIGA; Liège, Belgium). Selected DNA samples of maternal (vaginal and faecal extracts), neonatal (faecal extracts at days 1, 3 and 5) and cell-culture origins were subjected to random shotgun sequencing (Supplementary Data 1). Metagenomic libraries were constructed with an optimized low-quantity DNA library preparation kit and sequenced on a HiSeq 2500 platform (Illumina) at GATC Biotech (Konstanz, Germany). For neonatal samples collected from C105, C109, C110 and C119 metagenomic libraries were prepared using TruSeq DNA Nano kit (Illumina) and sequenced on a NextSeq 500 platform (Illumina) at LCSB Sequencing Platform. A total of 84% of the study samples (63 of 75) collected from the mother–neonate pairs yielded sufficient DNA for metagenomic sequencing and sufficient artefact-curated metagenomic data for subsequent analyses.
Metagenomic data processing
Metagenomic data sets were processed with the Integrated Meta-omic Pipeline (IMP; version 1.3), which performs pre-processing, assembly, functional annotation of predicted genes and downstream analyses of Illumina next-generation sequencing metagenomic data in a single, reproducible workflow28. Illumina TruSeq3-PE-2 adapter sequences were trimmed from the reads in the pre-processing step (including the removal of human reads), and the de novo assembly step used the MEGAHIT54 metagenome assembler. The IMP parameters were customized for different sample types: default parameters were retained for maternal faecal samples; for low-biomass samples (maternal vaginal swabs, neonatal faecal samples from days 1, 3 and 5 and cell culture sample), the integrated VizBin29 sequence cut-off length was set to 1.
Curation of metagenomic data from artefacts
To identify and exclude artefactual sequences in the low biomass samples, contigs were assembled from the sequencing reads obtained from the DNA extracts of the Caco-2 cells after the removal of human reads. Given that the Caco-2 cells were cultured in the presence of 1% penicillin–streptomycin, that the routine surveys for Mycoplasma were negative, and that the metagenomic sequencing data did not include any Mycoplasma sequences, any bacterial contamination of the mammalian cell culture could be confidently excluded. Then, metagenomic reads from each study sample were mapped against these contigs using Bowtie 255 (version 2.0.2). Matching sequences were excluded prior to taxonomic profiling of metagenomic reads by phylogenetic markers35. As the artefactual sequences identified in the control samples did not represent full genomes, we further used a binning-based approach to identify additional potential artefactual sequences of the same organism among the de novo assembled contigs of the study samples. After removing the rRNA sequences from the contigs56, we performed joint binning of control cell-culture contigs with each of the samples’ contigs individually using VizBin29 without any length cut-off. Bins were identified based on VizBin embeddings56 using density-based spatial clustering of applications with noise (DBSCAN), without correction for the depth of coverage and completeness. All distinct bins (total length < 10 Mbp) that contained > 0.01% of the total contig length of the cell-culture control sample were considered putative reconstructed genomes of artefactual DNA, and the corresponding contigs were removed from the study samples in silico.
Functional profiling
Genes were predicted from contigs assembled with IMP and, after removing artefactual contigs, these genes were functionally annotated with hidden Markov models (HMMs)56 trained for all KO57 groups. The functional KO HMMs were aligned using HMMER 3.158,59. The best hit KO (if multiple KOs could be assigned to a gene, the KO with the highest bit score was chosen) for every gene was assigned if the bit score was higher than the binary logarithm of the number of target genes. The FeatureCounts60 tool with arguments –p and –O was used to extract the number of reads per KO (Supplementary Data 4; representing mean ± standard deviation 77 ± 13 % of all mapping reads).
Linking genome reconstructions by marker gene sequence homology
The curated contigs were binned based on the VizBin embeddings using DBSCAN as well as correction by the depth of coverage and completeness56. The reconstructed genomes of all samples belonging to a mother–neonate pair were merged into a union set. For each sample set, predicted amino acid sequences were searched against and annotated using a defined set of essential marker genes61 using HMMER 3.158. Protein sequences assigned to 35 specific marker genes that form the cross-section of previously suggested sets of phylogenetic marker genes61,62 were selected. These marker amino acid sequences were clustered with CD-HIT63 at 97.5% identity. The frequencies of genes from different genome reconstructions co-occurring in the same clusters were determined. A simple graph network representation was constructed with the reconstructed genomes as nodes and counts of co-occurrences between two reconstructed genomes as weighted, undirected edges. Highly interlinked sub-networks, representing related reconstructed genomes, were detected with the cluster_fast_greedy algorithm64 implemented in the R package igraph (v.1.0.1). The resulting reconstructed genomes from a given sub-network were manually inspected, and the taxonomy of reconstructed genomes was assigned using PhyloPhlAn30 (Supplementary Data 5).
Strain-level analysis
Strains that occurred in multiple samples were determined with StrainPhlAn31, using the pre-processed sequencing read data and reconstructed genomes. For each sample, taxonomic profiles were generated from pre-processed reads with MetaPhlAn265 using default settings. Strain reconstructions were extracted with the sample2markers.py script in StrainPhlAn with default arguments. StrainPhlAn was used to extract the clades detected in all samples and to construct reference databases for each clade. The sample-based strain reconstructions and reference databases of each clade and all reconstructed genomes were analysed with StrainPhlAn to build multiple sequence alignments and phylogenetic trees. The neonatal samples were considered to share strains with maternal samples if the cophenetic distance between the neonatal microbiome read-based or reconstructed genome-based markers and the maternal markers was less than the distance to the markers of any other individual. Trees were visualized with GraPhlAn (https://bitbucket.org/nsegata/graphlan/wiki/Home). To visualize the positions of markers in genome reconstructions, the reference markers of the species assigned to the reconstructed genomes in StrainPhlAn were aligned to the genome reconstructions post hoc, using blastn and an E value cut-off of 1 × 10−10, as in StrainPhlAn.
Fixation index and intra-population diversity calculation
For all neonatal reconstructed genomes that were estimated to be > 65% complete and linked to at least one other sample of the same neonate or their mother, the fixation index (FST) and the intra-population diversity (π) were assessed by the presence of SNVs. Metagenomic sequencing reads were mapped against the reconstructed genomes using MOSAIK66 (version 2.2), with default parameters. A minimum alignment identity of 95% was applied to restrict the mapping to reads of the same species67. Genome–sample combinations generating alignments with a median coverage < 20X and/or a breadth < 40% were not included in downstream analyses. To reduce bias stemming from variation in coverage, alignments were down-sampled to a median coverage of 20X using Picard tools (version 1.85; http://broadinstitute.github.io/picard/). SNV calling was performed with FreeBayes68 (version 1.1.0) using the -pooled-continuous option on the merged alignment files containing all samples for the same genome. Potential SNVs were required to be supported by four or more reads and to have an allele frequency ≥ 1%.
The output from FreeBayes (VCF-file) was used as input for POGENOM (https://github.com/EnvGen/POGENOM), a Perl-based tool that enables population-genomic analysis of metagenome samples. POGENOM was used to calculate the intra-population nucleotide diversity (π), which is defined as the average number of nucleotide differences per site between any two sequence reads chosen randomly from the sample population (0 ≤ π < 1). When reads of two or more samples mapped with sufficient coverage to the same genome, the fixation index (FST) was calculated, reflecting the population differentiation between a pair of samples. FST is defined as one minus the average intra-population diversity of the samples divided by the nucleotide diversity between the samples (inter-population diversity). POGENOM was tuned to include only the loci recovered in all samples mapped to the same reference genome, assuring a valid comparison of the intra-species variation.
Processing of amplicon sequencing data
Analysis of the 16S rRNA gene amplicon sequences was performed with NG-Tax27, with default parameters. Operational taxonomic units (OTUs) were assigned to the taxonomy in an open reference approach, using USEARCH69 against the SILVA70 16S rRNA gene amplicon reference database (version 128; Supplementary Data 6). To exclude sequencing artefacts, only dominant phylotypes were examined by removing OTUs that were represented by fewer than 10 reads in the study samples.
Analyses of taxonomic profiles
To determine the Gram staining of the bacteria, we used the NCBI microbial attributes, which can be downloaded from http://www-ab2.informatik.uni-tuebingen.de/megan/taxonomy/microbialattributes.zip. Final Gram staining was assessed by main staining trends per genus and manually curated at the family and order levels. Functional community profiles were predicted based on OTU abundances using PanFP34.
Statistical data analysis
The R statistical software package (version 3.3.3) was used for statistical analyses and visualization of the taxonomic profiles derived from metagenomic and amplicon sequencing. Sum normalization and calculations of taxon richness (number of metagenomic OTUs (mOTUs) for metagenomic data or OTUs for amplicon sequencing data), diversity (Shannon), evenness (Pielou) indices and Spearman correlation coefficients were performed using the vegan R package. To discover differences in the data sets between the birth modes at the different collection time points postpartum, Wilcoxon rank-sum tests were applied, with FDR multiple-testing adjustment if applicable. Differential taxonomic abundances (according to delivery mode) were also calculated using ANCOM71 with Benjamini-Hochberg multiple testing correction at 0.05 false-discovery rate. To determine the effect of the variables within the metadata, differentially abundant taxa were also determined using MaAslin32 with default parameters and a q < 0.05 threshold for multi-testing correction. The model used was genus ~ sampling day + maternal antibiotic intake + feeding regime + gestational age. Differential analysis of KO abundance, comparing VD to CSD and VD to CSD + SGA with a linear model, which considered the different collection time points containing at least 1000 KOs (days 3 and 5) as covariates, was performed with the R package DESeq2 version 1.10.133. KOs were considered significantly differentially abundant in VD and CSD (±SGA) if the FDR-adjusted P value of the Wald test was < 0.05 for at least one comparison (CSD vs. VD or CSD + SGA versus VD) and the directionality of change in both comparisons was the same. Principal coordinate analysis (PCoA) graphs were generated using Jensen-Shannon distances as implemented in the R package phyloseq72. Differentially abundant pathways were detected through pathway enrichment analysis using a custom R script56. Tests for the enrichment of reconstructed genomes with differentially abundant KOs were performed using Fisher’s exact test and FDR-adjustment for multiple testing in R.
LPS isolation from neonatal faecal samples
LPS was isolated from 16 selected neonatal faecal samples on the basis of availability of sufficient material. Samples (7 VD, 7 CSD, 2 CSD + SGA; Supplementary Data 12) were collected on day 3 after birth, and from overnight cultures of Escherichia coli strain K-12 (sub-strain MG1655). To maximise yields, LPS was purified from three aliquots of 50 mg of each neonatal faecal sample using the hot phenol–water method73 and further purification was performed using a modified phenol re-extraction protocol74. For the E. coli control samples, three 5 ml overnight cultures were diluted to an optical density (600 nm) of 0.5 and centrifuged. LPS was isolated from cell pellets by the same protocol as above. LPS for each individual was pooled and quantified using an ELISA-based endotoxin detection assay (Endolisa; # 609033, Hyglos GmbH, Germany). From the 16 neonatal faecal samples, 11 produced measurable amounts of LPS, whereas 5 were under the detection limit (Supplementary Data 12). An extraction blank was generated using the same LPS isolation protocol.
Quantitative real-time PCR to determine bacterial loads
DNA from all neonatal faecal samples used for LPS isolation was diluted (when applicable) to a concentration of 1 ng l−1 and amplified in duplicates with universal prokaryotic 16S rRNA gene primers 926F and 1062R75 and with specific Escherichia coli primers Ec461F and Ec780R76 . Primer sequences, annealing temperatures and cycle details are specified in Supplementary Data 14. Genomic DNA isolated from Salmonella Typhimurium LT2 and E. coli strain K-12 (sub-strain MG1655) was used to prepare standard curves for universal prokaryotic and specific E. coli primers, respectively. Reaction mixture, measurements and calculations of bacterial load (nanograms bacterial DNA per milligram stool and nanograms E. coli DNA per milligram stool) were performed as previously described7 (Supplementary Data 14). The proportion of E. coli DNA in comparison to total bacterial DNA was subsequently calculated.
In vitro immunostimulation using LPS from neonatal faecal samples
Primary human monocytes were isolated from blood samples obtained from the Luxembourg Red Cross originating from twelve healthy adult donors. Human neonatal dendritic cells (DCs) were previously shown to be competent in MHC class I antigen processing and presentation to the same extent than adult DCs77. Most importantly, the NF-κB-dependent pathway in TLR-4 signalling is intact in neonatal MoDCs as they produce pro-inflammatory cytokines upon LPS stimulation, while adult and neonatal DCs are both able to produce comparable levels of TNF-α, IL-6 and IL-8 in response to LPS78. Isolated monocytes were differentiated into dendritic cells (MoDCs) in 12-well plates for 5 days in RPMI 1640 medium (Thermo Fisher Scientific) supplemented with 10% foetal bovine serum (Thermo Fisher Scientific), 20 ng ml−1 each of granulocyte-macrophage colony-stimulating factor (Peprotech, London, UK), 20 ng ml−1 IL-4 (Peprotech) and 1% penicillin–streptomycin (Invitrogen). To assess the immune stimulatory potential of isolated LPS, we treated MoDCs for 24 h with LPS extracted from VD or CSD (±SGA) neonatal faecal samples using two different methods; one based on LPS volume and one based on the normalization of LPS concentration with the bacterial load (see below for more information).
As we started from the same amount of material for all the neonatal stool samples and used the exact same extraction protocol to isolate all LPS fractions for all samples, we assumed that if we treated MoDCs from the same donor with the exact same volume of yielded LPS (independent of the concentration of LPS present), we would realistically emulate the microbial LPS load which immune cells would be exposed to in vivo and thus be representative of the immunostimulatory potential of a given sample at 3 days postpartum. To stimulate MoDCs, 7.5 µl of LPS extract per 105 MoDCs was added per well. For the negative control, MoDCs were incubated with 7.5 µl of LPS extraction blank, and for the positive control, MoDCs were treated with 15 endotoxin units (EU) LPS isolated from E. coli cultures. MoDCs were treated for 24 h to assess the immunostimulatory potential of the isolated LPS. Treatments were performed in duplicates and tested on at least three different donors. Culture supernatants from stimulated MoDCs were diluted 1/10 (or 1/50, if above standard curve range) and analysed for the presence of TNF-α using a commercial ELISA reagent set (Human TNF alpha uncoated ELISA, Life Technologies, Belgium) and a microplate reader (Biotek instruments, Germany).
For the second method, we verified our results using the bacterial load for normalizing LPS concentration values. Naturally, all faecal samples have a different bacterial load within the 150 mg of starting material that is used to isolate LPS. In order to assess if the differences that we observed before with equal volumes of LPS (see above) were due to the fact that some samples have a much lower bacterial load or if also the bacterial composition (and proportion of Gram-negative bacteria) plays a role in the immunostimulation, we normalized the amount of LPS used to stimulate MoDCs with the bacterial load. For example, the bacterial load was highest for VD neonate C105 (Supplementary Data 15; Supplementary Fig. 14), and the corresponding bacterial load was 51.5 µg DNA per 150 mg stool. Therefore, this load was divided by the load in each other sample to yield a normalization factor. To stimulate MoDCs with 100 EU of LPS, 2.51 µl C105 LPS was added. For other samples, the LPS load was calculated by multiplying 2.51 µl by the previously determined bacterial normalization factor. For the negative control, 2 × 105 MoDCs were incubated with 15 µl of LPS extraction blank, and for the positive control 2 × 105 MoDCs were treated with 100 EU LPS isolated from E. coli cultures. Treatments were performed on cells from four distinct MoDCs donors (2 × 105 MoDCs/donor), except for LPS isolated from C120, which was only sufficient to stimulate donor 4’s MoDCs in duplicate. MoDCs and isolated LPS samples were incubated for 24 h. Culture supernatants from stimulated MoDCs were diluted twofold and analysed for the presence of seven cytokines (CXCL8/IL-8, IL-1β, IL-6, IL-10, IL-12p70, IL-18 and TNF-α) using a Human Premixed Multi-Analyte Kit (R&D Systems Europe; UK) and a MagPix multiplex reader (Luminex, Netherlands), according to the manufacturers’ instructions (Supplementary Data 16). Statistical significance between the different cohorts was determined using the Wilcoxon rank-sum test.
Coomassie blue and silver staining of LPS extracts
On the basis of availability of sufficient extracted LPS material, 0.5 µg of extracted LPS from the stool samples of two VD neonates (C007 and C111) collected on day 3 after birth, were prepared with Laemmli sample buffer (Bio-Rad, Belgium), heated for 5 min at 95 °C and separated on 12 % Bis-Tris precast gel (Bio-Rad, Belgium) at 200 V for 45 min. As positive controls, 0.5 µg, 1 µg and 10 µg of commercially available LPS (Escherichia coli O55:B5, gel-filtration chromatography; Sigma-Aldrich, Belgium) and 10 µg of E. coli protein extract were used. A precast gel was loaded with the LPS samples and stained with Coomassie (Imperial protein stain, ThermoFisher, Belgium) to check for protein contaminations. Silver staining of the gel was performed using a corresponding kit (SilverQuest, ThermoFisher, Belgium) according to the manufacturer’s instructions.
Ethidium bromide staining of LPS extracts
To check if LPS extracts were contaminated with immunostimulatory nucleic acids, 0.5 µg of extracted LPS from the stool samples of two VD neonates (C007 and C111), which presented highly concentrated LPS fractions that could be visualised on agarose gel, were prepared with DNA loading dye (ThermoFisher, Belgium) and loaded onto a 1% agarose gel. In addition, 0.5, 1 and 10 µg of commercially available LPS (Escherichia coli O55:B5, gel-filtration chromatography; Sigma-Aldrich, Belgium) were used to compare the purity of the LPS samples. As a positive control, 100 ng of E. coli DNA extract was used. The gel was stained with ethidium bromide, separated at 100 V for 50 min and analysed using a BioDocAnalyse system (Biometra, Germany). To check if nucleic acid contaminations could be identified in isolated LPS samples and would result in a TNF-α response, agarose bands were cut out (Supplementary Fig. 12) and purified using NucleoSpin Gel and PCR Clean-up kit (Macherey-Nagel, France). As controls, bands of E. coli DNA and commercially available LPS (10 µg) were cut out and purified. In addition, a purification blank was generated. The purified DNA fractions were used to stimulate MoDCs following the same protocol as for the stimulation with extracted LPS.
HEK-Blue™ cell assay
In order to verify the purity of the extracted LPS fractions, HEK-Blue™ reporter cell lines overexpressing one of the receptors hTLR2, hTLR4, NOD1 or NOD2 (InvivoGen, France), were stimulated with LPS extracted from five selected neonatal faecal samples (three VD and two CSD), which presented sufficient amounts of extractable LPS. HEK-Blue™ TLR and NOD cells are designed to detect stimulants of the human receptors by induction of secreted embryonic alkaline phosphatase (SEAP). For all the cell lines, the levels of SEAP were determined with HEK-Blue™ Detection (InvivoGen, France), a cell culture medium that allows for real-time detection of SEAP.
While the hTLR4 receptor only recognizes LPS, hTLR2 recognizes peptidoglycan, lipoteichoic acid and lipoprotein from gram-positive bacteria, lipoarabinomannan from mycobacteria, and zymosan from the yeast cell wall, the receptor NOD1 binds to bacterial molecules containing the D-glutamyl-meso-diaminopimelic acid (iE-DAP) moiety and NOD2 recognizes bacterial molecules (peptidoglycans) and stimulates an immune reaction. HEK-Blue™ cells were grown and maintained in DMEM (4.5 g l−1 glucose, l-glutamine, Sigma-Aldrich, Belgium), supplemented with 10% foetal bovine serum (Thermo Fisher Scientific), 1% penicillin–streptomycin (Sigma-Aldrich, Belgium), 100 µg ml−1 Normocin (InvivoGen, France) and respective selective antibiotics according to the user’s manual.
To monitor the activation of NF-κB, HEK-Blue™ cells were seeded according to the user’s manual in HEK-Blue™ Detection medium (InvivoGen, France), in flat-bottom 96-well plates and stimulated for 22 h with LPS samples. We used two conditions: first, using the same concentration of LPS, where 1 µl of extracted LPS (0.01 ng µl−1) was added per well, and second, using the same volume of LPS, where 7.5 µl extracted LPS was added to 105 HEK-Blue™ cells. To convert endotoxin activity (EU) into mass (ng), we considered that around 10 EU are equivalent to 1 ng endotoxin79. For positive controls, HEK-Blue™ NOD1 cells were stimulated with 1 µl TriDAP (10 µg µl−1, InvivoGen, France), HEK-Blue™ NOD2 cells with 1 µl Murabutide (10 µg µl−1, InvivoGen, France), HEK-Blue™ hTLR2 cells with 1 µl of Pam3CSK4 (1 µg µl−1; InvivoGen, France) and HEK-Blue™ hTLR4 cells with 1 µl ultrapure LPS (5 µg µl−1, source strain: ATCC 12014; CDC 5624-50 [NCTC 9701], InvivoGen, France). In addition, all cell lines were treated with 1 µl ultrapure LPS (5 µg µl−1, InvivoGen, France) and 1 µl ultrapure LPS (0.01 ng µl−1, InvivoGen, France) as well as with commercially available LPS (standard LPS; Escherichia coli O55:B5, gel-filtration chromatography; Sigma-Aldrich, Belgium): 1 µl of 5 µg µl−1 and 1 µl of 0.01 ng µl−1. For the negative control, HEK-Blue™ cells were incubated with 1 µl of endotoxin-free H2O (InvivoGen, France). All conditions were performed in duplicates and SEAP expression was monitored using a microplate reader at 655 nm (Biotek instruments, Germany) except for LPS isolated from C117 where only 7.5 µl extracted LPS/105 HEK-Blue™ cells was added to the cells and tested in duplicates.
Cytokine profiling of neonatal plasma samples
Plasma samples (n = 31) collected 3 or 5 days postpartum (13 VD, 13 CSD and five CSD + SGA; 28 samples collected at day 3, 3 samples collected at day 5 postpartum; Supplementary Data 1) were diluted twofold and analysed for 18 cytokines using a Human Premixed Multi-Analyte Kit (R&D Systems Europe) and a Bio-Plex analyser multiplex reader (Bio-Rad, Belgium), according to the manufacturers’ instructions. The kit is able to detect CXCL8/IL-8, IL-1β, IL-6, IL-10, IL-12/23 p40, IFN-β, IL-15, IL-21, IL-5, Galectin-1, IFN-γ, IL-18, IL-27, Granzyme B, IL-13, IL-2, IL-4 and TNF-α. Of these cytokines, 11 were above the detection limit (CXCL8/IL-8, IL-6, IL-10, IL-15, IL-21, Galectin-1, IL-18, IL-13, IL-2, IL-4 and TNF-α; Supplementary Data 17).
Code availability
All custom scripts written for this study are available online at https://git-r3lab.uni.lu/Cosmic/Earliest.