Thirteen preterm infants (≤32 weeks of gestation and/or ≤1,500 g of weight) of 31 recruited in a previous study (36) born at the Hospital Universitario 12 de Octubre of Madrid (Spain) were randomly selected for this study. Preterm infants with malformations, metabolic diseases, or severe conditions were excluded. Relevant demographic and clinical data such as antibiotherapy, enteral and parenteral nutrition, use of nasogastric tube, need of mechanical ventilation, hospital stay, birth weight, gestational age, gender, or delivery mode are described in Tables 1 and 2.

Following the routine NICU feeding protocols, all infants were preferably fed with their OMM and, when this was not possible, with pasteurized human milk from the Milk Bank Unit DM. When the weight of the infants was ≥1,500 g and both types of human milk were unavailable, they received adapted preterm formula. In general, the feeding patterns of the recruited infants were very heterogeneous, a fact that prevented the formation of well-defined feeding groups.

Continuous nasogastric feeding was ordinarily administrated with a pump and intermittent feeding by gavage or pump. The syringe barrels were used as reservoirs that were connected through an external feeding tube (EFT). Feeding tubes were routinely replaced every 24 h, which means that different feed types could pass through the same tube during such a period.

None of the infants received antiacid treatment during their participation in the study.

Gastric content samples (~2 mL) were collected weekly by aspiration, using a sterile syringe, through the NEFT inserted into each baby’s stomach before new milk administration when intermittent feeding.

The feeding (OMM and/or DM) samples collected were the last fraction obtained after their passage through the EFT, immediately before entering the nasogastric tube at the connector, as previously reported (36).

All samples were collected 12 h after the replacement of the nasogastric feeding tubes and stored at −20°C until analysis.

Adequate dilutions of GC samples were spread onto Man, Rogosa and Sharpe (MRS; Oxoid, Basingstoke, UK) and MRS supplemented with l-cysteine (0.5 g/L) (Sigma, St. Louis, MO, USA) (MRScys) for isolation of lactic acid bacteria, MacConkey (BioMérieux, Marcy l’Etoile, France) for isolation of Enterobacteriaceae, Baird Parker (BioMérieux) for isolation of staphylococci, Sabouraud Dextrose Chloramphenicol (SDC; BioMérieux) for isolation of yeasts, and Brain Heart Infusion (BHI; Oxoid), Wilkins-Chalgren (WC; Oxoid), and Columbia Nalidixic Acid Agar (BioMérieux) as general media for isolation of other bacterial groups. Plates were aerobically incubated at 37°C for up to 48 h, with the exception of SDC plates, which were incubated at 32°C for 96 h, and WC and MRScys plates, which were incubated (85% nitrogen, 10% hydrogen, and 5% carbon dioxide) in an anaerobic workstation (Mini-MACS Don Whitley Scientific Limited, Shipley, UK) at 37°C for 48 h.

Bacterial identification was performed as previously reported (36). The identifications were confirmed by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (Vitek MS, Biomerieux) at the facilities of ProbiSearch, SL. (Tres Cantos, Spain).

The genetic diversity of all the isolates belonging to the species E. faecalis, S. aureus, and K. pneumoniae was assessed by pulsed-field gel electrophoresis (PFGE) in a CHEF DR II apparatus (Bio-Rad, Birmingham, UK). To separate SmaI-digested fragments of enterococci and staphylococci, different protocols were applied (2–28 s for 24 h and 5–15 s for 10 h, and then 15–60 s for 13 h). The chromosomal DNA of K. pneumoniae isolates was digested with XbaI enzyme, and the electrophoresis conditions were 1–40 s for 20 h. The analysis of PFGE profiles was performed using the UPGMA method based on the Dice similarity by the Phoretix 5.0 software.

Multilocus sequence typing (MLST) schemes were applied for PFGE-unrelated strains of E. faecalis and S. aureus,¹ E. coli,² and K. pneumoniae.³

Gastric content samples were thawed at room temperature and centrifuged at 13,000 rpm and 4°C for 10 min. Then, the pellets were washed with TE buffer and centrifuged under the same conditions. DNA extraction protocol was carried out as previously described (23). To detect similarities and, more interesting, differences between GC and infant’s feeding, the bacterial composition of some OMM and DM samples was also assessed. DNA extraction of those samples was made following the same protocol.

PCR amplifications were performed using primer 27F-DegL (5′-GTTYGATYMTGGCTCAG-3′) in combination with an equimolar mixture of two reverse primers, 338R-I (5′-GCWGCCTCCCGTAGGAGT-3′) and 338R-II (5′GCWGCCACCCGTAGGTGT-3′), generating ~345 bp amplicons from the V1 to V2 hypervariable regions of 16S rDNA genes. Barcodes used for Illumina sequencing were appended to 3′ and 5′ terminal ends of PCR amplicons to allow separation of forward and reverse sequences. Subsequently a bioanalyzer (2100 Bionalyzer, Agilent) was used to determine the concentration of every sample in the region of interest.

Barcoded PCR products from all samples were pooled at approximately equal molar DNA concentrations and run on a preparative agarose gel. The correct sized band was excised, and the DNA was purified. One aliquot of pooled, purified, barcoded DNA amplicons was sequenced on an Illumina MiSeq pair-end 2 bp × 250 bp protocol (Illumina Inc., San Diego, CA, USA) at the Unidad de Genómica of the Fundación Parque Cientifico de Madrid (Spain).

Raw sequences were processed according to their quality using the program TRIMOMMATIC by filtering those reads that showed a window of 50 bp with an average of quality values below 25. MOTHUR v 1.33.0 and UCHIME programs were used to eliminate chimeras and ambiguous bases. Resulting reads of quality controls were assembled and classified taxonomically by comparison with databases Greengenes, Ribosomal Database Project, and SILVA using a Bayesian classification method and a level of similarity of at least 97%.

The statistical analysis was performed using R 2.15.3 (R-project).⁴ When data were not normally distributed, median and interquartile ranges (Q1 and Q3) were calculated for all sampling times, and mean and 95% confidence interval (95% CI) were used for normal distributed data. The Kruskal–Wallis test for non-normal data or one-way ANOVA test when data were normally distributed were used to evaluate the differences between sampling times. In all cases, P values of <0.05 were considered to be significant.

The 13 infants enrolled in this study had a mean gestational age of 28 weeks (ranging from 24 to 32 weeks) and a mean birth weight of 1,192 g (from 720 to 2,030 g) (Table 1). Approximately half of the infants (n = 7) were born by cesarean section (Table 1). All of them, except one, received antibacterial prophylaxis for, at least, the first 3 days of life and eight infants needed mechanical ventilation (Table 2). All infants were fed either with their OMM, DM, and/or preterm formula by nasogastric feeding tube for, at least, 21 days after delivery (mean of 62 days) (Table 2).

A total of 38 GC samples collected during the first month of life were analyzed by culture-based methods: 6 samples from the first day of life (day 0), 8 from the first week of life (day 7), 11 from the second week of life (day 14), 11 from the third week of life, and 2 from the fourth week of life (day 28). Samples were very heterogeneous in texture and color, ranging from milky to mucous.

No microorganism could be isolated from 25% of the samples (particularly from those collected at day 0). When bacterial growth was detected, total bacterial counts oscillated between 4.20 and 5.95 log¹⁰ CFU/mL at day 0 and between 6.00 and 8.13 log¹⁰ CFU/mL at day 28 after birth in BHI media (Table S1 in Supplementary Material). The dominant bacterial genera detected were Enterococcus, Staphylococcus, and Lactobacillus, among Gram-positive bacteria, and Klebsiella, Serratia, and Escherichia, among Gram-negative ones. All of them showed a trend to increase, in both concentration and frequency, from birth onward, but only the genus Enterococcus showed a statistically significant increase during the study period (P = 0.027) (Table 3).

A total of 241 isolates, belonging to 22 different species, were identified. The number of species per sample ranged from 1 to 17 species, and the bacterial profiles showed a high interindividual variability (Figure 1). Globally, the dominant species were E. faecalis and Staphylococcus epidermidis, followed by Serratia marcescens, E. coli, and K. pneumoniae.

MLST analysis revealed the presence of five different ST clones among the E. faecalis isolates. Most of the infants (67%) were colonized by the clone ST 64, while clone ST 56 was detected in two infants (6 and 9) and clones ST21, 34, and 40 only in one infant each (2, 7, and 10, respectively) (Figure 1). All these clones belonged to high-risk clonal complexes and are associated with hospital environments. In relation to S. aureus, ST378 and ST34 were isolated in GC of only two infants (1 and 2, respectively), but both clones are also associated with high-risk clonal complexes. The ST393, ST1978, and ST69 of E. coli were detected in gastric samples from four infants; two of them were colonized by the same clone (ST393). Finally, clones ST641 and ST505 of K. pneumoniae were detected in GC of two infants (3 and 11, respectively), and, again, both clones belonged to high-risk clonal complexes (Figure 1).

DNA of good quality was obtained from 15 gastric samples of 13 infants. Two of them from the first week after birth (day 0), four from the second week of life (day 14), eight from the third week of life (day 21), and one from the fourth week of life (day 28). After processing the quality of the reads, a total of 3,613,468 sequences were analyzed that represented an average of 240,898 ± 32,493 sequences per sample. A high interindividual variability was detected among the samples. The Shannon Index ranged between 1.158 and 2.563 independently to the sample collection time with a mean of 1.863 ± 0.476 (Figure 2).

Three phyla predominated in the analyzed samples, Proteobacteria (49.61% ± 14.69), Firmicutes (30.11% ± 8.86), and Actinobacteria (15.66% ± 9.44) (Figure 3). Other phyla were detected but with a relative abundance below 1%. Similarly, to the culture results, the most abundant genera observed in GC samples with 16S rRNA sequencing were Corynebacterium, Streptococcus, Staphylococcus, Lactobacillus, Enterococcus, and Serratia. However, 11 other genera represented that ≥1% of the communities were observed across samples (Figure 4). The genera Serratia, Klebsiella, and Streptococcus were most abundant in GC with a median relative abundance of 15.68, 9.59, and 9.07%, respectively (Table 4).

Gastric content samples were collected through the nasogastric tubes by syringe aspiration. Since this route was shared with OMM or DM samples, analyses of feeding samples by sequencing the 16S rRNA gene were also performed. Globally, a high degree of similarity between GC and milk samples was observed at the genus level, although the bacterial pattern of OMM samples were more similar to GC than DM (Figure 4). However, the statistical analysis did not show any significant difference (Table 4) between these three types of samples.