Mode of delivery and weight shape the intestinal microbiome composition progression in preterm infants: results of a prospective study

Background: In preterm infants the intestinal microbiome differs markedly from term infants. However, it is unclear whether the microbiome follows infant specific maturation patterns or whether it is mainly characterized by varying states of dysbiosis. We investigated the development of the intestinal microbiome in extremely preterm infants over time by 16S rRNA amplicon sequencing. We analysed the first meconium and faecal samples from the 2 nd , 3 rd and 4 th weeks, and (clinical) metadata to identify the main factors influencing the microbiota composition development. Results: The study included 41 extremely preterm infants (gestational age 25-30 weeks; birth weight (BW) 430-990g). Birth via Caesarean section (CS) was associated with placental insufficiency during pregnancy and lower BW. In meconium and in weeks 2 and 3 an increased combined abundance of Escherichia and Bacteroides (maternal aerotolerant fecal bacteria) was associated with vaginal delivery (p=0.039, p=0.0002, p=0.034, respectively) while Staphylococcus epidermidis (skin bacterium) was associated with CS (p=0.001, p=0.0003 p=0.048, respectively). Secondly, a switch was observed from a microbiome dominated by S. epidermidis (Bacilli) towards a microbiome dominated by Enterobacteriaceae (Gammaproteobacteria, mainly represented by Klebsiella and Escherichia ), in which the stage of progression appeared to be dependent upon the current weight of the infant, irrespective of the week of sampling or the mode of birth. Conclusions: Our data shows that the mode of delivery does affect the meconium microbiome composition. It also suggests that the weight of the infant at the time of sampling is a better predictor for the stage of progression of the intestinal microbiome development/maturation than gestational/postnatal age. We hypothesize that impaired growth, for example due the effects of diminished placental function during pregnancy, is a key factor in the maturation of the intestinal microbiome in extreme

The process of microbiota maturation, a pattern where microbiome maturation is mirrored by the maturation of the infant, is important in understanding differences between the microbiome of preterm-and term infants. In preterm infants, the prolonged absence of bacteria that usually colonize and protectterm infants (bifidobacteria) is a clear indicator that the intestinal maturation process is either severely disturbed or altered [5,6,21,[26][27][28][29].
Besides host biology, such as low birth weight (BW) and immaturity of the multiple organs including the gastrointestinal tract as a result of low gestational age (GA), there are multiple other exogenous factors that could affect the intestinal microbiome development/maturation of preterm infants (i.e. mode of delivery, neonatal feeding regime, the neonatal intensive care unit (NICU) environment and peri-and postnatal antibiotic exposure) [5,6,14,16,19,29].
While interest in this subject is growing, the focus of research is typically, with a few exceptions [15][16][17][18][19]23], on the relation between the intestinal microbiome of the preterm infant and disease instead of on microbiome composition development itself. We aimed to determine whether the early microbiome development in extremely premature infants 4 born between 25 and 30 weeks of gestation is dependant by infant maturity/maturation or whether it is mainly characterized by varying states of dysbiosis. Secondly, we aimed to analyse whether exogenous factors were intricately linked with microbiome maturation.

Patient characteristics
In this study 41 preterm born infants were included. The GA ranged between 25 and 30 weeks (median 27.6 weeks IQR 26.0-28.1). Table 1 summarizes the baseline characteristics. Detailed patient characteristics per week are shown in Table 2. Important for understanding this cohort, the mode of delivery was strongly associated with the birth weight Z-score for GA and gender ( Figure 1a). Infants delivered by CS had a higher GA on average than their vaginally born counterparts (median 28.1 IQR 26.7-28.2) vs 26.0 IQR 25.3-26.9 weeks, p = 0.002), yet were of lower BW (median 778 IQR 641-923 vs 835 IQR 800-980 grams, p = 0.03). Underlying causes of prematurity for all vaginally delivered infants (13/13) were intra-uterine infection, (prolonged) premature rupture of membranes and/or cervix insufficiency while most infants delivered by C-section were born preterm because of placental insufficiency (22/27) (Table 1). Placental insufficiency was negatively associated with the Z-score for birthweight (p = 0.0005, Figure 1b).

Intestinal microbiome development over time
We analyzed 142 samples (3.5 samples per patient on average, range 2-4), including the first meconium. An overview of the abundances (% of reads per sample) of the most important bacterial groups (species/genera) in these 4 timepoints is shown in Figure 2.
Within meconium samples Staphylococcus epidermidis was most frequently the most abundant bacterial species ( Bifidobacteria and lactobacilli were groups of minor importance within this cohort.

Mode of delivery and intestinal microbiota development
One of the most striking patterns in the data, alluded to in Figure 2, is that the mode of delivery has a significant influence on the microbiome composition. This was most evident within the first three weeks after birth (

Mode of delivery and infant weight development
In this cohort the mode of delivery was significantly associated with BW and GA ( Figure 1).
The association between the infant's current weight at sampling time and the mode of delivery remains largely unchanged during the first four weeks (p = 0.04, p = 0.03, p = 0.02 and p = 0.04 for weeks 1-4, respectively), as no difference was present in weight gain in g/week of infants delivered vaginally and by CS (p = 0.9; p = 0.8; p = 0.6 for weeks 2-4, respectively). Z-scores from BW were strongly correlated with absolute weight in all 4 weeks (Spearman correlation coefficients of 0.70, 0.67, 0.66 and 0.52, respectively).

Infant weight and intestinal microbiome development
While the mode of delivery was the most important determinant for the infants' initial microbiome (Figures 2 and 3), the increase of Enterobacteriaceae in weeks 3 and 4 appeared associated with absolute weight at sampling time ( Figure 4). When comparing infants with an above-median weight (835g) with their lighter counterparts, little difference was observed in the microbiome composition of the meconium (week 1). In week 2, infants with an above median weight (860g) contained significantly less S. epidermidis (p = 0.013). This particular association was however partially indirect, as infants delivered by CS had a lower median BW and were more frequently colonized during delivery by S. epidermidis ( Figure 3a). The weight of infants increased from a median of 860g in week 2 to 969g in week 3 to 1095g in week 4, respectively. In week 3 Enterobacteriaceae became more dominant again at the expense of S. epidermidis, as the influence of the mode of delivery declined. In week 4 the correlation coefficient between abundance of Enterobacteriaceae and current weight was significant ( = 0.4, p = 0.04).
More specifically, at a body weight of >1100g nearly all samples were dominated by Enterobacteriaceae. In general, the absolute weight of infants, irrespective of the week of sample collection, appeared to be (independently) associated with the shift from a S.
epidermidis dominated gut towards one dominated by Enterobacteriaceae ( Figure 5).
When combining all samples from all 4 timepoints, S. epidermidis was negatively associated with absolute weight measured at sampling time ( = 0.39, p = 0.000001) while the abundance of Enterobacteriaceae correlated positively with absolute weight measured at sampling time ( = 0.25, p = 0.003).

Intestinal microbiota development and health
While staphylococci were associated with low absolute weight in this cohort, they did not appear detrimental to health as 1) the amount of weight gain during any single week did not correlate with the gut microbiome composition (or any of the individual species) at the start of that week and 2) their previously reported negative association with necrotizing enterocolitis development was similarly found within this cohort in meconium samples (p = 0.034). Overall infant mortality in this cohort, in part caused by necrotizing enterocolitis (n = 8 in total; 3 caused by NEC), was not significantly associated with the gut microbiome composition or with BW (p = 0.36) but it was negatively associated with GA at birth (p = 0.005) and positively but not significantly (p = 0.08) with prolonged premature rupture of membranes (PPROM).

Exogenous factors and microbiota development
Exogenous factors such as antibiotics use and/or feeding regime were found to be of ancillary importance in comparison with patterns associated with the mode of delivery or with absolute body weight. Associations of bacterial groups with antibiotics use were either found to be non-significant or disappeared when adjusting the analyses for mode of delivery. The current number of subjects was insufficient to unravel significant associations between microbiome development with antibiotics and feeding regimes.

Discussion
This study, which prospectively investigated the development of the microbiome of extremely preterm infants during the first four weeks of life, has three main findings.
First, confirming current data [30], differences caused by mode of delivery have a strong dominance tot an Enterobacteriaceae dominated microbiome in extremely preterm infants.
We however found that gestational/postnatal age was an inaccurate descriptor of maturity/host biology with regard to the development of the gut microbiome. A reason for this inaccuracy is highlighted by our comparison of infants delivered vaginally or by CS. In our cohort CS delivered infants had on average a higher GA at birth but a lower BW (Zscore) than their vaginally delivered counterparts (Figure 1). In this dataset the type of pregnancy complication (placental insufficiency vs intra-uterine infection/spontaneous preterm birth) was not only significantly associated with the mode of delivery but also with Z-scores for BW, representing fetal growth restriction.
In the cohort observed by Ho et al. [16] they similarly found that bacilli and Enterobacteriaceae formed the dominant groups but they ascribed their findings to a dichotomous development of the gut microbiome. The dichotomous development of the gut includes one cluster (I) of samples starting off with a high abundance of staphylococci which in time gave way to an increased abundance of Enterobacteriaceae and the other cluster (II) starting off with a high abundance of Enterobacteriaceae that declined slowly as clostridia became more prominent [16]. The developments in these two clusters fit perfectly into the aforementioned patterned progression if weight is used as a marker of intestinal microbiome maturity instead of gestational/postnatal age. Instead of a threshold for GA, which La Rosa [26] suspected, there might be a weight-determined threshold, which influences the gut microbiome maturation. In the cohort of Ho et al., infants from cluster I and II had a similar GA at birth on average (28.0 vs. 27.9) but infants assigned to cluster I had a BW of 1053 g while those assigned to cluster II had a BW of 1176 g [16]. In our study we found that the switch (threshold) between staphylococci and Enterobacteriaceae was particularly evident around 1100 g, consistent with the difference between these 2 clusters.
The main driver for this difference is the cause of prematurity; placental insufficiency is typically accompanied by C-section and is logically negatively associated with lower BW Zscores due to fetal growth restriction of the infant. Furthermore, some infants thrive and concomitantly gain weight after birth while others do not, although both increase equally in gestational/postnatal age. We observed that absolute weight, as a logical key marker for host biology, does not suffer from these confounders and could be used to clearly visualize the patterned progression of {1} staphylococci to {2} Enterobacteriaceae ( Figure 5) in this dataset. The underlying mechanism why weight influences the gut microbiome maturation is not understood. We hypothesize that fetal growth restriction / low (birth) weight influences immune response modulation and altered intestinal development (i.e. influencing maturation of paneth cells, mucus production) that could cause weight dependent microbiome maturation differences [34].
Exogenous factors other than the mode of delivery such as antibiotics, neonatal feeding regime were also analyzed with regard to their association with the intestinal microbiome but no significant correlations were found in this cohort. While such external factors have been found to be associated with certain aspects of the gut microbiome development, they do not appear to represent the main drivers of gut maturation in preterm infants.
More samples would be required to study these factors. Maternal data, such as the results of vaginal swabs and/or the use of maternal antibiotics, were not available for this study but could be of possible relevance for initial colonization.
This study highlights that the intestinal gut microbiome development in the extremely preterm infants is mainly driven by weight but is initially strongly affected by the mode of delivery. It also stresses the importance of pregnancy complications with diminished placental function as it directly affects the actual physical maturity level of the infant in which in turn directly affects the stage of progression [26; 33] from which the gut microbiome development starts. Underlying mechanism behind weight thresholds is not yet understood. There is a need for more knowledge on the affect of (birth) weight on immunological responses and organ maturation.
Before interventions are implemented, such as targeted antibiotic therapies or the use of pre-and probiotics, it is critical to understand which organisms are to be considered normal (part of the maturation process), in regards to the level of gut development of the 11 infant at a certain weight, and which ones are indicative of potential dysbiosis/disease.
For example, K. oxytoca or K. pneumonia, which belong to the family Enterobacteriaceae, have been frequently associated with, amongst other things, sepsis in newborns, but they appear to be a 'normal' part of the patterned progression of the bacteria composition as the infant gut matures [16,26].

Patients
We selected infants from the CALIFORNIA trial who were born at a GA of ≤ 30 weeks and / or who had a BW of ≤ 1000 grams, from whom more than two samples during the first 4 weeks were available. Patient characteristics and demographic variables were derived from the CALIFORNIA database.

Demographic and clinical variables
Data from each sample day were used. Variables consisted of mode of delivery, BW, GA at birth, z-score BW (which represents the standard deviation in SD units from the Dutch reference growth curves) [41], bodyweight on the sample day, the administration of mothers' milk and / or of formula milk in milliliters/kg on the sample day, the antibiotic use on the sample day and if antibiotics were administered for more than 48 hours after birth. Complications during pregnancy were classified as placental insufficiency (preeclampsia/HELLP, fetal growth restriction and fetal distress) and intra-uterine infection/spontaneous preterm birth (chorioamnionitis, PPROM, cervical insufficiency and premature contractions).

13
Faecal samples We intended to analyze one fecal sample per week, starting with the first meconium, and afterwards the first fecal sample of every week during the first four weeks after birth.
Faecal samples were stored at-80°C prior to the start of this study. The first sample of each week was used for analyses of the current sub-study.
DNA extraction and sequence library preparation Faecal DNA was extracted from a 0.25g faecal sample by double bead beating in combination with the QIAamp DNA Mini kit (Qiagen; Hilden, Germany), as described by a study that used the same technique [42]. Polymerase chain reaction (PCR) amplification targeted the V3 and V4 region of the 16S rRNA gene by using modified 341F and 806R primers [43][44]. The 806R primers contained a 6-nucleotide barcode. An detailed description of the PCR reaction, DNA cleanup, and MiSeq library preparation is found in the Appendix file 1.

Analyses of sequence reads
The software used to analyze the data received from Illumina paired-end sequencing included PANDAseq [45] and ARB [46]. Readouts with a quality score lower than PANDAseq as standard practice to increase the quality of the sequence readouts discarded 0.9. ARB was used to identify sequences to the species level. As particularly meconium samples, but also many later samples have low bacterial biomasses, reagent contamination needs to be accounted for. Reagent contamination recognition analyses were subsequently performed as described by de Goffau et al [42], using the Spearman's rank correlation coefficients method.

Availability of data and materials:
The datasets during and/or analyzed during the current study are available on reasonable request.
Author consent and contribution: Acknowledgments: We would like to thank all the parents of subsequent neonates for participation in the study. Also, we would like to thank all the staff and students who were 21 involved in the inclusion and sample collection of the study. We greatly acknowledge the help of C. Bus-Spoor and R.H.J. Tonk for technical assistance of the study. Tables   Due to technical limitations, Table 1