Characterising the bacterial gut microbiome of probiotic-supplemented very preterm infants

Background: 27 The gut microbiome plays a critical role in the healthy development, immunity and 28 metabolism of infants. Preterm birth disrupts microbiome development and can contribute to 29 acute and chronic disease. To promote microbial and infant development, and to mitigate the 30 risk of disease, premature infants may be treated with probiotics. Here we used 16S rRNA 31 high throughout sequencing to characterize the bacterial microbiome of probiotic- 32 supplemented premature infants. The study aimed to identify and understand variation in 33 bacterial gut flora, including changes from admission to discharge, and the effect of several 34 clinical variables using a combination of univariate and mixed effects analyses. Results: faecal 63). Our research builds on previous research and supports significant changes over time in the preterm infant microbiome, and in response to several variables. Univariate analysis showed admission and discharge samples had significantly different microbial populations, with Staphylococcus 41 enriched at admission and Enterobacter , Lactobacillus , Colstridium sensu stricto 1 and Veillonella at discharge. From the mixed effects modeling we observed significantly lower alpha diversity in infants diagnosed with either sepsis or retinopathy of prematurity (ROP), and those that only received formula milk. Chorioamnionitis, preeclampsia, sepsis, necrotizing enterocolitis and ROP were also all associated with differential abundance of several taxa. Conclusions: Our study builds on previous research and supports significant changes in the preterm microbiome over time and in association with several factors. The fact that several associations were observed, and some in ways that counter previous work, highlights the complexity of microbiome ecology.

Backwards selection (69) was then implemented to simplify the model by comparing 220 Akaike's Information Criterion (AIC) scores between regression models and removing 221 predictors that were not contributing to the model. The process was repeated until the least 222 complex adequate model was identified, when no more predictors could be removed without 223 significant effects. The final model was Shannon ~ (Sepsis + Feeding Type + 224

Chorioamnionitis + (Mode of Delivery + Gestation Days + NEC + Preeclampsia + ROP)) * 225
Type + (1|URN). The significance of the fixed effects variables in this final model was then 226 assessed using analysis of deviance (Type II Wald Chi-square test) from the car package 227 (73), and post-hoc pairwise Tukey comparisons (correcting for multiple comparisons) from 228 the emmeans package (74). 229 For differential taxonomic abundance, two negative binomial generalized linear 230 models were created using the package DESeq2. A combination of previous literature and 231 exploratory analysis, including PCoA plots, PCA and scatterplots, were used for model 232 selection. Again, continuous predictors were scaled and centered, and multicollinearity was 233 assessed. Taxa were agglomerated at the genus level, due to the limited sequencing depth of 234 short amplicon sequencing. To reduce the number of false positives, two separate models 235 were run; one each for admission and discharge samples. The resulting model Taxonomic 236

Abundance ~ Sepsis + Feeding Type + Chorioamnionitis + Mode of Delivery + Gestation 237
Days + NEC + Preeclampsia + ROP was created to assesses the effect of the 9 independent 238 predictors at both time points on all genera present. Low abundance and low frequency taxa 239 were then removed, and a Wald Test with the Benjamin-Hochberg multiple inference 240 correction was then performed to determine significant differentially abundant taxa. More 241 information on the analysis can be found in Additional   Figure 2C). 258 For beta diversity, although there was limited separation between admission and discharge 259 samples, there was clustering that resulted in a significant difference between the two groups 260 based on abundance and phylogeny ( samples. Significant pairwise differences in diversity were observed for feeding type, sepsis 269 and ROP (Figure 3), and chorioamnionitis, sepsis, NEC, ROP and feeding type were all 270 associated with changes in taxonomy (Table 1) Both the mode of delivery and type of milk the baby received had significant associations 289 with diversity, but only mode of delivery was associated with differential taxonomic 290 abundance. Diversity was significantly higher in cesarean born infants at discharge than those 291 born vaginally, relative to the difference observed at admission ( Figure 3A; ꭓ 2 = 4.18, df = 1, 292 p<0.05). However, subsequent post-hoc analysis showed no significant pairwise comparisons 293 within the delivery variable. The type of milk the infant received also had a significant effect 294 15 ( Figure 3B; ꭓ 2 = 7.29, df = 2, p<0.05), with subsequent post-hoc pairwise comparisons 295 finding a significant difference between formula-fed infants ( ̅ = 2.10 ± 0.17) and those 296 breastfed ( ̅ = 1.56 ± 0.11) ( Figure 3B; p<0.05). For differential abundance, infants who 297 were fed only breastmilk had significantly higher abundances of both Bifidobacterium (Table  298 1; p<0.01) and Klebsiella (Table 1; p<0.01) relative to those only fed formula. 299

Pregnancy complications 300
Both preeclampsia and chorioamnionitis had a significant impact on the infant gut 301 microbiome. Of the two complications, only preeclampsia influenced infant microbial 302 diversity. A significant difference exists at discharge between infants whose mothers were 303 diagnosed with preeclampsia ( ̅ = 1.68 ± 0.27) and those infants whose mother did not have 304 the disease ( ̅ = 1.83 ± 0.10) (ꭓ 2 = 4.96, df = 1, p = 0.03), relative to the difference observed 305 at admission ( Figure 3C). However, no significant pairwise differences were found in 306 subsequent analyses for diversity. 307 Both preeclampsia and chorioamnionitis also significantly influenced taxonomy 308 (Table 1). In infants whose mothers were diagnosed with Chorioamnionitis before or during 309 labor, Staphylococcus was significantly higher at admission (p<0.05) and Streptococcus 310 significantly lower at discharge (p<0.01). For infants whose mother was diagnosed with 311 Preeclampsia there were no differences at admission, but significantly lower 312 The aim of this study was to understand and identify variation in gut flora development in a 333 unique cohort of probiotic-supplemented preterm infants. Specifically, we set out to assess 334 how the bacterial microbiome differs between two time points in the hospital, admission and 335 discharge, and the effect of several clinical variables (both maternal and infant) in a cohort 336 of preterm infants from North Queensland, Australia. To do so, we utilised 16S rRNA high 337 throughput sequencing. We then conducted univariate comparisons to examine the 338 difference between the infant microbiome at admission and discharge, and mixed effects 339 models to explore the influence of several clinical variables, including Sepsis, Feeding Type, 340 Chorioamnionitis, Mode of Delivery, Gestation, NEC, Preeclampsia and ROP. 341 Exploring changes in composition and diversity from admission to discharge 342 Although median alpha diversity increased between admission and discharge, the difference 343 was not significant. As previously mentioned, other work has shown the preterm infant 344 microbiome changes significantly with time, eventually becoming more similar to that of 345 full-term infants. Our contrary findings could be due to a large spread of diversity scores at 346 admission, and a reduction in diversity for some infants. While the cause of this large 347 variation in diversity during admission is unclear, the decrease in diversity could be for in higher abundance at admission, which has also been observed previously (22). As time 364 progresses, most other microbes also increase in absolute abundance (5), due to further 365 colonisation and replication of microbes. 366 At discharge from the NICU, several taxa appear to dominate, with Bifidobacterium, 367 Lactobacillus and Enterobacter found in high abundance across most of the cohort. 368 Significant differences in abundances between admission and discharge were found for 369 Lactobacillus and Enterobacter, but not for Bifidobacterium (p = 0.11). This is surprising 370 considering preterm infants are known to experience delayed and limited colonization of 371 common commensals like Bifidobacterium and Lactobacillus (9, 22, 79). Although changes 372 in Bifidobacterium did not reach a level of significance in our cohort, it is worth noting that 373 99 of 134 samples contained the genus, in a cohort of infants born <32 weeks gestation, that 374 were also receiving a probiotic (Infloran TM ) containing both Lactobacillus acidophilus and 375 Lactobacillus bifidus (Bifidobacterium bifidum). So, although significant changes were not 376 observed, the presence of Bifidobacterium in the majority the cohort suggests the probiotic 377 may be having an impact. 378 When comparing variation of microbial communities between samples using beta 379 diversity, there appears to be limited separation by sample type (admission/discharge) 380 ( Figure 2A). However, there is still a significant association between when the sample was 381 collected (admission or discharge) and beta diversity, which is unsurprising considering 382 microbial populations change significantly over time during infancy. However, the poor R 2 383 suggests that although when the sample was collected is associated with beta diversity, there 384 is still a lot of variation unexplained in the model. This is likely due to environmental 385 variables that also influence the developing infant gut microbiome. 386

Exploring the effect of clinical variables on alpha diversity and taxonomic abundance 387
Mixed effects modelling was used to explore the impact of several clinical variables on alpha 388 diversity (Shannon Index) and taxonomic differential abundance. Some of these variables 389 have previously been implicated in shaping the gut microbiome, and others associated with 390 disease. Our data builds on previous findings, but does not support all previously made 391 observations, highlighting the complexity of gut microbiome ecology. 392

Mode of delivery and diet 393
In contrast to previous work, we observed no significant pairwise differences in diversity or 394 taxonomy between vaginally and caesarean delivered infants at admission or discharge. 20 Klebsiella was also found to be significantly higher in breastfed infants. The genus 415 Klebsiella contains well known pathogen species, such as Klebsiella pneumoniae, previously 416 associated with NEC (88). The transfer of this pathogen from mother to infant via breastmilk 417 has also been implicated in sepsis in clinical observations (89). However, as pathogens like 418 K. pneumoniae only constitute a small proportion of the genus Klebsiella, and Klebsiella is 419 also a member of normal gut flora there is little need for concern. Additionally, despite 420 clinical reports linking maternal-infant translocation of microbes to breastfeeding, breastmilk 421 is the most cost effective preventative intervention from infection (90). In fact, the presence 422 of microbes, specifically commensal microbes like Bifidobacterium and Lactobacillus could 423 by why breastfeeding reduces the risk of diseases like sepsis (91). 424

Pregnancy complications 425
Infants whose mother was diagnosed with chorioamnionitis had higher abundances of the 426 genus Staphylococcus at admission, but fewer Streptococcus at discharge. A significant 427 relationship between chorioamnionitis and the altered infant gut microbiome has also been 428 observed previously for differential abundance of different taxa (92), as well as a close to 429 significant difference in diversity (25). As chorioamnionitis is a bacterial infection of the 430 membrane surrounding the fetus, occurring before or during labor, translocation of pathogens 431 from the membrane to the fetus may occur. Unfortunately, the translocation and resulting 432 increased abundance of Staphylococcus may be why exposure to chorioamnionitis increases 433 the risk of preterm infants to adverse neonatal outcomes (92), like sepsis, which has also been 434 associated with Staphylococcus (93, 94). 435 For infants whose mother was diagnosed with preeclampsia, Escherichia/Shigella was 436 significantly lower at discharge, and although no significant pairwise comparisons were 437 found, preeclampsia did appear to have some effect on alpha diversity at discharge as 438 determined by analysis of deviance. As preeclampsia can alter the maternal microbiome (30), 439 21 and a large proportion of infant microbial colonization is from a maternal route it is 440 unsurprising that preterm infants whose mothers were diagnosed with preeclampsia can have 441 significantly different microbiomes. Additionally, there was also a close to significant 442 decrease in alpha diversity (p = 0.08) from admission to discharge for infants whose mother 443 was diagnosed with the disease, which may explain why several individuals experience a 444 reduction in diversity from admission to discharge.  For NEC, we observed significantly lower abundances of Bifidobacterium, but in 476 contrast to previous work, no enrichment of any taxa. As previously mentioned, 477 Bifidobacterium is a common commensal microbe, that is also found in the probiotic 478 Infloran TM . It is uncommon in preterm infants born <33 weeks gestation (79), and has 479 previously been shown to be protective against NEC (98). Although our work does not 480 support previous evidence of a single infectious pathogen, the plethora of microbes that have 481 previously been associated with NEC (39, 99, 100) , in combination with studies showing 482 reduced commensal microbes (101, 102) and diversity (103, 104), suggests the aetiology is 483 far more complicated than just the presence of a single pathogen. Rather, the disease appears 484 to result from microbial dysbiosis, that includes reduced commensal microbes like 485

Bifidobacterium. 486
We also observed significant enrichment of Staphylococcus (of the 487 Staphylococcaceae family) at admission for infants diagnosed with ROP, as well as 488 significantly lower diversity. An association between the gut microbiota and ROP has only 489 23 been explored once before, by Skondra et al (44). They observed significant enrichment of 490 the family Enterobacteriaceae in preterm infants with the disease at 28 weeks postmenstrual 491 age (44). The discrepancy in our results is not necessarily a product of error, but rather, as 492 seen with NEC, due to the complex aetiology characterised by more than just the presence of 493 a particular group of taxa. This complexity makes it difficult to hypothesise the specific role 494 that the microbiome could be playing in ROP. However, if a role is established there is 495 potential for the microbiome to become a target for intervention, and thus this should be the 496 target of further research. 497 498

Limitations 499
Limitations of our work include low sequencing depth and only sampling in early infancy. 500 The use of 16S metabarcoding limited our detection power to the genus level, resulting in no 501 identification of species or functional genes. Additionally, only collecting samples at 502 admission and discharge means we have no insight into the longevity of the differences 503 observed, which may impact their clinical significance. Our future work will use a 504 combination of 16S metabarcoding and shotgun metagenomic techniques to both characterize 505 species and genes and to explore if the differences observed in this study, and others, persist 506 in the long-term. 507 508

Conclusion 509
This prospective observational study used 16S rRNA high throughout sequencing to 510 characterize the bacterial microbiome of probiotic-supplemented infants. It aimed to identify 511 and understand variation in bacterial gut flora between two time points and as the result of 512 several clinical variables. Our study builds on previous research and supports significant 513 changes in the preterm microbiome over time and associations with several factors. 514