Housing conditions, level of feeding and presence of antibiotics in the feed shape rabbit cecal microbiota

Background: the effect of the production environment and different management practices in rabbit cecal microbiota remains poorly understood. While previous studies have proved the impact of the age or the feed composition, research in the housing conditions and other animal management aspects, such as the presence of antibiotics in the feed or the level of feeding, is still needed. Characterization of microbial diversity and composition of growing rabbits raised under different conditions could help better understand the role these practices play in cecal microbial communities and how it may result in different animal performance. Results: four hundred twenty-ve meat rabbits raised in two different facilities, fed under two feeding regimes (ad libitum or restricted) with feed supplemented or free of antibiotics, were selected for this study. A 16S rRNA gene-based assessment through the MiSeq Illumina sequencing platform was performed on cecal samples collected from these individuals at slaughter. Different univariate and multivariate approaches were conducted to unravel the inuence of the different factors on microbial alpha diversity and composition at phylum, genus and OTU taxonomic levels. The animals raised in the facility harboring the most stable environmental conditions had greater, and less variable, microbial richness and diversity. Bootstrap univariate analyses of variance and sparse partial least squares-discriminant analyses endorsed that farm exerted the largest inuence on rabbit microbiota since the relative abundances of many taxa were found differentially represented between both facilities at all taxonomic levels characterized. Furthermore, only ve OTUs were needed to achieve a perfect classication of samples according to the facility where animals were raised. The level of feeding and the presence of antibiotics did not modify the global alpha diversity but had an impact on some bacteria relative abundances, albeit in a small number of taxa compared with farm, which is consistent with the lower sample classication power according to these factors achieved using microbial information. Conclusions: this study reveals different degrees of inuence attributable to environment and animal management. It highlights the importance of offering a controlled breeding environment that reduces differences in microbial cecal composition that could be causative of different animal performance.

Housing conditions, level of feeding and presence of antibiotics in the feed shape rabbit cecal microbiota Background Microbial communities that inhabit the gastrointestinal tract (GIT) of animals constitute a complex ecosystem whose members constantly interact between them and with their host [1-Gaskins, 1997].
These interactions ensure homeostatic balance maintenance since GIT ecosystem components are involved in many physiological and immunological processes [2-Belkaid and Hand, 2014]. In the case of the domestic meat rabbit (Oryctolagus cuniculus), a small herbivorous mammalian belonging to the family Leporidae, cecum is the main organ for microbial fermentation. Thus, it is not surprising that the rabbit cecum hosts the richest and the most diverse microbial community of its GIT [3-Gouet and Fonty, 1979]. For this reason, the cecum has been the organ preferably chosen in previous rabbit gut microbiota assessments [ Thanks to the development of the next generation sequencing (NGS) technologies, and their rapidly decreasing costs, it is currently possible to characterize the gut microbiota of a large number of animals.
This characterization allows a deeper comprehension of the differences between animals concerning their microbial composition and diversity. It is hypothesized that the production environment could partially mediate these differences. Our general aim is to provide further evidence of the effect of different management and environmental factors on the cecal microbial composition and diversity. In relation to this topic, there is a certain amount of information already published. A growing number of studies have revealed changes in rabbit cecal microbial communities exerted by the age [8-Combes et al., 2011] or the type of feed provided to the kits after weaning    [11-Gidenne et al., 2009] demonstrated that feed restriction, despite penalizing animal growth, improves feed e ciency and reduces mortality due to enteric disorders. It is hypothesized that these positive effects could be partially explained by changes in gut microbial composition or activity originated by the application of feed restriction. However, techniques used so far to study this possible association have not found evidence of it [11-Gidenne et al., 2009].
This study, which comprises a large number of animals in an experimental design involving different management and environmental factors, is intended to unravel changes in diversity and composition of rabbit cecal microbial communities associated with these factors. It will allow for a better understanding of how the housing conditions associated with farm where the animal was raised, the presence of antibiotics in the feed, and feed restriction shape the cecal microbiota of growing rabbits.

Results
Sequence processing OTUs and 0.17 (0.03) Shannon indexes; P FDR < 0.001). Furthermore, larger variability in both indexes was observed in farm A than in farm B. No signi cant differences for the two alpha diversity indexes were found between feeding regimes within both farms ( Figure 2, P FDR > 0.05), nor between the presence and the absence of antibiotics in the feed within farm B (Figure 2, P FDR > 0.05).
Animal management and farm environment shaping cecal microbial composition According to the taxonomic assignment of representative sequences (Additional le 4) performed with the UCLUST consensus taxonomy assigner on the Greengenes reference database gg_13_5_otus, Firmicutes (76.74%), Tenericutes (7.22%) and Bacteroidetes (6.26%) were the predominant phyla, accounting for more than 90% of the microbial diversity, in the rabbit cecal samples studied ( Figure 3).

Differential growth and cecal microbial composition across farms
The facility where the animals were raised affected their growth performance. Animals raised in farm B exhibited a faster growth (47.11 grams/day) than those raised in farm A (44.19 grams/day). The estimated average daily gain difference between farm B and farm A was 2.92 ± 0.94 grams per day (P < 0.005). Cecal samples of rabbits raised in farm A showed an overrepresentation of phyla Bacteroidetes, Proteobacteria and Verrucomicrobia while phyla Euryarchaeota, Cyanobacteria and Firmicutes were found to be overrepresented in cecal samples of rabbits raised in farm B (Table 1).  The analyses on the CSS-normalized OTUs revealed that 648 out of the 946 OTUs showed signatures signi cantly different between farms. Out of these, 276 were overrepresented in farm A, while 372 were overrepresented in farm B. Table S1 shows the estimated difference between farms for these OTUs, their sequences and their assignment at the lowest taxonomic level. Only 9 of them could be assigned at species level and 129 were assigned to known genera. These results showed remarkable coincidences with those obtained from the analyses directly performed on the relative abundance of taxa at phylum and genera levels. An example that illustrates this match is the overrepresentation of genus Akkermansia in farm A. This genus is encompassed by phylum Verrucomicrobia that was also overrepresented in rabbits raised in farm A, as well as 6 out of the 7 OTUs assigned to this phylum.
Differential growth and cecal microbial composition across feeding regimes The feeding regime affected the rabbits' growth performance in both facilities. Animals fed AL had a higher growth (48.74 and 55.77 grams/day in farms A and B, respectively) than those fed R (38.95 and 38.65 grams/day in farms A and B, respectively). The estimated average daily gain difference between AL and R groups was 9.79 ± 0.58 and 17.12 ± 1.08 grams per day in farms A and B, respectively (P < 0.001). An overrepresentation of phyla Cyanobacteria (estimated difference R -AL = 0.11 ± 0.04; P FDR = 0.04) and Verrucomicrobia (estimated difference R -AL = 0.11 ± 0.05; P FDR = 0.04) was found in cecal samples of rabbits fed R and raised in farm A. On the other hand, phylum Euryarchaeota was overrepresented in animals fed R and raised in farm B (estimated difference R -AL = 0.14 ± 0.04; P FDR < 0.001). At genus level, the only signi cant contrast was observed for rc4-4 which resulted overrepresented in samples from animals fed AL in farm A (estimated difference R -AL = -0.03 ± 0.01; P FDR < 0.001) while in farm B none of the genera resulted differentially represented (P FDR > 0.05) between feeding regimes.
The contrasts based on the CSS-normalized OTUs revealed 51 and 9 OTUs differentially represented between feeding regimes within farms A and B, respectively. Within farm A, 32 OTUs were overrepresented in cecal samples of rabbits that were fed AL and 19 OTUs in the samples from rabbits fed R. Within farm B, 7 OTUs were overrepresented in cecal samples of rabbits that were fed AL and 2 OTUs were overrepresented in rabbits that were fed R. Table S2 shows the estimated difference between feeding regime within farm of these OTUs, their sequences and their assignment at the lowest taxonomic level. The analyses based on the CSS-normalized OTUs within farm A were in full accordance with the analyses performed at genus level given that all OTUs assigned to genus rc4-4 (phylum Firmicutes) were overrepresented in cecal samples of rabbits fed AL.

Effect of the presence of antibiotics in the feed
The effect of the presence of antibiotics in the feed could only be assessed within farm B given that all rabbits raised in farm A received feed supplemented with antibiotics. Animals that received antibiotics had a slightly higher growth (47.29 grams/day) than those that did not (46.59 grams/day). The estimated average daily gain difference between groups was not signi cant (0.69 ± 2.43 grams per day; P = 0.78). Cecal samples of rabbits that received feed free of antibiotics showed an overrepresentation of phyla Cyanobacteria compared to those that received feed supplemented with antibiotics (estimated difference without antibiotics -with antibiotics = 0.49 ± 0.09; P FDR < 0.001). In addition, the analyses on the CSS-normalized OTUs revealed an overrepresentation of 15 and 29 OTUs in cecal samples of rabbits that received a feed supplemented or free of antibiotics; respectively. Table S3 shows the estimated difference between the presence and the absence of antibiotics in the feed for the OTUs in which the differences reached the signi cance threshold. The OTU sequences as well as their assignment at the lowest taxonomic level are also shown in Table S3. Only 1 of these OTUs could be assigned at species level (Bacteroides fragilis) and 2 OTUs at genus level (Oscillospira and Coprococcus).
Microbial information as a classi er of cecal samples according to farm environment and animal management Sparse partial least squares-discriminant analyses (sPLS-DA) on the CSS-normalized OTUs were conducted to discriminate samples according to the factors considered in this study (i.e., the farm where the animal was raised, the presence or the absence of antibiotics in the feed and the feeding regime). The tuning process of the sPLS-DA conducted to discriminate samples according to the farm where the rabbits were raised selected 5 OTUs for component 1 and 1 OTU for component 2 ( Figure 4). Component 1 explained 7.00% of the total variance while component 2 explained 0.67%. The classi cation performance of this sPLS-DA could be said to be perfect since its overall and balanced error rate (BER) per class across 1000 replicates of 5-folds cross-validation runs was 0.00 (0.00). Furthermore, two OTUs of component 1 had a stability higher than 0.9.
The sPLS-DA performed to discriminate samples across feeding regimes within farm A selected 70 OTUs for component 1 and 65 OTUs for component 2 ( Figure 5). Component 1 explained 2.34% of the total variance while component 2 explained 5.58%. The cross-validation assessment of the classi cation performance of this sPLS-DA showed an overall and BER per class of 0.27 (0.02). The stability of 18 and 5 OTUs selected in components 1 and 2, respectively, across the different cross-validation folds was higher than 0.9.
Finally, the sPLS-DA conducted to discriminate samples of animals raised within farm B according to the combination of the presence or not of antibiotics in the feed and the feeding regime selected 9 OTUs for component 1 and 70 OTUs for component 2 ( Figure 6). Component 1 explained 3.05% of total variance and de ned the discrimination between samples from animals fed with antibiotics and those fed without antibiotics. On the other hand, component 2 explained 3.05% of total variance and de ned the discrimination between samples from animals fed R and those belonging to animals fed AL. The crossvalidation assessment of the classi cation performance of this sPLS-DA showed an overall BER of 0.32 (0.15). The BER per class was 0.34 (0.12) for samples fed R without antibiotics, 0.46 (0.14) for samples fed AL without antibiotics, 0.29 (0.11) for samples fed R with antibiotics, and 0.20 (0.07) for samples fed AL with antibiotics. The stability of 3 and 11 OTUs selected in components 1 and 2, respectively, across the different cross-validation folds was higher than 0.9.

Discussion
The in uences of farm environment and common commercial practices of animal management on their gut microbiota are not yet well known in many livestock species. In this study, we have aimed to disentangle potential changes in microbial diversity and composition of meat rabbit cecal communities as a result of being raised in different farms and subjected to different handling during their growing period. To shed light on this matter, we conducted a microbiota comparison of a large number of rabbits raised under different housing conditions, feeding regimes, and fed with feed supplemented or free of antibiotics.
16S rRNA gene-based characterization of meat rabbit cecal microbiota The Illumina MiSeq sequence processing of samples collected from these animals revealed that phyla Firmicutes, Tenericutes and Bacteroidetes dominate the growing meat rabbit cecal ecosystem representing more than 90% of its entire microbial composition. This fact is in accordance with previous studies that have characterized the rabbit cecal microbiota    [12] reported Firmicutes (76.42%), Tenericutes (7.83%) and Bacteroidetes (7.42%) as the predominant phyla of meat rabbit fecal and cecal microbial communities. These discrepancies found across studies could be attributed to technical issues (e.g., pair of primers, sequencing platform, bioinformatic pipeline employed to process raw sequences or reference database used for the taxonomic assignment of the representative sequences) or to purely biological reasons (e.g., breed, age or section of the GIT sampled). Nonetheless, Kylie et al. (2018) [13] depicted that the relative increase in less bene cial phyla, such as Proteobacteria, could be related to seasonal climate changes that directly impact rabbits' health. This impact affects the susceptibility to enteritis and possibly feed conversion e ciency. In any case, this phylum was more prevalent in farm A where the animals were more exposed to changes in climate conditions.

Farm environment modify alpha diversity
Regarding the alpha diversity assessment, Shannon and the observed number of OTUs indexes revealed the existence of signi cant differences between housing conditions (i.e., the experimental farm where the rabbits were raised). Cecal samples collected from rabbits raised in farm B had greater richness and diversity than those belonging to animals raised in farm A. This could be explained by more stable environmental conditions in farm B (i.e., facility better insulated) than in farm A. It has been already shown that intestinal health is positively associated with microbial diversity [14-Larsen and Claassen, 2018]. In our case, this better health could be said to be granted by the more stable environmental conditions offered by farm B. The most exposed environmental conditions of farm A, combined with the fact that samples of animals raised in this facility were collected from rabbits produced in 4 different batches, could also explain the larger variability in both indexes observed in this farm [13-Kylie et al., 2018]. Despite not having observed signi cant differences between the presence or not of antibiotic in the feed, nor between feeding regimes, it is noteworthy to mention that samples collected from animals fed AL in both farms had a greater, although not signi cant, richness than those fed R. This fact is consistent with previous studies in mice that observed a lower alpha diversity in animals with a restricted level of between animals fed on diets with antibiotics with respect to those on diets free of antibiotics. Nevertheless, these studies were able to detect differences in the relative abundances of some speci c species between diets. For example, Kumar et al. 2018 [19] found that the inclusion of bacitracin in the feed did not affect the chicken bacterial phyla. However, they observed differences between the control and the bacitracin-fed group in the ileal and cecal bacterial populations at lower taxonomic levels. It is worth noting that the antibiotic withdrawal at the beginning of the last week of the rabbits' lives equalized the diets of both groups and possibly their microbial populations, which may explain some lack of differences between them.
Farm environment has a large impact on rabbit cecal microbiota Despite the lack of differences in microbial diversity and richness across management factors; univariate studies revealed differential microbial composition across the studied factors. In addition, the performed multivariate analysis evidenced a certain classi cation power of the samples on the different levels of management and environment factors based on the microbial composition of the samples.
As it might be expected, analyses of variance con rmed that the largest modi cation of meat rabbit cecal microbial composition is generated by the housing conditions (in this case represented by farm factor). Our results revealed that the relative abundances Within the phylum Verrucomicrobia, genus Akkermansia is an anaerobic Gram-negative bacterium that encompasses mucin degrader species [28-Belzer et al., 2012]. In the cecum, a proper enrichment of this genus could maintain a suitable mucosal turn-over, thus exerting a protective effect that could help the animal to deal with in ammatory processes.
It is worth mentioning that we have detected genera Epulopiscium, p-75-a5, Phascolarctobacterium, Campylobacter and Desulfovibrio only in the cecal samples of rabbits housed in farm A. The rst three are encompassed within the phylum Firmicutes. Genus Epulopiscium is a large size Gram-positive bacterium that has a nutritional symbiotic relationship with surgeon sh that eats algae and detritus. This bacterium is physically similar to the phylogenetically related Metabacterium polyspora which is an endospore-producing bacterium isolated from the cecum of guinea pigs [29-Angert et al., 1996]. On the other hand, genera Campylobacter and Desulfovibrio are Gram-negative bacteria that belong to phylum Proteobacteria. Some species of these genera are pathogens responsible for infections and diarrheas in mammals. The exclusive presence of these genera in farm A could indicate the existence of a potential dysbiosis of the animals raised in that facility that could affect their sanitary status and growth. While farm A was a semi-open-air facility, farm B was arti cially ventilated and offered more controlled environmental conditions that favor animal growth. Moreover, the presence of sulfate-reducing bacteria (SRB) such as Desulfovibrio could be enhanced by sulfate-secreting bacteria (SSB) such as Rikenella in farm A where this genus is signi cantly more predominant. It is noteworthy to mention that SRB could also obtain sulfate via cross-feeding mediated by Bacteroides-encoded sulfatases [30-Rey et al., 2013], and interestingly, this phylum is more prevalent in farm A.
Regarding sample classi cation based on the sPLS-DA study, given the important differences in gut microbial composition found between farms, a perfect classi cation of the samples can be achieved with only 5 OTUs. One of these 5 OTUs was overrepresented in farm B and belonged to family S24-7 (phylum Bacteroidetes). The remaining 4 were overrepresented in farm A and belonged to family Barnesiellaceae (phylum Bacteroidetes), order Bacteroidales (phylum Bacteroidetes), and genera Desulfovibrio (phylum Proteobacteria) and Bacteroides (phylum Bacteroidetes). It is worth mentioning that these 5 OTUs were also declared as differentially represented between farms by the univariate analyses.
Administration of antibiotics impact on some taxa relative abundances However, some signi cant differences were observed at phylum and OTU levels. An overrepresentation of phylum Cyanobacteria was found in rabbits fed without antibiotics. The detection of this bacterial phylotype, commonly assigned to photosynthetic activity, in the rabbit cecum could suggest contamination during the GIT sampling. However, Zeng [13], revealed that rabbits fed without antibiotics exhibited higher abundances of OTUs assigned to phylum Bacteroidetes than those fed with antibiotics. In addition, samples of rabbits that received antibiotics had a signi cant increase of an OTU taxonomically assigned to genus Coprococcus. Interestingly, a study that evaluated the differences in bacterial communities of Rex rabbits fed with different antibiotics also found an overrepresentation of this bacterium in animals treated with zinc bacitracin [5-Zou et al., 2016]. Coprococcus is an anaerobic bacterium that may protect against colon cancer in humans by producing butyric acid [35-Ai et al., 2019]. We hypothesized that the administration of antibiotics could modulate the abundance of some Coprococcus species to provide intestinal protection on meat rabbits. However, it is important to recognize that the reduced sample size of the group of rabbits fed without antibiotics may have limited the statistical power to detect microbial composition differences associated with this factor.
Feed restriction modify Euryarchaeota and some bacteria relative abundances

Gastranaerophilales
Regarding the sample classi cation based on the sPLS-DA study conducted within farm B, component 1 and component 2 discriminated between animals that did or did not received antibiotics in the feed and between feeding regimes, respectively. It is worth mentioning that 8 out of 9 OTUs selected in component 1 were also declared as differentially represented between the presence or the absence of antibiotics in the feed by the univariate analyses. Within farm A, an sPLS-DA was also performed to classify samples according to the feeding regime using microbial information. Although a large number of OTUs were selected as classi er variables in the tuning process of this sPLS-DA, the classi cation error rate was high. It implied a poor discrimination capacity of samples according to the feeding regime the animal received. Nevertheless, bootstrap univariate analyses of variance detected some signi cant differences at all taxonomic levels analyzed between feeding regimes within farm A. At genus level, rc4-4 was overrepresented in animals fed AL. This genus belongs to phylum Firmicutes and it is known as an obesity-associated bacterium [41-Ziętak et al., 2016] and as a pathogenic candidate identi ed in mice with multiple sclerosis [42-Gandy et al., 2019]. A potential pro-in ammatory role has been proposed for this genus [42] what could be related to a reduced incidence of enteric disorders when feed restriction is applied. It is worth mentioning that family Peptococcaceae, which encompasses genus rc4-4, is strongly related to total rabbit weight gain from weaning to 12-week old [43-North et al., 2019]. Although in our study this genus was prevalent in animals fed AL, its association with weight gain is not clear since the greater growth exhibited by these animals was consequence of higher feed intake.
Rabbit cecal microbiota is shaped by farm environment and animal management Different approaches have been applied in this study to evaluate the effect of different environments and management practices, commonly used in rabbit production, in their cecal microbial composition and diversity. Our results con rmed that the most important effect is exerted by the environment provided by the farm where the animals were raised. Those raised in the best insulated facility (farm B) appear to have a microbiota characteristic of healthier animals than those raised in the open-air facility (farm A). It is worth mentioning that the rabbits were housed in cages interspersed with feeding regime. This fact could make possible the exchange of microorganisms between animals of different feeding regimes and therefore have reduced the differences observed between regimes. However, the joint consideration of 70 OTUs in the sPLS-DA made possible a certain discrimination power of samples according to the level of feeding received by each animal raised in farm A. It implies the existence of cecal microbiota content patterns characteristic of each regime which could be revealed thanks to the univariate analyses conducted at different taxonomic levels. Similarly, the sPLS-DA performed within farm B also involved the consideration of 70 OTUs to discriminate samples according to the amount of feed consumed. Within this farm, the classi cation of samples regarding the presence or the absence of antibiotics in the feed needed a smaller number of OTUs than the feeding regime but greater than the farm. This suggests that the effect of the presence of antibiotic in feed is stronger than the feeding level. The lack of a group of samples collected from animals that did not receive antibiotics precluded the evaluation of the magnitude of importance of this effect over the feeding level on the cecal microbiota of animals raised farm A. It might have been possible that the magnitude of the effect of the presence of antibiotics in the feed was larger in farm A than that observed in farm B. The experimental design of this study prevented the comparison of the effect of antibiotic treatments across farms on rabbits' microbial communities. The implication of the discussed microbial composition and diversity differences originated by the studied management and environmental factors on the animals' performance still needs to be investigated. In future studies the role of speci c groups of bacteria in rabbit growth and feed e ciency will be analyzed.

Conclusions
The analysis of a large number of animals from a paternal rabbit line has allowed a deeper comprehension of the role played by different management and environmental factors shaping the composition and diversity of cecal microbial communities. It reveals that the housing conditions offered to the rabbits during their growth play a key role that can result in different microbial alpha diversity and composition of almost all species that inhabit the rabbit GIT. This highlights the importance that a stable and controlled environment could have in the intestinal health and, consequently, in animal performance.
It seems clear that the better insulated conditions of farm B favored the presence of a gut microbiota characteristic of healthier animals. Although the level of feeding and the presence of antibiotics in the feed did not modify the global diversity of cecal microbial communities, these factors can increase or decrease the prevalence of speci c bacteria which could lead to a microbial composition potentially bene cial for the animal or, at the other extreme, to an origin of future intestinal dysbiosis. once per day in a feeder with three places for the 4-5 weeks that the fattening lasted. Water was provided ad libitum during the whole fattening period. The animals were under two different feeding regimes: (1) ad libitum (AL) or (2) restricted (R) to 75% of the AL feed intake. The amount of feed supplied to the animals under R feeding regime in a given week for each batch was computed as 0.75 times the average feed intake of kits on AL from the same batch during the previous week, plus 10% to account for a feed intake increase as the animal grows. Kits were randomly assigned to one of these two feeding regimes after weaning (32 days of age). They were categorized into two groups according to their size at weaning (big if their body weight was greater than 700 g or small otherwise) aiming to obtain homogenous groups regarding animal size within feeding regime. A maximum of two kits of the same litter were assigned to the same cage in order to remove the possible association between cage and maternal effects on animal growth during the fattening period. The distribution of these animals across the different levels of management factors is shown in Table 3. The body weight of each animal was weekly recorded. The individual average daily gain was computed as the slope of the within animal regression of all body weight measurements recorded during the growing period. Sample processing, DNA extraction and sequencing Animals were slaughtered (at 66 and 60 days of age in farm A and farm B, respectively) and cecal samples of each rabbit were collected in a sterile tube, kept cold in the laboratory (4ºC) and stored at -80ºC. DNA extraction, ampli cation, Illumina library preparation and sequencing followed methods  Greengenes reference database gg_13_5_otus with the UCLUST consensus taxonomy assigner (QIIME default parameters). The raw sequence data were deposited in the sequence read archive of NCBI under the BioProject accession number PRJNA524130. Metadata, the pre ltered and normalized OTU tables, and corresponding taxonomic classi cations are also included as Additional les 1, 2, 3 and 4, respectively.

Models and statistical methods
In order to study differences in diversity and richness between rabbits grouped according to farm environment and management that they received, two alpha diversity indexes (Shannon and the observed number of OTUs) were computed from the OTU table rari ed to 10,000 sequences per sample with "phyloseq" R package [52-phyloseq]. The statistical method chosen to assess alpha diversity differences between these groups of animals was an analysis of variance that included a factor resulting from the combination of four factors (the farm where the animal was raised, the batch, the presence or the absence of antibiotics in the feed and the feeding regime). The signi cance threshold was set at 0.05 for type I error.
Different approaches were considered to assess the in uence of the environments and management factors on microbial composition. A bootstrap analysis of variance was individually implemented for each OTU to test whether it was differentially represented between the different categories of the factors studied. This univariate analysis was conducted by normalizing the OTU table with the cumulative sum scaling (CSS) method [53-Paulson et al., 2013] and only for those OTUs which were detected in at least 5% of the samples and had a sum of its counts resulting in a frequency greater than 0.01% of the total sum of all OTUs counts across all samples. It was implemented by tting a model de ned by the combination of the four aforementioned factors by using lm() function in R [54-R]. Then, the differences between the CSS-normalized OTUs counts in the different levels of the studied factors were tested. The signi cance between the levels of the main factors: farm, presence of antibiotics in the feed and feeding regime was assessed using an F statistic. When the involved interaction terms were signi cant, the contrasts of interest were studied nested within the levels of other interacting factors, i.e. feeding regime were studied within farm levels. When the interaction terms were not signi cant, the effects of the different levels were averaged, i.e. the effects of the levels of the batches within farm A were averaged to present the effect associated with this farm. In the performed F tests instead of relying on the theoretical distribution of the statistic under the null hypothesis to de ne the p-values, they were empirically computed using bootstrap after 1,000 permutations of the dependent variable with respect to the design matrix of factors in the model. The use of bootstrapping enabled the hypothesis test to be done without the need of assuming that data are normally distributed, which is an assumption that fails for OTUs counts. P-value was de ned as the proportion of bootstrap rounds having an F statistic value equal or greater than that obtained with the original dataset. P-values were corrected de ning a false discovery rate (FDR) of 0.05 [55-Benjamini and Hochberg, 1995]. This bootstrap analysis of variance approach was also implemented to study the effect of the management factors on the relative abundance of bacteria at phylum and genus levels.
The value of the microbial information to classify samples into the three factors considered in our study was explored using multivariate techniques. In particular, sparse partial least squares-discriminant analysis (sPLS-DA) [56-Le Cao et al., 2008] was used to nd the combination of OTUs that allowed the best classi cation of cecal samples according to: (1) the farm where the animals were raised, (2) the feeding regime within farm A and (3) the combination of feeding regime and the presence or absence of antibiotics in the feed for the animals raised in farm B. This approach was implemented through the R package "mixOmics" [57-mixomics]. In a rst step, the function tune.splsda() was used to select the optimal sparsity parameters of the sPLS-DA model: the number of components and the number of variables (OTUs) per component. For the tuning process, a 5-fold cross-validation repeated 10 times was performed one component at a time, with a maximum of 4 components, on an input grid of values that indicate the number of variables to select on each component. The sparsity parameters were de ned, based on the BER and centroids distance, and then included in the nal sPLS-DA model. Samples were represented on the rst two components and colored according to their class (e.g., R or AL in the case of the feeding regime) in a sample plot with the function plotIndiv(). The performance of the sPLS-DA model was assessed with a 5-fold cross-validation repeated 1,000 times that randomly split the data in training and validation sets. In this data partition, it was ensured that 20% of the samples within each level of the discriminant factor were assigned to the validation set. Five different partitions were performed for each replicate to guarantee a different sample distribution in each validation set. The sPLS-DA model with the sparsity parameters previously de ned was adjusted in the training set and its classi cation performance was assessed in the validation set using the overall and BER per class as criteria. The stability of the OTUs selected on each component was also assessed in the cross-validation by computing the selection frequency of each variable across the replicates.

Consent for publication
Not applicable.