Salmonella Phage cocktail, its effects and benets on the gut of chickens in a commercial farm.


 BackgroundThe microbiota of broiler chicken gut affects positively or negatively the health, metabolism and immunity of chickens thus it has a significant impact on the animal productivity. Phages, host-specific parasites of bacterial cells, are a promising antimicrobial alternative that selectively target pathogens without disturbing the microbiota. The aim of the study is further characterizing the behaviour of phages and its effect on the commensal microbial community.MethodsWe used 16S rRNA gene amplicon sequencing to evaluate the effect of SalmoFree® (a phage cocktail against Salmonella) on the microbiota of the cecum of broilers reared in a commercial farming system. Two field trials were carried out including 2 control and 2 experimental houses. Phages were incorporated in the broilers’ drinking water using three doses at the grower stage. The core microbiome, differentially abundant taxa between treatments as well as taxa associated with age were identified using Qiime2 and the R-Project for statistics.ResultsAnalyses of similarities among communities over time allowed us to identify two stages of microbiota development at the last stage of the production cycle. The core microbiome identified some key species in the adaptation of the microbiota at the last stage of the cycle. Among these there are important degraders of complex polysaccharides and producers of short chain fatty acids such as Eisenbergiella and Lachnoclostridium. Additionally, the phage cocktail did not affect the normal development of the microbiota structure while its application contributes to reduce the presence of pathogenic Enterobacteria including Salmonella. The addition of the phage cocktail led to a significant reduction in the abundance of Campylobacter and Helicobacter and increase in Butyricimonas and Rikenellaceae abundance which are common inhabitants of the chicken gut with known negative and positive effects in the health, respectively.ConclusionsOur results suggest that phages can contribute to improve the chicken health and reduce pathogens burden at the end of the production cycle of broilers. Furthermore, it presents for the first time a detailed look of the microbiota composition and changes in broilers in a commercial scenario, information that will valuable towards the implementation of phage therapy technologies in the field.


Abstract Background
The microbiota of broiler chicken gut affects positively or negatively the health, metabolism and immunity of chickens thus it has a signi cant impact on the animal productivity. Phages, host-speci c parasites of bacterial cells, are a promising antimicrobial alternative that selectively target pathogens without disturbing the microbiota. The aim of the study is further characterizing the behaviour of phages and its effect on the commensal microbial community.

Methods
We used 16S rRNA gene amplicon sequencing to evaluate the effect of SalmoFree® (a phage cocktail against Salmonella) on the microbiota of the cecum of broilers reared in a commercial farming system. Two eld trials were carried out including 2 control and 2 experimental houses. Phages were incorporated in the broilers' drinking water using three doses at the grower stage. The core microbiome, differentially abundant taxa between treatments as well as taxa associated with age were identi ed using Qiime2 and the R-Project for statistics.

Results
Analyses of similarities among communities over time allowed us to identify two stages of microbiota development at the last stage of the production cycle. The core microbiome identi ed some key species in the adaptation of the microbiota at the last stage of the cycle. Among these there are important degraders of complex polysaccharides and producers of short chain fatty acids such as Eisenbergiella and Lachnoclostridium. Additionally, the phage cocktail did not affect the normal development of the microbiota structure while its application contributes to reduce the presence of pathogenic Enterobacteria including Salmonella. The addition of the phage cocktail led to a signi cant reduction in the abundance of Campylobacter and Helicobacter and increase in Butyricimonas and Rikenellaceae abundance which are common inhabitants of the chicken gut with known negative and positive effects in the health, respectively.

Conclusions
Our results suggest that phages can contribute to improve the chicken health and reduce pathogens burden at the end of the production cycle of broilers. Furthermore, it presents for the rst time a detailed look of the microbiota composition and changes in broilers in a commercial scenario, information that will valuable towards the implementation of phage therapy technologies in the eld.

Background
The microbiota is de ned as the complete microbial community, including commensal, symbiotic and pathogenic microorganisms that reside on or within a complex multicellular organism, including plants, Salmonella can also contribute to a reduction in the broilers' feed intake and their growth rate. This effect on growth rate reduction has been estimated that could be up to 29% [20]. Additionally, two restricted host range serotypes, Salmonella enterica serovar Pullorum and Salmonella enterica serovar Gallinarum cause septicemic diseases in the chicken causing pullorum disease and fowl typhoid, respectively. These diseases have been largely controlled in Europe and North America, but still cause substantial economic losses in South America and Asia where the poultry is continuing to intensify and where open sided housing is common [21,22]. Controlling Salmonella outbreaks is thus a priority due to its health impacts and large economic losses. The pre-harvest stage in poultry is a potentially relevant control point, since it would aid preventing the introduction of the pathogen into the food chain and consequently reduce food poisoning among consumers [23]. However, to date, the most common practice to control the pathogen is with antibiotics, at the cost of the aforementioned undesirable side effects.
Phages, as host-speci c parasites of bacterial cells, are a promising antimicrobial alternative. Particularly, the use of lytic bacteriophages is an alternative that complies selective targeting of particular pathogens without disturbing the microbiota [24]. Phage therapy has been reported to show additional advantages, for example, the modulation of the immune system as well as the modulation of the microbiota of the host with the potential of improving host health [25]. Furthermore, phages have been proposed for animal therapy, prophylaxis and reduction of pathogen loads in food products of animal origin [26], thus, resulting an appropriate alternative for the control of Salmonella contamination in poultry. The use and research on Salmonella phages in poultry is not new, with reports including isolation and characterization of the phages, safety assessment and effectiveness of selected phages in chicken meat and in chickens in laboratory or controlled environments [27][28][29][30][31][32]. Although, most of these studies have been successful; however, to date there is only one report on the use of phages targeting Salmonella at a productive scale [33].
We have previously isolated, characterized and developed a phage cocktail that selectively targets Salmonella strains [33]. The cocktail, called SalmoFree ® , has the ability to control a broad range of Salmonella serotypes. The phages present in the cocktail have been characterized by host range, infection assays, stability in chlorine, transmission electron microscopy, genome sequencing and a safety assessment in broilers kept in cage batteries [33]. We have recently demonstrated the effectiveness of the cocktail to reduce the presence of Salmonella in a commercial farm [34], without affecting animals nor the production parameters, thus demonstrating its innocuity at the productive scale.
In order to further characterize, the behaviour of phages and its effect on the commensal microbial community, we studied the effect of SalmoFree ® in the cecum microbiota of broilers in a commercial farm, when phages were incorporated in the drinking water. This study is based on the same experimental set up recently published [34], and extended the analysis through the use of 16S rRNA gene amplicon sequencing. Altogether, this work expands the knowledge about the effect of phage therapy in the microbiota in broiler chickens and will help to reveal the performance of phages in the commercial farming conditions. Additionally, this work is the rst study to characterize the composition of broilers Two eld trials under commercial rearing conditions were carried out in a commercial broiler farm in Colombia [34]. This farm belongs to an integrated poultry company that typically handles the entire production and processing cycle of a chicken (hatching, feed, production, processing, and marketing).
Four production houses (labeled as houses 1, 2, 3 and 4) were selected due to the existing record of Salmonella presence detected during two previous productions cycles (data not shown). Chickens in houses 1 and 2 were treated with a control suspension (see below) whereas houses 3 and 4 were treated with the bacteriophage cocktail SalmoFree®. Treated ocks were separated from the controls by a distance of 300 m approximately. Information about houses' size, breed line, sex, and antimicrobial therapy per house and trial are presented in detail in Clavijo et al, 2019 [34]. Houses 1, 2, 3 and 4 correspond to 4, 8, 9 and 10 respectively in Clavijo et al, 2019.
The broiler production cycle in Colombia is carried out in two stages. The rst stage comprises the period from which one-day old chickens are received at the farm (day 1), until day 13-17. Chickens at this stage are fed with a starter diet. Following this diet, the grower stage spans from days 14 to 18 until day 35 to 42, where chickens receive a grower diet. Variation in the length of each period depends on several factors (market demand, weight of the chickens, schedule at slaughterhouse, among others). At the end of the second stage, chickens are sent to the slaughterhouse. SalmoFree® and the control suspensions were delivered to the animals in the drinking water during the second stage in three doses (one per week): at the beginning (day 18 for both trials), at the middle (day 27 and 26 for the rst and second trial, respectively), and one day before slaughter (day 35 and 34 for the rst and second trial, respectively). (Fig. 1).

Phage and control treatments preparation
SalmoFree® was prepared following a standard liquid lysate procedure using Salmonella Enteritidis s25pp. This strain was donated by Dr. Pilar Donado from the Colombian Integrated Program for the Antimicrobial Resistance Surveillance (COIPARS -CORPOICA) [35]. Typically, the preparation was carried out individually for each phage in nutritive broth (Sharlau), using an MOI of 0.1. Each lysate was centrifuged at 4 °C at 13,000 x g for 20 min and the supernatant was ltered through a 0.22-µm lter.
Approximately, 5.5L of each phage was produced per trial and stored at 4 °C. Individual phages were mixed in the adequate concentration and volume immediately before administering each treatment.
To evaluate the quality of the cocktail, a quality standard of sterility (zero bacteria) and a concentration of at least 10 10 PFU/mL were established. All doses were evaluated under these criteria. Sterility was evaluated incubating a 10 mL aliquot of the cocktail in 50 mL tubes at 37 °C for 24 h. Additionally, streaking from this culture was performed onto a nutritive agar plate (Sharlau). After 24 h of incubation, the absence of any type of growth was veri ed. Cocktail concentration was determined by carrying out serial dilutions from the cocktail suspension and plating the dilutions by the double agar overlay plaque assay [35].
The control suspension was elaborated using a fully-grown Salmonella culture that was lysed by the addition of chloroform at a nal concentration of 0.1% (v/v). The lysate was centrifuged and ltered as described above in the phage lysate procedure. The nal suspension was veri ed to be free of bacteria and phages. This control allows estimating if the cell residuals found in a normal lysate have any effect on the observed results.

Delivery of the treatments to the animals
The drinking water supply was removed 30 min before administering the treatments. This is a regular practice in the poultry production that is carried out in order to facilitate the uptake of any treatment due to the temporary shortage of hydration, but does not compromises the health of the chickens in any way.
The water supply tanks found in each farmhouse can store up to 1000 L of water. SalmoFree® and control suspensions were added to these tanks in a 100:1 water to treatment ratio. Thus, the nal concentration of the phage suspension was 10 8 PFU/mL. Treatments were delivered to the animals for 2.5 h, time in which is estimated that the animals drink the full 1000 L water supply.

Sampling methods
Cecum samples were taken one day before and after delivery of SalmoFree® treatments, according to Fig. 1. Additionally, samples one day before the start of the growing phase (day 15 and 14 for the rst and second trial, respectively) and at the abattoir, after the slaughter, were collected. To get all samples, ve female chickens from each of the four houses were randomly selected. These chickens were sacri ced and their cecum was removed in the most aseptic possible way in the farm. Each cecum was placed into sterile plastic bags (Nasco, USA) and transported in liquid nitrogen to the laboratory where they were stored at -80 °C and promptly processed for DNA extraction. A total of 320 samples were collected corresponding to 160 per trial (5 ceca per house per 4 houses per 8 sampling days). Since sampling days occurred at different points of the growth cycle between trial 1 and trial 2 ( Fig. 1), we renamed the samples according to each treatment dose. The sampling point corresponds to the number of dose (1)(2)(3) followed by a letter indicating whether it was taken one day before (b) or after (a) the corresponding dose.
Besides the ceca samples, individual cloacal swaps samples were taken in order to detect Salmonella using a genus-speci c PCR as described in Clavijo et al, 2019 [34]. Each individual swab corresponded to the same individual chicken sacri ced for cecum extraction.
DNA Extraction and 16S rRNA gene ampli cation From ceca samples, 180-200 mg of the cecal content were aseptically collected in dry ice, avoiding the thawing of the samples. Immediately, the samples were processed through the QIAamp DNA Stool Mini Kit (Qiagen, Valencia, CA) according to manufacturer's instructions. Samples were measured in a Nanodrop ND-1000 spectrophotometer (Thermo Scienti c, Wilmington, USA) to assess the DNA quality and were quanti ed using a Qubit uorometer (Life Technologies, Paisley, UK). After DNA quanti cation, samples were diluted with elution buffer (Qiagen, Hilden, GM) to a concentration of 5 ng/µL. From these diluted samples ampli cation of the V4 hypervariable region of 16S rRNA gene was achieved using the following primers: Primer1 5´TACACGACGCTCTTCCGATCTGTGCCAGCMGCCGCGGTAA 3´ and Primer2 5´AGACGTGTGCTCTTCCGATCTGGACTACHVGGGTWTCTAAT 3´.
The bold region of the primers corresponds to the universal 515F and 806R primers. Each PCR reaction used buffer 1X, dNTP´s 10 mM, Primer1 10 µM, Primer2 10 µM, Phusion High-Fidelity DNA Polymerase (0.02U/µL) and 1.5 µL DNA (7,5 ng on average) in a nal volume of 20 µL. The con guration of the temperature pro le for the reaction was as follows: Initial denaturation 94 °C for 3 min; 30 cycles at 94 °C for 45 s, 56 °C for 30 s y 72 °C for 30 s; and the nal extension at 72 °C for 7 min. The PCR procedure was carried out in triplicate with a negative control in which water was added instead of DNA sample. PCR products were visualized using Gelred through 1.5% (w/v) agarose gel electrophoreses. Finally, after the con rmation of the ampli cation, the PCR triplicates were mixed in one pool and were kept at -20 °C until further processing.

Libraries preparation and sequencing
After the rst ampli cation, a second PCR was carried out using a pair of primers with the Illumina adapters and indexes to demultiplex the samples. Sequences of the primers are: The underlined region corresponds to the location of a particular index (the sequences for all primers are presented in Supplementary material Table 1); Each sample is ampli ed with a pair of primers with a different sequence in this region, in order to allow pooling all samples in the same sequencing run and demultiplexing afterwards. The second PCR was carried by adding 5 µL of the pooled product from the rst PCR to a mixture with 4 µL of water, 10 µL of GoTaq Green Master mix (Promega) and 0.5 µL of each Illumina index primer (0.25 µM) which was ampli ed using the following PCR conditions: 3 min at 94 C, and 12 cycles of 45 s at 94 C, 60 s at 55 C and 30 s at 72 C and a nal period of 7 min at 72 C and kept at 4 C. Then, a puri cation process was performed by using 18 µl of AMPure beads (Beckman Coulter) and eluting samples with 15µL of Tris buffer (10 mM, pH 8.5). Next, the concentration of this mixture of puri ed amplicons was determined with the Qubit uorometer (Life Technologies, Paisley, UK) followed by pooling all the libraries into equimolar concentrations. Paired-end sequencing (2 × 250) of this pool was conducted on an Illumina MiSeq platform at Washington University in Saint Louis, Center for Genome Sciences and Systems Biology.

Bioinformatics' and statistical analyses
Sequences were pre-processed, quality ltered and analyzed using QIIME2 version 2018.11 [36] (https://qiime2.org) and its plugins. The input les used were the demultiplexed paired-end fastq les generated in Casava format (Illumina) and a mapping le. Raw sequencing data was imported and demultiplexed through the Casava 1.8 paired-end demultiplexed fastq protocol. Then, adapters were removed using the cutadapt plugin [37] and subsequently the fastq sequences were merged using FLASH software [38]. DEBLUR software package [39], included in QIIME2, was used for modelling and correcting Illumina sequences. This process integrates chimera removal, truncation of reads and the collapse of reads into amplicon sequence variants (ASVs). All parameters were used by defaults except for truncation of reads: -p-trunc 214. The detection limit of the method (1 × 10 − 4 ) was established based the maximum number of rare ed reads per sample (10,000 per sample) implying that any ASV at an abundance less than 1/10,000 will not be detected.
ASVs were ltered using QIIME2 q2-feature-table lter features command, keeping only features with a frequency higher than 10, in general corresponding to a minimum relative abundance of 0.001 and present in at least 2 samples. A second ltration was done retaining samples with more than 3,000 sequences after ASV ltration. Taxonomy assignment to the ASVs was performed using QIIME2 q2feature-classi er plugin and the Naïve Bayes classi er that was trained on the SILVA database (version SSUParc_100) [40]. Alpha-and beta-diversity analyses were performed with the q2-diversity plugin at a sampling depth of 10,000. Alpha diversity was calculated using Shannon's diversity index, observed OTUs and Faith's Phylogenetic diversity. Kruskall-Wallis test was used to test for differences in mean alphadiversity between experimental treatments and trial, farmhouse, genetic line, sampling point and dose variables. Distance matrices for beta-diversity were constructed using Bray Curtis and weighted Unifrac metrics. Permutational multivariate analysis of variance (PERMANOVA, P < 0,05) using group signi cance command was used to analyse spatial variation in beta-diversity and effects of experimental treatments and the other variables. Additionally, Unweighted Pair Group Method with Arithmetic Mean (UPGMA) clustering analysis based on Bray Curtis and weighted UniFrac distance was performed. Signi cant differences in taxons between the groups were assessed with the ANCOM test in QIIME2. Speci c analyses were carried by week collapsing samples between cycle days 15-21 in week 3, days 22-28 in week 4 and days 29-36 in week 5 ( Fig. 1). Additionally, the phyloseq package [41] in R was used to determine the core microbiome and also to analyse the abundance of the predominant phyla over time.

Bioinformatics' Salmonella analyses
Given that none of the identi ed ASV's were assigned to the genus Salmonella, but several of them were annotated as Enterobacteriaceae family, the next step was to determine if some of those ASVs in the feature le were closely related to Salmonella. For this purpose, a collection of Salmonella, Escherichia coli and Shigella 16S rRNA genes deposited in the NCBI Refseq collection were retrieved. These sequences (Supplementary material table 2) were used along with the sequences of ASV's assigned to the Enterobacteriaceae family for a multiple alignment using MUSCLE [42]. Following, the alignments were manually inspected and edited using the sequence editor Jalview (version 2) [43], looking for speci c positions that constitute speci c markers for the Salmonella group. Afterwards, a maximum-likelihood phylogenetic tree was constructed using FastTree version 2.1 [44] and visualized in Figtree [45]. From this analysis, it was possible to identify the ASV that most likely corresponds to Salmonella.
A similar characterization of the ASVs assigned to Helicobacter was performed since some species of the genera Helicobacter are considered human or animal pathogens. These ASVs were aligned with the collection of Helicobacter 16S rRNA gene sequences deposited in the NCBI Refseq collection (Supplementary material table 3)

Results
The current study focused on the evaluation of the effect of the Salmonella phage cocktail on the cecum microbiota of broilers in a commercial scenario while being incorporated as treatment in the broilers' drinking water. The experimental set-up was the same as previously published, where it was performed the evaluation of the effectiveness of the phage cocktail by comparing the presence of Salmonella and the production parameters between the two treatment groups; also, detection of phages was performed [33].
The summary of those published results is presented in Table 1 and Table 2.

Sequencing information
From a preliminary analysis, we determined that the sequence variants generated for the F2 farmhouse two (Control farmhouse) in trial 2 were signi cantly different in composition and diversity from the other samples (Supplementary material Fig. 1), likely re ecting a technical rather than a biological effect, thus all samples from that farmhouses were excluded from further analysis. Additionally, sequences generated from most of the samples taken one day before the beginning of the growing phase (day 15 and 14 for the rst and second trial, respectively) had a very low extraction yield of DNA leading to a high frequency of failed ampli cations and sequencing; likely a consequence of low biomass due to the early age of the chicks. Thus, all the samples from these time points were excluded from further analysis (Supplementary material Fig. 1). Finally, 244 samples were kept for the rest of analysis.
After quality control, 228 samples were retained and we obtained 4,995,664 sequences with a total of 3,993 amplicon sequence variants (ASVs). The frequency of reads per sample varied from 3,527 to 210,082 with an average of 21,911 and a median of 18,130 per sample, which is comparable to previous studies [46,47].

Microbial diversity
The evaluation of rarefaction curves based on rare ed Shannon indices indicates that a su cient sequencing effort was obtained to characterize the most abundant bacteria affecting the diversity index for all samples, as represented by a plateau in the curve (Supplementary material Fig. 2A and B), however, the observed OTUs metric shows that even at a sequencing depth of 11,000 reads per samples, it was still possible to detect new ASVs as a function of sequencing, likely transient or very low abundant species, as they didn't have an effect on the Shannon indices (Supplementary material Fig. 2C and D).
Alpha diversity analyses for both trials using Faith's phylogenetic diversity and number of observed OTUs suggested a slight increase over time. This temporal phenomenon is more prominent in trial 2. However, there was not a high variation trend in diversity between the days of the experiment (17-36) since their Faith indices fall within a comparable range of 20 to 33. This small range might occur because the microbial community is reaching a relatively stable, yet dynamic, state (Fig. 2). This behaviour is in agreement with previous reports where the microbiome stabilizes at approximately day 12 [48]. Regarding the maturation process of the microbiota, the microbial communities have a similar diversity at day 17 between samples (3rd week), at the beginning of the experiment. Also, it is possible to notice that between the 4th and 5th week, the microbiome increases its variability and stabilize again at the last day of the trial, as it remains constant at the slaughterhouse (Fig. 2). Comparing trials, higher alpha-diversity index values (Faith, Shannon and Observed OTUs) were observed in trial 1 compared to trial 2 (Fig. 2).
Alpha diversity indices changed signi cantly (Kruskall Wallis test, P < 0.05) in response to the trial, farmhouse, treatment and markedly by the age of the animal. Conversely, the genetic line did not show signi cant differences (Kruskall Wallis test, P > 0.05) in microbial diversity. Although the treatment showed signi cant differences, there is no observable pattern to discriminate among treatment groups.
Alpha-diversity results are complemented by the beta-diversity comparison between weeks, treatments and trials (Fig. 3A, B, C, D). Principal Coordinates Analysis (PCoA) plots based on treatment did not reveal any clustering pattern (Fig. 3A) while PCoA plot based colored by age and trial showed that microbial communities were driven mainly by these two variables (Fig. 3B, 3C). Interestingly, regardless of the trial, the microbial community showed signi cantly higher similarity at the beginning of the experiment (Fig. 3B, 3D), and diverged as a function of time, becoming more distant at the second dose of the treatments. The trial dependency suggests a high contribution of the environment in the development of the microbiota, which is expected due to the uncontrolled variables and their variation in a broiler farm, such as temperature, humidity, feed composition, litter replacement, feeding and antibiotic intervention, among others. The observation of complex but highly similar diversity pattern in older birds agrees with previous studies where microbial communities exhibit similar patterns as chickens reached their marketing age [48].

Taxonomic Composition of the Bacterial Community
We performed a general analysis of the taxonomic composition at phylum level of the community in order to characterize their behaviour in a commercial scenario throughout the last stage of the production cycle.
In addition, we analysed the minimum community of microbes that is essential for the host (i.e. core microbiome) for both trials, at genus level.
First, the taxonomic composition of trials 1 and 2 indicated that Firmicutes is the most predominant phylum in the gut of broilers (46.4% abundance), followed by Bacteroidetes (37%). Together, both phyla account for more than 80% of the relative abundance within the community. Then, in lower proportion, the phylum Epsilonbacterota (4.47%), Proteobacteria (3.2%), and Tenericutes (1.21%) are found, while unassigned bacteria at phylum level accounted for 6.41% of the abundance.
Regarding phyla dynamics, Firmicutes and Bacteroidetes remain relatively constant throughout the experiment (Fig. 4), only a slight decrease in Firmicutes, with a corresponding increase in Bacteroidetes, was observed, the day after the second dose (Day 28 and 27 for trial 1 and 2, respectively) suggesting a complementation between this two phyla. Proteobacteria behaves similarly in the two trials, maintaining relatively constant numbers over time where the lowest point was found after the second dose as well.
The presence of bacteria of the phylum Tenericutes is higher at the beginning of the experiment with a slight reduction over time (Fig. 4). Epsilonbacteria phylum was present at a very low abundance at the beginning of the experiment but increased rapidly starting at the 4th week of the cycle (before the second dose) and remain constant until the end of the experiment (Fig. 4).
To better understand the shared taxa occurring over time, we analysed the core microbiome discriminated by trial and week (Table 3). Core microbiomes were de ned as those taxa that were present in more than 50% of samples of each week sampling point. When comparing the core microbiome for both trials over time, it was observed that the genera conforming the broilers' microbiome is highly conserved since most genera (78%) were present in both trials (Table 3). This result is interesting because despite being different in terms of a) alpha and beta diversity; b) breeds; and c) antimicrobial regime, both trials share the core genera.
Likewise, the analysis of the core microbiome over time showed that the microbiota, during the period examined, was also conserved: 65% of the genera were found in most of the samples during the three weeks ( Table 3).
The core microbiome at the slaughterhouse contained all the genera found during the experiment.
Consequently, the main members of the community were maintained during transportation of the broilers to the slaughterhouse (Table 3).
Taxonomic dynamics at the grower stage of the production cycle The differential analysis discriminated by week allowed the identi cation of some key species in the adaptation of the microbiota during the grower phase of broilers. For instance, in both trials several genera revealed an increasing abundance trend over time, including Alistipes, Rikenellaceae, Phascolarctobacterium, Desulfovibrionaceae and Megamonas; while Bacillales, Coprobacter, Barnesiellaceae and Ruminococcaceae decreased. Another taxon with an intriguing abundance was Odoribacter, which increases between the 3rd and 4th week and then remains relatively constant until the end of the cycle. In contrast, Hydrogenoanaerobacterium is only detected at the beginning of the cycle (3rd week) and then disappears (Fig. 5).
Abundance differences in the microbial taxa between the treatment groups As the structure of the microbiota was markedly in uenced by the trial, and additionally, given the antibiotic intervention in the rst trial (Fig. 1), a differential analysis was carried out separately in each trial in order to de ne the differences between the farmhouses treated and not treated with the phage cocktail.
We performed and independent PcoA analysis by Trial. This analysis showed a separation by treatment after the second dose but this clustering disappeared towards the end of the experiment. This separation is clearer for trial 1 (Fig. 6), although a more compact clustering of all samples can be observed towards the last days of the experiment for both trials.
The differential abundance analysis among doses revealed four genera as signi cantly associated with the treatment; the taxa Campylobacter, Helicobacter, Rikenellaceae and Butyricimonas (Fig. 7). Campylobacter in the rst trial appeared between the rst and second dose in both control and treated samples, however, the abundance in treated farmhouses slowly increases its abundance in contrast to the abrupt increase of the control groups during the second dose with a subsequent decrease, leading to a nal convergence (for treatments and controls) in abundance at the end of the experiment. Interestingly, opposing to Campylobacter´s under-representation in treated samples, the closely related Helicobacter seems to increase its abundance. Butyricimonas abundance increased signi cantly respect to the control after the second and third doses in trial 1 (ANOVA, p < 0.005). Similarly, Rikenellaceae increased its abundance after the second doses and is signi cantly different, compared to the control group (Fig. 7A).
Abundance patterns of particular taxa displayed similarities in both trials. In both cases, Rikenellaceae was present at a higher abundance in treated chickens compared to the control in the last week of the experiment. Even though the tendency to increase abundance over time is observed for treatment groups in both trials, trial 2 showed overall lower abundances than trial 1, as seen for Campylobacter and Helicobacter (Fig. 7B). In addition to these taxa, the genera Parasutterella that was not detected in trail 1, increased its abundance over time in trial 2.
Given the differential abundance of Helicobacter seen on the treatment groups and since some species of the genera Helicobacter are considered human and animal pathogens, a further characterization of the ASVs assigned to Helicobacter was performed. These ASVs were aligned with the collection of Helicobacter 16S rRNA gene sequences deposited in the NCBI Refseq collection. After the bioinformatics analysis, these ASVs were identi ed as Helicobacter pullorum (Supplementary material Fig. 4).

Enterobacteriaceae abundance Analysis
Throughout the analysis implemented for Salmonella, one ASV from 32 was clustered in the clade that grouped Salmonella enterica subsp. enterica and was separated from the clades that grouped Salmonella bongori and Escherichia coli-Shigella (Supplementary material Fig. 5). Thus, we established that this ASV corresponds to Salmonella sp. The others ASVs assigned to Enterobacteriaceae family were grouped in the E. coli-Shigella clade.
The abundance of the Salmonella-related ASV across samples was rather low, in a range of 2.5 × 10 − 5 to 6.8 × 10 − 4 , and close to our estimated con dence detection limit of the method (1 × 10 − 4 ). Moreover, the ASV was detected in a small number of samples (n = 22). Thus, further inferences based on its abundance were discarded. In addition, no correlation was found with the abundance of this ASV and the molecular detection of Salmonella (p > 0.01) (Supplementary material Fig. 6). This result indicates that, despite identifying speci c regions to discriminate Salmonella from the other members of the Enterobacteriaceae family, due to its very low abundance it would be necessary to carry out a deeper sequencing in order to detect it within accurate detection levels.
We resorted to analyze patterns of the family Enterobacteriaceae due to the di culties of detecting Salmonella sp. at accurate detection levels. This choice was justi ed based on two observations: (1) family Enterobacteriaceae showed a positive correlation with the molecular detection of the genus Salmonella (ANOVA, p < 0.001) (Fig. 8A), and (2) a negative correlation with the detection of phages (ANOVA, p < 0.001) was observed, meaning that the abundance of the family was lower in samples where phages were detected (Fig. 8B). Here the Salmonella and phage detection are referred to the molecular detection carried by the ampli cation of the speci c nucleic acid sequences of Salmonella sp. and the phage tail ber gene, respectively [34].
The abundance of Enterobacteriaceae throughout the days, discriminated by trials and treatments, is shown in Fig. 8. In trial 1 at the beginning of the experiment, the abundance of the family was lower in the farmhouses treated with the phage cocktail. This result was expected due to the antibiotic intervention that occurred in week 1 because the antibiotic used is active against several gram-negative bacteria. In treated farmhouses the low abundance of Enterobacteriaceae is maintained until the last week in the farm and also at the slaughterhouse. Control farmhouses also exhibited a signi cant reduction after the second dose (Fig. 8C). In trial 2 after the second dose, a high reduction of the Enterobacteriaceae abundance is identi ed for both treated and non-treated houses. This abundance was lower than the results obtained in trial 1 and was also re ected in the samples taken at the slaughterhouse. Notably, the abundance of this group is increased before the second dose for treated houses in trial 2 (Fig. 8C). This increase might be due to a high mortality rate in the farmhouse 4 (a treated house) in this period of time that could lead a higher abundance of pathogens of this group re ecting in the abundance of Enterobacteriaceae (data shown in [33]).

Discussion
The aim of this study was to determine the effect of the Salmonella phage cocktail SalmoFree® in the composition of the microbiota of the cecum of broilers. The study was performed supplying the phage cocktail to chickens reared with commercial purposes at a farm under production scale conditions. Our study also included the characterization of the dominant bacterial microbiota present at the grower stage of broilers in production conditions. Our results suggest a process of normal microbiota maturation characterized by a transition towards a higher diverse community. This observation was independent of the treatment applied, further demonstrating that phages do not affect the normal development of the microbiota.
Surprisingly, our results suggest that the behaviour of the microbiota in the period of the experiments (age 17-36 days), corresponding to the grower phase of the chickens, is similar to what have been previously reported in experimental chickens reared in a controlled environment [48,49]. For instance, our observations support the stabilization dynamic of the microbiota at this developmental stage. Likewise, the observation of a microbial community similar in older chickens to the one observed at slaughter age was consolidated (Figs. 2 and 3). Additionally, the age of the animal was the variable that had the highest in uence over the microbiota variation, in agreement with previous studies [48,50]. This is a key evidence suggesting that microbiota approximations done under controlled environments do not differ largely from farming conditions. Analyses of similarities among communities over time allowed us to identify two main moments of microbiota development at the last stage of the production cycle (PERMANOVA, P < 0,05) (Fig. 3). The 3rd week, representing the rst week of the experiment and the week just after the change to the grower diet, is where the community seems to be more uniform and with a signi cantly higher abundance of bacteria such as Ruminococcaceae, Bacillales, Coprobacter, Hydrogenoanaerobacterium and Barnesiellaceae (Fig. 5) compared to Parasuterella and Flavobacteriaceae. The other two weeks, the 4rd and 5th weeks, exhibit a higher variation in abundance and diversity, which could be attributed to the change in diet.
Nevertheless, further studies are necessary to con rm the generalization of our current results to other farms and conditions. Analysis of the dynamics of the core microbiome identi ed members that were reported previously in the literature as being part of the most abundant genera in the microbiome of chicken cecum [48]. In one previous study, authors performed a comprehensive day-to-day microbiome analysis of the chicken cecum from day 3 to 35 using experimental chickens in a controlled environment. Authors identi ed the most abundant genera, also identi ed in our core analysis, as: Escherichia.Shigella, Eisenbergiella, Ruminiclostridium, Flavonifractor, Anaerotruncus, Faecalibacterium, Lachnoclostridium, Megamonas, Intestinimonas, Shuttleworthia, Subdoligranulum, Tyzzerella, Lactobacillus, Blautia and Erysipelotrichaceae, among others ( Table 3).
The core composition of the microbiota identi ed key members at the last stage of the production cycle ( Fig. 3) (Table 3). Those microorganisms may be responsible for important metabolic processes in the microbiota of broiler chickens. Among these, there are some important degraders of complex polysaccharides and producers of short chain fatty acids (SCFA). For instance, Eisenbergiella, Lachnoclostridium of the family Lachnospiraceae play an important role in the production of butyrate which is the preferred energy source for the gut epithelial cells [51]. Another butyrate producer found in the core was Intestinimonas [52]. Megamonas and Bacteroides were detected as well; these bacteria are known to produce propionate as the main end product of the degradation of complex plant polysaccharides. Although propionate is a less preferred energy source than butyrate, its production might represent an e cient balance between energy acquisition from available nutrients and sustained growth [50]. Other bacteria present in the core microbiome involved in producing SCFA were Subdoligranulum, Faecalibacterium Alistipes, Coprobacter, Blautia and Butyricimonas [53].
On the other hand, Campylobacter, Helicobacter and Megamonas are bacteria carrying hydrogenases, which have the potential to serve as hydrogen sinks that facilitate succinate production [54]. Succinate is an important metabolite in both host and microbial processes [55]. Meanwhile, the presence of Oscillibacter, a Clostridium cluster IV member, has been identi ed as an anaerobe producer of valerate and associated with diet-induced obesity [56]. Surprisingly, Bi dobacterium, a butyrate producer, was not detected while it has been reported consistently as a dominant member of the chicken microbiota [48][49][50].
Comparison analyses between treated and control farmhouses must be done with caution, because of the conditions and characteristics of the current trials (variation of temperature, humidity, feed composition; the antibiotic intervention, and phage's cross-contamination). Altogether, it seems that the treatment patterns visualized in trial 1 are not evident in trial 2. For example, in trial 1 the diversity seems to be higher in the control group, also in trial 1 the relative abundance of Campylobacter sp. is lower in the treatment group while in trial 2 this abundance is low in both experimental groups. We proposed that these dissimilarities might be an effect of the presence of phages due to cross-contamination starting at the beginning of the trial 2 in all treatments. This event could generate a bias in trial 2.
In this line, our data suggest a low trend of alpha diversity in treated farmhouses. Although an increase in microbial diversity in the gut has been linked to improved health in the elderly [57], a reduced alpha diversity in samples subjected to phages has been reported previously as well as in trials with probiotics in chickens, where treated groups exhibit a lower diversity compared to non-treated groups [58]. An alternative explanation for the observed reduction of diversity could lie in the presence of opportunistic pathogens [59]. In some cases, the microbiota associated with disease is more diverse due to the presence of opportunistic pathogens that have the capacity to colonize this niche. Thus, the alpha diversity reduction could be attributed to a lower presence of opportunistic pathogens.
Differential analyses by treatment group determined that few taxa are signi cantly associated with the addition of phages, supporting the statement that phages are not affecting the structure of the microbiome. Particularly interesting is the evidence of the reduction of Campylobacter in treated farmhouses. Campylobacter is also considered an important food-borne pathogen associated to the consumption of poultry products and its importance in public health is substantial [60]. In addition, when this opportunistic pathogen is highly abundant in chickens, it has been demonstrated to cause damage to the gut [61]. The observed reduction of Campylobacter might suggest a bene cial modulation of phages while the reduction in Salmonella and pathogenic Enterobacteriaceae such as Escherichia coli represents a great added value to the application of SalmoFree® in broilers. Additionally, the presence of Campylobacter jejuni has been associated with a higher abundance of E coli not only in broilers [62] but also in humans [63]. It is important to highlight that the reduction of Campylobacter is likely occurring as a secondary effect of the phage lysis on Salmonella, since is not expected that phages infect strains of this genera given the speci city of the cocktail towards Salmonella sp. strains. Furthermore, the correlated increase in Helicobacter abundance with the decrease in Campylobacter supports the proposal of a competitive dynamic between these two genera [62].
The increase of Butyricimonas and Rikenellaceae following the phage treatment is noteworthy. These two genera are reported as bene cial bacteria in chickens due to their enrichment in samples treated with probiotics [58].
Since the relative abundance of Salmonella could not be obtained, the analysis of Enterobacteriaceae family is highly important to elucidate the performance of the phage cocktail. We consider this analysis useful since members of this family are phylogenetically closely related. Enterobacteria such as E. coli and Salmonella enterica are facultative anaerobic pathogens that occupy the same niche [64,65], hence knowing the behaviour of the family will help us to understand what happened with Salmonella. Furthermore, the Enterobactereaceae family has been reported as one consistent biomarker of poor gut health in poultry [66], as it has been negatively correlated with performance of production variables [67].
Regarding the Enterobateriaceae analysis, our data con rmed that in the presence of the phages the abundance of this taxon is reduced. Although the reduction of this group was observed for both experimental groups in trial 1 and trial 2, it might be a consequence of the cross contamination of phages that occurred in control houses (Table 2). Apparently, the higher effect of phages is around the 4th week of the production cycle; this time point also corresponds to the moment where more microbial changes are occurring. This nding identi ed the importance of the second dose and suggests it as a relevant dose in the regime proposed for SalmoFree®. The result that the reduction on the second trial was even better, let us to hypothesize that the reduction of the target could be higher with the application of phages by consecutive cycles.
Equally important is the maintenance of low abundance at slaughter. Hence, the application of phages is achieving its main objective that is to reduce the presence of Salmonella (an Enterobacteriaceae member) when chickens arrive in the slaughterhouse in order to impact directly the risk of food poisoning and reduce infection cases in humans.

Conclusion
This study con rmed the safety of the SalmoFree® cocktail on the health and microbiome health in chickens. Its delivery to broilers did not affect the normal development of the microbiota structure.
Compelling evidence was obtained on the SalmoFree® contribution to reduce the presence of pathogenic Enterobacteria including Salmonella. Additionally, our results suggest that the use of the phage cocktail might be proposed as a bene cial microbiota modulator improving host health. This differential effect could contribute to improve the chicken health and reduce pathogenic burden at the end of the production cycle of broilers. It is important to highlight that our results were obtained not via an experimental setup, but in situ at the production farm, giving a closer idea of what the real effect of the phages under commercial settings will be. These ndings reinforce the knowledge that phages targeting Salmonella may constitute an e cient prevention measure to avoid food poisoning outbreaks associated with the pathogen and reduce the problematic widespread use of antibiotics in the poultry industry  Trial I  Trial II   1b  1a  2b  2a  3b  3a  1b  1a  2b  2a

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. AdditionalFile1.pdf AdditionalFile2.pdf AdditionalFile3.docx