A total of 120 female pigs (Pietrain × [Large White × Landrace]) with an initial body weight (BW) of 25.4 ± 3.7 kg were used in this experimental study. Pigs were sourced from SEARA Foods (SEARA Alimentos), Santa Catarina, Brazil. The pigs were distributed in a randomized block design in a 2 × 2 factorial arrangement composed of two sanitary conditions (Barn-SC): good (GOOD) or poor (POOR—S. Typhimurium inoculation + poor hygiene) and were fed with one of two diets [control (CN), formulated to meet the nutritional requirements according to (85); or surplus supplemented (AA) with Thr, Met+Cys, and Trp ratios to Lys adjusted to 20% above the CN diet], as previously described (9). The experiment was conducted in two similar barns located at the facilities of the Swine Research Laboratory at São Paulo State University (UNESP-Jaboticabal, São Paulo, Brazil). Pigs were identified and randomly assigned to one of two similar growing-finishing barns (0.9 m²/pig), where they had been housed for 14 days in the same facilities prior to the beginning of the experiment. The full concrete floor of both barns underwent the same cleaning and disinfection protocol before allocating all pigs. For both barns, the temperature was controlled (set to 22°C) with the aid of exhaust fans and an automated evaporative system of pad cooling (Big Dutchman, Araraquara, SP, Brazil) and monitored using data loggers (Hobo, Onset Computer Corporation, Bourne, MA, United States). Artificial lights were used to maintain a 12-h photoperiod (7:00 a.m.–7:00 p.m.). Additionally, each room was equipped with four Automatic and Intelligent Precision Feeders (AIPF, University of Lleida, Lleida, Spain) and six nipple drinkers, allowing ad libitum access to feed and water, respectively. AIPF’s feed delivery is achieved through recognition of the individual radio frequency ear tag (Allflex, Joinville, SC, Brazil) attached to the right ear of each pig, enabling individualized data collection, as previously described (9, 46).

The S. Typhimurium strain (RLO971/09, originating from a past swine clinical salmonellosis case) was used to prepare the experimental inoculum. This strain is stored at the Laboratory of Ornitopathology, Department of Veterinary Pathology, UNESP-Jaboticabal, SP, Brazil (9, 11). A single colony was grown overnight to make a total volume of 5 mL of brain heart infusion (BHI, CM 1135, Oxoid, Thermo Fisher Scientific, NH, England) broth, which was used to inoculate 2 × 10⁹ colony-forming units (CFU) of S. Typhimurium selected for antibiotic resistance to nalidixic acid (25 μg/mL) by oral gavage in all 60 pigs allocated into the POOR (S. Typhimurium inoculation + poor hygiene) SC barn, as previously described (9). The counterpart pigs allocated in the GOOD SC barn were sham-inoculated via oral gavage with 5 mL of brain heart infusion broth solution as negative controls, following the same procedures done on pigs allocated in the POOR (S. Typhimurium inoculation + poor hygiene) SC barn.

Under both inoculation conditions, all pigs were fasted for 6 h and had no water consumption for 1 h prior to inoculation. After pigs in the POOR (S. Typhimurium inoculation + poor hygiene) SC barn was inoculated with S. Typhimurium, fresh manure from a commercial pig farm was spread across all pens to establish the initial sanitary challenge and remained uncleaned throughout the study, as previously described (9). In contrast, in the GOOD SC barn, no manure was spread on the floor, while it was cleaned twice a day with a water jet stream, and potassium monopersulfate (1:200; Virkon; Lanxess, Colony, Germany) was pulverized into the air once a week as part of the biosecurity protocol to improve the hygiene condition. The management teams were required to wear clean clothing and footwear with a bleach solution (1:10 dilution) when entering the building to avoid cross-contamination (9). Regardless of sanitary conditions, pigs did not receive any growth promoters or any other treatment prior to or during the experiment. Cross-contamination between barns was assessed by testing fecal samples at day post-challenge (DPC) 0, 10, and 21 for Salmonella spp. growth using selective media (9), and none was found. All experimental procedures applied in this trial followed the guidelines of the Brazilian National Council of Control of Animal Experimentation (CONCEA) and were reviewed and approved [protocol no. 4784/20] by the Ethical Committee on Animal Use (CEUA) of Faculdade de Ciências Agrárias e Veterinárias (FCAV/UNESP—Jaboticabal, SP, Brazil) (9).

As previously described (9), all experimental diets were based on corn-soybean meal and were formulated using the reported nutrient content and analyzed AA composition of ingredients to meet or exceed the nutrient requirements based on pig weight at the beginning of the study, according to (85) and AMINODat^® 6.0 (Evonik Operations GmbH, Essen, Germany). AMINODat is a web-based, internationally referenced database for the nutritional composition of feed ingredients. The diets were steam-pelleted (2.5 mm) and provided ad libitum without the inclusion of in-feed antibiotics as growth promoters throughout the trial.

Rectal samples were collected from all pigs at DPC 0, 10, and 21. Specifically at DPC 0, fecal samples were collected prior to S. Typhimurium inoculation, and sample aliquots were separated for bacterial counts (fresh samples) and DNA extraction for subsequent 16S rRNA sequencing. All aliquots used for DNA extractions were frozen in liquid nitrogen and then stored at −80°C. DNA extraction was performed using the Qiagen DNeasy® PowerSoil® Pro kit in accordance with the manufacturer’s instructions (12). All DNA extracts were eluted in 60 μL of elution buffer and then frozen at −80°C prior to quality assessment and sequencing. DNA concentration was checked using a Nanodrop One spectrophotometer (Thermo Fisher Scientific Inc., Middletown, VA). For 16S rRNA sequencing, the V4 region was amplified from each sample using an in-house modified dual-indexing sequencing strategy on the Illumina MiSeq platform, as previously described (12, 47, 48). Paired-end sequences were analyzed using the Quantitative Insights Into Microbial Ecology (QIIME) program (version 2) (49). Sequences were truncated (220 bases for forward reads and 160 bases for reverse reads) and denoised into amplicon sequence variants (ASVs) using DADA2 (50). They were then rarefied to 5,000 reads per sample, as previously done (12). All ASVs were assigned taxonomic information using the pre-fitted sklearn-based taxonomy classifier from the SILVA database (release 138) (12, 48, 51). Only bacterial taxa (contained domain = “Bacteria”) with genus-level information in the name (g from QIIME2 output) were filtered for both diversity and taxonomic analyses prior to statistical analysis. Each fecal sample was treated as a unique experimental unit for microbiome sequencing with no replicates added.

Network construction and visualization were performed using the NetCoMi package within the R statistical framework (version 4.2.2) (52). Overall, samples were rarefied to 5,000 reads and grouped by sampling at DPC 21 and sanitary conditions [GOOD vs. POOR (S. Typhimurium inoculation + poor hygiene)], as previously described (12). In brief, each network was constructed using a centered log-ratio transformation and Spearman-based correlations, employing the 50 most abundant microbial taxa. Data interpretation was summarized based on each node representing a bacterial taxon, and the size of each node is scaled according to its centrality. Edges represented associations between taxa, with positive correlations being colored in green and negative ones in red. The thickness of each edge corresponds to the strength of the association, and edges representing a value less than 0.5 were excluded from the analysis. Taxa were colored based on clusters calculated in network construction. Therefore, the taxa with the highest degree of centrality were selected to highlight cluster-specific dominance (12).

All bacterial taxonomic outputs were processed for quality control (data distribution) to statistical modeling using R version 4.4.0. All data exploration and statistical analysis were conducted using bacterial genus-level taxonomy only, except for an examination of the most important taxa (>1%) over time, which was also performed using family-level taxonomic resolution when necessary for data amalgamation. In brief, the tidyverse library (version 2.0.0) was used for data exploration, analysis, and plotting. Statistical analysis was conducted at two levels: (1) by comparing Barn-SC GOOD vs. POOR (S. Typhimurium inoculation + poor hygiene) by DPC (in the absence of interaction between diet and Barn-SC, with Barn-SC being the main significant effect); and (2) by comparing all treatments [combination between Barn-SC and diet by DPC—GOOD CN, GOOD AA, POOR (S. Typhimurium inoculation + poor hygiene) CN, and POOR (S. Typhimurium inoculation + poor hygiene) AA]. All taxa data were rarefied to 5,000 counts, as previously described (12). A first-glance analysis of the sampling distribution using mean and median was conducted for taxa with a proportion greater than 1% (major taxa) at both the family and genus levels (hierarchical levels) by DPC, regardless of treatments, to assess data skewness. For alpha-diversity (Simpson’s D and Shannon’s indexes), either a one-way analysis of variance (ANOVA) (more than two treatments—Barn-SC, dietary condition, and interaction) or a Welch T-test (two treatments only—Barn-SC effect, regardless of diet provided) was used to assess significance (p < 0.05). Both Shannon and Simpson’s D indexes of alpha diversity were calculated using rarefied counts, with the diversity function in R from the vegan library (version 2.6-6.1).

If no significant effects arose from the interaction between Barn-SC and Diet, only the significant main effect would be further examined. This approach was taken for alpha- and beta-diversity. The rest of the analysis was conducted based on the main significant effect on community diversity, composition, and structure, as previously established by Gomes-Neto et al. (32).

Beta-diversity was calculated using the vegdist function from the vegan library (version 2.6-6.1), employing a Bray-Curtis distance matrix (non-binary data) within a framework previously established (32). For the principal coordinates analysis (PCoA), a classical multidimensional scaling model was first used to reduce the distance matrix to two dimensions (2 principal coordinates—PCs), using the cmdscale function (k = 2) in R from the stats library. Specifically, for DPC 21, principal component analysis (PCA) was employed to reduce the data to three dimensions, thereby enhancing the clustering assessment. Ultimately, PC1 and PC2 were used to present a biplot that captures the effects of either sanitary conditions or all treatment effects on beta diversity each day. A PERMANOVA was used to calculate the effect of treatment groups (p-value and R-squared) on beta-diversity (Bray-Curtis distances), using the ‘adonis2’ function in R from the “vegan” library (version 2.6-6.1). A volatility analysis (microbiome compositional assessment using each PC independently as an outcome variable—PC1, PC2, or PC3 from the PCoA) was conducted by assessing significant differences between Barn-SC groups across each DPC or across all treatments by DPC as well (a Welch T-test was used to compare Barn-SC by DPC); an ANOVA model was used for all treatments by DPC, with a Tukey Honest Significant Differences test (Tukey HSD) applied for pairwise comparisons, provided a significant effect of treatments (p < 0.05) was found in the ANOVA model. An analysis of similarity (ANOSIM) was used to calculate the effect of treatment groups (Barn-SC or all treatments) on beta-diversity (Bray-Curtis distances), using the anosim (method = “bray”—Bray-Curtis’s distance matrix) functions in R, respectively, from the vegan library (version 2.6-6.1). Finally, a PCA (without a distance matrix) analysis was also conducted using log2-transformed proportions across bacterial taxa, with rarefied counts as input data, to assess clustering between Barn-SC groups and across all treatments by DPC.

Keystone taxa were identified based on significant (p < 0.05) differences in community structure at DPC 21 across Barn-SC groups [GOOD vs. POOR (S. Typhimurium inoculation + poor hygiene)]; by filtering all bacterial genera with (1) mean proportion > 1%; (2) using a LEfSe analysis; (3) using a co-occurrence network analysis to identify central nodes; and (4) using a random forest classifier. Mean and median proportions were compared for skewness, and mean proportions were used to examine the distribution of keystone taxa using either proportion or log2-transformed proportions (adding a pseudo-value of 1 for all data to ensure proper transformation with zero counts). LEfSe analysis for identification of differentiating taxa was done with the Galaxy platform with a Wilcoxon (p < 0.05), and upon removal of taxon or features with zero counts across fecal samples, only contrasting sanitary conditions [GOOD vs. POOR (S. Typhimurium inoculation + poor hygiene) Barn-SC] were considered (53). The random forest classifier was also applied for DPC 0 and 10. For this purpose, the caTools and randomForest libraries were utilized in R. All random forest models were constructed using rarefied counts (without further transformation), 100 trees, a classification model for a binary outcome, and a 70/30 random split of the training and testing data, respectively. The Gini importance was used to identify the most important predictors of the random forest model, which differentiates between POOR (S. Typhimurium inoculation + poor hygiene) and GOOD Barn-SC across each DPC.

Statistical differences across keystone taxa were calculated using both proportion and log2-transformed proportion, all derived from rarefied counts. When comparing GOOD vs. POOR (S. Typhimurium inoculation + poor hygiene), Barn-SC, a Welch T-test (p < 0.05), was used by DPC. If comparing across all four treatment groups, then an ANOVA model was used, followed by a pairwise T-test, provided the ANOVA model yielded significant results (p < 0.05). Box-and-whiskers and a clustering heatmap were used to visualize the distribution of taxa across Barn-SC groups or all treatments by DPC. For the clustering heatmap visualizations, a nested approach was applied, considering the 2 × 2 factorial design (diet groups within each barn-SC). All cluster heatmap analysis was performed using log2-transformed proportions (a pseudo-value of 1 was input when zero was present), followed by data scaling (normalization) and visualization using a heatmap.2 functions in R. The correlation structure between keystone taxa was also calculated using log2-transformed proportions, with input values of 1 for zeros. This analysis was conducted using the cor function to calculate the Pearson correlation matrix, followed by the cor.mtest to identify significant correlations (p < 0.05). The results were then visualized using a corrplot in R, clustered using a hierarchical clustering approach (hclust). Fold-change (ratio) differences between keystone taxa across Barn-SC or all treatments were made from the endpoint (DPC 21) and baseline (DPC 0). For that, a pseudo-value of 0.01 was initially input for zeros in the mean proportional data (to avoid infinity values in the denominator), followed by a log₂ transformation of the ratio with a pseudo-value imputation of 1 for zeros. Pearson’s correlation between keystone taxa relative abundance (log₂ transformed) ratio (DPC 21/DPC 0) and the average daily gain (host-associated trait) for the corresponding period was also calculated. All tests for Pearson’s correlation coefficient were calculated using the cor_test function in R (p < 0.05). For the average daily gain between DPC 0 and 21 (host-associated trait), an ANOVA model was used to assess the effect of diet, Barn-SC, and interaction between the two (p < 0.05). All ANOVA models were done using the aov function in R, while two-group comparisons were made using a T-test by applying a two-sample t-test function in R from the stats library. A sample was removed from the analysis when a missing value (NA) was consistently present across all taxa. Illustrative figures were created using BioRender.

In our companion study, several aspects of growth, including performance and body composition, among others, were captured in group-housed growing pigs under contrasting sanitary conditions [GOOD SC vs. POOR (S. Typhimurium inoculation + poor hygiene) SC] and fed two diets, control (CN) or supplemented with AA (9). Pigs housed under POOR SC (S. Typhimurium inoculation + poor hygiene) had a significantly lower average daily gain compared to those in GOOD SC (S. Typhimurium inoculation + good hygiene) (p < 0.05), underscoring a potential difference in fecal microbial communities. Figure 1 presents the experimental design, sampling, and key microbiome outcomes. Supplementary Table 1A,B describe the fecal sampling scheme throughout the study across all treatments. More specifically, this study assessed microbial diversity, community structure, and enrichment of differentiable taxa across treatments over time using 16S rRNA sequencing from fecal samples. The aim was to identify differences between sanitary conditions and the potential impact of AA surplus in the diets of growing pigs. A first-glance assessment of the keystone fecal taxa (> 1%) both at the family and genus level, over time, regardless of treatment conditions, was done to determine data skewness (mean vs. median) prior to subsequent statistical data mining with all samples rarefied at 5,000 counts (Supplementary Figures 1, 2).

Differences in alpha diversity across treatments were analyzed using both the Simpson’s D (evenness) and Shannon (richness) indexes of diversity. Figures 2A,B demonstrates the results of the ANOVA and T-test-based differences across diet groups by DPC for each SC condition for both Simpson and Shannon indexes, respectively. Significant differences were found for both Simpson (p = 0.00835) and Shannon (p = 0.0104) indexes of diversity only in barn SC (Figures 2A,B), with no significant differences between diet groups across SC by DPC (p < 0.05). Alpha diversity was significantly lower in the POOR SC group (S. Typhimurium inoculation + poor hygiene) at both DPC 21 (p = 0.048) using Simpson’s D index and at DPC 10 (p = 0.042) using the Shannon index, in comparison to the GOOD SC group (Figures 2C,D).

Initial assessment of beta diversity across fecal communities showed no clustering by diet across SC conditions by DPC (Figure 3A). A PERMANOVA analysis revealed only a significant effect for barn SC (p < 0.05) and not for diet or the interaction between the two, with the R-squared going from ~ 5% at DPC 0 to ~17% at DPC 21, demonstrating a significant biological effect of SC on the community clustering (Figure 3B). Based on this, Figure 3C highlights a clear separation of the fecal communities by SC, starting at DPC 10, with more discrete clustering at DPC 21. A 3D representation of PCoA analysis at DPC 21 confirmed the separation between GOOD vs. POOR (S. Typhimurium inoculation + poor hygiene) SC fecal communities (Figure 3D). ANOSIM results mirrored those of the PERMANOVA, underscoring a dissimilarity between GOOD vs. POOR (S. Typhimurium inoculation + poor hygiene) SC fecal communities, starting at DPC 10 (R = 0.196, p = 0.01) and reaching a higher degree at DPC 21 (R = 0.352, p = 0.001) (Supplementary Figures 3–5). When stratified by all treatment combinations, ANOSIM results corroborated the differences between SC groups, with fecal community dissimilarities most pronounced at DPC 21, irrespective of diet (R = 0.219, p = 0.001) (Supplementary Figures 6–8). A follow-up analysis using the individual PCs (PC1, PC2, and PC3) as the response variable recapitulated the beta-diversity results, demonstrating significant differences across SC groups at DPCs 10 and 21, not driven by diet (p < 0.05) (Supplementary Figures 9A–D, 10). Notably, the POOR (S. Typhimurium inoculation + poor hygiene) SC community exhibited higher dispersal, as indicated by PC1- and PC3-based analysis (Supplementary Figures 9A, 10A). An independent PCA analysis (not using a distance matrix) corroborated all results by showing clustering driven by barn SC at DPC 10 and 21 (Supplementary Figures 11–13) and not by diet (Supplementary Figures 14–16).

Given the beta-diversity results clearly demonstrating a separation between fecal communities between SC conditions at DPC 21 (Figures 3B–D) and the absence of a significant dietary effect (Figure 3A), a combination of multiple analytical methods was used to identify all taxa differentiating GOOD vs. POOR (S. Typhimurium inoculation + poor hygiene) SC groups at DPC 21 first, including network analysis (Figures 4A,B), LEfSe analysis (Figure 4C and Supplementary Figures 17–23), random forest (Supplementary Figures 24–26), and t-test (Figure 5 and Supplementary Figure 29). Additionally, an exploratory analysis was conducted using the overall mean abundance and a cut-off of 1% to select the major taxa across the dataset (Supplementary Figures 1, 2), which allowed for contextual interpretation of the data for the most significantly different taxa between groups (p < 0.05).

Based on that, the most important taxa differentiating the two groups are as follows: (1) on average lower in relative abundance in the POOR (S. Typhimurium inoculation + poor hygiene) group at DPC 21: Acidaminococcus, and (2) on average, higher in relative abundance in the POOR (S. Typhimurium inoculation + poor hygiene) group at DPC 21: Clostridium sensu stricto 1, Dorea, Intestinibacter, Lactobacillus, Romboutsia, Ruminococcus torques, Subdoligranulum, Terrisporobacter, and Turicibacter. Network analysis revealed two unique sub-clustering structures (with positive interactions) formed in the POOR (S. Typhimurium inoculation + poor hygiene) SC group (Figure 4B): Sub-cluster 1 (Romboutsia, Turicibacter, Clostridium sensu stricto 1, Terrisporobacter, and Intestinibacter) and Sub-cluster 2 (Dorea, Subdoligranulum, Ruminococcus torques, Blautia, Holdemanella, and Solobacterium). LEfSe analysis also indicated enrichment in Roseburia for POOR (S. Typhimurium inoculation + poor hygiene) SC at DPC 21 (Figure 4C). Average community abundances (proportion or log-transformed) were also examined (Supplementary Figures 27, 28). Following the identification of the most differentiating taxa, a temporal analysis was conducted using relative frequencies (log-transformed or not) for the barn SC effect or across all treatments (Figure 5 and Supplementary Figures 29–31).

The initial clustering separation between GOOD vs. POOR (S. Typhimurium inoculation + poor hygiene) SC groups for the beta-diversity analysis at DPC 10 (Figure 3C) could be explained in part by the significantly higher average abundance for Clostridium sensu stricto 1, Intestinibacter, Lactobacillus, and Terrisporobacter in the POOR (S. Typhimurium inoculation + poor hygiene) SC group and significantly lower abundance for Prevotella, some Prevotellaceae taxa, and Roseburia in the same group (Figure 5 and Supplementary Figure 29), despite AA surplus (Supplementary Figures 30, 31). Additionally, a fold-change analysis demonstrated that Clostridium sensu stricto 1, Intestinibacter, Romboutsia, Terrisporobacter, and Turicibacter were among the top enriched taxa in the POOR (S. Typhimurium inoculation + poor hygiene) SC group at DPC 21 in relationship with DPC 0 (Supplementary Figure 32), despite AA surplus in the diet (Supplementary Figure 33). The host phenotype of interest, average daily gain, was significantly lower in the POOR (S. Typhimurium inoculation + poor hygiene) SC group and remained unchanged by AA surplus (Supplementary Figure 34). In the GOOD SC group, Prevotellaceae UCG 001 and Sphaerochaeta were significantly positively correlated with average daily gain for the duration of the study (Supplementary Table 2A). On the other hand, Oscillibacter and Turicibacter were significantly negatively correlated with average daily gain for the study duration in the POOR (S. Typhimurium inoculation + poor hygiene) SC group, while no significant association was found for the GOOD SC for these two taxa (Figure 6 and Supplementary Table 2B). In aggregate, Prevotella was more likely to be positively correlated with the host phenotype for the GOOD SC group (Supplementary Table 1A,B). When stratifying by barn SC, it was found that for GOOD CN, Moryella and Sphaerochaeta were significantly positively correlated with the outcome (Supplementary Table 3A), while Lactobacillus was significantly negatively correlated with it for GOOD AA (Supplementary Table 3B). Finally, for POOR (S. Typhimurium inoculation + poor hygiene), AA Prevotellaceae UCG 003 was found to be significantly positively correlated with average daily gain, while no significant associations were found across taxa for POOR (S. Typhimurium inoculation + poor hygiene) CN, even though both Romboutsia and Turicibacter trended in demonstrating a strong negative correlation in comparison to POOR AA (Supplementary Table 4A,B).

A temporal clustering and correlation analysis using only the keystone differential fecal taxa confirmed a separation between communities based on barn SC status, most prominently at DPC 21 (Figures 7A–C), with no predictive signatures based on dietary modification (AA surplus) (Supplementary Figures 35–46). At DPC 21 (Figure 7C), there were two groups of taxa more likely to be enriched in the POOR (S. Typhimurium inoculation + poor hygiene) than the GOOD SC group: (1) a major sub-cluster comprised of Clostridium sensu stricto 1, Intestinibacter, Romboutsia, Terrisporobacter, and Turicibacter as previously captured by the temporal analysis of relative abundances (Figure 5 and Supplementary Figure 29); and (2) a secondary sub-cluster comprised of Blautia, Butyricoccus, Dorea, Faecalibacterium, Ruminococcus torques, and Subdoligranulum, with bimodal distribution-like patterns across animals in the POOR SC group. In accordance, these two sub-clusters demonstrated a correlated pattern within themselves, starting at DPC 10 and more clearly at DPC 21 for the POOR (S. Typhimurium inoculation + poor hygiene) SC group (Figures 8A–F). The correlation structure among these keystone fecal taxa did not appear to be predictably affected by dietary changes (Supplementary Figures 41–46).