This study was comprised of two separate blinded, randomized, and controlled trials of steers conducted at large commercial beef feedlots, one in Colorado and another in Texas. The feedlots and rearing conditions were typical of large North American feedlots with >5,000 animal capacity. Complete details about the sample population, cattle handling, base diets, and experimental design of both trials have been described previously (Weissend, 2018; Huebner et al., 2019).

Briefly, in the Colorado-based trial, steers (n = 4,689 cattle housed in 28 pens) were enrolled in a randomized, block-controlled study at a commercial feedlot in Northern Colorado. Upon arrival cattle were administered injectable and oral anti-parasitic treatment (Noromectin, Norbook Laboratories, Newry, United Kingdom; Synanthic, Boehringer Ingelheim, Ingelheim am Rhein, Germany; Standguard, Elanco Animal Health, Greenfield, IN, United States) and two injectable bacterin-toxoid vaccines protecting again BRD (Titanium 5 + PH-M, Elanco Animal Health; Vision 7, Intervet/Merck Animal Health, Omaha, NE, United States) and randomly assigned into two adjacent pens based on equal weight distribution. Pens were open-air, dirt-floor with a central feed alley. All cattle received base diets consisting predominantly of steam flaked corn, and cattle in adjacent pens were randomly assigned to receive either (1) a no treatment diet containing no additives for prevention of LAs (negative control) or (2) a diet supplemented with Saccharomyces cerevisiae fermentation product (SCFP; 18 g·d⁻¹; Diamond V, Cedar Rapids, IA, United States). A treatment and control pen designated one block, and each block contained cattle with similar source locations, bread, headcounts, and weight distributions.

In the Texas-based trial, steers (n = 5,481 housed in 40 pens) were enrolled in a randomized, block-controlled study at a commercial feedlot located in West Texas. Following arrival, cattle received a standard processing regimen consisting of drench dewormer, IBR and BVD vaccines, lot tags, and ear implants (Revalor-XS, Merck Animal Health, Madison, NJ, United States). Cattle were then randomly assigned to pens within randomization blocks; each block consisted of four pens with an equal number of cattle, and each pen was randomly assigned one of the four treatments groups. All cattle were provided with corn-based rations ad libitum with one of the following four modifications: (1) a no treatment diet containing no additives for prevention of LAs (negative control), (2) a diet supplemented with tylosin (90 mg·d⁻¹; Elanco Animal Health), (3) a diet supplemented with the essential oil limonene (1 g·d⁻¹; Victus-LIV; manufactured by Prinova Flavors for DSM Nutritional Products), or (4) a diet supplemented with SCFP (18 g·d⁻¹; Diamond V). Additionally, cattle in all four treatment groups received monensin (Rumensin, Elanco Animal Health) in their intermediate and finishing rations and ractopamine (Optaflexx, Elanco Animal Health) during the 14 days prior to harvest.

Methods involving animal care and use in both trials were approved by and carried out in accordance with the Colorado State University Research Integrity and Compliance Review Office (Protocol number 102616).

At the end of both feeding trials, all livers from cattle were evaluated at the time of harvest for the presence and severity of LAs. The prevalences of A and A+ abscesses within each pen were recorded, and pens were subsequently categorized according to tertiles of the prevalence distribution (i.e., highest-, middle-, and lowest-third of the prevalence values). Up to 5 abscessed livers per pen (n ≤ 5 per pen) were selected for sampling. The LAs and surrounding tissue were excised, and this tissue block was placed into sterile sample bags, placed on ice, and transported to the Center for Meat Safety and Quality Food Safety Microbiology Lab at Colorado State University where they were processed within 24 h.

To facilitate aseptic recovery of purulent material, individual tissue blocks containing abscesses (n = 259) were immersed in 100% ethanol and then flamed to sterilize the outer surfaces. Abscesses were then aseptically dissected to access the purulent material, and 0.1–0.4 g was harvested. Genomic DNA was isolated using the QIAamp PowerFecal DNA Isolation kit (Qiagen, Hilden, Germany), according to the manufacturer’s instructions. Following isolation, DNA was quantified (ng μl⁻¹) using a Qubit flex fluorometer (Thermo Fisher Scientific, Waltham, MA, United States).

16S rRNA gene amplicon library preparation and sequencing were carried out by Novogene Corporation Inc. (Chula Vista, CA, United States). The V4 region of the 16S rRNA gene was amplified using the 515f (5′-GTG CCA GCM GCC GCG GTA A-3′) and 806r (5′-GCA CTA CHV GGG TWT CTA AT-3′) primer pair and Novogene’s proprietary amplification conditions. Following successful PCR, amplicon libraries were prepared and pooled using Novogene’s proprietary process and sequenced on an Illumina HiSeq 2,500 instrument (Illumina Inc., San Diego, CA, United States) using 2 × 250 bp paired-end chemistry. As negative controls, equal volumes of nuclease-free sterile water were included with each batch of samples processed. These controls were included in procedures for PCR amplification and in preparation of sequencing libraries. Sequencing of these negative control samples did not yield product and therefore were not included in further downstream analysis. The number of reads per liver abscess sample ranged from 14, 966 to 973, 255, with an average of 286, 429 reads per sample and there were no differences in read depth between any of the treatment groups (pairwise Wilcoxon rank-sum with Benjamini–Hochberg correction, p > 0.05).

Demultiplexed paired-end reads were imported into QIIME2 version 2020.11 (Bolyen et al., 2019) and DADA2 (Callahan et al., 2016) was used to filter reads for quality, remove chimeric sequences, merge overlapping paired-end reads, and generate amplicon sequence variants (ASVs). Forward and reverse reads were trimmed at 25 and 26 bp and truncated at 244 and 249 bp, respectively. Taxonomy was assigned using a Naïve Bayes classifier trained on the Greengenes version 13_8 99% OTUs database (DeSantis et al., 2006), where sequences had been trimmed to include only the base pairs from the V4 region bound by the 515f/806r primer pair. Reads that mapped to chloroplast and mitochondrial sequences were filtered from the representative sequences and ASV table using the “filter_taxa” function, and a midpoint-rooted phylogenetic tree was then generated using the “q2-phylogeny” pipeline with default settings, which was used to calculate phylogeny-based diversity metrics. The proportion of ASVs classified at each taxonomic rank can be found in Supplementary Table S1.

Data and metadata were then imported into phyloseq (McMurdie and Holmes, 2013). Richness (Observed ASVs) and Faith’s Phylogenetic Diversity (FPD) were calculated for all samples with phyloseq and the btools package. ASV counts for each sample were then normalized using cumulative sum scaling (Paulson et al., 2013) and beta-diversity was analyzed using generalized UniFrac distances (Lozupone et al., 2011; Chen et al., 2012). From these distances, non-metric multidimensional scaling (NMDS) was performed and plotted, and a permutational multivariate analysis of variance (PERMANOVA) was used to test for significant differences in community structure using the vegan (Oksanen et al., 2019) and pairwiseAdonis (Arbizu, 2017) packages. To ensure significant differences were not the result of unequal dispersions of variability, permutational analyses of dispersion (PERMDISP) were conducted for all significant PERMANOVA outcomes using the vegan package. Hierarchal clustering was performed using Ward’s agglomeration clustering method (Murtagh and Legendre, 2014) on generalized UniFrac distances and the “hclust” function. Further, the relative abundances of ASVs within each sample were calculated and plotted using phyloseq.

To further explore the dichotomy between liver abscess microbial communities dominated by Fusobacteria or Bacteroidetes, samples were classified as belonging to 1 of 5 groups: high Fusobacteria (Fusobacteria relative abundance [RA] > 75%, n = 180), high Bacteroidetes (Bacteroidetes RA > 14%, n = 61), high Proteobacteria (Proteobacteria RA > 15%, n = 9), high Firmicutes (Firmicutes RA > 11%, n = 5), or other (n = 4). Group cutoffs were set based on relative abundance values that resulted in the least number of samples cross-appointed to two groups (i.e., Fusobacteria >75% and Bacteroidetes >14%). Though largely mutually exclusive, samples (n = 3) cross-appointed between high Fusobacteria and high Bacteroidetes were classified as high Fusobacteria and samples (n = 3) cross-appointed between high Bacteroidetes and high Proteobacteria were classified as high Bacteroidetes. All other samples (n = 253) were mutually exclusive.

To identify the higher-level taxa discriminating Fusobacteria-dominated abscesses (n = 180) from Bacteroidetes (n = 61) dominated abscesses, linear discriminant analysis effect size (LEfSe) was performed using the online LEfSe tool on the Galaxy server¹ under default settings, except for using a stricter alpha value and higher logarithmic LDA score threshold (0.01 and 3.0, respectively). Genera making up greater than 0.05% of the overall community across each community type (i.e., high Bacteroidetes or high Fusobacteria) and present in at least 20% off all LA communities were considered.

To identify potential anatomic sources for bacterial communities that were translocated from the GIT to seed the liver, the relative abundances of genera discriminant of both Fusobacteria and Bacteroidetes-dominated abscesses were visualized and compared to bacterial communities of nine GIT locations from a total 56 cattle, using 16S rRNA data from previous research that was publicly available at the NCBI’s Sequence Read Archive (SRA) through BioProjects PRJEB28377 and PRJNA399771. PRJEB28377 represents the largest collection of 16S rRNA sequence data covering multiple GIT locations publicly available at the time of writing and represents nine different segments (rumen, omasum, abomasum, duodenum, jejunum, ileum, caecum, colon, and rectum) of the GIT in healthy cattle and cattle with hemorrhagic bowel syndrome. For the purposes of this study, only samples collected from healthy cattle were included (n = 373). Similarly, PRJNA39971 sampled four different segments (rumen, jejunum, caecum, and colon) of dairy calves. All samples collected from this study were included (n = 49) here. Demultiplexed sequence reads from both studies (n = 422) were downloaded from the SRA and processed as described above in the Bioinformatics subsection.

Unless otherwise stated, R version 3.6.3 (R Core Team, 2017) was used for statistical analysis of data. The impact of dietary inclusion of tylosin and antibiotic alternatives on microbial community composition were analyzed independently for the Colorado and Texas trials, because of potential confounding related to location and cattle source, and because monensin and ractopamine were only included in the diets of cattle enrolled in the Texas trial. Pairwise Wilcoxon rank-sum tests were performed with a Benjamini–Hochberg correction for multiple comparisons. Differences in beta-diversity were tested using pairwise PERMANOVA with a Benjamini–Hochberg correction for multiple comparisons and 9,999 permutations. Additionally, pairwise PERMDISPs were carried out for all significant PERMANOVA outcomes using 9,999 permutations to test for differences in the variability of dispersions. Correlation between the relative abundance of Fusobacteria and Bacteroidetes was tested using Spearman’s rank correlation coefficient.

All sequence reads were made available through BioProjects PRJNA472428 (Colorado trial) and PRJNA804092 (Texas trial) at the NCBI’s Sequence Read Archive.

Previous analyses using the same trials as this study demonstrated that diets supplemented with tylosin decreased the prevalence of LAs (Weissend, 2018), but supplementation with SCFP and EO had no impact on LA prevalence (Huebner et al., 2019). In the Colorado trial, there were no significant differences in the richness or diversity within liver abscess microbial communities from cattle given the control diet (n = 59) or the diet supplemented with SCFP (n = 74; Figure 1A; pairwise Wilcoxon rank-sum with Benjamini–Hochberg correction; n = 59–74, p > 0.05). Similarly, there were no significant differences in the richness or diversity of liver abscess communities between cattle given the control diet (n = 36), diet supplemented with SCFP (n = 35), diet supplemented with essential oil (EO; n = 32), or diet supplemented with tylosin (n = 23) during the Texas trial (Figure 1B; pairwise Wilcoxon rank-sum with Benjamini–Hochberg correction; n = 23–36, p > 0.05).

Based on generalized UniFrac distances, there were no significant differences in microbial community composition between cattle fed the control diet versus those receiving SCFP-supplemented diet in the Colorado trial and NMDS illustrated no clustering of abscess microbial communities collected from cattle given each diet (Figure 2A; PERMANOVA, n = 59–74, p > 0.05). There were also no significant differences in abscess-associated microbial community structure between cattle fed the control diet and diet supplemented with SCFP, EO, or tylosin in the Texas trial and NMDS showed no clustering of communities by treatment (Figure 3A; PERMANOVA, n = 23–36, p > 0.05).

Hierarchal clustering further revealed that treatment (diet supplemented with SCFP) within the Colorado trial had little effect on abscess microbial community structure, as communities from cattle given different diets were intermixed across all clades (Figure 2B). Visualization of microbial community structure at the taxonomic rank of phyla demonstrated that members of Fusobacteria, Bacteroidetes, Proteobacteria, and Firmicutes comprised most of the microbial community, representing over 90% of all microbial taxa in 128 of 133 samples (~96%; Figure 2B). However, the ranges of relative abundances for dominant phyla were extremely large: Fusobacteria, 14.8%–99.3%; Bacteroidetes, 0.02%–82.0%; Proteobacteria, 0.2%–30.0%; and Firmicutes (0.08%–31.3%). Some communities were almost exclusively comprised of Fusobacteria while others are largely composed of Bacteroidetes. Interestingly, clades appeared to form based largely on the abundance of members Fusobacteria and Bacteroidetes, with communities dominated by the latter forming their own distinct clade (Figure 2B).

Within the Texas trial, hierarchal clustering demonstrated that treatment (diet supplemented with SCFP, EO, or tylosin) had little effect on the microbial community structure of LAs, as abscess communities from cattle receiving the different diets were intermixed across all clades (Figure 3B). Visualization of the microbial community composition at the rank of phyla illustrated that members of the phyla Fusobacteria, Bacteroidetes, Proteobacteria, and Firmicutes comprised almost the entirety of the microbial community, representing over 90% of all microbial taxa in 125 of 126 samples (~99%; Figure 3B). Once again, these four phyla exhibited extremely variable ranges of abundance among the individual abscess communities (Fusobacteria, 2.0%–99.8%; Bacteroidetes, 0.03%–88.0%; Proteobacteria, 0.02%–85.1%; and Firmicutes, 0.01%–14.4%). As was the case in the Colorado trial, the formation of clades appeared to be largely influenced by the abundance of Fusobacteria and Bacteroidetes, with Bacteroidetes-dominated communities forming their own distinct clade (Figure 3B).

Of the 259 LA microbial communities, the vast majority (241, 93.1%) was dominated by either Fusobacteria (180, 69.4%) or Bacteroidetes (61, 23.6%). The remaining 18 LA communities were classified as high Proteobacteria (9, 3.5%), Firmicutes (5, 1.9%), or other (4, 1.5%). The proportions of LAs with communities considered high Fusobacteria or high Bacteroidetes were similar between the two trials, while communities not dominated by Fusobacteria or Bacteroidetes were far more abundant in the Colorado trial. Importantly, while there was a slightly higher proportion of Fusobacteria-dominated abscess communities in pens with higher liver abscess prevalence, the difference was not statistically significant (Supplementary Figure S1; pairwise Wilcoxon rank-sum with Benjamini–Hochberg correction; n = 21–22, p > 0.05). Conversely, there was a slightly higher proportion of Bacteroidetes-dominated abscess communities in pens with lower liver abscess prevalence, but the difference was not statistically significant (Supplementary Figure S1; pairwise Wilcoxon rank-sum with Benjamini–Hochberg correction; n = 21–22, p > 0.05). There were no significant differences in the proportions of Fusobacteria or Bacteroidetes-dominated liver abscess communities between any of the treatment groups, though there was a slightly lower proportion of Bacteroidetes-dominated communities in cattle given tylosin (Supplementary Figure S2; pairwise Wilcoxon rank-sum with Benjamini–Hochberg correction; n = 9–24, p > 0.05).

Hierarchal clustering of LAs from both trials showed that the predominance of Fusobacteria or Bacteroidetes was the most important factor in differentiating between abscess microbial communities, with virtually every community dominated by Bacteroidetes forming its own clade that was most distinct from all other clades (Figure 4A). Communities not dominated by Bacteroidetes of Fusobacteria (high Proteobacteria, high Firmicutes, or other) largely clustered with themselves inside of a larger clade comprised of Fusobacteria-dominated communities (Figure 4A). Visualization with NMDS clearly illustrates that communities with high Fusobacteria and high Bacteroidetes both form distinct clusters with communities of their own type, while those with high Proteobacteria, Firmicutes, or other are much more highly variable and do not form distinct clusters (Figure 4B). Both Fusobacteria and Bacteroidetes-dominated communities were significantly different from all other community types, even when only partially weighting taxa abundance (Supplementary Table S2; PERMANOVA, n = 61–180, p < 0.05). However, the other community types (high Proteobacteria, Firmicutes, or other) were not significantly different from one another (Supplementary Table S2; PERMANOVA, n = 4–9, p > 0.05). Communities rarely contained high abundances of both Fusobacteria and Bacteroidetes, and their relative abundances within liver abscess microbial communities were strongly negatively correlated (Figure 4C; Spearman’s rho = −0.857, n = 259, p < 0.0001).

Fusobacteria, and the genus Fusobacterium specifically, was the most abundant taxa in LAs considered high Proteobacteria or high Firmicutes but was in lower proportion than LAs classified as high Fusobacteria using criteria described for this analysis (Supplementary Table S3). The relative abundance of other community members within these community types was highly variable. In high Proteobacteria LAs, the genera Halomonas (2.9% ± 1.53 SEM), Agrobacterium (6.6% ± 6.56 SEM), and unclassified Enterobacteriaceae (2.1% ± 1.53 SEM) were the most abundant members of Proteobacteria (Supplementary Table S3). However, the high relative abundance of Agrobacterium was the result of it dominating (59%) a single LA and it comprised less than 0.2% of communities in the other high Proteobacteria LAs. The increased abundance of Firmicutes in LAs considered high Firmicutes was largely the result of unclassified Veillonellaceae (5.5% ± 1.33 SEM), unclassified Ruminococcaceae (2.8% ± 1.18 SEM), and unclassified Clostridiales (1.8% ± 0.30 SEM), which were all among the top five most abundant taxa at the level of genus (Supplementary Table S3).

In addition to Fusobacterium, taxa discriminant of Fusobacteria-dominated liver abscess microbial communities included the genera Acinetobacter, Lactobacillus, Pseudomonas, and Psychrobacter (Figure 5; LEfSe, n = 61–180, p < 0.01). Of these genera, Fusobacterium was unsurprisingly the most abundant, comprising 92.2% ± 0.48 SEM of the overall community in Fusobacteria-dominated abscesses. The other four discriminant taxa comprised only 0.1%–0.3% of the overall community (Supplementary Table S3). Two genera of Bacteroidetes, Bacteroides and Porphyromonas, along with the genera Atopobium, Campylobacter, Filifactor, Helcococcus, Parvimonas, and Trueperella were discriminant of Bacteroidetes-dominated liver abscess microbial communities (Figure 5; LefSe, n = 61–180, p < 0.01). Not surprisingly, Bacteroides was the most abundant of the three, representing 51.3% ± 2.67 SEM of the overall community in Bacteroidetes-dominated LA communities. Porphorymonas (4.9% ± 1.37 SEM) and Campylobacter (2.6% ± 0.78 SEM) were moderately abundant, while the other six discriminant taxa comprise 0.1%–0.3% of the overall community (Supplementary Table S3). The genus Salmonella was not identified in any sample.

Genera discriminant of the community composition types assigned to LAs were similarly investigated in 422 samples collected from nine locations within the bovine GIT in two previous studies (rumen, omasum, abomasum, duodenum, jejunum, ileum, caecum, colon, and rectum). In general, discriminant taxa that were abundant in LAs (i.e., Fusobacterium, Bacteroides, and Porphorymonas) were considerably less abundant in the GI tract. In aggregate, taxa discriminant of Fusobacteria-dominated abscesses were significantly more abundant within the jejunum than all other GIT locations while taxa discriminant of Bacteroidetes-dominated abscesses were significantly more abundant within the cecum, colon, and rectum than all other locations (Figure 6; pairwise Wilcoxon rank-sum with Benjamini–Hochberg correction; n = 34–88, p < 0.05). Within the rumen, omasum, and abomasum, the relative abundance of both sets of discriminant taxa were incredibly low (<0.08%). However, within more proximal portions of the GIT (i.e., abomasum, duodenum, jejunum, and ileum), taxa discriminant of Fusobacteria-dominated abscess communities were more abundant than those discriminant of Bacteroidetes-dominated abscesses (Figure 6). Conversely, genera discriminant of Bacteroidetes-dominated LAs were more abundant than taxa discriminant of Fusobacteria-dominated LAs in the cecum, colon, and rectum (Figure 6).