The sources and transmission routes of microbial populations throughout a meat processing facility

The study addresses the global challenge of microbial food spoilage, which contributes substantially to food waste and poses significant public health risks. Animal-derived foods, including pork, are key vehicles for food-borne diseases, and microbial contamination during primary processing remains difficult to control due to multiple potential sources. Traditional ISO culture-based hygiene indicators (e.g., aerobic colony counts and Enterobacteriaceae) are useful but insufficient to characterize complex microbial communities and transmission routes in slaughterhouses. The authors hypothesize that a large portion of the microbial community found on pork meat after processing is not animal-associated but instead is transmitted during cutting via personnel, equipment, machines, or the slaughter environment. The purpose is to map microbial diversity and identify transmission routes throughout a pork-processing facility using full-length 16S rRNA gene sequencing and microbial source tracking, thereby enabling targeted hygiene interventions to reduce spoilage and improve food safety.

The authors situate their work within efforts to apply next-generation sequencing in food processing environments. Prior studies primarily relied on ISO culture-dependent methods for hygiene monitoring, which do not capture complex microbial community dynamics. Recent microbiome studies in meat processing and other food sectors have demonstrated the suitability of sequencing to map microbial ecosystems and spoilage-associated microbiota, including psychrophilic spoilers dominating pig musculature microbiomes. SourceTracker has been used across diverse environments (e.g., coastal waters, drinking water systems, public surfaces, NICUs) but rarely in food processing (with one brewery example). The polishing tunnel has previously been identified as a critical contamination step in pork slaughtering, while disinfection protocols and WGS have been used to track pathogens (e.g., L. monocytogenes) in facilities. However, WGS is costlier; full-length 16S rRNA can offer higher taxonomic resolution than short-read amplicons and may suffice for routine monitoring and source tracking to identify general transmission routes.

Facility and sampling: An Austrian slaughterhouse (capacity 200–250 pigs/h) was sampled along the processing line (stunning/sticking → scalding → dehairing → pre-singeing → singeing ~6 s → polishing (whips, nozzles, water) → declawing → clean area: evisceration (gloves, knives, aprons) → splitting (saw) → classification (gloves, railing) → shock shower → cooling chamber (16 h at 7 °C; wall) → truck (gloves, wall)). Twelve pigs from three farms (four each) were ear-tagged and followed; 84 carcass surface swabs (100 cm² at the back) were collected at multiple positions (skin at “Sticking”; meat thereafter). Seventy-five environmental swabs were taken from equipment, staff, and infrastructure, totaling 159 samples collected in one day. Sterile polyurethane sponge swabs were used; each sample swabbed 10 s horizontally and 10 s vertically; sponges were chilled, transported on ice to the lab (~2 h), and split for culture and molecular analyses. Microbiological enumeration: Aerobic mesophilic counts (AMC; ISO 4833-2:2013), Enterobacteriaceae (EB; ISO 21528-2:2017), and Pseudomonadaceae (PS) were enumerated via serial dilution in BPW and plating on PCA (30 °C, up to 72 h), VRBG (37 °C, 24–48 h), and GSP (25 °C, 24–48 h). Counts expressed as CFU/cm² (10–300 CFU included). Presumptive EB and PS isolates (n=5 each) were confirmed by oxidase and API 20E profiling (species for EB; genus for PS). DNA extraction and qPCR: Samples were concentrated (3220× rcf, 20 min), pellets resuspended in PBS; DNA extracted (QIAamp DNA Stool Mini Kit) with modified elution (2×25 µl DEPC water). DNA quantitated by Qubit. Total bacterial cell equivalents (BCE) were determined by 16S rRNA gene qPCR (duplicates, with negatives) and extrapolated using an average of four 16S rRNA gene copies per genome (rrnDB). Normality was assessed (Shapiro-Wilk, QQ plots, histograms); non-normal groups tested via Wilcoxon; FDR controlled by Benjamini-Hochberg. Spearman correlations between BCE and AMC were computed per position. Amplicon sequencing: Full-length 16S rRNA gene libraries (133 samples, including 3 negatives) amplified with 27F/1492R; barcoded with PacBio universal primers; sequenced on PacBio Sequel (2.1 chemistry) across three SMRT cells (~50 GB raw data and ~642k sequences per cell). Additionally, the V3–V4 region of 52 samples was sequenced on Illumina MiSeq (2×300 bp) using 341f/785r. Bioinformatics: PacBio circular consensus reads (ccs) generated (min predicted accuracy 0.99; ≥3 passes) and demultiplexed. Both PacBio and MiSeq datasets were processed with DADA2 to infer ASVs at single-nucleotide resolution, taxonomically assigned using a DADA2-formatted Genome Taxonomy Database (GTDB, release 03-RS86). Post-filtering removed samples with <200 reads, ASVs with <5 reads, and contaminants via decontam (prevalence method, threshold 0.5). Community analyses (alpha diversity indices on rarefied datasets; beta diversity) were conducted using phyloseq and tsnemicrobiota, visualized with ggplot2. Normality and appropriate paired tests (Wilcoxon or t-test) were applied with Benjamini-Hochberg adjustment. Source tracking and sequencing depth assessment: SourceTracker (v1.0.0; default parameters) estimated contributions from sources (environmental samples and animal skin) to sinks (meat samples). To evaluate sequencing depth effects, Illumina datasets (min 7,712; mean 19,119; max 30,340 reads/sample) were rarefied to 7,712; 5,000; 1,000; 500; and 200 reads/sample with replicate rarefactions (3× for larger sizes; 10× for 200) and compared to the full dataset. Goodness-of-fit was assessed via squared differences from the original dataset and Spearman correlations; distributional assumptions checked (Shapiro-Wilk; Levene). Kruskal-Wallis with Dunn’s post hoc (BH-adjusted) tested differences in fit, hit ratio (percentage of correctly assigned sources, including unknown), and unknown classification rates. Beta diversity reliability at different depths was also checked (BAT package).

• Bacterial load dynamics: Skin at entry (“Sticking”) showed highest contamination; singeing significantly reduced microbial loads. After polishing, AMC and Pseudomonadaceae counts significantly increased; EB remained undetected until evisceration/classification and in environmental samples. AMC ranged from 4.14×10^3 CFU/cm² (after singeing) to 5.21×10^6 CFU/cm² (sticking). Polishing tunnel whips had the highest environmental counts (2.19×10^7 CFU/cm²). BCE ranged from 1.64×10^2 to 1.20×10^5 BCE/cm². BCE and AMC correlated on skin samples (R=0.76, p<0.01) but not on final meat (R=0.38, p=0.32) or environmental samples.
• Community diversity and composition: Alpha diversity decreased from start to end, with a transient increase after polishing. Beta diversity revealed major shifts at singeing and dispersion after classification/truck. 91% of full-length sequences classified to genus and 74% to species. Early positions (anal swab, sticking) featured Helicobacter and Curvibacter; after singeing, Anoxybacillus, Chryseobacterium, and Moraxella dominated meat samples (>50% of sequences). Many ASVs detected on meat were also found on facility surfaces, with some taxa widely distributed (e.g., Bacillus_S, Moraxella) and others location-specific (e.g., Luteimonas_A, Helicobacter_F at polishing tunnel; W16RD/Sphingomonas at cooling chamber wall).
• Long-read vs short-read resolution: Full-length sequencing improved species-level classification by 30.3% over short-read MiSeq data (species-level: 74% vs 56.8%; genus-level: 91% vs 80.6%), while both platforms captured consistent community transitions.
• SourceTracker results (transmission map): Major contamination sources to meat were the polishing tunnel (especially whips, also nozzles and water), gloves of employees, and the classification railing. Example: at the final position (Truck), 11.4% of meat-associated bacteria originated from evisceration gloves; gloves also contributed at Truck (8.3%) and classification (2.8%); the classification railing contributed 4.4%. Anal swab and sticking samples contributed minimally, consistent with effective singeing and rectum sealing. Non-contact environmental sites (locks, walls) were not major contributors, suggesting direct contact as the dominant transfer route. Unknown sources accounted for 31.6% of bacteria on meat.
• Taxon-specific transmission: Escherichia originated almost exclusively from anal swabs. Lactococcus, Staphylococcus, Chryseobacterium, and Moraxella had multiple sources. Taxa clustered by shared-source similarity (e.g., Flavobacterium and Lactobacillus_H from evisceration gloves; Lactococcus and Bacillus_L from splitting saw). Different Chryseobacterium populations traced to distinct sources (e.g., polishing tunnel vs locks). Moraxella spp., abundant on meat, most likely transferred from polishing whips, employee gloves, and the classification railing.
• Sequencing depth simulation for SourceTracker: Results from rarefied datasets strongly correlated with the original dataset (Spearman’s rho 0.95–0.99). Mean squared differences decreased with larger dataset sizes; significant differences across depths (Kruskal-Wallis χ²=125.9, p<0.0001). Hit ratio increased with depth: 53.6% at 200 reads up to 89.8% at 7,712 reads; significant differences across depths (χ²=90.9, p<0.0001). Unknown classification rate decreased from 45.4% (200 reads) to 28.4% (7,712 reads), with significant differences (χ²=93.9, p<0.0001). Datasets with ≥1,000 reads/sample yielded comparable classification rates and community shifts to deeper datasets.

The findings support the hypothesis that much of the microbial community on processed pork meat is introduced during processing rather than being animal-derived. Singeing effectively reduced initial skin-associated microbes, while the polishing step reintroduced and spread microorganisms across carcasses, increasing both counts and diversity, consistent with equipment acting as reservoirs and dispersal points. SourceTracker identified key transmission hotspots—polishing tunnel equipment, employee gloves, and the classification railing—revealing contamination routes not apparent from standard culture-based hygiene indicators alone. Taxon-resolved analyses showed that different species occupy niche-specific habitats within the facility and have distinct transmission patterns, underscoring the value of high taxonomic resolution for source attribution. Full-length 16S rRNA sequencing provided substantially improved species-level classification compared with short reads and, combined with SourceTracker, offered actionable insights for targeted interventions (e.g., enhanced cleaning/disinfection of polishing whips, glove hygiene practices, and railing sanitation). Simulation results indicated that moderate sequencing depth (~1,000 reads/sample) can be sufficient for robust source tracking and beta diversity assessments in this context, enabling cost-effective routine monitoring. While strain-level resolution (via WGS/metagenomics) remains critical for molecular epidemiology and diagnostics, the presented approach is well-suited for ongoing environmental monitoring and identification of general transmission routes in food-processing facilities.

This study mapped microbial diversity and transmission routes along a pork-processing line using high-throughput full-length 16S rRNA gene sequencing and SourceTracker. It pinpointed critical contamination sources—polishing tunnel equipment, employee gloves, and the classification railing—and linked specific spoilage-associated taxa (e.g., Moraxella, Chryseobacterium) to these sources, enabling targeted hygiene measures. Full-length 16S rRNA sequencing improved taxonomic resolution over short-read approaches and, at moderate sequencing depths, delivered reliable source tracking results suitable for routine monitoring. The approach can be extended to other food-processing environments to improve hygiene standards, enhance food safety, and reduce waste. Future work should include standardized sampling and bioinformatics workflows, expanded source sampling (e.g., soil, human skin), multi-day and multi-facility studies, and integration with strain-level genomics when epidemiological precision is required. Advances in long-read technologies (e.g., Sequel II; full rRNA operon sequencing) will further increase throughput and resolution, facilitating broader implementation in monitoring systems.

• Source attribution uncertainty: Approximately 31.6% of meat-associated bacteria were assigned to unknown sources, potentially due to shallow sequencing in some samples or missing primary sources (e.g., soil, human skin) in the sampling design.
• Directionality cannot be inferred: SourceTracker presumes source/sink relationships without establishing directionality; some transfers could occur bidirectionally between meat and surfaces.
• Sequencing depth and resolution: While 1,000 reads/sample were adequate for this facility, lower depths (e.g., 200–500 reads) reduced hit ratios and increased unknown rates; strain-level resolution was not achieved with 16S rRNA sequencing, limiting pathogen/source precision.
• Scope and generalizability: Sampling occurred over a single day in one facility; results may not generalize across facilities, shifts, or seasons. Airborne transfer was inferred as marginal based on sampled sites; unmeasured routes may exist.
• Cost and workflow variability: Long-read sequencing can be more expensive than short-read platforms; lack of standardized sampling and bioinformatics workflows hinders cross-study comparability.