Between January and March 2024, a cross-sectional study was conducted at four artisanal cheese distribution centers in Tijuana, Baja California, Mexico. From each site, three sample types were collected: fresh cheese (Q), food-contact surfaces (S), and ambient air (A). Samples totaled 12 (4 Q + 3 S + 5 A), and all assays were performed in duplicate ( Table 1 ).

Fresh cheese samples (100 g) were placed in sterile bags and kept at 4˚C - 8˚C for up to 4 h. Food-contact surfaces (20 - 100 cm²) were swabbed using pre-moistened swabs in tryptic soy broth, following ISO 18593: 2004 [12] . Ambient air samples were obtained by settle plate method on Plate Count Agar for 10 min at positions determined by p = 0.15 √S (where S is room area in m²) according to UNE 171330-1: 2008 [13] .

After incubation for 24 h at 37˚C on tryptic soy agar (TSA) and MacConkey agar (BD Difco, USA), each sample was processed in accordance with the relevant Mexican Official Standards ( Table 2 ). From every plate, no more than five colonies that differed in morphology or pigmentation were sub-cultured onto brain–heart infusion (BHI) agar; duplicate morphotypes were discarded to avoid over-representation. Pure cultures were characterised by Gram stain, oxidase and catalase tests, followed by species-level identification with the VITEK-2 system (bioMérieux).

The analyser was checked every 48 h with E. coli ATCC 25922 and S. aureus ATCC 29213, and results had to fall within CLSI M100-S34 acceptance limits [14] .

Presumptive Gram-negative isolates were identified using the VITEK 2 GN system (bioMérieux, Mexico) with a 0.5 McFarland inoculum. Results were available after approximately 18 h. Pure cultures were cryopreserved in TSB with 20% glycerol at −80˚C.

Antibiotic susceptibility was determined by VITEK 2 AST-GN cards and interpreted according to CLSI M100 (2024) breakpoints. Table 3 lists the antibiotics tested.

To characterize the growth kinetics of five selected Gram-negative genera (Escherichia coli, Serratia marcescens, Pantoea agglomerans, Raoultella planticola and Citrobacter freundii), each strain was grown for 18 h in BHI at 37˚C and adjusted to 0.5 McFarland (≈1 × 10⁸ CFU/mL) with a DensiCHEK Plus (bioMérieux). Only Gram-negative isolates were included because the study did not plan to analyze the rpoS gene in Gram-positive cocci. Erlenmeyer flasks (250 mL) containing 100 mL BHI were inoculated to ~1 × 10⁶ CFU/mL (1% v/v) and incubated at 37˚C, 120 rpm. At regular intervals ( Table 4 ), 2 mL samples were withdrawn and the OD₆₀₀ was measured in duplicate on a Genesys 20 spectrophotometer (Thermo Scientific) using sterile BHI as blank. The exponential phase was identified from OD₆₀₀-versus-time plots, and the specific growth rate (µ) was calculated with Equation (1).

µ = [ln(OD₂) − ln(OD₁)]/(t₂ – t₁) (1)

and generation time (g) as Equation (2)

g = ln2/µ. (2)

Data were compared by one-way ANOVA (p < 0.05). E. coli ATCC 25922 served as quality control, and any curve with R² < 0.95 was repeated.

These kinetic parameters (µ, g) were used to identify the onset of stationary phase for subsequent assays, ensuring all cultures were sampled at comparable physiological states.

Biofilms were quantified using the microtiter plate method of O’Toole [18] and classified per Stepanović et al. [19] . Stationary-phase cultures were adjusted to ≈10⁶ CFU/mL (1:100 dilution of 0.5 McFarland cultures). Seventeen Gram-negative isolates were tested in condition C1 and additionally under C2-C4 ( Table 5 ). Briefly, 200 µL of each suspension was inoculated into 96-well polystyrene plates, washed three times with PBS, fixed at 60˚C for 15 min, and stained with 0.1% crystal violet for 15 min. After rinsing and air-drying, dye was solubilized in 200 µL 30% (v/v) glacial acetic acid, and OD₆₃₀ [20] was measured with a Multiskan FC (Thermo Fisher, USA) at the crystal violet absorption maximum in acidic medium. Biofilm production was categorized using ODc = mean blank + 3 SD: non-producers (OD ≤ ODc), weak (ODc < OD ≤ 2 × ODc), moderate (2 × ODc < OD ≤ 4 × ODc) or strong (OD > 4 × ODc). Statistical analysis was performed by one-way ANOVA with Tukey’s post-hoc test (p < 0.05).

Specific primer sets were carefully designed for each targeted genus, as detailed in Table 6 , resulting in PCR products of varying lengths: 226 bp for Escherichia coli, 201 bp for Serratia marcescens, 242 bp for Pantoea agglomerans 236 bp for Raoultella planticola, and 157 bp for Citrobacter freundii. As illustrated in Figure 4 , agarose gel electrophoresis (1.2% TBE) confirmed that all Gram-negative isolates generated bands corresponding to their respective genera. Notably, the positive controls—E. coli K-12 MG1655 and C. freundii ATCC 8090—produced PCR product sizes aligned with those observed in the field isolates. In contrast, the negative control (PCR-grade water) yielded no amplification. A purification protocol was optimized to isolate high-quality DNA before molecular testing, adapted from Weerakkody et al. [21] . This procedure was conducted in a biosafety cabinet under aseptic conditions, using sterilized materials and strict sanitation practices to prevent cross-contamination. Approximately 20 μL of bacterial biomass from fresh cultures (TSA plate scraping) was transferred to a 1.5 mL microtube for cell lysis. 300 μL of lysis buffer (containing NaOH and SDS) was added and mixed for 15 seconds on a vortex mixer. The samples were incubated for 5 minutes at 80˚C in a water bath to denature proteins and disrupt the cell wall. Following incubation, the tubes were briefly cooled, and 1.5 μL of RNase A (10 mg/mL) was added, followed by a 10-minute incubation at 37˚C to degrade residual RNA. Subsequently, 100 μL of precipitation buffer (4M potassium acetate, pH 4.0) was added and mixed gently. The tubes were centrifuged for 3 minutes at 13,000 rpm, discarding the pellet containing precipitated detritus and proteins. The clarified supernatant was transferred to a clean tube, and 300 μL of isopropanol was introduced to precipitate the DNA, which was mixed by inversion. The samples were stored at −20˚C for 24 hours to ensure complete precipitation. Afterward, the samples were centrifuged for 1 minute at 13,000 rpm, with the isopropanol discarded. The resultant pellet was washed with 100 μL of 70% ethanol and centrifuged again. The pellet was air-dried at room temperature until all residual ethanol evaporated. Ultimately, the DNA was resuspended in 100 μL of nuclease-free water and stored at −20˚C until further use [22] .

The oligonucleotides were designed from the sequences reported in the National Library of Medicine ( https://www.ncbi.nlm.nih.gov/ ) Gene ID: 947210, (for Escherichia coli), Gene ID: 57426935 (Raoultella planticola), Gene ID: 66824848 (Pantoea agglomerans), Gene ID: 86999769 (Citrobacter freundii), Gene ID: 93695333 (Serratia marcescens) and using the software Primer3web version 4.1.0. [23] .

For the detection of the rpoS gene, conventional PCR was performed on the extracted genomic DNA. Each 25 μL reaction mix comprised 12.5 μL of 2 × Taq PCR Master Mix (Bioneer, USA), five μL of a primer mixture (both forward and reverse primers at 10 μM; refer to Table 2 ), two μL of template DNA, and 5.5 μL of nuclease-free water. Amplification was carried out in a ATC 201 Thermal Cycler (Nyx Technik, Inc. San Diego, CA, USA). following this protocol: initial denaturation at 95˚C for 10 minutes, followed by 30 cycles of denaturation at 95˚C for 30 seconds, annealing at 66.8˚C for 30 seconds, and elongation at 72˚C for 30 seconds, concluding with a final extension at 72˚C for 10 minutes. The PCR products were resolved on a 1.2% agarose gel (in TBE buffer) at 120 V for 25 minutes and visualized under UV light post-staining with ethidium bromide, using a 100 bp ladder (Invitrogen) for size estimation of the amplicons. Positive controls included genomic DNA from Escherichia coli K-12 MG1655 and Citrobacter freundii ATCC 8090, while PCR-grade water was a negative control. The detection of rpoS was confirmed by the presence of a band of the expected size, correlating this finding with the intensity of the biofilm.

A total of 30 viable isolates were obtained from fresh cheese, food-contact surfaces and ambient air: 17 Gram-negative (56.7%) and 13 Gram-positive (43.3%). Sampling was carried out in duplicate per matrix and each sample was labelled with a unique alphanumeric code (date-site-matrix-replicate). Fresh-cheese wedges (~100 g) were placed directly into sterile Whirl-Pak^® bags by the sampling team wearing new nitrile gloves, which were changed between vendors; the outer 1 cm of rind was aseptically trimmed inside the bag with a sterile disposable scalpel to minimise carry-over from retail knives. Bags were stored at 4˚C - 8˚C in insulated coolers equipped with a calibrated thermometer and were processed within 4 h. Surfaces (20 - 100 cm²) were swabbed with single-use, pre-moistened swabs and transported on ice; ambient air was collected by 10-min settle plates sealed with Parafilm^TM. A complete set of field blanks (bag, swab, plate) accompanied every round, and none showed growth after incubation. All subsequent handling was performed in a class II biosafety cabinet under aseptic conditions.

Figure 1 summarizes their distribution by sample type. The Gram-negative isolates included Escherichia coli (10), Raoultella planticola (3), Pantoea agglomerans (2), Serratia marcescens (1) and Citrobacter freundii (1). The Gram-positive isolates comprised Staphylococcus aureus (6), Kocuria kristinae (3), Propionibacterium acnes (1), Staphylococcus epidermidis (1), Staphylococcus auricularis (1) and Eggerthia catenaformis (1). Fresh cheese yielded the greatest diversity (21/30 isolates), followed by surfaces (8/30); only one Gram-positive cocci (S. epidermidis) was recovered from air ( Figure 1 ).

Using the VITEK 2 AST-GN system, only 3 of 17 Gram-negative isolates (17.6%) exhibited resistance ( Table 6 ): two P. agglomerans (cheese and surface) were resistant to cefoxitin (MIC = 16 µg/mL), and one E. coli (cheese) was resistant to ampicillin/sulbactam (MIC ≥ 32 µg/mL). All other Gram-negatives—including R. planticola, S. marcescens and C. freundii—were susceptible to the full panel. Among the 13 Gram-positive isolates, only a single Staphylococcus epidermidis from air exhibited tetracycline resistance (MIC ≥ 16 µg/mL). All S. aureus, K. kristinae, P. acnes, S. auricularis and E. catenaformis strains were susceptible to every antibiotic tested.

Figure 2 displays the growth curves at 37˚C for representative strains of E. coli, S. marcescens, P. agglomerans, R. planticola and C. freundii. All followed a typical sigmoid pattern, with a short lag phase (<45 min) and transition to stationary phase between 6 and 10 h, depending on the genus. Also during the development of the growth curve, the growth rate and generation time were determined ( Table 7 ).

A one-way ANOVA revealed significant differences in µ among genera (F = 11.6; p < 0.01). Tukey’s post-hoc test showed that E. coli grew significantly faster than P. agglomerans, R. planticola and C. freundii (p < 0.05), while differences between S. marcescens and P. agglomerans were not significant (p = 0.08). These parameters defined the precise sampling point at OD₆₀₀ ≈ 1.2 ± 0.1 for stationary phase prior to biofilm assays, ensuring comparable physiological states.

All 17 Gram-negative isolates were tested under four conditions: C1 (BHI + 2% glucose, 37˚C), C2 (peptone water, 37˚C), C3 (BHI + 2% glucose, 22˚C) and C4 (BHI + 2% glucose, 4˚C). Biofilm biomass was measured as OD₆₃₀ after crystal violet solubilization in 30% glacial acetic acid.

Figure 3 shows the mean ± SD of OD₆₃₀ for each genus and condition. Overall, C1 produced the highest biomass (mean 0.62 ± 0.14), followed by C3 (0.45 ± 0.12). C2 and C4 reduced biomass to 0.31 ± 0.10 and 0.27 ± 0.09, respectively. A two-way ANOVA (genus × condition) confirmed significant effects of both factors (p < 0.05) and their interaction (p = 0.03), indicating genus-dependent responses to temperature and nutrient availability.

Using ODc = mean blank + 3 SD (0.12), isolates were classified as non-producers (OD ≤ ODc), weak (ODc < OD ≤ 2 × ODc), moderate (2 × ODc < OD ≤ 4 × ODc) or strong (OD > 4 × ODc) ( Table 8 ).

A chi-square test of homogeneity (χ² = 14.2; df = 6; p = 0.028) showed that intensity distributions differed by genus: C. freundii was the only “strong” biofilm former in all conditions, while R. planticola was mainly “non-producer” (67%). Most E. coli and S. marcescens were classified as “weak”, although two E. coli strains reached “moderate” in C1.

All isolates tested positive for the rpoS gene confirming the presence of the stationary-phase sigma factor. However, rpoS presence alone did not predict biofilm robustness: R. planticola carried rpoS but produced little biomass, whereas C. freundii combined slow growth with dense biofilms under all conditions, suggesting that additional extracellular matrix components and regulatory mechanisms govern biofilm strength.

Specific primer sets were meticulously designed for each targeted genus, as outlined in Table 9 , resulting in PCR products of distinct lengths: 226 bp for Escherichia coli, 201 bp for Serratia marcescens, 242 bp for Pantoea agglomerans, 236 bp for Raoultella planticola, and 157 bp for Citrobacter freundii.

As demonstrated in Figure 4 , the agarose gel electrophoresis (1.2% TBE) revealed that all Gram-negative isolates produced bands corresponding to their respective genera. Notably, the positive controls—E. coli K-12 MG1655 and C. freundii ATCC 8090—exhibited PCR product sizes consistent with those of the field isolates. At the same time, the negative control (PCR-grade water) showed no amplification, effectively ruling out any potential reagent contamination. Gram-positive strains were excluded from this analysis, as the rpoS gene is specific to Gram-negative bacilli [19] . The rpoS gene is a critical regulator of biofilm formation, enhancing bacterial resilience in challenging environments. Factors influencing rpoS activity include oxidative stress, osmotic pressure, and nutrient availability. A thorough understanding of the regulatory networks governed by rpoS is pivotal for devising targeted biofilm management strategies, ultimately aimed at augmenting food quality and safety. 100% of Gram-negative isolates carry rpoS, although biofilm intensity varied among genera.

The detection of only three resistant phenotypes among Gram negatives (17.6%) and one among Gram positives confirms the low prevalence of antimicrobial resistance (AMR) in fresh cheeses observed by Cabrera-Díaz et al. [23] in central Mexico. Nevertheless, cefoxitin resistance in Pantoea agglomerans and ampicillin/sulbactam resistance in one E. coli suggest local selective pressure, possibly due to β-lactam use in dairy herds as documented by the FAO [24] . Although no ESBL or multidrug-resistant profiles were found, the WHO warns that even low percentages can rapidly amplify within biofilm niches [2] .

Growth curves showed that E. coli (µ ≈ 0.28 h⁻¹) multiplied significantly faster than P. agglomerans, R. planticola and C. freundii. However, biofilm biomass did not correlate linearly with growth rate: C. freundii, despite being the slowest grower (µ ≈ 0.19 h⁻¹), produced the only “strong” biofilms under all conditions. This supports the concept that prolonged stationary phase enhances matrix gene expression and three-dimensional architecture, as described for Enterobacter spp. by Pires et al. [6] .

Biofilm biomass was the highest in BHI at 37˚C (C1), yet incubation at 22˚C (C3) retained 73% of the biomass, indicating that typical retail-display temperatures still allow substantial biofilm formation. Cold storage at 4˚C reduced biomass by ≈ 55%, confirming that refrigeration slows but does not eliminate adherence [8] . The significant genus × condition interaction (p = 0.03) highlights genus-specific responses: for example, P. agglomerans doubled its OD₆₃₀ in peptone water compared to BHI, possibly driven by a “biofilm-by-starvation” mechanism reported for environmental Enterobacterales [6] .

The 100% positivity for rpoS confirms a universal stress-response potential among the isolates. However, rpoS presence alone did not predict biofilm strength: R. planticola carried the gene but was largely a non-producer, while C. freundii combined slow growth with dense biofilm formation under all conditions, suggesting that upstream regulatory mutations or interaction with bolA modulate matrix synthesis [7] . It is plausible that R. planticola has less efficient rpoS regulation or produces an EPS with weaker affinity for polystyrene.

No significant correlation was found between biofilm biomass (OD₆₃₀) and the number of antibiotics to which an isolate was resistant (ρ = 0.21, n = 30, p = 0.32). Similar results were reported by Robbe-Saule et al. for Salmonella [9] and by Priego-Salado et al. for dairy Citrobacter isolates [24] . Although oxidative stress can activate the AcrAB efflux pump via rpoS and slightly increase multidrug tolerance in E. coli [25] , recent genomic studies indicate that rpoS chiefly aids environmental adaptation and contributes little to clinically relevant AMR [26] . Our data support this view: in artisanal cheeses, rpoS enhances persistence through biofilm formation, whereas antimicrobial resistance depends mainly on other factors

Although AMR prevalence was low and most strains formed weak biofilms, the isolation of C. freundii—a strong biofilm former carrying rpoS—poses a potential hazard. Pipeline model studies demonstrate that dense Citrobacter biofilms protect coliform pathogens and hinder alkaline detergent penetration [27] [28] . Furthermore, tolerance at 22˚C implies that open-market display surfaces facilitate persistence and dissemination.

Fresh, raw-milk cheeses sold at room temperature can carry rpoS-positive Gram-negative bacteria that form biofilms on knives and display cases. These biofilms may survive routine rinsing for at least two days, seeding the cheese with stress-tolerant cells and occasional β-lactam resistance. Similar strains were traced from artisanal cheeses to patients during recent outbreaks in Europe and Latin America [29] . Even low-level resistance can complicate treatment in vulnerable consumers, yet Mexico’s NOM-243 focuses only on coliform counts and overlooks biofilms. Based on our findings, we recommend: 1) disinfectants validated against biofilms, 2) regular disassembly and scrubbing of utensils, and 3) targeted monitoring of rpoS and key AMR genes along the producer-to-retailer chain. These measures would reduce microbial load and the selection pressure for resistance, aligning local practice with the One Health guidance of the FAO and WHO [30] .

1) This study was limited to a single municipality and a winter–spring sampling period, so seasonal variation may alter microbial diversity. 2) Biofilm assays were performed on polystyrene, whereas stainless steel predominates in dairy equipment [31] [32] . 3) Biomass was estimated solely by the crystal-violet OD₆₃₀ assay, which stains both living cells and extracellular matrix; differences in matrix composition, washing intensity and plate surface chemistry can over- or under-estimate viable biofilm [33] [34] . Complementary methods such as viable-cell counts or confocal microscopy should be included in future work. 4) Upcoming studies should therefore assess rpoS and EPS-gene expression via RT-qPCR. 5) evaluate dual-species biofilms involving Gram-positive cocci. 6) validate cleaning strategies such as enzyme-based disinfectants.