Strain sharing and persistence of microbial pathogens colonizing the skin of residents in a regional nursing home network Authors: Yaovi M. G. Hounmanou, Gabrielle M. Gussin, Sean Conlan, Raveena D. Singh, Clay Deming, Diana M. Proctor, Marco Teixeira, Ashlee M. Earl, Colin J. Worby, Heidi H. Kong, Susan S. Huang, Julia A. Segre Published: 2026 ABSTRACT Antimicrobial resistance (AMR) is a health threat disproportionately affecting nursing home (NH) residents. Surveillance and infection control in NHs are restricted to nares or perirectal cultures, overlooking skin colonization and multidrug-resistant organisms (MDROs) not recovered by selective media. Here, within the PROTECT trial NCT03118232, we show, that NH residents’ skin serves as a reservoir of transmissible MDROs. We analyzed 207 groin and axilla swabs from 38 residents across 15 California NHs using metagenomics, culturing, and genome sequencing. Culture detected MDROs in 10 of 38 residents (26.3%), including 4 (10.5%) with ESBL-producing Escherichia coli sequence type (ST)131/ST648 and 7 (18.4%) with methicillin-resistant Staphylococcus aureus. Skin microbiome analysis by metagenome-assembled genomes identified broader MDRO colonization, including 27 (71.1%) with E. coli ST93, 14 (36.8%) with Staphylococcus epidermidis ST2, 16 (42.1%) with Proteus mirabilis, 7 (18.4%) with Providencia stuartii, 7 (18.4%) with Enterococcus faecalis, and 5 (13.2%) with Pseudomonas aeruginosa. Colonization persisted after bathing. Clonal E. coli ST93 was shared by 27 residents across 9 facilities, and 5 resident pairs carried clonally related strains of ≥2 MDRO species, suggesting polymicrobial transmission. We confirmed skin as a reservoir of MDROs, utilizing metagenomics to detect colonization and transmission pathways, supporting AMR surveillance in long-term care. Introduction Antimicrobial resistance (AMR) threatens decades of medical progress, rendering once-treatable infections increasingly difficult or impossible to manage. Multidrug-resistant organisms (MDROs) colonizing and infecting patients undermine treatment efficacy, increase morbidity and mortality, and escalate healthcare costs. In 2021, an estimated 4.71 million deaths were associated with bacterial AMR, with global disability-adjusted life years projected to reach 46.5 million by 20501. The economic toll of direct healthcare costs associated with AMR is estimated at 66 billion USD and predicted to reach 159 billion USD without improved interventions2. Older adults are disproportionately affected, with AMR-related deaths increasing by more than 80% in those aged 70 years and older between 1990 and 20211. Many of these vulnerable individuals reside in nursing homes (NH)3. Nursing homes are increasingly recognized as persistent reservoirs for MDROs4. In the USA, more than 1.3 million people live in 15,300 nursing homes, where MDRO prevalence is four to six times higher than in hospitals3,5. Prolonged stays, frequent antibiotic use, hospital transfers, and close-contact communal living amplify opportunities for colonization and onward transmission6. Colonization is often asymptomatic but increases the risk of infection, making detection essential for prevention. Most studies of MDRO carriage in NHs have relied primarily on culture-based detection from perirectal sites, providing only a partial view of MDRO ecology and transmission risk7,8. These approaches overlook skin colonization, fail to capture non-resistant or co-colonizing strains on selective media, and lack strain-level resolution of bacterial diversity and persistence. Recent shotgun metagenomic data primarily of residents in a single nursing home showed that the skin can harbor MDROs, including Candida auris and ESKAPE pathogens (Enterococcus faecium, Staphylococcus aureus, K. pneumoniae, A. baumannii, P. aeruginosa, Enterobacter species)9. The inguinal crease and axilla are recognized hotspots for microbial colonization and potential transmission7,10,11. As moist, occluded niches with frequent contact with textiles and caregivers, they represent epidemiologically relevant sites of microbial colonization amongst NH residents. Yet, AMR surveillance focuses almost exclusively on nares and perirectal sites and relies on culture-based methods that can miss co-colonizing strains, morphologically similar lineages, or organisms lacking expected resistance phenotypes10. Shotgun metagenomics analysis enables recovery of metagenome-assembled genome (MAGs)12, providing strain-level resolution directly from complex samples10,13. Integrated genomic approaches combining culturomics and metagenomics now offer an unprecedented ability to track MDRO persistence and spread. Few studies have performed skin microbiome analyses of NH residents9,14 and none have combined metagenomics with bacterial isolate genome sequencing to study MDRO skin ecology. This study therefore, advances the field by applying an integrated genomic approach to reveal the extent, persistence, and strain-sharing patterns of MDRO skin colonization in nursing home residents. Project PROTECT trial, a cluster-randomized study of nursing homes in California, which demonstrated that MDRO colonization was highly prevalent in NH residents and strongly associated with increased hospital transfers5. During this trial, skin swabs were collected from a convenience sample of residents for bacterial isolation and microbiome analysis. Here, we show that integration of shotgun metagenomics and isolate genome sequencing identifies extensive MDRO colonization, persistence, and transmission among 38 residents in 15 nursing homes. We define the extent of MDRO skin colonization, characterize inter-resident and inter-facility strain sharing, resolve the resistome structure, and assess strain persistence. Our findings reveal cryptic, skin colonization, and polymicrobial strain sharing of MDROs among nursing home residents, informing potential future infection control strategies for long-term care settings. Results Selective culture reveals diverse MDROs on NH residents’ skin To test for skin carriage of MDROs on nursing home residents, the axilla and groin of 38 residents of 15 NHs (2–3 residents per NH) were sampled and cultured on antibiotic selective media. Of the 10 NH residents (26.3%) with skin MDRO carriage, five residents (13.2%) carried extended-spectrum β-lactamase (ESBL)-producing gram-negative bacteria. Three residents (7.9%) in 3 different NHs carried E. coli sequence type (ST)131, a globally disseminated uropathogenic ESBL lineage, one resident co-carried ESBL-producing E. coli ST648 and Providencia stuartii and the fifth resident carried ESBL-producing Proteus mirabilis. Vancomycin-resistant Enterococcus faecalis (VRE) was recovered from 2 residents (5.3%), including ST109 and ST778 lineages (Table 1). In addition, methicillin-resistant Staphylococcus aureus (MRSA) was cultured from the skin of seven residents (18.4%). These findings confirm that skin can act as a reservoir for clinically relevant, antimicrobial-resistant organisms, with potential implications for infection risk and resident-to-resident dissemination.Table 1Skin-colonizing MDROs isolated from NH residents in this studySpeciesST#NHsResidents colonizedPhenotypic profileEscherichia coliST1313 (20%)3 (7.9%)ESBLST6481 (6.7%)1 (2.6%)ESBLEnterococcus faecalisST1091 (6.7%)1 (2.6%)VREST7781 (6.7%)1 (2.6%)VREProteus mirabilisNA1 (6.7%)1 (2.6%)ESBLProvidencia stuartiiST831 (6.7%)1 (2.6%)ESBLStaphylococcus aureus-6 (40%)7 (18.4%)MRSAaOverall Skin MDRO carriage7 (46.7%)10 (26.3%)aThe MRSA isolates were not sequenced as part of this observational study. Table 1Skin-colonizing MDROs isolated from NH residents in this studySpeciesST#NHsResidents colonizedPhenotypic profileEscherichia coliST1313 (20%)3 (7.9%)ESBLST6481 (6.7%)1 (2.6%)ESBLEnterococcus faecalisST1091 (6.7%)1 (2.6%)VREST7781 (6.7%)1 (2.6%)VREProteus mirabilisNA1 (6.7%)1 (2.6%)ESBLProvidencia stuartiiST831 (6.7%)1 (2.6%)ESBLStaphylococcus aureus-6 (40%)7 (18.4%)MRSAaOverall Skin MDRO carriage7 (46.7%)10 (26.3%)aThe MRSA isolates were not sequenced as part of this observational study. Skin-colonizing MDROs isolated from NH residents in this study aThe MRSA isolates were not sequenced as part of this observational study. Opportunistic pathogens persist within a dysbiotic skin microbiome To place these culture-based findings in broader ecological context, we next analyzed the skin microbiome of these 38 residents from 15 NHs from two epidemiologically relevant skin sites (inguinal crease (Ic) and axilla (Ax)), sampled at 2–3 visits per resident between 12 h and 48 h after bathing (Fig. 1A). In total we sequenced and analyzed 223 samples including 207 skin swabs, and 16 sample controls (Supplementary Data 1 and Supplementary Fig. 1A). Analysis of the skin shotgun metagenomic data revealed a predominantly bacterial skin microbiome with typical commensal taxa (Cutibacterium, Corynebacterium, Staphylococcus spp.), coexisting with opportunistic pathogens such as Enterococci, E. coli, P. mirabilis, P. aeruginosa and Providencia spp that are not expected in a normal skin microbiome. Fungal (Malassezia spp.) and viral taxa (Betapapillomavirus, Uroviricota phages) were consistently detected but at low relative abundances (Fig. 1B).Fig. 1Overview of study design and skin microbiome composition analysis.A Study workflow for sample collection, sequencing, and genomic analysis of skin swabs collected from nursing home residents. B Stacked bar plots show the relative microbial abundances of individual nursing home residents, at two skin sites (axilla and inguinal crease) at three visits (Timepoint (T), T1, T2, T3). A was created using Biorender under publication license Segre, J. (2026) https://BioRender.com/krcae4k. Overview of study design and skin microbiome composition analysis. A Study workflow for sample collection, sequencing, and genomic analysis of skin swabs collected from nursing home residents. B Stacked bar plots show the relative microbial abundances of individual nursing home residents, at two skin sites (axilla and inguinal crease) at three visits (Timepoint (T), T1, T2, T3). A was created using Biorender under publication license Segre, J. (2026) https://BioRender.com/krcae4k. Across samples, the overall composition of NH residents’ skin microbiome appeared broadly similar between body sites and over time, with no significant differences detected in Shannon diversity between Ic and Ax samples (p > 0.05). Taken together, these findings point to the presence of persistent opportunistic pathogen populations on nursing-home residents’ skin rather than large compositional shifts over the short sampling period. We therefore reconstructed MAGs to resolve organism-specific colonization patterns at higher genomic resolution. MAG recovery reveals extensive skin colonization by opportunistic pathogens From the skin shotgun metagenomic data generated from 207 samples, we recovered 1185 (1160 bacterial and 25 fungal) MAGs, with a median of four MAGs per sample. Taxonomic classification revealed both skin commensals and clinically relevant microbes such as Escherichia coli (n = 68 MAGs; 60 near-complete), Staphylococcus epidermidis (n = 85; 77 near-complete), Proteus mirabilis (n = 40; 34 near-complete), Providencia stuartii (n = 17; 16 near-complete), and Pseudomonas aeruginosa (n = 11; 10 near-complete) (Fig. 2A; Supplementary Figs. 1B and 2A). Fungal MAGs were largely Malassezia spp., consistent with typical skin mycobiota (Fig. 2A; Supplementary Fig. S2B). MAG abundances correlated with read-level detection across samples (Fig. 2B), confirming analytic robustness.Fig. 2MDRO colonization of nursing home residents’ skin characterized by MAG recovery and abundance.A Recovery per resident at any timepoint of near-complete bacterial and fungal MAGs colored for representative species. Detailed overview of MAGs per resident per species are displayed in Supplementary Fig. 2A. B Relative abundance of same representative bacteria and fungi in shotgun metagenomic sequencing samples. Colors are defined in (A). C Frequency of MAG recovery from both inguinal crease and axilla of the same resident. MDRO colonization of nursing home residents’ skin characterized by MAG recovery and abundance. A Recovery per resident at any timepoint of near-complete bacterial and fungal MAGs colored for representative species. Detailed overview of MAGs per resident per species are displayed in Supplementary Fig. 2A. B Relative abundance of same representative bacteria and fungi in shotgun metagenomic sequencing samples. Colors are defined in (A). C Frequency of MAG recovery from both inguinal crease and axilla of the same resident. Dual-site colonization of a resident was common for E. coli, with concurrent detection in axilla and inguinal crease in 11 of 27 colonized residents (40.7%), and for P. mirabilis in 6 of 16 residents (37.5%). S. epidermidis and P. stuartii displayed lower dual-site co-occurrence rates (18.8% and 14.3%, respectively) (Fig. 2C). These findings suggest that high-burden colonizers can simultaneously occupy multiple skin niches, potentially increasing transmission risk within long-term care settings. Given the clinical importance of Enterobacterales, we next performed strain-resolved analyses of E. coli colonization dynamics. Metagenomics reveals dominant MDR E. coli ST93 missed by culture Genomic analyses of matched bacterial isolates and MAGs derived from shotgun metagenomic sequencing elucidated the E. coli strain that colonized resident skin. ESBL-selective culturing recovered E. coli from 4 residents (Fig. 3A), with genomic sequence analysis further delineating that 3 of these residents were colonized by E. coli ST131 and 1 resident was colonized by E. coli ST648. These four E. coli genomes carried multiple antibiotic resistance genes including blaCTX-M-15 and blaOXA-1 genes, which are known to confer an ESBL phenotype (Red asterisk in Fig. 3A; Supplementary Data 2). By contrast, MAG assembly from the skin metagenomic samples recovered E. coli ST131 MAGs from 4 residents, E. coli ST93 MAGs from 19 residents and additional singleton sequence types from 4 residents (Fig. 3A and Supplementary Data 3). None of the E. coli ST93 MAGs carried blaCTX-M genes, nor was the blaCTX-M gene detected in the corresponding skin metagenomes (Supplementary Data 4), providing a likely explanation of why these organisms did not grow on ESBL selective plates (Fig. 3B, red box). However, the E. coli ST93 MAGs still carried a broad MDR repertoire such as blaTEM, AmpC-family β-lactamases, aminoglycoside, fluoroquinolone, macrolide, sulfonamide, and tetracycline resistance genes (in the absence of other Proteobacteria on the contigs as confirmed by strainGE and BLASTn) (Fig. 3B; Supplementary Fig. 3; Supplementary Data 4). Moreover, an E. coli ST93 isolate was recovered in the wider PROTECT study from a California nursing home resident (not in our microbiome cohort) carrying blaCTX-M-1 on an IncI1 plasmid, demonstrating the lineage’s genetic capacity to acquire ESBL determinants. Together, these data reveal a two-tiered colonization landscape: low-prevalence ESBL-producing E. coli ST131 detected by cultivation and metagenomics and more prevalent MDR E. coli ST93 detected via metagenomics (Fig. 3).Fig. 3Nursing home residents’ skin colonization with multidrug resistant E. coli.A Detection of E. coli across residents with metagenome-assembled genomes (MAGs; circles), cultured isolates from ESBL-selective media (triangles), or both (squares). The red asterisk marks the β-lactamase gene content (i.e., carried blaCTX-M-15 and blaOXA-1.) of cultured E. coli isolates from ESBL-positive residents. B Presence/absence matrix of β-lactamase gene families recovered from metagenomic sequencing across all residents. Detection of blaCTX-M genes (highlighted in red box) was confined to residents with cultured ESBL-producing isolates. In contrast, E. coli ST93 MAGs lacked blaCTX-M but encoded other β-lactamase family genes and multidrug resistance determinants (see Supplementary Fig. 3). Figure panels were assembled using Biorender under publication license, Segre, J. (2026) https://BioRender.com/vgvb49x. Nursing home residents’ skin colonization with multidrug resistant E. coli. A Detection of E. coli across residents with metagenome-assembled genomes (MAGs; circles), cultured isolates from ESBL-selective media (triangles), or both (squares). The red asterisk marks the β-lactamase gene content (i.e., carried blaCTX-M-15 and blaOXA-1.) of cultured E. coli isolates from ESBL-positive residents. B Presence/absence matrix of β-lactamase gene families recovered from metagenomic sequencing across all residents. Detection of blaCTX-M genes (highlighted in red box) was confined to residents with cultured ESBL-producing isolates. In contrast, E. coli ST93 MAGs lacked blaCTX-M but encoded other β-lactamase family genes and multidrug resistance determinants (see Supplementary Fig. 3). Figure panels were assembled using Biorender under publication license, Segre, J. (2026) https://BioRender.com/vgvb49x. Regional clonal transmission of E. coli ST93 and diverse ST131 introductions Consistent with sequence typing, a combined phylogenetic analysis of E. coli MAGs and isolate genomes (Fig. 4A, B) revealed two distinct clusters (ST93, ST131). A high-resolution SNP-based phylogeny of E. coli ST93 genomes showed a single clade circulating regionally across 15 NHs (Fig. 4C), with ≤30 SNPs between any pair of MAGs and ≤5 SNPs between MAGs from different timepoints or body sites of the same individual (Fig. 4D and Supplementary Data 5). Notably, 27 residents (71.1%) across 9 facilities shared clonally related E. coli ST93 strains (≤10 SNPs), suggesting widespread transmission or common-source exposure. The clonal, regional strains from this study were however >100 SNPs away from ESBL-producing E. coli ST93 obtained in another California NH in 2019, as well as other E. coli ST93 genomes from elsewhere in the country (Fig. 4C).Fig. 4Comparative genomic analysis of E. coli strains from NH resident skin.A Network analysis of pairwise average nucleotide identity (ANI) among E. coli MAGs and isolate genomes from this study and the four isolate genomes from outside the cohort. Each node represents the highest-quality E. coli genome per resident, colored by resident. Circles denote MAGs and squares denote isolate whole genomes sequence (wgs). Edges connect genomes with ≥99.95% ANI, indicating high genetic relatedness. B Phylogenetic tree based on single-copy marker genes, including E. coli isolate genomes, MAGs from this study, and public reference genomes. Clades corresponding to E. coli ST131 and ST93 are highlighted in red and purple, respectively. C Phylogenetic tree of E. coli ST93 genomes from this study alongside representative references, inferred from core SNPs. The purple blocks along the tree mark samples from the same resident. gem10417s1v1_2006_Ic-Ax is the ESBL-producing E. coli ST93 obtained in another California NH in 2019. MAGs are labeled with residentID_bodysite_timepoint; D Histogram of pairwise core genome SNP distances among study-derived E. coli ST93 genomes. The red dashed line marks a 30-SNP threshold for strain-level relatedness. E Phylogenetic tree of E. coli ST131 genomes from study samples and public references, inferred from core SNPs. The red blocks along the tree mark samples from the same resident. F Histogram of pairwise core genome SNP distances between study-derived E. coli ST131 genomes and public E. coli ST131 references. The red dashed line denotes a 30-SNP threshold used to infer potential strain sharing. Abbreviations: Diff Subj different resident, Diff Fac different facilities/NHs. Figure panels were assembled using Biorender under publication license Segre, J. (2026) https://BioRender.com/s82td3a. Comparative genomic analysis of E. coli strains from NH resident skin. A Network analysis of pairwise average nucleotide identity (ANI) among E. coli MAGs and isolate genomes from this study and the four isolate genomes from outside the cohort. Each node represents the highest-quality E. coli genome per resident, colored by resident. Circles denote MAGs and squares denote isolate whole genomes sequence (wgs). Edges connect genomes with ≥99.95% ANI, indicating high genetic relatedness. B Phylogenetic tree based on single-copy marker genes, including E. coli isolate genomes, MAGs from this study, and public reference genomes. Clades corresponding to E. coli ST131 and ST93 are highlighted in red and purple, respectively. C Phylogenetic tree of E. coli ST93 genomes from this study alongside representative references, inferred from core SNPs. The purple blocks along the tree mark samples from the same resident. gem10417s1v1_2006_Ic-Ax is the ESBL-producing E. coli ST93 obtained in another California NH in 2019. MAGs are labeled with residentID_bodysite_timepoint; D Histogram of pairwise core genome SNP distances among study-derived E. coli ST93 genomes. The red dashed line marks a 30-SNP threshold for strain-level relatedness. E Phylogenetic tree of E. coli ST131 genomes from study samples and public references, inferred from core SNPs. The red blocks along the tree mark samples from the same resident. F Histogram of pairwise core genome SNP distances between study-derived E. coli ST131 genomes and public E. coli ST131 references. The red dashed line denotes a 30-SNP threshold used to infer potential strain sharing. Abbreviations: Diff Subj different resident, Diff Fac different facilities/NHs. Figure panels were assembled using Biorender under publication license Segre, J. (2026) https://BioRender.com/s82td3a. In contrast, E. coli ST131 genomes displayed greater diversity. While longitudinal samples within individuals were clonal (0–1 SNPs; Supplementary Data 6), intra-facility comparisons between residents typically exceeded 40 SNPs, suggesting multiple introductions. A few inter-facility comparisons (e.g., residents 0303 and 0802) approached potential transmission thresholds (28–52 SNPs) (Fig. 4E, F). Comparisons with public genomes from California highlighted occasional proximity to clinical E. coli ST131 isolates from California, implying regional overlap but overall high diversity. These results revealed a dominant, clonally conserved E. coli ST93 lineage spreading widely, contrasted with sporadic, diverse E. coli ST131 introductions. Methicillin-resistant S. epidermidis ST2 persists on skin and spreads across facilities To determine whether transmission patterns extended beyond Enterobacterales, we next explored the prevalent methicillin-resistant Staphylococcus epidermidis (MRSE) ST2 in nursing home residents. S. epidermidis MAGs were detected in the skin microbiome of 84% of residents, with ST2 MRSE MAGs found in 14 residents (36.8%) (Fig. 5A). The mecA gene was detected in 23 of 28 S. epidermidis ST2 MAGs as well as in the remaining 5 samples co-colonized with S. aureus. Within-individual SNP distances ranged from 0 to 2, indicating clonal persistence. Average nucleotide identity (ANI) and SNP-based networks revealed closely related S. epidermidis ST2 strains circulating within and across facilities (Fig. 5B–D). Among the many examples, highly related strains (4–6 SNPs) were shared by residents 0601 and 0602 of the same NH, strains from resident 0903 differed by 3 SNPs from those of resident 0502, as well as related strains (41–43 SNPs) in residents 1001 and 1002 who are also in the same NH. Strains from different NHs were also highly related (e.g., residents 0503 and 0601 shared strains differing by 4 SNPs, 0104 and 0602 shared strains with only a maximum of 10 SNPs difference; 0704 and 0602 with 13–14 SNPs; 0903 and 0602 with 6 SNPs, 0104 and 0704 with 17 SNPs), suggesting regional dissemination of a predominant MRSE ST2 clone persisting on skin and potentially spreading between nursing homes.Fig. 5Persistence and resident strain sharing of methicillin resistant Staphylococcus epidermidis ST2 within and between nursing homes.A Detection of the mecA gene across metagenomic samples. Each circle represents a metagenomic sample. Red-filled circles indicate detection of mecA; mecA + S. aureus is indicative of co-presence by S. aureus in the metagenomic reads as identified by StrainGE. B Average nucleotide identity (ANI) network of S. epidermidis ST2 MAGs. Nodes represent individual S. epidermidis ST2 genomes, colored by resident, and edges connect genomes sharing ≥99.95% ANI. The clustering of genomes from different residents and facilities reflects the circulation of a highly conserved S. epidermidis ST2 lineage across the nursing homes. C Maximum-likelihood phylogenetic tree of S. epidermidis ST2 rooted to the S. epidermidis type strain (GCF_001922515.1). D Histogram of pairwise SNP-based genomic distance between S. epidermidis ST2 samples. Red dashed lines indicate the 30 SNP threshold for defining closely related strains. Figure panels were assembled using Biorender under publication license, Segre, J. (2026) https://BioRender.com/30fcvv9. Persistence and resident strain sharing of methicillin resistant Staphylococcus epidermidis ST2 within and between nursing homes. A Detection of the mecA gene across metagenomic samples. Each circle represents a metagenomic sample. Red-filled circles indicate detection of mecA; mecA + S. aureus is indicative of co-presence by S. aureus in the metagenomic reads as identified by StrainGE. B Average nucleotide identity (ANI) network of S. epidermidis ST2 MAGs. Nodes represent individual S. epidermidis ST2 genomes, colored by resident, and edges connect genomes sharing ≥99.95% ANI. The clustering of genomes from different residents and facilities reflects the circulation of a highly conserved S. epidermidis ST2 lineage across the nursing homes. C Maximum-likelihood phylogenetic tree of S. epidermidis ST2 rooted to the S. epidermidis type strain (GCF_001922515.1). D Histogram of pairwise SNP-based genomic distance between S. epidermidis ST2 samples. Red dashed lines indicate the 30 SNP threshold for defining closely related strains. Figure panels were assembled using Biorender under publication license, Segre, J. (2026) https://BioRender.com/30fcvv9. Proteus, enterococcus and providencia exhibit persistence and facility-level spread Having established persistence and dissemination among the dominant lineages, we broadened the analysis to additional opportunistic pathogens detected on NH resident skin. Proteus mirabilis was detected in 18 residents (47.3%), including three ESBL-producing isolates carrying blaCTX-M-15 (Fig. 6A, B and Supplementary Data 2). E. faecalis was detected in 7 residents (18.4%), with VRE isolates typed as ST109 and ST778 (Supplementary Data 2). E. faecalis MAGs and isolates from 1001 and 1003 clustered phylogenetically (22–41 SNPs), consistent with NH-associated circulation, while the relatedness of a MAG from 1901 and an isolate from 0101 with 27 SNPs pointed to a possible regional reservoir (Fig. 6C, D). For P. mirabilis, SNP distances of 0–7 were observed among genomes recovered from the same resident, from residents within the same facilities and from residents in different facilities (e.g., 0104-Ax identical to 0901-Ax and 0902-Ic; 1003-Ax/Ic identical to 2401-Ax/Ic) (Fig. 6E, F and Supplementary Data 7). Extending this analysis to another species for which we recovered isolates and MAGs, Providencia stuartii was detected in 7 residents (18.4%), showing clonal persistence within individuals and near-identical strains (≤13 SNPs) shared between facilities (Fig. 6G, H; Supplementary Data 8). Additional MDROs included P. aeruginosa and M. morganii (Supplementary Fig. 4), often showing within-host clonality and occasional inter-resident sharing (<30 SNPs), while K. pneumoniae (ST20, ST1842) and S. pettenkoferi colonization remained largely resident-specific (Supplementary Figs. 5 and 6).Fig. 6Genomic relatedness of healthcare-associated microbes colonizing nursing home residents.A Detection of Proteus mirabilis and Enterococcus faecalis on skin across NH residents, based on recovery as metagenome-assembled genome (MAG) from shotgun metagenomics, cultured isolate, or both. B Antimicrobial resistance gene detection in isolate genomes. blaCTX-M-15 was found in P. mirabilis isolates, while vanHAX operon genes (vanH, vanA, vanX) were identified in E. faecalis. Resident 0510 was not a participant in our skin microbiome sequencing cohort but resides in one of the NHs included in this study. C Pairwise SNP distance heatmap among E. faecalis strains. Lower values indicate greater genomic similarity. Samples starting with the identifier “gem” are isolate genomes. MAGs are labeled with ResidentID-bodysite-timepoint-ST. Light blue shading is 10-100 SNPs; Blue shading is 100–1000 SNPs. D Maximum likelihood phylogeny of E. faecalis isolates and MAGs, displaying clustering by resident and timepoint; the blue shade marks E. faecalis from the same resident. E Phylogenetic tree of P. mirabilis strains shows resident-specific clustering, highlighted with black side bars. F Histograms for P. mirabilis of SNP distances of genomes from the same resident, genomes of different residents within the same NH and genomes from residents of different nursing homes. G Phylogenetic tree of Providencia stuartii MAGs shows resident-specific clustering, highlighted with green side bars. H Histograms for P. stuartii comparing SNP distances of genomes from the same resident with between residents. Red dashed lines indicate the 30 SNP threshold for defining closely related strains. Figure panels were assembled using Biorender under publication license, Segre, J. (2026) https://BioRender.com/8nr8pf3. Genomic relatedness of healthcare-associated microbes colonizing nursing home residents. A Detection of Proteus mirabilis and Enterococcus faecalis on skin across NH residents, based on recovery as metagenome-assembled genome (MAG) from shotgun metagenomics, cultured isolate, or both. B Antimicrobial resistance gene detection in isolate genomes. blaCTX-M-15 was found in P. mirabilis isolates, while vanHAX operon genes (vanH, vanA, vanX) were identified in E. faecalis. Resident 0510 was not a participant in our skin microbiome sequencing cohort but resides in one of the NHs included in this study. C Pairwise SNP distance heatmap among E. faecalis strains. Lower values indicate greater genomic similarity. Samples starting with the identifier “gem” are isolate genomes. MAGs are labeled with ResidentID-bodysite-timepoint-ST. Light blue shading is 10-100 SNPs; Blue shading is 100–1000 SNPs. D Maximum likelihood phylogeny of E. faecalis isolates and MAGs, displaying clustering by resident and timepoint; the blue shade marks E. faecalis from the same resident. E Phylogenetic tree of P. mirabilis strains shows resident-specific clustering, highlighted with black side bars. F Histograms for P. mirabilis of SNP distances of genomes from the same resident, genomes of different residents within the same NH and genomes from residents of different nursing homes. G Phylogenetic tree of Providencia stuartii MAGs shows resident-specific clustering, highlighted with green side bars. H Histograms for P. stuartii comparing SNP distances of genomes from the same resident with between residents. Red dashed lines indicate the 30 SNP threshold for defining closely related strains. Figure panels were assembled using Biorender under publication license, Segre, J. (2026) https://BioRender.com/8nr8pf3. Polymicrobial strain sharing reveals complex transmission networks To integrate findings at the resident level, we aggregated our findings to explore possible polymicrobial strain sharing among nursing home residents. Five resident pairs harbored clonally related strains from ≥2 MDRO species, underscoring transmission networks. For example, residents 0902 and 0903 shared near-identical P. mirabilis (3–4 SNPs) and E. coli ST93 (1–3 SNPs) in NH 09 (Fig. 7). Cross-facility polymicrobial sharing was observed between NHs 08 and 22 (P. stuartii, M. morganii, ≤30 SNPs). Overall, 27 residents across 9 NHs (71.1%) shared at least one clonally related MDRO strain with another resident (≤10 SNPs), highlighting extensive intra- and inter-facility dissemination of multidrug-resistant skin colonizers.Fig. 7Persistent colonization of NH residents’ skin post-bathing.A Persistent recovery of E. coli ST131 and ST93 MAGs before and after bathing, stratified by body site (axilla [Ax] and inguinal crease [Ic]) and organized by resident. Shape represents STs, with the color indicating time since last bath ranging from 12 h (lighter) to 48 h (darker). The x-axis lists NH residents, and the y-axis indicates sampling events. Red asterisk (*) indicates when bathing occurred between sampling and is only applied for ST131 and ST93. Gaps represent no E. coli MAG detected. B Persistent recovery of S. epidermidis ST2 MAGs before and after bathing, stratified by body site (axilla [Ax] and inguinal crease [Ic]) and organized by resident. Each tile represents a sample; colored circles indicate recovery of an S. epidermidis ST2 MAG and are colored by time since last bath. Red asterisks mark persistent detection of S. epidermidis ST2 across multiple timepoints in the same resident. In the facet “Inguinal Crease” for resident 0602 there is an asterisk but no isolate indicated for Sample 2 or 3 because the individual had a bath and a subsequent MAG but this MAG did not meet the genomic criterion for completeness. Figure panels were assembled using Biorender under publication license, Segre, J. (2026) https://BioRender.com/eo9rqi6. Persistent colonization of NH residents’ skin post-bathing. A Persistent recovery of E. coli ST131 and ST93 MAGs before and after bathing, stratified by body site (axilla [Ax] and inguinal crease [Ic]) and organized by resident. Shape represents STs, with the color indicating time since last bath ranging from 12 h (lighter) to 48 h (darker). The x-axis lists NH residents, and the y-axis indicates sampling events. Red asterisk (*) indicates when bathing occurred between sampling and is only applied for ST131 and ST93. Gaps represent no E. coli MAG detected. B Persistent recovery of S. epidermidis ST2 MAGs before and after bathing, stratified by body site (axilla [Ax] and inguinal crease [Ic]) and organized by resident. Each tile represents a sample; colored circles indicate recovery of an S. epidermidis ST2 MAG and are colored by time since last bath. Red asterisks mark persistent detection of S. epidermidis ST2 across multiple timepoints in the same resident. In the facet “Inguinal Crease” for resident 0602 there is an asterisk but no isolate indicated for Sample 2 or 3 because the individual had a bath and a subsequent MAG but this MAG did not meet the genomic criterion for completeness. Figure panels were assembled using Biorender under publication license, Segre, J. (2026) https://BioRender.com/eo9rqi6. E. coli ST93 and ST131 persist on skin after bathing As bathing was documented for this study, we leveraged our sampling on multiple days to assess whether E. coli strains identified by near-complete MAGs persisted on nursing home residents’ skin after bathing. Post bathing, persistent detection was observed in half the residents who were colonized by E. coli, including residents 0204, 0205, 0801, 0802, 1501, and 1502 carrying E. coli ST93, and 0802 and 0902 carrying E. coli ST131 (Fig. 7A). In every case, the same sequence type of E. coli was observed prior to and after the bath. For example, E. coli ST93 was detected on the axilla and inguinal crease of resident 1502 at both 12 and 24 h post-bath. Resident 0802 is remarkable with E. coli ST93 persisting on the axilla and E. coli ST131 persisting on the inguinal crease after bathing (Fig. 7A). Next, we sought to assess if there was a microbial community feature that distinguished residents with persistent skin colonization. Community-level microbiome profiles were broadly similar between persistent and non-persistent residents, indicating that post-bathing persistence occurred against a largely stable skin microbiome background. As described earlier, E. coli relative abundance was consistently higher in colonized compared to non-colonized samples across both body sites. In contrast, no clear differences in relative abundance were observed between E. coli persistent and non-persistent states. Alpha diversity at the pre-bathing timepoint did not differ between samples that remained colonized versus those that did not (Ax: p = 0.85; Ic: p = 0.53). Similarly, persistent S. epidermidis ST2 colonization was observed over 12–48 h post-bathing in residents 1001, 1002, 0104, 0601, and 0602 (Fig. 7B; Supplementary Figs. 7–8, and Supplementary Data 9). As observed above, S. epidermidis relative abundance and skin microbiome alpha diversity did not differ between persistent and non-persistent sites. However, these analyses are limited by the co-occurrence of additional S. epidermidis strains within the same site. Discussion AMR represents one of the greatest challenges to global health, with its burden projected to rise substantially in coming decades. Surveillance and response strategies have largely centered on acute-care hospitals and the livestock sector, leaving long-term care facilities, particularly nursing homes, underappreciated despite their role as persistent MDRO reservoirs5,9. Using a combined approach of shotgun metagenomics, MAGs, and isolate genome sequencing, we characterized MDRO colonization on nursing home residents’ skin and revealed ecological persistence, regional dissemination, and multi-species sharing events that are largely invisible to current culture- and peri-rectal-based diagnostics. Our data demonstrate that MDRO colonization on skin is persistent rather than transient. Across multiple timepoints, clonal strains of E. coli (ST93 and ESBL-producing ST131), methicillin-resistant S. epidermidis (MRSE ST2), and vancomycin-resistant E. faecalis (ST109 and ST778) persisted despite bathing intervals of 12-36 h, consistent with previous findings that skin acts as a long-term reservoir for clinically relevant microbes capable of causing opportunistic infections7. Such persistence has implications for autoinoculation of wounds or medical devices, increased risk of infection, and ongoing environmental contamination that may sustain transmission cycles in NH settings. One of the most striking findings was the widespread presence of E. coli ST93, detected in 71% of residents via MAGs yet absent from ESBL-selective culture plates. These strains lacked blaCTX-M genes but carried a broad resistome, including blaTEM, AmpC β-lactamases, and resistance determinants for aminoglycosides, sulfonamides, and fluoroquinolones. The detection of an E. coli ST93 isolate in a neighboring nursing home carrying blaCTX-M-1 on a conjugative IncI1 plasmid bordered by IS1380 elements provides direct evidence of this lineage’s potential to acquire ESBL genes, a configuration well documented in past ESBL epidemics15,16. Public data reveal additional high-risk variants of E. coli ST93, including isolates with blaCTX-M-55 and blaNDM-1 (SAMN28179333, SAMN39296765), as well as colistin resistance gene mcr−1 in Asia17, underscoring the lineage’s evolutionary potential to become a clinically significant MDRO. These results highlight that current resistance-targeted culturing may underestimate E. coli reservoirs in long-term care facilities, leaving silent dissemination unchecked. Similarly, S. epidermidis ST2, a healthcare-adapted clone linked to device-associated infections in immunocompromised hosts18,19 was detected in 37% of residents, often shared clonally within and across facilities, with S. epidermidis ST2 MAGs differing by ≤10 SNPs. These strains frequently carried mecA and circulated regionally, yet none were captured by culturing since methicillin-resistant S. epidermidis was not specifically sought after during culturing. These findings suggest that dominant, clinically relevant lineages may be silently spreading within NHs, escaping detection by routine clinical workflows focused on hospital pathogens that rarely involve skin sampling. Our metagenomic analysis further demonstrated that nursing homes act as interconnected nodes in regional transmission networks rather than isolated microbial ecosystems. Clonally related E. coli ST93, E. faecalis ST109/ST778, Proteus mirabilis, and Providencia stuartii strains were shared both within and between nursing homes, consistent with recent genomic epidemiology studies reporting MDRO dissemination among NH residents9,20,21. Frequent resident transfers between NHs and hospitals5, combined with limited infection prevention resources and delayed implementation of mandatory programs22, may contribute to this silent, inter-facility spread. Our data suggest that colonized residents, even in the absence of active infection, may serve as long-term, mobile reservoirs bridging care settings and contributing to regional MDRO circulation. In several cases, we identified polymicrobial strain sharing where resident pairs carried clonally related strains from two or more distinct MDRO species. These events occurred both within facilities and across different nursing homes, involving combinations of E. coli ST93, S. epidermidis ST2, E. faecalis, Providencia stuartii, and P. mirabilis. These polymicrobial colonization patterns highlight that strain sharing patterns in nursing homes are not pathogen-specific but involve broader microbial consortia that may co-evolve, persist, and potentially exchange resistance genes on skin. Infection control strategies focused on single priority pathogens may therefore fail to interrupt the true ecological networks sustaining AMR in long-term care facilities. While our study did not integrate environmental samples like was done in other studies8, the use of shotgun metagenomics here gives a much wider overview on the constellation of culturable and non-culturable MDROs in the nursing homes. Shotgun metagenomics provided critical insights into the skin resistome, detecting ARGs spanning multiple antibiotic classes, including β-lactams (blaTEM, blaCTX-M, blaOXA, mecA), aminoglycosides, sulfonamides, fluoroquinolones, phenicols, macrolides, and tetracyclines. Many of these genes were detected without corresponding culture isolates, highlighting the limitations of diagnostics dependent on phenotypic growth under selective conditions. For example, E. coli ST93 MAGs lacked ESBL determinants yet encoded multiple β-lactamases, illustrating a latent capacity for resistance that would remain undetected in standard screening. Although current binning algorithms limit direct plasmid-to-host ARG attribution23,24, read-level mapping confirmed key resistance determinants in high-risk lineages such as E. coli ST131 and ST93 (in the absence of other Proteobacteria on the contigs as confirmed by strainGE and BLASTn), supporting their potential clinical significance. Collectively, the ecological and genomic data indicate that resistance-targeted culturing alone underestimates MDRO burden on skin. Integrating metagenomic surveillance alongside targeted culturing could close detection gaps and guide more effective infection-prevention strategies across long-term care and the healthcare networks they connect to. This study has limitations. We sampled a maximum of three residents per nursing home, preventing detailed analysis of within-facility networks, including between roommates. Environmental sampling was not performed, restricting source tracking. Also, SNP-based thresholds (<30 SNPs) align with genomic epidemiology standards9 but cannot establish transmission directionality without denser temporal and epidemiological data. Despite these limitations, our findings establish nursing home resident skin as a stable and overlooked reservoir of MDROs with potential for persistence and dissemination within and across facilities. The recurrent detection of polymicrobial strain sharing underscores that long-term care residents are exposed to overlapping reservoirs and transmission routes involving multiple resistant taxa. Culturing confirmed that these MDROs were viable and replicating on the skin while metagenomics enabled detection of additional pathogens of concern that were missed by selective culturing. Expanding surveillance frameworks to include skin as relevant site and nursing homes, employing genome-informed, multi-species strategies that integrate metagenomics with traditional culturing, could close critical detection gaps and guide more effective infection control. As the population of frail, immunocompromised individuals in long-term care grows, addressing colonization and transmissions within nursing homes will be crucial to curb the spread of AMR within healthcare ecosystems. Methods Study design, participants, and skin sample collection This skin microbiome study was nested within Project PROTECT (NCT03118232), a cluster-randomized trial of 28 nursing homes in California, USA5. Between September and December 2018, a convenience sample of residents from facilities assigned to both routine care and decolonization arms were enrolled. A total of 38 residents from 15 NH were enrolled and contributed usable samples (Supplementary Data 1). Two skin sites (inguinal crease and axilla) were sampled at 2–3 visits per resident between 12 h and 48 h after bathing (Fig. 1A). These intertriginous sites provide robust and reproducible colonization signal and are well-suited for strain-resolved metagenomic investigation while reducing participant burden. In total, we analyzed 223 samples, including 207 skin swabs, 6 sample controls (air swabs at sampling site), 6 negative controls (reagent-only controls), and 4 positive controls (Mock Community standard ZymoD6306, ZymoBIOMICS, Zymo Research, California, USA). The study was approved by the Institutional Review Board at the University of California, Irvine, and informed consent was obtained from all participants or their representatives. Culture-based microbiology All 207 skin swabs and the 6 air swabs were plated on CHROMagar™ ESBL, CHROMagar™ VRE and CHROMagar™ MRSA to selectively isolate ESBL-producing Enterobacterales, vancomycin-resistant Enterococcus (VRE) and methicillin-Resistant Staphylococcus aureus (MRSA). Colonies were sub-cultured, and DNA extraction was performed using the QIAmp 96 DNA QIAcube HT kit or the Omega Bio-Tek Universal Pathogen Core Kit with modifications. Library preparation used Nextera XT kits, and sequencing (of 23 isolates excluding MRSA isolates) was performed on Ilumina NovaSeq X+ (Supplementary Data 2). Sequence reads were assembled with SPAdes25 v3.14.1, and quality assessed using QUAST26 v5.2.0 and CheckM227 v1.0.2. Isolate genomes were analyzed for AMR genes with Abricate (using ResFinder v4.0 database28). Skin shotgun metagenomic sequencing and analysis DNA from 207 skin swabs, 6 negative air swabs, 6 reagent-only controls, and 4 samples of the mock microbial community standard was extracted as previously described10. Nextera libraries were prepared and shotgun metagenomic sequencing was performed as paired-end reads (151 bp) on Illumina NovaSeq6000. Shotgun metagenomic reads were trimmed of adapters and low-quality bases with Cutadapt v4.0, and mapped to the T2T-CHM13v2.0 human reference genome using Bowtie2 v2.4.5 to remove host reads29. Quality metrics were assessed with FastQC and MultiQC. MAGs were reconstructed using metaSPAdes v3.15.0 and MetaWRAP v1.2.230, integrating MetaBAT2, MaxBin2, and CONCOCT for binning. MAGs were retained if completeness ≥50% and contamination ≤5% (CheckM227 v1.0.2) as previously defined10. However, for strain sharing and persistence analyses near-complete MAGs were used (Median completeness of 97.45% and a median contamination of 0.45%). Chimeric assemblies were identified with GUNC v1.0.531 and removed. Taxonomic assignment used GTDB-Tk (release 20732). Eukaryotic MAGs were screened with EukCC v2.1.133 and classified using Mash v2.334 and Mummer v3.2335 against fungal genome references (after downloading all fungal genomes in RefSeq on September 25th 2024). MAG assembly statistics, gene content metrics, and quality scores are provided in Supplementary Data 3. Taxonomic composition of skin metagenomic samples was determined using Kraken2 v2.1.236 with a custom database of the standard Refseq database (Refseq-Release226) enriched with dereplicated MAGs from this study. Abundance estimates were generated with Bracken v2.8.0 and imported into R (phyloseq v1.38). Antibiotic resistance genes (ARGs) were detected from the MAGs as well as from shotgun metagenomic reads by mapping with the resistance gene identifier (RGI) v5.2.1 integrated with the comprehensive antibiotic resistance database (CARD) v3.1.237. When ARGs were not captured within near-complete MAGs, read-based detections were interpreted conservatively and evaluated in the context of strain-resolved MAG recovery. For these cases genes were considered present with ≥40% allele coverage and ≥70% nucleotide identity. To minimize potential cross-species misassignment of plasmid-borne resistance genes, ARG signals were conservatively attributed only in samples containing strain-resolved MAGs and lacking evidence of plausible alternative Proteobacteria species as assessed by StrainGE38 and BLASTn. Methicillin resistance was inferred by detection of mecA in Staphylococcus epidermidis MAGs in the absence of S. aureus. Strain-level analyses Strain level analysis characterized MAGs and bacterial isolate genomes from this study, as well as dereplicated public genomes from RefSeq database for each species of interest downloaded on September 2nd, 2024. The genomic characterization of the ESBL Enterobacterales and VRE isolates recovered from NH residents and four comparative genomes from additional residents from the same facilities as part of the wider PROTECT trial5 are provided in Supplementary Data 2. We performed Multi-locus Sequence Typing (MLST) for lineage assignment with mlst v2.23.0, using default settings and published pubMLST schema (https://pubmlst.org/, MLST Achtman). Phylogenomic analyses based on single-copy marker genes were built with GToTree39 and core-genome SNP based phylogenetic analysis for maximum likelihood trees with RaxML40 following previously described procedures9. Pairwise average nucleotide identity (ANI) was computed using FastANI v1.34.141 to infer genomic similarity thresholds. Core-genome SNP profiling was performed with Snippy v4.4.1, with recombination masking using Gubbins v2.4.142. SNP distances were calculated with snp-dists v0.8.2. Read-based strain detection from metagenomic sequences was performed using StrainGE v1.3.338 with a database constructed with the combination of isolate genomes and dereplicated public genomes. A threshold of 30 SNPs was considered for phylogenetic relatedness based in the targeted species in NH settings from previous studies9. Statistics and reproducibility This study is based on metagenomic and isolate genome sequencing analyses of skin swabs collected from nursing home residents enrolled in the Project PROTECT trial5. Sample sizes were determined by sample availability from the parent longitudinal cohort and no statistical method was used to predetermine sample size in this microbiome study. No data were excluded from the analyses other than standard microbiome filtering steps, where taxa present at <0.1% relative abundance or in <5% of samples were removed to reduce sparsity and sequencing noise. The experiments were not randomized, as this was an observational genomic analysis nested within an existing clinical trial cohort. Investigators were not blinded to sample allocation during downstream genomic and microbiome analyses. All microbiome analyses were performed in R v4.3.2. Alpha and beta diversity metrics were computed from metagenomic profiles, and differences across body sites and time points were assessed by PERMANOVA. Pairwise SNP distances and phylogenetic relationships were reconstructed from whole-genome sequencing data and visualized using ggplot2 and iTOL43. Reported frequencies and percentages represent descriptive summaries of colonization prevalence and strain sharing patterns. Enrolled patient characteristics are shown in Supplementary Data 10. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Data Transparent Peer Review file Supplementary Information Description of Additional Supplementary Files Supplementary Data Transparent Peer Review file