Epidemiological investigations identify raw pet food and bovine exposure as main infection sources

Between July 2020 and April 2023, 20 human cases were included in a cluster, labelled as 2009STWGS-2QC-MP based on PNC surveillance data (Supplementary Data 1). These cases resided in 10 (56%) of the 18 administrative regions in the province of Quebec, with one to 3 cases per region, and a male-to-female gender ratio of 1.5. Ages ranged from 2 months to 91 years (median of 9 years), and 10 (50%) were 2 years old or younger, including 6 under one year. Eight (62%) of the 13 cases for which information was available were hospitalized for a length of time ranging from 3 to 31 days (median of 4.5 days), and none died. Data on exposures was available for 16 (80%) cases (results provided are non-mutually exclusive). All (100%) had contact with at least one animal: 12 (75%) with dog(s), 3 with cat(s), 4 with bovine(s) (3 with calf/calves and one with cows), one with other farm animals, and one with a reptile. Among the 14 dogs and cats in contact with human cases, 8 dogs were fed with a RMBD (6 with beef, 4 with other meat and 2 with an unknown type of meat); 3 were fed with pet treats, beef or pig ears, or beef chewing sticks. The 8 cases reporting no contact with pets fed with a RMBD stated the following exposures: 3 had direct or indirect contact with calves, 2 had contact with manure, one did gardening, 2 ate undercooked beef or veal meat, one ate at a restaurant, one drank water from an underground well, and one lived in a long-term health care facility (reporting contact with a dog with unknown diet).

Compared to the 2015 Foodbook report²⁴, the proportion of the following exposures among cases was statistically higher than among Quebec’s general population using binomial probability: contact with animals, their excrements, habitat or food (p = 0.001); contact with dogs (p = 0.0008); handled any raw pet food (store-bought or home-made; p < 0.00001); handled any treats derived from animal parts (p = 0.03); contact with bovine cattle (p = 0.0001); visited a farm or barn (p = 0.001).

Cases phylogenetically related to the cluster not only circulated in Quebec, but clinical cases also started appearing in other Canadian provinces (Fig. 1). The first 3 Quebec cases are shown during the summer of 2020, followed by a rise in cases in 2021 and again in 2022, then in an irregular pattern, suggesting they were sporadic. According to the genomic investigations described in the following sections, the emergence of new cases coincides with the acquisition of an IncHI2A plasmid between the summers of 2020 and 2021, which increased the strain’s MDR phenotype to XDR.

The 2009STWGS-2QC-MP cluster includes 20 Quebec (QC) clinical cases (navy), 23 QC animal isolates (blue), 2 QC raw meat isolates (from meat rendering plants; light blue), 13 clinical cases from outside Quebec (dark green), and 2 RMBD isolates collected from Ontario (ON-Home; light green). The sample collection dates were used when provided. Antimicrobial resistance profiles (multidrug-resistant [MDR] and extensively drug-resistant [XDR]) and the acquisition of the XDR IncHI2A plasmid are shown. Two MDR isolates within the XDR area (July 2021–July 2023) are indicated with an asterisk.

Animal health and food investigations indicate bovine as principal strain source

A total of 25 strains isolated from 21 animals: 16 veal calves, 3 dogs, 1 piglet and 1 moose, in addition to 4 RMBD samples were included in the 2009STWGS-2QC-MP cluster (Supplementary Data 1). Table 1 summarizes the associations between isolates by common household (QC-Home-A to D and ON-Home) or farm (QC-Farm-A to C). Seven veal farms for meat production, tested positive for this strain, including one with four different locations (QC-Farm-B). The 2 dog feces isolates were associated with two homes (QC-Home-B and QC-Home-C) who fed commercial RMBDs of raw beef and chicken meats to their dogs (asymptomatic), while the third dog isolate came from a deceased puppy fed a RMBD with unknown composition by a dog breeder from a kennel. Veal calf isolates were obtained from the tissues of 14 deceased animals and from the feces of two other calves. The piglet had an indirect link with bovine via the neighbouring farm site and via an employee whose spouse works on a cattle farm. These farm samples were mostly submitted in the context of mortalities in the herd that were unresponsive to antibiotic treatment. The deceased captive moose also had indirect contact with cattle because the zoo owner kept both animals. We collected two RMBD isolates from an Ontario home with an infant case (ON-Home; case isolate not included in this study). The 2 other RMBD isolates (raw chicken meat; raw ground chicken-beef-veal meat mix with bones) were collected from separate meat rendering plants producing raw pet food for dogs. Both plants transformed multiple types of meat, notably meat unfit for human consumption. Figure 2 illustrates the probable and confirmed transmission pathways between humans, animals and the environment.

Possible reservoirs, vehicles and routes of transmission of S. 4,[5],12:i:- are shown and based on case and animal health investigations and inclusion of bacterial isolates in the 2009STWGS-2QC-MP cluster. When the link between both isolate sources at each end of the arrows have been confirmed both epidemiologically and genomically (e.g., Table 1 associations), pathways are illustrated by full orange arrows. Possible pathways based on epidemiology alone are illustrated by grey dashed arrows. All studied isolates are portrayed, including the 2 raw meat-based diet (RMBD) isolates collected in Ontario. Only exposures reported by at least 2 human cases are illustrated. The number of studied isolates (yellow boxes) and exposure reports based on 16 investigated cases (blue boxes) are indicated next to their respective image.

Phylogenetic analysis confirms common strain between livestock, dogs, RMBD and humans

Figure 3 shows the phylogenetic tree of the 45 cluster-related strains alongside available isolate information, case exposures, and predicted plasmid and ARG presence. At least 5 high-quality single nucleotide variants (hqSNVs) separate the 4 MDR isolates, including the 3 isolates collected in 2020 before the rise of cases in 2021, from the other 41 isolates having acquired the IncHI2A plasmid (pAA738) responsible for causing the XDR phenotype. These 41 isolates form an XDR phylogroup with 99% branch support and display 0-13 hqSNV differences. Five hqSNVs differentiated the 2 RMBD isolates obtained from separate meat rendering plants. Another phylogroup with 91% support revealed close genetic relationships ( ≤ 3 hqSNVs) between the QC-Home-B dog and human case and the QC-Home-C human case. Both homes fed RMBDs to their dogs. These close phylogenetic relationships support the notion of a shared contamination source between livestock, dogs and humans.

The tree was generated based on 57 core high-quality single nucleotide variants (hqSNVs) detected among the 45 cluster-related isolates when compared to the chromosome of our PNCS017413 strain used as the reference genome (CP140755). The tree was rooted from the first collected isolate, PNCS017401. The aLRT SH-like internal branch support values above 90% are indicated by orange circles. Isolates referred to antimicrobial susceptibility testing (AST) are highlighted in pink. The predicted drug-resistance phenotype, sample source, patient age, home and farm associations, contact exposures, key plasmids, antimicrobial resistance genes (ARGs) and quinolone resistance-determining region (QRDR) mutations are also shown. The exposure “Animals + ” represents “Animals, their excrements, habitat or food”. Only the 3 confirmed ubiquitous plasmids (AA738, AB062 and AA147) are illustrated. The ARGs and quinolone resistance-determining region (QRDR) mutations are organized by genomic location based on the hybrid assembly of PNCS017412 and coloured by drug class. ARGs providing resistance to at least one of the 5 empirically recommended antibiotics (azithromycin, ciprofloxacin, ceftriaxone, trimethoprim/sulfamethoxazole and ampicillin) are identified with an asterisk.

Antimicrobial resistance genes and confirmation of XDR phenotype

Five isolates (highlighted in pink in Fig. 3) were selected for antimicrobial susceptibility testing (AST; Supplementary Table 1) based on their distinct predicted ARG profiles. Only the PNCS017412 (QC-Home-A baby) AST results are herein described. The familiar ampicillin, streptomycin, sulphamethoxazole and tetracycline (ASSuT) resistance locus provides chromosomally encoded resistance to 4 antibiotics: ampicillin (bla_TEM-1B; > 32 mg/L), streptomycin (aph(3”)-Ib and aph(6)-Id; not tested), sulfisoxazole (sul2; > 256 mg/L) and tetracycline (tet(B); > 32 mg/L). In addition, this strain contains 13 plasmid-encoded ARGs providing resistance to 13 antibiotics: ampicillin (bla_TEM-1B and bla_CTX-M-55; > 32 mg/L), azithromycin (mph(A); 64 mg/L), ceftriaxone (bla_CTX-M-55; > 64 mg/L), chloramphenicol (floR; > 32 mg/L), ciprofloxacin (qnrS1; 1 mg/L), gentamicin (aac(3)-IId; 32 mg/L), kanamycin (aph(3’)-Ia; not tested), lincomycin (lnu(F); not tested), rifampicin (ARR-2; not tested), streptomycin (aadA22; not tested), sulfisoxazole (sul3; > 256 mg/L), tetracycline (tet(A); > 32 mg/L) and trimethoprim-sulfamethoxazole (dfrA14; > 4 mg/L). In total, 18 ARGs conferring predicted resistance phenotypes to 13 antibiotics from 9 antimicrobial classes were detected, including all commonly recommended empiric and alternative antibiotics: azithromycin, ceftriaxone, ciprofloxacin, trimethoprim-sulfamethoxazole, and ampicillin. These AST results therefore infer the XDR phenotype for 41 isolates with common critical ARGs. A single calf isolate (PNCS017508) collected in June 2022 procured a mutation in a quinolone resistance-determining region: gyrA D87G. An MIC increase for ciprofloxacin (2 mg/L) and nalidixic acid ( > 32 mg/L) was observed, thus shifting the phenotype of nalidixic acid from susceptible to resistant for this isolate.

Plasmid profiling and acquired fitness discovery

Our 45 cluster-related isolates had a median of 4 (range: 3–8) in silico plasmid predictions (i.e., MOB-cluster IDs; Supplementary Fig. 1; Supplementary Data 4). Four primary MOB-cluster IDs: AA738, AB062, AA147 and AD672, were detected in over 91% of the isolates studied. The complete hybrid assembly of PNCS017412 (QC-Home-A baby) confirmed 3 of these plasmids: pAA738, pAB062 and pAA147, as AD672 was identified as a prophage. Figure 4 illustrates and contrasts these 3 key plasmids (pAA738, pAB062 and pAA147) with other similar sequences described below.

a The confirmed pAA738, pAB062 and pAA147 plasmids are described by their MOB-cluster, predicted mobility, incompatibility (Inc) groups, sequence length, depth relative to chromosome (i.e., copy number), antimicrobial resistance genes (ARGs), other genes of interest, mobile genetic elements and predicted host range by MOB-suite. b The three complete and closed plasmids from the hybrid assembly of the PNCS017412 QC-Home-A isolate are shown. The pAA738, the extensive drug resistance-carrying IncHI2A/IncN plasmid is compared with pVPS18EC0801-1 (CP063719), pL1 (CP071712), PJM1-chr (CP045038), pR18.0877_278k (CP037959), p16-6773.1 (CP039861), pST34VN3 (OQ658821), pVNB151 (NZ_LT795115) and p01 (OU015338). The pAB062, encoding Protein L and an ATP-binding protein (putative abortive infection AbiEii toxin), is compared with p16-6773.3 (CP039863), pST35-6035.1 C (CP051283), pRHB09-C20_4 (CP057935) and pMTY2805_NA4 (AP026536). The pAA147, encoding aph(3’)-Ia (kanamycin resistance), is compared with p16-6773.4 (CP039864), pF18S040-2 (CP82392), pN18S2154-2 (CP082528), pMRSN346355_5.3 (CP018126) and pPK8241-5kb (CP080144). Abbreviations: IS insertion sequence, TA toxin-antitoxin, RM restriction-modification, T4SS type IV secretion system, ncRNA non-coding RNA, tRNA transfer RNA.

Plasmid 1: XDR and possible reduced biocide susceptibility

The pAA738 (CP140756) is a 258,912 bp conjugative IncHI2 plasmid found among Proteobacteria and is identified as plasmid sequence type pST2. The ARG content of this plasmid varied across the 41 isolates (see Fig. 3) but is autonomously responsible for the XDR phenotype observed. The plasmid contains typical plasmid-borne genes involved in plasmid maintenance (5 complete or partial toxin-antitoxin systems), replication (repB, repE and repHI2) and transfer (3 type IV secretion system gene clusters). BLASTn analysis showed that this plasmid has been frequently detected. We compared pAA738 to eight other IncHI2 plasmids (245-278 kbp; 6 pST2 and 2 pST3) representative of the major gene content variances detected by BLASTn (Supplementary Fig. 2). The pVPS18EC0801-1 recovered from Escherichia coli in retail ground veal in 2018 USA⁴³ showed the most sequence similarity to our pAA738. We included two plasmids recovered in clinical Canadian isolates of E. coli and S. 4,[5],12:i:- from 2016-2017: pL1 and p16-6773.1, respectively. We also included pR18.0877_278k from a clinical outbreak isolate of Salmonella Goldcoast from 2018 Taiwan¹⁷, and the plasmids described from the thoroughly investigated ST34 lineages of Southeast Asia (pST34VN3 [pST2]; pVNB151 [pST3])¹³ and Australia (p01 [pST3])¹⁶. In addition to the innately diverse ARG and insertion sequence element regions of our plasmid, we also observed gene content variances owing to efflux pump activators (ramAp), toxin-antitoxin systems (agrB/dqlB; Hok family toxin; vbhA), replication initiation genes (repE), metabolite exporters (eamA), restriction-modification systems (dcm), peptidoglycan synthesis genes (ftsI) and copper resistance determinants (copG). Some of these genes (ramAp, eamA, ftsI, copG) in addition to others more preserved hint at reduced biocide susceptibility functions by encoding reductases, efflux pump regulators (marR), tellurium metal resistance (terX, terW, stiP and terZABCDE), DNA repair (umuDC), disulfide bond repair (Dsb family proteins), acidification-induced genes (inaA) and protective DNA-binding proteins (hns, hha). These additional functions may provide increased fitness and promote a stable maintenance of this IncHI2 plasmid in bacterial populations at the farm or in meat processing environments via co-selection despite an absence of continuous antimicrobial pressure, which was the case for the RMBD products containing this XDR strain.

Plasmid 2: putative phage defense

pAB062 (CP140757) is a 6306 bp plasmid predicted to be mobilizable and observed among Enterobacteriaceae. This plasmid encodes Protein L and an ATP-binding protein as a putative abiEii toxin associated with innate phage immunity through abortive infection (Abi)^44,45, in addition to 3 hypothetical proteins. No incompatibility groups or insertion sequence elements were detected. Only 8 plasmids showed at least 65% coverage and 97% identity to pAB062 (Supplementary Fig. 3), suggesting a certain scarcity. We compared pAB062 with 4 plasmids (6.0-6.3 kbp): S. 4,[5],12:i:- p16-6773.3 from a human in 2016 Ontario, S. Typhimurium pST35-6035.1 C from a wild bird in British-Columbia, E. coli pRHB09-C20_4 from cattle in 2017 United Kingdom (UK) and E. coli pMTY2805_NA4 from 2006 Japan.

Plasmid 3: kanamycin resistance

The pAA147 (CP140758) is a 5310 bp plasmid predicted to be mobilizable, observed among Enterobacteriaceae, and encodes aph(3’)-Ia (kanamycin resistance) next to a Tn3 transposon fragment with no incompatibility groups. BLASTn resolved 16 sequences containing aph(3’)-Ia with over 75% coverage and 99% identity (Supplementary Fig. 3). We compared pAA147 to 5 plasmids (5.3-14.5 kbp): S. 4,[5],12:i:- p16-6773.4 from a 2016 clinical isolate from Ontario, S. 4,[5],12:i:- pF18S040-2 from swine in 2017 USA, S. Hadar pN18S2154-2 from retail turkey in 2008 USA, E. coli pMRSN346355_5.3 from a clinical isolate in the USA and E. coli pPK8241-5kb from chicken in 2020 Pakistan.

Local retrospective analysis illustrates plasmid clonality in ST34 while identifying IncHI2A plasmids in all XDR cases

We analyzed the phylogenetic distribution of common primary MOB-cluster IDs to our 2009STWGS-2QC-MP cluster among all other S. 4,[5],12:i:- and S. Typhimurium Quebec PNC surveillance isolates (n = 1202; Fig. 5) sequenced since June 2017. We also added 7 bla_CTX-M-55-positive animal isolates from the Canadian Integrated Program for Antimicrobial Resistance Surveillance to the analysis. Counts and percentages below exclude the 45 2009STWGS-2QC-MP cluster-related isolates.

The S. 4,[5],12:i:- clades and isolates are indicated by pink partitioning. The isolates containing a predicted XDR phenotype (a) or the same key primary MOB-clusters (b AA738/739, (c) AB062 and (d) AA147) as the 2009STWGS-2QC-MP cluster are illustrated according to the legend. Most isolates are ST19, but other STs with a common XDR phenotype or MOB-cluster IDs to the cluster are indicated by labels.

We found the AA738/739 primary MOB-cluster IDs, responsible for the XDR phenotype of our cluster, in 67 isolates (5.6%): ST34 (n = 41; 11.8%), ST19 (n = 21; 2.5%), ST313 (n = 2; 66.7%) and ST36 (n = 1; 4.5%) in either sporadic isolates or small clusters. A total of 62 (5.1%) isolates, including all 11 XDR isolates outside of the 2009STWGS-2QC-MP cluster (9 animal isolates 2018-2023 and 2 human cases 2019-2021), exhibit an AA738:AJ047 or AA739:AJ047 MOB-cluster ID prediction with IncHI2A. However, animals may harbor this plasmid more frequently than estimated here, as non-human isolates are rarely targeted in PNC surveillance unless an investigation was specifically requested by the province, and only bla_CTX-M-55-positive non-human isolates from the Canadian Integrated Program for Antimicrobial Resistance Surveillance were included.

Regarding the 2 remaining plasmids, we detected the primary MOB-cluster ID AB062 (encoding Protein L in our cluster), in 114 (9.3%) isolates: ST34 (n = 108; 31.0%), ST19 (n = 4; 0.5%), while the AA147 primary MOB-cluster ID (carrying aph(3’)-Ia in our cluster) showed a similar distribution, where 115 (9.5%) were found in ST34 (n = 95; 27.3%), ST19 (n = 17; 2.0%) and ST302 (n = 1; 100%). Among 142 AB062 and/or AA146-carrying isolates, 87 (61.3%) carried both plasmids, indicating a certain plasmid clonality, which was observed in ST34 since the beginning of our genomic surveillance.

Global phylogenetic and ARG analysis demonstrates genetic link to USA porcine and rarity of our XDR strain

All 45 isolates were submitted and included in the NCBI Pathogen Detection system, which consolidates public genomic data from various global surveillance and research initiatives for comprehensive analysis and monitoring of bacterial pathogens. Our isolates formed a phylogroup in the SNP cluster PDS000159488.16 encompassing a total of 13,571 sequences from 2001-2023 collected by 36 different countries at the time of analysis with up to 89 SNPs separating all sequences (average: 51 SNPs).

Within 10 SNPs of our 45 isolates, only 3 other isolates from outside Canada were detected: 2 clinical isolates from the USA (PNUSAS355579 and PNUSAS362933) collected in 2023, and 1 raw pork product from the USA (FSIS11922734) collected in 2019. While the 2 clinical USA isolates only shared 12 of the 18 ARGs detected in our XDR cluster, thus lacking resistance determinants for gentamicin, lincomycin and chloramphenicol, they were still considered XDR.

Within 25 SNPs of our 45 isolates, 59 clinical isolates from Canada (2008–2018), 68 from the USA (2014–2022) and 7 from the UK (2012–2016) are added. One Canadian poultry isolate (2013), 4 Canadian pig isolates (2018-2022), 10 USA pork products (2016–2023), and 2 USA pig isolates (2020) were also found.

Out of the 13,571 sequences included in this SNP cluster, only 296 (2.2%) contain bla_CTX-M-55, of which 30 (0.2%) also contain mph(A) and qnrS1 after filtering our study isolates and the Vet-LIRN duplicates (see methods). Of these 30 isolates, none have the same 18-ARG profile as our emerging XDR cluster under study, which demonstrates the unique and rare extent of our strain.

Global retrospective plasmid subtype comparisons indicate shared plasmid pool among food-producing animals

Robertson et al. (2023) used MOB-suite, a bioinformatic tool allowing the identification and typing of plasmids in draft assemblies, on 150,767 publicly available Salmonella sequences covering 1204 distinct serovars uploaded to the NCBI database up until mid-October 2019⁴². Here, we recycled their results to determine the serotype and source distribution of the 3 key plasmids detected in our emerging cluster (Supplementary Fig. 4). Detailed observations can be found in Supplementary Note 2.

To summarize, serovars S. 4,[5],12:i:- and S. Typhimurium were among the top 3 serovars in which these 3 key plasmids (based on secondary MOB-cluster ID) were historically detected. Other than humans, they were mainly found among chicken, turkey and porcine sources. The secondary MOB-cluster IDs seem to be able to differentiate between different serotype and source associations, which may coincide with niched plasmid-host adaptations reflected in the plasmid sequence. For exemple, the increased presence of plasmid AA738:AJ047 detected in serotypes S. 4,[5],12:i:- and S. Typhimurium compared to S. Kentucky and S. Heidelberg, and in animal feed/pet food compared to chicken sources for other AA738 secondary MOB-cluster IDs is of interest. This may indicate a possible environmental reservoir of XDR S. 4,[5],12:i:- supporting the notion that this plasmid provides an increased fitness advantage beyond AMR, thus strengthening its maintenance in bacterial cells via other environmental pressures during food processing.