Public health implications of Yersinia enterocolitica investigation: an ecological modeling and molecular epidemiology study

Background Yersinia enterocolitica has been sporadically recovered from animals, foods, and human clinical samples in various regions of Ningxia, China. However, the ecological and molecular characteristics of Y. enterocolitica, as well as public health concerns about infection in the Ningxia Hui Autonomous Region, remain unclear. This study aims to analyze the ecological and molecular epidemiological characteristics of Y. enterocolitis in order to inform the public health intervention strategies for the contains of related diseases. Methods A total of 270 samples were collected for isolation [animals (n = 208), food (n = 49), and patients (n = 13)], then suspect colonies were isolated and identified by the API20E biochemical identification system, serological tests, biotyping tests, and 16S rRNA-PCR. Then, we used an ecological epidemiological approach combined with machine learning algorithms (general linear model, random forest model, and eXtreme Gradient Boosting) to explore the associations between ecological factors and the pathogenicity of Y. enterocolitis. Furthermore, average nucleotide identity (ANI) estimation, single nucleotide polymorphism (SNP), and core gene multilocus sequence typing (cgMLST) were applied to characterize the molecular profile of isolates based on whole genome sequencing. The statistical test used single-factor analysis, Chi-square tests, t-tests/ANOVA-tests, Wilcoxon rank-sum tests, and Kruskal–Wallis tests. Results A total of 270 isolates of Yersinia were identified from poultry and livestock (n = 191), food (n = 49), diarrhoea patients (n = 13), rats (n = 15), and hamsters (n = 2). The detection rates of samples from different hosts were statistically different (χ2 = 22.636, P < 0.001). According to the relatedness clustering results, 270 isolates were divided into 12 species, and Y. enterocolitica (n = 187) is a predominated species. Pathogenic isolates made up 52.4% (98/187), while non-pathogenic isolates made up 47.6% (89/187). Temperature and precipitation were strongly associated with the pathogenicity of the isolates (P < 0.001). The random forest (RF) prediction model showed the best performance. The prediction result shows a high risk of pathogenicity Y. enterocolitica was located in the northern, northwestern, and southern of the Ningxia Hui Autonomous Region. The Y. enterocolitica isolates were classified into 54 sequence types (STs) and 125 cgMLST types (CTs), with 4/O:3 being the dominant bioserotype in Ningxia. The dominant STs and dominant CTs of pathogenic isolates in Ningxia were ST429 and HC100_2571, respectively. Conclusions The data indicated geographical variations in the distribution of STs and CTs of Y. enterocolitica isolates in Ningxia. Our work offered the first evidence that the pathogenicity of isolates was directly related to fluctuations in temperature and precipitation of the environment. CgMLST typing strategies showed that the isolates were transmitted to the population via pigs and food. Therefore, strengthening health surveillance on pig farms in high-risk areas and focusing on testing food of pig origin are optional strategies to prevent disease outbreaks. Supplementary Information The online version contains supplementary material available at 10.1186/s40249-023-01063-6.


Background
Yersinia enterocolitica is a common zoonotic pathogen that is widely found in soil, water, animals, and various foods [1,2]. Y. enterocolitica is heterogeneous and consists of six biotypes: 1A, 1B, 2, 3, 4, and 5. Biotype 1B is regarded as highly pathogenic, and biotype 1A is regarded as being non-pathogenic in humans; other biotypes are considered to have low pathogenicity [3]. Enteric yersiniosis has been reported globally, with the most severe epidemic in Europe, where it was the fourth most common zoonotic disease in 2019 [4]. New Zealand has a high rate of yersiniosis notification, which is on the rise [5]. There were two outbreaks of Y. enterocolitica in China in the mid-1980s, resulting in over 500 infections [6]. Surveillance of the disease in China ceased for nearly 20 years, leading clinicians to rarely consider yersiniosis in the diagnosis of gastrointestinal diseases, and testers rarely being able to provide a basis for accurate diagnosis [7]. Y. enterocolitica was included in the national food contaminant surveillance network until 2016 [8]. The lack of mandatory government oversight in many countries may lead to an underestimation of the incidence of gastrointestinal disorders caused by Y. enterocolitica [9,10]. Such cases reflect a general lack of attention to the disease at a domestic and international level, which can lead to untimely treatment and misdiagnosis, resulting in chronic and prolonged infection. As such, research on Y. enterocolitica is essential and urgent.
The emergence and re-emergence of zoonotic diseases have been recognized as being driven by variability at the human-animal-environment interface [11]. However, this comprehensive investigation is still lacking in almost all zoonotic systems, as we lack empirically validated models of ecological interactions between humans, animal reservoirs, and key environmental drivers [12]. In order to clarify the complexity of zoonotic spills, field monitoring and modelling methods that link human-animal contact with disease dynamics within animal reservoirs are essential [13]. The surveillance survey results of diseases caused by Y. enterocolitica in China through consecutive years have indicated that its hosts include livestock, poultry, and rodents [14]. Pigs are an essential source of Y. enterocolitica infections [15]. Y. enterocolitica is transmitted via the faecal-oral route as well as through contact with animals (including farm animals, domestic pets, and wild animals) and contaminated food [14,16]. Previous studies have shown that the separation rate of domestic animals (pigs and dogs) was significantly and positively correlated with altitude, mean temperature, and precipitation, significantly and negatively correlated with altitude and mean temperature, and significantly and positively correlated with rainfall [14]. However, the integrated picture of host-environment interactions and the resulting spread and spillover of Y. enterocolitica is far from clear. Yersiniosis is an example of a health threat that affects humans, animals, and our shared environment, and controlling outbreaks of yersiniosis highlights the importance of a sustainable One Health approach. A crucial component of the One Health approach is identifying and monitoring the prevalence of zoonotic diseases in an area. However, information about the epidemiology of Y. enterocolitica in Ningxia is scarce. In this area, Y. enterocolitica has been sporadically recovered from animals, foods, and human clinical samples.
The primary objective of this study was to explore the ecological factors and molecular characteristics that influence the spread and pathogenicity of Y. enterocolitica. An exogenous association between ecological factors and the pathogenicity of isolates in Ningxia was demonstrated using a predictive modelling approach. The distribution and molecular characteristics within Y. enterocolitica populations in the region were explored through correlated genomic analyses. The public health implications of the host-human-environment relationship were explored in terms of the endogenous and exogenous causes of Yersinia transmission and provided a reference for the tracking and control of related zoonotic diseases. Health threats at the human-animal-environment interface are best addressed through an efficient and sustainable One Health approach.

Data sources
The 270 isolates in this experiment were obtained from five prefectures in Ningxia Hui Autonomous Region (Yinchuan, Shizuishan, Wuzhong, Guyuan, and Zhongwei) from 2007 to 2019. The isolates were obtained from animals, food, and patients. Animal hosts included pigs, sheep, rats, cattle, chickens, and hamsters. Pharyngeal swabs, anal swabs, intestinal contents, faeces and food samples were collected. Patient samples were obtained from faeces of diarrhoea patients collected at monitoring sites. Collection of pharyngeal swabs, anal swabs, and intestinal contents from healthy sheep and fattening pigs in slaughters. Pharyngeal swabs, anal swabs, and faeces collected from animals on farms.
Intestinal contents of rats and hamsters from hunted rats and hamsters. Foods were purchased at butchers, retail markets and supermarkets in the region. All samples were transported to the laboratory in separate sterile containers using a cooler. Isolates enrichment was performed in phosphate-buffered saline with sorbitol and bile salts (PSB) at 4 ℃ for 21 days.
Environmental and climatic data obtained from open-access data sources, including elevation, normalized difference vegetation index (NDVI), monthly mean temperature, and rainfall (Additional file 8: Table S1). The NDVI and elevation data were aligned over a grid with a spatial resolution of 5 × 5 km. Data from the survey samples were extracted from the corresponding geo-referenced locations. The data processing procedures described above were using software R 4.6.0 (Lucent Technologies, Jasmine Mountain, USA) with the 'raster' package.

Whole-genome sequencing and assembly
Genomic DNA was extracted using a Wizard ® Genomic DNA Purification Kit (Promega, United States) according to the manufacturer's protocol. DNA concentration, quality, and integrity were determined using a 5400 Fragment Analyzer System (Agilent, United States) and a NanoDrop 2000 Spectrophotometer (Thermo, United States). All 270 isolates were sequenced using Illumina Nova technology. Raw reads were quality filtered and assembled. The conditions for filtering were as follows: (1) reads containing more than 40% low-quality bases (mass value ≤ 20) were removed; (2) reads with N beyond a certain proportion (the default was 10%) were removed; (3) reads that overlapped with the adapter sequence by more than 15 bp with fewer than three mismatches between them were removed. Assemblies were performed using SOAP denovo (http:// soapd enovo2. sourc eforge. net/) [20], SPAdes (http:// bioinf. spbau. ru/ spades), and Abyss (https:// www. bcgsc. ca/ resou rces/ softw are/ abyss) software. Assembly results were integrated using CISA software (http:// sb. nhri. org. tw/ CISA/ en/ CISA). The preliminary assembly results were complemented with gap-closing software to filter reads with low sequencing depths (less than 0.35 of the average depth) to remove homolane contamination, resulting in the final assembly results. The reference genomes of 26 species of the Yersinia genus were shown in Additional file 8: Table S2. The ANI was estimated for each pair of isolates using the 270 Yersinia genomes, the 26 reference genomes (see above), and fastANI v1.33 (https:// github. com/ ParBL iSS/ FastA NI) [21]. The reference genes for the Yersinia were the same as those listed in the previous section.

Virulence profiles
The assembled genomes of the isolates in this study were inserted into the Virulence Factor Database (VFDB; MOH Key Laboratory of Systems Biology of Pathogen, Institute of Pathogen Biology, Beijing, http:// www. mgc. ac. cn) with default parameter selection.

Statistical analysis
Single-factor analysis was conducted to compare the differences of ecological environment factors between the host, biotype, serotype, and pathogenicity of Yersinia. Chi-square tests were used for categorical variables, while t-tests or ANOVA-tests were applied for normal distribution of continuous variables. If there exists non-normal distribution, Wilcoxon rank-sum tests and Kruskal-Wallis tests were used. The significance level was set to 0.05. The study also using the ecological environment factors to predict the risk of pathogenicity of Yersinia. We have split our data into training dataset (80%) and testing dataset (20%), and utilized general linear model (GLM), random forest model (RF), and XGBOOST (eXtreme Gradient Boosting) models to evaluate methodological improvements (Additional file 8: Table S4). The area under the receiver operating characteristic (ROC) curve was used to evaluate the model performance. The model was simulated by 'caret' package, and AUC (area under the curves) was performed by 'PRROC' package with statistical software R 4.6.0.

Epidemic profile of Yersinia during 2007-2019
A total of 9031 samples were monitored from 2007 to 2019, with the detection rate of Yersinia ranging from 0.9% to 7.6% ( Table 1). The highest detection rate was in 2014 (7.6%), eightfold higher than in 2013 (0.5%). The difference in positivity rates between years was statistically significant (χ 2 = 40.282, P < 0.001). No significant upward or downward trend in detection rates between years. Samples from different hosts were detected from 0.3% to 5.9%, with the highest detection rate of 5.9% for samples from poultry and livestock sources ( Table 2). The detection rates of samples from different hosts were statistically different (χ 2 = 22.636, P < 0.001).

Phylogenetic analysis
Total of 270 Yersinia genomes were evaluated according to the standard 95-96% ANI [27]. Twelve species were delineated using a 95% ANI cut-  (Fig. 2). The clustering analyses was consistent with the results of the ANI analysis and identical separation into 12 distinct species as determined by BAPS (Bayesian analysis of population structure) (Fig. 2). The relatedness clustering presented by the Y. enterocolitica ML tree showed a direct relationship with the biotype. Biotypes 1A, 1B, and 5 isolates formed discrete clusters, whereas biotypes 2, 3, and 4 isolates consisted of closely related but distinct lineages, confirmed by BAPS clustering. Pathogenic isolates from Ningxia were genetically more distant from the reference genome than isolates from other countries and regions (no biotype 1B isolates in Ningxia). Isolate LC20 from Zhejiang had the longest genetic distance with the reference genome compared to the other isolates in China. Biotype 1A isolates exhibited no geographical differences and a broader range of host species and serotypes. The hosts of biotypes 4 and 5 isolates in Ningxia were pigs and sheep, respectively, whereas the hosts in the other regions were humans and hares. Biotype 3 isolates were O:3 in Ningxia, compared to O:5,27 and O:6,30 in the other countries (Fig. 2).

Associations between ecological factors and pathogenicity
Temperature, precipitation, altitude, and NDVI were highly statistically significant to pathogenicity (P < 0.001) (Fig. 3a-d). The ambient temperature of the collection locations for pathogenic isolates was lower than that of the non-pathogenic isolates (12.70 ± 6.39 ℃ vs. 16.85 ± 6.05 ℃); the precipitation of the collection locations for the pathogenic isolates was lower than that of the non-pathogenic isolates (1.34 ± 1.29 mm vs. 2.54 ± 1.42 mm). The median elevation of the collection locations for pathogenic and non-pathogenic isolates was 1111.00 m and 1816.00 m, respectively; the median NDVI was 0.35 and 0.19, respectively. Also, except for the host of the isolates in precipitation factors, the ecological factors and pathogenicity were statistically highly significant with biotype, serotype of the isolates, and host of the isolates (Fig. 3, Additional file 1: Fig. S1, Additional file 2: Fig. S2, Additional file 3: Fig. S3). Then, we used three models to predict the pathogenic of Y. enterocolitica with ecological environment factors. The AUC of the training and test sets of all three models exceeded 0.9, particularly the AUC of RF, and XGBOOST, indicating good evaluation performance (Fig. 3e, f ). The contributions of ecological factors to the risk of Y. enterocolitica pathogenicity were shown in Additional file 4: Fig. S4, which indicated the prediction of Y. enterocolitica pathogenicity was the combination of all ecological factors. Finally, we used those performed models to evaluate the risk of Y. enterocolitica pathogenicity among our study regions (Additional file 5: Fig. S5). The results indicated that the highest risk area was concentrated in Yinchuan City and Shizuishan City in northern Ningxia, Zhongning City, and Zhongwei County in central Ningxia, and the southeastern part of Guyuan City and Haiyuan County in southern Ningxia. The predicted risk also varied between sources, with animals and humans suggesting a wider range of risk areas.

Virulence profiles
Y. enterocolitica isolates were annotated with 130 virulence genes in 5 categories: flagella, invasion, O-antigen, Yersiniabactin, and T3SS (Fig. 4). There were 53 flagella genes, most of which were che, flh, flg and fli genes. The O-antigen was represented by 27 genes, mostly mrk and wbc genes. The Yersiniabactin was made up of 9 genes, the majority of which were ybt gene. T3SS included 41 genes. The major ones were ysc, yop, icr, and syc genes. The distribution of virulence factors was closely related to the biotype. Most biotype 1A isolates lacked the virulence factors of invasion, O-antigen, Yersiniabactin, and T3SS. In contrast to pathogenic isolates of other biotypes, biotype 1B isolates were presented in Yersiniabactin.

ST and cgMLST typing
In 187 Y. enterocolitica isolates, 54 STs were detected using multilocus sequence typing. ST429 was the most common, accounting for 42.3% of all STs (79/187) (Fig. 5 and Additional file 8: Table S6). ST429 isolates included biotype 4 (n = 79) and biotype 3 (n = 2), and they were closely related to serotype O:3. ST3 (n = 9), ST278 (n = 6), ST178 (n = 5), ST637 (n = 5), ST640 (n = 4), ST643 (n = 4), and ST216 (n = 3) were biotype 1A isolates with sero-  Fig. S6). An NJ tree and an MST based on cgMLST analysis of 187 Y. enterocolitica isolates were constructed. These 1,553 cgMLST target genes were randomly distributed across the genome. Y. enterocolitica isolates were divided into 125 cgMLST types (CTs) (Additional file 8: Table S6 and Fig. 6). CgMLST analysis revealed the core genome diversity of isolates with the same ST from 0 to 84 allelic differences. The NJ tree of 187 Y. enterocolitica isolates from Ningxia indicated the names of two microclades of the HC1490 (Hierarchical clustering) cluster (Fig. 6). HC1490_10 and HC1490_2 were the primary phylogenetic branches. These two microclades of the HC1490 cluster were consistent with the results of the Y. enterocolitica ML tree. HC1490_2 was strongly associated with biotypes 3, 4, and 5 isolates. HC1490_10 was closely related to biotype 1A isolates. The 125 CTs present in the 187 isolates from the Ningxia region clustered to form 54 microclades of HC100. Of these, HC100_2571 was the principal microclades, and the isolates were all from this study. The NJ tree constructed from isolates from the Ningxia region and those publicly available in the database showed that several microclades of the HC100 cluster were significantly associated with serotypes, hosts, and countries. HC100_2571 isolates were all obtained from this study and were of bioserotype 4/O:3 and 3/O:3, hosts were pigs and humans. HC100_406 isolates were 4/O:3, mostly from patients in New Zealand. The serotype of HC100_397 isolates was O:5,27, with hosts of pigs, humans, and food. HC100_2 and HC100_111 isolates were 3/O:9 and 4/O:9, respectively, mainly from pigs and humans in the UK and New Zealand. HC100_4570 isolates were biotype 5, obtained from sheep in Ningxia. HC100_466 isolates were biotype 1B, which were isolated from patients. HC100_150 isolates from pigs and patients

Discussion
Ningxia Hui Autonomous Region is located in the upper reaches of the Yellow River in western China, where the largest Hui nationality in the country resides. Of the five cities in Ningxia, Guyuan and Wuzhong are Muslim-concentrated areas where livestock farms are not allowed to raise pigs, resulting in a small sample size in these two areas. Yinchuan and Zhongwei are areas with scattered Muslim populations, and 95.9% of the samples (259/270) in this study were collected from these two regions.
Yersinia presented a diversified epidemiological trend in Ningxia over a 12-year period, with a pattern of the diminished predominance of prevalent isolates and the coexistence of multiple subtypes. There is no significant upward or downward trend in detection rates between years. The detection rates of samples from different hosts were statistically different (χ 2 = 22.636, P < 0.001), as well as 72.1% of poultry and livestock samples originated from pigs, implying that pigs were the primary source of Yersinia in Ningxia. Y. enterocolitica is present in the oral cavity (especially the tonsils and submaxillary lymph nodes), intestine, and feces of pigs [28,29]. Y. enterocolitica was easily transmitted between pigs via the faecal-oral route and may also be spread through direct contact between humans and animals or the environment (water and soil) [30][31][32]. Therefore, people involved in swine farming and processing or those with livestock farms in the vicinity of their living area are at greater risk of Y. enterocolitica infection. Food (18.2%) was another major source of Yersinia infection in the Ningxia region, which is associated with the consumption of undercooked contaminated pork [29,33]. It is potential for Y. enterocolitica to contaminate carcasses and pork products during the slaughter process [34]. Cross-contamination of other foods is possible when handling and preparing raw pork [35]. Because the transmission of Y. enterocolitica involves multiple pathways, the health department should continuously monitor and strengthen epidemiological investigation and training; the animal health department should improve quarantine and hygiene supervision of pigs and regulate the operation of workers in the breeding industry; the environment department should improve the supervision of breeding areas and their surrounding environment; the food regulatory department should continuously focus on and strengthen the testing of food with pig origin; the public should be concerned about food safety and avoid eating raw or undercooked meat products.
Public health and epidemiology face increasing challenges in understanding how environmental trends affect zoonotic disease dynamics [36]. An effective approach to zoonotic disease prevention and control that integrates animal, human, and environmental ecological principles is required [37]. The distribution of Y. enterocolitica was influenced by a variety of factors. Correlation analyses revealed that altitude, mean temperature and precipitation were important factors influencing the epidemiological diversity of Y. enterocolitica in animals in China, and were strongly related to climatic and agricultural Fig. 6 The neighbor-joining (NJ) tree of Yersinia enterocolitica isolates based on cgMLST. The circle size was proportional to the number of isolates. Clusters generated using the hierarchical clustering method from EnteroBase and using a 100 cgMLST allele distance (HC100) were represented by circles. a Isolates from this study. b Isolates from this study and the public database characteristics [14]. It was evident that temperature and precipitation of the collection locations were closely related to the biotype, serotype, and pathogenicity of the isolates. The mean temperature and mean precipitation for pathogenic isolates were 12.70 °C and 1.34 mm, respectively, whereas the mean temperature and mean precipitation of non-pathogenic isolates were 16.85 °C and 2.54 mm, respectively, with the differences being highly statistically significant. This was because that Y. enterocolitica was a psychrotrophic bacteria that grew more slowly than most Gram-negative bacteria at room temperature and was unable to compete in mixed populations at high temperatures, but was able to grow, multiply and survive for long periods at low temperatures [38]. Y. enterocolitica must acclimate to the temperature of the human host before infection, therefore temperature-dependent chromosomal and plasmid-associated virulence determinants play an essential function [39]. In terms of precipitation, Ningxia has a typical continental semi-humid and semi-arid climate, with the rainy season lasting from June to September. Therefore we hypothesized that the significant differences in pathogenicity and precipitation were directly related to temperature. We used both unifactor and multifactor analysis in the prediction model, and it was obvious that the multifactor prediction result (temperature + precipitation + host) was the most optimal. Because neither humans nor microorganisms exist as independent individuals, various environment factors inevitably influence how organisms survive and contribute to the formation of specific ecological niches, necessitating a multifactorial approach. The random forest model outperformed the other three models in both the training and validation sets. This could benefit from the algorithmic advantages of the random forest, such as high performance and generalisation ability. Accuracy can be maintained even in the presence of missing data. Predictive models are powerful tools for public health and epidemiology, predicting the spread and spillover of pathogenic bacteria based on external environmental factors like temperature, precipitation, altitude, and vegetation distribution, as well as the characteristics of isolates like serotypes and biotypes, and monitoring and alerting on relevant zoonotic diseases.
Through robust and reproducible comparisons, WGS provides unprecedented resolution for characterising Y. enterocolitica [40], allowing it to be applied to public health management as well as real-time surveillance and outbreak investigation of foodborne diseases [41]. The ML tree of Y. enterocolitica isolates demonstrated that clustering was tightly linked to the biotype, which was consistent with previous findings [42,43]. Biotype 3, 4, and 5 isolates displayed tight clusters with short terminal branches compared with biotype 1A isolates. This clade represented the earliest diverging branches of Y. enterocolitica, supporting the hypothesis that pathogenic members evolved from non-pathogenic ancestors [42]. Genome analysis reveals different evolutionary processes in pathogens that potentially influence their lifestyles and contribute to specific ecological niches. The typing results revealed regional differences in the distribution of STs and CTs, biotype 1A isolates contained 46 STs (85.2%, 46/54) and 73 (58.4%, 73/125) CTs, indicating a high degree of heterogeneity. Notably, isolates from patients in this study were classified into the same microclades as isolates from food and pigs, implying that Y. enterocolitica had the potential to infect humans via pigs and food. According to Karlsson et al. [44], an outbreak of Y. enterocolitica 4/O:3 in fresh prewashed spinach occurred in Sweden in 2019. This was completely consistent with the dominant bioserotype of the isolates in this study. Some studies have shown that this biological serotype is associated with approximately 80% of human infections of Y. enterocolitica worldwide [45]. Despite the fact that no Yersinia outbreaks have been reported in Ningxia, it is worthwhile to raise the alarm.
The distribution of virulence genes was closely related to biotype in the current study. The presence of the virulence factor for iron uptake is attributed to the high pathogenicity of biotype 1B isolates. Iron supply and production of the siderophore transport system were significant factors in the infection of Y. enterocolitica [46]. Except for biotype 1B isolates, other pathogenic strains utilise exogenous iron carriers such as ferroflavin B and ferroflavin E [47]. Different pathways of iron uptake may explain differences in the virulence among biotypes [48]. In comparison to biotype 1A isolates, biotypes 2, 3, 4, and 5 isolates had virulence genes associated with the invasion and T3SS. T3SS is primarily composed of multiple Yersinia outer proteins (Yops), low calcium responsestimulating proteins (LCRS), and heat-stable enterocolitica toxin (Yst) [49]. Gram-negative pathogens inject virulence proteins into the cytoplasm of target eukaryotic cells via Type III secretion system (T3SS) [50]. Biotype 1A strains are regarded as non-pathogenic. However, increasing epidemiological and experimental evidence indicates that biotype 1A strains can cause gastrointestinal disorders [51] and reactive arthritis [52] as well as being implicated in foodborne and nosocomial outbreaks around the world [53], implying a role for the ystB gene in the pathogenic process [54]. In the present research, 88.8% (n = 79) of the Biotype 1A isolates possessed the ystB gene. Although there was no evidence that biotype 1A isolates were pathogenic in this study, researchers should remain vigilant for non-pathogenic Y. enterocolitica as well.
There were several limitations to our study. Firstly, the patient sample size was small, and it was collected in 2018 and 2019. The reason is that symptom of diarrhoea caused by Y. enterocolitica are usually mild and self-limiting. Patients rarely visit outpatient clinics. Secondly, the collection of samples from both food and patient sources lacked quantitative homogeneity and temporal consistency. Therefore, statistical data and predictive models may be affected. Thirdly, we used ecological epidemiological study to explore the relationship between macro-level environmental factors and pathogenicity of Y. enterocolitica. Ecological fallacies in ecological studies are the most significant drawback of ecological studies and are difficult to avoid. Finally, there is a scarcity of epidemiologically relevant information on isolates from diarrhoea patients. These constraints emphasized the significance of the One Health approach. Future work in the human, animal, environmental, and other relevant sectors should be followed by the Tripartite Zoonoses Guide to optimally support countries and regions in addressing zoonotic diseases through a multisectoral, One Health approach [55]. Concrete measures such as early warning and early detection through the Global Early Warning System for health threats and emerging risks at the human-animal-ecosystems interface (GLEWS +), linking surveillance systems, establishing standardised detection methods, sharing laboratory resources, improving laboratory capacity, expanding epidemiology training, and conducting joint sectoral risk assessments [56,57]. One Health offers a promising and valuable approach to successfully achieving optimal human and animal health while considering the protection of ecosystems and our shared environment.

Conclusions
The distribution of STs and CTs was related to biotype, geographical area, serotype, and host. The distribution of virulence factors was closely related to the biotype. The distribution of pathogenic isolates was correlated to the environment factors of temperature and precipitation, and machine learning models predicted that the epic of Y. enterocolitica areas in the Ningxia region was concentrated in the north, northwest, and south. Our research integrated the ecological modeling and molecular epidemiology study which lead to the public health of Y. enterocolitica. It also served as a guide for epidemiological and public health analyses. Zoonotic diseases and other public safety threats can be better addressed using this approach, which addresses health threats at the humananimal-environment interface in a synergistic manner.