Detailed materials and methods are provided in the Appendix. In brief, we performed whole-genome sequencing (WGS) on clinical (n = 1,046) and nonclinical (n = 1,325) L. monocytogenes isolates collected across NY, as previously described (8). We deduplicated clinical isolates from the same patient or mother–infant pairs, resulting in 1,006 clinical isolates. Likewise, we deduplicated multiple nonclinical isolates from the same sampling event, resulting in 1,089 nonclinical isolates.

We used whole-genome SNP distances to assess the genetic relatedness among isolates, as previously described (8). We used genetic relatedness of clinical and nonclinical isolates to define 3 types of isolate grouping. When only clinical isolates were involved, we used the term clustered, and when clinical and nonclinical isolates were involved, we used the term linked. Hence, the term linked was used to describe genomic similarity between clinical and nonclinical isolates on the basis of SNP thresholds, rather than confirmed epidemiologic associations. Therefore, NY clinical isolates were considered clustered if clustered with >1 other NY clinical isolate by <20 SNPs (9–11), NY nonclinical isolates were considered cluster-linked if linked by <50 SNPs with >1 clustered NY clinical isolate (12), and NY clinical and nonclinical isolates were considered sporadic-linked if a NY clinical isolate was not clustered with any other NY clinical isolate by <20 SNPs but was linked with >1 NY non-clinical isolate by <50 SNPs. The terms clustered, cluster-linked, and sporadic-linked refer only to genetic relatedness on the basis of SNP thresholds, without interpretation of epidemiologic data. Isolates not meeting those criteria were considered unclustered NY clinical isolates and unlinked NY nonclinical isolates.

Our analysis of L. monocytogenes isolates in NY during 2000–2021 reveals clear differences in CC distributions between clinical and nonclinical isolates. Six CCs were significantly overrepresented among clinical isolates. Of those, 2 CCs (CC1 and CC4) have previously been identified as hypervirulent CCs on the basis of experimental evidence showing that they exhibit enhanced gut colonization, intestinal invasion, and infection of internal organs in a humanized mouse model (4). CC1, a well-established hypervirulent CC associated with invasive listeriosis (14,15), was the most frequently isolated CC among clinical isolates in our study, similar to a previous analysis of NY clinical isolates from 2000–2020 (16). The other CCs significantly overrepresented among clinical isolates in our study (i.e., CC217, CC388, CC389, and CC554) have not been phenotypically classified in terms of their virulence potential but might represent candidates for additional hypervirulent clones, as supported by studies suggesting that the virulence of L. monocytogenes CCs is positively associated with their epidemiologic association with human clinical cases (3). CC6 was the second most frequently isolated CC among clinical isolates but was not among the CCs significantly overrepresented among clinical isolates. That CC has previously been classified as hypervirulent and has been implicated in multiple large listeriosis outbreaks, including outbreaks linked to turkey deli meats, hot dogs, and French-style cheese (17). Previous studies in France and the Netherlands also identified CC1, CC2, CC4, and CC6 as frequently isolated CCs among clinical isolates (3,18). Overall, our data support that certain CCs are more likely to cause human disease.

We also identified 6 CCs (CC5, CC9, CC121, CC199, CC288, and CC321) that were significantly overrepresented among nonclinical isolates. CC9, CC121, CC199, and CC321 were the CCs with the highest proportion of isolates carrying inlA PMSC, a frequent hallmark of hypovirulent L. monocytogenes strains (13,19,20). CC9 and CC121 have previously been described as hypovirulent and environmentally adapted (4); those CCs were also reported to show increased tolerance to benzalkonium chloride, a commonly used sanitizer (4). Two CCs (CC5 and CC288) were overrepresented among nonclinical isolates but exhibited low frequencies of inlA PMSCs. Although CC5 was overrepresented among nonclinical isolates, it was the third most common CC among clinical isolates, suggesting that CC5 isolates are unlikely to be hypovirulent. The observed frequency of CCs among nonclinical isolates differs from observations in Europe. For example, CC121 was the most isolated CC among food and food processing environment–related isolates in France and Norway (3,22), as was CC9 in the Netherlands (18); CC121 and CC9 were the ninth and fifth most common CCs among nonclinical isolates characterized in this study. In contrast, CC5 was the most common CC among nonclinical isolates, but that CC was infrequently found among clinical and nonclinical isolates in France and the Netherlands (3,18). That discrepancy suggests a geographic difference in the distribution of this CC, which could be driven by possible differences in production systems, food preferences, or regional dynamics in L. monocytogenes transmission between NY and those countries in Europe. Despite likely regional differences in prevalence of different CCs, our data further support frequent global occurrence of certain clinically associated CCs (i.e., CC1 and CC4), whereas the occurrence of nonclinically associated CCs seems to show larger regional differences.

Overall, we found 7 statistically significant associations between CCs and food categories. For example, CC9 was significantly overrepresented among processed meat isolates and CC121 was significantly overrepresented among processed fish isolates. Those findings are consistent with previous reports indicating frequent recovery of those CCs from ready-to-eat meat and fish products and infrequent occurrence in dairy products (4,22–24). In addition, CC5 showed overrepresentation among raw meat isolates compared with processed meat isolates. That finding might suggest that different CCs could be introduced at distinct points along the food production chain; some CCs might be more closely linked to primary production or raw materials and others could be more likely to persist in food processing facilities. For the other 5 CCs associated with specific food categories, associations were based on smaller numbers (<10 isolates for the associated food category) and represented associations that had not been previously reported. Thus, those associations are possibly driven by specific sources (rather than general associations with a product type). Overall, our findings support that some CCs might be associated with certain food categories or with raw versus processed foods in a given category, which highlights opportunities to use WGS data to improve outbreak investigations and enable more rapid identification of a likely food source. However, although some CCs showed overrepresentation within specific food categories (e.g., CC9 with processed meat), others (e.g., CC321) were observed across multiple food sources with no overrepresentation in specific food categories. That finding suggests that certain CCs might not be specific to a single reservoir, and their presence should be interpreted cautiously in the context of source attribution.

Overall, inlA PMSCs were significantly overrepresented among isolates from processed meat and overrepresented, but they were not significantly overrepresented among processed fish isolates. Although our findings are consistent with previous studies that have reported frequent occurrence of inlA PMSC in isolates from processing plant environments and ready-to-eat foods (13,19,21,25), our analysis further highlights significant differences between processed and raw food isolates in the frequency of inlA PMSCs. Of note, a previous study in France also reported that hypovirulent CCs were associated with meat products in general (without clear indications on raw vs. processed sources) and proposed that hypovirulent CCs might be adapted to the processing environment, whereas hypervirulent isolates might be adapted to hosts (4). In contrast, we found that intact inlA were overrepresented among dairy isolates, which is also consistent with previous studies, such as a study in France that reported hypervirulent CCs to be strongly associated with dairy products (13). Finally, we found that intact inlA was borderline overrepresented among produce isolates (14 of 15 produce isolates showed intact inlA). That pattern is consistent with produce- and mushroom-focused surveys in the United States (5,26), United Kingdom (27), and China (28), which consistently reported low frequencies of inlA PMSCs in produce-associated isolates (5). In addition, we also found that sanitizer tolerance genes were significantly overrepresented in processed fish isolates and underrepresented in raw dairy isolates, consistent with a previous study that sanitizer tolerance genes were frequent among processed fish isolates (5). Overall, our findings suggest that food categories might differ in their likelihood of carrying virulence-attenuated L. monocytogenes and indicate that using WGS data to identify genetic markers (e.g., inlA PMSCs, sanitizer tolerance genes) could help with source attribution of raw versus finished products for some specific food categories.

We hypothesized that food category and presence of inlA PMSC were associated with the time between a food isolate collection and subsequent detection of a genetically related clinical isolate. Of note, we found that food category, but not the presence of inlA PMSC, was a significant predictor of that time; differences in time interval were observed across food categories. Specifically, isolates from produce had a shorter time for subsequent detection of a genetically related clinical isolate than did isolates from all other food categories.

Several factors likely contribute to the early detection of clinical isolates that are genetically related to produce isolates. Fresh produce typically has a shorter shelf life than other ready-to-eat products (e.g., cheese), possibly creating a narrow window in which contamination can directly lead to human exposure and illness. Contaminated produce typically would be consumed <20 days after processing, whereas cheese could be consumed >6 months after processing. However, the observed difference is unlikely to be explained solely by the short shelf life of fresh produce, because the time scale observed extends well beyond typical produce shelf life. Differences in consumer storage practices across different food categories (e.g., persons would more likely freeze raw meat than fresh produce) might extend the exposure window for certain food categories. In addition, L. monocytogenes contamination might be more frequently driven by repeated reintroduction rather than long-term persistence in packing and fresh-cut facilities (29–31). Persistent L. monocytogenes strains have also been detected less frequently in produce packinghouses (31–33) than in other food-associated facilities, although persistence over multiple seasons in some produce packing houses has been reported (34). Further research is needed to elucidate the persistence of L. monocytogenes in produce environments (29), but our findings suggest somewhat unique L. monocytogenes transmission pathways for produce. Of note, our data also showed that the time between the first food isolate and last subsequent detection of genetically linked clinical isolate is shorter for produce than for other food categories. That finding suggests that produce-linked clusters and outbreaks might often cover relatively short time frames, although long-term multistate outbreaks linked to produce have been reported (35). Even extremely small outbreaks putatively linked to produce might need to be further prioritized for outbreak investigations.

The first limitation of our study is that our dataset only included isolates obtained in NY during 2000–2021. Therefore, the proportion of unclustered and unlinked isolates might be overestimated because some clinical and nonclinical isolates could be clustered or linked with isolates from outside NY. Second, the SNP thresholds used to define clustering and linkage (i.e., <20 SNPs for clinical–clinical isolates and <50 SNPs for clinical–nonclinical isolates) are supported by previous studies as useful initial cut-offs for screening of potential clusters and sources but might need adjustment based on available epidemiologic data, particularly during outbreak investigations. Consequently, genetic relatedness inferred using those thresholds should be interpreted with caution. Third, genetic similarity alone does not establish epidemiologic linkage. Isolates with low SNP distances might reflect persistence in food production environments or widespread dissemination rather than direct transmission events. This study did not incorporate epidemiologic data (e.g., exposure histories), which limits the ability to confirm specific food sources for clinical cases. Similarly, a limitation of the cluster analysis is that some observed intervals between food isolate collection and subsequent detection of a genetically related clinical isolate were long. Those long intervals reduce confidence in interpreting the food as a continuous source of human exposure. Instead, such associations might reflect repeated reintroduction or cross-contamination across different reservoirs or food commodities. That interpretation is further supported by the observation that some SNP clusters included isolates from multiple food categories (Table 5). Therefore, the time intervals should be interpreted as reflecting temporal patterns of genomic relatedness rather than direct evidence of long-term source attribution to a single reservoir or commodity. Fourth, the dataset might be subject to biases, including uneven representation across food categories and time periods. In particular, the dataset includes isolates collected through targeted surveillance and research activities, which might result in overrepresentation of certain food categories or production environments. In addition, although thought to be rare, underdiagnosis or underreporting of listeriosis cases might lead to incomplete capture of clinical isolates. Finally, this study relied solely on SNP-based analyses. Incorporating approaches such as core-genome/whole-genome MLST could further strengthen the robustness of the findings, which highlights the need for more comprehensive integration of clinical and nonclinical isolates into unified databases to enable consistent genomic analyses, clustering, and outbreak investigations. In the future, advances in artificial intelligence and machine learning might further enhance the interpretation of surveillance datasets by identifying potential source attribution that are not easily analyzed through conventional approaches.