Correlation of annual prevalence between cephems resistance and bla_CMY-2 in Salmonella enterica isolated from retail meat sources in the United States

Abstract

The objective of this study was to correlate the annual prevalence of cephems resistance and bla_CMY-2 in Salmonella enterica using surveillance data in the United States. Using datasets retrieved from the surveillance programs of the United States National Antimicrobial Resistance Monitoring System (NARMS) for Enteric Bacteria from 2002 to 2018, we performed Spearman’s correlation analysis to correlate the annual prevalence data. We observed a near-perfect positive correlation in the annual prevalence between cefoxitin (ρ = 0.97, P < 0.0001), ceftiofur (ρ = 0.96, P < 0.0001), ceftriaxone (ρ = 0.95, P < 0.0001) resistance and bla_CMY-2 in S. enterica recovered from chicken retail meat. Similarly, we observed a very high positive correlation in the annual prevalence between cefoxitin (ρ = 0.94, P < 0.0001), ceftiofur (ρ = 0.91, P < 0.0001), ceftriaxone (ρ = 0.82, P < 0.0001) resistance and bla_CMY-2 in S. enterica recovered from turkey retail meat. Using Autoregressive Integrated Moving Average (ARIMA) modeling, the forecasted annual prevalence of beta-lactam resistance for the years 2019–2021 was similar to the NARMS-reported data for these periods. Correlation between the annual prevalence of cephems resistance and bla_CMY-2 suggests either data can be used as a proxy for decision-making in retail meat surveillance programs.

Introduction

Antimicrobial resistance (AMR) poses a global health threat, driven by the use of antimicrobials in humans and food animals, which promotes the emergence, selection, and spread of resistant bacteria [1]. Food contaminated with antimicrobial-resistant bacteria can serve as a source of resistance for human pathogens, either through direct colonization by zoonotic bacteria or by the transfer of beta-lactamase genes and other mobile genetic elements. Extended-spectrum cephalosporins are critically important antimicrobials and are primarily mediated by extended-spectrum beta-lactamase (ESBL) and AmpC beta-lactamase [2]. Numerous studies have identified that beta-lactamase-producing Salmonella enterica occurs in chicken and turkey retail meats [3, 4]. In S. enterica, resistance to extended-spectrum cephalosporins is often associated with the plasmid-encoded bla_CMY-2. Animals, particularly poultry, have been linked to the spread of resistant Salmonella to humans, posing significant food safety concerns.

Poultry consumption in the United States has significantly risen over the past few decades with a per capita intake of 50.1 kg in 2019 [5]. Antimicrobials are commonly used in the poultry industry [6], which leads to the selection of resistant strains with resistance genes [7]. The United States has a National Antimicrobial Resistance Monitoring System (NARMS), where the United States Department of Agriculture (USDA) contributes data on AMR in retail meat, including both phenotypic and genotypic information (https://www.cdc.gov/narms/index.html). Such data availability provides an opportunity for secondary data usage that may support surveillance activities and improve decision-making. Therefore, our objective in this study was to explore the correlation between the annual prevalence of cephems resistance (ceftriaxone, cefoxitin, and ceftiofur) and the bla_CMY-2 in S. enterica isolated from retail meats collected by the national surveillance programs in the United States from 2002 to 2018.

Results

We observed a near-perfect positive correlation in annual prevalence between cefoxitin (ρ = 0.97, P < 0.0001), ceftiofur (ρ = 0.96, P < 0.0001), ceftriaxone (ρ = 0.95, P < 0.0001) resistance and bla_CMY-2 in S. enterica recovered from chicken retail meat. (Figs. 1A and 2A, 2B and 2C).

Fig. 1

Correlation of annual prevalence between cephems resistance and bla_CMY-2 in Salmonella enterica. A reflects the correlation of cephams resistance and bla_CMY-2 in chicken retail meat, and B reflects the correlation of cephams resistance and bla_CMY-2 in turkey retail meat. There was near-perfect correlation between cephems and bla_CMY-2 resistance in S. enterica isolated from retail meats. "R" denotes "Resistance"

Fig. 2

Correlation between cephems resistance and bla_CMY-2 in Salmonella enterica in chicken and turkey retail meat samples from the National Antimicrobial Resistance Monitoring System (NARMS) database. A shows the linear relationship between the annual prevalence of bla_CMY-2 and cefoxitin resistance, B shows the linear relationship between the annual prevalence of bla_CMY-2 and ceftiofur resistance, and C shows the linear relationship between the annual prevalence of bla_CMY-2 and ceftriaxone resistance in chicken retail meat, D shows the linear relationship between the annual prevalence of bla_CMY-2 and cefoxitin resistance, E shows the linear relationship between the annual prevalence of bla_CMY-2 and ceftiofur resistance, and F shows the linear relationship between the annual prevalence of bla_CMY-2 and ceftriaxone resistance in turkey retail meat. "R" denotes "Resistance"

For turkey retail meat, we similarly observed a very high positive correlation in annual prevalence between resistance to cefoxitin (ρ = 0.94, P < 0.0001), ceftiofur (ρ = 0.91, P < 0.0001), ceftriaxone (ρ = 0.82, P < 0.0001), and bla_CMY-2 in S. enterica (Figs. 1B and 2D, 2E and 2F).

We performed Autoregressive Integrated Moving Average (ARIMA) models to predict the annual proportions of both bla_CMY-2 and selected cephems resistance in S. enterica isolated from both chicken and turkey. For S. enterica from chicken, the ARIMA models (ARIMA (0, 1, 0)) for cefoxitin and bla_CMY-2 in S. enterica corresponds to random walk models with one level differencing to stationarized the time series of the annual proportions. Therefore, the forecasted proportion corresponds to the last year of the time series (2018), with annual proportions of cefoxitin and bla_CMY-2 being 3.87% (13/336) and 3.00% (10/333), respectively. Therefore, the predicted proportions for the subsequent years were 3.87% and 3.00% for the years 2019–2025 (Fig. 3A).

Fig. 3

Prediction of the annual proportions of bla_CMY-2 and cephems resistance in S. enterica. A shows the time series and forecasts of annual proportions of beta-lactam resistance in S. enterica in chicken retail meat based on Autoregressive Integrated Moving Average (ARIMA) models and B shows the time series and forecasts of annual proportions of beta-lactam resistance in S. enterica in turkey retail meat based on ARIMA models. "R" denotes "Resistance"

We identified a significant autoregressive term ARIMA (1, 0, 0) of order lag (1) year with non-zero mean in the forecasting of the annual proportions of ceftriaxone resistance. The predicted proportion of ceftriaxone-resistant S. enterica for the years 2019–2025 is 13.45%, 15.27%, 16.49%, 17.30%, 17.85%, 18.21%, and 18.45%. We identified a significant moving average component of order (1) year ARIMA (0, 0, 1) with non-zero mean in the forecasting of the annual proportions for ceftiofur resistance. The predicted proportion of ceftiofur-resistant S. enterica for the years 2019–2025 is 21.34%. This reflects the infinite sum of exponentially weighted past annual proportions of the ceftiofur resistance (Fig. 3A).

For S. enterica from turkey, we identified a similar significant autoregressive term of order lag (1) year with ARIMA (1, 0, 0) with non-zero mean in the forecasting of the annual proportions for bla_CMY-2, ceftiofur, cefoxitin, and ceftriaxone resistance. The predicted proportion of cefoxitin-resistant S. enterica for the years 2019–2025 is 3.09%, 4.50%, 5.40%, 5.97%, 6.33%, 6.57%, and 6.73%. The predicted proportion of ceftriaxone-resistant S. enterica for the years 2019–2025 is 6.77%, 7.08%, 7.35%, 7.34%, 7.39%, 7.42%, and 7.44%. The predicted proportion of ceftiofur-resistant S. enterica for the years 2019–2025 is 7.49%, 7.59%, 7.65%, 7.67%, 7.69%, 7.69%, and 7.70%. The predicted proportions of bla_CMY-2 producing S. enterica for the years 2019–2025 are 4.03%, 5.87%, 6.94%, 7.57%, 7.94%, 8.15%, and 8.28% (Fig. 3B).

Measures of accuracy of the ARIMA models predicting the annual proportions of beta-lactam resistant S. enterica in retail meats of chicken and turkey are presented in Table 1. For all the ARIMA models, the Mean Absolute Scaled Error (MASE) values (0.82–0.98) were < 1, indicating the ARIMA models were better than the naïve forecast. The Mean Absolute Percentage Error (MAPE) values varied and ranged from 22.97% to77.60%, indicating the ARIMA models were accurate 32.4–77.03% of the time.

Table 1 Predictive accuracy of Autoregressive Integrated Moving Average (ARIMA) models of annual proportions of beta-lactam resistant Salmonella enterica in retail meats of chicken and turkey

We compared the forecasted annual prevalence from the ARIMA models to the reported data by NARMS for the years 2019–2021 (Table 2). Bland–Altman test of the limit of agreement between ARIMA prediction and NARMS-reported data was not statistically significant different for the bla_CMY-2 producing S. enterica (P = 0.495), and ceftriaxone-resistant Salmonella (P = 0.769) for chicken, ceftriaxone-resistant Salmonella for turkey (P = 0.523), and cefoxitin-resistant Salmonella for turkey (P = 0.096). Therefore, the ARIMA predictions were not statistically different from the NARMS-reported data. However, the Bland–Altman test of the limit of agreement between ARIMA prediction and NARMS-reported data was statistically significant different for cefoxitin-resistant Salmonella in chicken (P = 0.031), and a marginal statistically significant difference was observed for the CMY-2 producing S. enterica in chicken (P = 0.052).

Table 2 Validation of ARIMA predicted proportions with actual reported proportions by NARMS for the years 2019–2021

Discussion

The findings from our study reveal significant positive correlations between resistance to cephems (cefoxitin, ceftriaxone, and ceftiofur) and the presence of the bla_CMY-2 in S. enterica isolated from both chicken and turkey retail meats. These results highlight the critical role of the bla_CMY-2 in mediating resistance to extended-spectrum cephalosporins in S. enterica, reinforcing previous findings that have identified this gene as a key driver of AMR in foodborne pathogens [8]. The ARIMA models forecast stable or increasing trends over the next seven years in the potential future prevalence of cephems resistance and bla_CMY-2 presence in S. enterica from retail meats, indicating that resistance is likely to remain a significant concern [9]. The predicted increase in cephems resistance in chicken and turkey in the following years reflects a worrying trend that could compromise the efficacy of this critical antimicrobial and make it more challenging to control and manage the infections of S. enterica [10]. Forecasting is essential in the context of surveillance data, offering valuable insights to support decision-making processes and prepare surveillance programs to minimize future occurrences and development of AMR through policy interventions [11]. The high correlations and prediction of stable or increasing trends of resistance suggest that interventions targeting AMR development and dissemination at farm and processing levels are needed. These interventions could involve strategies such as phasing out critically important antimicrobials such as extended-spectrum cephalosporins in poultry production, improving biosecurity measures to prevent the spread of AMR strains at the farm level, improving poultry processing activities that minimize cross-contamination and reduce bacterial load, and enhanced surveillance activities that continue to track AMR development and dissemination at both phenotypic and genotypic levels [12, 13]. While the use of bla_CMY-2 data as a surveillance marker for beta-lactam resistance offers a complementary and, in some cases, a definitive approach to the phenotypic resistance, the results of correlation analysis from this study showed that either bla_CMY-2 genotypic or the phenotypic data for cephems resistance can be used as a proxy for decision-making concerning extended-spectrum cephalosporins resistance in retail meat surveillance programs.

Understanding antimicrobial resistance requires a comprehensive analysis of both phenotypic and genotypic data. Phenotypic detection methods, such as minimum inhibitory concentration (MIC) testing, offer direct measurement of resistance profiles, providing valuable insights into practical resistance scenarios. However, phenotypic methods can be time-consuming, with results often taking several days to obtain, which may delay response efforts, as noted by previous studies [14]. In contrast, genotypic detection methods, which identify specific genetic mutations or resistance genes, offer greater precision and faster results. These methods can detect resistance genes that phenotypic tests might overlook, providing a more thorough understanding of resistance mechanisms [15]. However, genotypic methods require advanced laboratory equipment and expertise, making them more costly and complex compared to phenotypic methods. Moreover, the detection of resistance genes does not always perfectly correlate with observed resistance levels, which can lead to discrepancies [16].

In resource-limited scenarios where there is no sufficient budget to cover both phenotypic and genotypic AMR surveillance, data generated from either phenotypic and genotypic AMR surveillance may be used for estimating annual prevalence since the correlation in the annual prevalence between the phenotypic and genotypic data is almost similar as shown in this study. This could lead to more efficient resources use in national surveillance programs, facilitating for quicker and more accurate detection of emerging AMR trends.

For the retail meat surveillance program, we recommend integrating both phenotypic and genotypic methods to leverage their respective strengths. Genotypic methods can be employed for initial screening, while phenotypic methods should be used for confirmation and detailed characterization of resistance profiles. Analyzing the correlation between phenotypic and genotypic data will enhance understanding of how genetic markers relate to practical resistance profiles, ultimately refining surveillance strategies and improving accuracy.

This study has certain limitations, the ARIMA model used to forecast the annual proportion of bla_CMY-2 and cephems resistance in S. enterica relies on historical data trends and may not effectively predict significant deviations or extreme values. Additionally, gaps or inconsistencies in the data could impact forecast accuracy. Thus, caution is needed when interpreting the forecasts, particularly in the context of abrupt changes or incomplete data. Since this study relies on previously available data for forecasting, missing data could limit the model's ability to achieve full validation.

Conclusions

The strong correlation of annual prevalence between the cephems resistance and bla_CMY-2 presence in S. enterica from retail meats in the United States suggests either data can be used as a proxy for decision-making in retail meat surveillance programs. In addition, by leveraging genotypic and phenotypic resistance data with analytical approaches, we can enhance our understanding and control of antimicrobial resistance in foodborne pathogens, ultimately protecting public health and improving decision-making by the surveillance programs.

Materials and Methods

We retrieved the dataset from the NARMS for Enteric Bacteria from 2002 to 2018 as available on 1 February 2022 (accessed at https://www.cdc.gov/narms/index.html on 1 February 2022). We collected yearly frequency data and estimated the annual prevalence of cephems phenotypic resistance (cefoxitin, ceftriaxone, and ceftiofur) and bla_CMY-2 genotypic resistance in S. enterica isolates from chicken, and turkey retail meat sources in the United States. USDA collected retail meat samples of chicken and turkey and tested them at the reference laboratories in the United States. bla_CMY-2 genotypic resistance was selected from all the beta-lactamases reported by NARMS because CMY-2 AmpC beta-lactamase is the most detected in S. enterica from the NARMS. This also ensured the availability of a sufficient sample size for this study. To conduct our statistical analysis, the dataset was imported into the R open-source computing environment (version 4.4.2). We conducted Spearman’s correlation analysis to explore the association between the annual prevalence of bla_CMY-2 and the annual prevalence of cefoxitin, ceftriaxone, and ceftiofur resistance in S. enterica isolated from chicken and turkey retail meats. The linear relationship was presented graphically using both line and scatterplots. In addition, we used the ARIMA model to demonstrate the time series of the annual prevalence from 2002 to 2018 and forecasted annual prevalence of both phenotypic and genotypic resistance in S. enterica for the next 7-year period (2019–2025) using the automatic selection method of ARIMA parameters. The ARIMA model is characterized by three parameters: (p, d, q) and the presence or absence of a non-zero mean or constant in the equation. The p represents the number of autoregressive terms (the weighted moving average over past annual prevalence observations), d is the degree of differencing (the number of differences needed to make the series stationary), q is the number of moving average terms (the weighted moving average over past annual prevalence errors), and the non-zero mean which indicates that the data fluctuates around an average value that is not zero [17]. The ARIMA model parameters were estimated using the Maximum likelihood estimation method [18]. We used MAPE and MASE to determine the accuracy and forecasting effectiveness of the ARIMA models. The MAPE represents the magnitude of error produced by the ARIMA model and a value close to 0% represents a model with better predictability. While MASE represents the performance of the ARIMA model compared to a naïve forecast, a value < 1 means our ARIMA model performed better than a naïve forecast [19].

We further validated the ARIMA models by comparing the forecasted prevalence with the annual prevalence of both phenotypic and genotypic beta-lactam resistance reported by NARMS from 2019 to 2021. We used the Bland–Altman test of the limits of an agreement to determine the concordance between the ARIMA predictions and NARMS reported data for the three years. The Bland–Altman test shows if the bias difference between the proportions from the ARIMA prediction and NARMS reported data is equal to zero or not. If the P-value of the Bland–Altman test was greater than 0.05, the average bias/difference was not statistically significant different from zero, therefore, the ARIMA predictions were not statistically different from the NARMS-reported data. Ceftiofur data was not available from NARMS and therefore, no validation analysis was performed. For all analyses, statistical significance was considered at P-value less than 0.05.