Investigating the association between socioeconomic and agricultural risk factors and the incidence of Salmonella Heidelberg and Salmonella Typhimurium in Ontario in 2015: A multi‐level ecological approach

The incidence of salmonellosis, and other enteric zoonotic diseases, has been associated with various food and environmental exposures and socioeconomic factors. Increasingly, there is interest in exploring differences among serotypes of Salmonella to better inform public health prevention efforts. Consequently, we investigated whether rates of Salmonella Heidelberg and Typhimurium in Ontario communities in 2015 were influenced by household characteristics, agricultural factors, and the presence of meat plants. Data for each reported case of S. Heidelberg and S. Typhimurium in Ontario in 2015 were collected. Expected cases of each serotype were calculated, stratified by age group and sex, according to the underlying population distribution from the Canadian census. Socioeconomic, agricultural census data, and data concerning provincial and federally inspected meat plants were combined with observed and expected case counts. The association between community‐level agricultural, meat processing, and socioeconomic variables, serotype, and the rate of salmonellosis in each census subdivision (CSD) was explored using multilevel Poisson models, with random intercepts for CSD and census division (CD). Rates of S. Heidelberg and S. Typhimurium were associated with the proportion of married individuals in a CSD, and were higher in CSDs with the highest quantile of labour participation compared to those in the lowest quantile. There was an interaction effect between cattle, poultry and swine farm density in a CD and serotype, with rates of either serotype decreasing as cattle, poultry, or swine farm density in the encompassing CD increased. The rate of the decrease varied by serotype. Our findings concerning community‐level household characteristics may be explained by the influence of family structure and occupation on food consumption patterns and environmental exposures. Rates of S. Heidelberg and S. Typhimurium may be lower in areas with increased animal farm density due to naturally acquired immunity from routine exposure to Salmonella via livestock.


| INTRODUC TI ON
Enteric pathogens are a frequent cause of sporadic gastrointestinal illness in Canada, and less commonly, are associated with outbreaks of illness.Individuals may be exposed to enteric pathogens via their occupation, indirect or direct animal contact, or via consumption of contaminated food or water (Currie et al., 2005;Hale et al., 2012;Hanning et al., 2009;Tangkonda et al., 2021;Whitfield et al., 2017).
Enteric illnesses can place a large burden on the healthcare system, both in terms of hospitalizations and lost work time.Ideally, actions taken locally, provincially, and nationally to prevent the occurrence of these illnesses will reduce their incidence and impact.
Salmonellosis is one of several enteric illnesses that is considered a reportable disease of public health significance in Ontario, and Salmonella is consistently the second most common bacterial enteric pathogen identified in Ontario, after Campylobacter spp.(Ontario Agency for Health Protection and Promotion (Public Health Ontario).
(PHO), 2021).In 2015 for example, there were 3299 reported cases of Campylobacter enteritis in Ontario (24.1 cases per 100,000 population) and 2900 reported cases of salmonellosis (21.2 cases per 100,000 population) (Ontario Agency for Health Protection and Promotion (Public Health Ontario).(PHO), 2021).While there are over 100 serovars commonly associated with human salmonellosis, three serovars are consistently responsible for the bulk of identified infections in Ontario each year: Salmonella Enteritidis, S. Typhimurium and S. Heidelberg.In 2015, these serovars collectively accounted for 52.7% (35.7%, 9.9% and 7.1%, respectively) of reported cases of salmonellosis in Ontario (Ontario Agency for Health Protection and Promotion (Public Health Ontario).(PHO), 2021).While S. Enteritidis is consistently the most identified serotype responsible for human salmonellosis, from 2006 to 2011 the incidence of S. Typhimurium was observed to decrease in Ontario, while the incidence of S.
Heidelberg increased, making the changing epidemiology of these two serotypes of particular interest, especially since the exposure responsible for individual illness is often reported by investigating public health units as unknown (Ontario Agency for Health Protection and Promotion (Public Health Ontario).(PHO), 2021).
Many previous studies have examined the epidemiology of S. Enteritidis, and individual illness is often attributed to travel or to consumption of eggs or poultry (Hobbs et al., 2017;Middleton et al., 2014;Tighe et al., 2012).The potential sources of S. Heidelberg and S. Typhimurium are believed to be more variable, including many different animal species that act as natural reservoirs for the bacteria, including food animal species such as cattle, poultry, and swine, with people being exposed to the bacteria domestically or via travel (Callaway et al., 2008;Kingsley & Bäumler, 2002).Salmonella Typhimurium was isolated from all sources but most commonly from swine (Flockhart et al., 2017).By comparison, S.
Heidelberg was isolated from cattle, broiler chickens and water, but most commonly from broilers and not from swine (Flockhart et al., 2017).Transmission of the same serotypes of Salmonella between animal species and from animals to people was hypothesized to have occurred through contact between animals and people, between animal species (and specifically from exposure to poultry), and/or from faecal contamination of the environment (Flockhart et al., 2017).
Previous studies have identified an association between the incidence of enteric pathogens, such as Campylobacter, Salmonella, and E. coli O157:H7, and environmental factors, including living in areas of high agricultural density, or with numerous bodies of freshwater, or near a licensed abattoir (Arsenault et al., 2012;Graziani et al., 2015;Paphitis et al., 2021;Pearl et al., 2009).Various factors are believed to have contributed to the identified geographic associations, including working in close contact with live animals, or visiting seasonal locations such as petting zoos and recreational bodies of water, particularly during warmer summer months (Arsenault et al., 2012;Graziani et al., 2015;Paphitis et al., 2021;Pearl et al., 2009).
Researchers have also investigated the associations between socioeconomic risk factors, such as education, household size, access to health care, moving from the usual place of residence (home address within or outside of Canada) within the last year, or being part of a low-income household, and observed rates of enteric illness, and the interactions between agricultural and socioeconomic variables (Pardhan-Ali et al., 2013;Pearl et al., 2009;Younus et al., 2007).
Recent studies investigating the role of community-level socioeconomic variables on observed rates of illness found that increasing education was associated with increased rates of laboratory-confirmed salmonellosis at the neighbourhood level, potentially due to differing health-seeking behaviours (Younus et al., 2007), and that after adjustment for other socioeconomic variables at the community level, the rate of confirmed salmonellosis generally increased as the number of households with inadequate or unsuitable housing increased (Pardhan-Ali et al., 2013).

Impacts
• Rates of Salmonella Heidelberg and Typhimurium decreased in areas with increased animal farm density, which may at an individual level be due to increased natural immunity from environmental exposure.
• Rates of both serotypes were influenced by the proportion of individuals in a community who were married or participating in the labour market, which may reflect the influence of factors such as occupation and family structure on food consumption patterns and environmental exposures.
• Living in an area with one or more meat plants was not associated with increased rates of illness at a population level; previously identified clusters around meat plants may indicate localized risk factors or exposure events.
Given the previously identified associations between agricultural and socioeconomic factors and enteric illness, the goal of this ecological study was to further investigate proposed agricultural and socioeconomic risk factors, and their contribution at a community level to the observed incidence of salmonellosis due to S. Typhimurium and S. Heidelberg in Ontario.Specific objectives were to investigate whether the rates of illness due to S. Typhimurium or S. Heidelberg were higher (a) in communities with one or more meat plants or with increasing agricultural density, influenced by (b) socioeconomic and demographic variables, such as the proportion of households comprised of three or more individuals and the proportion of those in a community who had completed some level of post-secondary education, and (c) whether the impact of these agricultural and socioeconomic factors at a community level varied by serotype.

| MATERIAL S AND ME THODS
Case data were obtained from Public Health Ontario for all human cases of S. Heidelberg and S. Typhimurium that had been reported to public health units in Ontario in 2015.Where an individual had more than one reported episode of salmonellosis involving the same serotype in 2015, only the first episode was retained for analyses.
Individual-level data available for aggregated analyses included the date of symptom onset or the date on which the case was reported to the local public health unit, the age in years of each case, gender, responsible serotype, and the 6-digit postal code of their home address (where available).Public health units are comprised of numerous municipalities (cities/towns) or their equivalent, which are referred to as census subdivisions (CSDs) for statistical census reporting purposes (Statistics Canada, 2010).CSDs are further aggregated upwards into Census Consolidated Subdivisions (CCSs; groupings of adjacent CSDs) and then to census divisions (CD; neighbouring municipalities) for regional planning and large-scale service delivery purposes (Statistics Canada, 2010).CDs aggregate to provinces or territories.
The 6-digit postal code of the home address for each case and the Statistics Canada Postal Code Conversion File Plus (PCCF+, Version 5E, 2009;Wilkins, 2009) were used to assign cases to the relevant CSD, CCS, and CD using SAS version 9.4 (SAS Institute, Cary NC).
Population data from the 2006 census were retrieved from Statistics Canada (Statistics Canada, 2007-2008) for each CSD.
The following census variables were selected for subsequent analyses: the proportion of residences occupied by the usual owner; the proportion of respondents reporting their marital status as 'married'; the proportion of all residences (owner or tenant occupied) where 30% or more of household income was spent on shelter; the proportion of respondents who reported living and working in the same CSD; the proportion of respondents who had obtained a certificate, diploma or degree of higher education; the proportion of those aged 15 years and older who reported being in the labour force; the proportion of those who had not changed their home address (moved) in the previous year; and the proportion of private households with 3 or more people (compared to 1 or 2 people).
Statistics Canada classified each CSD into one of the following seven categories according to its statistical area classification (SAC type): census metropolitan area (CMA), tracted census agglomeration, non-tracted census agglomeration, strongly influenced zone, moderately influenced zone, weakly influenced zone, no influenced zone, and territories (Statistics Canada, 2006).For our analyses, data for tracted and non-tracted census agglomerations were combined into a single category (collapsing SAC type into six categories) to obtain roughly similar sample sizes across SAC types for analyses.
Where applicable, census population variables were converted to proportions for regression analyses (Table 1).Census data for 2006 were used as this was the most recent full census data available at the time of data processing and analysis.The total number of reported cases for each serotype (S.Heidelberg or S. Typhimurium) was summed for each age and sex category within each CSD.The expected number of cases of each serotype was calculated for each age and sex stratum using indirect standardization based on the age and sex stratum-specific rates for the entire province in 2015, and the age and sex stratum population numbers for each CSD.For regression analyses, only those cases were included that could be linked by postal code to the relevant CSD.
Publicly available online data regarding the location of provincially and federally inspected meat plants conducting slaughter that were operational in all or part of 2015 were obtained from the Ontario Ministry of Agriculture, Food and Rural Affairs (OMAFRA) and from the Canadian Food Inspection Agency (CFIA), and the total number of meat plants in each CSD was determined using the postal code of each abattoir and the PCCF+ file, as outlined previously.Where available, data regarding the overall total number of farms in each CD (including those producing livestock and crops), and the total number of farms producing the following animal species were obtained from the 2006 Canadian Census of Agriculture: cattle (dairy cattle and milk production and/or beef cattle ranching and farming), swine (hog and pig farming), and poultry (chicken egg producers, poultry hatcheries, and broiler and turkey meat producers) (Statistics Canada, 2006).Agricultural density, cattle, poultry, and swine farm density were calculated as the total number of all farms (including those producing animals and crops), cattle farms, poultry farms, and swine farms, respectively within a CD, divided by the total area of the CD (km 2 ).All data described above were collated into a single dataset for analyses using Stata v. 15.0 (STATACORP, College Station, TX).
Multilevel Poisson regression modelling was applied to explore the relationship between the incidence rate of S. Typhimurium and S. Heidelberg in a CSD and the above-mentioned explanatory variables.The linearity assumption for each continuous predictor variable was examined visually using a locally weighted regression, plotting each continuous predictor against the natural log of observed over expected cases of both serotypes combined, and by introducing a quadratic term for each continuous variable.Where a non-linear relationship was identified, it was modelled as a quadratic relationship if appropriate.Otherwise, the variable was categorized TA B L E 1 Descriptions of dependent and independent variables used to examine the associations between the rates of Salmonella Typhimurium and Heidelberg and agricultural and socio-economic variables in Ontario (2015).Initially, multilevel univariable Poisson models were fitted with random intercepts for CSD and CD to examine the associations between the rate of salmonellosis in a CSD (without differentiating between the two serotypes) and the demographic, agricultural, and slaughter plant variables examined.For each serotype, the observed number of cases was included as the dependent variable and the natural log of the expected number was the offset (Lawson et al., 2003).
In addition, we examined if there were significant differences between the two serotypes by including serotype as a fixed effect (SH vs. ST).Following this process, we examined whether independent variables interacted with serotype and whether the removal of serotype confounded the association between the independent variable and the outcome.Confounding was defined as a 20% or greater change in the coefficient of a significant variable following removal of the explanatory antecedent or distorter variable.A variable was considered for inclusion in the multivariable modelling process if it had a significant interaction with serotype or remained significant after the inclusion of serotype in the model.
Poisson models were fitted to examine the associations between the rate of salmonellosis with serotype and the other variables and interactions initially identified from the variable screening process described above.A manual backwards elimination process was used, and variables were retained if they were significant, were part of a significant interaction, or acted as an explanatory antecedent/distorter variable (i.e., confounder) on a significant variable in the model.A significance level of 5% (i.e., α = 0.05) was used for all analyses.
However, recognizing that this is an observational study where the sample size was determined by the availability of the data, and acknowledging concerns over the misuse of the term "statistical significance" and misinterpretation of the threshold of p < 0.05 to indicate a significant association (Wasserstein & Lazar, 2016), these p-values were used in an exploratory sense.All models initially included random intercepts for both CSD and CD.In the final multivariable models, a random intercept was removed if its variance component was extremely small (<1 × 10 −5 ), its removal did not change our interpretation of the fixed effects in the model, and the fit of the model was not improved with its inclusion based on reductions to the Bayesian Information Criterion (BIC).
The fit of the multilevel Poisson models were assessed via examination of the best linear unbiased predictors (BLUPs) to determine if these met the assumptions of normality and homogeneity of variance.Pearson and deviance residuals were examined to identify potential outliers.Where outlying observations were identified, their impact on the main effects model was assessed via their removal from the model and reported.For all models, we examined whether model fit was improved by fitting a mixed negative binomial model using BIC.
Ethics approval for this study was received from the University of Guelph Research Ethics Board (REB#14JN010).All analyses were performed in STATA v. 15.0 (STATACORP).

| Descriptive statistics
There were 490 cases of S. Typhimurium and S. Heidelberg (282 S. Typhimurium, 208 S. Heidelberg) reported by public health units via the provincial electronic reporting system in 2015.These cases were reported from each of the then 36 public health units, and from 95 of A total of 161 federally or provincially licensed abattoirs were reported to have been operational in all or part of 2015.These were located across 103 CSDs, with the number of abattoirs in a single CSD ranging from 0 (424 CSDs) to 7 (1 CSD), with a median of 0.
These processed one or more food animal species, including cat-

Univariable models
Univariable models assessing the association between the rates of these serotypes and individual agricultural census, meat plant, and census variables identified significant associations with labour participation rate and with the proportion of individuals within a CSD who reported their marital status as married (Table 2).

Examining effect of serotype on univariable associations
Serotype did not act as an explanatory antecedent or distorter variable for the census or meat plant variables identified in our univariable models.However, significant interactions were identified between serotype and each of cattle, poultry, and swine farm density (Table 3).

Multivariable models
Poultry, cattle, and swine farm density were highly correlated with one another (R > 0.60), so separate multivariable models were fitted with each variable.Regardless of the farm animal density variable included in each multivariable model, the incident rate of Salmonella of either serotype was significantly higher in CSDs with the highest quantile of labour participation compared to the lowest quantile (Table 4).Similarly, the rate was significantly higher for CSDs within the second highest quantile of married individuals (i.e., proportion married) vs. the lowest quantile in the model that included swine density and in the model that included cattle density (Table 4).
Although marital status was not significant in the model for poultry farm density, this variable was retained in the model due to its confounding effect on other variables.Significant interactions were identified between serotype and each of cattle, poultry, and swine farm density (Table 4).For each animal farm density variable, the rate of Salmonella declined as farm density increased, but the rate of decline was faster for S. Heidelberg than for S. Typhimurium (Figures 1-3).
Assessment of the best linear unbiased predictors (BLUPs) for each model found that assumptions for normality and homogeneity of variance for CSDs were not completely met, but model fit was substantially improved based on BIC with the inclusion of the random intercepts compared to models with no random effects.Assessment of Pearson and deviance residuals for each model identified several outlying observations, but their removal did not change the interpretation of the model, so all observations were retained in each model.

| DISCUSS ION
Previous research has demonstrated the impact of various socioeconomic, environmental, and agricultural risk factors on the rates of enteric illness in a community, including factors such as occupation, food preparation and consumption patterns, agricultural density, recreational water exposure, the use of contaminated water for crop irrigation, and animal contact (Arsenault et al., 2012;Currie et al., 2005;Flockhart et al., 2017;Graziani et al., 2015;Hale et al., 2012;Hanning et al., 2009;Tangkonda et al., 2021;Whitfield et al., 2017).A previous study estimated that approximately 11% of nontyphoidal salmonellosis cases in the United States from 2000 to 2010 were likely attributable to animal contact (Hale et al., 2012), and a similar study in Ontario estimated that approximately 26% of enteric diseases between 2010 and 2012 were likely associated with animal contact (Whitfield et al., 2017).However, ingestion of contaminated food, particularly produce and food products derived TA B L E 2 (Continued) from animals, and not indirect or direct contact with animals is believed to be the primary source of most human infections (Wegener et al., 2003;Whitfield et al., 2017).
Interestingly, our community-level analyses found that the rate of reported cases of S. The finding that the rate of S. Heidelberg cases decreased more rapidly than S. Typhimurium cases as swine, cattle, or poultry farm density increased may also be influenced by pathogen or host effects, including the normal host range of each serotype.For example, the more profound decline observed for S. Heidelberg may reflect the narrower host range for S. Heidelberg, which is primarily found in poultry, compared to S. Typhimurium, which has a broader range of host species, including poultry, swine, sheep and cattle (Kingsley & Bäumler, 2002).Additionally, the on-farm prevalence of various serotypes may differ from overall human population prevalence of each serotype, due to various factors including host-specificity, and on-farm herd management practices.Future studies could explore on-farm prevalence of various serotypes compared to those responsible for human illness, similar to the work of Flockhart et al. (2017).
Although previous research using the spatial scan statistic identified several clusters of human S. Heidelberg cases around meat plants conducting slaughter in Ontario in 2015 (Paphitis et al., 2021), the present study did not find a significant association between the rate of observed S. Heidelberg or S. Typhimurium in a CSD and the number of meat plants in the same CSD.This might suggest that any previously identified clusters associated with meat plants were likely due to one or more localized risk factors or exposure events (e.g., within-plant sanitation and food handling practices, or the presence of a public facing retail counter) and not to a broader increased risk of human illness associated with living near one or more meat plants.
As individuals working in meat plants may have ongoing exposure to animals, they may also develop immunity to infection, similar to farm workers.If meat plant workers live close to their place of employment (i.e., within the same CSD) this may contribute to a decreased incidence of salmonellosis in these CSDs.However, we found no association between the presence of meat plants and the rate of S.
Heidelberg or S. Typhimurium in CSDs.Additionally, it is important to note that the previous spatial analysis (Paphitis et al., 2021)

Flockhart
et al. (2017) compared Salmonella serovars found in humans within an Ontario sentinel site from 2006 to 2011 to those found in cattle (beef and dairy), poultry, swine, and the local watershed in the same area, and during the same time frame.

585
CSDs in Ontario (18.0%).Dropping the second reported episode of the same serotype for each of 2 cases left 488 cases for further analyses.Twelve of these cases were linked to known outbreaks in 2015 (nine cases of S. Heidelberg and three cases of S. Typhimurium) with the remainder considered to be sporadic cases.Forty sporadic cases (all cases from each of 5 PHUs (collectively comprised of 148 CSDs): Kingston-Frontenac-Lennox & Addington (n = 7), Lambton County (n = 3), Middlesex-London (n = 13), Northwestern (n = 1), and Renfrew County and District (n = 6)) and an additional 10 cases from various PHUs with no postal code data were lost to analyses involving census geography, as missing postal code information for these individuals meant that they could not be assigned to the relevant CSD or CD.A total of 448 cases, including those 12 cases linked to identified outbreaks, were subsequently available for regression analyses(198 SH, 250 ST).
Heidelberg and S. Typhimurium decreased as poultry, cattle, or swine farm density measured at the CD-level increased.While this finding may initially seem paradoxical, at an individual level, it is plausible that long-term exposure to animals may reduce the risk of salmonellosis and other enteric, zoonotic diseases.Previous research found that individuals working in poultry meat plants developed naturally acquired immunity to Campylobacter jejuni due to their ongoing exposure to poultry in this setting, and that individuals who lived on or had visited a farm were less likely to become ill with S. Heidelberg (Currie et al., 2005; Tangkonda et al., 2021).If individuals living in areas of high agricultural density are more likely to work in occupations involving direct or indirect contact with animals and their environment, including farms and meat plants, or to live on a farm property, these individuals may (after initial exposure or infection) have some level of immunity to subsequent illness due to Salmonella and other enteric zoonotic bacteria.

a
Animal farm density used the total number of swine farms/CD km 2 (model A), poultry farms/CD km 2 (model B) or cattle farms/CD km 2 (model C). b Do not interpret main effects independent of their interaction terms.Refer to Figures 1-3 for interpretation of interaction effects involving a continuous variable.* p-value significant at p < 0.05.consumption patterns at an individual level.Labour participation rates may also differ based on the type of labour being performed, which would also influence environmental and other workplace related exposures.Although we examined marital status as one of several characteristics of the home environment, including household size and proportion of the households spending 30% or more of their monthly income on shelter costs, it was the only factor significantly associated with the rates of Salmonella for either serotype in two of our models and was retained in another due to its confounding effect on other associations.We suspect the proportion of married individuals in a community is capturing other cultural factors that influence potential exposures to these serotypes.It is likely that the community-level associations explored here were surrogates for important exposures that could not be directly assessed.For instance, if married individuals were more likely than non-married individuals to have young children or pets, this may increase the potential for within-household zoonotic transmission of disease, or person-to-person transmission.Future studies could further explore individual-level and household-level factors that may be associated with marital status, and subsequent exposure risk.It should be noted that associations based on rates and characteristics measured at a population level are not necessarily reflective of associations or risk factors at an individual level, and caution should be applied in generalizing these community-level findings F I G U R E 1 Predicted log of standardized morbidity ratio (SMR) of observed vs. expected Salmonella Typhimurium (ST) and Salmonella Heidelberg (SH) cases in Ontario CSDs in 2015 by serotype and swine farm density in the encompassing census division (CD) based on a multivariable mixed Poisson regression model.F I G U R E 2 Predicted log of standardized morbidity ratio (SMR) of observed vs. expected Salmonella Typhimurium (ST) and Salmonella Heidelberg (SH) cases in Ontario CSDs in 2015 by serotype and poultry farm density in the encompassing census division (CD) based on a multivariable mixed Poisson regression model.F I G U R E 3 Predicted log of standardized morbidity ratio (SMR) of observed vs. expected Salmonella Typhimurium (ST) and Salmonella Heidelberg (SH) cases in Ontario CSDs in 2015 by serotype and cattle farm density in the encompassing census division (CD) based on a multivariable mixed Poisson regression model.
Total number of cattle and ranching farms in each CD, divided by the area of the CD in km 2 .
The significance of each variable and interaction terms were assessed using likelihood ratio tests.If independent variables were highly correlated, either the most biologically plausible variable was included in the model or separate models The results of univariable multilevel Poisson models with random intercepts for CSD and CD, examining the associations between the rate of Salmonella (Heidelberg or Typhimurium) in a CSD in Ontario, Canada in 2015, serotype, and census and agricultural census variables.Each model included data for 379 CSDs, except the model examining labour participation rate, which only included data for 378 CSDs.
TA B L E 2 b *p-value significant at p < 0.05.

level -with fixed effect for swine farms B. Poisson 2-level -with fixed effect for poultry farms C. Poisson 2-level -with fixed effect for cattle farms
The results of two-variable multilevel Poisson models with random intercepts for CSD and CD, examining the associations between the rate of Salmonella (Heidelberg (SH) or Typhimurium (ST)) in a CSD in Ontario, Canada in 2015, serotype, and census and agricultural census variables, and where a significant interaction was identified between serotype and a predictor variable.The results of multilevel Poisson models with a random intercept for CSD, examining the association between the rate of Salmonella (Heidelberg (SH) or Typhimurium (ST)) in a CSD in Ontario, Canada in 2015, and serotype and agricultural census variables, with model A including swine farm density, model B including poultry farm density and model C including cattle farm density.
The associations between rates of Salmonella Heidelberg and Typhimurium and labour participation and marital status may reflect unmeasured exposures and behaviours.The finding that increased labour participation in a CSD was associated with a corresponding increased rate of salmonellosis may be due to differing diet and food TA B L E 3 a Do not interpret main effects independent of their interaction terms.*p-valuesignificant at p < 0.05.TA B L E 4