Plant‐eating carnivores: Multispecies analysis on factors influencing the frequency of plant occurrence in obligate carnivores

Abstract Plant‐eating behavior is one of the greatest mysteries in obligate carnivores. Despite unsuitable morphological and physiological traits for plant consumption, the presence of plants in scat or stomach contents has been reported in various carnivorous species. However, researchers’ interpretations of this subject are varied, and knowledge about it is scarce, without any multispecies studies. This study assessed the extent of variation in the frequency of plant occurrence in scat and stomach contents, as well as its relationship with various factors in 24 felid species using data from 213 published articles. Since the frequency of plant occurrence has not always been reported, we created two‐part models and estimated parameters in a Bayesian framework. We found a significant negative relationship between the frequency of plant occurrence and body mass. This may be because plant‐eating behavior reduces the energy loss caused by parasites and increases the efficiency of energy intake, which has a greater importance in smaller animals that have relatively high metabolic rates. This exploratory study highlights the importance of considering plant consumption in dietary studies on carnivorous species to understand the adaptive significance of this behavior and the relationship between obligate carnivores and plants.

different from that of animals. Contrary to animal cell membranes, which mainly consist of proteins, plant cell walls are rich in carbohydrates, especially cellulose, which is difficult for animals to digest (Tomme et al., 1995;Watanabe & Tokuda, 2001). In addition, several plants contain toxic compounds as an antipredatory defense strategy (Dearing et al., 2005). Thus, herbivores have developed special digestive systems to detoxify secondary compounds and obtain nutrition from a plant-based diet (Hofmann, 1989;Vallentine, 2000).
In contrast, carnivores have predatory and scavenging feeding strategies, possessing numerous traits suitable for hunting and/or eating other animals. Feeding on other animals is nutritionally more efficient than eating plants, since the chemical composition of the food item is quite similar to that of the consumer (Hayami, 1967).
Functional carnivores also have morphological and physiological trait characteristics of this diet. For instance, their dentition is better suited to slicing (Hamper et al., 2012;Van Valkenburgh, 1991), and their digestive tracts are shorter than those of herbivores (Stevens & Hume, 2004) owing to a decreased requirement for fermentation when digesting animal tissue as opposed to plant tissue. Additionally, taste receptor function is altered in many carnivores, including in felids; there is a loss of sensitivity to sugar in fruits and heightened sensitivity to amino acid and bitter compounds (Bosch et al., 2015;Jiang et al., 2012;Kim et al., 2016;Li & Zhang, 2014).
All members of Felidae are considered obligate carnivores, whose diets consist almost entirely of animal flesh, based on their dentition and physiological specialization (Legrand-Defretin, 1994;Morris, 2002;Sanquist & Sanquist, 2002;Van Valkenburgh, 1991;Van Valkenburgh & Gittleman, 1989). These species are widely distributed and inhabit various environments, from the tropics to the frigid zones (Johnson, 2006;Kitchener et al., 2017). In several regions, these obligate carnivores [e.g., tigers (Panthera tigris) (Kapfer et al., 2011), snow leopards (Panthera uncia) (Shehzad et al., 2012)] eat plants even though their diet is considered to be exclusively carnivorous, and despite the aforementioned morphological and physiological traits that are not suitable for plant consumption. Yet, researchers' interpretations of the presence of plant tissues in scat samples or stomach contents are varied, possibly owing to the difficulties associated with observing this plant-eating behavior and because the amount of plant content present in these samples is often small. Some researchers believe that the presence of plant content is caused by unintentional intake (Avenant & Nel, 2002;De Villa Meza et al., 2002;Krofel et al., 2011), while others argue that there might be some advantages of plant eating (Hoppe-Dominik, 1988;Sueda et al., 2008;Tatara & Doi, 1994;Xiong et al., 2016). Indeed, observational studies indicate that felids eat plants voluntarily (Montalvo et al., 2020;Yoshimura et al., 2020) both in the captivity and in the wild, which indicates that this behavior is relatively common and natural among felids. However, experimental studies suggest that cellulose intake can be disadvantageous, since it decreases dry matter and energy digestibility (Edwards et al., 2001;Prola et al., 2010). In addition, because of pseudogenization of the gene encoding a specific detoxification enzyme, felids are unable to detoxify phenolic compounds found in plants (Shrestha et al., 2011). Therefore, there may be some advantage for the existence of plant-eating behavior in felids. Currently, three major hypotheses have been proposed to explain the adaptive significance of plant-eating in carnivores.
First is the self-medication hypothesis (Hart, 2008). Many animals are known to use plants to counter parasites or diseases (Hart & Hart, 2018;Huffman, 2003;Huffman & Canon, 2000). Sueda et al. (2008) reported in a questionnaire survey of owners of dogs under one year of age that these dogs ate plants more frequently, and the authors suggested that plant consumption may be a way for individuals with low immunity to fight parasites and pathogens.
Second is the hair evacuation hypothesis (Shultz, 2019;Yoshimura et al., 2020). Functional carnivores often ingest their own hair while grooming, as well as the hair of their prey. Ingested plants are considered to aid in excreting hairballs (Herbst & Mills, 2010). Third is the food source hypothesis. DNA extracted from leopard cat scats included Solanum and Rosoideae species that produce berry fruits rich in sugar and nutrients (Xiong et al., 2016). Although the replacement of animal food by fruits may be subject to physiological constraints (Larivière et al., 2001), fruits may help obligate carnivores endure starvation or periods when prey animals are scarce.
Currently, knowledge about the plant-eating behavior of felids is scarce, and no comprehensive multispecies analyses have been performed. In this study, we attempted to explore and investigate factors that drive plant-eating behavior of felids in order to understand the common features of this unique behavior among felid species. To clarify whether plant eating is conserved through the evolution of Felidae, we need to evaluate the relationship of this behavior with phylogenic history. Environmental factors also need to be considered since Felids are widely distributed throughout diverse habitats (Johnson, 2006;Kitchener et al., 2017). In addition, given that the body mass of animals affects their diet (Carbone et al., 1999;Kleiber, 1947), its effect should be examined as well. Therefore, in this study, we focused on the aforementioned factors to elucidate their relationship with the frequency of plant consumption in extant feline species.

| Literature search
A literature search using Web of Science (www.webof knowl edge.com) was conducted on 15 September 2020 with the following keywords: "[common name of each species]" OR "[scientific name of each species]" AND "diet" OR "food." Target species were all 41 extant felid species. Common names and scientific names were obtained from the International Union for Conservation of Nature (IUCN)/Species Survival Commission (SSC) cat specialist group (Kitchener et al., 2017).
This search returned 4,100 research articles. The final output was based on the following exclusion criteria: review articles, captive studies (including domesticated animals), studies that were not based on scat or gut contents (e.g., an isotope study using body hair), and noncomprehensive studies (i.e., covered only specific food items). To assess the extent of variation in the frequency of plant occurrence in the diet of carnivores, we additionally sorted these studies according to the following exclusion criteria: sample size of less than 10 and species for which no studies calculated the frequency of plant occurrence. We separated data on fruits and other plants because fruits are different from other plants in terms of energy and nutrients. We only analyzed the data of nonfruit plants because the data of fruits were too scarce to be analyzed by itself. In all, 316 records from 213 studies of 24 felids (some references included records of several species) were used in the analyses (Appendix S1; Yoshimura et al., 2021).

| Environmental factors
We included seven environmental attributes: absolute latitude, island size index, mean monthly precipitation, mean maximum daily temperature, mean minimum daily temperature, mean monthly normalized difference vegetation index (NDVI), and season (spring, summer, autumn, winter, dry, wet). In addition, we added sample type (scat or the digestive tract) because the remains present in the digestive tracts may be greater or lesser than those present in a single scat and may not be directly comparable. Latitude, precipitation, and temperature represent climate parameters of the habitat of subject animals. Since obligate carnivores live in diverse habitats, we added these factors to know whether frequency of plant occurrence relates to specific habitats. Animals on islands often show unique traits due to limited habitat and resources (Foster, 1964); therefore, we added "Island" as a binary variable, which reflects whether the sampling site was an island or mainland including a large island with area over 10,000 km 2 . We attempted to determine the effect of the abundance of vegetation on the frequency of plant occurrence in carnivores' scat and stomach contents through NDVI. Season is mainly characterized by precipitation and temperature; thus, we used the mean values of the studied season for monthly precipitation and daily temperature to consider the seasonal difference.
Where there was seasonal difference independent of precipitation or temperature, we added seasons as binary variables. Climate data were obtained from the MeteoBlue database (Cano-Cruz & López-Orozco, 2015). NDVI data from the Moderate Resolution Imaging

| Phylogenetic factors and body mass
Phylogeny of felids was based on Li et al. (2016). To test the phylogenetic signals in the mean frequency of plant occurrence in each species, phylogenetic eigenvector regression (PVR) was conducted (Diniz-Filho et al., 1998). After extraction of pairwise phylogenetic distances from the branch duration information, the distance matrix was subjected to a principal coordinates (PCo) analysis. Following a broken-stick model (Diniz-Filho et al., 1998;Sakamoto et al., 2010), the first to fifth PCo axes (phylogenetic eigenvector 1-5, PV1-5) were retained (Appendix S3: Table S1 and Figure S2; Yoshimura et al., 2021). These five axes cumulatively explained 86% of the total variance and were included in the analysis as predictor variables for measuring phylogenetic similarity. Additionally, log-transformed body mass values were included as species-specific factors. Body mass data of all species were according to Sakamoto et al. (2010). Since data concerning the body mass of the African wildcat (Felis lybica) were absent, we used the same value as that for the European wild cat (Felis silvestris), according to International Society for Endangered Cats Canada (International Society for Endangered Cats (ISEC) Canada, 2020).

| Statistical analysis
All analyses were performed in R v.3.6.1 (R Development Core Team, 2019). To explain the number of samples that contained plant materials in each study, we constructed two-part binomial (TPB) models. Since the frequency of plant occurrence has not always been reported in dietary studies on carnivores, several records in our dataset lacked values for frequency of plant occurrence. If we ignore records with missing values and apply ordinary regression models, it is likely to lead to imprecise estimation of parameters (Minami & Lennert-Cody, 2013;Minami et al., 2007). Two-part models are considered to be effective when dealing with data with many zero values or data generated from a combination of different mechanisms (Barry & Welsh, 2002;Matsuura, 2016a;Minami & Lennert-Cody, 2013;Minami et al., 2007;Welsh et al., 1996). We assumed that the absence of reported plant material did not necessarily indicate that no plant material was found in the samples, as some reports mentioned that they ignored plant materials in scat or stomach samples (e.g., Abreu et al., 2008;Moleón & Gil-Sánchez, 2003;Silva-Pereira et al., 2011). Specifically, our models assumed that the frequency of plant occurrence has not always been reported irrespective of whether the samples included plant materials, and that the probability of reporting the frequency of plant occurrence follows a Bernoulli distribution with a parameter . Thus, where y i is the number of samples that contained plant materials, N i is the sample size, and p i is the frequency of plant occurrence in record i.

| Model 1: Variation in the frequency of plant occurrence in obligate carnivores
In this model, we assumed that the extent of intraspecies variation in the frequency of plant occurrence differs between species. Thus, where j represents the mean frequency of plant occurrence in species j, i represents the random effect which explains the overdispersion between records, and j is a hyperparameter vector with a length of the number of species (Appendix S2: Table A1).

| Model 2:
Environmental and nonenvironmental factors affecting variation in the frequency of plant consumption in obligate carnivores In this model, we explored the factors that affect the frequency of plant occurrence observed in each study. We assessed the effect of each variable using an approach similar to the hierarchical Bayesian models: where j represents the species-specific intercept of species j, are coefficients of environmental factors X_env, I is the speciesindependent intercept, are coefficients of nonenvironmental factors X_sp (i.e., body mass and phylogenetic eigenvectors), explains the overdispersion between species with hyperparameter , and explains the overdispersion between records with hyperparameter (Appendix S2: Table A1). The number of environmental and nonenvironmental factors is expressed as s and t, respectively. When considering the overdispersion between records, the standard deviation of was assumed to differ between species since different species had different distribution areas, number of references, etc. Thus, hyperparameter is a vector with a length corresponding to the number of species. To consider the effect of collinearity in Model 2, we examined the correlation between environmental factors and between nonenvironmental factors using Pearson's product-moment correlation (r), but |r| < 0.80 (Elith et al., 2006;Matsuura, 2016b) in all pairs.

| Data imputation
We estimated parameters in the models mentioned above using the original dataset (Model 1_1 and Model 2_1). In these models, missing values in the frequency of plant occurrence are treated as the same NA. However, the presence of plant material in samples has been reported in some studies even if the frequency of plant occurrence has not been reported. These descriptions are informative since they mean that missing values were at least above zero. Therefore, we attempted to impute the missing data concerning the frequency of plant occurrence so that there was no waste of information. First, we sorted the literature without information regarding the frequency of plant occurrence into two groups: literature reporting the presence of plant materials in samples and those in which the presence of plant materials has not been reported. We then imputed and replaced the 23 records from 14 references in the first group using two different methods.

| Model 1_2 and Model 2_2: Data imputation with random values
First, random values were sampled from a sequence of 0.01 to 1 in increments of 0.01 to impute the frequency of plant occurrence.
Then, the number of samples containing plant materials (y) was calculated as a product of random values and sample size N for each record that required imputation. Therefore, the parameter was sampled from the posterior distribution of models without data imputation (Model 1_1 and Model 2_1).

| Parameter estimation
We sampled all parameters using the No-U-Turn Sampler (Hoffman & Gelman, 2014) within an MCMC. We ran four parallel chains and calculated the potential scale reduction factor (Rhat; Gelman et al., 2013;Kruschke & Liddell, 2018) to check convergence. The number of iterations was set as 5,000 with 2,000 warm-ups in the models without data imputation (Model 1_1 and Model 2_1). In models with data imputation (Model 1_2, Model 1_3, Model 2_2, and Model 2_3), MCMC sampling was repeated 10 times to reduce the potential effect of specific random value set. Thus, the number of each iteration was set as 2,000 with 1,500 warm-ups to reduce computational load for these models, and posterior distributions from each trial were cumulated. This rate was 1/2, meaning that one of every two consecutive values of posteriors was taken to reduce autocorrelation. If Rhat was 1.0 or less, the model was considered successfully converged. In addition, we conducted graphical posterior predictive checks to determine whether our models were a good fit (Appendix S3: Figures S3 and S4; Yoshimura et al., 2021). Models coded in Stan were compiled into C++ and run using the "rstan" package (Carpenter et al., 2017).
We used a mode of posterior distribution (maximum a poste- ROPE were set to the effect size at half of Cohen's conventional definition of a small effect (Cohen, 1998), that is, [−0.1, 0.1], proposed by Makowski et al. (2019) and Kruschke et al. (Kruschke, 2018;Kruschke & Liddell, 2018). The "rope" function was used to calculate the overlap of HDI and ROPE. Additionally, estimated values were considered significant when the 95% CI did not include zero (Kubo, 2018).

| RE SULTS
Within the 316 records that passed the exclusion criteria, the number of records dedicated to each species varied from 1 [African

| Limitations
Our data relied on the frequency of occurrence data from previous studies. Therefore, we should acknowledge the biases and limitations of the frequency of occurrence method (reviewed in Klare et al. (2011)). The frequency of occurrence method tends to overestimate the importance of small food items as it weighs the presence of small and large food items in the scats equally (Klare et al., 2011;Weaver, 1993

| Phylogenetic factors
The results showed that the frequency of plant occurrence was observed to be higher in Panthera and Caracal, the two earliest F I G U R E 3 Expected a posteriori (EAP; the mean of the posterior distribution) estimates of coefficients of fixed effects. The light and thick error bars represent 95% and 90% credible interval (CI), respectively. The black line indicates zero. Estimated parameters were considered as significant if the 95% CI did not include zero diverging lineages of Felidae (Kitchener et al., 2017;Li et al., 2016) than other felids. This might indicate that plant-eating behavior in felids is a trace of omnivorous ancestral traits (Bradshaw, 2006;Tseng & Flynn, 2015a,b

| Body mass
We found that body mass showed a significant negative correlation with the frequency of plant occurrence, meaning that smaller carnivore species engaged in plant-eating behavior more frequently than larger species. The correlation was significant in Model 2_1 and Model 2_3 according to both the HDI + ROPE rule and 95% CI, but not in the model with random data imputation (Model 2_2).
However, the percentage of posteriors in the ROPE was only 3.9%, and 90% CI did not include zero in Model 2_2 (Figure 3, Appendix S3: Table S4; Yoshimura et al., 2021). In this model, the frequency of plant occurrence was imputed completely at random; therefore, unrealistic values such as 1 might have been applied and affected the posterior distribution. Hence, judging from the overall results, we concluded that body mass has a significant negative correlation with the frequency of occurrence.
One possible explanation for this correlation relates to selfmedication. Kleiber's law states that relative energy consumption is higher in smaller species (Kleiber, 1947). Maintenance metabolism (i.e., the energy required to maintain homeostasis) scales fractionally with body size; as such, smaller animals require more metabolic energy per unit of body mass (Demment & Van, 1985). Therefore, energy loss caused by parasites has higher consequences for smaller carnivores. Moreover, Gregory et al. (Gregory et al., 1996) suggested that host species with higher metabolic rates for their body size may show a greater number of parasite species due to increased food intake. A multispecies study of mammals in Mexico revealed that the order Carnivora showed the greatest occurrence of parasitic helminths and that the host body mass has significant negative correlation with parasite richness (Villalobos-Segura et al., 2020).
These studies support that the cost of parasites is higher in smaller felids than larger species. However, the association between parasite species richness and body weight varies depending on the subject species (Dáttilo et al., 2020); hence, further quantitative study is required to confirm the relationship between host body mass and parasite richness in felids. Several animal species are known to utilize plant physical or chemical aspects against parasites or pathogens (Bosch et al., 2015;de Roode et al., 2013;Hart & Hart, 2018;Huffman, 2003). Consumption of grasses is considered to work as scouring agent against intestinal parasites such as roundworms and tapeworms in canids (Bosch et al., 2015). Small carnivores might eat plants for parasite control, since the energetic costs of parasite load are relatively high. Leopard cat (Prionailurus bengalensis) scat has been reported to contain parasites on Arundinella hirta plant (Lee et al., 2014). Nonetheless, to our knowledge, this is the only study reporting the presence of plant and parasite in the same scat of felids.
Evacuation of hair or undigested materials can be another explanation. Plant-eating behavior in felids is hypothesized to have an effect on hairball evacuation (Herbst & Mills, 2010;Shultz, 2019).
Similar to the aforementioned endoparasites, a greater frequency of plant occurrence in small felids may relate to the high energy cost of an ectoparasite load. Fleas are the main ectoparasite that affect cats, and self-grooming using cornified papillae on the tongue is one of the removal strategies (Hart & Hart, 2018). As the cost of ectoparasite load increases, the intensity of grooming increases, which is likely to result in increased ingestion of its own hair by the animal.
Carnivores weighing less than 21.5 kg generally consume animals consisting of 45% or less of their own mass, while those weighing more than 21.5 kg prey mostly on animals larger than themselves (Carbone & Gittleman, 2002). Small prey consumption often includes the ingestion of indigestible parts such as fur, skin, bone, and connective tissue, besides muscle and organs, while large carnivores can selectively eat digestible parts (Clauss et al., 2010;Stirling & McEwan, 1975). In humans, dietary fiber intake is known to promote digestion and bowel movements by stimulating peristalsis and mucus secretion in the digestive tract (Chutkan et al., 2012;El-Salhy et al., 2017). Plant consumption might promote digestion or excretion of indigestible food items, which are consumed by small carnivores at a high frequency. Sugar cane-derived fibers reduced the size of hairballs in the scat of domestic cats (Loureiro et al., 2014). However, cellulose, one of the main insoluble fibers, did not have such an effect (Loureiro et al., 2014), and plant intake had little effect on hair evacuation in captive snow leopards (Yoshimura et al., 2020). Owing to the aforementioned attributes of prey items, smaller carnivores are considered to be more tolerant to indigestible food items (Jethva & Jhala, 2004;Rühe et al., 2008). Indeed, Vester et al. (2008) demonstrated that small felids have higher digestion ability of dietary fiber, and Kerr et al. (2013) showed that tract dry matter, organic matter, fat, and energy digestibility coefficients decreased linearly with body weight in four medium-to-large cats [jaguar (Panthera onca), cheetah (Acinonyx jubatus), Malayan tiger (Panthera tigris corbetti), and Siberian tiger (Panthera tigris altaica)] fed cellulose and beet pulp diets. Although cellulose intake reduces dry matter and energy digestibility both in large (Kerr et al., 2013) and small felids (Edwards et al., 2001;Prola et al., 2010), smaller animals may be less affected, which could explain their increased tolerance to more frequent plant consumption. Nevertheless, this can be true whether or not plant intake has some adaptive significance for obligate carnivores, and thus, this does not negate the self-medication hypothesis or the hair evacuation hypothesis.  (Monterroso et al., 2019) are important to infer the role of plant intake. Hypothesis-centered studies will provide direct evidence about the adaptive significance of plant eating as well. By unraveling the relationship between carnivores and plants, we will be able to understand not only their behavioral ecology but also their interactions within ecosystems.

ACK N OWLED G M ENTS
We thank Dr. Sota Inoue for providing helpful advice on statistics and Annegret M. Naito-Liederbach for reviewing a draft of the paper and providing helpful comments. We would like to thank Editage (www.edita ge.com) for English language editing.

CO N FLI C T O F I NTE R E S T S
We have no competing interests.

A PPE N D I X 1
Here, we describe the characteristics of plant eating in each lineage.
However, only seven lineages are described here because we did not have the data from Bay cat lineage.

Domestic cat lineage
This represents the most recent lineage and consists of smaller species . Among the 67 records found for this group, 55 described feral cats. There were 6 studies reporting fruit detection (Biró et al., 1999;Carvalho & Gomes, 2004;Ferreira et al., 2014;Lanszki et al., 2016;Meckstroth et al., 2007;Spencer et al., 2014), with fruit possibly having been consumed as food. However, fruit was detected more frequently in domesticated cats than in feral cats living on a Croatian island (Lanszki et al., 2016), suggesting that the detection of fruit content might be associated with proximity to human activity [e.g., food provisioning or scavenging garbage (Yamane et al., 1994)]. Additionally, there were several studies showing the presence of nonfruit-bearing plants, which may have other benefits, such as parasite control (Hart, 2008;Hart & Hart, 2018;Sueda et al., 2008).

Leopard cat lineage
This group consists mainly of small species inhabiting Central to South-East Asia. The two species used in this analysis had a relatively high frequency of plant occurrence. Parasites, together with A. hirta, were detected in leopard cat scat in Korea (Lee et al., 2014), implying that plants likely contributed to antiparasite measures or promoted gastrointestinal tract movement (Tatara & Doi, 1994).
Although no cases of fruit detection have been reported in the leopard cat, a DNA-based study of its scat contents in China showed fruit-bearing species, suggesting its use as food (Xiong et al., 2016).

Puma lineage
This
The ocelot's larger body mass might have caused its relatively low frequency of plant occurrence compared with that for smaller species.
Additionally, ocelots, pumas, and jaguars (Panthera onca) have been observed eating wild rice containing high levels of cyclooxygenase inhibitors (Montalvo et al., 2020), which works as an anti-inflammatory agent in dogs and cats (Jones & Budsberg, 2000). However, it should be noted that studies on this topic are scarce and there is a high degree of uncertainty in the estimates.

Caracal lineage
This lineage consists of medium-sized species that live mainly in has glands that secrete acidic substances (Dippenaar-Schoeman et al., 2018). Caracals may eat this plant to ingest these compounds possibly for self-medication (Hart & Hart, 2018;Huffman, 2003) or for pH control in the digestive tract (Kerr et al., 2013), although it is unclear whether these compounds have a beneficial effect. This finding further suggests that these animals might use plant odor as one of the selecting factors for consumption.

Panthera lineage
These so called "big cats" constitute one of the basal lineages of extant felid species . Fruit consumption has not been reported for them; however, the presence of grasses and shrubs has been detected in numerous cases (e.g., Jumabay-Uulu et al., 2014;Ott et al., 2007;Tkachenko, 2012). Hoppe-Dominik stated that leopards (Panthera pardus) may eat grasses to keep their digestive tract moving during starvation (Hoppe-Dominik, 1988). However, captive snow leopards also ate plants regularly even though they were fed daily (Yoshimura et al., 2020), suggesting that starvation is not always the trigger for plant eating. Furthermore, it has been suggested that grasses are selectively eaten because they are free of secondary plant compounds, unlike those in other plant groups (Hoppe-Dominik, 1988). Indeed, undigested Poaceae and Cyperaceae plants were detected in 40%-50% of the scat of leopards (Hart et al., 1996) and tigers (Tkachenko, 2012), similar to that in the scat of puma and jaguarundi (Rocha-Mendes et al., 2010).
Therefore, these plant species may be consumed not for medicinal secondary compounds but for physical traits such as hairs on their surface (Hoppe-Dominik, 1988).

Snow leopards and leopards have been reported to eat Myricaria
shrubs in addition to grasses (Jumabay-Uulu et al., 2014;Lovari et al., 2013;Wegge et al., 2012). Tamaricaceae plants (the family that includes Myricaria) have been detected in 4.1%-16.9% of scat and constituted the bulk of hairballs (Lovari et al., 2013), although it is uncertain that hairballs were caused by plant intake. These Myricaria plants have anti-inflammatory properties and have been used as traditional medicines (Chernonosov et al., 2017;Liu et al., 2009). Cold and dry climates restrict the transmission and growth of parasites (Morris, 2002), whereas low temperature increases the probability with NDVI, support the possibility that plant consumption has some advantage for carnivores.

A PPE N D I X 2
A list of parameters in the models.

N_all
Integer 1 The total number of records