Interspecific interactions play a critical role in shaping ecosystem structure, species assemblages, and food web dynamics. Specifically, carnivores often exert top-down regulatory control on ecological communities, and the loss of apex predators can lead to profound changes in community composition and ecosystem function. Thus, carnivore reintroduction (trophic rewilding) has gained prominence as a strategy to restore ecological balance. However, the success of such efforts is contingent upon the availability of a sufficient prey base capable of supporting growing carnivore populations, underscoring the significance of bottom-up regulatory processes. Furthermore, species composition and assemblage patterns are strongly influenced by environmental variables and landscape characteristics. In this study, we examined the mammalian community of Panna Tiger Reserve (PTR) to evaluate the relative roles of top-down and bottom-up regulation in shaping mammalian assemblages. We utilized camera trap data from two seasons (2019) and constructed hypothesis-driven interaction pathways representing both regulatory processes within a piecewise structural equation modeling (SEM) framework. The results demonstrated that both top-down (carnivore-driven) and bottom-up (prey-driven) mechanisms significantly influence community structure. Additionally, environmental factors and habitat features were found to be critical drivers affecting the spatial distribution of both predators and prey. Importantly, our findings highlight the dominant influence of top-down effects in species community; however, bottom-up regulation in facilitating the successful reintroduction and population recovery of tigers in PTR, despite the presence of a high density of co-predators. This underscores the essential role of prey abundance and habitat quality in supporting large carnivore conservation. Overall, this study provides key insights into trophic rewilding and emphasizes the importance of maintaining intact habitats, conserving prey populations, and promoting coexistence within multi-species communities.

Data collection

Field methods

We conducted camera trap (CT) surveys in PTR during the summer and winter seasons in the year of 2019. Camera traps were deployed in January-February for winter sampling (mean duration: 32 days, 475 CT stations) and May-June for summer sampling (mean duration: 38 days, 338 CT stations). Each station was equipped with a pair of motion-triggered digital camera traps (Cuddeback C1; www.cuddeback.com), installed within a 2 km² grid-cell framework (​Jhala et al., 2018​). The average inter-station distance was 1.05 km, enabling intensive spatial coverage. Cameras were placed on either side of animal trails, forest roads, riverbeds, or mid-slope locations, at a height of 30–40 cm above ground level, to optimize the detection of large carnivores (​Chen et al., 2009; Johnson et al., 2009; Evans et al., 2019​). All units were operational for 24 h a day and configured to capture a single photograph per trigger, with a 5-second delay between triggers. Camera traps were checked at 15-day intervals for maintenance and data retrieval. The study was conducted under permit number Technical/4301 dated 09/06/2015, issued by the Principal Chief Conservator of Forests (Wildlife Division), Government of Madhya Pradesh, India.

Predictor variables

Environmental data were extracted for each camera trap location using a 100-meter buffer zone from remotely sensed raster layers. Continuous covariates included tree canopy density (%), distance to the nearest village (m), distance to the nearest water source (m), and slope (in degrees; Table S1). These spatial layers were developed and resampled to a 100 m resolution using the ‘raster’ package (Hijmans, 2018) in R (​R Core Team, 2023​). Canopy density and slope were included as proxies for vegetation cover and terrain complexity, respectively. Distance to village served as an indicator of anthropogenic disturbance, while distance to water represented a fundamental ecological resource. All covariates were standardized using z-transformation before analysis. To assess multicollinearity, we applied Pearson’s pairwise correlation test using the cor() function from the ‘corrplot’ package (​Wei and Simko, 2017​) in R (​R Core Team, 2023​), with a correlation threshold of |r| ≥ 0.70 (Figure S1; ​Dormann et al., 2013​).

Analytical methods

To assess the influence of top–down and bottom–up processes on the mammalian community in PTR, we applied a piecewise Structural Equation Modelling (SEM) approach using the ‘piecewiseSEM’ package in R (​Lefcheck, 2016; R Core Team, 2023​). Compared to traditional SEM, piecewise SEM offers a more flexible framework, as it allows for the independent estimation of a set of structured linear equations (​Lefcheck, 2016​). This method is particularly well-suited for ecological datasets that are often smaller in size and exhibit varied distributions (​Shipley, 2009​). In our analysis, we incorporated both environmental variables (as independent predictors) and biological variables (species occurrences as dependent responses). We considered the independent capture events at 30-min intervals obtained from the camera trap data (​O’Brien et al., 2003​), and thereafter, species occurrences were calculated by summing all independent records for each species per CT location (​Dorresteijn et al., 2015​). We adopted a stepwise model-building strategy, developing three distinct models: a top–down model, a bottom-up model, and a combined model integrating both regulatory pathways. Initially, we constructed a priori top-down and bottom-up models to represent hypothesized causal relationships among species and between species and environmental factors. All models consisted of multiple generalized linear models (GLMs) with a Negative Binomial distribution to address the overdispersion of the data and to model the level of interaction among species.

Hypothesis Testing

Top-down processes

Tigers and leopards, the two sympatric carnivores in PTR, formed the base of our top–down hypothesis. We predicted a strong negative causal relationship between tiger occurrence and the abundance of its primary prey species, chital and sambar, as supported by previous studies (​Karanth and Sunquist, 1995; Biswas and Sankar, 2002; Bagchi et al., 2003; Ramesh et al., 2009; Schaller, 2009; Mondal et al., 2012​). Additionally, we hypothesized negative relationships between tigers - nilgai and cattle, which are also preyed upon to varying extents (​Harihar et al., 2011; Navaneethan et al., 2019; Basak et al., 2020​). Wild pig constitutes a substantial portion of the tiger’s diet, warranting a negative causal pathway (​Harihar et al., 2011; Navaneethan et al., 2019; Basak et al., 2020​). Although hares have occasionally been reported in tiger diet studies, their contribution is minimal; hence, we did not predict a strong negative causal link between tigers and hares. Leopards, which have a more generalist and broader dietary niche compared to tigers (​Ramesh et al., 2009; Harihar et al., 2011; Mondal et al., 2012​), were predicted to exhibit negative causal relationships with hare and wild pig, which form key prey items (​Ahmed and Khan, 2008; Basak et al., 2020​). Chital and sambar are also important components of the leopard’s diet, and we thus hypothesized negative pathways toward these species (​Ahmed and Khan, 2008; Ramesh et al., 2009; Harihar et al., 2011; Zehra et al., 2017​). Leopards are occasionally known to prey on nilgai and cattle; hence, we included weak negative pathways to these species as secondary prey (​Ahmed and Khan, 2008; Zehra et al., 2017; Basak et al., 2020​). Environmental factors were also hypothesized to exert a significant influence on species ecology, interactions, and behavioural responses (​Kolipaka et al., 2018​). Water scarcity has been identified as a limiting factor for large carnivore occurrence, and proximity to human settlements is known to influence habitat use negatively (​Dutta et al., 2024​). Therefore, we predicted negative causal pathways between distance to water and both tiger and leopard occurrence, and a positive association between distance from village and tiger occurrence. We also hypothesized that canopy cover and slope would have a positive influence on leopard distribution (​Dutta et al., 2024; Dutta, 2024). Hyenas, as obligate scavengers in this ecosystem, are highly dependent on the presence of apex predators for carrion availability (​Panda et al., 2023; Dutta et al., 2024​). Thus, we predicted positive causal pathways from both tiger and leopard occurrence to hyena occurrence. Given their preference for more arid, open habitats, we hypothesized a negative relationship between hyenas and canopy cover (​Bhandari et al., 2021; Panda et al., 2022​). The golden jackal, a meso-carnivore and facultative scavenger, exhibits a generalist omnivorous diet that includes plant material and carrion of large herbivores such as chital, sambar, and nilgai (​Chourasia et al., 2012; Negi, 2014​). As scavenging opportunities for jackals may also depend on hyena’s presence (​Panda et al., 2022​), we hypothesized positive causal associations between jackals and other carnivores. Furthermore, given that hares are a significant component of jackal diet, we predicted a negative causal pathway between jackal and hare occurrence (​Chourasia et al., 2012; Alam et al., 2015​). Previous studies have shown that jackals tend to prefer areas in close proximity to water sources and actively avoid regions with high human activity (​Katna et al., 2022​). Based on this, we hypothesized a negative causal pathway between jackal occurrence and increasing distance from water, and a positive causal relationship with increasing distance from human settlements.

Bottom-up processes

Prey availability and diversity are critical parameters for sustaining viable carnivore populations (​Dyck et al., 2025​). A robust prey base directly influences predator populations by shaping their spatial distribution and abundance. Previous studies have shown a high degree of spatio-temporal overlap between prey species and top predators, supporting the hypothesis of positive causal pathways from prey to predator (​Hayward et al., 2007; Chaudhary et al., 2025​). As the apex predator, the tiger has access to a wide range of prey species. Earlier research has demonstrated strong spatial and temporal overlap between tigers and prey species such as sambar (​Ramesh et al., 2012; Selvan et al., 2013​), chital, wild pig, and cattle (​Shameer et al., 2021; Biswas et al., 2023; Sharma et al., 2025​). Consequently, these species were hypothesized to have positive causal pathways with tigers. A weaker positive relationship was proposed between tigers and nilgai, based on previous findings (​Bayani and Watve, 2016​). Leopards, known for their broader dietary niche (​Balme et al., 2020​), show spatial association with nearly all available prey species (​Ramesh et al., 2012​). Based on literature, positive causal pathways were constructed between leopards and multiple prey species, including chital (​Sehgal et al., 2022​), sambar (​Mondal et al., 2022; Yadav et al., 2024​), wild pig (​Vinitpornsawan and Fuller, 2020; Shameer et al., 2021; Sehgal et al., 2022​), and hare (​Edgaonkar and Chellam, 2002​). In arid landscapes, cattle have also been shown to contribute positively to leopard diets (​Babrgir et al., 2017; Khorozyan et al., 2020​). Although our top-down model previously hypothesized a negative causal pathway between leopards and nilgai, the bottom-up framework proposed a positive causal relationship between these two species (​Mondal et al., 2012​). Hyenas, being obligate scavengers, were excluded from the bottom-up analysis due to their indirect reliance on prey species. Conversely, we hypothesized a positive causal pathway from hare to jackal, as hare constitutes a key component of the jackal's diet (​Baskaran et al., 2020​). Water availability is a well-documented ecological constraint influencing species interactions (​McCluney et al., 2012​). In earlier studies, it was observed that water acts as a limiting factor for several key prey species, including chital, sambar, nilgai, and hare (​Jhala et al., 2009; Habib et al., 2010; Gupta and Krishnamurthy, 2023​). Thus, we predicted a negative causal relationship between these prey species and increasing distance from water sources. Additionally, both nilgai and wild pig have emerged as significant crop raiders across the Indian subcontinent (​Gautam and Bissa, 2014; Bayani and Watve, 2016; Chauhan et al., 2009; Rao et al., 2015; Milda et al., 2023b​). Accordingly, we hypothesized a negative relationship between their occurrence and distance from villages. Habitat preferences among sympatric prey species further influenced our modeling. For instance, chital are known to prefer flat terrain (​Kumar et al., 2023​) and tend to avoid areas with high anthropogenic disturbance (​Rajawat and Chandra, 2018; Gupta and Krishnamurthy, 2023​), similar to sambar (​Kushwaha et al., 2004​). Thus, a negative association with slope and a positive association with increasing distance from human settlements were predicted for this species. In contrast, sambar typically prefer dense, closed-canopy forests (​Ilyas and Khan, 2005; Gupta and Krishnamurthy, 2023​) and are more frequently associated with undulating terrain (​Ramesh et al., 2013b​). A positive causal relation was assumed in both cases. Cattle grazing is a widespread practice in the central Indian landscape; therefore, we hypothesized a negative causal relationship between cattle occurrence and both canopy cover and increasing distance from villages, reflecting their preference for open habitats close to human settlements (​Nayak et al., 2013​). In contrast, small mammals, such as the Indian hare, are particularly sensitive to anthropogenic disturbance. Previous studies have shown that hares tend to avoid steep terrain and areas of high human activity (​Marinković et al., 2023; Mayer et al., 2023​). Accordingly, we predicted a negative causal relationship between hare occurrence and slope, and a positive relationship with increasing distance from village sites.

Statistical analysis

All previously hypothesized causal pathways: top-down (​Fig. 2a) and bottom-up (​Fig. 3​a), were evaluated using structural equation modelling (SEM). The piecewiseSEM package was used to conduct tests of directed separation, which assess independence claims (i.e., non-hypothesized or missing pathways) among variables. Each independence claim was evaluated based on its statistical significance (p-value). When a claim was found to be both statistically significant and biologically credible, it was incorporated into the optimized model (​Lefcheck, 2016​). Independence claims that were statistically significant but lacked biological relevance or did not improve model fit were excluded. Conversely, any statistically significant and ecologically justifiable claim that enhanced model fit (i.e., resulted in a lower Chi square [χ ]and Fisher’s C value) was retained in the final model (​Stenegren et al., 2017​). This procedure was consistently applied to both species and environmental variables, even in cases where no prior hypothesis existed. To minimize model complexity, non-significant or ecologically irrelevant pathways were eliminated.

Following the separate application of SEM for both the top-down and bottom-up models, a combined model was constructed by integrating only the retained paths from the two optimized models. Subsequently, all three models (top-down, bottom-up, and combined) were evaluated based on goodness-of-fit using Chi-square (χ2) and Fisher’s C statistics, along with their associated p-values (​Lefcheck, 2016​). In both cases, a lower value, accompanied by a non-significant p-value (p > 0.05), was interpreted as indicative of a well-fitting model. Additionally, Nagelkerke’s pseudo R² values were used to assess the model fit for each structural equation. The pseudo R² represents the proportion of explained variance in each structural equation relative to a null model (​Nagelkerke, 1991​). Standardized coefficient estimates (Std. estimates) were used to determine the direction and strength of individual pathways, serving as indicators of effect size.