Parasite‐Mediated Competition Limits Dominant Cervid Competitor

ABSTRACT

Species interactions structure ecological communities through direct and indirect pathways with ecosystem‐wide implications. Despite mounting interest in the importance of indirect interactions, empirical evidence remains limited. Here, we demonstrate the critical role of parasite‐mediated competition in driving community outcomes in a multi‐species system of conservation and management concern. We leveraged 2 years of detection/non‐detection data of moose ( Alces alces ) and white‐tailed deer ( Odocoileus virginianus ) and parasite loads in faecal samples within a hierarchical abundance‐mediated interaction model to test hypotheses regarding interactions between these cervids and their shared parasites (Parelaphostrongylus tenuis, Fascioloides magna ). We demonstrate that moose occupancy was limited by parasite‐mediated competition, with no evidence of population‐level effects of direct competitive interactions between moose and white‐tailed deer. Such evidence of the importance of indirect interactions and resulting community outcomes is critical for species conservation and managing range contractions due to increasing pressures from habitat loss, disease and climate change.

We leveraged 2 years of detection/non‐detection data of moose ( Alces alces ) and white‐tailed deer ( Odocoileus virginianus ) and parasite loads in faecal samples within a hierarchical abundance‐mediated interaction model to test hypotheses regarding interactions between these cervids and their shared parasites (Parelaphostrongylus tenuis, Fascioloides magna ). We found that moose occupancy was limited by parasite‐mediated competition rather than direct competitive interactions between moose and white‐tailed deer. Such evidence of the importance of indirect interactions and resulting community outcomes is critical for species conservation under changing conditions.

1. Introduction

Direct and indirect species interactions play a crucial role in driving wildlife community structure and co‐existence patterns. While direct effects like competition, predation and herbivory have well‐established ecological impacts (Holt 1977; Bonsall et al. 2003), parasites and pathogens also play critical roles in many systems (Tompkins et al. 2003; Hatcher et al. 2006). Competing species can share parasites and pathogens, and impacts cannot be understood or accurately predicted without considering resulting indirect interactions (Holt and Bonsall 2017). Determining how direct and indirect effects shape community dynamics is pivotal for informing wildlife population management.

Parasites can shape ecological communities, substantively influencing host distribution, abundance and community structure (Hatcher et al. 2012). Ecological theory posits that parasites affect species coexistence via two main pathways: (1) in apparent competition, coexistence of hosts is destabilised by a shared natural enemy (Hudson and Greenman 1998). Without shared parasites, definitive and alternative hosts may not directly interact; however, parasites generate negative symmetric or asymmetric indirect effects on hosts (Holt and Bonsall 2017). (2) In parasite‐mediated competition, definitive and alternative hosts directly compete, and parasites mediate the outcome (Hatcher et al. 2006). Specifically, parasite‐mediated competition can drive community outcomes when differential virulence either (i) favours the dominant competitor and facilitates further advantage over the subordinate species, as observed in interactions between native and invasive Sciurid spp. (Santicchia et al. 2020) or (ii) reduces the competitive advantage of the dominant competitor, as observed in Anolis lizards (Schall 1992). Despite the theoretical foundation, empirical evidence on the role and strength of direct vs. indirect interactions in driving community dynamics is lacking.

A classic example of parasites altering extant interspecific competition is the infection of cervid hosts with meningeal worm (Parelaphostrongylus tenuis). White‐tailed deer ( Odocoileus virginianus , hereafter deer) are the definitive hosts of meningeal worm and suffer minimal morbidity from infection (Lankester 2010), while sympatric cervids such as elk ( Cervus elaphus ), mule deer ( Odocoileus hemionus ), caribou ( Rangifer tarandus ) and moose ( Alces alces ), are abnormal hosts that frequently experience severe neurological disease and death from infection (Samuel et al. 1992). Rapid onset of disease following infection by meningeal worm (Lankester 2002) has been hypothesised to alter the outcomes of competition, causing population declines (Schmitz and Nudds 1994) and limiting densities or co‐occurrence of other cervids with white‐tailed deer (Hatcher et al. 2006). Quantifying factors influencing species interactions in disease systems remains largely intractable because of complex and often unobservable interactions. Previous studies have addressed hypotheses related to parasite‐mediated competition through theoretical modelling (Schmitz and Nudds 1994), quantification of parasite loads (Tompkins et al. 2000), and associations with host species abundance or body condition (Irvine 2006). Nevertheless, the testing of theoretical models with empirical evidence from natural experiments may help disentangle host and parasite interactions across a range of species densities.

Individual investigations of direct competition, as well as parasite‐mediated competition, have greatly improved our understanding of the potential for parasites to impact community outcomes, yet joint examination of these mechanisms remains rare. In systems with deer, other cervids, and meningeal worm, early work argued for the role of apparent competition in mediating community outcomes (Anderson 1972). Alternative explanations have included type I parasite‐mediated competition, in which the dominance of deer over other cervid competitors is exacerbated by the parasite (Holt and Pickering 1985), and type II parasite‐mediated competition, wherein alternative cervid hosts are superior competitors over deer but parasites diminish or reverse this advantage (Price et al. 1988). Identifying the underlying mechanism and relative contributions of direct or indirect interactions requires empirical evidence from a system with these cervids in allopatry and sympatry, both in the presence and absence of shared parasites.

In northern New York, moose and deer co‐occur across gradients of shared parasite intensities. In this system, moose persist at low densities within the Adirondack Park (hereafter, Adirondacks; Hinton, Wheat, et al. 2022) and suffer deleterious health effects from meningeal worm and giant liver fluke ( Fascioloides magna ; Grauer 2024). Giant liver flukes fail to complete their life cycle in moose and can cause extensive organ damage and death given heavy infestations (Lankester and Foreyt 2011). Spatial overlap with white‐tailed deer exposes moose to both parasites through intermediate gastropod hosts (terrestrial for meningeal worm, aquatic for giant liver fluke) that become infected from parasites shed in deer faeces, causing moose morbidity and mortality that is hypothesised to limit moose densities and distribution in the state (Figure 1; Hinton, Wheat, et al. 2022). While deer historically moved seasonally in the Adirondacks to avoid areas of deep snow (Hurst and Porter 2008), their abundance and spatiotemporal overlap with moose have potentially increased with recent declines in winter severity and total snowfall (Hinton, Hurst, et al. 2022). Little empirical evidence exists on the strength or direction of direct competition between moose and deer (Schmitz and Nudds 1994), though moose are presumed to be competitively dominant due to adaptations to colder climates (Schwartz 1992) and their ability to forage higher in the canopy (Hodder et al. 2013). While black bears ( Ursus americanus ) and coyotes ( Canis latrans ) are present and may prey upon white‐tailed deer (Kautz et al. 2019), predation on moose is rare (Benson et al. 2017), and apex predators that could prey on moose have been extirpated from the region. Thus, while predator‐mediated competition may influence cervid dynamics, it does not contribute to spatial patterns or population limitation of moose in this system. Despite the absence of predators in the system, moose have failed to reach densities similar to neighbouring populations. The primary yet untested hypotheses underpinning moose limitation in the region are (1) direct competitive interactions with deer and (2) indirect interactions via parasite‐mediated competition.

FIGURE 1.

Hypothesised interactions between white‐tailed deer ( Odocoileus virginianus ), moose ( Alces alces ), and shared parasites (Parelaphostrongylus tenuis, Fascioloides magna ) that use deer as primary hosts and incidentally infect moose. Line thickness denotes strength of hypothesised interaction and dashed lines indicate interactions explicitly incorporated in model structure used in this study.

While this system provides a unique natural experiment of the relative contributions of hypothesised direct versus indirect processes, modelling approaches that address the ubiquitous issue of imperfect detection in large‐scale surveys of wildlife are needed. Hierarchical models, which explicitly model the observation process jointly with the state process of interest, are frequently used to account for imperfect detection in monitoring species distributions, abundances and interactions across scales (Kéry and Royle 2015). Species interactions have commonly been explored using hierarchical multispecies occupancy models (e.g., Rota et al. 2016), which consider the occupancy probability of one species conditional on the presence or absence of another. These models are expected to perform well when species interact strongly, such that the presence of one species is sufficient to exclude another (e.g., Twining et al. 2022). Nonetheless, interaction strength is frequently underpinned by population abundances. For example, parasite intensities can critically affect host demography (Balstad et al. 2021), and information on heterogeneity in intensities is lost when only prevalence is considered, potentially altering the conclusions of ecological inference (Brian and Aldridge 2021). A recently developed framework that incorporates species abundance, environmental factors, and interaction terms between species (Twining et al. 2025) facilitates formal testing of direct (e.g., competition) versus indirect interactions (e.g., parasite‐mediated competition) in driving species co‐existence.

Here we leverage a natural experiment and multi‐species sampling across a heterogeneous landscape, modelled using an abundance‐mediated interaction framework (Twining et al. 2025), to test whether the regional limitation of moose is due to direct interactions with deer or indirect interactions with shared parasites. We tested whether moose occupancy was best explained by: (1) the direct effect of an index of deer abundance (hereafter, abundance), (2) the indirect effect of deer via parasite‐mediated competition, with moose occupancy being negatively associated with parasite abundance or (3) habitat factors, including both biotic (timber harvest, forest cover) and abiotic (snow depth) landscape effects. We hypothesised that, due to adequate local habitat for moose (Peterson et al. 2022) and the potential for niche partitioning between deer and moose (Ludewig and Bowyer 1985), there would be limited direct effects of deer abundance on moose occupancy despite direct competition for resources. Rather, we hypothesised that moose occupancy would be negatively affected by deer indirectly via parasite‐mediated competition through local intensities of meningeal worm and giant liver fluke, altering the direction of interspecific competition in favour of white‐tailed deer.

2. Methods

2.1. Study Area

We selected a 4050 km² study area in the northern portion of the Adirondack Park in New York, USA that contained the core moose population in the state. Elevation in the area ranged from approximately 95 to 1430 m, with primary browse species including red maple ( Acer rubrum ), yellow ( Betula alleghaniensis ), grey ( Betula populifolia ) and paper birch ( Betula papyrifera ) and balsam fir ( Abies balsamea ; Peterson et al. 2020). We used K‐means clustering to distribute 105, 9‐km² grid cells representing the range of predicted moose and deer densities (0.1–0.4 moose/km², Hinton, Wheat, et al. 2022; 0.4–2 deer/km², Hinton, Hurst, et al. 2022) and available forest habitat cover types (Kramer et al. 2022). This grid size was selected as an intermediate between moose and deer home ranges [25 km² moose (Leptich and Gilbert 1989); 2 km² deer (Tierson et al. 1985)] for subsequent analysis of moose site use and deer abundance. Considering moose could be detected at multiple sites, model estimates of occupancy more accurately reflected moose site use, however we refer to occupancy throughout.

2.2. Camera Trap Surveys

We conducted camera trap surveys continuously from September 2021 to September 2023 using one camera per grid cell. We set camera stations within 100 m of randomly generated points within accessible areas, selecting sites near animal trails or cervid sign when possible. Placement was required to be at least 100 m from trails, roads and buildings to minimise human disturbance to camera sets. Remotely activated cameras (Reconyx HyperFire 2) were mounted ~1.5 m high on trees to maximise detection of cervids and minimise snow coverage. Cameras were set to medium sensitivity and captured five photos per trigger separated by a 15 s delay between triggers. We visited camera stations at least twice per year to replace lithium batteries and SD cards, trim vegetation, and restore any loss of camera function. All photographs were screened for the presence of moose and deer by at least two trained viewers. Periods of camera malfunction or blocked view were screened from analysis. We assigned species tags to images using digiKam (7.3.0) and processed image data using program R (version 4.2.1; R Core Team 2021) and package camtrapR (version 2.2.0; Niedballa et al. 2016). Detection data for each species (1 for detection, 0 for non‐detection) was pooled into weekly occasions.

2.3. Parasite Data Collection

From May to July 2022 we walked transects within each grid cell and collected deer pellet groups to quantify meningeal worm and giant liver fluke intensities in the samples and sites. Pellet transects, randomly initiated within 500 m of camera locations, consisted of four lengths of 100 m transects along two perpendicular connected lines each 400 m in length (Figure S1). Observers monitored 1 m on either side of the marked transect and collected pellet group samples in plastic freezer bags. We recorded transect length, observers, GPS location of each pellet group, relative condition of pellet group (fresh, average or dry), percent ground visibility, dominant land cover type, and whether it had rained within the last 24 h. We collected site‐level covariate data to account for heterogeneity in abundance and detectability of parasites through hierarchical modelling; however, initial model exploration demonstrated that Royle–Nichols models (Royle and Nichols 2003) were unsuitable due to extreme overdispersion in the parasite count data (Appendix S1). We collected multiple pellet groups at a single location along the transect if pellets varied in size (potentially indicating unique individual deer) but did not collect pellet groups within the next 10 m along the transect to reduce collection of pellet groups from the same individuals (Van der Waal et al. 2014). Pellet groups were chilled in the field and frozen within 8 h.

We quantified the intensity of giant liver fluke eggs in deer pellet samples using a modified quantitative faecal sedimentation (FlukeFinder Visual Difference, Moscow, Idaho, USA) and the intensity of meningeal worm larvae using a modified Baermann technique (Forrester and Lankester 1997). Eggs of giant liver fluke were distinguished from other species such as rumen flukes (Paramphistomum spp.) using colour and morphology (Bouvry and Rau 1984). It was not possible to distinguish eggs of F. magna from Fasciola hepatica due to challenges in species identification from egg morphology alone (Loginova et al. 2024). To confirm whether the dorsal spined nematode larvae detected were meningeal worm, we conducted PCR on a subset of samples and sequencing of the ITS‐2 region (Verocai et al. 2013). All parasitology tests were conducted at the Cornell College of Veterinary Medicine Animal Health Diagnostic Center. We used the numbers of each species per gram of faeces to quantify the total parasite burden at each site, representing cumulative exposure risk from multiple parasite species ( F. magna and P. tenuis ), and investigated how this aggregate parasite risk influenced moose site use. We explored the possibility of incorporating parasites as additional species within our interaction model (Appendix S2); however, insufficient data prohibited model convergence and necessitated the inclusion of parasite counts as a covariate for moose occupancy.

2.4. Abundance‐Mediated Interaction Model

We used an abundance‐mediated interaction model (Twining et al. 2025) to formally test whether occurrence of moose was mediated by (i) abundance of deer, (ii) intensities of parasites, (iii) environmental covariates or (iv) a combination of factors. Our formulation (Grauer and Fuller 2025) included a Royle–Nichols model to estimate mean site‐level deer abundance using detection/non‐detection data for deer from repeat surveys, where deer abundance N ^D at site i was drawn from a Poisson distribution of parameter ${\lambda }_i$. We estimated ${\lambda }_i$ as a function of site covariates using a log‐link:

$$N_i^{\mathrm{D}}~\text{Poisson}\left({\lambda }_i\right)$$

We used Poisson as the naïve distribution for deer assuming random distribution and group sizes that were not statistically overinflated (Hinton, Hurst, et al. 2022). We assumed that observations were independent and the site‐level population detection probability, p ^D, at site i on occasion j was a function of individual deer detection probability r ^D and the number of individuals at site i, $\left(N_i^{\mathrm{D}}\right)$:

$$p_{\mathit{ij}}^{\mathrm{D}}=1-{\left(1-r_{\mathit{ij}}^{\mathrm{D}}\right)}^{N_i^{\mathrm{D}}}$$

In the moose submodel we estimated the occupancy of moose, z ^M at site i as a Bernoulli random variable with probability ${\phi }^{\mathrm{M}}$, estimated as a function of site covariates using a logit‐link (MacKenzie et al. 2002):

$$z_i^{\mathrm{M}}~\text{Bernoulli}\left({\phi }_i^{\mathrm{M}}\right)$$

Moose detection/non‐detection data, y ^M at site i and occasion j was modelled as a function of site‐by‐occasion level detection probability $\left(p_{\mathit{ij}}^{\mathrm{M}}\right)$ conditional on presence at a site $\left(p_{\mathit{ij}}^{\mathrm{M}}|z_i^{\mathrm{M}}=1\right)$:

$$y_{\mathit{ij}}^{\mathrm{M}}~\text{Bernoulli}\left(p_{\mathit{ij}}^{\mathrm{M}}\times z_i^{\mathrm{M}}\right)$$

Finally, we incorporated directional interactions between deer, moose and parasites (Figure 1) by estimating moose occupancy as a function of (i) deer abundance $N_i^{\mathrm{D}}$, (ii) parasite intensities $C_i^{\mathrm{P}}$ and (iii) environmental covariates.

To determine the number of covariates that could be supported in our analysis, we ran simulations with data generated from distributions approximating the structure of our observed data (Appendix S2). These simulations supported inclusion of up to three covariates on each species state submodel and two covariates on each species observation submodel. We explored biotic and abiotic factors hypothesised to be important for the observation and state processes of both species (Table 1; Appendix S3). Specifically, we included the percentage of timber harvest (predicted removal and regeneration present in 2018; Kramer et al. 2022) in the state model for both species and predicted a positive relationship due to cervid reliance on these areas for forage. For deer abundance, we also included mature conifer cover and total weekly snow depth (National Operational Hydrologic Remote Sensing Center [NOHRSC] 2004). We predicted that conifer cover would provide winter refuge for deer, resulting in a positive relationship with deer abundance, while snow depth would negatively impact deer abundance because of constraints on movement and energetics (DelGiudice et al. 2013). For moose occupancy, we included mature deciduous cover and parasite counts from the transect surveys. We predicted that moose occupancy would increase with deciduous cover that is also used for forage (Peterson et al. 2020), while parasites would negatively impact moose occupancy because of deleterious health impacts and potential moose avoidance of high‐risk habitats (Escobar et al. 2019). All habitat covariates consisted of the percentage of specific land cover classes (Kramer et al. 2022) within a 2‐km radius around each camera location to characterise regional home range of deer (Tierson et al. 1985) and site use by moose. Observation models for both species included the number of operational camera days and ordinal date of the year. All covariates were centered on zero and scaled. Covariates were included in analyses if Pearson correlation coefficients were within ±0.6 (Dormann et al. 2013). We interpreted results from a single global model including all uncorrelated covariates.

TABLE 1.

Covariates testing interactions among moose ( Alces alces ), white‐tailed deer ( Odocoileus virginianus ), and two shared parasites ( Fascioloides magna ; Parelaphostrongylus tenuis) at camera trap locations in New York, USA between 2021 and 2023.

Species	Parameter	Covariate	Description	Prediction
Deer	Abundance	Timber harvest	Overstory and intermediate removal of forested land cover types approximately 5–18 years old	+
		Conifer	Mature conifer cover	+
		Snow depth	Total weekly snow depth in cm	−
	Detection	Ordinal date	Numbered (1–365) day of the year	+
		Effort	Operational camera days	+
Moose	Occupancy	Timber harvest	Overstory and intermediate removal of forested land cover types approximately 5–18 years old	+
		Deciduous	Mature deciduous cover	+
		Parasite	F. magna eggs and P. tenuis larvae per gram of deer faeces	−
	Detection	Ordinal date	Numbered (1–365) day of the year	+
		Effort	Operational camera days	+

Note: Predictions included whether the covariate was expected to be negatively (−) or positively (+) associated with the indicated model parameter. Forest cover covariates (timber harvest, conifer, deciduous) were calculated as percentage within a 2 km radius circle around camera locations.

As sampling occurred continuously over multiple years, thus violating modelling assumptions of closure, we used a stacked design (Kéry and Royle 2020), treating each season and year combination as a unique site. We defined seasons as calving (April, May, June, July), rut (August, September, October, November) and winter (December, January, February, March). To account for spatial dependence and artificially reduced standard error created by the stacked data structure, we estimated fixed effects (intercepts) for each of the six season‐by‐year combinations for both species in their respective state and observation models. We used uninformative priors for all parameters (Table S1). All modelling was conducted in C++ via the NIMBLE package (version 0.13.1, de Valpine et al. 2024). We ran three chains for 200,000 iterations with a thinning rate of five and burn in of 40,000, resulting in 10,000 posterior samples per chain. Model convergence was assessed using visual inspection of chain mixing and Gelman–Rubin diagnostics (convergence when $\widehat{r}<1.1$; Gelman and Rubin 1992). We used this model to formally test our hypotheses regarding direct abundance‐mediated interactions between moose and deer, parasite‐mediated competition between moose and deer, and environmental covariate effects on both cervids. We calculated 95% credible intervals (CI) for model posterior estimates and inferred biological importance for covariates whose CI did not include zero.

3. Results

3.1. Species Detections

We obtained detection/non‐detection data for deer and moose across 2 years at 620 sites. Total effort was 64,013 sampling days (24 h periods), which resulted in 3242 independent detections of deer and 167 independent detections of moose. Naïve occupancy was 0.86 for deer and 0.15 for moose. Deer and moose were jointly detected at 89 sites, deer were detected alone at 455 sites and moose were detected alone at 5 sites.

We collected and analysed 658 deer pellet groups for the presence of giant liver fluke eggs and meningeal worm larvae. Fascioloides magna were the only parasites identified during faecal sedimentation; the method may also detect amphistome eggs, though none were found. PCR identification of dorsal‐spined larvae was conducted on four samples previously collected within the same study area. All samples were positively identified as P. tenuis ; however, unconfirmed samples cannot be easily distinguished by morphology from Parelaphostrongylus andersoni, which could also potentially occur in the region. In total, 83 samples were infected with both parasites, 127 samples contained meningeal worm larvae only, 176 contained giant liver fluke eggs only and 272 samples contained neither. This translated to 85% of sites infected with giant liver flukes and 81% infected with meningeal worms. Counts of total giant liver fluke eggs per site ranged from 0 to 183.5, with a mean for all samples of 22.3 eggs per gram of deer faeces (SD = 34.9, median = 6.5). Counts of total meningeal worm larvae per site ranged from 0 to 198.8, with a mean of 12.2 per gram of deer faeces (SD = 27.6, median = 2.2). When pooled, joint parasite counts per gram of deer faeces ranged from 0 to 233.7, with a mean of 34.5 per site (SD = 44, median = 17.8), with 96% of sites infected by at least one parasite. Within sites where moose and deer co‐occurred, joint parasite intensities varied widely (range = 0.5–121.5, mean = 20.3, median = 8.8 per gram of deer faeces).

3.2. Abundance‐Mediated Interaction Model

All covariates were sufficiently uncorrelated and included in the hierarchical model. Visual inspection revealed acceptable mixing, and $\widehat{r}$ values were < 1.1, indicating convergence for each parameter. Posterior estimates (Figure 2; Table 2) indicated that deer abundance (Figure 3a,b) was positively associated with mature conifer cover (β = 0.07, 95% CI = 0.01, 0.12) and timber harvest (β = 0.16, 95% CI = 0.11, 0.22), and independent from snow depth (β = −0.09, 95% CI = −0.43, 0.24). Detection of individual deer increased with the number of functional camera days (β = 0.06, 95% CI = 0.01, 0.11) and ordinal date (β = 0.13, 95% CI = 0.07, 0.18). Moose occupancy was independent of deer abundance (Figure 4; β _deer = −0.11, 95% CI = −0.31, 0.07), but negatively associated with parasite abundance (β = −0.77, 95% CI = −1.19, −0.39). In addition, moose occupancy (Figure 3c,d) was positively associated with mature deciduous cover (β = 1.31, 95% CI = 0.90, 1.78) and timber harvest (β = 0.70, 95% CI = 0.30, 1.13). Moose detection probability was independent from ordinal date (β = 0.25, 95% CI = −0.03, 0.52) and the number of camera functional days (β = 0.13, 95% CI = −0.07, 0.41).

FIGURE 2.

Posterior estimates of factors influencing white‐tailed deer ( Odocoileus virginianus ) abundance (a) and moose ( Alces alces ) occupancy (b) including the interaction between deer abundance and moose occupancy. Points are mean estimates and error bars represent 95% credible intervals from hierarchical modelling of camera trap data in New York, USA collected from 2021 to 2023.

TABLE 2.

Output including β estimates and credible intervals (CI) for the occupancy‐abundance model exploring the effect of white‐tailed deer ( Odocoileus virginianus ) abundance on moose ( Alces alces ) occupancy at camera traps in New York, USA between 2021 and 2023.

Species	Parameter	Covariate	β	Lower 95% CI	Upper 95% CI
Deer	Abundance	Timber harvest	0.16	0.11	0.22
		Conifer	0.07	0.01	0.12
		Snow depth	−0.09	−0.43	0.24
	Detection	Ordinal date	0.13	0.07	0.18
		Effort	0.06	0.01	0.11
Moose	Occupancy	Timber harvest	0.70	0.30	1.13
		Deciduous	1.31	0.90	1.78
		Parasite	−0.77	−1.19	−0.39
		Deer abundance	−0.11	−0.31	0.07
	Detection	Ordinal date	0.25	−0.03	0.52
		Effort	0.13	−0.07	0.41

FIGURE 3.

Predicted relationships between significant factors included in modelling white‐tailed deer ( Odocoileus virginianus ) and moose ( Alces alces ) direct and indirect interactions from camera trap data collected in New York, USA from 2021 to 2023. Presented are the effects of mature conifer cover (a) on deer expected abundance, timber harvested forest cover (b) on deer expected abundance and moose probability of occupancy, and parasite abundance (c) and percent mature deciduous cover (d) on moose probability of occupancy. Shading and dashed lines represent 95% credible intervals around state variable estimates.

FIGURE 4.

Conditional estimates of moose ( Alces alces ) occupancy as a function of white‐tailed deer ( Odocoileus virginianus ) abundance with (a) zero shared parasites (Parelaphostrongylus tenuis, Fascioloides magna ) and (b) conditional estimates of moose occupancy as a function of deer abundance when parasites were high across the study area in New York, USA. White‐tailed deer abundance had no significant effect on moose occupancy, and parasite counts were significant and negatively correlated with moose occupancy.

4. Discussion

We provide empirical evidence of parasite‐mediated competition underpinning outcomes of species interactions using a combination of field observation, laboratory testing and hierarchical modelling that accounts for imperfect detection. High abundances of meningeal worm and giant liver fluke were associated with lower moose occupancy probabilities, likely driven by parasite effects on moose morbidity and mortality and resulting in the exclusion of moose from sites with high parasite abundances. Concurrently, we observed independence of moose occupancy with deer abundance, demonstrating that moose persist, at least presently, in areas of overlap with white‐tailed deer, and experience minimal population‐level impacts from direct competition with deer. The lack of a significant association between deer abundance and moose occupancy, coupled with a strong negative association between parasite abundance and moose occupancy, suggests that parasites rather than direct competition mediate the interaction between deer and moose in this system. Continued monitoring of both cervid populations could help discern if parasite‐mediated exclusion of moose will be exacerbated by changing conditions that potentially increase parasite infection risk for moose (Feldman et al. 2017), and if increasing overlap with white‐tailed deer (Dawe and Boutin 2016) and subsequent pressure from parasites will alter the southern limits of moose distribution.

Considering the approximate 20% reduction in moose occupancy at sites with average parasite numbers, and 90% reduction at sites with the highest parasite numbers, the persistence of moose in New York state is likely contingent upon parasite‐mediated competition and potential alterations to parasite abundance in response to changing conditions, rather than changes in direct competition with deer. Parasite responses may depend upon the co‐occurrence and abundance of deer and gastropod hosts, and environmental factors impacting the reproduction and survival of parasites at various life stages (Lankester 2018). While white‐tailed deer are necessary hosts in the life cycles of meningeal worm and giant liver fluke, parasite loads may be frequency dependent rather than density dependent (Ryder et al. 2007). Other factors such as intermediate host population dynamics (Martinez and Merino 2011), microclimate conditions and intermediate host habitat suitability (Escobar et al. 2019) or parasite dynamics within hosts (Ezenwa and Jolles 2011; Vannatta and Moen 2016) may be more important in determining parasite numbers on the landscape. Specifically, longer growing seasons with wet, mild conditions are predicted to favour gastropod intermediate hosts and increase meningeal worm infections in white‐tailed deer (Lankester 2018). Increased wetland use has been observed to increase giant liver fluke intensities in elk (Normandeau et al. 2020). Warmer, wetter predicted conditions in the Adirondacks (Lamie et al. 2024) may increase moose exposure risk by increasing gastropod populations and moose thermoregulatory use of aquatic habitats where they can acquire fluke infections. Frequency of moose site use in areas of parasite contamination is likely a critical driver of infection risk, independent of direct temporal overlap with white‐tailed deer (McGraw et al. 2021). Additionally, the monotonic decline in moose occupancy with increasing parasite numbers indicates potential compounding deleterious effects of parasite infection for moose (Lankester 2010) along with higher probabilities of infection. The combined effects of parasite dynamics and climate change will likely be critical in shaping moose responses to the northward expansion of white‐tailed deer.

Despite similar responses to timber harvest and high rates of co‐occurrence between deer and moose, our models provided evidence of limited population‐level effects of direct interactions between the two cervids. Body size, physiology and feeding height in cervids may function as means of ecological character displacement (Dayan and Simberloff 2005) that potentially minimise impacts of direct competition. This type of competition, whereby moose overlap deer in resource use but also exploit resources not available to deer, is known as inclusive niche competition. A similar dynamic has been observed in competitive interactions between moose and snowshoe hare (Belovsky 1984). Moose and deer may differentiate in diet and space based on winter severity and snow depth (Hurst and Porter 2008; Ratkiewicz et al. 2024), though competition is expected to increase during severe winter conditions (Jenkins and Wright 1988). Both species are browsers that exhibited positive associations with forests undergoing timber management, which is limited within the Adirondacks and likely drives the high spatial overlap observed in our study area. Additionally, an increasing frequency of milder and shorter winters is expected to perpetuate the northward shift in white‐tailed deer distribution and increase overlap with conspecifics (Weiskopf et al. 2019). While moose may be the presumed dominant competitor in direct interactions because of larger body size (Ferretti and Mori 2020), we demonstrate that the outcome of competition shifted in favour of deer due to indirect interactions mediated by parasites. Although parasite‐mediated and predator‐mediated competition could both lead to reduced moose occupancy, the region lacks major shared predators that could induce predator‐mediated apparent competition, strengthening confidence in our evidence that parasites are the primary driver of the patterns observed.

Use of the abundance‐mediated interaction model (Twining et al. 2025) enabled us to leverage multiple years of host detection data and samples of parasite infectious stages to test the strength of both direct and indirect interactions that may have been missed using occupancy‐based interaction frameworks. Our results revealed ecological interactions important for effective management in a system where disease was known to be present but heretofore underappreciated with respect to its population‐level impacts. Deer were widespread across the study area and group size varied ($\overline{x}=3.21$ deer per group, SD = 2.89; Hinton, Hurst, et al. 2022), which invalidated typical multi‐species frameworks that model interactions as a function of occupancy. Incorporating the effect of deer abundance and relative parasite abundance on moose was critical in testing hypotheses of abundance‐mediated species interactions while accounting for habitat and environmental effects. We detected a weak association between deer abundance and conifer cover, likely diverging from historical patterns (Hurst and Porter 2008) due to increasingly mild winter conditions. Moose are frequently associated with areas of timber management for sources of protein‐rich forage (Peterson et al. 2020), and while we saw a strong positive association with higher proportions of timber areas, we found an even stronger association with mature deciduous cover. Mature forests could be important sources of intermediate levels of both forage (Peterson et al. 2022) and thermal cover (Dussault et al. 2004), potentially exposing moose to increased risk of parasite exposure if both cervid hosts are selecting for these habitats.

Quantifying parasite‐mediated competition remains challenging due to complex parasite life cycles, understudied intermediate hosts, and difficulty in quantifying host infections non‐invasively (O'Brien et al. 2022). Ample opportunity across disease systems exists to test direct species relationships between parasites and detrimentally impacted host species while accounting for imperfect detection. Extensions to other systems include ungulates such as caribou ( Rangifer tarandus ) and muskox ( Ovibos moschatus ) that share abomasal nematode parasites (Hughes et al. 2009), and systems where pathogens or predators rather than parasites mediate species interactions. Additionally, future studies could disentangle the overall direction of effect resulting from the multiple ways that natural enemies alter interactions within communities (Hatcher et al. 2012; e.g., negative direct effects but positive indirect effects). Understanding the outcomes of parasite‐mediated competition and other interspecific interactions is critical to anticipating and helping ensure species coexistence under a range of environmental conditions.

In our study, temporally limited sampling of parasites and low overall detections of moose precluded our ability to incorporate parasites as their own submodel within the abundance‐mediated interactions framework. Our simulations demonstrated that accurate estimates for species interaction terms between white‐tailed deer and parasites could not be estimated, likely due to high sampling requirements needed to support interaction terms and extreme overdispersion in the parasite data. However, the flexibility of the model structure allowed us to incorporate parasite counts as a covariate on moose occupancy to explore the indirect effect of deer on moose. While we sampled parasites in only one season, our use of a stacked single‐season modelling framework including parasite counts across seasons allowed us to evaluate potential delayed impacts of parasite intensities. Parasite infectious stages may persist in intermediate hosts or the environment beyond a single season, suggesting that site‐level parasite abundance reflects cumulative exposure risk rather than a strictly seasonal phenomenon. Nevertheless, seasonal shifts in moose and deer movements could influence spatial overlap and transmission risk, and future studies could sample parasite burdens and host site use across multiple years to refine these linkages. Due to the impact of both parasite species on moose in New York (Grauer 2024), we used pooled parasite data in the analysis. Continued sampling of both giant liver fluke and meningeal worm over time, along with exploration of environmental factors important for intermediate hosts, could expand our ability to differentiate the effects of individual parasites or use multi‐season frameworks.

As changing conditions have the potential to modify both interspecific competition and threats from parasite‐mediated competition, future monitoring will be critical to assess whether interactions between hosts, parasites, and climate drive the southern range contractions of incidental hosts such as moose (Feldman et al. 2017). Fine‐scale information could inform whether incidentally infected hosts are able to persist through their use of habitat refugia that lack parasites (Nudds 1990), and what cues, behaviours or physiological responses allow the minimisation of infection or disease. Similarly, the ecology of intermediate hosts and within‐host parasite dynamics remain key areas of study in understanding the eco‐evolutionary impact of disease. We demonstrated how the use of hierarchical models applied to large empirical datasets can aid in the investigation of these questions and allow researchers to explicitly test hypotheses on important interspecific interactions.

Author Contributions

J.A.G., J.P.T. and A.K.F. conceptualised the study. J.A.G. collected data and wrote the original draft. J.A.G., J.P.T. and M.L. conducted formal analysis. A.K.F., J.L.F. and K.L.S. acquired funding. All authors contributed resources and edited the manuscript.

Peer Review

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer‐review/10.1111/ele.70159.

Supporting information

Appendices S1–S4.

Grauer, J. A. , Twining J. P., Lejeune M., et al. 2025. “Parasite‐Mediated Competition Limits Dominant Cervid Competitor.” Ecology Letters 28, no. 6: e70159. 10.1111/ele.70159.

Data Availability Statement

All data and code used in this study are available at https://doi.org/10.5066/P13MZ4WZ (Grauer and Fuller 2025).