Landscape and tree community effects on the infestation of an endemic fungal pathogen in riparian forests were mediated by leaf traits

Plant pathogens are regarded as crucial agents shaping the dynamics of natural forest communities. Marssonina leaf spot of poplar is induced by an endemic pathogenic fungus Drepanopeziza populi, causing increased damage to riparian poplar stands in recent years. However, such endemic fungal diseases have received little attention at the landscape scale, despite the key role of landscape heterogeneity in the development and spread of emerging forest diseases. Moreover, most studies have insuciently captured multiple ecological factors driving the infestation of an endemic pathogen acting at the landscape, community, and individual scales. We measured pathogen load, disease prevalence, and disease severity of Marssonina leaf spot in poplars in riparian forests. We explored the direct and indirect effects of multiple ecological factors on pathogen infestation using a path analysis. Specically, we rst assessed the effects of landscape and community factors on leaf traits including leaf area, specic leaf area (SLA) and leaf dry matter content (LDMC), and then examined the role of these factors in shaping disease dynamics.


Results
Path analysis showed that landscape features had no direct impact on leaf traits and pathogen infestation, but directly affected tree community composition. Landscapes with higher forest cover resulted in higher host density and tree diversity. Host density was the most important factor of pathogen load, with higher host density resulting in more symptomatic leaves. Tree diversity had direct effects on disease prevalence, with poplars growing in mixed forest stands far less affected by pathogens than in pure stands. Moreover, disease prevalence was positively related to pathogen load. Tree diversity strongly reduced SLA, but increased LDMC. Higher SLA was found to increase pathogen load and disease severity, but higher LDMC was found to reduce both of them.

Conclusions
Our results show that the effects of landscape and tree community on Marssonina leaf spot disease are mediated by leaf traits. Disentangling the effects of biotic and abiotic factors affecting pathogen infestation contributes to reduce the overall impact of this disease, which can provide policy makers with sustainable management of endemic plant diseases in natural forests.

Background
Loading [MathJax]/jax/output/CommonHTML/jax.js Plant pathogens play an important role in shaping natural forest ecosystems, and most pathogens of forest trees have co-evolved for eons with their hosts, reaching a sort of balance between them and populations of susceptible tree species (Castello et al. 1995; Burgess and Wing eld 2017). In this sense, endemic fungal pathogens can affect many forest ecosystem processes and contribute to maintaining the dynamics and diversity of forests by reducing abundance of host trees or by altering competition outcomes among host and non-host tree species (Hansen 1999;Holdenrieder et al. 2004; Paap et al. 2018). However, anthropogenic activities and climate changes are enhancing occurrence and severity of certain endemic plant diseases, which are either the type not normally thought of as highly virulent or are those that have not been previously reported as a serious problem on a particular host (Sturrock et  Ecologists have increasingly focused on the reciprocal interactions between spatial patterns and ecological processes at a landscape scale over the past 30 years (Turner 1989(Turner , 2005. But this approach has not been applied su ciently to forest pathology of endemic diseases, as most ecologists were interested in landscape epidemiology of emerging infectious diseases, aiming to determine how spatial patterns shape disease severity (Castello et al. 1995;Condeso and Meentemeyer 2007;Plantegenest et al. 2007;Meentemeyer et al. 2012). The spatial composition and con guration of host vegetation is of great importance for pathogens affecting long-lived hosts such as forest trees (Condeso and Meentemeyer 2007), and thus landscape features have been increasingly recognized as important factors of pathogen spread and severity in a spatially heterogeneous environment (Holdenrieder et al. 2004;Ferrenberg 2016;Cobb and Metz 2017). In this way, landscape features not only affect the spread of invasive pathogens (Meentemeyer et al. 2012;Grosdidier et al. 2020), but are crucial to disease expression of endemic pathogens. A landscape-scale study, for instance, revealed that local biotic and abiotic processes contribute to the incidence of an endemic canker pathogen in marri in Australia (Paap et al. 2017a).
Within habitat patches, fungal pathogen infestation on host trees is affected by tree community composition which could change focal tree's physical and chemical apparency and anti-pathogen defences (Hantsch et  tree community effects on fungal pathogens are crucial to our understanding of multitrophic interactions and biodiversity functioning (Hansen 1999;Scherber et al. 2010), only a few studies on forest disease have explicitly assessed the role of tree diversity and density effects on endemic pathogen infestation (Benítez et al. 2013;Nguyen et al. 2016). Furthermore, landscape features and land-use changes are known to affect host density by in uencing where hosts aggregate, and the aggregation of tree hosts is particularly determined by landscape structure since trees don't move (Meentemeyer et al. 2008 Here, we tested the direct and indirect effects of landscape features, tree diversity, host density and leaf traits on pathogen infestation using Marssonina leaf spot of poplars as a case study of an endemic fungal disease. Marssonina leaf spot is induced by Drepanopeziza populi (formerly Marssonina populi), a specialist foliar pathogen, and is one of the most widespread and severely damaging diseases of poplars. This pathogen infection is more commonly known as 'black leaf spot' due to the way that it looks. Populus laurifolia Ledeb is the most commonly infected poplar host in study area, but other poplar Loading [MathJax]/jax/output/CommonHTML/jax.js species (i.e., Populus alba L.) are resistant to suffering from this pathogen. We hypothesized that biotic factors (i.e., tree diversity and host density) would affect disease prevalence/severity more strongly than abiotic landscape features, but their effects on pathogen infestation were mediated by leaf traits, i.e., leaf area, SLA, and leaf dry matter content (LDMC). Speci cally, we test the following hypotheses: (1) greater density of poplar hosts (P. laurifolia) will generate greater pathogen load and directly increase disease prevalence; (2) disease prevalence in poplars will be lower in mixed stands than in pure forest stands; (3) leaf area and SLA will have a positive effect on pathogen load and disease severity of D. populi, but LDMC will have a negative effect on pathogen load and disease severity; (4) landscape features (e.g., forest cover) will indirectly affect pathogen load, disease prevalence and severity by in uencing host density, tree diversity and leaf traits. By elucidating the relative effects of biotic and abiotic factors on pathogen infestation, our results will offer key insights into risk predictions and control strategies for damaging endemic diseases.

Study site and plot selection
This study was conducted in natural forests along the Irtysh River in Altay region of the Northern Xinjiang, Northwest China (Fig. 1). This region, a typical ecoregion with a temperate continental climate, contains a large number of forest patches, which are generally dominated by P. laurifolia, accompanied by other tree species such as P. alba. Poplar trees (Populus spp.) are one of the most important components of riparian ecosystems throughout most of the North China, shaping the environment of special habitats and providing many ecosystem services (Philippe and Bohlmann 2007; Wang et al. 2021). But in recent decades, the region has undergone extensive degradation of native forest resulting in a highly fragmented landscape due to the enlarged farmland size and the increase of grazing intensity.
The special climate conditions (e.g., drought, high temperature, and large temperature difference in the growing season) within study area also threaten the survival and regeneration of native forest trees. In addition to the anthropogenic disturbances and environmental changes, a wide variety of endemic insect herbivores and fungal pathogens had adverse effects on dominant poplar species in this region (Wang et al. 2021). In recent years, there has been a decline in the health of P. laurifolia associated with Marssonina leaf spot, though tree death is rare unless there are several consecutive years of other stresses such as drought and Cytospora canker. The disease cycle of D. populi takes one year, and its incidence rate is highest after high temperatures and periods of precipitation when spores are generally available during August every year.
To identify suitable locations for study sites in areas with known tree presence and disease infection, a desktop study was conducted based on recent satellite images (European Sentinel-2A, from 2020). Potential study sites were visited and visually assessed to con rm suitability. A ground survey was then carried out in 2020 on sites with tree presence to check for the presence of host poplars, and thus a total of 30 study sites within P. laurifolia range were found to be appropriate locations for the eld survey. We Loading [MathJax]/jax/output/CommonHTML/jax.js established a 30 × 30 m sampling plot in each of 30 study sites harbouring one of two forest stand types: pure stands (n = 15 plots) in which P. laurifolia was the main species, and mixed stands (n = 15 plots) consisting of P. laurifolia and P. alba where the former was abundant (Fig. 1). Plot locations were spaced at least 600 m apart to avoid sampling at the spatial scale within which pathogen infection is known to be clustered (Wang et al. 2021). Furthermore, plots were located in forest stands with varying levels of forest cover across the surrounding landscapes, and the remaining landscape is characterized by grasslands, shrublands, agricultural areas and human settlements.

Landscape and tree community characteristics
Around each woodland plot, we quanti ed landscape heterogeneity within a circular buffer of 300 m radius using a sample site-landscape approach (Fahrig 2013). We selected a 300 m radius area to measure landscape features because previous research had shown that spatial pattern of host vegetation most strongly predicted D. populi disease level at this scale (Wang et al. 2021). This radius distance was also chosen to avoid sampling overlap between adjacent plots, which is needed to make accurate landscape-scale inferences (Eigenbrod et al. 2011). We classi ed land-cover types into poplar woodland, farmlands, grasslands, shrublands, and human settlements by processing remotely sensed data acquired from the Sentinel-2A satellite in July 2020. We de ned human settlements as buildings, roads, and man-made streams. Analysis of landscape heterogeneity requires quanti cation of the con guration of land-cover types, and a large number of landscape metrics have been developed for this purpose. However, no single metric captures all information about landscape pattern, and strong correlations between many metrics make choosing an appropriate metric di cult. Thus, we quanti ed total area (i.e., forest cover) and perimeter : area ratio (PARA) of poplar woodland using Fragstats 4.2 to describe the woodland size and what degree trees were isolated or within a tree stand. Furthermore, the distance of each plot from the river (i.e., distance to river) was computed from remotely sensed data in ArcGIS 10.5 and de ned as the shortest linear distance between the plot edge and the river bank.
All trees within each plot with a diameter at breast height (DBH) > 5 cm were measured, and the species identity was recorded to determine tree species composition (i.e., pure or mixed forest stands). We also estimated poplar host density as the number of P. laurifolia stems per hectare based on the number of stems with a DBH > 5 cm rooted in the plot area.

Leaf sampling and measurements
As small brownish spots appear on infected leaves in early August, a eld survey was undertaken during the peak D. populi symptom expression between middle August and early September in 2020 to measure pathogen load, disease prevalence and severity in P. laurifolia trees within each 30 × 30 m plot. First, we recorded the number of P. laurifolia trees with symptomatic leaves, and calculated disease prevalence (proportion of symptomatic stems) by dividing symptomatic stems by the total number of P. laurifolia trees within each plot. Given that symptomatic stems in study area exhibited distinctly different levels of D. populi infection, assessing an individual as only infected or uninfected was deemed too coarse to examine variations in disease severity among different plots. Thus, ve symptomatic stems were Loading [MathJax]/jax/output/CommonHTML/jax.js selected randomly in each plot, with locations being chosen in the central area of each plot to avoid edge effects. Four branches were sampled growing in opposite directions from the lower and upper-tree layer of each individual (Hantsch et al. 2014a), and 30 leaves were collected per branch. We quanti ed pathogen load by counting the number of symptomatic leaves out of 120 leaves on each individual. Then, ve leaves were haphazardly chosen on each branch, and the percentage of leaf area infected by pathogen was visually estimated and classi ed into seven classes: (1) 0%, (2) 1-5%, (3) 6-10%, (4) 11-25%, (5) 26-50%, (6) 51-75%, (7) 76-100%, after previous authors (Hantsch et  Due to the lack of theoretical and empirical evidence of multiple drivers of this endemic disease, the rst step of our analyses was intended to determine if changes in landscape, tree community characteristics and poplar leaf traits could result in differences in pathogen infestation. In a preliminary step, we tted separate linear mixed-effect models (LMM) to identify the effects of landscape and tree community characteristics on leaf traits as well as their effects on pathogen load, disease prevalence, and disease severity. Fixed effects were forest cover, PARA, distance to river, host density, and forest type, and additional xed effects (leaf area, SLA and LDMC) were added to the full models of pathogen load, disease prevalence and disease severity. Plot identity was assigned as a random factor to account for the non-independence of samples from the same plot. For each response variable, we rst built the full model and then applied model simpli cation by sequentially removing non-Loading [MathJax]/jax/output/CommonHTML/jax.js signi cant terms to determine the best-tting model via Akaike's Information Criterion corrected for small sample size (AICc) (Grueber et al. 2011). Finally, we estimated model coe cients of the simpli ed model and calculated marginal R 2 and conditional R 2 for xed effects (R 2 m ) and xed plus random effects (R 2 c ). Response variables were log-transformed to satisfy the assumptions of normality.
Predictor variables with signi cant coe cients were used in SEM to determine whether landscape and tree community effects act directly or indirectly through changes in leaf traits. Using the results of preliminary analyses (Table S1) and published knowledge of endemic foliar fungal disease dynamics, we rst developed a structural model describing hypothesized direct and indirect relationships between forest cover, tree community, leaf traits, and disease (Fig. 2) and assessed this model structure using path analysis (Shipley 2009). We considered forest cover as exogenous predictors. Subsequently, tree diversity, host density, and two leaf traits (i.e., SLA and LDMC) served as endogenous predictors. We then used the piecewiseSEM package to evaluate our model because it permits the inclusion of hierarchical data by piecing multiple mixed-effect models into one causal framework (Lefcheck 2016). Based on the preliminary analyses, we modeled all response variables as normally distributed. We assessed the overall t of the initial piecewise SEM using directional separation (d-separation) tests, which determines the probability of informative path missing from hypothesized network (Shipley 2009). Models were considered rejected if χ 2 -test of Fisher's C-statistic fell below the signi cance level (P < 0.05), indicating that the model is inconsistent with the data. Finally, the standardized path coe cients (β) and p values were used to assess the signi cance of individual predictors within the nal model.

Results
Overall, 67% of the sampled leaves (n = 18000 leaves) showed signs of pathogen damage, with an average of 14.33% ± 6.93% (mean ± SD) of infected area per leaf. Biotic factors of tree diversity, host density, and leaf traits had stronger effects on pathogen infestation than abiotic factors, and abiotic landscape structure signi cantly in uenced tree community characteristics (Fig. 3). Forest cover at the landscape scale directly increased tree diversity and host density (Fig. 4a), but had no direct effects on pathogen infestation. The results of d-separation test indicated that only leaf traits directly affected disease severity. SLA had a positive effect on disease severity (Fig. 4b), whereas LDMC had a negative effect (Fig. 4c). And we found that disease prevalence was most strongly affected by the biotic factors of tree diversity, but also was in uenced by pathogen load. However, they had opposite direct effects on disease prevalence with greater tree diversity resulting in lower disease prevalence and greater pathogen load resulting in greater disease prevalence (Fig. 4d). Host density had the strongest direct effect on pathogen load (Fig. 4e), but pathogen load was also in uenced directly by leaf traits with greater SLA resulting in greater pathogen load (Fig. 4f) and greater LDMC resulting in lower pathogen load (Fig. 4g). Loading [MathJax]/jax/output/CommonHTML/jax.js Table 1 Standardized estimates of path coe cients from the tted path model in Fig. 3 for pathogen load, disease prevalence and disease severity. Estimate, estimated path coe cient; SE, standard error; pvalue, p-value of the estimated coe cient; Std. Estimate, standardized estimate of the path coe cient; SLA, speci c leaf area; LDMC, leaf dry matter content. Signi cant effects are in bold text. prevalence was also positively related to pathogen load (β = 0.15). Finally, disease severity was positively related to SLA (β = 0.46) and negatively to LDMC (β = −0.46) ( Fig. 3; Table 1).

Discussion
Our study is the rst to analyse the effects of landscape and tree community characteristics on important leaf traits and to explore their role in driving endemic pathogen damage in a riparian forest ecosystem. We demonstrated that pathogen infestation in the Marssonina leaf spot disease system was more strongly affected by tree community characteristics and leaf traits, but tree community was shaped by abiotic landscape features, i.e., plots with higher surrounding forest cover had higher host density and tree diversity. Higher host densities increased pathogen load, which resulted in greater disease prevalence in susceptible poplars, but this was countered by a relatively strong negative effect of tree diversity. Tree diversity also reduced SLA but increased LDMC. Leaf traits simultaneously in uenced pathogen load and disease severity, with SLA increasing both of them, but LDMC reducing both of them.
We found that, although host poplar density was the strongest driver of pathogen load (symptomatic leaf count) in the plot, there is no strong support for another part of our rst hypothesis, which stated that an increase in host density of P. laurifolia will result in an increase in the degree of disease prevalence. Several studies indicated that individual variation in host tree susceptibility can be obscured at the plot scale by tree community characteristics, reinforcing our ndings of host density effects on pathogen load , the detected absence of host density effects on disease prevalence was unexpected. As our results regarding host density effects oppose the previous ndings on density dependency of infection rates of fungal pathogens, it seems that local host density does not affect the degree of disease prevalence through declining resource availability but rather through modifying the pathogen load.
Our second hypothesis, stating that disease prevalence on susceptible poplars will be lower in mixed than in pure forest stands, could largely be con rmed. The negative relationship between tree diversity and disease prevalence is evidence for the dilution effect, i.e., species diversity reduces disease risk, which was found in many plant disease systems ( Increasing tree diversity could reduce the encounter rate between dispersed spores and susceptible trees. Interestingly, the dilution effect did not extend to pathogen load on host poplar, and this may be because a single symptomatic host tree with thousands of leaves could support a very large pathogen load even as the only host tree among non-host trees. The magnitude of the dilution effect may be in uenced by variations in individual leaf traits, but we did not detect obviously in our models. The effects of local tree diversity on fungal infestation in our study may also be subsumed in associational resistance hypothesis (Field et al. 2020), which states that a diverse tree community reduces pathogen development due to structural heterogeneity, especially through a higher microclimatic heterogeneity. As microclimatic properties can be affected by varying canopy size and structure of neighbouring plant species in a particular tree community, diverse tree size may be considered as one of the most important factors determining pathogen development (Jactel et al. 2009;Calonnec et al. 2013;Hantsch et al. 2014a). Our study is the rst to demonstrate associational resistance to endemic fungal pathogens by tree species in natural riparian forests at the stand level, and reinforces the previous nding in the foliar disease system by using a real-world data set from spatially diverse plots with different forest stand types.
Alongside tree community characteristics, leaf trait is also an important driver of the disease dynamics. This study demonstrated a positive effect of SLA on pathogen load and disease severity and a negative effect of LMDC on them. These correlations could be largely explained by leaf morphology. As tree leaves with low SLA and high LMDC contain more structural carbohydrates, such as lignin and cellulose in their cell walls (Mediavilla et al. 2008), these leaves may be not colonized by spot fungi as easily as leaves with thinner cell walls. In contrast, it can be speculated that leaves with high SLA, low LMDC and thin cell walls contain more non-structural carbohydrates in their cell sap, which are very important nutrients for obligate plant pathogens (Berger et al. 2007). Once infection occurs, these carbohydrates and others suitable nutrients (e.g., nitrogen and amino acids) are easily available for pathogens and therefore will promote disease severity in leaves with high SLA and low LMDC (Toome et al. 2010). Additionally, local tree diversity had strong effects on SLA and LMDC. Since host trees are the tallest species in this study, mixed stands have a higher proportion of shorter tree species. Thus, host trees surrounded by short heterospeci c neighbouring trees experience the higher light intensities and produce leaves with lower SLA, and they even experience the unfavourable conditions with water stress and produce leaves with higher LDMC (Muiruri et al. 2019). However, within pure stands, higher proportion of taller host trees may shape a shady environment, and shaded leaves found lower in the host canopy have larger SLA and lower LMDC (Legner et al. 2014) and are preferred by fungal pathogens. Therefore, variations in leaf traits with tree diversity as a result of reduced canopy cover around host tree may mediate tree community effects on this endemic fungal disease.
Predicting spatial dynamics of disease spread and severity in forest ecosystems requires knowledge of not only tree community characteristics and individual leaf traits, but also the spatial arrangement and disease expression has been shown to be in uenced by landscape features including the spatial structure of host populations (Meentemeyer et al. 2008), as well as spatial heterogeneity in the abiotic environment including topographic and climate effects (Haas et al. 2016;Dillon et al. 2019). However, we found that landscape characteristics had no direct in uence on pathogen infestation. Our path analysis revealed that locations with greater forest cover surrounding a plot indirectly reduced pathogen load by having lower host density, and exerted an indirect negative effect on disease prevalence via tree diversity. The in uence of land-cover changes on plant-pathogen interactions in forests has rarely been studied, but a previous research suggested that landscape features such as woodland cover in uences disease spread through forest structure, host density, and understory microclimate (Meentemeyer et al. 2008

Conclusions
Although indirect effect of tree community characteristics on leaf damage in forests was also found in previous studies, our unique contribution is that we demonstrated that tree community-pathogen infestation relationship is most probably mediated by leaf traits in a landscape-scale study. Despite the complex nature involved in these ecological processes, and several limitations for controlling many ecological variables shaping disease expression in natural patches across multiple local and landscape scales, we demonstrates that forest cover surrounding host trees affects pathogen infestation indirectly, through the variations of tree community composition and the associated changes in the poplar leaf traits. As landscape-scale changes can limit the growth and regeneration of plant assemblages, we would expect that forest remnants in landscapes with different forest cover would have different successional trajectories, potentially leading to divergence in plant species composition and leaf traits. More research on leaf traits accounting for functional differences between forest stands under different landscape-scale context can improve our understanding of tree diversity-disease resistance relationships, and enhance our ability to predict and reduce pathogen infestation of endemic plant diseases in natural forests.

Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests
The authors declare that they have no known competing nancial interests or personal relationships that could have appeared to in uence the work reported in this paper.  Conceptual path model describing the hypothesized causal relationships between factors in uencing pathogen infestation in the Marssonina leaf spot disease system. Positive and negative hypothetical pathways are indicated by "+" and "−" near the arrows, respectively. SLA, speci c leaf area; LDMC, leaf dry matter content.

Figure 3
Loading [MathJax]/jax/output/CommonHTML/jax.js Standardized coe cients for each relationship in the path model. Positive and negative pathways are indicated by blue and red lines, respectively. Dotted lines are statistically insigni cant at p < 0.05. The coe cient of determination (R2) is shown in green circles for all response variables (i.e., host density, tree diversity, SLA, LDMC, pathogen load, disease prevalence, and disease severity). Arrow thickness is scaled to the absolute value of the standardized path coe cient. SLA, speci c leaf area; LDMC, leaf dry matter content. Signi cant effects are in bold text.

Figure 4
Plots of the predicted relationships (smooth line) for signi cant paths between response and explanatory variables from tted models of the path analysis in Figure 3;