Host evolutionary relationships explain tree mortality caused by a generalist pest–pathogen complex

Abstract The phylogenetic signal of transmissibility (competence) and attack severity among hosts of generalist pests is poorly understood. In this study, we examined the phylogenetic effects on hosts differentially affected by an emergent generalist beetle–pathogen complex in California and South Africa. Host types (non‐competent, competent and killed‐competent) are based on nested types of outcomes of interactions between host plants, the beetles and the fungal pathogens. Phylogenetic dispersion analysis of each host type revealed that the phylogenetic preferences of beetle attack and fungal growth were a nonrandom subset of all available tree and shrub species. Competent hosts were phylogenetically narrower by 62 Myr than the set of all potential hosts, and those with devastating impacts were the most constrained by 107 Myr. Our results show a strong phylogenetic signal in the relative effects of a generalist pest–pathogen complex on host species, demonstrating that the strength of multi‐host pest impacts in plants can be predicted by host evolutionary relationships. This study presents a unifying theoretical approach to identifying likely disease outcomes across multiple host‐pest combinations.

threats and optimize the use of limited resources for management, decision-makers require robust analytical tools that help determine in which plant communities emergent pests are most likely to establish and cause damage during critical early stages of invasions. As a necessary first step to developing predictive models of pest spread in novel habitats, we take an evolutionary ecology approach and examine how the host range structure of different pest-pathogen combinations can be used to better understand mechanisms of their establishment, spread and impacts.
Evolutionary tools show promise as a way to understand invasions and predict host range of pests in novel locations (Briese, 2003;Fountain-Jones et al., 2018;Gilbert et al., 2012). For plants and their pathogens, evolutionary constraints in physiological, morphological and chemical traits that confer host susceptibility or pathogen virulence produce a phylogenetic signal for host range; hence, closely related plants are more likely to share pests and pathogens (Gilbert & Webb, 2007;Young et al., 2017). Phylogenetic signal in host range has been used to predict the likely host range of generalist plant pests in local communities not yet invaded by such pests (Gilbert & Parker, 2016;Parker et al., 2015). Patterns of phylogenetic signal in host range have been well documented for plant-pest relationships involving a single pest interacting with their host plants (e.g., plantpathogen, plant-insect), but not for those exhibiting multiple interactions (e.g., pest-pathogen complexes) where the traits shaping the relationships may differ among the multiple partners and their interactions. As such, the patterns and strength of the signal as a basis for risk analysis for more complex plant-pest problems are less well understood. Here, we use an emergent invasive pest-pathogen complex affecting a diversity of tree hosts in Southern California to test the utility of this phylogenetic tool in evaluating host range for novel plant-insect-pathogen interactions. Further, we assess whether we can use information on the phylogenetic structure of the pest-pathogen host range in California, where the complex has been intensively studied, to guide an understanding of likely patterns in South Africa and inform priorities for phytosanitary surveillance, where the invaders have only recently established.
Since the appearance of ISHB in California in 2012, the combined effects of ISHB and their fusaria symbionts have killed or caused dieback on 77 tree species on which the beetles can reproduce, but the beetles make attempted attacks on an additional 247 tree species ( Figure 1, Table S1; Eskalen et al., 2013). The two pest-pathogen complexes that form FD-ISHB have indistinguishable host ranges.
Critically, the recent introduction of one of those complexes to South Africa, the polyphagous shot hole borer (PSHB, Table 1; Paap et al., 2018a) has been cause for concern given the severe damage these invasive species have caused in California. The known host range in California and South Africa continues to grow, pointing to the need for a sound scientific understanding of the complexity of the FD-ISHB host range to inform risk assessments and focus phytosanitary actions in areas where the beetles have established, and in noninvaded locations worldwide that have favourable conditions for their establishment.
While a large body of work has established there is a phylogenetic signal in overall host ranges of pests and pathogens (Gilbert & Parker, 2016), the phylogenetic signal of competence and severity among hosts is much less well understood . In addition to distinguishing between hosts that do not support reproduction of the beetle-pathogen (non-competent) and those that do (competent), phylogenetic relatedness may also predict those hosts that are killed by the beetle-pathogen (killed-competent; Figure 1). For FD-ISHB, different host types (non-competent, competent, and killed-competent) are based on nested types of outcomes of interactions between host plants, the beetles and the fungi (Figure 1) Africa is therefore imperative.
In this study, we tested the hypothesis that hosts supporting ISHB-Fusarium reproduction are more strongly phylogenetically constrained than non-competent hosts. As such, we expect that the probability of finding ISHB on two host species declines with F I G U R E 1 Representation of the expected phylogenetic effects on different host types impacted by Fusarium dieback-invasive shot hole borers. The left panel (a-e) depicts examples of nested types of outcomes of interactions between host plants, the beetles, and the fungi. Non-competent hosts (a-c) represent tree species that do not support beetle reproduction or fungal transmission. For host types on which the beetle attempts an attack (a-b), entry holes are observed but removal of the bark reveals healthy tissue and no signs of a gallery.
Removal of the outer bark on hosts susceptible to Fusarium colonization (c) reveals necrotic tissue caused by the pathogen, but no signs of a gallery. On competent hosts (d), the beetle is able to establish a natal gallery and produce offspring and on some of these (e), the beetle and pathogen can kill the host (i.e., killed-competent). Successfully established breeding galleries in competent hosts contain a "fungal garden" and beetles at all life stages (eggs, developing larvae, adults), demonstrating the beetles' ability to cultivate their nutritional symbiotic fungi and complete their life cycle. Colours around each image correspond to the host type represented by the nested boxes in the middle panel (f), the sizes of each which correspond to the relative proportion of tree species for each host type. The phylogenetic tree in the right panel (g) depicts our hypothesis that hosts are a nonrandom, closely related, subset of all available tree species and that this phylogenetic signal is more pronounced for each of the nested interaction outcomes. The icons represent the examples of the nested types of interaction outcomes from most inclusive to least inclusive phylogenetic distance between the hosts, and this decline is steeper for competent hosts. Moreover, we expect that phylogenetic signal in host range is stronger on competent hosts that are killed when attacked.

| Host range assessment
The FD-ISHB host range comprises 77 host species that support beetle reproduction (competent hosts), 18 of which are killed when attacked ( Figure 1, Table S1). The adult beetles make attempted attacks on another 247 species in 61 families that do not support their reproduction (non-competent hosts), although the fungi can colonize and cause necrosis on 137 of these non-competent hosts ( Figure 1, Table S1; Eskalen et al., 2013). These non-competent hosts are never killed when attacked. The specific definitions and details for each of these categories are provided in Figure  For each individual tree, surveyors recorded at minimum the tree location, species, and the presence or absence of FD-ISHB based on the unique symptoms caused by the beetles and fungi as described in Eskalen et al. (2013). Tree species not exhibiting FD-ISHB symptoms, but in areas with active infestations, were classified as apparent nonhosts. In all cases of new tree species exhibiting symptoms characteristic of FD-ISHB, fungal and beetle identities were confirmed using morphological and molecular identification techniques described in Eskalen et al. (2013). Suitability for reproduction was confirmed by the presence of eggs, larvae, pupae or teneral females, or by the presence of males in the galleries of infested trees.

| Analyses
To estimate the time of independent evolution between plant species (phylogenetic distance), we first created a hypothesis for the phylogenetic relationships among tree and shrub species in California and South Africa using the R2G2_20140601 supertree of; see Data S1 for newick file). This tree includes dated nodes for all angiosperm families given by the Angiosperm Phylogeny Group classification III (APG III; Bremer et al., 2009) as well as gymnosperm and monilophyte families; the tree was dated using Wikström ages (Davies et al., 2004;Wikström et al., 2001) and additional consensus dates from the literature, with all nodes in the tree given stable dates (Parker et al., 2015). We used this tree rather than basing our phylogenetic tree on APG IV (Byng et al., 2016) to be consistent with and comparable to the validated work on phylogenetic signal in host ranges in the previous studies. All 2717 taxa for which the beetles could encounter in California or South Africa include native and nonnative trees and shrubs found across agricultural, urban and wildland landscapes, and were compiled using the CalFlora, West Coast Arborists, The Plant List, and Dendrological Society of South Africa curated databases (Data S1). We used Phylomatic version included in Phylocom v4.2 (Webb et al., 2008) to create a pruned ultrametric tree of all genera in the database, with branch lengths that reflected the estimated time between branching events (Data S1).
The case of zero phylogenetic distance (distance from a known host species to itself) was included in the analysis.
We performed a phylogenetic dispersion analysis of phylogenetic distances for all examined tree species, confirmed nonhosts, non-competent hosts (attempted host attack only and attacked hosts suitable for fungal colonization), and all competent host species and their subsets of those that are killed or not killed when attacked. We followed approaches used in previous publications and inspected the cumulative distribution of phylogenetic distances between species pairs (CDPD), which provides useful information on the depth of trait conservatism in plant-pathogen interactions (Gilbert & Parker, 2016;Parker et al., 2015). Overlap of CDPD curves between all examined tree species and host tree species indicates that hosts are a random subset of all available tree species (no phylogenetic signal). A downward shift in the host CDPD curve indicates that host species are a more closely related subset of all available tree species than expected at random, because the removal of more distantly related clades retains shorter distances (phylogenetic signal). We expect these downward shifts to be more dramatic with hosts that are increasingly more severely impacted by the beetle-fungal interactions. Measures of mean phylogenetic distance in pest host ranges across broad plant phylogenies tend to be dominated by the influence of many long phylogenetic distance pairings (Gilbert & Parker, 2016). Additionally, nearest phylogenetic distance measures can be unstable because they do not reflect the plant community as a whole. In addition to examining the overall CDPD, we follow Parker et al. (2015) and compare distances at the 10th quantile, which were found to be more informative than mean distances for plant-fungal interactions because it reduces the structural swamping effect of many distantly related pairs in phylogenies.
In addition to phylogenetic dispersion analysis, we measured

| Phylogenetic patterns of host-pest interactions
The distribution of non-competent and competent hosts exhibited a nested pattern across the phylogeny of potential host species in California and South Africa. Species that were attacked by the beetles clustered within 62 families and 170 genera within our geographic ranges (Figure 2). These taxa cover the range of angiosperm and some gymnosperm tree species. For gymnosperms, beetle attack attempts occurred on species within the "crown conifer" clade (Cupressaceae, Podocarpaceae, Pinaceae) but not species within other more distantly related groups (e.g., Ginkgoaceae or Cycadales;  (Table 2).

| Phylogenetic dispersion analysis
The phylogenetic distances for all pairs of the 2717 observed tree species and confirmed nonhosts from California and South Africa ranged between 1.4 and 806 Myr ( Figure S1). This range decreased notably with increasingly severe nested types of outcomes of interactions between host plants, the beetles, and the fungus ( Figure   S1). We ranked the phylogenetic distances for all species pairs and their respective subsets (Figure 3a and S2,S3). Consistent with results in Parker et al. (2015), inspection of the full CDPD curves indicated that affected phylogenetic distances tend to be much shorter than the overall median because of the swamping effect of many distantly related pairs ( Figure S2). As such, we focused our analysis at the scale of the 10th quantile of pairwise phylogenetic distances between species, where the depth of conservatism of important traits that confer host susceptibility is most informative (Figure 3b).
As phylogenetic distance represents time of independent evolution (Myr), shorter distances indicate species are more closely related to one another. Species that were attacked by beetles were a nonrandom subset of all the available hosts as indicated by a downward shift in their CDPD curve; the phylogenetic distances among the attacked F I G U R E 2 Phylogenetic tree of families representing all examined tree species in the present study. Stacked columns at the tree tips depict the nested types of outcomes of interactions between host plants, beetles, and fungi for genera within each family. Segments within each column represent the number of attacked genera with tree species that are Fusarium-colonized, competent, and killed-competent hosts within each family  for all the hosts attacked by the beetles (i.e., killed host species are much more closely related to each other than are all the species attacked by the beetles). Removal of gymnosperms from the host data revealed a shift in the CDPD for non-competent hosts, but distances were still longer than competent hosts ( Figure S3). Patterns were not different when South African trees were removed from the analysis ( Figure S3).

| D ISCUSS I ON
In this study, we quantified the degree of phylogenetic signal in the host range of a new invasive generalist pest and pathogen complex from southeast Asia that elicits different effects across different host tree species. As we expected, the 327 tree species attacked by

F I G U R E 3
Phylogenetic distances for all species pairs of each host type (a-b). Intervals represent the 95% confidence interval envelope generated from 10,000 bootstrap simulations on a random sample of 90% of the species within each host type. (a) Cumulative distribution of phylogenetic distances (CDPD) from quantiles 1%-15%. (b) Boxplots of phylogenetic distances at the 10th quantile. Grey dots represent actual data from the simulations species, demonstrating that the strength of multi-host pest impacts in plants can be predicted by host evolutionary relationships. These findings form the basis for developing predictive models of multihost pest spread in novel habitats using tools in phylogenetic ecology.

| Estimations of phylogenetic signal
Both phylogenetic dispersion analysis and the D statistical measure of phylogenetic signal (Fritz & Purvis, 2010) detected a phylogenetic effect on the most severely affected competent hosts. Phylogenetic dispersion analysis was potentially more sensitive in detecting a signal for non-competent and all competent hosts than D because while there are "jumps" in the signal (i.e., roughly 25% of competent hosts occur outside the Rosids), we see high clustering within groups containing competent host species. Within the Rosids, there is another jump in the signal between the Fabids and Malvids, but a high degree of clustering occurs within those two groups, particularly in the Fabids (i.e., Salicaceae, Fagaceae and Fabaceae) and the Malvids (i.e., Sapindaceae). The D measure in phylogenetic signal is based on an underlying threshold model, which assumes that patterns of a binary trait across the phylogeny are based on one or more evolved, continuous traits (Fritz & Purvis, 2010). However, although many traits important in plant-enemy interactions show a phylogenetic signal (Agrawal, 2007;Boller & Felix, 2009;Gilbert & Parker, 2016;Pearse & Hipp, 2009), there are exceptions (Becerra, 1997;Pichersky & Lewinsohn, 2011;Wink, 2003). Thus, our results suggest there are many ways for hosts to be susceptible. Those ways are moderately constrained phylogenetically, but susceptibility clusters within phylogenetic groups and this clumping becomes more restricted with more impactful interactions.

| Phylogenetic signal in multi-host pest interactions
Quantitative measures that leverage an understanding of the evolutionary ecology of host-pest interactions to assess the relative impacts of generalist pests on their hosts provide important and novel tools to predict threats to ecosystems. By utilizing multiple invasion pathways, multi-host pests present inherently different epidemiological dynamics than single host pests when introduced to naïve plant or animal communities. In particular, generalist pests do not rely on density-dependent transmission of a single host species, which thereby increases the likelihood of pest-induced host extinction (De Castro & Bolker, 2005;Smith et al., 2006). As the majority of plant and animal pests attack multiple host species (Cleaveland et al., 2001;Gilbert et al., 2012;Gilbert & Webb, 2007;Malpica et al., 2006;Novotny et al., 2002;Pearse & Hipp, 2009;Weiblen et al., 2006), these essential evolutionary tools in species conservation efforts are also broadly applicable. For domesticated mammals, Farrell and Davies (2019) demonstrated that evolutionary distance from an infected host to another mammal host species is a strong predictor of multi-host disease-induced mortality. Similarly, Gilbert et al. (2015) reported that the relative amount of damage done by a natural enemy on plant species declines predictably with increasing evolutionary distance from highly susceptible hosts.
Our study affirms that the use of host evolutionary relationships presents a unifying theoretical approach to predicting disease outcomes across multiple host-pest combinations.

| Epidemiological implications of host evolutionary relationships
In addition to determining which species are prone to pest-induced

| Caveats
One limitation to our analysis is that our information on which hosts the Fusarium pathogens can grow is not independent of beetle attack. Experimental inoculations of the fungi on confirmed nonhost tree species (no symptoms of beetle attack) would indicate whether the Fusarium host range is truly constrained phylogenetically.
However, the relationship between the beetles and their fungi is tightly coupled. The Fusarium species belong to the monophyletic Ambrosia Fusarium Clade (AFC; Kasson et al., 2013) and the ~22

Myr old mutualism between AFC members and beetles in the genus
Euwallacea represents 1 of 11 known evolutionary origins of fungiculture by ambrosia beetles (O'Donnell et al., 2015). These closely related wood-inhabiting Fusarium species are transmitted in mycangia and cultivated by females in galleries as a source of nutrition for the beetle (Kasson et al., 2013;O'Donnell et al., 2015). Phylogenetic models based on evolutionary distances between hosts of generalist pests can be used to evaluate which host species are potentially most vulnerable to pest impacts and most important to their establishment and spread. Certainly, other essential factors that drive host-pest interactions influence host outcomes. Changes in environmental conditions, pathogen virulence or the host microbiome can amplify or inhibit host susceptibility or damage. In particular, the phylogenetic structure and host abundance of local communities strongly influence the severity of impact on focal hosts (Parker et al., 2015). Although phylogenetic signal in host range cannot fully explain overall epidemic patterns, it can be used as a first approximation to understanding complex novel pest invasions, serving as a powerful tool to assess risk and guide response priorities.

ACK N OWLED G EM ENTS
We thank our funders for supporting this research: San Diego