Woodland wildfire enables fungal colonization of encroaching Douglas-fir

1. Self-reinforcing differences in fire frequency help closed-canopy forests, which resist fire, and open woodlands, which naturally burn often, to co-occur stably at landscape scales. Forest tree seedlings, which could otherwise encroach and overgrow woodlands, are killed by regular fire, yet fire has other effects that may also influence these feedbacks. In particular, many forest trees require symbiotic ectomycorrhizal fungi in order to establish. By restructuring soil fungal communities, fire might affect the availability of symbionts or the potential for symbiont sharing between encroaching trees and woodland vegetation. 2. To investigate this possibility, we performed a soil bioassay experiment using inoculum from burned and unburned oak woodlands and Douglas-fir forests. We examined how fire, ecosystem type, and neighboring heterospecific seedlings affect fungal root community assembly of Douglas-firs and oaks. We asked whether heterospecific seedlings facilitated fungal colonization of seedling roots in non-native soil, and if so, whether fire influenced this interaction. 3. External fungal colonization of oak roots was more influenced by fire and ecosystem type than by the presence


| INTRODUC TI ON
Plant communities vary widely in species composition, giving rise to numerous different kinds of ecosystems. Identifying the drivers of plant and ecosystem distributions is of both basic and applied importance, as enumerating and comparing them enables us to better predict how things might change in the future. Several early investigations into this subject suggested that climate might lead to the deterministic development of particular ecosystem types in particular places (Clements, 1936). According to such a framework, the vegetation of a given location ought to be broadly predictable from, for example, its average rainfall and temperature (Whittaker, 1975).
Since those pioneering first steps, however, the prevailing view has grown more nuanced (Beadle, 1951).
It is now well established that plant communities are dynamic, stochastic and often in apparent persistent disequilibrium with climate (Bond, 2008;Sankaran et al., 2005). Numerous other factors, such as disturbance (Bond et al., 2005), human introduction of novel species (Mack et al., 2000) and interactions with other organisms (Peay, 2016;Pringle et al., 2009), also affect vegetation distributions at fine and broad scales. These other factors may be especially important where the climatic envelope allows for multiple different ecosystem types, as in regions where closed-canopy forest and savannahs with low tree density co-occur at landscape scale (Staver et al., 2011b).
Indeed, regular wildfire helps to maintain ecosystems with sparse tree cover, like woodlands or savannahs, where closed-canopy forests might otherwise predominate (Bond et al., 2005). In part due to the traits of their constituent plants (Ellair & Platt, 2013), these ecosystems catch fire more frequently than do dense forests (Archibald et al., 2009;Staver et al., 2011b). This hinders woody encroachment by killing forest tree seedlings (Hoffmann & Solbrig, 2003) and preventing recruitment into adult populations (Higgins et al., 2000).
However, factors besides fire may additionally constrain tree density in these kinds of ecosystems (Pacé et al., 2019). In particular, the availability of root-associated mycorrhizal symbionts, which help plants take up nutrients from soil, also constrain growth and range sizes (Delavaux et al., 2019), especially of trees (Briscoe, 1959).
In fact, the majority of our planet's forest trees acquire nutrients through symbiotic relationships with mycorrhizal fungi (Brundrett & Tedersoo, 2018) and many cannot sustain populations without suitable ectomycorrhizal partners in particular (Briscoe, 1959).
Ectomycorrhizal fungi are obligately biotrophic and dispersal limited, which constrains their spatial distributions at both fine and broad scales (Peay et al., 2012;Smith et al., 2018;Talbot et al., 2014). As a result, tree seedling recruitment at forest edges or outside of a species' native range can be limited by low ectomycorrhizal inoculum abundance in surrounding soil (Ashkannejhad & Horton, 2006;Nuñez et al., 2009;Pringle et al., 2009). This can occur because plant communities adjacent to ectomycorrhizal forests are often dominated by grasses and shrubs that instead associate with incompatible arbuscular mycorrhizal fungi (Smith et al., 2018). Even more broadly, root-associated fungi in general, whether specifically mycorrhizal or not, can affect the health and growth of plants (Grelet et al., 2017).
Thus, forces that are top-down (as in the case of wildfire's combustion of tree biomass) and bottom-up (as in the case of fungal influences on tree growth) both affect spatial vegetation patterning across scales, tempering climate's apparently deterministic influence. These complex factors vary spatiotemporally in their relative importance and may also interact. For example, though a lack of ectomycorrhizal inoculum can constrain seedling growth outside of forests in some regions (Richardson et al., 2000), scattered woodland trees and shrubs in others may support active populations of ectomycorrhizal fungi, which can play a facilitative role if they are compatible with forest tree seedlings Richard et al., 2009). Similarly, while land management choices have reduced the influence of fire in some areas, others face mounting wildfire pressure due to global change (Westerling et al., 2006). Finally, since fire affects ecosystems not only above but also below-ground (Pressler et al., 2019;Smith et al., 2021), changing fire regimes could even reshape soil fungal communities in both forests and savannahs, with knock-on effects on plant dynamics.
All of these factors are potentially at play in the north coast bioregion of California, United States, where closed canopy mixed evergreen forests intergrade with open, grassy oak woodlands (Van Wagtendonk et al., 2018). Here, many common forest trees, like Douglas-fir, are ectomycorrhizal, just like the oaks that populate adjacent woodlands (Brundrett & Tedersoo, 2018). This shared mycorrhizal type could enable mycorrhizal facilitation between oaks and evergreen forest trees (Rasmussen et al., 2017), which may support ecosystem transitions over time from woodland to forest. Moreover, while woody encroachment of Douglas-fir into oak woodlands does occur where fire is suppressed, prescribed and natural burns effectively halt this process (Barnhart et al., 1996;Hastings et al., 1997).
These fires may be maintaining oak woodlands not only by killing encroaching tree seedlings but also by reducing their ability to benefit from established oak-associated fungal populations.
To evaluate and compare the roles played by these different potential factors in structuring vegetation distributions, we collected soil from Douglas-fir Pseudotsuga menziesii forests and oak woodlands in Sonoma County, California, United States, directly after a spate of damaging wildfires in 2017 (Nauslar et al., 2018), pairing sites affected and unaffected by fire. We used these soil samples as inocula for a fully factorial growth chamber bioassay experiment with Douglas-fir and Oregon white oak Quercus garryana (hereafter 'oak') as bait seedlings. We sought to thereby determine the composition of the root-associated fungal communities recruited by Douglas-fir and oak seedlings, the extent to which these communities are affected by growing the seedlings of the two species together instead of separately (indicating potential for direct interspecies facilitation via partner sharing) and how root-associated fungal community assembly for each seedling interacts with fire and ecosystem type. We chose a soil bioassay approach because resistant propagules, such as those recovered by bioassays, are known to play an important role | 3 Functional Ecology SMITH and PEAY in the fungal colonization of tree seedlings establishing after wildfire (Baar et al., 1999).
We expected that (H1) Douglas-fir and oak would each recruit a distinct fungal community, with levels of taxonomic (amplicon sequence variant; ASV) richness and root colonization highest when inoculated with home soil. Yet, as even host-specific ectomycorrhizal fungi may colonize secondary hosts via hyphal spread from a primary host (Lofgren et al., 2018), we also expected that (H2) growing a seedling with a heterospecific (that is, an oak if the focal seedling were a Douglas-fir, or a Douglas-fir if the focal seedling were an oak) would expand its root-associated fungal community and increase root colonization. We expect that this fungal facilitation effect should be particularly strong in soil not sourced from a seedling's own ecosystem type, as a larger proportion of fungi in these samples are likely to prefer the other tree species as a primary host.
Since fire is a strong selective filter for fungal communities (Baar et al., 1999), we hypothesized that (H3) fungal colonization of roots and root-associated fungal species richness would be greatest when seedlings were inoculated with soil from ecosystems unaffected by fire. Finally, since fire hampers woody encroachment of Douglas-fir into oak woodlands (Barnhart et al., 1996;Hastings et al., 1997), we also hypothesized that (H4) expansion of Douglas-fir root-associated fungal communities by oaks would be greater in oak woodland soil that was unaffected rather than affected by fire. Trione-Annadel. Oak woodlands at the park have historically burned as frequently as every 6 years, but fire frequency decreased in the 19th and 20th century (Finney & Martin, 1992), allowing encroachment by Douglas-fir trees (Barnhart et al., 1996;Hastings et al., 1997). In October 2017, portions of Trione-Annadel burned in the Nuns wildfire, which consumed nearly 23,000 hectares in total (Nauslar et al., 2018).

| Field site
In the following December, we established three sites each in burned and unburned Douglas-fir forest and oak woodland, for a total of 12 sites. Full site descriptions and a map are available in Smith et al. (2021). Study sites were chosen in cooperation with and according to the requirements of the California State Park Service.
These requirements, as well as the inevitable spatial pattern of the naturally occurring wildfire, constrained the placement of sites. At each one, five trees of the target species (either P. menziesii or Q. garryana) located at least 5 m from one another were selected. We chose this distance because prior investigation of ectomycorrhizal fungi found that spatial autocorrelation was limited to approximately 3 m distance (Lilleskov et al., 2004). Moreover, a prior study in this system found that the minor influence of geography on soil fungal community composition relative to other predictors increased most rapidly among the shortest pairwise distances (Smith et al., 2021).
This indicates that even small spatial distances between samples create differences in community composition that are large relative to the maximum potential impact of geography, minimizing the potential statistical challenge of spatial autocorrelation in this system.
Using an ethanol-sterilized trowel, we collected a sample from the top 10 cm of soil approximately 1 m from the trunk of each selected tree, for a total of five samples per site. In total, 60 soil samples (2 ecosystem types × 2 burn histories × 3 sites × 5 samples) were collected. Samples were transported at 4°C to Stanford University for processing.

| Growth chamber experiment
Each soil inoculum sample was diluted 1:7 by volume with an autoclaved 1:1 mixture of organic potting soil and coarse sand, which had been sterilized by autoclaving at 121°C for 45 min. For each of our 60 soil samples, three 164 mL Cone-tainer pots were prepared (Stuewe & Sons Inc.), representing three different seedling treatments: either P. menziesii or Q. garryana growing alone, or both species together.
Seeds and acorns were purchased from Sheffield's Seed Co. Inc. and surface sterilized in 6% H 2 O 2 for 20 min, followed by a 24-h soak in distilled water prior to planting. Because the five soil samples from each site were not pooled, every seedling replicate received a unique inoculum treatment and thus represented a unique biological (not technical) replicate. Control seedlings receiving autoclaved soil inoculum were maintained in the growth chamber for the duration of the experiment. Less than 3% of control seedlings showed fungal colonization; contaminated control seedlings were not included in further analyses. In total, 180 treatment pots were prepared (2 ecosystem types × 2 burn histories × 3 sites × 5 samples × 3 seedling treatments).
Pots were arranged in a blocked design in the growth chamber and received five Douglas-fir seeds, a single oak acorn, or both, depending on their seedling treatment. The position of the blocks in the growth chamber was rotated monthly in order to ensure that they did not each experience a consistent different abiotic environment over the course of the experiment. Pots were watered thrice weekly to saturation. Replicates where seedlings failed to emerge after approximately 4 weeks were replanted; at this time point, pots where multiple Douglas-fir seedlings had emerged were thinned to a single individual. Pots where replanted Douglas-fir still did not sprout were removed from the experiment. After 6 months, we harvested the seedlings (see Table S1 for final replicate counts). Because the harvest was time-consuming, different seedlings grew for different amounts of time; to control for this variation, we included days of growth as a term in downstream models.
Then, a spatially stratified subset of roots consisting of equal sections taken from the top, middle, and bottom of the root system of each plant were scored for fungal colonization using the gridline intersect method (Giovanetti & Mosse, 1980). Since non-mycorrhizal fungi can colonize and interact with roots (Grelet et al., 2017;Smith et al., 2017), we considered all visible external fungal colonization rather than restricting our investigation to known ectomycorrhizal morphology. Internal fungal colonization was not assessed. A random 0.1 g portion from this subset was then frozen at −80°C for DNA extraction, amplification and sequencing. Root samples taken from oaks and Douglas-firs growing together in the same pot were kept separate. Oaks generally develop a large tap root with low fungal colonization, so root samples were taken from the fine roots and laterals associated with the tap root, but not from the tap root itself. From SFGF, we received a total of 5,237,549 demultiplexed reads across 130 visibly fungal colonized root samples, with an average of 40,289 reads per sample. We processed these reads bioinformatically using a DADA2 pipeline optimized for the fungal ITS region (Callahan et al., 2016). After quality filtering, denoising, merging of forward and reverse reads, and removal of chimeric sequences, we were left with a total of 4,182,231 reads and an average of 32,171 per sample. We rarefied all of our samples to an even sequencing depth before analysis and removed samples with fewer than 10,000 reads or from control seedlings. We used the DADA2 default naïve Bayesian classifier (Wang et al., 2007) to assign taxonomy to our amplicon sequence variants (ASV) based on UNITE (Kõljalg et al., 2013).

| Bioinformatics
We did not perform post-hoc operational taxonomic unit clustering since ASV analyses preserve information relative to the latter and recover comparable ecological signals (Glassman & Martiny, 2018).
Our taxonomic designations refer, thus, to ASVs rather than operational taxonomic units and thereby capture fine-scale diversity that could be masked by clustering (Tipton et al., 2021). In total, we retained 116 samples containing a total of 1074 unique ASVs, to which we assigned functional guild classifications using FUNGuild (Nguyen et al., 2016) following the protocol of Smith et al. (2021).

| Statistical analyses
All statistical analyses were carried out using R statistical programming language (R Core Team, 2021). Some samples intended to be two-species treatments instead became single species treatments because one seedling failed to germinate. If these incidental replicates were not pseudoreplicates (i.e. there did not already exist a seedling of that species grown alone in the same soil inoculum due to germination failure), we recoded them appropriately and kept them for analysis; otherwise, they were discarded. In total, we retained 174 replicate seedlings across 137 pots. The number of replicates for each treatment group is given in Table S1. Metabarcoding data generated by DADA2 were processed using r package phyloseq (McMurdie & Holmes, 2013). Plots were generated with r packages ggplot and cowplot (Wickham, 2016;Wilke, 2019). Though investigations of fungal root communities often focus only on known mycorrhizal taxa, other fungi may also frequently colonize and functionally interact with roots (Grelet et al., 2017;Smith et al., 2017), so we did not filter the community prior to analysis. We did, however, perform several analyses specifically on the subset of the community classified by FUNGuild (Nguyen et al., 2016) as mycorrhizal. When discussing these analyses, we refer to the 'mycorrhizal' communityotherwise, we refer to the unfiltered 'fungal' community.
To examine variation in percent root fungal colonization across treatments, we used generalized linear models as implemented in glm, with interacting terms for ecosystem type and fire status of the soil inoculum, as well as whether the focal seedling was grown alone or with a heterospecific. We also included a term for the number of days each seedling grew. We chose not to employ mixed effects models because prior research, both at our study site (Smith et al., 2021) and elsewhere (Lilleskov et al., 2004), indicates that our sampling strategy minimized the potential impact of spatial autocorrelation (see Section 2.1) and each of our seedlings represented a unique biological (not technical) replicate because soil samples were not pooled. Because our response variable was a proportion, we used a quasibinomial family. To assess the significance of model terms, we used Anova from the r package car (Fox & Weisberg, 2011) with type III sum of squares if interaction terms were significant and type II sum of squares if they were not. To enable examination of marginal effects, we used treatment contrasts for our models, assigning the unburned home ecosystem inoculum for each seedling, growing without a neighboring heterospecific, as the reference state.
We used zero-inflated Poisson models, as implemented in zeroinfl from r package pscl (Zeileis et al., 2008), to compare total ASV richness of fungi as well as of mycorrhizal fungi only across different treatments, utilizing the same set of model terms as employed for percent root fungal colonization. For these models, we included data from uncolonized seedlings, which were assigned a richness measurement of zero; these zero measurements from uncolonized seedlings aren't included in Figure 2. Since zero-inflated Poisson models comprise a separate count model and zero-inflation model, these measurements from uncolonized seedlings do not influence any findings regarding richness. Instead, with this approach we are able to examine both the factors affecting whether a seedling becomes colonized by externally visible fungi at all (the zero-inflation model) and the factors affecting the root-associated fungal richness of colonized seedlings (the count model). We point out that even seedlings that appear outwardly uncolonized (which we did not sequence) are likely to contain fungal endophytes.
This approach did not work when examining oak fungal communities due to complete separation of colonized and uncolonized seedlings across treatments (in other words, some treatments had no colonized seedlings), which biased estimates and inflated standard errors in the zero-inflation model. Here, we instead used a bias-reduced generalized linear model, as implemented in r package brglm2 (Kosmidis et al., 2020), with a Poisson distribution and including zero measurements representing uncolonized seedlings. We also ran a regular generalized linear model as detailed above, excluding measurements from uncolonized oak seedlings.
To enable depiction of significant differences between groups in figures, we performed post-hoc tests separately examining each seedling species and soil inoculum combination. In each case, we first ran quasibinomial generalized linear models for percent colonization data and Poisson generalized linear models for richness data, examining the interacting effects of fire and the presence of a heterospecific while controlling for number of growth days. For richness, zeroes were excluded from the post-hoc comparisons, just as they are excluded from the figure. Then, we performed post-hoc pairwise comparisons using r package lsmeans (Lenth, 2016), with a multivariate t distribution correction for multiple comparisons.
To investigate variation in root fungal community composition, we used nonmetric multidimensional scaling ordination and permutational analysis of variance as implemented in command adonis from r package vegan (Oksanen et al., 2019). First, we performed a permutational analysis of variance to determine whether there were significant differences between oak and Douglas-fir communities.
Then, we assessed the interacting effects of fire, ecosystem type, and the presence of a heterospecific on each species separately.
In all tests, we also included a term for growth days. To examine the affinity of mycorrhizal fungi (as classified by FUNGuild;Nguyen et al., 2016) for different treatment combinations and seedling species, we generated genus-and species-level heatmaps using plot_ heatmap from phyloseq (McMurdie & Holmes, 2013).
To compare the degree to which each tree species associates with host-specific fungi, we examined the proportion of reads belonging to fungal ASVs unique to Douglas-fir or oak when grown with the inoculum from the same soil sample, either in two separate pots or together in the same pot. First, we ran generalized linear models (as described above for colonization percent), with a quasibinomial link function due to proportional response data, separately for Douglasfir and oak. We used these models to examine whether the proportion of reads belonging to unique ASVs was affected by ecosystem type, fire, the presence of a heterospecific, or interactions between these terms, with an added term to control for any difference in the number of growth days for the paired seedlings.
We then tested whether the proportion of reads belonging to unique ASVs differed between seedling species when they were grown separately with the same inoculum treatment. Finally, as prior research finds host-specificity in Pinaceae-associated ectomycorrhizal fungi, we examined whether the proportion of reads belonging to unique ASVs was greater in Douglas-fir seedlings than in oaks when they were grown together, which enables hyphal spread from a primary to a secondary host (Lofgren et al., 2018). For these two tests, we first used linear regression to examine whether the difference in number of growth days between the oak and Douglas-fir seedlings in question predicted variation in the proportion of unique reads found in either one. We performed this preliminary check separately for oaks grown with a heterospecific, oaks grown alone, Douglas-firs grown with a heterospecific, and Douglas-firs grown alone. We did not find any significant (α = 0.05) effect of differences in growth days in any of these cases, so we felt comfortable using a simpler statistical framework that omitted this factor. Accordingly, we verified normality of differences using Shapiro-Wilk tests then performed a paired t-test if differences were normal.
In all cases, our models were intended to identify the statistical effect of specific predictors based on biologically motivated hypotheses formulated a priori, following the example of Hobbs et al. (2012). We therefore did not perform model comparison in order to maximize, for example, Akaike information criterion or any other information-theoretic measures.

| RE SULTS
Visible fungal root colonization varied significantly across the different treatments and seedling species (Figure 1). For Douglas-fir seedlings, percent root colonization was significantly lower in oak woodland soil (p < 0.0001, χ 2 = 22.5567; Table S2), but there were also significant interactions: between ecosystem type and fire (p = 0.0006, χ 2 = 11.7617), indicating that fire can attenuate the negative effect of oak woodland soil on visible colonization, and between the presence of a heterospecific and ecosystem type (p = 0.0020, χ 2 = 9.5388), indicating that a neighboring oak seedling can similarly increase colonization in this soil-about as strongly, in fact, as fire does (ß = 3.1820 vs. ß = 3.1826). There was also a marginal negative interaction between fire and the presence of a heterospecific (p = 0.0513; χ 2 = 3.7996), which would suggest that when fire and a neighboring oak co-occur, they marginally reduce colonization in Douglas-fir soil.
In contrast, visible colonization of oak seedlings was marginally affected by ecosystem type (p = 0.0560, χ 2 = 3.6525; Table S3), where Douglas-fir forest soil was associated with lower colonization.
There was also a significant negative interaction between Douglasfir forest soil and fire (p = 0.0022, χ 2 = 9.3710), indicating that fire had a negative effect on colonization in this ecosystem type, but not in oak woodlands. Thus, visible colonization of Douglas-fir root systems was more affected by the biotic influence of a neighboring heterospecific relative to oaks, which were instead largely sensitive only to variation in local fungal species pools (Smith et al., 2021) linked to differences in fire or ecosystem type.
The richness of root fungal communities recruited by Douglasfir seedlings in Douglas-fir forest soil was significantly reduced by fire (p = 0.0001, χ 2 = 16.3824; Figure 2; Table S4) and increased by the presence of a heterospecific (p = 0.0005, χ 2 = 12.0786). We also observed a significant positive interaction term between ecosystem type and fire (p = 0.0044, χ 2 = 8.1293) underscoring that the negative effect of fire on richness only occurred in Douglas-fir forests.
While we found no effect of ecosystem type on richness in the count model (p = 0.8150, χ 2 = 0.0547), the paired zero-inflation model indicated that visible fungal colonization was significantly more likely to occur in Douglas-fir forest soil than in oak woodland soil (p = 0.0007; Z = 3.3829). There was also a significant interaction between ecosystem type and presence of a heterospecific (p = 0.0043, Z = −2.8573), which indicates that a neighboring oak seedling fully ameliorates the negative effect of oak woodland soil on Douglas-fir seedling colonization likelihood (ß = 5.3392 vs. ß = −5.4517). There was also an interaction between ecosystem type and fire (p = 0.0211, Z = −2.3055), indicating that fire also does this to some extent, though not as strongly (ß = −5.4517 vs. ß = −4.1947). Additionally, we found a three-way interaction between ecosystem type, fire, and the presence of a heterospecific (p = 0.0173, Z = 2.3801), which indicates that specifically in oak woodland soils affected by fire, neighboring oak F I G U R E 1 Root colonization of Douglas-fir and oak seedlings grown together or alone, inoculated with soil collected from Douglas-fir forests (DF forest) or from oak woodlands. Red points and boxes represent soil inoculum from fire-affected ecosystems, while blue represents unaffected ecosystems. Pairwise post-hoc comparisons showing significant differences (α = 0.05) were performed within each panel, using a multivariate t distribution adjustment for multiple comparisons; within each panel, groups that do not share a letter are significantly different.  Table S6), but a significant interaction term indicated that this negative effect only occurred when seedlings were grown without a neighboring oak seedling (p = 0.0291, Z = −2.1815). Interestingly, in Douglas-fir forest soil, mycorrhizal richness of Douglas-fir seedlings was increased by the presence of a neighboring oak (p = 0.0037, χ 2 = 8.4028).
In contrast to Douglas-fir, oak root system fungal richness wasn't significantly reduced by fire in their home ecosystem (p = 0.6698, χ 2 = 0.1819; Figure 2; Table S5). We did, however, find a significant negative interaction term between ecosystem type and fire (p < 0.0001, χ 2 = 140.4994), underscoring that fire-affected Douglas-fir soil is a very poor source of fungal inoculum for either host plant species. We found a significant negative effect of a heterospecific (p = 0.0001, χ 2 = 14.3943), coupled with a positive interaction between ecosystem type and heterospecific (p = 0.0035, χ 2 = 8.5057)-together, this means that neighboring Douglas-fir reduce the root-associated fungal diversity of oak seedlings in oak woodland soil, but that in Douglas-fir soil this negative effect is mostly ameliorated (ß = −0.3653 vs. ß = 0.3564). Finally, we found a positive three-way interaction between ecosystem type, fire and the presence of a heterospecific, whether seedlings that were not visibly colonized were included (p < 0.0001, χ 2 = 29.5698) or unincluded (p = 0.0018, χ 2 = 9.7822). Combined, these interactions mean that for oak seedlings establishing in Douglas-fir forest after fire, a neighboring Douglas-fir has a positive effect, increasing both likelihood of visible colonization and total fungal diversity.
Taking both visibly colonized and uncolonized oak seedlings into account, richness specifically of mycorrhizal fungi on oak was significantly reduced by fire in Douglas-fir soil, as indicated by an interaction between ecosystem type and fire (p = 0.0000, χ 2 = 16.7822, Table S7). These trends were not apparent when excluding seedlings that were not visibly colonized, which suggests that fire caused mycorrhizal diversity reductions by affecting likelihood of visible mycorrhizal root colonization.
Fungal community composition varied across treatments for both seedling species (Figure 3). As we hypothesized, there was a significant difference between the communities of Douglas-fir and oak seedlings (p = 0.001; F = 7.8720;  fire and ecosystem type (p = 0.014, F = 1.8905). We also observed a significant three-way interaction between all factors (p = 0.032,

F = 1.7310).
When grown separately but with the same soil inoculum, there was no significant difference between the proportion of reads belonging to ASVs unique to each tree species in the root-associated fungal communities of Douglas-fir and oak seedlings (Figure 4; p = 0.76, t = −0.307), indicating that comparable proportions of the two species' root-associated communities were composed of host-specific fungi. Grown together in soil from either ecosystem, however, Douglas-fir seedlings had significantly higher proportions of reads belonging to unique ASVs compared to neighboring oak seedlings (p = 0.047, t = 1.7643). This suggests that while oakassociated fungi tend to colonize co-occurring Douglas-fir seedlings, it may be rarer for Douglas-fir associated fungi to colonize nearby oaks. Nevertheless, for both Douglas-fir and oak, the presence of a heterospecific significantly reduced the proportion of reads belonging to unique ASVs (respectively, p = 0.012, χ 2 = 6.2856; p < 0.0001, χ 2 = 16.1071). Additionally, for Douglas-fir, fire independently and significantly increased the proportion of reads belonging to unique ASVs (p = 0.01, χ 2 = 6.6638).
We observed host-specificity in several mycorrhizal genera.
Suillus and Rhizopogon were key root symbionts for Douglas-fir seedlings ( Figure 5), the latter especially in soil from Douglas-fir forests or oak woodlands after fire. Rhizopogon was only found with oak seedlings grown in Douglas-fir soil when a neighboring Douglas-fir seedling was present. Similarly, the Hebeloma, Inocybe, Tomentella, and Elaphomyces at our study sites only occurred with Douglas-fir seedlings when an oak seedling was present. Though a prominent component of oak woodland mycorrhizal communities, Elaphomyces was absent from Douglas-fir soil. The other three oak-associated genera had a wider distribution, but Tomentella never associated with Douglas-fir in fire-affected soil even when grown with an oak while, in contrast, Inocybe was only found with Douglas-fir in fireaffected soil.

| DISCUSS ION
Where climatic conditions neither preclude the development of closed-canopy forests nor result necessarily in the formation of open savannah or woodlands, self-reinforcing variation in fire frequency can create heterogenous patchworks of these two ecosystem types at landscape scales (Staver et al., 2011b). Prior work examining the mechanisms that drive these patterns has emphasized fire's role in directly restricting recruitment of juvenile trees into adult populations and increasing seedling mortality (Higgins et al., 2000;Hoffmann & Solbrig, 2003). Here, we find evidence that a further factor could also play a role, with fire altering soil fungal communities, which may then influence seedling establishment. Specifically, we observe clear patterns in root-associated fungal community assembly linked to fire, differences in ecosystem type, and biotic facilitation. Prior literature suggests that these sorts of differences in root fungal community assembly have the potential to create knock-on differences in plant growth (Nuñez et al., 2009;Peay, 2018).
In our system, evidence for such potential is able to arise, first, because there is differentiation and host-specificity in the rootassociated fungal communities of Douglas-fir and oak seedlings. In agreement with H1, Douglas-fir and oak trees developed distinct rootassociated fungal communities ( Figure 3) and levels of root system fungal colonization were higher when seedlings were inoculated with soil from their own ecosystem types (Figure 1). Yet fungal richness was largely unaffected by ecosystem type for either seedling, though Douglas-fir seedlings were more likely to be associated specifically with mycorrhizal fungi in their home soil. A handful of ectomycorrhizal genera frequently occurred with both Douglas-fir and oak, especially F I G U R E 4 Venn diagrams depicting the mean proportion of reads either belonging to ASVs that are either unique to each tree species or shared, occurring in root-associated fungal communities of Douglas-fir and oak seedlings that were grown together (w/ heterospecific) or separately (alone) while using the same inoculum. Green circles correspond to Douglas-fir seedlings, while yellow circles correspond to oak seedlings. The percentages listed refer to the mean proportion of reads belonging to unique ASVs for each species and treatment group. A visual presentation of unaggregated data for this analysis is available in the SI ( Figure S1).

DF forest soil
Oak woodland soil Tuber, which was common across both ecosystem types. Others displayed host specificity, including Tomentella, Inocybe, Hebeloma and Elaphomyces, all of which preferred oak, as well as Rhizopogon and Suillus, which showed affinity for Douglas-fir ( Figure 5). These results indicate potential in our study system for mycorrhizal limitation of tree seedling establishment outside of the home ecosystem sensu, for example, Nuñez et al. (2009), especially for Douglas-fir.
Second, we found that growing a seedling along with a heterospecific supported root colonization and expansion of the rootassociated fungal community in soil from the other ecosystem type (as predicted in H2), particularly with Douglas-fir. For oaks, root fungal colonization was unaffected by a Douglas-fir neighbor ( Figure 1) and fungal community richness was only thereby increased in fireaffected Douglas-fir soils (Figure 2). In contrast, Douglas-fir root fungal and mycorrhizal community richness and colonization were increased by a neighboring oak, especially in oak woodland soil (Figures 1 and 2). We find evidence, thus, that in order to be fungally colonized in unburned oak woodlands, Douglas-fir seedlings benefit from biotic introductions to fungal communities facilitated by the oaks themselves. We do not observe that Douglas-firs facilitate oak colonization to the same extent, potentially due to stricter hostspecificity in Douglas-fir root fungal communities ( Figure 4). Third, we observed significant but variable effects of fire on root fungal colonization (Figure 1), richness ( Figure 2) and community structure (Figure 3), demonstrating that fire-driven shifts in soil fungal communities (Smith et al., 2021) translate to differences in root-associated fungal community assembly. In support of H3, we found negative effects of fire on fungal species richness with both Douglas-fir and oak when growing in Douglas-fir forest soil, and a negative effect of fire on root fungal colonization in oak seedlings in Douglas-fir forest soil. However, while fire lowered fungal richness of both plant species in Douglas-fir forest soil, this trend was either reduced or entirely reversed in oak woodland soil ( Figure 2).
Indeed, Douglas-fir visible fungal root colonization was increased in oak woodland soil both by fire and the presence of a neighboring oak (Figure 1). Yet while fire in this ecosystem type reduced the observable positive effect of oaks on Douglas-fir root fungal colonization likelihood (that is, the chance that a plant's roots are colonized at all by visible fungi; Table S4), as predicted in H4, this difference did not result in Douglas-fir failing to become F I G U R E 5 Heat map of fungal genera classed as mycorrhizal (not only ectomycorrhizal) by FUNGuild (Nguyen et al., 2016) found in root-associated communities of colonized Douglas-fir and oak seedlings treated with soil inoculum from differing ecosystem types and fire statuses, grown alone or in combination with a heterospecific. Degree of brightness indicates the mean abundance of each genus. Green represents communities found on Douglas-fir roots, while yellow represents communities found on oak roots. For the dendrogram, the left branch indicates the first of two given options (e.g. Douglas-fir rather than oak seedling species), while the right branch indicates the second. In the case of binary conditions (e.g. no fire vs. fire), a dashed branch indicates absence (e.g. no fire), while a solid branch indicates presence. A heatmap of ASVs classified to species level is available in the SI ( Figure S2). colonized. Instead, for Douglas-firs establishing in oak woodlands, fire appears to remove the fungal colonization barrier that otherwise would need to be alleviated through oak facilitation, perhaps by activating dormant spores from fire-adapted Rhizopogon species (Figure 5;Peay et al., 2009). In contrast, the effect of Douglas-fir on oak root system fungal richness was most strongly positive with soils specifically from fire-affected Douglas-fir forests (Table S5), where Douglas-fir seedlings facilitated secondary Rhizopogon affiliation with oaks ( Figure 5; Lofgren et al., 2018). In other words, we find that in post-fire landscapes, facilitation by oaks is no longer necessary for Douglas-firs to become fungally colonized in oak woodlands, and the ability of Douglas-fir to diversify oak rootassociated fungal communities in Douglas-fir forests is increased.
These patterns in fungal community assembly and facilitation raise the interesting possibility that our prior understanding of fire's role in bistable forest/savannah landscapes may have been incomplete. A failure to find suitable fungal symbionts can prevent Pinaceae establishment (Dickie et al., 2010;Nuñez et al., 2009).
As members of the Pinaceae, Douglas-fir are considered obligately ectomycorrhizal (Molina et al., 1992). We observe that in our system, fire reduces this potential barrier for Douglas-fir seedlings in oak woodlands just like oak facilitation does (Figures 1 and   2). We thus speculate that a single isolated fire may strengthen subsequent growth of Douglas-fir seedlings through its effect on root-associated fungal community assembly. However, other work shows that frequent fire would likely negative impact Douglasfir populations via direct impacts on seedling mortality and recruitment (Higgins et al., 2000;Hoffmann & Solbrig, 2003). It's not yet clear what these opposing potential effects might mean downstream for the relationship between fire regime and woody encroachment in these types of systems overall.
Indeed, further research is still needed to conclusively determine the consequences that these differences in root-associated fungal community assembly have on the long-term survival and growth of these key tree species, which we do not assess here. Moreover, conditions in situ could differ from those of our growth chamber experiment. Communities of soil microbes besides the externally visible fungi that we investigated can be impacted by fire (Pressler et al., 2019), and such communities of, for example, bacterial and fungal endophytes could in turn also influence plant establishment and growth. Additionally, soil chemistry is altered by fire (Neary et al., 1999), which could affect microbial community assembly, fungal function, and plant growth in situ in ways that our experimental setup does not capture. Translating patterns in fungal community assembly, such as those we have found, into reliable predictions of realized vegetation distribution will require careful consideration of these factors as well as further empirical work.
Positive fire-frequency feedbacks can support stable cooccurrence at landscape scale of closed-canopy forests and ecosystems with lower tree density like woodlands and savannahs (Staver et al., 2011a), but fire suppression is now contributing to a global increase in woody plant encroachment into these ecosystems (Stevens et al., 2017). Though forests can help to buffer climate change by absorbing carbon (Pan et al., 2011), such woody plant encroachment can cause to losses of soil carbon large enough to overwhelm aboveground gains (Jackson et al., 2002). As such, uncontrolled encroachment of trees and shrubs represents a major management challenge for these ecosystem types (Hastings et al., 1997). Our results indicate (1) that fungal symbiont availability could limit growth or establishment of invasive seedlings in some contexts, rendering them dependent on facilitation from native vegetation, but (2) that fire can lift this barrier. Since growth and establishment of invasive seedlings is a proximal driver of woody encroachment, further research investigating potential links between these fungal community patterns and landscape-scale effects may helpfully inform sustainable woodland and savannah management.

AUTH O R CO NTR I B UTI O N S
Gabriel Reuben Smith and Kabir G. Peay designed the project.
Gabriel Reuben Smith carried out research and analysis. Both authors interpreted the findings. Gabriel Reuben Smith wrote the paper with input from Kabir G. Peay.

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors declare no conflict of interest.

S U PP O RTI N G I N FO R M ATI O N
Additional supporting information can be found online in the Supporting Information section at the end of this article.