Bacterial, but not fungal, communities show spatial heterogeneity in European beech (Fagus sylvatica L.) deadwood

Abstract Deadwood decomposition and other environmental processes mediated by microbial communities are generally studied with composite sampling strategies, where deadwood is collected from multiple locations in a large volume, that produce an average microbial community. In this study, we used amplicon sequencing to compare fungal and bacterial communities sampled with either traditional, composite samples, or small, 1 cm3 cylinders from a discrete location within decomposing European beech (Fagus sylvatica L.) tree trunks. We found that bacterial richness and evenness is lower in small samples when compared to composite samples. There was no significant difference in fungal alpha diversity between different sampling scales, suggesting that visually defined fungal domains are not restricted to a single species. Additionally, we found that composite sampling may obscure variation in community composition and this affects the understanding of microbial associations that are detected. For future experiments in environmental microbiology, we recommend that scale is explicitly considered as a factor and properly selected to correspond with the questions asked. Studies of microbial functions or associations may require samples to be collected at a finer scale than is currently practised.


Introduction
Forests cover 35% of the land area of Europe with deadwood accounting for, on average, 1.5 m 3 /ha or 7% of the total carbon pool (FOREST EUROPE 2020). Deadwood decomposition is initiated by fungi that are able to colonize deadwood due to their filamentous growth, lower nitrogen requirements, and enzymes necessary to digest recalcitrant molecules such as lignin and cellulose. This initial colonization appears to be primarily stochastic and depends on the species either already present in the living tree or recruited from the soil , Baldrian 2017, Hagge et al. 2019. The fungi grow through the deadwood, encounter hyphae from other species and compete, eventually establishing zones of control from which other fungal species are excluded (Boddy 2000, Heilmann-Clausen and Boddy 2005, Hiscox and Boddy 2017, O'Leary et al. 2018. During decomposition, simpler sugars are freed, which allow prokaryotes to grow in the deadwood and provide nitrogen and other compounds (Hoppe et al. 2014, Tláskal et al. 2021a. Ecological interactions, such as those involved in deadwood degradation, can not be fully understood without taking scale into account (Ladau and Eloe-Fadrosh 2019). At a large scale, measured in kilometres, we might recognize forests as a single type of environment but, inside a forest, there are multiple smaller environments, including deadwood, soil, water, rock surfaces, and air (Baldrian 2016). Even in these smaller environment, such as soil, there can be a large amount of heterogeneity at the metre (Štursová et al. 2016) or centimetre (O'Brien et al. 2016) scales, leading to very different microbial communities and biological processes existing just a short distance from one another. Furthermore, even samples taken several centimetres apart are more likely to represent meta-communities than the actual microbial communities wherein real interactions take place (Cordero and Datta 2016). This is supported by recent work showing that regularly mixing soil samples reduces the microbial richness, emphasizing the importance of divergent microbial communities that exist on the micro scale (West and Whitman 2022).
Organisms interact and influence one another; invertebrates influence fungal communities by influencing the outcome of fungal-fungal interactions (Crowther et al. 2011) and spreading fungi to new environments (Eskalen et al. 2013, Paap et al. 2018, Seibold et al. 2019. In turn, fungi influence bacteria in ways that includes antagonistic interactions (Deveau et al. 2018, Carmona-Hernandez et al. 2019, providing niches for bacterial growth (Boer et al. 2005, Xiong et al. 2022, facilitating bacterial movement through the environment (Christofides et al. 2020, Xiong et al. 2022) and as a carbon source (Brabcová et al. 2016). While the exact nature of the interaction between fungi and bacteria will change depending on the current environmental state (Zhang et al. 2016), we expect that, at small scales, volumes of deadwood will be controlled by a limited number of dominant fungi and those fungi will be associated with a specific bacterial community. Small-scale sampling may thus allow us to analyse the nature of pairwise interactions between the dominant fungus and individual bacterial taxa.
In this study, we compared the microbial communities of deadwood objects from the Žofín forest-an unmanaged, mixed, Euro-pean forest-sampled in the form of approximately 1 cm 3 cylinders to existing composite samples of pooled sawdust from multiple drillings that were collected to represent the microbial community within a whole deadwood object. As bacterial community composition is driven by environmental factors, we hypothesized that fine-scale sampling may reveal the deadwood microenvironment whereas composite sampling will contain communities from several microenvironments. Given current knowledge of fungal zones in deadwood, we hypothesized that fine-scale samples would show lower levels of fungal diversity due to the restricted sampling size and would be dominated by one or two main fungal taxa.

Sampling, DNA extraction, and sequencing
The study was conducted in the Žofínský Prales National Nature Reserve (48 • 39 57" N, 14 • 42 24" E; Figure S1, Supporting Information) in the south of the Czech Republic, which is unmanaged and protected since 1838. Composite samples were harvested from European beech (Fagus sylvatica L.) deadwood objects (decomposing tree trunks) in November 2016 and consisted of pooled sawdust from drilling four holes per log using a 10 mm diameter drill bit. A detailed description of the selected logs, sampling process and sequencing can be found in our previous publications (Tláskal et al. 2021a,b).
In June 2019, 10 logs, five of which were estimated to have died and fallen between 1997 and 2008 and five of which were estimated to have died and fallen between 2008 and 2013, were resampled to acquire fine-scale samples. For each log, three drillings were made with a hollow, 10 mm diameter drill bit to a depth of ∼10 cm. As compression of the wood occurred during the drilling process, individual judgement was used to remove a 1-cm 3 cylinder from the deadwood core. Each cylinder was ground into powder with a mortar and pestle before being processed in the same manner as the composite samples.

Preprocessing and ASV calling
The sequencing data was imported and processed with QIIME2 (Bolyen et al. 2019). The quality scores were visualized and used to inform the trimming thresholds. For the bacterial 16S reads, the first 20 base pairs were trimmed from both the forward and reverse reads, forward reads were truncated at length 235 bp and forward and reverse reads were joined with DADA2 (Callahan et al. 2016) during the ASV calling process. In order to maximize the diversity of microbes detected, DADA2 was run with pooled processing (Kleine Bardenhorst et al. 2022). For fungal ITS sequences, only the forward, full-length amplicon sequences were used. The first 20 base pairs of the fungal reads were trimmed and the DADA2 settings were adjusted to allow a maximum of eight expected errors and to truncate reads when quality scores dropped below 8 in accordance with recommendations for DADA2 processing of ITS sequences (Rolling et al. 2022). Once the ASVs had been identified, taxonomy was assigned by a naïve Bayes classifier. The classifier was pretrained on the Silva 138.1 database (Quast et al. 2013, Yilmaz et al. 2014, with reference sequences extracted using the 515F/806R primers, for bacterial sequences and on the UNITE 8.3 database (Nilsson et al. 2019, Kõljalg et al. 2020) for fungi. After processing, 958 457 total bacterial reads were kept with individual samples containing between 6342 and 108 146 reads with a mean of 23 961 reads and a median of 14 585 reads. A total of 1419 637 fungal reads were kept with the number per sample ranging from 9783 to 193 144 with a mean of 35 491 and a median of 21 257 reads.
ASVs with fewer reads than 0.1% of the mean number of reads per sample were removed as these are more likely to contain cross-contaminants. This removed 4097 bacterial and 2327 fungal ASVs. Bacterial ASVs were further filtered to remove ASVs, which were classified as eukaryotic, mitochondrial, chloroplasts, or unassigned at even the domain level. After filtering, 3297 prokaryotic (one archaeal and 3296 bacterial ASVs) and 956 fungal ASVs were retained.

Diversity metrics
Alpha diversity was measured in terms of the observed number of ASVs (richness), Simpson's index (1-D) and the Shannon index. In order to eliminate the effects of different library sizes, all samples were rarefied to size of the smallest library. Rarefaction has been criticized as statistically invalid (McMurdie and Holmes 2014) and introducing more biases than it removes (Willis 2019), but was used as pipeline changes were unable to fully eliminate the correlation between diversity and sequencing depth (Kleine Bardenhorst et al. 2022).
Beta diversity was calculated using the weighted Jaccard index and, additionally, with the Euclidean distance of Hellingertransformed abundances. Ordination plots were constructed by nonmetric multidimensional scaling (NMDS) with beta diversity distances calculated on microbial relative abundance using all ASVs (McKnight et al. 2019).

Clustering
Clustering of samples was performed using hierarchical clustering of the Jaccard distances and the Ward.D2 method. The optimal number of clusters was the most common number returned from a list of 20 clustering indices. Clusters were bootstrapped with 1000 iterations using the bootstrapped hclust function from the Shipunov R package (Shipunov et al. 2022).

Association networks
To estimate how sampling scales may influence interpretation of microbial communities, we constructed association networks independently for bacteria and fungi. Taxa with fewer than 60 reads were excluded. These networks were generated using the compositionally aware CCREPE algorithm (Faust et al. 2012). We compared the overall structure of the composite and fine-scale networks with the netCompare function of the NetComi R package (Peschel et al. 2020) and identified the overlap of associations with an edge weight greater than 0.5.

Results
Alpha diversity was assessed through three metrics; ASV richness provides a measure of the number of individual ASVs, which were identified, the Simpson index provides a measure of how evenly distributed the ASVs are and the Shannon index provides a metric, which takes into account both evenness and richness. We observed significantly lower levels of bacterial alpha diversity in the fine-scale samples when compared to the composite samples across all metrics (Fig. 1). In contrast to bacteria, there was no significant difference between composite and fine-scale fungal communities (Fig. 1). Bacterial alpha diversity was considerably higher than fungal diversity, e.g. there were approximately five times as many bacterial ASVs as there were fungal ASVs.
Fine-scale samples from each log tended to cluster together on an ordination plot while the majority of the composite samples clustered in the centre of the plot (Fig. 2). This was confirmed by hierarchical clustering analysis, which showed that composite samples generally clustered among one another and not with finescale samples from the same deadwood object ( Figure S2, Supporting Information). Ordination patterns were robust to the distance metric used ( Figure S3, Supporting Information).
We generated CCREPE association networks from the composite and fine-scale bacterial and fungal data ( Figure S5, Supporting Information). Comparison of the networks showed that the composite and fine-scale networks differed for both taxonomic groups with the fine-scale networks showing higher levels of connectivity and lower average dissimilarity with significant differences in clustering. While we observed that a large proportion of associations in the composite networks were also identified in the finescale networks (86% for bacteria, 75% for fungi), there were many associations that were specific to either the bulk or fine-scale networks. In the case of the bacterial association networks, the vast majority of fine-scale associations were not detected in the composite network.

Fine-scale samples show lower diversity than composite samples for bacterial, but not fungal, communities
In line with our hypothesis about bacterial communities, we found fine-scale samples showed significantly lower levels of bacterial diversity compared to composite samples (Fig. 1). Furthermore, the increased community evenness in composite samples suggests that spatial heterogeneity among the bacterial community is obscured when composite samples are formed by pooling sawdust; e.g. combining different bacterial communities results in the increased ASV richness that we observed. We believe that this heterogeneity is caused by differences in the deadwood microenvironment (Moll et al. 2018) but, due the limited sample material, we were unable to test this. This is important for biodiversity surveys as it shows that repeated sampling of the deadwood should result in more ASVs being detected as different microenvironments are sampled.
In contrast to the bacterial results and our expectations, we observed no significant differences in fungal alpha diversity (Fig. 1) between the composite and fine-scale samples. This could be a consequence of the differences in fungal and bacterial biology; fungal hyphae extend throughout the deadwood and nutrients may be transported throughout the hyphal network, allowing filamentous fungi to fully colonize a deadwood object while bacteria are forced to move through the deadwood as nutrients are exhausted or the environmental conditions, e.g. pH, change. This hyphal growth also represents that multcellular fungal structures operate on a much greater physical scale than unicellular prokaryotes and so will be less affected by sampling at different points along the same object. The lack of difference in alpha diversity between the composite and fine-scale fungal further supports the view of complete colonization by a hyphal network and suggests that repeated sampling would not capture any more fungal diversity.
While some samples were dominated by one, or a few, fungal ASVs (e.g. sample_069A-C, Fig. 3), we observed between 5 and 111 (mean: 45.62, median: 43) genus-level fungal ASVs in each sample. Our expectation of a single fungal species dominating a small area of deadwood comes from studies of fungal-fungal interactions on deadwood and the observation of dark interaction zone lines in deadwood (Boddy 2000, Heilmann-Clausen and Boddy 2005, Hiscox and Boddy 2017, O'Leary et al. 2018. These experiments are obviously biased towards those fungi, which grow rapidly and can be cultured. Our amplicon-sequencing results suggest that while certain fungi may compete and exclude one another, the situation in deadwood is more complex and those fungal domains often do not belong to a single fungal species. Alternatively, fungal domains may exclusively belong to a single species but some of those domains may be too small to observe with the naked eye. Technical challenges connected with fine-scale sampling have precluded reaching the necessary spatial resolution to distinguish between these possibilities.

Microbial heterogeneity must be accounted for in order to identify associations
In most environmental microbiology studies, samples are pooled in order to reduce variation (Alteio et al. 2021) and ensure that the sample gives a representative view of the microbial community. This is very useful when conducting a survey of microbial diversity in an environment, although there is evidence that pooling samples can mask rare species and obscure diversity in patterns in various taxa (Avis et al. 2010, Manter et al. 2010, Engel et al. 2012). In addition, pooled samples may give an inaccurate view of the actual microbial communities and their interactions at fine-scales where the exact composition of the subsamples may be relevant (Alteio et al. 2021, Fleishman et al. 2022).
In the NMDS ordinations (Fig. 2) and clustering analyses ( Figure  S2, Supporting Information), the composite samples are located centrally, indicating that they represent an average community of European beech deadwood. However, these average communities fail to capture the heterogeneous distribution of deadwood microbes. For example, two fine-scale replicates from sample_049 show a relative abundance of the bacterial genus Pseudomonas above 74% (Fig. 3) while, in the third replicate, Pseudomonas only makes up 20% of the community. Similarly, the fungal genus Kretzschmaria has a relative abundance of at least 85% in two replicates from sample_007 ( Fig. 3) but makes up only 63% in the third. This variation is likely to play a major role in accurately identifying microbial associations when using abundance data.
We compared the association networks generated from the composite and fine-scale samples and found evidence that the detected associations change according to sampling scale with networks constructed from fine-scale samples to show much higher levels of natural connectivity. We found that, of the total highly weighted associations observed, 96% of bacterial and 34% of fungal associations would be missed if using composite data. In addition, 1% of bacterial and 17% of fungal associations only occur in the composite data and are likely false associations due to pooling taxa, which are physically separated in the deadwood. This suggests that deadwood studies attempting to show interactions between taxa need to adjust their sampling strategies to reflect heterogeneous deadwood microcommunities.
Fine-scale samples are important for understanding possible microbial interactions as the scales over which most sampling is conducted may be orders of magnitude larger than the scales on which microbes interact (Cordero and Datta 2016, Ladau and Eloe-Fadrosh 2019, Alteio et al. 2021. While the composite deadwood samples were collected from three 40-cm drillings and the finescale samples were from a 1-cm 3 cylinder, recent work has shown that microbial communities at the soil-water interface may stratify at millimetre distances with Gammaproteobacteria decreasing from ∼20% to less than 5% relative abundance over only a 6-mm distance (Cai et al. 2022). Xiong et al. (2022) showed that, even in an overall aerobic environment, fungal hyphae can reduce the oxy- gen concentration in the surrounding water film at a distance of up to 13 μm from the hyphal surface. These oxygen-depleted microniches provided the conditions necessary for obligate anaerobic bacteria to germinate and grow. Anoxic microniches might play important role in maintaining active diazotrophs in deadwood (Tláskal et al. 2021a). Sampling at the relevant scale for a particular community is necessary to identify the interactions which are possible and the potential functions which are realized.

Strengths and limitations
As with all studies, there are several limitations and strengths associated with this work. The major strength of this study is the deadwood sampling at a centimetre scale, which allows a much finer grain analysis of microbial communities. However, this also comes with a drawback that material was too limited to perform both DNA extraction and analyse the chemistry of the samples. While we might assume that the chemistry reflects that of the whole deadwood object, there is likely to be variation of the con-ditions within the deadwood, which may explain the patterns that we observe, e.g. wood density (Rinta-Kanto et al. 2016). Potentially, different bacterial communities could be used as an indicator of the deadwood microenvironment but more sensitive techniques will be needed to test this.
As the sampling is necessarily destructive, we lack information about deadwood microbial dynamics at the centimetre scale and of forest microbial communities more generally (Baldrian 2017). The community may be stable at the scale of the whole deadwood object but, through microbial growth and migration and changing deadwood composition during degradation, there may be substantial variation over time in the fine-scale communities. Unfortunately, addressing this problem will likely require whole new techniques as current sequencing approaches require destructive sampling, preventing studying the dynamics of specific regions, and, as a solid and opaque substance, there is little opportunity for nondestructive monitoring techniques such as fluorescent microscopy in deadwood. . Only ASVs with a mean relative abundance greater than 1% are identified, all ASVs with a mean relative abundance less than 1% are included in the grey 'Rare_Taxa' category.

Conclusion
Our results suggest that fungi are evenly distributed in the deadwood but that bacteria are heterogeneously distributed. The bacterial community detected will likely differ depending on the sampling scale, with variation occurring at the centimetre, and probably even finer, scale. This is of particular importance for studies, which wish to identify microbial associations. Fungi do not show a change in diversity with sampling scale and likely do not form domains, which contain exclusively a single species.
Laboratory-based studies of microbial interactions are extremely valuable but do not reflect the diversity found in the environment and there is a need to confirm laboratory concepts in real-world communities. We also identified large numbers of microbial associations that would be missed or incorrectly identified when using composite data. Care should be taken in association studies where fine-scale, heterogeneous community composition matters more than the average community composition.
We recommend that sampling scale is explicitly considered when designing future experiments, both for deadwood specifically and for environmental microbiology more generally. We agree with previous calls to explicitly consider scale during sampling (Ladau and Eloe-Fadrosh 2019, Alteio et al. 2021, Fleishman et al. 2022) and would recommend pooling samples in silico whenever possible as pooling physically samples is irreversible and results in both a loss of fine-scale data and community variability. This may not always be practical due to the amount of time, effort, and/or money required and, in such cases, it is important to properly record the metadata for researchers to understand how the data was sampled in order to assess the validity of different analytical methods. lm.nih.gov/bioproject/PRJNA672674). The fine-scale sequencing samples are available from NCBI BioProject PRJNA910110 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA910110/). The sample metadata and scripts used for the analysis are available on Github (https://github.com/jasonbosch/Bacterial-but-not-fu ngal-communities-show-spatial-heterogeneity-in-European-be ech-Fagus-sylvatica).