Endophytic bacterial communities are associated with leaf mimicry in the vine Boquila trifoliolata

The mechanisms behind the unique capacity of the vine Boquila trifoliolata to mimic the leaves of several tree species remain unknown. A hypothesis in the original leaf mimicry report considered that microbial vectors from trees could carry genes or epigenetic factors that would alter the expression of leaf traits in Boquila. Here we evaluated whether leaf endophytic bacterial communities are associated with the mimicry pattern. Using 16S rRNA gene sequencing, we compared the endophytic bacterial communities in three groups of leaves collected in a temperate rainforest: (1) leaves from the model tree Rhaphithamnus spinosus (RS), (2) Boquila leaves mimicking the tree leaves (BR), and (3) Boquila leaves from the same individual vine but not mimicking the tree leaves (BT). We hypothesized that bacterial communities would be more similar in the BR–RS comparison than in the BT–RS comparison. We found significant differences in the endophytic bacterial communities among the three groups, verifying the hypothesis. Whereas non-mimetic Boquila leaves and tree leaves (BT–RS) showed clearly different bacterial communities, mimetic Boquila leaves and tree leaves (BR–RS) showed an overlap concerning their bacterial communities. The role of bacteria in this unique case of leaf mimicry should be studied further.

www.nature.com/scientificreports/ and leaf mimicry has been established 8 . However, deciphering the mechanism behind the exceptional capacity of leaf mimicry in Boquila is indeed a challenging, complex task. Two hypothetical explanatory mechanisms for leaf mimicry in Boquila were outlined in the original study 8 : (i) volatile compounds emitted by trees could modulate gene expression in Boquila, and (ii) microbial vectors could carry genes or epigenetic factors from trees to Boquila that would alter the expression of leaf traits. The first hypothesis could explain the observed leaf mimicry without direct contact and is generally supported by the fact that volatile plant communication is widespread and multi-purpose 18 . Nonetheless, to our knowledge, there is no documented evidence of changes in leaf shape elicited by volatiles and, more importantly, known volatilemediated responses in receiver plants are rather general [18][19][20][21] , while leaf mimicry in Boquila is highly specific. The second hypothesis, the horizontal gene transfer (HGT) hypothesis, has been deemed implausible [22][23][24] . However, evidence from other study systems suggests that the HGT hypothesis is not too speculative. First, HGT has been demonstrated for a number of plant species [25][26][27][28][29] , including transposable elements 30,31 . Second, HGT may have an adaptive value in natural populations: in a grass species, a transgene acquired from a distantly related grass contributes to local adaptation to microenvironmental variation 32 . Third, vector-mediated HGT has been reported for several plant species. Published cases involve transposons as well as microbial vectors such as fungi, bacteria and viruses 29 . Fourth, bacteria themselves may affect epigenetic factors, ultimately influencing gene expression [33][34][35] .
As a first step to unravel the mechanisms behind leaf mimicry in Boquila, and in the context of the HGT hypothesis, here we addressed whether leaf endophytic bacteria are associated with the mimicry pattern. Leaf endophytic bacterial communities have been shown to play significant roles in plant metabolism and ecological interactions 36,37 . Using 16S rRNA gene sequencing, we determined the taxonomic richness and composition of the leaf endophytic bacterial communities in a common Boquila-tree association at a temperate rainforest in southern Chile. Specifically, we compared the endophytic bacterial communities in three groups of field-collected leaf samples: RS = leaves from the model tree species, Rhaphithamnus spinosus (Verbenaceae), BR = Boquila leaves mimicking the tree leaves, and BT = Boquila leaves from the same individual vine but not mimicking the tree leaves (Fig. 1). We hypothesized that, if bacterial vectors are involved in the leaf mimicry phenomenon, the bacterial community from group BR (mimetic Boquila) would be more similar to that from group RS (model tree) than the bacterial community from group BT (non-mimetic Boquila).

Results
A total of 45 bacterial phyla were detected across leaf samples, with Proteobacteria, Thermotogae and Actinobacteria comprising over 75% of taxa (Supplementary Figure S1). Overall, we identified 1571 bacterial OTUs (Operational Taxonomic Units). The average number of bacterial OTUs differed among the three groups of leaf samples (F 2,8 = 11.59, P = 0.004; one-way ANOVA). ANOVA assumptions of data normality and homoscedasticity were met. Tukey HSD tests showed that the number of bacterial OTUs was significantly lower in BT (269.4 ± 38.5, mean ± SE) compared to both BR (435.6 ± 26.4) and RS (589.2 ± 59.2), whereas no significant differences were found between BR and RS in the number of bacterial OTUs. A Venn diagram shows that BT and BR shared 33 unique OTUs, whereas BT and RS shared 79 unique OTUs (Fig. 2). Remarkably, BR and RS (i.e., mimetic Boquila and the model tree) shared 255 unique OTUs (Fig. 2).  www.nature.com/scientificreports/ The PERMANOVA results indicate that there were significant differences in the endophytic bacterial communities among the three groups of leaf samples ( Table 1). The NMDS patterns, based on Bray-Curtis dissimilarity, indicate that-despite the small number of replicates-our hypothesis was verified, i.e., the bacterial community from mimetic Boquila (BR) was more similar to that from the model tree (RS) than the bacterial community from non-mimetic Boquila (BT) (Fig. 3). Whereas non-mimetic Boquila leaves and tree leaves (RS-BT) showed different endophytic bacterial communities, mimetic Boquila leaves and tree leaves (RS-BR) showed a slight overlap in the 95% confidence areas concerning their endophytic bacterial communities (Fig. 3). Accordingly, the distance between the centroids of RS and BT was 0.50, while the distance between the centroids of RS and BR was 0.14 ( Fig. 3). The distance between the centroids of BT and BR was 0.44 ( Fig. 3). There was concordance between observed interobject distances and those predicted from the dissimilarities (final stress = 0.063). Interestingly, the dispersion of points was much greater in the potentially multi-phenotype Boquila (BT) than in the model tree (RS) or in Boquila mimicking the tree (BR) (Fig. 3).

Discussion
We found that mimetic Boquila were closer to the model tree Rhaphithamnus spinosus than non-mimetic Boquila in terms of the composition of endophytic bacterial communities, with over three-times more shared unique OTUs and less than one-third the distance between centroids in the NMDS analysis. Our results suggest the involvement of bacterial agents in leaf mimicry by Boquila, yet we are still far from proving the HGT hypothesis. Thus, here we validate-and promote further research on-the role of bacteria in this unique case of leaf mimicry. Although there were 255 bacterial OTUs exclusively shared by mimetic Boquila and the model tree, and it could be tempting to delve further into this group, here we refrain from attempting to identify particular bacterial taxa that presumably could play a role in the leaf mimicry phenomenon. Such a specific question should be tackled with a different experimental approach, e.g., sequentially excluding particular bacterial taxa and measuring the expression of leaf traits. More importantly, our hypothesis is that bacteria could be just the vectors carrying genetic or epigenetic factors from the tree to the vine. Thus, it would be of little use for the purpose of testing   26,27,29 , which is likely a consequence of their intimate and long-standing contact. Furthermore, known examples of HGT between plants are discrete events that occurred once-or a few times-in evolutionary history 26,29 . These spatial and temporal features of HGT in plants pose significant challenges to the HGT hypothesis in Boquila. On the one hand, to account for leaf mimicry in Boquila we need, regarding spatial aspects, a mechanism that can be effective without contact between plants, and this is why we considered a microbial airborne vector. On the other hand, regarding temporal aspects, we need a mechanism similar to the above described "historical" HGT, but operating at an ecological time-scale, and this is why we considered potential epigenetic roles of such microbial vectors.
There are several cases of crop mimicry in weeds driven by unintentional selection by farmers, also known as "Vavilovian mimicry" 3,38,39 . For instance, mimetic populations of the weed Echinochloa crus-galli and cultivated rice are indistinguishable at the seedling stage, particularly sharing an upright habit of both tillers and leaves 38,40 . In a genomic study comparing mimetic and non-mimetic populations of E. crus-galli in rice paddies, Ye et al. 40 reported that genomic regions harbouring 87 putative plant architecture-related genes were under selection during the differentiation between mimetic and non-mimetic populations, which occurred ≈ 1000 years ago. This study illustrates that, even for a rather simple and widely known mimicry case, elucidation of the underlying mechanisms is a complex task. Therefore, concerning the mechanisms behind mimicry capacities of Boquila, we envision a long road ahead of us.
We need to explain not only how Boquila is able to mimic over a dozen species in terms of leaf shape and size, even without direct contact, or how a single individual vine can mimic two different tree species 8 . We also need to elucidate how this vine can develop a small spine at the leaf tip when twining around-or being close to-species with such mucronate leaves, which include Luma apiculata 8 , Cissus striata 10 , and Rhaphithamnus spinosus (Gianoli, personal observations: a video footage showing this feature is included in the Supplementary Video S2); importantly, the botanical description of Boquila does not include spiny leaf tips 41 . Moreover, concerning the temporal axis of the mimicry phenomenon, and unlike the model-mimic associations lasting for centuries or millennia, we have detected that trailing Boquila vines are able to mimic the exotic herb Ranunculus repens 10 , which was introduced in the study area a few decades ago 42 . A comprehensive research programme aiming to test the HGT hypothesis for leaf mimicry in Boquila will likely include genetic, metagenomic, transcriptomic, proteomic, metabolomic and epigenetic studies, tied to both field and greenhouse experiments. We suggest that such a research programme would eventually crack the code of this amazing plant, and beyond leaf mimicry, help further our understanding of plant phenotypes in general. www.nature.com/scientificreports/

Methods
Study system. The study was carried out at Anticura, Puyehue National Park, southern Chile (40°39′S, 72°11′W; 350-400 m). In this cold temperate rainforest 43 the dominant trees are broadleaf evergreen species and woody vines are fairly abundant 9,44,45 . The main herbivores are slugs, snails, weevils and leaf beetles 46,47 . The woody vine Boquila trifoliolata (Lardizabalaceae, a monotypic genus) is distributed along the whole light gradient in the mature forest 9 . This twining vine has slender stems when young, and leaves are composed of three pulvinated leaflets 41 . The central leaflet is slightly larger than the lateral ones, and leaflets show significant variation in size (10-100 mm) and shape: ovate-elliptical, oblong-elliptical or obovate; leaf tips are lobulated or emarginated 8,41 . The small tree Rhaphithamnus spinosus (Verbenaceae), endemic to the temperate rainforest of southern South America 48 , is commonly found in advanced regeneration stands 49 , but is distributed across the entire light gradient 44 . It is armed with thorns on leaf axils 48 and has simple, opposite leaves (7-35 mm long, 5-25 mm wide) with ovate or cordate-lobed blades and spiny tips 50 .
Field sampling. We located five adult individuals of R. spinosus (height range: 100-170 cm) climbed by Boquila vines. Distance between individual trees was 50-900 m. In each of those five tree-vine associations we collected three groups of samples: RS = two leaves from R. spinosus, BR = a single Boquila leaf (composed of three leaflets) mimicking the tree leaves, and BT = a single Boquila leaf from the same individual vine but not mimicking the tree leaves. In order to carry out a proper comparison, in all cases the mimetic and non-mimetic Boquila leaves were very close (< 60 cm) and their respective distances to tree leaves were nearly the same (video footage included in the Supplementary Video S1). Collected leaves were placed in paper bags with silica gel within a cooler with ice packs, stored at 4 °C and transported to the lab within 48 h for DNA extraction. This study complies with local and national regulations concerning research and field studies on plants in protected areas. Permission for research and collection of plant material was granted by CONAF (Corporación Nacional Forestal) permit No. 012/2018 to EG. Plant species were identified by the first author, who has over ten years of experience working in the study system.
DNA extraction, amplification, and sequencing. For DNA extraction, leaves were sterilized with washes of ethanol (70%), sodium hypochlorite (1%) and water. The success of surface sterilization was confirmed by the absence of any microorganism growing on PDA (potato-dextrose-agar) (Phyto Technology Laboratories) plates from the plating of last washing water. Genomic DNA was extracted from 100 mg dry leaf material using a CTAB-based method 51 . DNA samples were amplified by PCR using the primers 515F (5′-GTG CCA GCMGCC GCG GTAA-3′) and 806R (5′-GGA CTA CHVGGG TWT CTAAT-3′). Primers 515F-806R target the V4 region of the 16 SSU rRNA gene. Since sequencing the V4 region from plant tissues can lead to host-derived plastid and mitochondrial sequences 52  The Illumina MiSeq Platform was used to sequence the amplified V4 region of the 16S rRNA gene from metagenomic DNA samples. The reads obtained were paired-ended with a read length average of 300 bp. The TrimGalore wrapper application [http:// www. bioin forma tics. babra ham. ac. uk/ proje cts/ trim_ galore/] was used to remove adapters and low-quality sequences. The sequencing data was analysed using Mothur software (version 1.38.1) 53 with the default options, unless otherwise stated. Reads shorter than 200 bp were discarded. Reads were denoised using the "pre.cluster" command in Mothur platform to remove sequences that were likely due to errors and assemble reads that differed only by 2 bp. Chimeric sequences were identified and removed, and the remaining sequences classified against the SILVA database 54 using the RDP Classifier program 55 . The table of OTUs obtained from all high-quality sequences was defined at 97% similarity level. Customized perl scripts were used to create OTUs and taxonomy tables.
Statistical analyses. The total number of OTUs was compared among the three groups of leaf samples with a one-way ANOVA (n = 5 individuals per group, with each tree-vine association considered a block), followed by a post-hoc Tukey HSD test. A Venn diagram was constructed using the "VennDiagram" package in R; OTUs present in a single replicate were considered as present for the whole group. We tested for differences in the community structure of endophytic bacteria among the three groups using non-parametric Permutational Multivariate Analysis of Variance (PERMANOVA). PERMANOVA was performed through the R-function "adonis()" from the vegan package on a matrix based on Bray-Curtis dissimilarity from the presence/absence data, taking blocks (= five tree-vine associations) into account using the "strata" argument. We also implemented a two-dimensional non-metric multidimensional scaling (NMDS) based on Bray-Curtis dissimilarity to visualize community structure changes 56

Data availability
The dataset supporting this article is available as part of the electronic supplementary information. Sequences were deposited in the sequence read archive (SRA) of the National Center for Biotechnology Information (NCBI; BioProject PRJNA479681).