A thallus forming N-fixing fungus-cyanobacterium symbiosis from subtropical forests

Fungi engage in diverse symbiotic relationships with phototrophs. Lichens, symbioticcomplexes involving fungi and either cyanobacteria, green algae, or both, have fungi forming the external layer and much of the interior. We discovered an erect thallus resembling a lichen, yet with a surprising thallus structure composed of interwoven cyanobacterial filaments with numerous fungal hyphae inserted within individual cyanobacterial sheaths, contrasting with typical lichen structure. Phylogenetics identified the fungus as a previously undescribed species, Serendipita cyanobacteriicola, closely related to endophytes, and the cyanobacterium belongs to the family Coleofasciculaceae, representing a genus and species not yet classified, Symbiothallus taiwanensis. These thalli exhibit N-fixing activity similar to mosses but lower than cyanolichens. Both symbiotic partners are distinct from known lichen-forming symbionts, uncovering a phylogenetically and morphologically unprecedented thallus-forming fungus-cyanobacterium symbiosis. We propose the name “phyllosymbia” for these thalli, to underscore their unique symbiotic nature and leaf-like appearance. This finding marks a previously unknown instance of fungi solely residing within structures generated by cyanobacteria.

Methods
We utilized the ITS+28S rRNA dataset to reconstruct the fungal phylogenetic tree. Species and samples for the ITS+28S dataset were selected based on recent studies. Sequence alignment was performed using MAFFT v.7 (https://mafft.cbrc.jp/). For the bacterial phylogenetic tree, we used a second dataset consisting of 16S rRNA gene sequences. The selection of 16S sequences was informed by Cydrasil (https://www.cydrasil.org/) and recent studies. Phylogenetic analyses, including Bayesian inference (BI) and maximum likelihood (ML), were conducted using MrBayes 3.2.7a and RAxML-HPC 8.2.12 via the CIPRES Science Gateway (http://www.phylo.org/). Evolutionary models for each gene region of both datasets were predicted using jModelTest2 2.1.6, based on the Akaike information criterion (AIC). The BI analyses were run for 10 million generations with four chains, sampling every 1000 generations. The initial 25% of trees were discarded as burn-in, and the remaining trees were used to generate the 50% majority-rule consensus phylogram with Bayesian posterior probabilities (BPPs). ML analyses were performed under a GTRCAT model with 1000 bootstrap replicates, resulting in a best-scoring tree with proportional bootstrap values (BS). Phylograms were visualized and edited using the Interactive Tree of Life (iTOL) (https://itol.embl.de/) and Adobe Illustrator. BS ≥ 70 and BPPs ≥ 0.9 were reported above the branches of the phylogenetic tree.