Invertron plasmids are thought to be the most abundant group of mitochondrial plasmids in fungi (Griffiths 1995). Of the 54 Lyophyllaceae strains sampled for this study, over half (28) had at least one associated plasmid strain. Studies in Podospora (Van Der Gaag et al. 1998) and Neurospora (Arganoza et al. 1994; Maas et al. 2005) have previously shown intraspecific variation in plasmid presence of fungal isolates. Consistent with this, we found intraspecific variation in plasmid infection for the seven strains we analysed of one Termitomyces species, T. cryptogamus (T. sp. J132, sp. T132, sp. T8, etc.), which has been shown to be a single biological species (De Fine Licht et al. 2006; Nobre et al. 2014). As such, it is likely that in at least some of the species for which our sequenced strains showed no plasmids in this study, strains in the wild exist that do have plasmids.

Our phylogenetic tree (Fig. 1) combines the 54 newly discovered plasmids from the Lyophyllaceae with 18 that were previously reported in other fungi, and shows significant support for clades incongruent with the host–fungal phylogeny. Several clades contain a mixture of plasmids derived from ascomycetes and basidiomycetes. The majority of plasmids analysed in this study were found in the genus Termitomyces, and many of these cluster into different clades with plasmids from other fungi. These incongruencies with the host phylogeny can partly be explained by a long history of differential loss of plasmids among hosts. However, considering the fast mutation rate of these plasmids (Warren et al. 2016) and the relatively low genetic divergence between plasmids of different host genera, horizontal transmission of plasmids between different fungal species probably plays a significant part as well.

A couple of plasmids we discovered in two Calocybe species (pCGr and pCC2) did not conform to the expected invertron structure, with the DNA polymerase and RNA polymerase genes positioned on the same strand. Both plasmids were still capped on each end by terminal inverted repeats. In our phylogenetic reconstruction (Fig. 1), these plasmids form a clade with one of the plasmids associated with Moniliophthora roreri (pMR1), which also has a non-invertron structure (Costa et al. 2012).

Sequence comparison of 12 Lyophyllaceae mitochondrial genomes to our plasmid data set revealed numerous and sizeable plasmid insertions in mtDNA, even in species for which no autonomous plasmid was detected (Fig. 1). Furthermore, in species with known plasmid infections, mitochondrial inserts were often not homologous to the autonomous plasmid. Since most plasmid inserts were fragmented and degraded, these findings suggest most inserts represent ancestral insertion events that occurred between the ancestral mtDNA and an autonomous plasmid that in many cases no longer occurs in the isolate.

Transfer of mitochondrial DNA to linear and circular plasmids has previously been described in fungi (Akins et al. 1986, Akins et al. 1989, Kempken 1989, Mohr et al. 2000) and in plants (Leon et al. 1989). In circular plasmids, such transfers may occur by the same reverse transcription process that drives plasmid replication (Chiang Lambowitz 1997). The transfers we identified in invertron plasmids may also result from replication errors by the plasmid, or a more general process of insertion. The replication process of invertron plasmids is thought to involve a protein primer, using the proteins covalently bound to the terminal inverted repeats. This differs significantly from circular plasmids, which are thought to be able to replicate without a primer but initiate close to a tRNA-like structure.

We found at least two separate occasions of mitochondrial loss of a tRNA gene following the transfer of a copy of that gene to a plasmid (Fig. 1). Leon et al. found that in the maize relative Teosinte, the transfer of a tRNA-Trp gene coincided with its loss in the mtDNA. Warren et al. (2016) suggested that transfer of mtDNA to plasmids contributes to accelerated sequence evolution of such genes due to the higher mutation rate of plasmids. Our findings of conserved tRNA sequences in plasmids show that such sequences may in fact remain relatively free from mutations when the mitochondrial copy is lost and the transferred sequence encodes an essential function.

Mitochondria import many tRNAs from the nucleus, which could be an alternative explanation for mitochondrial gene loss from mtDNA. However, the sudden loss of tRNAs co-occurring with the appearance of a functional copy in a mitochondrial plasmid that appears to be fixed in the population suggests the tRNA function has transferred to the plasmid and not to the nucleus. In the strains that showed loss of a mitochondrial tRNA gene, we found no evidence of gene copies in the nuclear genome assemblies for these lost mitochondrial tRNA genes. In addition, we only observe loss of mitochondrial tRNA genes in Lyophyllaceae mtDNA when a copy of that tRNA gene is present on an autonomous plasmid. If tRNA transcripts were imported from the nucleus, loss of a mitochondrial tRNA would occasionally be expected in the absence of a plasmid copy. Yet, in all 12 complete mitochondrial genome sequences of Lyophyllaceae species available to us, the only two instances of mitochondrial tRNA gene loss occurred when the lost tRNA gene had been transferred to a plasmid.

Ferandon et al. (Ferandon et al. 2008) reported a tRNA-Met gene located on a linear plasmid that had integrated in the mtDNA of its host, the fungus Agrocybe agerita, and suggested that the tRNA gene may have been captured by the plasmid prior to integration. Mitochondrial insertion of plasmids carrying mitochondrial tRNA genes or other mitochondrial genes poses an interesting evolutionary process. First of all, plasmids can be horizontally transferred between different species, creating the potential for indirect horizontal transfer of mtDNA between these species. Second, following mitochondrial loss of a tRNA gene that transferred to a plasmid, the plasmid could reintroduce the tRNA gene to the mtDNA through insertion. This would likely alter the location of the tRNA gene in the mtDNA. It would also nullify any selective benefit the plasmid enjoyed from the transferred tRNA gene if the mtDNA is able to recover the tRNA function.

Plasmid pT123-4 also contains portions of nad2, a mitochondrial protein-coding gene. This shows plasmids are capable of capturing other parts of mtDNA besides tRNA genes, but since captured sequences are generally limited in size, capturing a complete functional gene may be restricted to tRNA genes and small genes like atp8, atp9, and nad4L.

We observe at least five independent acquisitions of tRNA genes by plasmids in the genus Termitomyces, twice of tRNA-Cys and three times of tRNA-Arg (TCG). Why do we only observe transfer of these tRNA genes and no others? It is possible that the capture of mitochondrial tRNAs by plasmids is dependent on certain sequence motifs found only nearby these tRNA genes. However, the intergenic sequences are not well conserved between Termitomyces species making this unlikely. Looking at the codon usage of mitochondrial genes (Fig. 3), tRNA-Cys and tRNA-Arg (TCG) encode anticodons for the least frequently used of all codons in the corresponding mtDNA. If the production of tRNA transcripts by plasmids is lower than that of mitochondrial tRNA genes (as they are not adapted for their transcription), it is possible that plasmids can only meet the mitochondrial demand of rare anticodons. Such captured tRNA genes can then be selectively beneficial if the mitochondrial copy is lost or mitochondrial transcription of that tRNA gene is otherwise inhibited. However, this would not explain why we also find only low-frequency codon-coding tRNA genes in plasmids when the host mtDNA has not lost its own copy. Fungal mtDNA is thought to be generally transcribed in large polycistronic transcripts (Kolondra et al. 2015), which might result in roughly equal transcription rates for tRNA genes. It is possible that low-frequency codons have more unused transcripts available for integration by a plasmid, increasing their chance of transfer to a plasmid.

Capture of tRNA genes coding for codons commonly used in the host mtDNA was shown by Akins et al. (Akins et al. 1989) in Neurospora crassa, where circular plasmids were found containing tRNA-Trp, tRNA-Val, and tRNA-Gly. These three tRNA genes code for codons that appear in fungal mtDNA at relatively high frequencies (Fig. 3). This shows that it is possible for plasmids to adopt high-demand tRNA genes, but for some reason, we did not observe any such cases in linear plasmids.