Sources of genome sequence datasets are listed in Supplementary Table 1.

Genomic DNA was extracted using a standard extraction method (Starnes et al., 2012) and further purified using the Zymo genomic DNA Clean-up kit (Zymo Research, Irvine, CA, United States). Preparation of libraries for Illumina sequencing were performed using the Illumina Nextera or Roche KAPA HyperPlus kits according to the manufacturers’ instructions. Sequencing was performed using MiSeq machines at the University of Kentucky or Bluegrass Community and Technical College; or HiSeq2000 at Novogene (Sacramento, CA, United States). For MinION sequencing, the extracted genomic DNAs were subsequently purified using the MagAttract HMW DNA purification kit (Qiagen, Germantown, MD, United States) and libraries were constructed using the Ligation Sequencing Kit according to the manufacturer’s instructions. Reads were acquired for ∼ 16 h.

Illumina reads were trimmed and residual adaptor sequences were removed using Trimmomatic. Genome assembly was performed using Velvet version 1.2.10 (Zerbino and Birney, 2008) using VelvetOptimiser¹ to determine the optimal kmer length and coverage. The VelvetOptimiser parameters employed were -s 89 -e 129 -× 2 -shortPaired.

MinION reads were assembled using Canu (Koren et al., 2017) with default parameters. Nucmer and mummerplot (Kurtz et al., 2004) were then used to align the draft assemblies with a chromosome-level reference genome for strain LpKY97, which allowed the contigs to be re-ordered in co-linear fashion. The resulting chromosome-level draft assemblies were then interrogated to identify rare translocations that caused telomeres to align with “internal” positions on the reference genome. If an internally aligned telomere defined the end of a contig in the draft chromosome, the assemblies was “broken” at this position, so that the telomere correctly defined the new chromosome end. To identify breaks at interstitial telomeres, MinION reads were aligned with the US71 reference genome using Minimap2 (Li, 2018) and visualized in IGV (Robinson et al., 2011).

Assembled genomes were masked using a custom masking algorithm from the TruMatch blast post-processor (Li et al., 2005). Masked genomes were then aligned in pairwise fashion and SNPs were called using the iSNPcaller program². The resulting normalized pairwise distance data (SNPs per megabase of uniquely aligned sequence) were used to build a neighbor joining tree using MEGA X software (Kumar et al., 2018). The phylogenetic lineages thus identified were named according to the majority host from which the constituent members were isolated, with numerical suffixes being used to distinguish different lineages from the same host.

Telomere-containing sequence reads were retrieved from raw, paired-end, FASTQ format Illumina datasets using a simple grep search to retrieve reads with exact matches to CCCTAA₃ and TTAGGG₃. Reads containing the latter motif were reverse complemented before clustering the reads using wcdest (Hazelhurst et al., 2008). The resulting clusters were then assembled with MUSCLE (Edgar, 2004) using default parameters, and consensus sequences (telConsensus) were called using a custom script. Nucleotide positions not reaching a majority consensus were represented with Ns. Internal telomere-like sequences were identified as telConsensus that did not start with any permutation of the CCCTAA motif and were discarded from further analysis. Information about the raw read datasets and the strains from which they were derived are provided in Supplementary Table 1.

Raw reads that started with telomere repeats but did not cluster with other telomeric reads, or belonged to very small clusters were considered candidates for de novo telomeres that formed while the strain was being cultured for DNA extraction. These reads were searched against the telConsensus using the low stringency default blast parameters to filter out reads that escaped clustering due to poor quality sequence. The resulting reads were then searched against the relevant genome to identify their original chromosomal positions.

To detect the presence of individual telomere junction (TJ) sequences (Figure 1A) in different Pyricularia strains, we used SeqOthello - a program that accelerates searching of very large raw (FASTQ) sequence datasets using a fast and efficient indexing structure (Yu et al., 2018). Using telConsensus as inputs, “SeqOthello,” iterates through all possible 21-mer subsequences and tests for their presence in raw read datasets, thereby eliminating issues associated with poor telomere assembly. Here it is important to note that single nucleotide polymorphisms in a given SeqOthello search window can prevent detection of TJ conservation. However, for this to be a problem in the present study, a SNP would have to occur within 21 nucleotides of the junction (the query k-mer length used by SeqOthello). This is unlikely when most Pyricularia isolates within a species show much less than 1% sequence divergence, and such instances would be detectable in the outputs. Telomeres attached to MoTeR sequences were removed from the analysis because repeated MoTeR insertion events can give false impressions of TJ preservation. Annotated data outputs are summarized in Supplementary Datasheets 1, 2.

Telomere repeats were stripped off the start of the telConsensus sequences to generate a set of telomere-adjacent sequences (TAS, Figure 1A) which were then used as queries in BLAST (RRID:SCR_001598) searches of more than 200 Pyricularia strains. Information about these strains and access to the sequence data is provided in Supplementary Table 1. To avoid overcalling matches caused by repeated sequences which are often enriched in subterminal regions, we used genomes where all sequences with a copy number > 1 were masked. Valid hits were scored as alignments that covered ≥ 90% of the query. For comparison, a dataset comprising equivalent numbers of randomly selected 280 bp sequences were retrieved from each of the repeat-masked test genomes. Sequences containing NNNs (indicative of gaps/repeats) were discarded and resampled. The random sequence queries were searched against the subject genomes using parameters that were identical to the TJs. As before, valid hits were scored as alignments that covered ≥ 90% of the query. The BLAST results that matched the filtering criteria are provided in Supplementary Datasheet 3.

Telomere-adjacent sequences were used as queries in BLAST searches of the CD156 MinION reference genome (NCBI: BioProject PRJNA320483). Sequences with multiple matches to the genome were filtered out, as were redundant TAS, and the positions of non-redundant sequences having unique “hits” were plotted along the chromosomes using a custom R script³. Vertical displacement of overlapping datapoints was performed using the ggplot “jitter” function to resolve hits with similar positions. The unique BLAST hits used for plotting are provided in Supplementary Datasheet 4.

BLAST searches (-e-value 1e-1 -task BLASTn-short) were used to reveal the locations, lengths, and orientations of MoTeR relics in genomes. Matching sequences were considered relics if they did not belong to telomeric MoTeRs arrays. UNIX grep searches of the MoTeR 3′ end was used to detect matches that were missed by BLAST due to short match lengths. The positions of MoTeR relic-associated duplications are provided in Supplementary Table 2.

BLASTn (e-value = 1e-20) interrogations of the CD156 genome against itself were used to search for duplicate sequences. Duplications linked to MoTeR relics were detected using the Integrative Genomics Viewer (Thorvaldsdottir et al., 2013). Duplications were considered MoTeR-associated if the secondary alignment started within 20 nt of a relic boundary and were retained for further analysis if they were at least 500 bp in length. MoTeR relics and adjacent duplicate flanking sequences were visualized using the R package Circos (RRID:SCR_011798) (Krzywinski et al., 2009). The positions of MoTeR relic-associated duplications are provided in Supplementary Table 3.

Candidates for former telomeres that had been internalized (telomere relics) were identified using a python script to search MinION assemblies for all permutations of the repeat motif (CCCTAA/TTAGGG, CCTAAC/GTTAGG, … etc.) that contained at least 2, 3, or 4 repeat units see text footnote 3. An inexact matching algorithm was employed to account for occasional sequence errors and matches to telomeres and interstitial telomeres in MoTeR arrays were filtered out. The positions of matches on the respective chromosomes were then scaled to allow their relative distances to the telomere to be plotted on a single reference assembly. The relative positions of telomere relics are provided in Supplementary Datasheet 5.

A gene disruption construct was created by sandwiching a hygromycin B resistance cassette between 1.5 kb sequences that flanked the telomerase gene (MGG_01617). Briefly the flanking sequences were amplified from genomic DNA, using the following primers, TERT1-F: 5′ GTTCTCTTCCCGCATTTCAG 3′; TERT1-R: 5′ GGCAAGCTTGGAAAGAACTGCT 3′; TERT2-F: 5′ ATTGAGGATCCCGCATTTCAGTCACG 3′; and TERT2-R: 5′ CCATCTCTAGAGCGAGGCTA 3′ and then purified using a PCR cleanup kit (Invitrogen). The 5′ flank was digested with HindIII and the 3′ flanking fragment with BamHI. The fragments were gel-purified (Qiagen, Inc.) and then ligated with the hygromycin B resistance cassette that had been released from the backbone vector with HindIII and BamHI, using a reaction volume of 5 μl. After overnight incubation, the ligation mix was diluted 100-fold and then the full-length disruption cassette was amplified using the TERT1-F and TERT2-R primer pair. The desired fragment was gel-purified and used directly for transformation.

Protoplasts were transformed using previously reported methods (Peyyala and Farman, 2006). The hygromycin selection plates were surveyed daily to identify transformants (colonies with smooth, hazy borders), and colonies were carefully monitored for sudden cessation of growth (indicative of telomere crisis). Hyphal tips were then carefully excised, being extremely careful not to include hyphae from neighboring, untransformed individuals. A number of normally growing, “ectopic” transformants were picked as controls.

Routine fungal growth was performed using oatmeal agar. Single spore isolation was performed by massaging a colony with a sterilized, sealed glass pipette and then streaking across the surface of water agar. After overnight incubation, a single, germinated spore was isolated under the dissecting microscope and transferred to fresh oatmeal agar plate. For DNA extraction, fungal cultures were grown with shaking for 5 to 7 days in test tubes containing 10 ml of liquid complete medium (6 g casamino acids, 6 g yeast extract, 10 g sucrose per liter).

DNA was extracted as previously described (Rahnama et al., 2020). For Southern hybridization, electrophoretically fractionated DNAs were electroblotted to Pall Biodyne B membranes (Pall Corp., Port Washington, NY, United States) using the manufacturer’s instructions (Idea Scientific, Minneapolis MN, United States). α³²P dCTP- labeled probes were prepared by oligolabeling (Thermo Fisher Scientific, Waltham, MA, United States) according to the manufacturer’s instructions and hybridization was performed using previously described conditions (Starnes et al., 2012).

Seeds of rice cultivar 51583 were soaked in de-ionized (DI) water overnight, sterilized for 10 min by shaking in a 50% bleach solution, and then rinsed in DI water. Pots (2.35″ × 2.15″ × 2.33″) were filled with moistened coarse ground vermiculite and 12-15 seeds were sowed in each pot. The pots were labeled and placed in plastic trays flooded with DI water and covered with a transparent plastic lid. The trays were incubated in a growth chamber using a 27°C, 16 h/21°C, 8 h light (400-500 μE/m-2s-1)/dark cycle with humidity at < 80%. After seedling emergence (∼7 days), the plastic cover was removed, and the trays were watered daily with Hoagland’s solution. Plants were inoculated when the third leaf emerged (approximately 2 weeks post-planting).

Fungal cultures were activated from frozen stocks by placing paper disks on oatmeal agar. Cultures were then grown at 25°C under continuous illumination. After 14 days, the plates were flooded with 10 ml of 0.25% gelatin, the surface of the colony was massaged with a sterilized cell spreader to liberate conidia and the solution was filtered through Miracloth (EMD Millipore, San Diego, CA, United States). Spores were quantified using a hemocytometer and adjusted to 10⁵/ml with 0.25% gelatin.

Plants were placed in a single Mycobag (Garland, TX, United States) containing a few milliliters of water to increase humidity. Aerosol inoculation was conducted using a glass sprayer at 20 psi. The bags were sealed and incubated in the dark for 20 h at 22°C. The bags were then moved to the growth chamber and opened slightly to equilibrate humidity. After 1 h, the pots were removed and maintained in the growth chamber as above. Disease observations and ratings were made 7 days post-inoculation.

A previous analysis of P. oryzae chromosome ends performed in the early days of genome sequencing revealed that the sequences found immediately adjacent to the telomeres (telomere-adjacent sequences, or TASs, Figure 1A) in one strain were often absent from other strains of the fungus; or, if they were present, they were not located at telomeres (Farman et al., 2014). On the other hand, the fact that each strain had “unique” TAS indicated that the chromosome ends are regions of enhanced genomic innovation. Here, we wished to explore the full extent of this telomeric variability by examining genome sequence data from a comprehensive collection of Pyricularia strains/species (P. oryzae, 138 isolates; P. grisea, 4 isolates; P. pennisetigena, 2 isolates; and P. urashimae, 2 isolates). Altogether, these represent 25 phylogenetic lineages (Figure 1B and Supplementary Figure 1) and are variously specialized on 17 different host plant genera (Supplementary Table 1).

First, we focused on the very tips of the chromosomes, with the goal of determining how often different strains have precisely the same telomere structure at a given chromosome end. To this end, we first performed targeted assemblies of telomeric sequences for each strain. Then for each resulting contig, we used SeqOthello (Yu et al., 2018) to ask if the sequence that defines the junction between the telomere and the telomere-adjacent sequences (TAS) is present in any of the other isolates. Among 689 TJs analyzed, we found that 331 (48%) were “private” - i.e., they occurred only in the strain of original discovery. Another 140 TJs showed lineage-specific conservation and, even when junctions were shared, they were not widely distributed because the median number of strains possessing a given TJ was just 3.2 (Figure 1C).

One possible explanation for the poor conservation of TJs is that the P. oryzae chromosomes experience occasional telomere failure. This, in turn, would make them prone to terminal truncations which would result in a corresponding loss of the telomere-adjacent sequences (TASs). To test for such occurrences, we took the TASs identified in each strain as queries, and used BLAST to test for their presence in 204 Pyricularia genome assemblies. As a control, we performed a parallel analysis with equivalent sets of sequences sampled randomly from each of the respective genomes. The randomly selected sequences showed high conservation with a median presence in 186 strains (91%). Only seven of the 516 random sequences analyzed showed lineage−specificity (Figure 1D). In comparison, the TASs were very poorly conserved because, among the 516 single-copy TASs analyzed, 38 were private, 471 were lineage-specific, and the median number of strains that possessed a given TAS was only 22 (Figure 1D). Strains from certain hosts had TASs that were much more poorly preserved than others, with those from rice (Oryza sativa, O) being prime examples. The median number of strains sharing an O-strain TAS was found only 15, and 39% of the TASs (39/98) showed lineage specificity. TASs found in P. grisea, P. pennisetigena and P. urashimae also showed extremely poor conservation in other species but this was to be expected given the significant sequence divergence between species (∼ 10%), which typically allows only ∼60% of their genomes to be aligned. That most of the TASs identified in this study were absent from a majority of strains is certainly consistent with a pattern of occasional telomere failure.

While 331 TJs were private to the strain in which they were originally identified, only 38 TASs showed this pattern. This naturally begs the question, if a TAS is present in a given genome but is not linked to a telomere, where is it? This was addressed by using the TASs as queries to search a chromosome-level genome assembly for the Eleusine pathogen, CD156. The resulting alignments are plotted above the chromosomes shown in Figure 2A. To simplify interpretation of the results, only unique matches were plotted, and the TAS from CD156 were excluded. Eighty-three of the 516 TAS had unique matches in the CD156 genome and most aligned with sequences found in subterminal locations. Predictably, none of the alignments abutted a telomere. In many cases, the positions and orientations of TASs in the CD156 chromosomes were consistent with their having arisen through simple terminal truncations that extended for variable distances in the chromosomes of the source strains (see Figures 2B,C). On the other hand, a number of TASs aligned with positions deep within the CD156 genome interior, or they were in inverted orientation with respect to a nearby telomere. As such, they were not explicable by terminal truncation. In most cases, however, these particular TASs had a second copy in the source genome (Figure 2A), which suggests that these sequences very likely arrived in telomeric positions after having been captured from internal loci during the repair of failed telomeres (Figure 2D; and see below).

Also notable were two situations where TASs from different strains mapped in clusters positioned firmly within the CD156 genome interior. Four strains had TASs that all aligned in the “sense” orientation at distinct positions centered around ∼775 kb from TEL1. Because these sequences were all single-copy in the source genomes, it is clear that chromosome 1 terminated at the corresponding positions in the respective strains. This suggests that they had a common ancestor that experienced a chromosomal break that was healed by de novo telomere formation. The new telomere then appears to have suffered variable length truncations (as illustrated in Figure 2C). A second hotspot for telomere-driven alterations lies in a region between 2 and 3 Mb on chromosome 3. Not only do several TASs map there but, CD156 also harbors a number of MoTeR relics and interstitial telomere sequences in this location (see below). The presence of so many “telomeric features” at an internal location, suggests that these might be footprints of an ancestral chromosome fusion event. Note that we were able to rule out mis-assembly of the CD156 genome as an alternative explanation because it was validated via alignment with chromosome-level assemblies for three other strains (B71, LpKY97-1, and FH).

The distributions of other strains’ TASs on the CD156 reference genome pointed to rearrangements driven by the healing and/or repair of chromosome ends that experienced temporary telomere failure. When telomere maintenance fails, replicative sequence attrition results in the progressive loss of terminal sequence until the chromosome ends are healed by de novo telomere formation, or via alternative telomere maintenance (ATM) processes that add internal sequences onto de-protected ends. If ATM is activated before the telomere repeats are completely lost, the vestigial telomeres will be internalized (see below). In P. oryzae, the canonical telomeres have lengths ranging from ∼20 to 30 repeats (Rehmeyer et al., 2006; Starnes et al., 2012) but, for internalized (interstitial) telomeres, we would expect their lengths to be shorter because repair is not activated until they reach a critical length (∼10 repeats). To identify possible telomere immigrations, we searched chromosomal assemblies of eight strains for interstitial telomere motifs with the various permutations of at least two, three and four repeat units (CCCTAACCCTAA, CCTAACCCTAAC, etc.). On average, bi-repeats are expected to have a random occurrence of approximately six times every 16.8 million nucleotides, or ∼ 36 times in a ∼ 37 - 45 Mb double-stranded genome. Actual occurrences far exceeded random expectations, with an average of 119 per genome (range 73 to 265, Table 1). Tri-repeats and larger also occurred much more frequently than by random expectation (1 in 68.7 Gb), with an average of 10 discrete copies per genome (range 2 to 21, Table 1). These overabundances suggest that a number of these motifs represent former telomeres that were internalized when they were repaired following deprotection. Plotting the relative positions and orientations of these motifs against the CD156 reference genome provided additional support for this hypothesis because it showed that the most of the interstitial sequences were in subterminal regions; and a majority of these were in the orientation expected had they been former telomeres that were repaired after experiencing failure (Figure 2A below chromosome and Figure 2B).

While overabundances of internal (CCCTAA)_n motifs are certainly consistent with the internalization of former telomeres, we cannot rule out the possibility that these motifs evolved independently and over-accumulated for functional reasons. However, additional evidence in support of telomere internalization comes from the presence of MoTeR relics (Figure 1A) at internal chromosome positions (Rahnama et al., 2020). MoTeRs insert specifically into telomeres (Figure 1A), so the presence of very short, 5′-truncated relics with associated telomere motifs at internal locations is best explained by their having origins in terminal positions (Starnes et al., 2012). CD156, has MoTeR insertions in most of its telomeres and 14 internal relics were found distributed among five of the seven chromosomes (Figure 3A). Approximately half were positioned in subterminal regions (within ∼500 kb from a chromosome end), while six were deeply embedded in the chromosomes, at least 2 Mb from the nearest telomere. Most of the relics in CD156 were associated with segmental duplications and, interestingly, mapping of the secondary copies revealed that most reside in subterminal regions and usually very close to the chromosome ends (Figures 3B,C).

MoTeRs cause telomeres to rearrange at spectacular frequencies (Rahnama et al., 2020), which suggests that relics might be continually generated by “illegitimate” repair. In this case, we might expect different P. oryzae strains to be highly polymorphic around relic loci. Using whole genome comparisons, we found that most strains did indeed vary widely in their relic constituency (Figure 3A). Strains B71, FH, and LpKY97, are somewhat exceptional because they share many relic loci with CD156 owing to the fact that they recently inherited significant portions of their genomes from a related Eleusine pathogen (unpublished data). Arcadia2 had 11 relics distributed among all chromosomes (Figure 3A), which was surprising because it doesn’t have any functional MoTeRs in its telomeres (just one drastically truncated insertion in TEL10). All 11 copies resided close to chromosome ends but none matched copies found in CD156. US71 - a strain related to Arcadia2 - also contained 11 relics but these were distributed across five chromosomes, and only five matched copies in Arcadia2. The rice pathogen, Guy11, had just four relics, while U233, from St. Augustinegrass - had only one (Figure 3A).

If the wide strain-to-strain variation in relic copy numbers and positions were due to independent internalization events, we would predict that polymorphic relics define duplication or translocation breakpoints detectable using genome alignments. However, a comparison of CD156 (14 relics) with U233 (1 copy) revealed that, in every case, U233 failed to possess the subterminal relics because the chromosome ends comprised strain/lineage-specific sequences. On the other hand, the more “internal” relics were usually absent due to deletions in the U233 genome where chromosomal synteny was preserved across the loci, but the relics and varying amounts of flanking sequence were simply absent (Table 2). The same pattern held true for most other instances where a relic was found in one genome but not another (Table 2). In a few cases, relic deletion was associated with a larger rearrangement such as a translocation or an inversion, which possibly could have been coincident with, and driven by, MoTeR internalization, although the absence of the relic precluded a definitive conclusion in this regard.

A nice example of relic-related polymorphism is shown in Figure 4. This shows a relic on chromosome 3 of the Lolium pathogen LpKY97 that is conserved in Arcadia2 and Um88324 (signalgrass) but absent in CD156 and Guy11. Alignments of the five loci revealed that overall chromosomal synteny was preserved in all cases but the relic was simply deleted in CD156 and Guy11 (Figure 4). In Guy11, the deletion breakpoints were defined by insertions of the MGL and Pyret retroelements, whereas in CD156, the breakpoints occurred in a nondescript, single-copy sequence. This suggested that the deletion in CD156 occurred independently of any transposon activity.

In addition to relic presence/absence, the chromosome 3 region in Figure 4 was also polymorphic due to strain-to-strain differences in occupancy by other transposons (Figure 4). In many cases, these transposon insertions were “clean” with obvious target site duplications pointing to their acquisition via transposition. Conversely, none of the MoTeR relics had target site duplications, so there was no evidence that any of the internal relics arrived through transposition. Thus, the original MoTeR internalization events would have necessarily required the addition of chromosomal sequences at the 5′ ends of telomeric copies. This, in turn, would have generated a duplication or a translocation - either of which would have created new syntenic relationships. Therefore, to explain the absence of novel synteny in any of our strain comparisons, most internalization events must have occurred prior to the divergence of the various P. oryzae lineages.

The poor conservation of TJ and TAS reported above suggests that P. oryzae routinely experiences failures in telomere maintenance, which is predicted to cause replicative attrition and possibly exonuclease-mediated loss of terminal sequences. To explore the fates of deprotected chromosome ends in P. oryzae, we deleted the telomerase (TERT) gene from two lab strains, 60-3 and 2539, and characterized alterations in chromosome end structure, along with collateral genomic alternations. Both strains were generated through crosses between P. oryzae strains infecting different hosts (Leung et al., 1988; Starnes et al., 2012), such that some telomeres possess MoTeR insertions, while others lack them. In both strains, the deletion of TERT resulted in a period of telomere crisis characterized by a complete cessation of growth - a phenotype from which many strains never recovered. However, a small proportion of “survivor” strains eventually started re-growing, albeit slowly, and these were used to characterize alternative telomere maintenance mechanisms and the fates of the chromosome ends.

Telomerase reverse transcriptase KO lines derived from the strain 60-3 (Starnes et al., 2012), displayed dramatically altered telomere profiles in Southern blots, where the discrete telomeric PstI fragments in the parent strain were replaced with high molecular weight bands whose hybridization intensities were significantly greater than the sum of the individual bands in the parent (Figure 5A). This pattern signals amplification of the telomere repeat sequences. The high molecular weight fragments also showed intense hybridization to a MoTeR1 probe (Figure 5B), pointing to concomitant amplification of MoTeR sequences. MoTeR elements lack PstI sites such that digestion with this enzyme releases telomeric restriction fragments that contain entire MoTeR arrays. To explain the signal amplification and TRF size increases, we surmised that MoTeR sequences and the telomere repeats that separate them experienced tandem amplification at the chromosome ends. To confirm this, we blotted DNA samples that had been digested with MboI, which has sites in both MoTeR1 and MoTeR2, and then probed with the telomere repeat (Figure 5D). This produced strong hybridization signals at positions that correspond to MboI fragments expected for the four possible tandem MoTeR structures (M1-M1; M1-M2; M2-M1; and M2-M2) (Figure 5E). Tandem amplification of MoTeR arrays was confirmed using Nanopore sequencing data for two other 60-3 TERT KO strains (60-3ΔT15 & 17), which also revealed amplification of telomere plus MoTeR-derived, telomere-like sequence, separating each MoTeR copy - (CCCTAA)₂(CCCGAA)₂(CCCAAA)₈CCCGAA. Several reads contained tandem MoTeR arrays that are much longer than any found in the parent strain, with a particularly good example, being a read showing the addition of at least 15 MoTeR2 copies to what was originally a two-element array in TEL-A on the left arm of minichromosome 1 (Figure 5F). Each MoTeR copy was separated from the next by short telomere tracts, which explains the increased intensity of telomeric hybridization signals in Figure 5A. Here, the number of repeats between MoTeR copies will initially depend on the length of the invading (compromised) telomere, the numbers of TTAGGG units in the target, and the register of the invasion event.

Based on the above result, we propose that amplification is initiated when replicative telomere attrition results in the loss of telomeric protein binding, which in turn allows the deprotected end to form a D-loop and invade interstitial telomere motifs between MoTeR copies. This would then initiate cycles of telomere attrition-induced replication (TAIR) to produce tandemly amplified arrays (Figure 5G). Other de-protected, MoTeR-containing chromosome ends could then capture the amplified array through additional TAIR events. Interestingly, in KO strains ΔT6 and ΔT23, the intense signals were at different positions, consistent with the preferential amplification of a different, MoTeR-MoTeR junction in the two strains (Figure 5D). These differences could be because the different TERT KO lines experienced propagation of amplified arrays that were initially established at different MoTeR-containing chromosome ends.

Parallel experiments in other P. oryzae strains indicate that the tandem amplification of subterminal sequences following TERT KO requires the presence of interstitial telomere motifs (manuscript in preparation). This begs the question: what are the fates of chromosome ends that lack such motifs? This was addressed using strain 2539 which has very few MoTeR copies in its genome. TERT deletions in 2539 produced KO strains showing almost complete losses of telomeric hybridization signals (Figure 6A). Furthermore, amplification of MoTeR sequences did not appear to be a dominant ATM mechanism (Figure 6B). To determine the fate of the 2539 telomeres, we used inverse PCR to amplify the sequences adjacent to TEL12 in a number of transformants. This telomere was chosen because it was known to lack MoTeR insertions and would presumably be repaired via a mechanism other than D-loop formation. TAS were successfully amplified from four different TERT KO lines, with 2539ΔT10 yielding two different amplicons. Sequencing revealed that a small telomere vestige was preserved in all cases (Supplementary Figure 2), and the compromised ends were repaired via two distinct repair pathways, with both using other subterminal sequences during the repair process (Figure 6C). 2539ΔT4 had acquired MoTeR sequences - most likely from another chromosome end through TAIR, or possibly via transposition. The other molecules were apparently repaired either by micro-homology mediated TAIR - a telomere-specific variant of micro-homology mediated break-induced replication (MMBIR; Hastings et al., 2009) in the case of 2539ΔT2, or by non-homologous end-joining (NHEJ). Having said this, it is not clear how NHEJ would have allowed telomere vestiges to be joined with sequences occurring at distances that are around 20 to more than 120 kb from a chromosome end (Figure 6D), as this would involve one chromosome end losing approximately 150 bp of sequence, and the other losing tens of kilobases.

In general, it is extremely rare to find telomeric sequence motifs with more than four CCCTAA/TTAGGG repeats at internal genome locations in P. oryzae - probably because long, interstitial telomere tracts experience frequent breakage and healing via telomere addition (Florea et al., 2016; Rahnama et al., 2020). P. oryzae strains from foxtails (Setaria spp.) are notable exceptions because they contain variable numbers of long, interstitial telomere sequences. Remarkably, strain US71 has eight, with lengths of 6, 9, 11, 16, 22, 27, 32, and 34 repeat units - the latter four being native telomere length. Arcadia has a single, 23-repeat internal array (Table 1). A clue to the origin and persistence of these highly labile motifs came from the observation that all instances reside near chromosome ends and are interspersed with, and form the boundary of, a 4 kb repeat sequence that occurs, along with variously truncated copies, in subtelomeric, tandem arrays (Figure 7). This arrangement is strikingly reminiscent of the tandem arrays that are generated in TERT KO strains (Figure 5F) which suggests that they too might have been generated during an episode of spontaneous telomere failure. Normally, interstitial telomeres should be resolved quite quickly by break-healing. Indeed, such events were readily detected near US71 TELs 5, 7, and 12 by aligning MinION reads to the chromosomal assembly and identifying groups of reads that had one end anchored in single copy sequence but then terminated in telomeric sequence at the interstitial telomere (Figure 7). To explain the persistence of multiple extended, internal repeats in these strains, we speculate that they “suffered” telomere failure quite recently; and possibly experience such events in an ongoing and periodic fashion.

The idea that interstitial telomere formation and subtelomere amplification occurs through ongoing episodes of telomere dysfunction is intriguing because it has long been proposed that spontaneous losses in telomere function may serve as a mechanism to spur genomic alterations with adaptive benefits (McEachern, 2008); and, while it has been shown that induced telomere failure can accelerate adaptation in an engineered evolution experiment (Mason and McEachern, 2018), until now, there have been no documented examples of spontaneous telomere failure in a native organism. As part of a study to investigate MoTeR transposition, we transformed a foxtail pathogen, Arcadia2, with a construct containing a copy of MoTeR1, and then screened single spore cultures for alterations in telomere profiles that would signal new insertions. Although MoTeR1 mobilization was not detected, we found that serial culture of one of the transformants (A7) resulted in the disappearance of parental telomere fragments, and the appearance of one band with a particularly intense hybridization signal (Figure 8A) - i.e., a pattern strikingly reminiscent of the telomere profiles seen in TERT KO strains (see Figures 5B,D). While we believe that this was a spontaneous telomere failure event and incidental to the transformation procedure, we cannot not rule out the possibility that it was precipitated by expression of the MoTeR1 transgene. However, a second, truly spontaneous telomere failure event was subsequently identified in a wild-type strain, GrF5-2, when a similar telomere profile showed up in a single-spore from a culture that had recently been purified by single-spore isolation (Figure 8C). Targeted cloning and sequencing of a representative fragment from the intensely hybridizing band revealed that it comprised the previously described tandem repeat bordered by interstitial telomeres (Supplementary Figure 3). Thus, we were able to associate tandem amplification of the subtelomeric repeat, with a spontaneous telomere failure event.

To obtain evidence that tandem amplification of the subtelomeric repeat is a direct response to telomere failure, we deleted the TERT gene from the wild-type Arcadia2 strain. Transformants with KOs were identified based on a sudden arrestation of growth and development of a characteristic hyphal “bubbling” phenotype. DNA was extracted from surviving transformants, and telomere fingerprinting was performed. All of the TERT KO strains showed losses of most parental telomere fragments and the appearance of intense signals with varying amplitudes at the same position as the band seen in the original Arcadia2 transformant (Figure 8B).

The 4 kb sequence codes for a predicted reverse transcriptase (RT) and is bordered by direct repeats. Thus, its structure and content resembles that of a long terminal repeat (LTR) retrotransposon. However, this resemblance is likely coincidental because the RT sequence is absent from most P. oryzae genomes and single-copy in most of the genomes where it is found. The repeat unit is also much shorter than typical fungal LTR-retrotransposons, and the RT gene lacks the integrase and RNase H domains required for transposition. In fact, a blastx search of the nr database revealed the only domain similarity to retrotransposon RTs was in the nucleotide binding region which, intriguingly, showed the closest similarity to telomerase RTs. Given that the 4 kb “element,” and its terminal repeats, are all bordered by telomere motifs (Figure 7 and Supplementary Figure 3), this suggests that it first originated following a telomere failure.

Direct evidence that telomere failure might provide an adaptive benefit came from an unrelated, long-term project where we were investigating the genetics of host specificity. As a control for that study, we would routinely inoculate strain 2539, on a rice cultivar (51583) that it is normally unable to infect (Figure 9A). In one particular experiment (out of dozens), we identified a single lesion on a single plant (Figure 9B). Spores were harvested from this lesion and used to establish a number of pure single-spore (SS) cultures. Inoculation of spores from a representative SS culture (2539ss4) on 51583 produced abundant lesions (Figure 9C), thereby confirming that the strain had acquired a stable gain of virulence.

Due to the uniqueness of this event, we initially suspected that the lesion might have been caused by cross contamination from a virulent progeny of 2539 that was inoculated in the same experiment. In an attempt to rule this out, we compared the telomere fingerprints of 2539ss4 with its parent, 2539MH. At first, this pointed to contamination because nine of the 14 telomeres in the parent culture were missing in 2539ss4 (Figure 9D, black dots), and nine novel fragments were observed in their place (asterisks). In contrast, various 2539 cultures stored at various times over the past 23 years had almost identical profiles to 2539MH, with the only exceptions being due to changes in rDNA and MoTeR-containing telomeres that are known to be unstable (Figure 9D).

Although F1 progeny are expected to share ∼50% of telomeric fragments with the 2539 parent, none of the novel fragments matched telomeres in the other parent used in the cross (data not shown). Therefore, we were unconvinced that the 2539ss4 culture was a contaminant and proceeded to strip the blot and re-probe it with a highly repetitive DNA fingerprinting probe. All cultures had virtually identical fingerprints (Figure 9E), which would not be expected if the 2539ss cultures had inherited 50 % of their genomes from a distantly related strain. Finally, a genome assembly was generated for 2539ss4 and comparison with 2539 indicated that they were clones of one another because the SNP frequency (3.6E-5) was well within typical SNP calling error rates (Farman and Ramachandran, unpublished data). This absolutely ruled out the possibility of contamination, and lead to the conclusion that the dramatic alteration of telomere profiles in the 2539ss cultures must have resulted from a temporary failure in telomere maintenance. Here, it should be noted that intense telomere signals are not expected following spontaneous telomere failure in 2539 because these were not observed in the TERT KO experiments (Figure 6A).