DNA transposon expansion drives genome plasticity in Diutina catenulata

Abstract

Diutina catenulata is an ascomycetous yeast of environmental and emerging clinical relevance. Genome analysis of strains representing the major phylogenetic lineages of the species revealed a strong genetic structure, with deeply diverged clades displaying limited gene flow, likely reflecting clonal propagation or long-term historical divergence. A notable feature of D. catenulata genomes is an extensive degree of chromosomal instability associated with two DNA transposons, Dici1 and Dici2, present across all lineages, with up to 96 Dici1 copies in the type strain CBS 565^T. We experimentally demonstrate that both elements remain active and mobilize under laboratory conditions. High sequence conservation among Dici copies supports a recent and ongoing transpositional burst reshaping genome architecture. Related elements identified in multiple fungal species define a previously undescribed family of DNA transposons related to the IS630-Tc1-mariner superfamily. These findings provide a framework for studying DNA transposon proliferation and genome evolution in yeasts.

Genome evolution in Diutina catenulata is poorly understood. Here, the authors show that active DNA transposons drive extensive chromosomal instability and shape genome architecture across genetically structured lineages.

Introduction

Transposable elements (TEs) are DNA sequences capable of moving within and between genomes, significantly shaping genome architecture and evolution. They are broadly classified into two major classes based on their transposition mechanisms^1,2. Class I elements (retrotransposons), which transpose via an RNA intermediate through a “copy-and-paste" mechanism, are subdivided into long terminal repeat (LTR) elements, long interspersed nuclear elements (LINE), and short interspersed nuclear elements (SINE). For example, Ty3/gypsy (Metaviridae) and Ty1/copia (Pseudoviridae) are two major superfamilies of LTR retrotransposons that occur in all major groups of eukaryotes. In some large-genome plants, RNA transposons make up the majority of the nuclear genome like in maize (Zea mays) and wheat (Triticum aestivum)³.

Class II elements (DNA transposons) typically move via a “cut-and-paste” mechanism involving a DNA intermediate. In this process, the transposase recognizes terminal inverted repeats (TIRs) flanking the element, excises the transposon from its original genomic locus by generating double-strand breaks, and inserts it into a new target site¹. This mechanism is generally considered conservative, as the element is removed from the donor site prior to reintegration elsewhere. However, increases in copy number can nonetheless occur when transposition is coupled with DNA replication or DNA repair processes, such as gap repair at the donor site or replication-associated transposition^1,4. DNA transposons are commonly classified into four major groups: DDE transposases, Cryptons, Helitrons, and Mavericks/Polintons. Among these, DDE transposons are the most diverse and widespread, comprising multiple superfamilies defined by phylogenetically distinct transposases². Notably, members of the IS630-Tc1-mariner (ITm) superfamily of DDE transposons are broadly distributed across eukaryotes⁵.

Yeast are unicellular fungi that reproduce primarily asexually through budding or fission. They harbor a variety of TEs that have contributed to genome diversification, gene regulation, and adaptation to different environments. In Saccharomyces cerevisiae, TEs are predominantly represented by LTR-retrotransposons named Ty elements, which resemble retroviruses in structure and replication mechanism⁶. The abundance and activity of these elements vary among yeast species and strains, reflecting both phylogenetic divergence and ecological pressures⁷. Some yeasts also possess DNA transposons, although these are less common and often degenerate^8,9.

Through insertional mutagenesis, homologous recombination between TE copies, and generation of chromosomal rearrangements, TEs actively promote genome plasticity in yeasts, and can drive genetic innovation and adaptation^10,11. However, the same factors can threaten genome integrity^12,13. To mitigate these risks, host organisms have evolved sophisticated regulatory mechanisms, including RNA interference, epigenetic silencing¹⁴, and domestication of TE-derived sequences¹⁵. An interesting example in yeast is the domestication of two distinct DNA transposons for mating-type switching in Kluyveromyces lactis: α3 with homology to the transposase of mutator-like elements (MULE) and KAT1 to hAT transposases^16,17.

Diutina catenulata (formerly Candida catenulata) is an ascomycetous yeast belonging to the taxonomic order of Serinales (also known as the CUG-Ser1 clade) and to the Debaryomycetaceae family^18,19. This species is predominantly associated with environmental and food-related niches, particularly dairy products such as cheeses^20–24, where it contributes to the yeast diversity of various French cheeses^25–28. Its prevalence in the gastrointestinal tract of poultry²⁹ and fish³⁰ has also been well documented, as well as its occurrence in the feces of pigeons³¹.

In addition, D. catenulata has been identified in clinical specimens, including blood and mucosal surfaces. It has the capacity to colonize the digestive tract and trigger both superficial and deep infections^32–34. These findings underscore its emerging pathogenic potential in immunocompromised individuals. Despite limited reports of human infection by D. catenulata, the emergence of multidrug resistance, particularly against frontline antifungals (azoles and echinocandins), raises significant concerns for clinical management should invasive infections arise^34,35. The dual ecological role of D. catenulata underscores the importance of understanding its genetic diversity, adaptive strategies, and potential impact on food safety and human health.

To date, only two studies have characterized the D. catenulata genome. O’Brien et al.³⁶ produced the first draft genome assembly from a soil isolate (accession: PJEZ01000000), thus confirming its placement within the Debaryomycetaceae family. Subsequently, Boden et al.³⁷ reported the genome assembly of the type strain CBS 565^T (accession: JBKQXV010000000), identifying virulence factors commonly found in Candida albicans. However, although genome annotation was reportedly performed in both studies, the available datasets contain only raw genome sequences, limiting their utility.

In the present study, we assessed the genetic diversity and population structure of D. catenulata. High-quality genome assemblies and annotations were generated for five representative strains spanning the species’ phylogeny. We provide evidence of a genome invasion by a previously unidentified DNA transposon, named Dici, which defines a previously unrecognized family in the ITm superfamily and likely contributes to the extensive chromosomal fragmentation observed in D. catenulata.

Results

Marked population structure in D. catenulata

A collection of 61 D. catenulata strains was assembled from French cheeses with a Protected Designation of Origin (PDO)²³ and from international yeast collections and laboratories, representing eight geographical and five ecological origins (Supplementary Data 1). Flow cytometry revealed that most strains are haploid (n = 51), including the type strain CBS 565^T, while seven are diploid. Short Illumina sequencing reads were obtained for the 61 D. catenulata strains and were mapped to the assembled genome of strain CBS 565^T (see bellow). A total of 508,773 variants were identified, including 436,405 SNPs and 72,368 small insertions and deletions (indels). The overall nucleotide diversity (π, the average number of pairwise differences per nucleotide site) was estimated to be 6.2 × 10⁻³ (Supplementary Data 2). This value is close to the values obtained in other yeast species associated with anthropized environments, such as S. cerevisiae (π = 3 × 10⁻³) and Lachancea thermotolerans (π = 9.3 × 10⁻³), while lower than that observed for Kluyveromyces lactis (π = 2.8 × 10⁻²)^38–40. In the case of D. catenulata, the observed π value may be influenced by a sampling bias resulting from the fact that the majority of the strains analyzed were from French cheeses (n = 48) and by the limited number of clinical isolates (n = 9).

A high-quality subset of 351,942 SNPs was employed to construct a phylogenetic tree, which revealed six distinct clades, hereafter referred to as clade #1 through clade #6 (Fig. 1a). The population structure was analyzed using a pruned subset of 34,101 SNPs that are in approximate linkage equilibrium with each other (r² < 0.5) and the Bayesian model-based clustering approach implemented in fastStructure (Fig. 1b). The analysis yielded a most probable population structure comprising six subpopulations, which correspond to the phylogenetic clades identified.

Fig. 1. Phylogeny and population structure of the 61 D. catenulata strains.

a Maximum-likelihood phylogenetic tree based on 351,942 SNPs (IQ-TREE, evolutionary model: TVM+F) revealing six clades. Bootstrap values for major branches are shown in red. b Population structure inferred from a subset of 34,101 SNPs in approximate linkage equilibrium (r² < 0.5). The vertical axis represents the fractional assignment of each strain (horizontal axis) to the resolved genetic groups (colors) under the assumption of K = 5 ancestral populations. Numbers indicate the clades shown in (a).

The average SNP difference was used to quantify the divergence between clades (Supplementary Data 2). The lowest values were observed between the clades #1, #2, and #3, ranging from 114,148 to 132,156. In contrast, clade #5, which comprises four heterozygous diploid strains, showed the largest differences relative to the other clades, with values ranging from 196,002 to 235,860 SNPs.

Clade #1, which exhibited the highest level of nucleotide diversity (π = 1.9 × 10⁻³), comprised six strains (UMIP805.63 was a homozygous diploid of strain CBS 565^T) isolated from human subjects, with the exception of CBS 6174, which was isolated from a milk sample obtained from a cow with mastitis. Clade #2 comprised three distinct cheese strains. Two of these strains were isolated from cheeses of PDO43, and were obtained from two different producers (43B and 43F). These two strains differ by only 51 SNPs. The diversity within this group was the lowest observed (π = 7.1 × 10⁻⁶) across the clades. While clade #3 (n = 5) is the most diverse concerning its ecological origin (human, water, soil and flower), a low nucleotide diversity of 1.8 × 10⁻⁵ was observed, with less than 541 SNPs between strains.

Clade #4 (n = 12; π = 3.4 × 10⁻⁵) is almost exclusively composed of cheese strains, including three pairs originating from the same PDOs (PDO9, PDO18, and PDO25). Two of these pairs were obtained from the same producers but were isolated in different production seasons. The strains from producer 18B were labeled 18B1 and 18B2 for seasons 1 and 2, respectively, and the same labeling scheme was applied to the strains of producer 25B (25B1 and 25B2). The only strain of clade #4 that was not isolated from cheese, but from a human hemoculture sample (strain CNRMA20.419), differs from any of the clade #4 cheese strains by no more than 851 SNPs.

The four heterozygous diploid strains in our collection, all isolated from cheese, presented high levels of heterozygosity (heterozygous SNPs/homozygous SNPs = 0.8) and were found to cluster in the clade #5. This clade displayed the highest diversity (π = 5.7 × 10⁻³), with strains differing by up to 9,103 SNPs. The population structure analysis revealed that these strains exhibited a mosaic pattern, indicative of a genetic contribution from members of clade #6 (cheese), clade #1 (human/animal) and an as yet unidentified contributor (illustrated in green on Fig. 1b). The large clade #6 (n = 31; π = 7.2 × 10⁻⁵) was composed of strains isolated from cheese. Several were from the same PDO: e.g. PDO43 (n = 6) and PDO24 (n = 4), but from different producers or at different seasons. This clade contains two homozygous diploid cheese strains. It also includes an isolate obtained from a case of stomatitis (CBS 564) that does not grow at 35 °C, suggesting that this strain is opportunistic.

Interestingly, strains isolated from cheeses with the same PDO were found in different clades. For example, strains isolated from PDO18 and PDO25 cheeses had representatives in clades #4 and #6. Strains originating from PDO43 cheeses were found in clades #2 and #6. In the latter case, strains 43B2s-11 (clade #2) and 43B1s-15 (clade #6) were obtained from the same producer, 43B in two different seasons.

In order to validate the genetic structure of the yeast population, an Analysis of Molecular Variance (AMOVA) was performed based on the SNP data (Supplementary Data 2). The results indicated that a large proportion of the total genetic variance (93.6%) could be attributed to variations among the six predefined clades, while the variance among individuals within clades was found to be negligible. These findings indicate a robust genetic structure of the D. catenulata population, with clades representing well-differentiated lineages.

Pairwise fixation indices (F_ST) supported this observation, with high levels of genetic differentiation observed between clades, consistent with long-term divergence and limited gene flow between groups (Supplementary Data 2). In addition, elevated Tajima’s D values (D > 0.2 for clades #4 and #6; D > 0.8 for the others) indicate a decrease in population size and/or balancing selection. When clades were analyzed jointly, Tajima’s D was still consistently positive (D = 1.83). This pattern is most likely explained by a strong population structure, which inflates estimates of intermediate-frequency polymorphisms.

Overall, these results indicate that, despite the overrepresentation of cheese strains in the dataset, most D. catenulata clades show a clear pattern of specialization according to the environmental origin of the isolates, with the exception of clade #3. They also highlight a high level of genetic differentiation among clades. We hypothesize that these lineages may be reproductively isolated, potentially as a consequence of clonal propagation and historical divergence.

Comparative analysis of 5 representative genomes

Four D. catenulata haploid strains (43B2s-11, CLIB 1179, 25B2s-01 and 1F1c-01), in addition to the type strain CBS 565^T, were selected from the above genetic clades (excluding clade #5, which contains only heterozygous diploids). Their genomes were sequenced using long reads, assembled, and comprehensively annotated, providing a foundation for comparative genomic analyses. A review of the key genomic features is presented in the accompanying documents (Supplementary Information and Supplementary Data 3).

Analysis of the synteny surrounding the mating-type-like (MTL) locus in D. catenulata revealed a conserved genomic architecture similar to that described in Candida anglica⁴¹ and C. railenensis⁴², in which mating pathway genes are located between the RCY1 and PIK1 genes (Supplementary Fig. 1). Two distinct versions of the MTL locus were identified among the five assembled D. catenulata genomes. The first version, containing a single gene homologous to the transcriptional activator MTLα1, was detected in strains CBS 565^T and 1F1c-01. The second version, present in strains 43B2s-11, CLIB 1179, and 25B2s-01, lacked detectable a1/a2 or α1/α2 genes. To assess the distribution of these two locus configurations (solo α1 versus no detectable MTL gene), short reads from the 61 D. catenulata strains were mapped to the corresponding locus sequences (Supplementary Data 1). The α1-containing locus was detected in four strains of clade #1 and all strains of clade #6, whereas the alternative locus version was observed in strains of clades #2, #3, and #4. Both locus types were detected in the heterozygous diploids of clade #5. A gene coding for a protein with similarity to α2, a transcription factor that plays a central role in mating-type-specific gene regulation, was identified in all five assembled genomes, but it was located outside the canonical MTL locus. Taken together, the presence of incomplete and alternative MTL locus configurations, combined with the apparent absence of the a1/a2 counterparts, suggests that the mating pathway of D. catenulata may be impaired. We hypothesized that this limitation could lead to clonal propagation and contribute to reproductive isolation, potentially reinforcing the pronounced genetic structuring observed among clades.

Despite limited reports of human infection by D. catenulata, the emergence of multidrug resistance, particularly against frontline antifungals (azoles and echinocandins), prompted us to survey genes associated with antifungal resistance and virulence (Supplementary Information and Supplementary Data 4). Core drug targets, including ERG11 (lanosterol 14-α-demethylase, target of azoles) and FKS1 (β-1,3-glucan synthase, target of echinocandins), are present as single-copy genes. Their presence alone does not confer resistance; specific mutations that alter drug binding or target activity are required. However, genes encoding multidrug efflux transporters (CDR1, MDR1) and hydrolytic enzymes (PLB1, SAP5) were identified in multiple copies. Gene copy-number variation affecting efflux pumps and their regulators has been shown in pathogenic yeasts to enhance drug tolerance and, in some cases, virulence⁴³. Together, these observations indicate that D. catenulata harbors a conserved genetic repertoire with the potential to support antifungal tolerance and host interaction, despite its comparatively low clinical prevalence relative to other Candida species. Furthermore, this repertoire does not differ between the clinical isolate CBS 565^T and the cheese-derived strains, suggesting that the genetic potential for antifungal resistance and pathogenicity is conserved in environmental and clinical isolates.

A DNA transposon has invaded the D. catenulata genome

Like many fungal genomes, D. catenulata harbors a diverse repertoire of transposable elements, including both retrotransposons and DNA transposons. The main Class I transposable elements (retrotransposons) identified in the D. catenulata genomes belong to two superfamilies: Ty3/gypsy (Metaviridae), for which four families were found in D. catenulata, and Ty5/copia (Pseudoviridae), for which two families were detected (Supplementary Data 5). Together, these elements represent a relatively small fraction of the genomes of the 61 D. catenulata strains (Fig. 2a). Retrotransposons of the Ty5/copia superfamily are the most abundant, being particularly prevalent in seven strains, accounting for up to 0.5% of the genome (copy numbers ranging from 6 to 13). In contrast, some strains (CBS 2821, CBS 2014, and CNRMA20.419) are almost devoid of retrotransposons. LINE retrotransposons (L1 superfamily) were detected, while mainly found as truncated or degenerate copies (Supplementary Data 5).

Fig. 2. Transposon content in D. catenulata genomes.

For each strain, short-read sequences were mapped to individual TE sequences using BWA. Abundance and copy number were estimated based on normalized read depth, as described in the Methods. a Abundance (left y-axis) and estimated copy number (right y-axis) of LTR retrotransposons (Ty5/copia and Ty3/gypsy). The upper panel shows the abundance of TE internal regions, whereas the lower panel shows the abundance of the corresponding long terminal repeats (LTRs), permitting estimation of total LTR content, including solo LTRs. b Abundance (left y-axis) and estimated copy number (right y-axis) of the Dici1 and Dici2 DNA transposons. In both panels, strains are grouped by clade using the same color scheme as Fig. 1b.

During the annotation curation of the five D. catenulata genomes, a repeated sequence containing three coding sequences (CDSs) was identified (Fig. 3a). It is flanked by two terminal inverted repeats (TIRs) of 8 or 33 bp, starting with the sequence CAGTCGGT. A systematic duplication of TA dinucleotides was observed on each side, a consequence of a target site duplication (TSD) commonly found at the insertion sites of transposons of the ITm superfamily⁴⁴.

Fig. 3. Characteristics of Dici DNA transposons detected in D. catenulata.

a The transposon is flanked by short terminal inverted repeats (TIRs), and a target site duplication (TSD) was observed at both extremities (TA). The three coding sequences (CDSs) are drawn to scale. The transposase gene is located in the center of the element. b Schematic representation of the main features of Dici1 and Dici2 transposases: a pair of helix-turn-helix motifs (PAIRED domain, orange and red regions), each consisting of three α-helices (H in blue). A conserved GRPR-like motif (also known as an AT-hook) is present between the two HTH motifs. Four putative nuclear localization signals (NLS) were detected in conserved regions. Finally, the catalytic triad DDE (highlighted with purple circles) is present within the catalytic domain (blue region), with 34 amino acids between the second aspartic acid (D) and glutamic acid (E). A full-length alignment of the Dici transposases can be found in Supplementary Fig. 7. c Nucleotide composition at the insertion points was visualized using sequence logo. The number of sequences used is indicated (n = 157).

The CDS located at the center of the element was found to encode a transposase (see below). This putative transposable element (TE) was designated Dici, for D. catenulata invader. Two types of Dici were identified, designated Dici1 and Dici2, with lengths of 4.0 kb and 4.2 kb, respectively. The two elements showed marked divergence, with 52.2% nucleotide identity, but shared an identical GC content of 42.7%, considerably lower than the 53.5% observed for the D. catenulata nuclear genome.

Dici1 was predominantly identified in CBS 565^T (96 full-length copies), while Dici2 was found in the genomes of CLIB 1179, 43B2s-11, 25B2s-01 and 1F1c-01, with 8, 28, 39 and 45 full-length copies, respectively (Supplementary Data 6). Four cases of insertion of a Dici1 transposon into another Dici1 element were observed for strain CBS 565^T. Miniature inverted-repeat transposable elements (MITEs) derived from Dici1 were identified in the genomes of the five strains, with a length of 1053 bp (Supplementary Data 6). Neither Dici1 nor Dici2 were detected in other Diutina species (see Methods; nt database downloaded on 23 May 2025).

Dici elements are widely distributed in D. catenulata

We assessed the diversity of Dici elements across the D. catenulata assembled genomes (Supplementary Data 6). Within individuals, Dici1 and Dici2 were highly conserved. For instance, the 96 full-length copies of Dici1 of CBS 565^T exhibited an average pairwise nucleotide identity of 99.9%, while Dici2 exhibited a comparable level of conservation, with identities ranging from 99.6 to 99.8% in strains 43B2s-11, CLIB 1179, 25B2s-01, and 1F1c-01. Likewise, the comparison of Dici2 elements across strains (CBS 565^T was excluded as it only contains two copies of Dici2) revealed low levels of nucleotide divergence, with identities ranging from 99.4 to 99.6%. These values are more than 1% above those obtained from whole-genome comparisons, where identities ranging from 97.7 to 98.2% were observed when comparing the genomes of strains 43B2s-11, 1F1c-01, 25B2s-01 and CLIB 1179 (Supplementary Data 6). These results indicate that Dici elements were highly conserved within strains and exhibited limited diversification between clades. This finding suggests that Dici have undergone recent amplification and have exhibited minimal evolutionary divergence among clades.

We estimated the genomic abundance of Dici1 and Dici2 elements across the 61 D. catenulata strains in our collection (Fig. 2b). Overall, Dici2 appeared to be the most prevalent element, with an estimated copy number exceeding 10 in 54 strains. Notably, strains CBS 564 (clade #6), CBS 6092 (clade #3), and CBS 2014 (clade #1) showed particularly high copy numbers, reaching 85, 69, and 63 copies, respectively. The clinical isolates CNRMA12.1409 and CBS 6092 (clade #3), which are differentiated by a mere 255 SNPs, exhibited markedly divergent Dici2 contents (6 versus 69).

In contrast, Dici1 was generally less abundant than Dici2. The only exception was observed in strains CBS 565^T and UMIP805.63 from clade #1, in which the copy number exceeded 90. Interestingly, the four heterozygous diploid strains (clade #5) displayed similar copy numbers of Dici1 and Dici2, with approximately 20 copies of each element. Conversely, low copy numbers of either Dici1 or Dici2 were observed in strains CBS 2015 (clade #1) and PYCC 7045, CNRMA12.1409, and CLIB 1179 (clade #3). In strain CBS 2821 (clade #1), only a single Dici1 element was detected.

The structural organization of Dici1 and Dici2—including their similar length (~4 kb), nearly identical TIRs (first 8 bp fully conserved), and three CDSs arranged in the same order and orientation—indicates a shared origin. Despite this, the two elements are highly divergent at the nucleotide level (52.2% identity), reflecting long-term independent evolution. Within individual strains, however, each element shows remarkable sequence conservation, consistent with recent amplification events that have allowed limited accumulation of mutations between copies.

Massive genomic rearrangements in the D. catenulata genome

The five D. catenulata strains selected as representative genomes exhibited marked variation in their karyotype profiles (Fig. 4a). This prompted us to further investigate the chromosomal patterns among strains within the same clade (Supplementary Fig. 2). Strains from clade #1 exhibited several variations in their karyotype profiles. For example, strains CBS 2014 and CBS 565^T displayed distinct patterns despite only 1302 pairwise SNP differences between their genomes. Within clade #2, strains 43B2s-11 and 43F2s-01, isolated from cheese samples of the same geographical origin (PDO43), showed identical chromosomal patterns, whereas strain 22C1s-06 displayed a markedly different karyotype, despite a low number of pairwise SNP differences (188 and 197, respectively). Similarly, strains isolated from the same dairy in two different seasons (25C1s-04 and 25C2s-03) within clade #6 displayed identical karyotypes, whereas the three other strains exhibited distinct patterns. The diploid strains of clade #5 showed similar karyotype patterns, although slight size differences were observed among the largest chromosomes. Conversely, identical karyotypes were observed among all strains examined within clade #3. Collectively, these results reveal extensive inter- and intra-clade chromosomal rearrangements in D. catenulata, with the extent of karyotype variation appearing to be clade-specific, ranging from no detectable differences in clade #3 to marked chromosomal reshuffling in clades #1, #2, #5, and #6.

Fig. 4. Chromosomal rearrangements in the D. catenulata genomes.

a Karyotypes of five D. catenulata strains showing distinct chromosome size patterns. Numbers on the left indicate the sizes of S. cerevisiae chromosomes used as markers. b Visualization of synteny blocks along chromosomes. High-quality genome assemblies generated from long-read data and their annotations were used to infer synteny blocks using SynChro (see Methods). The six chromosomes of strain CBS 565^T were used as a reference and assigned distinct colors. These colors indicate the positions of synteny blocks in the other genomes. The numbers of synteny blocks, synteny block breakpoints, and breakpoints associated with Dici elements are indicated above each plot. Scale bar, 1 Mb. c Reciprocal translocation events inferred from shared synteny blocks involving chromosomes B and E of strain CBS 565^T and chromosome E of strain 1F1c-01. Dici1 transposons are shown as pale blue triangles. A close-up view of breakpoints involving a Dici1 element on chromosome B of CBS 565^T is shown in the lower part. Synteny blocks are colored according to the corresponding chromosomal regions in the schematic above. d Schematic representation of an inversion arising from recombination between identical Dici1 copies arranged in inverted orientation. Graphical symbols are as in (c).

This observation was supported by whole-genome alignments and comparative synteny analyses among the five reference genomes, which revealed substantial chromosomal reshuffling (Fig. 4b and Supplementary Fig. 3). The highest number of synteny blocks (n = 66) was observed in the comparison between CBS 565^T (96 Dici1 copies) and 1F1c-01 (45 Dici2 copies), with Dici elements involved in more than 20% of the observed breakpoints (Supplementary Data 7). These breakpoints are consistent with ectopic recombination between Dici copies at distinct genomic locations (Fig. 4c; Supplementary Fig. 4). We also used the CLIB 1179 genome as a reference, as it contains the fewest Dici elements (2 Dici1 and 8 Dici2), making it less likely to be affected by ectopic recombination between copies, which is consistent with the absence of karyotype variation between at least four strains in this clade (Supplementary Fig. 2). Similarly, a high proportion of breakpoints were associated with Dici elements, with comparisons against CBS 565^T and 1F1c-01 yielding 18 and 50 breakpoints, respectively, and Dici elements present at 72% (13/18) and 20% (10/50) of the boundaries (Supplementary Fig. 3; Supplementary Data 7). Although most rearrangements are interchromosomal, others result from ectopic intrachromosomal events, as illustrated by an inversion arising from recombination between Dici1 copies in an inverted orientation (Fig. 4d).

In some cases, breakpoints lack any detectable Dici elements, particularly in strains with the fewest Dici copies. For instance, a comparison of the CLIB 1179 and 43B2s-11 genomes revealed seven breakpoints without associated Dici elements. We hypothesize that simultaneous excision events generating double-strand breaks (DSBs) on different chromosomes may subsequently be repaired by error-prone pathways, such as non-homologous end joining (NHEJ), without leaving a detectable sequence signature at the junction. A similar observation was made during experimental evolution, which demonstrated Dici transpositional activity (see below). However, further research is required to confirm this hypothesis.

In summary, these observations suggest that Dici elements may contribute to structural plasticity in the D. catenulata genome. While such plasticity could potentially enhance adaptability, it may also reduce hybrid fertility and limit gene flow, which could, in turn, contribute to the pronounced genetic structure observed between clades. These results, combined with RNA-seq data showing the expression of all three CDSs under the conditions used for genome annotation (Supplementary Fig. 5), are consistent with the possibility that Dici elements undergo rapid, strain-specific propagation in D. catenulata genomes. This motivated direct experimental assessment of their mobilization and potential impact on genome evolution.

Experimental evidence for Dici mobilization

To test whether Dici1 and Dici2 retain mobilization capacity, we subjected two D. catenulata strains to experimental evolution. Strains CBS 565^T and 1F1c-01, which contain the largest number of Dici copies, were grown in rich medium for around 100 generations (Fig. 5a). For each strain, long-read sequencing was performed on pooled DNA extracted from three independent evolved lines to increase transposition detection sensitivity.

Fig. 5. Experimental evolution workflow and karyotype analysis of evolved clones.

a For each ancestral strain (D. catenulata CBS 565^T and 1F1c-01), three independent replicate lines were obtained, accumulating  ≈ 100 generations (see Methods). Genomic DNA was extracted from the three replicate lines, and pooled prior to ONT long-read sequencing. In parallel, several clones were isolated from replicate lines (n = 4 for evolved CBS 565^T populations and n=3 for evolved 1F1c-01 populations) and subjected to PFGE karyotype profiling as described above. Clones and evolved lines were tested for cross-contamination by PCR using primers specific to a chromosomal breakpoint discriminating the ancestral strains CBS 565^T and 1F1c-01 (see Methods). b Karyotypes obtained for the ancestral strain CBS 565^T and four evolved clones. c Same as (b) for the ancestral strain 1F1c-01 and three evolved clones.

Genomic signatures compatible with Dici transposition were detected in both strain backgrounds (Supplementary Data 6). For the evolved lines derived from CBS 565^T, seven Dici1 insertions were detected at sites unoccupied in the ancestral genome, along with two deletions consistent with element excision. Similarly, 20 Dici2 insertions and one excision were detected in the evolved lines derived from strain 1F1c-01. Although our approach—pooling DNA from independent lines prior to sequencing—does not permit estimation of transposition frequencies, these results provide direct evidence that Dici1 and Dici2 actively transpose in D. catenulata genomes under laboratory conditions.

To assess the susceptibility of D. catenulata genomes to chromosomal rearrangements, we examined karyotype variation among clones isolated from the evolution lines (Fig. 5b, c). Contrasting results were obtained from the two founder strains. All four clones derived from CBS 565^T displayed distinct chromosomal patterns from their founder and from each other, whereas none of the clones derived from 1F1c-01 showed karyotype differences relative to their founder.

We then investigated whether chromosomal rearrangements occurring at Dici insertion sites could be detected in the sequencing reads (Supplementary Fig. 6). A recombination between chromosomes B and E at a Dici1 insertion site was identified in the CBS 565^T evolved lines. The absence of Dici1 copies at the junction suggests simultaneous excision of both elements and repair by a non-homologous end joining (NHEJ) mechanism. This structure was supported by a single read, suggesting that either the event occurred recently, or was counter-selected in the evolved population. In the 1F1c-01 evolved lines, 14 read sequences were compatible with a reciprocal translocation between chromosomes D and F at a Dici2 insertion site, consistent with homologous recombination between copies distributed across the genome. Together, these observations suggest that Dici transposons may be in part responsible for the genome plasticity previously observed in D. catenulata (Supplementary Fig. 2).

Dici1 and Dici2 are therefore as-yet undescribed active transposable elements in D. catenulata genomes, whose unusual copy numbers and impact on genome structure prompted a detailed analysis of their coding sequence organization and evolutionary origins.

Dici elements belong to the ITm superfamily

The presence of a TA dinucleotide duplication flanking Dici elements (Fig. 3c) is commonly found at the insertion sites of DNA transposons of the ITm superfamily⁴⁴. However, the presence of three CDSs within Dici elements (Fig. 3a) is unusual for this family, which typically harbors a single transposase gene.

The smallest CDS of Dici (CDS1) encodes a protein comprising 143 or 146 (aa) in Dici1 and Dici2, respectively. No homologies were identified in protein databases (NCBI database: ClusterNR), and no function could be deduced from the detection of protein domains or sites using InterPro (Supplementary Data 6). For Dici1, the protein appears to contain a signal peptide that facilitates the transport of proteins across cellular membranes.

In the center of the Dici elements, CDS2 encodes a protein of 477 (Dici1) and 473 aa (Dici2). The search in protein databases revealed similarities with uncharacterized fungal proteins containing an IS630-transposase domain (domain NF033545 in the Conserved Domain Database). A DDE-catalytic domain (InterPro entry IPR038717, Tc1-like transposase, DDE domain) was detected (Supplementary Data 6). This domain has been identified in the transposases of DD(E/D)-transposons, a group that belongs to Class II DNA transposons². This group of TEs, which encompasses numerous superfamilies, is characterized by the occurrence of TIRs and TSDs, exhibiting variability in both their length and sequence¹. Yuan and Wessler⁴⁵ proposed a revised classification system for DD(E/D)-transposons by defining a superfamily-specific “signature" consisting of conserved residues within the transposase catalytic DD(E/D) domain. The protein sequence of CDS2 was found to match to the signature of the transposases of the ITm superfamily transposons. This superfamily is ubiquitous in eukaryotes and fungi, and it has a simple structure consisting of two TIRs of variable length and one CDS encoding a transposase with a DD(E/D)-catalytic domain. Transposons belonging to this superfamily have a marked preference for insertions at adjacent TA dinucleotides⁴⁶, as was observed for Dici1 and Dici2. Furthermore, a search in InterPro using the CDS2 protein sequence revealed similarity to homeodomain (HD)-containing proteins (InterPro entry IPR009057), a family characterized by a helix-turn-helix (HTH) DNA-binding domain found in a wide range of proteins, including transcription factors and transposases. The presence of both a DDE-catalytic domain and an HTH DNA-binding domain strongly supports CDS2 as the active transposase of Dici elements.

Consistent with this, structural analysis of the CDS2 protein sequence revealed additional characteristic features of ITm transposases⁴⁶. In particular, two HTH motifs forming a PAIRED domain, each composed of three α-helices, were identified in the N-terminal region of the protein (Fig. 3b and Supplementary Fig. 7). This domain was reported to mediate the specific binding of the transposase to TIRs. A GRPR-like motif was observed between the PAI and RED subdomains (GRTA in Dici1 and GPKT in Dici2). Four putative nuclear localization signals (NLS), composed of basic amino acids (K or R), were predicted by PSORT. Finally, a DD34E catalytic domain, with 34 aa between the second aspartic acid (D) and glutamic acid (E), was localized in the C-terminal part of the protein. It is responsible for the catalytic activity required for the “cut-and-paste" mechanism of transpositions of the DNA transposons.

Finally, a search of protein databases using the CDS3 protein sequence (347-398 aa) revealed similarity to uncharacterized fungal proteins annotated as containing an IS630-transposase domain. A search in InterPro further revealed the presence of a homeodomain (Supplementary Data 6). These observations initially suggested that CDS3 might encode a transposase. However, a multiple-sequence alignment with CDS2 and related transposases revealed that the DDE-catalytic triad was not conserved, indicating that CDS3 is likely a relic of a transposase gene (Supplementary Fig. 7).

In summary, the presence of three distinct CDSs contrasts with the typically compact architecture of ITm elements, which encode a single transposase. The degenerate state of CDS3 raises questions about its current functional role and long-term evolutionary fate, prompting an analysis of selective pressures acting on each CDS.

Evolutionary dynamics of Dici1 and Dici2

To address these questions, we analyzed non-synonymous (dN) and synonymous (dS) substitution rates. This analysis suggests that Dici1 and Dici2 may follow distinct evolutionary trajectories, potentially reflecting differences in their genomic distribution and copy-number dynamics (Table 1).

Table 1.

Evolutionary dynamics of Dici1 and Dici2

Dici	CDS	unique/totala (n0/n1/n2/n3/n4)b	Positive selection (p value)c	Sites under purifying selection (p < 0.1)	Evolutionary strategy
Dici1	CDS1	4/102 (2/0/1/0/1)	0.2777ns	0	Conserved backbone
	Tpase	5/86 (3/0/1/0/1)	0.3055ns	7	Functional constraint
	CDS3	4/102 (2/0/1/0/1)	0.2423ns	4	DNA-binding relic
Dici2	CDS1	13/115 (0/3/0/4/6)	0.0060**	0	Adaptive innovation
	Tpase	33/119 (1/5/1/10/16)	0.0007***	5	Red Queen hypothesis
	CDS3	25/122 (1/5/1/9/9)	0.5000ns	4	DNA-binding relic

^anumber of unique CDS sequences used in the analysis/total number of CDS sequences in the five strain genomes.

^bnumber of unique CDS sequences obtained from strains CBS 565^T, 43B2s-11, CLIB 1179, 25B2s-01 and 1F1c-01, respectively.

^cBUSTED’s p values (likelihood ratio test with boundary-corrected null distribution). Threshold: ^ns, p > 0.05;^**, p ≤ 0.01; ^***, p ≤ 0.001. No multiple testing correction.

Tpase: transposase CDS.

Dici1, which is largely restricted to CBS 565^T, displays low sequence diversity. No evidence of episodic positive selection was detected for any coding sequence using the branch-site model, although the limited number of unique sequences likely reduces statistical power. In contrast, site models identified several codons under pervasive purifying selection in the transposase (7 sites; p < 0.1), indicating strong functional constraints and suggesting the maintenance of transposition efficiency across copies during the recent or ongoing clonal expansion of Dici1 in CBS 565^T.

In contrast, Dici2, which is distributed in the last four reference genomes (43B2s-11, CLIB 1179, 25B2s-01 and 1F1c-01), shows signatures consistent with more heterogeneous selective pressures. Episodic positive selection was detected in CDS1 (p = 0.0060) and in transposase (p = 0.0007), suggesting that specific sites may experience adaptive changes in certain lineages, possibly in response to different host genomic contexts. At the same time, the transposase also contains multiple sites under a pervasive purifying selection (5 sites; with p < 0.1), indicating that the core functional constraints are maintained despite episodic diversification. These sites were further examined for their correspondence with catalytic residues or conserved motifs characteristic of transposase proteins; however, no clear association with the known catalytic triad or recognizable functional motifs was detected. These observations suggest that Dici2 may be engaged in an evolutionary arms race with its host, consistent with the Red Queen hypothesis model of host-pathogen interactions⁴⁷, whereby TE “offensive” components must continuously adapt to host defenses while maintaining essential functions⁴⁸.

Although the canonical catalytic triad of transposases is not conserved, CDS3 in both Dici1 and Dici2 retains a homeodomain, suggesting that it is not merely a relic of an ancestral transposase but rather a functionally repurposed derivative. Consistent with this, CDS3 shows signatures of purifying selection at four sites (4 sites; p < 0.1) and no evidence of positive selection. Although catalytic activity is unlikely to be retained, the persistence of constrained sites may indicate a retained DNA-binding function or a structural role.

Together, these observations suggest that different components of Dici elements are subject to distinct and partially decoupled selective regimes.

Dici, an as-yet undescribed family of ITm transposons

A systematic search was performed against all available prokaryotic and eukaryotic genomes using the protein sequences of Dici1 and Dici2 transposases as queries (see Methods). A total of 22 putative homologues were identified in the kingdom of Fungi (Supplementary Data 8). Of these, three were detected in species belonging to Ascomycota, which is the largest phylum of Fungi, while 19 were detected in the fungal phylum Mucoromycota. Among the 22 fungal sequences retrieved from databases, the Mucor piriformis transposase (FUN-MUPI-1) was nearly identical to the Tc1-MP1 sequence (identity: 432/434 aa) reported by Chang et al.⁴⁹. However, the other related Tc1 transposons described in that study were not recovered in our systematic search.

The mean size of these 22 fungal elements was found to be 1749 base pairs, with the exception of the transposon detected in Mucor piriformis (FUN-MUPI-2), which has an unusual size of 6758 bp (Supplementary Fig. 8 and Supplementary Data 8). This latter transposon exhibited a complex structure, characterized by two inverted-repeat regions of 1627 bp at both extremities, which harbor two identical transposases (438 aa). The TIR sequences of the fungal elements displayed a considerable degree of size variability, ranging from 50 to 187 base pairs, with the most prevalent TIR sequences found to start with CACTA (n = 6). Invariably, a characteristic dinucleotide TA duplication was present at each extremity, a common characteristic of the ITm transposon superfamily.

Their corresponding transposases are encoded by a single CDS ranging from 432 to 478 aa, with the exception of FUN-MUPI-2, which has two transposases. The characteristic domains previously observed in Dici1 and Dici2 transposases were conserved, namely the DNA-binding (PAIRED) and DD34E catalytic domains (Supplementary Fig. 7). The four putative NLS patterns were found to be conserved. Notably, the rightmost pattern (KKRK), located in the core of the DD34E catalytic domain, was detected in all transposases. The GRPR-like motif, located between the PAI and RED subdomains, was identified in multiple forms, with GR(P/K)(P/A) observed in 12 transposase sequences. Finally, the average pairwise sequence identity among the fungal transposases, including Dici1 and Dici2, was 40% and increased to 52% when the comparison was limited to the catalytic domain. (Fig. 6c, d).

Fig. 6. Comparison of Dici family transposases (n = 24) with transposases from the ITm superfamily.

a Phylogenetic tree inferred from full-length transposase sequences using the maximum-likelihood method; bootstrap support values are shown in red. The IS630 transposase was used as an outgroup. An expanded version of the tree, showing complete transposase names and clade assignments, is provided in Supplementary Fig. 9. b Enlarged view of the Dici, Gambol, and Hiker clades. The species in which each transposon was identified is indicated. A red asterisk indicates the position of Dici1 and Dici2. c, d Heatmaps showing pairwise protein sequence identities between transposases from different families. Values represent the mean of all pairwise identities calculated between members of two families. Comparisons were performed using either (c) full-length transposase sequences or (d) sequences restricted to the catalytic domain. Red squares highlight comparisons within the Dici family transposases. Green squares highlight comparisons between the transposases of the Dici family and those of the Gambol and Hiker families. Detailed information on transposase names and family assignments is provided in Supplementary Data 8.

These findings suggest that Dici1 and Dici2 may be part of a larger family that is also present in other fungal species. Consequently, we decided to name this newly identified fungal transposon family as Dici. A phylogenetic approach was used to classify the Dici transposases alongside representatives of the ITm superfamily of transposons (Fig. 6a). This analysis revealed that Dici elements form a well-supported (100% bootstrap support) distinct clade, separate from the Gambol/DD34E⁵⁰ and Hiker/DD35E⁵¹ transposase families, which themselves form distinct lineages within the ITm superfamily⁵¹ (Fig. 6b). Pairwise sequence comparisons between transposases from different families confirmed the close relationship between Dici and the Gambol and Hiker transposases (Fig. 6c, d). The highest conservation, when compared to the full-length Dici transposases, was observed with Gambol/DD34E (24%) and Hiker/DD35E (25%), rising to 31% identity when the analysis was restricted to their catalytic domains.

In summary, our results establish Dici as a previously undescribed family of DNA transposons, with members identified across multiple fungal species.

Discussion

The present study has revealed two previously unidentified transposable elements, Dici1 and Dici2, in D. catenulata genomes, adding to the scarce repertoire of DNA transposons identified in yeasts, particularly ITm transposons. Despite the report of Tc1/mariner elements in yeasts dating back to 2011⁸, such elements remain rare and poorly characterized in this group compared to filamentous fungi. A recent study reported seven Tc1 transposons in filamentous fungi, some of which showed sequence similarity to Dici1 and Dici2⁴⁹, although no comprehensive phylogenetic analysis was performed to place these elements within the highly diverse ITm superfamily of DNA transposons⁴⁴. Together with newly discovered homologous elements in other fungal species, Dici1 and Dici2 form an as-yet undescribed family, named Dici, so far only identified within the fungal kingdom. Phylogenetic analysis of transposase sequences places the Dici family as a monophyletic clade closely related to the Gambol/DD34E and Hiker/DD35E elements, which were previously considered distinct from the main Tc1/DD34E family⁵⁰. Genome annotation revealed that Dici elements reach relatively high copy numbers in some strains, up to 96 copies in CBS 565^T, which is unusual for a DNA transposon in yeasts. The experimental evolution approach employed in this study provides direct evidence that Dici1 and Dici2 actively transpose in D. catenulata genomes under laboratory conditions. This is particularly notable, as transposable element activity is typically counterbalanced by host genome defenses, including silencing mechanisms that limit transposition in fungi¹⁴. The activity of Dici elements therefore suggests that they may have evolved strategies to evade host defense systems, although the underlying molecular mechanisms remain to be elucidated. To our knowledge, this constitutes the first direct demonstration of endogenous DNA transposon activity in a yeast species, as research in yeasts has largely focused for decades on Saccharomyces cerevisiae LTR retrotransposons.

Transposons are commonly associated with genome rearrangements and distinctive chromosomal features¹³. This is well illustrated in Saccharomyces yeasts, where rearrangements frequently arise from ectopic recombination between Ty elements or other repeated sequences⁵². Comparative analysis of D. catenulata strain karyotypes revealed extensive large-scale chromosomal rearrangements, resulting in non-collinear chromosomes. Experimental evolution further provided compelling evidence of a high degree of genome plasticity in this species. Strikingly, all four clones derived from the ancestral strain CBS 565^T displayed distinct chromosomal patterns, demonstrating that chromosomal rearrangements can arise rapidly in D. catenulata. Many chromosomal breakpoints coincide with Dici insertion sites, suggesting that ongoing transposition activity may contribute substantially to the observed genomic instability. Consistent with this, strains from clade #3, which harbor few Dici elements, exhibit conserved karyotypes.

The current distribution and high sequence conservation of Dici elements across D. catenulata clades may be explained by two main scenarios. In the first scenario, both Dici1 and Dici2 transposons were acquired by a common ancestor prior to clade diversification, as suggested by their presence in all clades. A subsequent burst of Dici transposition may have contributed to early genomic rearrangements and potentially triggered lineage divergence. The absence of genetic recombination between rearranged genomes likely promoted the accumulation of mutations within each clade. The observed variation in Dici copy number among strains within the same clade further supports the occurrence of subsequent strain-specific transposition dynamics. In the second scenario, the high nucleotide conservation of Dici elements both within and between clades is consistent with a more recent acquisition event, which was subsequently followed by a rapid burst of transposition. The detection of both Dici1 and Dici2 elements across all lineages may be attributed to multiple independent introductions or a single acquisition event that was subsequently disseminated through horizontal transfer among lineages. This hypothesis is further supported by the observation that some strains, isolated from cheeses of the same PDO, were found in different clades, a pattern that may have facilitated such horizontal exchanges. In this scenario, the Dici elements likely invaded already differentiated genomic backgrounds and later expanded, contributing to the structural variations observed today.

The pronounced genetic structure observed for D. catenulata reveals distinct clades representing deeply diverged lineages with limited gene flow. This pronounced differentiation indicates that historical barriers to genetic exchange have persisted over extended evolutionary timescales. While the observed intra-clade karyotype polymorphism suggests that chromosomal structure is not fixed within lineages, we hypothesized that accumulation of large-scale rearrangements between clades likely acts as a post-zygotic barrier, further restricting gene flow in a highly plastic genomic landscape. Such chromosomal rearrangements have been proposed to promote hybrid infertility⁵³ and to lock in adaptive alleles⁵⁴, thereby accelerating lineage divergence.

Taken together, our findings establish D. catenulata as a powerful model to dissect the mechanisms of DNA transposon proliferation and to elucidate how mobile genetic elements drive genome plasticity. These processes may be pivotal in fungal evolution by facilitating rapid adaptation and promoting speciation. Beyond providing fundamental insights, these results may provide key information for investigating the emergence of opportunistic fungi in both clinical and food safety contexts.

Methods

Strains and culture media

The 61 D. catenulata strains used in this study were obtained from international collections (CLIB, CBS and PYCC strains), or isolated from cheeses as part of the MetaPDOcheese project²³. Others were kindly provided from their own collections by Françoise Irlinger from INRAE SayFood (Palaiseau, France), Marie Desnos-Ollivier from Pasteur Institute (Paris, France), and Nicole Desbois-Nogard from the University Hospital of Martinique (Fort-de-France, France) (Supplementary Data 1). Yeast strains were maintained on YPD medium (10 g/L yeast extract, 20 g/L bacto peptone, 20 g/L glucose, 20 g/L agar).

Isolation and identification of yeast strains from cheese

In a previous study, 2316 samples of rinds and cores from 1158 cheeses of 44 French Protected Designation of Origin (PDO) were collected, and the diversity of their microbial community was assessed using a metabarcoding approach²³. PDO is a label that refers to a category of products, such as cheeses that are produced and processed in a specific geographical area. Based on these results, we selected cheeses in which D. catenulata was present with an abundance greater than 10%, which represents 5.5% of the 2316 cheese samples. An aliquot of each cheese was plated on YPD and Wallerstein (Thermo Fisher Scientific) media containing 100 μg/mL chloramphenicol and incubated 48 h at 28 °C. Individual colonies were identified taxonomically by sequencing the D1/D2 and ITS regions using primers ITS1 (5′-TCCGTAGGTGAACCTGCGC-3′) and NL4 (5′-GGTCCGTGTTTCAAGACGG-3′) as previously described⁵⁵.

A total of 35 D. catenulata strains were selected from cheeses originating from 12 PDOs and 28 different producers. The PDOs were anonymized and designated as PDO1, PDO3, PDO9, PDO18, PDO22, PDO24, PDO25, PDO27, PDO28, PDO29, PDO30, and PDO43. Within each PDO, producers were anonymized using letters (e.g., 43B denotes producer B from PDO43). In some cases, cheese samples were obtained from the same producer across different production seasons; these samples were distinguished by appending a numerical suffix to the producer code (e.g., 18B1 and 18B2; 24C1 and 24C2; 24D1 and 24D2; 25B1 and 25B2; 25C1 and 25C2; 43A1 and 43A2; 43B1 and 43B2). Finally, only one D. catenulata isolate per cheese sample was analyzed.

This dataset was further supplemented with isolates obtained from international yeast collections and collaborating laboratories. These strains were selected to represent eight geographic origins and five ecological sources, primarily animal/human-associated (n = 9) or dairy-associated (n = 14). In total, the final collection comprises 61 D. catenulata strains (Supplementary Data 1).

Flow cytometry

Yeast cells were prepared for flow cytometry analysis as previously described⁵⁵. DNA labeled with SYTOX R green (Invitrogen) was quantified on a C6 Accuri (Ann Arbor, MI, United States) flow cytometer as previously described⁴¹, with an excitation wavelength of 488 nm and an emission wavelength of 530 nm. Acquisition was performed on 30,000 events observed with gating on forward scatter/side scatter signals. The flow rate was set to about 2000 events per second (medium flow, 35 mL/min; core, 16 mm).

Pulsed-field gel electrophoresis

Yeast karyotyping was achieved by contour-clamped homogeneous electric field (CHEF) gel electrophoresis. Plugs of yeast chromosomes were prepared as described elsewhere⁵⁶. The CHEF-DR III apparatus (Bio-Rad, Hercules, CA, United States) was set to 5.2 V/cm with pulses of 90–120 s for 12 h and then to 4 V/cm with pulses of 120-360 s for 24 h, and samples were run on 1% SeaKem^® Gold Agarose gels (Lonza, Rockland, ME, USA) in 0.5 × TAE buffer at 12 °C. The chromosomes of Saccharomyces cerevisiae CLIB 112 (=YNN295) were used as size markers. The agarose gels were stained with ethidium bromide (0.5 mg/mL) and washed with water before visualization under UV.

Genomic DNA extraction

Yeasts were cultured in 10 mL YPD medium at 28 °C with shaking at 220 rpm for at least 36 h⁴¹. Genomic DNA (gDNA) used for sequencing with Illumina technology was extracted based on an in-house protocol involving mechanical (vortexing for 4 min with 0.3 g glass beads) and chemical lysis (Tris-HCl 10 mM pH8, EDTA 1 mM, NaCl 100 mM, Triton 2% and SDS 1%), as previously described⁵⁵. To eliminate small fragments from the gDNA, a purification was performed using AMPure XP beads (Beckman Coulter Genomics, Danvers, MA, USA).

To extract yeast gDNA for long-read sequencing, strains were cultured overnight in 50 mL YPD medium at 28 °C with shaking at 220 rpm. The extraction method was adapted from a protocol developed at Genoscope⁵⁷. Briefly, yeast cells are lysed with zymolyase to generate spheroplasts, which are then chemically lysed in the presence of SDS. Proteins are then removed by precipitation with potassium acetate. Finally, the DNA is precipitated with isopropanol, washed with ethanol and resuspended in TE (10 mM Tris-HCl, 1 mM EDTA)⁴¹.

RNA extraction

To extract total RNA, strains were cultured overnight in 10 mL YPD medium at 28 °C with shaking (220 rpm). Cell concentration was then estimated by spectrophotometry and 10⁹ cells were harvested. The TRIzol method⁵⁸ was used to isolate total RNA as previously described⁵⁹. Briefly, cells were quickly washed with 750 mL cooled (4 °C) DEPC-treated water, pelleted, frozen in a −80 °C methanol bath and mechanically lysed through vortexing with glass beads in 400 μL TRIzol^™ (Life Technologies) at 4 °C for 15  min. The liquid phase was collected, and TRIzol added to a 4 mL final volume, with 800 μL chloroform. The mixture was then vortexed and centrifuged (9000 g for 15 min). The supernatant was centrifuged again (2000 g for 2 min). RNAs were precipitated with 2 mL cooled isopropanol and incubated for 10 min at −20 °C before centrifugation (9000 g for 10 min). The resulting pellet was washed twice with 75% ethanol and then dissolved in 150 μL of nuclease-free water (Qiagen). Total RNA from 100 μg aliquots was purified with a RNeasy mini kit (Qiagen). RNAs were eluted with 2 × 30 μL of the provided RNase-free water.

Illumina library preparation and sequencing

Libraries were prepared with NEBNext DNA Modules Products (New England Biolabs, Ipswich, MA, USA) and an “on beads" protocol developed by Genoscope. gDNA (250 ng) was used as starting material. After adapter ligation, the ligated product was amplified by 12 cycles of PCR with the Kapa Hifi Hotstart NGS library Amplification kit (Roche, Basel, Switzerland), followed by purification with AMPure XP beads (Beckman). Libraries were sequenced with an Illumina NovaSeq 6000 instrument (Illumina, San Diego, CA, USA), in paired-end mode, generating 150-base reads. Illumina short-read sequencing statistics are reported in Supplementary Data 9. Average sequencing depth across the 61 D. catenulata strains is 140X.

Following Illumina sequencing, we applied an in-house quality control process to the reads that passed the Illumina quality filters. The first step discarded low-quality nucleotides (Q < 20) from both ends of the reads. The Illumina sequencing adapter and primer sequences were then removed from the reads, and any reads shorter than 30 nucleotides after trimming were discarded. These trimming and removal steps were achieved using in-house software based on the FastX package (https://github.com/institut-de-genomique/fastxtend). In the final step, read pairs that mapped to the phage ϕX genome were discarded using the SOAP aligner⁶⁰, along with the reference genome of the ϕX174 enterobacterial phage (GenBank: NC_001422.1). Short Illumina reads were subjected to post-processing to remove low-quality data, as previously described^61,62.

Illumina RNA-seq library preparation and sequencing

Libraries were prepared from 100 ng total RNA extracted from five strains of D. catenulata (CBS 565^T, 43B2s-11, CLIB 1179, 25B2s-01, 1F1c-01) using the NEBNext Ultra II Directional RNA Library Prep for Illumina (New England BioLabs) according to the manufacturer’s protocol. The libraries were sequenced using an Illumina NovaSeq 6000 instrument (Illumina, San Diego, CA, USA) in paired-end mode, generating 150-bp reads.

Long reads library preparation and sequencing

Nine μ g gDNA was purified from five strains of D. catenulata (CBS 565^T, 43B2s-11, CLIB 1179, 25B2s-01, 1F1c-01) using the Short Read Eliminator XL Kit (Pacific Biosciences, Menlo Park, CA, USA). A library was prepared with 2 μ g purified gDNA as the starting material, following the 1D Native barcoding genomic DNA (with EXP-NBD104 and SQK-LSK109) protocol provided by Oxford Nanopore Technologies (ONT, Oxford Nanopore Technologies Ltd, Oxford, UK). The sample (pooled with five other barcoded samples) was sequenced using an R9.4.1 MinION flow cell. The reads were basecalled using Guppy version 4.2.5. ONT long-read sequencing statistics are reported on Supplementary Data 9.

Long read-based genome assembly

The raw Nanopore reads for the five strains of D. catenulata were assembled using NECAT with default parameters⁶³ and a genome size of 20 Mb. The resulting output was polished once with Racon⁶⁴ using Nanopore reads, then once with Medaka (model r941_min_high_g360) using Nanopore reads, and twice with Hapo-G 1.3.4⁶⁵ using Illumina short reads.

Genome annotation and manual curation

Tandem repeats in the genome assemblies were masked using Tandem Repeat Finder⁶⁶, while simple repeats and known repeats included in RepBase were masked using RepeatMasker (https://repeatmasker.org/) with the parameter -species Saccharomycetes⁶⁷.

Gene prediction was performed using several proteomes downloaded from NCBI, including C. railenensis, Debaryomyces hansenii, Diutina rugosa, Huiozyma naganishii (formerly Kazachtania naganishii) and Kazachstania africana, which were aligned against the genome in a two-step strategy. To rapidly locate the regions in the genome corresponding to these proteins, we used BLAT version 36 with default parameters⁶⁸. We retained the best match and matches with a score ≥ 90% of the best match score. The alignments were then refined using Genewise version 2.2.0 with default parameters⁶⁹, as this approach is more accurate for detecting intron-exon boundaries. We retained alignments if  > 75% of the length of the protein could be aligned with the genome.

RNA-seq short reads were used to detect expressed and/or specific genes. The reads were mapped onto the genome assembly using HISAT2 version 2.2.1 with default parameters⁷⁰. The resulting BAM file was sorted, and Stringtie version 2.2.1⁷¹ was applied with the following parameters: -p 16 -v -m 150. Only the most highly expressed transcript at each genomic locus was retained.

The protein and RNA-seq alignments were combined using Gmove, an easy-to-use predictor that does not require precalibration⁷² and is available on GitHub (https://github.com/institut-de-genomique/Gmove). Genes with >50% untranslated regions and a coding sequence (CDS) length of less than 300 bp were excluded.

The RNA-seq data were used for expert manual curation of gene models. This allowed for the identification of missing genes and the correction of intron splice sites. To this end, RNA-seq reads were mapped against the genome assemblies using HISAT2 with the following parameters: –rna-strandness FR –max-intronlen 3000. The results of RNA-seq mapping (BAM files) were displayed using the Artemis genome browser v18.2.0⁷³. Manual changes were made to the structural annotations within the browser to correct intron splice sites. The completeness of the annotation was assessed using BUSCO v5.3.1⁷⁴, with the the following parameters: -m protein and saccharomycetes_odb10 lineage dataset as the reference (2137 proteins). Functional annotation was performed using the in-house software go-FAnnoT (https://github.com/hdevillers/go-fannot), which assigns annotations based on homologies found in the curated UniProt database (release 2025_01).

In addition to predicting protein-coding genes, tRNA genes were identified using tRNAscan-SE v2.0.5⁷⁵. A BLASTN search was also conducted to detect the presence of a complete rDNA unit, including 18S, 5.8S, 26S and 5S rRNA genes. The reference sequences of Candida anglica⁴¹ were used for this analysis.

SNP calling

The Illumina reads for each D. catenulata strain were mapped using bwa-mem2 version 2.0pre1⁷⁶ to the genome obtained for the type strain CBS 565^T. The large segmental duplication (388 kb) observed in chromosome F (DICA0_F:3198109-3586457) was masked to prevent null mapping quality from being obtained in duplicated regions, which would impact subsequent variant detection.

We used Genome Analysis Toolkit (GATK) v4.5.0.0 for variant calling (HaplotypeCaller parameters: –sample-ploidy 2 –emit-ref-confidence GVCF), with hard filtering according to GATK best practice⁷⁷. The genotyping pipeline produced a VCF file of 508,773 variants, which included 436,405 SNPs and 72,368 indels identified in the 61 yeast samples.

Only SNPs were retained for phylogenetic analysis and population genomics. Those with a missing genotype above 10% or a minor allele frequency below 0.03 were removed. Genotypes with quality and read depths below 30 were masked. SNPs were filtered using vcftools version v0.1.16⁷⁸, using the following parameters: –minGQ 30 –minDP 30 –maf 0.03 –max-missing 0.9. The resulting filtered dataset contained 357,967 SNPs. Initial analysis revealed the presence of heterozygous SNPs even in haploid genomes (on average, 1150 positions per genome). These SNPs, which were mostly clustered, were removed if they fell within a region containing more than seven heterozygous SNPs within a 1000 bp window. The final dataset contained 351,942 SNPs.

Phylogenetic analysis

For the purpose of phylogenetic analysis, the SNP dataset (351,942 SNPs) was used to generate sequences in FASTA format, with IUPAC ambiguity codes (Y, R, W, S, K, M) used for heterozygous positions when necessary. Subsequently, a phylogenetic tree was generated using IQ-TREE v2.3.1 (parameters: -T AUTO -m MFP), with a maximum-likelihood approach⁷⁹. The best-fit nucleotide substitution model was automatically selected by ModelFinder⁸⁰. Branch support was assessed using an ultra-rapid bootstrap approximation (parameter -B 1000).

Population genomics

Population structure inference was performed using fastStructure⁸¹. The VCF SNP file (351,942 SNPs) was converted to PED format using PLINK v1.90b7.2⁸², filtering positions that are in linkage disequilibrium (LD_1/2 > 0.5, PLINK parameters: –indep-pairwise 50 5 0.5). This resulted in a set of 34,101 SNPs. Twenty runs of fastStructure were performed with the prior set to “simple" and summarized using CLUMPP v1.1.2⁸³. Plotting was carried out using R v4.5.0⁸⁴.

The AMOVA method⁸⁵ implemented in the R package poppr v2.9.7⁸⁶ was used for the detection of population differentiation. Permutation tests, which involve the random reallocation of individuals or haplotypes to different groups, were performed using the randtest function from the ade4 v1.7-23 package⁸⁷.

Population genetics statistics (π, d_XY, F_ST and Tajima’s D) were computed on the SNP dataset (351,942 SNPs) using the R package PopGenome v2.7.7⁸⁸ and a window of 10,000 bp. The values were averaged over all windows to obtain a genome-wide values.

Transposon content

Transposon content and copy number were estimated from Illumina short reads. For each strain, transposon content (relative genome abundance) was estimated as the ratio of reads mapped to each element relative to the total number of reads mapped to the genome. Reads were mapped to individual transposon sequences using bwa-mem2, and counts per element were obtained with featureCounts v2.0.6⁸⁹. The total number of reads mapped to the genome was determined using samtools v1.15.1 (samtools view -c -G 13 -F 256) on the BAM files generated by bwa-mem2 (Supplementary Data 9). Copy number per element was calculated as the product of its relative genome abundance and genome size (14 Mb), divided by the element length.

Chromosomal rearrangements analysis

SynChro was used to reconstruct synteny blocks from pairwise comparisons of the five annotated genomes⁹⁰. The MUMmer4 package was used to generate global alignments of the genomes⁹¹. The alignments were further filtered using the delta-filter utility (MUMmer4) with parameters -l 5000 (excluding regions ≤5 kb) and -m. Alignment coordinates were extracted with the show-coords utility to generate a tabular file. Circular representation of chromosomal rearrangements was performed in R using the package circlize v0.4.16⁹².

Dici transposons characterization

Functional analysis of the three protein sequences of Dici elements was performed using InterPro (https://www.ebi.ac.uk/interpro). Predictions of protein secondary structures were performed using PSIPRED version 4 (http://bioinf.cs.ucl.ac.uk/psipred). Nuclear localization signal (NLS) predictions were performed using the PSORT II (http://psort.hgc.jp).

Mining of Dici transposons

To assess the taxonomic distribution of Dici transposons in genomes, Dici1 (DICA0_E02476) and Dici2 (DICA4_A06172) transposase sequences were used as a query to search the National Center for Biotechnology Information (NCBI) non-redundant (nt) and eukaryotic reference genomes (Eukaryote Refseq Genomes) nucleotide databases (downloaded on 23/05/2025) using TBLASTN (version 2.15.0) with an E-value cutoff of 10⁻¹⁰. The reported sequences were kept only if a catalytic DDE domain was identified using InterProScan version 5.71-102.0 (parameters: –applications PFAM, search domain: PF13358).

To determine the boundaries of these elements, the best-hit sequences were recovered with 2-kb flanking regions, aligned using MAFTT v7.453⁹³ with parameters –maxiterate 1000 –genafpair. Transposon boundaries were then checked manually for the presence of terminal inverted repeats (TIR) and target site duplications (TSD). The resulting transposon sequence was searched against its host genome using BLASTN (E-value cutoff 10⁻⁴⁰) to estimate copy number. All hits obtained that were >40% in size and had identity >80% were used to calculate the copy number as proposed by ref. ¹.

Dici transposases phylogeny

Multiple sequence alignments of transposase sequences (full-length or catalytic domain) were generated using MAFFT (parameters: –maxiterate 1000 –globalpair). Alignment columns were trimmed using ClipKIT v1.3.0⁹⁴ to retain only phylogenetically informative sites (parameter: -m kpi-smart-gap). Phylogenetic trees were inferred using the maximum-likelihood method implemented in IQ-TREE v2.3.1⁷⁹. The best-fitting protein substitution model was selected using ModelFinder⁸⁰ as implemented in IQ-TREE (parameters: -T 10 -mset LG,JTT,WAG,Dayhoff,NQ.yeast,Q.yeast). Branch support was estimated using the ultrafast bootstrap approach with 1000 replicates (parameter: -B 1000).

Family-wise transposase comparisons were obtained using multiple alignments (full-length or catalytic domain) generated with MAFFT. The alignment files were imported, and the pairwise identity matrices were obtained using the seqinr R package (v4.2-36)⁹⁵. The identity values for each pair of transposon families were then averaged.

Analysis of Dici CDS molecular evolution

Nucleotide sequences for each CDS were de-duplicated to retain only unique haplotypes, and Dici1 and Dici2 were analyzed independently. Sequences were aligned using pytransaln v0.2.2 (https://github.com/monagrland/pytransaln), which performs codon-aware alignments by first translating sequences into amino acids, aligning them with MAFFT (v7.525), and then back-translating to maintain the reading frame. Phylogenetic trees were subsequently constructed with IQ-TREE (v3.0.1) using automated model selection (ModelFinder, parameter: -m MFP).

To characterize the selective pressures that act on Dici1 and Dici2 CDSs, we performed codon-level analyzes using the HyPhy software suite v2.5.93⁹⁶. BUSTED version 4.7⁹⁷ was used to detect episodic positive selection at the gene-wide level, identifying whether at least one site has experienced diversifying selection on at least one branch of the phylogeny (branch-site model). To identify specific sites under selection, we used FEL version 2.6⁹⁸ which estimates non-synonymous (dN) and synonymous (dS) substitution rates on a site-by-site basis (site model). Sites were considered to be under significant purifying selection (negative selection, dN ≪ dS) at a threshold p value of 0.1. Evidence of selection was correlated with the protein domains identified by InterProScan (see above).

Detection of Dici transposition events

Experimental evolutions were carried out using two strains, CBS 565^T and 1F1c-01 (Fig. 5a). For each strain, three independent lines were propagated in YPD medium at 28 °C and 220 rpm. Growth was maintained by daily inoculation at OD₆₀₀ ≈ 0.1 into fresh medium, resulting in 97 ± 1.0 (mean  ± standard deviation) and 98 ± 0.7 cumulated generations for CBS 565^T and 1F1c-01, respectively. Several clones were isolated from each independent line and subjected to PFGE karyotype profiling as described above.

Genomic DNA was extracted from each line as described above for long-read sequencing. Clones and evolved lines were tested for cross-contamination by PCR using primers specific to a chromosomal break point that discriminate the ancestral strains CBS 565^T (5′-CGGCCCCGCGTACTATATAT-3′; 5′-GTATTCCGATTGCTGGCCAG-3′) and 1F1c-01 (5′-ACCCACTCAATACAAGAGCC-3′; 5′-AACGAGGCTGTTGAGAAGGA-3′). Thermal cycling conditions were as follows: initial denaturation for 2 min at 94 °C, followed by 35 cycles of 30 s at 94 °C, 30 s at 56 ^∘ C, and 45 s at 72 °C, and a final extension of 2 min at 72 °C.

Since Dici mobilization events are expected to be rare, gDNA from the three replicate lines was pooled prior to sequencing to maximize sensitivity for detecting transposition. Genomic DNA libraries were prepared using the Rapid Barcoding kit (Rapid Barcoding 96 V14, ONT). The Short Read Eliminator kit (SRE EXP-SFE001, ONT) was used to enrich fragments larger than 25 kb. The barcoded libraries were pooled equimolarly and loaded onto ONT R10.4.1 flow cells. Sequencing was performed on a MinION Mk1B instrument with real-time Q9 filtering applied during acquisition to generate high-quality long-read datasets. The POD5 files were subsequently basecalled in super-accuracy (SUP) mode using Dorado basecaller v0.7.4 (https://github.com/nanoporetech/dorado).

Read sequences shorter than 5 kb were discarded prior to analysis. Summary statistics are provided in the Supplementary Data 9. Before read mapping, Dici elements were soft-masked in the CBS 565^T and 1F1c-01 reference genomes using RepeatMasker v4.1.7 (-nolow -no_is -lib dici-sequences.fasta) to prevent mapping bias towards pre-existing copies. Reads were then mapped to their respective genomes using minimap2 v2.26-r1175⁹⁹ with ONT-specific parameters (-x map-ont). Alignments were filtered using custom Bash scripts integrating samtools v1.16.1 (view -b -F 4 -F 256) to remove unmapped reads and non-primary alignments, and the awk utility to identify alignments consistent with Dici transposition events. Candidate events were defined as either: (i) an insertion (a stretch of I characters in the CIGAR string, SAM format field 6) spanning 3.5–4.5 kb, consistent with the insertion of a Dici element; or (ii) a deletion (a stretch of D characters) of comparable length, consistent with an excision event. Additionally, a custom Python script was used to detect Dici elements in the soft-clipped regions of mapped reads, enabling the identification of insertion events supported by a single junction. All candidate loci were further manually inspected to confirm read support and positional consistency.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Conceptualization: C.N.; Data curation: F.B. and C.N.; Formal analysis: F.B., C.N., M.W., and J.M.A.; Investigation: M.P.; Methodology: C.C., M.W., and J.M.A.; Project administration: C.N. and J.M.A.; Resources: M.P. and C.N.; Supervision: C.N. and J.M.A.; Visualization: F.B. and C.N.; Writing - original draft: F.B. and C.N.; Writing - review & editing: F.B., C.N., J.M.A., M.W., and C.C.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

The MetaPDOcheese project was funded by the Centre National Interprofessionnel de l’Economie Laitiére (CNIEL, France). This work was supported by the Genoscope, the Commissariat à l’Énergie Atomique et aux Énergies Alternatives (CEA, France) and the France Génomique national infrastructure, funded as part of the “Investissements d’Avenir” program managed by the Agence Nationale pour la Recherche (contract ANR-10-INBS-09).

Data availability

The data generated in this study are provided in the Supplementary Data files and have been deposited in the Zenodo open data repository (10.5281/zenodo.17613032,¹⁰⁰). Sequencing data and annotated genomes were deposited in the European Nucleotide Archive (ENA) at EMBL-EBI under the umbrella project accession number PRJEB94530. The corresponding accession numbers (Project, BioSample, and Run) for each strain are available in the Supplementary Data 1. Source data are provided with this paper.

Code availability

The latest version of the code has been deposited in the Zenodo open data repository (10.5281/zenodo.17613032,¹⁰⁰). The code is also available on an institutional forge, where it can be cloned (https://forge.inrae.fr/frederic.bigey/diutina-catenulata-code).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74608-6.