Clustering of inverted triplications in centromeric and subtelomeric chromosomal regions of Aspergillus flavus

ABSTRACT

The formation of inverted repeats frequently initiates rearrangements associated with gene amplification and disease states. Among these rearrangements are inverted triplications (TRP/INVs). We have used the filamentous fungus Aspergillus flavus as a model system to analyze spontaneous DNA rearrangements using high-depth third-generation sequencing of DNA isolated from vegetatively cultured strains. Analysis of sequence data identified a class of infrequent and transient rearrangements having structures typical of TRP/INVs. These TRP/INVs form both unprocessed and processed species and have a variably sized deletion at the junction between direct and inverted sequences. We found that TRP/INVs are enriched in heterochromatic centromeric and subtelomeric A + T-rich regions, suggesting that this process is a source of genetic instability in these domains. Consistent with this finding, A + T-rich regions contain elevated levels of direct, inverted, and perfect palindromic repeats. Inverted junctions contain palindromic sequences that are typically associated with TRP/INVs. A closer examination of the class of genomic palindromes found at inverted junctions revealed a broad distribution, with the majority lying in AT-rich centromeric sequences. The distribution of TRP/INVs in centromeric and subtelomeric domains mirrors the frequency of palindromes, indicating that palindrome abundance is likely to be a driver of TRP/INVs. A further examination of predicted pairing patterns in palindrome-like structures suggests roles in strand transfer and replicative stalling. These results are predicted from a replication-based model in which palindromes, under conditions of replication stress, facilitate the formation of TRP/INV structures. These studies represent, to our knowledge, the first characterization of spontaneous TRP/INV formation.

IMPORTANCE

Inverted triplications (TRP/INVs), coupled with gene amplification, are sources of genetic instability. However, little is known about how these species arise spontaneously prior to selective pressures. We have established a fungal model system that uses long-read third-generation sequencing to identify low-abundance rearrangements in the absence of an intentional selection. We identified and characterized multiple classes of TRP/INVs flanked by palindromic sequences. These rearrangements were concentrated in the AT-rich centromeric and subtelomeric regions, suggesting that these sites are regulated recombinational hotspots. Palindromes that are also associated with TRP/INVs are also most prevalent in A + T-rich centromeric regions, suggesting that the abundance of palindromes, both as a source of DNA damage and fork stalling, is a driver of TRP/INV formation. These studies provide the first evidence (to our knowledge) of the spontaneous formation of TRP/INVs during DNA replication, which may be relevant to the complex sources of disease states in higher eukaryotes.

INTRODUCTION

Gene duplication is a common theme in genetic change in all phyla ranging from bacterial to fungal to human cells and has been implicated in the genesis of disease states (1, 2). Indeed, the spontaneous formation of large duplication inversions is an underlying mechanism contributing to oncogenic states (2, 3).

Model systems are valuable tools to select and characterize gene duplications. Duplications can occur with long sequence elements through transposition and homologous recombination. Rearrangements, however, also occur in the presence of minimal homology, particularly in the presence of perfect paired and imperfect (quasi-) palindromes (4). Palindromes are particularly unstable due to their extrusion as either cruciform or hairpin structures. For example, breakage-fusion-bridge cycles can involve either the loss of a telomere or, alternatively, the processing of palindromic cruciform DNA to form inversions and subsequent DNA breakage, followed by fusion of DNA strands through processes such as non-homologous end joining (3, 5–7).

Inverted triplications (TRP/INVs) are another class of species mediated by the rearrangement of relatively short regions of homology (8–11). These latter species are observed in bacteria and yeast and can be initiated by palindromes under conditions of replicative stress and fork regression during bidirectional DNA replication (12, 13). Recent data have also suggested a role for TRP/INVs in the rearrangements involved in a multiplicity of disease states (14). In one model for the formation of these structures, palindromes flanking a gene can create a hairpin that primes DNA replication and the synthesis of the inverted copy. After reaching a second palindrome, template switching and DNA replication give rise to the direct repeat. The junctions between direct and inverted repeats are thought to reform the palindromes in the resulting TRP/INV, which may be resolved into asymmetric TRP/INVs containing deletions surrounding the palindromic junctional sequence. On a statistical basis, short palindromes are expected to form most easily in AT- or GC-rich sequences. AT-rich palindromes form frequently in the human genome and are involved in both chromosomal translocation and evolutionary hypervariability (15, 16). TRP/INVs and other rearrangements are typically observed only under selection. However, the array of rearrangements that may occur in the absence of any selection or in regions of simple sequence non-genic DNA has not been investigated.

Centromeric and telomeric regions of chromosomes are essential for the stable inheritance of genetic information. At the microtubule attachment site, centromeric domains can be present as point or regional structures, the latter including chromosome-specific satellite sequences (17). In Neurospora and other filamentous fungi, centromeres appear to be embedded within heterochromatic A + T-rich sequences, modeled in part by a centromere-specific histone H3 variant, CENP-A (18, 19). These centromeres are thought to undergo repeated cycles of repeat-induced point (RIP) mutation, suggesting that RIP contributes to the high A + T levels during evolution (20–22). In another filamentous fungus, Aspergillus flavus, centromeres are defined by genomic comparisons with the structure of the Neurospora crassa centromere (18, 19) and are composed of gene-free regions of approximately 100 kb of 90%–95% A + T-rich sequence on each of the eight chromosomes (23, 24). Generally, centromere sequences on homologs in different strains share extensive (88%–99%) but not complete homology (24). In contrast, the AT-rich sequences of non-homologs share minimal homology. Thus, competing forces appear to maintain a balance between variation and conservation.

Rearrangements near filamentous fungal telomeres are thought to drive genetic reshuffling, adaptive polymorphisms, and elimination of gene sequences (23, 25). The subtelomeric 100 kb, adjacent to the telomeric (TTAGGGTCAACA)_9–11 repeats (26), is also interspersed with A + T-rich elements (ATEs) of 2–20 kb in length that are distributed in different configurations among the chromosomal arms within and between strains. Studies in N. crassa have indicated that these regions are also heterochromatic, as defined by the presence of CENP-A (27). Our previous studies have revealed remnant transposable elements in subtelomeres, suggesting that repetitive cycles of transposition and RIP contribute to the evolution of these sequences (24). Nonetheless, we found highly conserved ATEs that lack any homology to transposable elements both on the same chromosome of different strains and on distinct chromosomes of the same strain. Hence, subtelomeric ATEs are also maintained in a homeostasis of divergence and conservation, indicating a process that acts to promote recombination. The processes acting to promote and to limit ATE destabilization are not understood.

Several issues are critical to advance our understanding of these processes. First, although short palindromes form readily in AT- or GC-rich contexts and can drive chromosomal translocation and hypervariability, how rearrangements occur in the absence of selection or in non-genic regions remains poorly characterized. Second, while our studies revealed a balance of divergence and conservation in both subtelomeric and centromeric ATE homeostasis, the forces that promote recombination are ill-defined. Third, since rare and complex rearrangements are challenging to detect with standard genetic tools, newer, more sensitive methods are required for their analysis (28).

To address these issues, we have employed a combination of three approaches to examine species arising from spontaneous events. First, the presence of binucleated cells during vegetative mycelial growth of A. flavus leads to the potential identification of rare deleterious mutations and rearrangements if they do not exert a dominant phenotype (29). Selection may eliminate such species. Second, we used analysis of high-coverage third-generation sequencing as a physical readout of genetic rearrangement using structural variant (SV) analysis coupled with rapid screening for inversions and individual read analyses. Third, we analyzed the association of palindrome sequences with the structural variants. Here, we report on the identification of infrequent and transient TRP/INVs clustered in the heterochromatic centromeric and subtelomeric regions in the absence of overt selection, a process that appears to be driven in part by the abundance of small palindromes. The data are consistent with a model in which the cleavage of palindromes initiates inverted triplications during DNA replicative stalling, a testable model for future studies.

RESULTS

A+T-rich elements in centromeric and subtelomeric regions have a non-random distribution of direct, inverted, and palindromic repeats

To analyze the accumulation of short repeated sequences in A. flavus AT-rich centromeric regions, we initially used a BLAST analysis to define direct, inverted, and palindromic sequences in centromeric and subtelomeric (and other) ATEs (Table S1; Supplemental text). Specifically, self-alignment of randomized or natural 90% A + T-rich 250 bp sequences by BLAST was used to identify the presence of repeats and statistically significant differences in their predicted abundance. These studies suggested that direct, inverted repeats and palindromes appear to be present at higher frequencies in genomic sequence versus randomized sequence.

To address this issue at a genomic level, we analyzed the entire centromeric sequences of chromosomes 4 and 8 before or after independent shuffling of the centromeres. We then used the repeat finder function of UGENE graphical interface algorithms (30) to identify the frequency of direct and inverted repeats (Fig. 1A; Fig. S2; see Materials and Methods). Minimal repeat sizes of 15 bp with 90% homology and a range of separation of 10–200 bp were used to detect both inverted (shown in pink) and direct repeats (shown in blue). In each case, the direct and inverted repeats are both greater in the unaltered centromere compared to a scrambled sequence of the same base composition (Sequence Suite [31]). The average number of total repeats across all centromeres is 130 per 100 kb of centromeric sequences (standard deviation [SD] = 17), while two independent shufflings of each centromere give rise to an average repeat number of 54 (SD = 9.7). This result is statistically significant (P < 0.05) for the sum of inverted and direct repeats. (Fig. S1). Similar results were obtained using these sequences in Inverted Repeat Finder (32) and direct repeats by EMBOSS tandem repeat finder (33) (Table S2). These data indicate that substructures within the centromere contribute to the frequency of repeats within the centromeric regions. This contribution is in addition to the high levels of repeats due to the high A + T sequence composition.

Fig 1.

Identification of direct and inverted repeats by UGENE Repeat Finder in intact centromeres and scrambled derivatives. (A) Direct and inverted repeats were identified in the chromosome 8 centromere or in two scrambled derivatives of chromosome 8 of identical size and base composition by UGENE Repeat Finder using the following parameters: window size = 15 bp, homology = 90%, minimum space between windows = 10 bp, maximum space between windows = 200 bp, and ran for either direct or inverted repeats. The viewer shows pairs of inverted repeats in pink and direct repeats in blue. The total number of repeats for each shuffled and unshuffled centromere is shown in Fig. S2. (B) The frequency of palindromes is reduced in shuffled centromeres on each chromosome. Palindromes were quantified at each centromere using the Palindrome Analyzer with the following parameters: palindrome repeat size = 12–100, spacer size = 4–19, and mismatches = 0. Light blue denotes unshuffled centromeres; dark blue denotes shuffled centromeres. The shuffled data are presented as the average with SD.

To analyze palindromes specifically, we introduced each of the above centromeres and scrambled derivatives into the Palindrome Analyzer (34) and set the parameters to look at the longest palindrome stems (11–100 bp) with a loop size of 4–19 bp and no mismatches. We found that on each chromosome, the number of palindromes was significantly reduced upon sequence shuffling (Fig. 1B).

Qualitative identification of a TRP/INV class in cultured cells

We were interested in identifying spontaneous rearrangements that may help to explain the non-random characteristic of repeated elements in subtelomeric and centromeric ATE sequences. To this end, we carried out high-coverage (500- to 1,500-fold), third-generation Oxford Nanopore Technologies (ONT) sequencing as a readout for infrequent SVs (35) after forced vegetative culturing. We optimized the Sniffles 2 and Minimap2 programs (36–38) for infrequent SVs and categorized them into apparent deletions, insertions, inversions, duplications, and translocations. We specifically characterized individual sequencing reads from the insertion class that predict sizes of ≥300 bp. BLAST analysis of these larger insertion sequences revealed two major classes: intact tandem duplications and duplications interrupted by a partial or complete inversion. This study focused on the latter class of structures that are similar to the TRP/INVs, previously described in gene amplification studies (8–10).

Our initial qualitative analysis compared this subset of the SV insertion class that appeared to arise in clusters from A. flavus genomic DNA prepared from a starter culture (growth cycle 0 [G0]) (generated from spores isolated from conidiophores) after 5 (G5), 10 (G10), and 15 (G15) weeks of forced vegetative (mycelial) growth. The TRP/INV structures of one of the major SV classes are shown in Fig. 2A. Nine TRP/INVs were present, five of which mapped in the ~1 Mb of ATE DNA, and four were dispersed in the remaining ~36.5 Mb of genomic DNA. When compared to all single-read TRP/INV candidates derived from other cultures (G5, G10, and G15), each TRP/INV was found only at the indicated stage of growth. Hence, within the limits of detection, TRP/INVs did not become fixed in the population and appeared to arise only transiently during growth (Fig. 2B).

Fig 2.

Initial analysis of TRP/INV from large clustered insertions in G5, G10, and G15 cells. (A) DNAs were isolated and sequenced after A. flavus NRRL3357 cells had been cultured for 5 (G5), 10 (G10), and 15 (G15) cycles and subsequently screened for large (≥300 bp) insertions, using the Sniffles 2 algorithm, as described in Materials and Methods, with a variant read number of 3 and characterized further by IGV analysis. Individual reads were then analyzed by BLAST alignment to the TRP/INV shown here. The listed coordinates are the approximate sites of TRP/INV formation relative to the chromosomal orientation. The yellow boxes refer to centromeric TRP/INV, while the orange boxes refer to subtelomeric TRP/INVs and TRP/INVs within ATE elements scattered at other sites in the genome. Uncolored boxes refer to TRP/INVs dispersed at non-ATE sites. The structure of the unprocessed TRP/INV for each species is shown. Arrows below the figure represent the relative orientations of TRP/INV direct and inverted repeats, while arrows above the figure represent the typical palindromic sequences at these sites. The corresponding sizes of the direct (blue) and inverted (red) repeats are shown below. Created in BioRender. Smith, A. (2026) https://BioRender.com/x2aqp0j. (B) The positions of the TRP/INVs are mapped on each chromosome for G5 (green), G10 (red), and G15 (blue). The dark gray region denotes the centromere. Created in BioRender (https://BioRender.com/x2aqp0j.

Quantitative characterization of the TRP/INV class in G15 cultured cells

To analyze the frequency of TRP/INVs, we extended the studies of all G15 structural variants identified as ≥300 bp insertions (relative to the G0 assembly) that also were associated with a linked inversion. We characterized each of these reads and identified those having a TRP/INV configuration. We found that 221 TRP/INV species were distributed over all eight chromosomes of A. flavus (Fig. 3). The low abundance and transitory nature of these species were further examined by quantifying all chromosome 1 TRP/INVs in DNA derived from G0, G5, G10, and G15. We found three cases of heritable sites of TRP/INV. In each case, the TRP/INV structure was unique (Fig. S1). We cannot rule out, however, the possibility that each of these had a common origin that subsequently underwent multiple rounds of rearrangement. These data reinforce the transient nature of TRP/INV.

Fig 3.

Chromosomal distribution of G15 TRP/INV among all chromosomes. The total number of TRP/INVs (y-axis) is plotted for each chromosome and as the summed total (x-axis).

Direct BLAST search of individual sequencing reads relative to the G0 telomere-to-telomere assembly revealed three TRP/INV subtypes (Fig. 4A). Eighty-nine species fall into the symmetric class (class 1) (Fig. 4A, top) and have intact head-to-tail configurations at both inversion junctions within the estimated margin of error of sequencing and sequence alignment (∆0–50 bp). One hundred thirteen species fall into an asymmetric class (class 2; Fig. 4A, middle), displaying an intact head-to-tail configuration (∆0–50 bp) at one of the two junctions and a more extensive deletion at the other end. In comparison, 19 species belong to another asymmetric class (class 3; Fig. 4A, bottom) consisting of TRP/INVs containing a >50 bp loss of sequence at both inversion junctions. The genomic distribution is depicted in Fig. 4B, while Fig. 4C shows examples of each of these classes. We also identified a series of complex forms that may be derived from multiple rounds of TRP/INV formation (data not shown). To test whether individual reads in the absence of BLAST alignment could be detected by algorithms established for the detection of direct and inverted repeats, we used the UGENE graphical interface program (30). Although sensitivity is lost in the absence of alignment, the structures of TRP/INVs were reflective of the direct and inverted repeats as expected, corroborating the individual structures based upon the BLAST analyses (Fig. 4D).

Fig 4.

Three structural classes of G15 TRP/INVs. (A) The three classes of TRP/INVs deduced by BLAST analyses of nanopore sequencing queries to genomic assemblies are presented here. The first class (class 1, symmetric) has little or no deletion at the junction between direct (blue) and inverted (red) repeats within the sequencing error range (≤50 bp). Class 2 (asymmetric) has a deletion at one of the two junctions >50 bp. Class 3 (asymmetric) has deletions of >50 bp at both junctions. The size ranges of both direct and inverted repeats are listed below for each class, and the number of observed species in each class is listed on the right. The orientation of hypothetical sequences A, B, and C (and complementary sequence A′, B′, and C′), as well as the directional arrows, is shown for visual clarity, as described in the legend of Fig. 2. Created in BioRender. Smith, A. (2026) https://BioRender.com/s70s676. (B) Karyotype of TRP/INV sites. TRP/INVs (green pentagons) are indicated on each chromosome. Dark gray region denotes the centromere. Created in BioRender. Lustig, A. (2026) https://BioRender.com/rf8y2mk (C) Examples of BLAST analyses for (left) non-ATE TRP/INVs and for (right) subtelomeric and centromeric TRP/INVs. The junctional structures of the two independent TRP/INVs from dispersed non-ATE regions on chromosomes were determined by BLAST analysis, and the length of direct (blue) and inverted (red) repeats was calculated. The detailed coordinates of direct and inverted (bracketed) repeats in selected TRP/INVs on chromosomes 1 and 5 are shown. The BLAST reads of the left junction and right junctions are shown together with the statistical output of the BLAST results from SequenceServer. +/− refers to a sequence complementary to the query sequence, while +/+ refers to a sequence in the same orientation as the query sequence. The query is the read sequence, and the subject is the G15 assembly. Blue and red boxes refer to the reads that define the duplicated and inverted sequences, respectively. (D) TRP/INV structures derived from an analysis of direct (blue, DR, top) and inverted repeats (other colors, IR) using UGENE Repeat Finder, with the following settings: window, 100 bp; homology, 85%; minimum distance between repeats, 10 bp; and maximum distance between repeats, 100 bp. The arrows indicate the positions of the direct (blue) and inverted (red) repeats as defined by single-read analysis using BLAST.

TRP/INV clusters at centromeric and subtelomeric ATEs

As noted above, >90% AT-rich sequences are present both in centromeres and in ATEs predominantly located within 100 kb regions associated with telomeric repeats. Given the sequence bias and unique chromatin packaging, we analyzed the TRP/INV distribution to determine their potential enrichment in specific centromeric, subtelomeric, and dispersed sequences. We found a 17.5- to 22-fold higher frequency of TRP/INV in ATEs present at centromeres (1/14.3 kb), subtelomeres (1/11 kb), and other classes of ATEs (1/11.5 kb), compared to the overall frequency at other genomic sites (1/251.6 kb) (Fig.5A). The latter class distributes to both coding (101/145) and non-coding (44/145) regions (Table S3).

Fig 5.

Enrichment and characterization of TRP/INVs in centromeric and subtelomeric ATEs. (A) The chromosomal location (x-axis), number (above bars), and frequency (y-axis) of TRP/INVs in ATE (centromeric, subtelomeric, and ATEs located elsewhere in the genome) and non-ATE (dispersed) regions, scattered among all chromosomes, are shown (24). (B) The direct and inverted repeat sizes (together with SDs) of TRP/INVs, located in centromeric and non-centromeric ATEs (subtelomeres + other ATEs) or dispersed non-ATEs, are shown.

Analysis of the size of the direct and inverted TRP/INV sequences (Fig. 5B) revealed that the range of values was reduced (as shown by the SD) in centromeres and subtelomeric (and other) ATEs, relative to dispersed non-ATE sequences. This reflects a tendency towards longer lengths in both the direct and inverted duplications in non-ATE TRP/INVs. For example, the lengths of such duplications extend up to 7.4 kb, compared to the maximum size of 0.92 kb at centromeres. Both the direct and inverted duplications in TRP/INV also trended to slightly smaller sizes in centromeric compared to non-ATE domains (Fig. 5B). These data suggest that selective forces may well restrict centromeric TRP/INV lengths.

To determine whether this result could be qualitatively reproduced in another sequencing methodology, we compared the TRP/INV patterns obtained from the analysis of the nanopore sequencing with the results from two independently PacBio (Revio) (39) sequenced G15 DNA to ascertain the influence of sequence methodologies on the TRP/INV generation (Table S4). Seven TRP/INVs were present, all at unique centromeric and subtelomeric sites. While substantiating the formation of TRP/INVs and their biased distribution towards AT-rich regions, TRP/INV levels were roughly 30-fold higher in ONT than in PacBio sequencing. QC analysis of the sequencing data (Fig. S3) revealed that the lengths present in the PacBio structural variant (Sniffles 2) output of PacBio sequencing were on average 1.5-fold longer than ONT reads. However, ONT structural variant output (tested on chromosome 1) was four times more abundant in centromeric regions and 30-fold more abundant overall, indicating that ONT output reflected the sampling of a far greater number of molecules, although underrepresenting centromeric sequences (40). The quantitative differences in the PacBio unique structural variants may be due to a more rigorous selection for low-error structural variants. Given that the BLAST and UGENE analyses in this study verified the structure of inversion triplications called by ONT, even with junctional deletions, we suggest that PacBio stringency may be overly restrictive for quantification of these events.

Mitochondrial genome accumulates TRP/INVs less frequently than the nuclear genome

As a control for the establishment of transient TRP/INVs and to test the contribution of library or sequencing anomalies that may generate these structures, we also measured SVs by the same methods in the high-copy AT-rich mitochondrial DNA present in the sequencing reactions of each culture. The 30 kb circularly permuted and high-copy mitochondrial genome is present in a sequencing depth of 20,000–30,000 copies. In contrast, centromeric sequences are generally covered in less depth (~250 copies) than other parts of the genome. The mitochondrial genome therefore can serve as a model for the accumulation of TRP/INVs in centromeric sequences. In the G15 nanopore sequencing reads, we found only two TRP/INVs at unique sites. Due to the small sample size, we also explored the frequency of mitochondrial TRP/INVs in G0, G5, and G10 nanopore sequencing experiments. G0, G5, and G10 reads gave rise to 1, 2, and 0 TRP/INVs, respectively. If mitochondrial and nuclear TRP/INVs were found at a comparable frequency, we would have expected 210 TRP/INVs per 30 kb genome. Furthermore, the frequency of TRP/INVs in dispersed sequences is >25-fold higher than in the mitochondrial genome, despite their similar palindrome density (Table 1). These results are inconsistent with a technical reason for TRP/INV formation and additionally suggest that aspects of nuclear DNA replication and recombination contribute to the formation of these species.

TABLE 1.

Observed frequencies of TRP/INVs at centromeres and mitochondrial sequences^a

	TRP/INV (observed)	Total genome sequenced (average)	Sequencing depth	TRP/INV frequency (/kb/read)	Predicted palindrome density (/100kb)
Centromere	56	800 (G15)	278	2.5 × 10−4	831
Mitochondria	5	120 (G0–G15)	26,140	1.6 × 10−6	140
Dispersed	145	36,500 (G15)	1,000	4.0 × 10−5	100

^a Out of the total of G0, G5, G10, and G15 DNA genomes sequenced (120 kb total), we found only TRP/INVs despite the higher depth of mitochondrial sequencing. Taking the predicted palindromes density of the 78% AT-rich mitochondrial genome into account, we estimate at least a 30-fold enrichment of TRP/INVs in the ~800 kb of centromeric sequence and a >25-fold enrichment in dispersed nuclear sequences.

Identification of palindrome sequences at the inverted/direct repeat junction

Previous studies revealed that TRP/INVs of bacterial and yeast genes associated with gene amplification are formed via palindrome formation and template switching, either during bidirectional DNA replication or during DNA repair, resulting in the restoration of the palindromic sequence in the symmetric product (8–10). Using BLAST analysis, we tested the prediction that palindromes are at the site of TRP/INV formation by identifying junctional read sequences that were complementary to both strands of the genomic assembly (examples are shown in Fig. 2C; Fig. S4). The palindromes were then screened further for their predicted thermodynamic stabilities using the UNAfold mfold 4.2 program (41–43) (Tables 2–5). Seventy-five junctions that had palindromic characteristics fell in centromeric (Table 2), subtelomeric (and other ATE) (Table 3), and dispersed (Table 5) locations.

TABLE 2.

Predicted junctional palindromes in centromeric TRP/INVs^a

Chr	Coord	Sequence	Length (bp)	∆G (kcal/mol)	Tm (°C)
1	4510493	TAATTTTAATTATATAATAAAAATTA	26	−4	48.4
1	4615841	AGAATTATAATTATATATTAACTACTAACTACTAAATAATATTATAATTAAAAATA	56	−5	44.6
2	2064783	TTATAATAAATATTATATAATAATATTATTTAA	36	−3.94	51.7
3	2732726	CTATAACTTATAAGTTATAG	20	−7	60.6
3	2739625	TATTTAATTTATAAATAGTTTATAAAATTAAATAT	35b	−6.58	45.8
4	2818566	TTATATATTTTATCTATTATTATAATATTTTTTATTATTTTTT TATAGTAATATTATTTAATATTAAAATCTATAT	73	−6.8	39.2
4	2865046	TAATAAAATATAATATTTTTTAAAATATATAAATATCTACTATATTTATATAAAATTTTTAAAAAAATTATATTTATAGT	75	−17.3	52.1
5	2447757	TAAATAATAATTAATATATTAATATCTTAATAATATTAACTA GTAGTTA	50b	−5.32	45.4
5	2472736	TTAAATATTCTAGAATCTTTCTAGAATATT	18	−5.08	60.8
5	2482236	TTTAATTTATAATTCTATAATTTTATTATTATATATTAAA	33	−9.61	65.8
6	2149500	TTTAATTTATAATTCTATAATTTTATTATTATATATTAAA	40d	−3.7	41.7
7	2618350	ATCTTTATATTAATAATAATAAAATTAATTAATATTTTAATATAATTAAT	19	−2.65	44.1
7	2630477	AATATAATAATTTTATATTAAAATAAATATTAATAAAAAATAAAATTATT	46	−8.83	53.8
7	2636529	ATTTTTAATTTTAAAGATTTTTAATTAATTAGCTAGTTTAAAATATTTTATAAAAATTAATTTATATAATATTTACTATTAA ATTTAAATTAAAAAAAATAAATAAAAAAT	100	−13.09	41.3
7	2646080	TTAATAGTATAATTAATCTTAACTTATTTATACTTTTTTAA	21	−5.32	52.5
8	1814185	TTTATTATTTTATTTTATCAATTTAAAAAAATAATTATAA	33c	−8.53	41.8
8	1851716	AATAAAATATTTTTTATAAAAATAATTAATT	25	−3.18	45.5
8	1870116	TATAAGTAGACTAGCGCCGGGCGCTAGTCTTAATA	24	−12.28	79.7

^a The chromosomal coordinates, sequence coordinates, sequence, length, and predicted ∆G (kcal/mol) and T_m values (°C) (calculated by UNAfold mfold 4.2 using the following default settings: 25°C, 3 mM Mg, and 50 mM NaCl) of centromeric TRP/INV palindromes. Values of ∆G >−3 kcal/mol are indicated in bold.

^b Palindromes associated with extended regions of pairing.

^c Multi-palindromic structures.

^d More complex pairing structures.

TABLE 3.

Predicted palindromes at subtelomeric and dispersed ATE TRP/INV junctions^a

Chr	Coord	Sequence	Length (bp)	∆G (kcal/mol)	Tm (°C)
1	75341	AAATTAGAATTATAAGATTAATTTTTAAGCAGCTTTTT AAATATAGATTCTAATTTC	57	−5.76	38.3
3	2693002	TTATAATTATATAGAATTATAA	22	−3.22	44.3
3	2705424	ATATTTTTTTAATTAAAATAT	21	−2.99	44.2
4	2071	TAATATTAATAATTATTATTTTTATAATATTAAATTCTATA	41b	−3.98	44.3
5	969	ATTAATTTAAATATTATTAATTTTATAATTTAATTAATATT ATATTTAAAATACTTTAAT	60	−7.92	42.7
5	1720	ATAAATATAAATTATATTTAT	21	−4.88	52.1
5	4510716	AATTATAAATACTTTTTTAATATATTATTA	21	−1.21	34
7	690480	CTATAAATTTAAATAGTTAATAAAAATTTTAATAATATATAT TAAATTTTTAAAATATTATTATTAATATTTAAATAAATAGATAA	89	−13.45	44
8	3238765	ATAAAATAGTTTAGTAAATATATATATAATTATTACTAGAAAAT AAATAATTTATTTTCTAGATATATAATTTTATTAGACTATTTTAT	90	−18.55	51.7

^a The chromosomal coordinates, sequences, length, and predicted ∆G and T_m values of palindromes at inverted repeat junctions of subtelomeric (and other ATE) TRP/INVs. Stabilities were determined as in Table 2. Bold highlight indicates >−3 kcal/mol.

^b Palindromes associated with extended regions of pairing.

TABLE 4.

Palindrome distribution identified by BLAST and UNAfold analyses^a

Location	Sequenced palindromes (total)	Sequenced palindromes (≤−3 kcal/mol, Tm ≥40°C)
Centromeres	20	18
Subtelomeres + other ATEs	9	7
Dispersed	46	44
Total	75	69

^a The distribution of palindromes from ATEs (centromeric and subtelomeric + ATEs located elsewhere in the genome) and non-ATE (dispersed) regions are listed together with the distribution of palindromes predicted to have ∆G values of ≤−3 kcal and T_m ≥40°C. The palindromes of lower stabilities may still form secondary structures but do not meet the limitations set in this study.

TABLE 5.

Predicted palindromes at dispersed non-ATE TRP/INV junctions^a

Chr	Coord	Sequence	Length (bp)	∆G (kcal/mol)	Tm (°C)
1	2266091	CCCAGAAATCCGCAGCAGCGGCTGGG	26c	−9.21	65.8
1	2286600	TCCCATGACGTCATGGGA	18	−7.27	73.1
1	5016955	AACGTTATAACGTT	14	−4.96	64.8
2	454491	ATAAACCATCGCTTTGCATTGACTGCATCCAGGTTTAT	39c	−6.82	50.1
2	454669	AAGCTTTGATTATACTACTGAATGAAAGCTT	31	−6.82	57.4
2	493210	CGGCCAAATTCGCCG	15d	−4.72	68.6
2	1804865	CATTTCACCACGGTGAACTGTGGTGAATTGTGCTTG	30	−10.36	62.4
2	3365568	ATTAGTATCTGATACTAAT	21b	−4.84	57
2	3840022	GCTAGCACCATCTGTCAGTTTGGCTGCAGAAAGGGGCTTCATAC AATAACCTACTCTTCATGGAACTCAT AGTCCAACGCCTTCCCCAGCCTGACCCGGA	100	−10.34	42
2	4116246	CCCGACAGTTACATACATTAGTAGTACTCGTACTGTACTCCCT GTATGTAACTTTCGGG	59	−17.44	59.7
2	4273901	GTAATGTCCTCAGCGGTGCGGACGTTAC	27	−7.05	59.7
2	5407963	CGACGGCGACAGCCTCGCCGTCG	23	−13.94	83.9
3	567045	TGTCGACGTCGACA	16	−4.75	61.1
3	573282	CACGCAAGAGCTTGGCCGGTTTGCGTGATCG	31	−8.35	69.9
3	616220	CAAAGTCCAGTGAGACGGCTCCGGGACTTTG	31	−9.18	59.7
3	1526699	ATTCCGTATCAATACGGAAT	20	−8.37	71.4
3	1572702	GATCGACGGAAGATGTAACGTCGATC	26	−9.18	68
3	1734350	TTTTTTTCTTTAAAAAAAAAAAAAAG	27	−6.42	50.8
3	3287741	TTCATTGAGGCAATGAA	17d	−4.78	56.9
3	3921549	TTACCTTATCACGGTAAAT	19	−1.84	40.5
3	4939733	GAATTCTAGATCC	14	−2.24	45.1
4	568414	AATTGCAGTCACTGATGTTCGCAATT	251	−4.05	59.2
4	794920	AACGAGACATGATCCATGTCGCGTT	24	−9.95	70
4	1194263	CAGGCGAAGCGCTGCAGCCTG	21d	−6.12	69
4	1273121	AAACCTCTCAGAGGTTT	17	−5.97	66.8
4	1274984	GCAGGACCATGAGCTCATGCGACATGTTGTGCA	31	−5.22	45.3
4	2528116	ATACTCCGGTATGCAATGCAATGCGGAGTAT	31	−7.88	55.4
4	4238747	ATCCCCCAATGCGGGGAT	18	−5.65	73.4
4	4238730	GTAGACTGGGCGCCCAGTCTAC	22	−9.27	69.6
4	4239051	TCTTGCGCGAGCGTCTGCAGGCGCTGCAGCAGA	34	−7.62	61.1
4	4240170	GTAGACTGGGCGCCCAGTCTAC	24	−9.54	68.3
4	4414478	TCGGTAGAGCCTCTACCGA	20d	−7.39	72.1
5	618917	TGTCAAAATTTCATGAAATTTTTGACA	27d	−6.54	56.3
5	2331951	AGAATGCTAAGAAAGACAATATCTATTTCTTAATTCTC	37d	−4.94	46.5
5	3445422	CATCGCGACGATGATG	16	−4.6	61.5
5	3772422	CAGGGCGCCTGCGACCTG	18d	−7.07	71.7
5	4111968	GACATCATCAAGTGATG	17	−4.91	62.2
6	1167663	TTTATAAATTTTGATATTTATAAA	26	−4.21	45.7
6	1661313	GTCCTCGCAATGTTACATCGAGGAC	25d	−7.71	66.7
6	2486550	ACGGAGCCGCAACGGCTCCGT	21	−12.1	87.5
6	2486789	TTGTCGCAGACACGGCAA	18	−4.13	62.2
6	2523428	ACTAGGAGTACCTAGT	16	−4.68	57.6
7	1656247	GTACCGTTTTCGGCTATGGGTTTCAAATCCCCCAAGTCCGAAAACCGTAC	50	−12.75	53.2
7	1814658	GCGGCGTTTGCACAGCCGCGT	19d	−4.36	67.9
8	256532	GTATGTCAACAGGTAACATAC	21d	−4.3	52
8	1428359	GAAGAACCAGACGATAGTTCGGTTCTTC	34	−7.82	63.6

^a The chromosomal coordinates, sequences, length, and predicted ∆G and T_m values of palindromes at inverted repeat junctions of non-ATE TRP/INVs. Stabilities were determined as in Table 2. Bold highlight indicates >−3 kcal/mol.

^b Multipalindromic structures.

^c More complex pairing structures.

^d Palindromes associated with extended regions of pairing.

Palindromes ranged in size from 13 to 100 bp in non-ATE sequences and 18–100 bp in centromeric and subtelomeric (and other ATE) sequences. Based on the UNAfold DNA secondary structure algorithms, most putative palindromes have stabilities of −3 kcal/mol or less (ranging from −3 to −20 kcal/mol) and T_m values of at least 40°C (ranging from 40°C to 84°C) (Table 2, right column). Based in part on the discussion below, −3 kcal/mol was used as the lowest stability cutoff if T_m was at or close to 40°C ( values in bold indicate the samples that did not achieve this level of stability). As a control, we compared the predicted stability of a previously isolated set of short palindromic junctions deduced to be at the sites of inverted sequence formation in yeast (7) (Table 6). These palindromes have ∆G values ranging from −2.31 to −7.06 kcal/mol and T_m values from 42°C to 67°C, substantiating the validity of the choice of parameters in this study.

TABLE 6.

Predicted stability of palindromes identified at junctions of inverted repeats by Rattray et al. (7)

Sequence	Length (bp)	∆G (kcal/mol)	Tm (°C)
GTACTATTGGTAC	13	−2.31	47.1
CGCCGGCCCAGTGGGCG	17	−5.44	72.3
TTGGTGCGGCCAA	13	−2.89	69.7
GGGCAATCACTTTTGCCC	18	−7.06	67
CTGGGGTTACCGCCAG.	16	−3.86	58.7
CCCAGGTGCCTGGG	14	−6.67	72.9

Not all junctional sequences contained identifiable palindromic sequences within the threshold defined in this study. Out of 75 sequenced junctions with no significant deletion, four fell below the threshold value. This could be an indication either that the threshold is somewhat arbitrary or that a stable palindrome may not always be necessary for TRP/INV formation.

Palindromes are present in genomic sequences at sites adjacent to the TRP/INV junctions

We sought to test more generally whether palindromes are present in the genomic sequences that predict the formation of TRP/INVs. To this end, we examined the 200 bp genomic sequence surrounding the junction in unrearranged sequences at TRP/INVs sites that had junctional deletions of 40 bp or less for the presence of simple palindromes, complex paired structures, and multipalindromic sequences. TRP/INV sites that contain larger deletions can obscure the actual site of palindromic sequences and were excluded from this study. We found that most junctions contain simple palindromes with or without extended associated pairing (147 [88%] in non-centromeric dispersed regions, 52 in centromeric regions [93%], and 24 [100%] in subtelomeric domains). The palindrome lengths (stem plus loop) averaged 33 bp in dispersed regions and 41–43 bp in AT-rich subtelomeric and centromeric regions. All stabilities averaged −6.7 kcal/mol, while T_m values averaged higher in dispersed (56.4°C) than in AT-rich regions (48°C), as expected due to differences in base composition (Fig. 6; Table S5A through C). A fraction of junctions (10%) had stabilities with ∆G of >3.0 kcal/mol. Some junctions were associated with long regions of extensive pairing (13%), and a fraction (12%) had more complex pairing patterns or multipalindromic structures (see Fig. S5 for examples). These findings raise the possibility that some homologous pairing patterns may have a role in perturbing localized structure with or without strand exchange. These predicted structures are unlikely to be static. Rather, they are likely to represent one among many palindromic structures that are present in an equilibrium.

Fig 6.

Palindrome stabilities in centromeric, dispersed ATE, and Non-ATE sequences. The number (y-axis) and stabilities (x-axis) of palindrome stabilities predicted by UNAfold analyses in TRP/INVs with a deletion of 40 bp or less. The distribution of TRP/INV palindromes in centromeric, subtelomeric, and other ATEs (CEN + ATEs) (green) and from dispersed non-centromeric sequences (blue) is shown as a function of stability (kcal/mol), as determined by UNAfold analysis. Also shown are the stabilities found at sites of inversion in yeast (Rattray et al. [7]) (red) for comparison.

Palindrome formation mirrors TRP/INV centromere enrichment

Regions of high AT or GC content produce a greater number of palindromes on a simple statistical basis, and we observed an even greater frequency of palindromes in centromeric, subtelomeric, and other ATEs (Table S1). We wanted to test whether the frequency of a palindrome, having the predicted characteristics of TRP/INV-associated palindromes, could be sufficient to explain the higher density of TRP/INVs at centromeres. We therefore estimated the relative number of palindromes that would be likely to generate a TRP/INV in centromeric and dispersed regions (see Materials and Methods). To this end, we calculated the most frequently observed stem and loop sizes, as well as ∆G and ∆G_cf–∆G_lin values, the latter a measure of the energetic favorability of cruciform formation (44). Due to the complexity of the centromeric palindrome patterns, we first determined that the most frequent classes of the terminal portions of the TRP/INV-associated palindromes in the centromere of chromosome 3 fell into a stem size of 12–100 bp and a loop size of 4–8 bp with 0–2 stem mismatches. The values were input into the program, Palindrome Analyzer (34), which is designed to estimate the frequency and stability of palindromes at the chromosomal level. The sites of the TRP/INVs were confirmed to be present in the output. The distribution of utilized palindromes had ∆G_cf–∆G_lin values that fell between 10 and 16 kcal/mol, which were applied proportionally to the output, since the distribution of stabilities in the observed and predicted palindromes was quite similar (data not shown). The frequency of the predicted palindromes was then corrected to account for the number of false positives. Based on this analysis, we estimate that an average 100 kb region of centromere contains over 800 palindromes on average (Fig. 7).

Fig 7.

Predicted distribution of TRP/INV-associated palindromes. (Top) Heat map of predicted distribution of peri-centromeric and centromeric TRP/INV-associated palindromes on chromosome 4 (before ∆G filtering). (Bottom) Predicted centromeric TRP/INV-associated centromeric palindromes are present at higher densities than dispersed TRP/INV-associated palindromes (measured on a 60 kb segment of chromosome 3). The assumptions used for the calculations of these values are described in the text. CEN, centromere.

However, these values could not be applied to non-centromeric sequences due to the differing TRP/INV-associated palindrome parameters in AT-rich and dispersed regions. Examination of non-centromeric TRP/INV-associated palindromes in chromosomes 1, 3, and 4 revealed stem lengths of 5–12 bp, loop sizes of –19 bp, and stem mismatches of 0–1. In this case, the observed and predicted patterns were significantly different, with ∆G_cf–∆G_lin values of the observed palindromes distributed as follows: 15 at 3.1–6.0 kcal/mol (class 1), 10 at 6.1–9.0 kcal/mol (class 2), and 14 at −9.1 to 12.0 kcal/mol (class 3). Since the predicted TRP/INV-associated palindrome output had a different relative distribution within a 60 kb region of centromere 3 (24 in class 1, 105 in class 2, and 507 in class 3), we normalized the observed recombination frequencies using a weighted eligibility model (45, 46) (see Materials and Methods). Under these conditions, Palindrome Analyzer predicted ~100 palindromes/100 kb. Although these values reflect the characteristics of most but not all palindromes, the overall frequency of palindromes clearly mirrors the enrichment of TRP/INVs in centromeric regions. We propose therefore that the abundance of short palindromes is a major driver of TRP/INV formation.

DISCUSSION

Here, using the A. flavus model system, we report the first evidence, to our knowledge, for spontaneously forming TRP/INVs, a process that is driven by the frequency of palindrome formation. Several lines of evidence support the identification of spontaneous but infrequent inverted triplications in Aspergillus flavus through structural variant analysis following high-coverage, third-generation sequencing. Although we did not apply any intentional selection during growth, selective pressures induced by vegetative growth are possible. Indeed, in studies to be presented elsewhere, we found reductions in conidiation levels between the 5th and 10th weeks of culturing (data not shown).

First, BLAST searches of individual reads of presumptive large insertions generated after two independent methods of genome sequencing (ONT and PacBio) predicted TRP/INV structures consistent with unprocessed symmetric and processed asymmetric TRP/INV forms (8–10), albeit differing in abundance. Asymmetric products with deletions at one or more junctions are consistent with a role for a classical or alternative non-homologous end joining (see below). We find many palindromes and quasi-palindromes at the site of genomic rearrangement, arguing for a mechanism that favors the use of a hairpin structure to generate the inverted sequence.

Second, we have shown that TRP/INVs are clustered at centromeres and subtelomeric AT-rich DNA. Interestingly, the abundance of palindromes in centromeric regions mirrors the clustering of TRP/INVs. In addition, these studies have found that palindrome formation (as well as direct and inverted repeats) is also significantly elevated in genomic DNA, compared to randomly generated DNA having the same AT content. These data argue for the central role of palindromes in driving TRP/INV formation and clustering in centromeric and subtelomeric regions. However, palindromes and AT-rich DNA are likely insufficient for TRP/INVs, since the mitochondrial genome has a significantly decreased TRP/INV frequency per unit size, despite being AT-rich. In addition, mitochondrial TRP/INVs were present at a lower abundance than in dispersed sequences, despite having a comparable density of predicted TRP/INV-associated palindromes. These data argue for a preference for inverted triplications in the nuclear genome.

Third, we found that some palindromes were predicted to be associated with more extended regions of homology or had more complex pairing structures. Although it is likely that predictive algorithms only reflect a snapshot of likely stable forms, the data do raise the interesting possibility that palindromes may have functions beyond or in addition to strand transfer (required for template switching [see below]) such as in structural perturbations contributing to fork stalling.

Fourth, we found that the TRP/INVs are infrequent and do not become fixed in the population after 15 cycles of forced vegetative growth. For the G15 cultures sequenced by ONT, we identified 221 reads out of 3.6 × 10⁶ reads of sizes (>2 kb) that were used in this analysis. Since each read is derived from an independent DNA molecule, this suggests an overall frequency of 6 × 10⁻⁵ TRP/INV/genome. This value is only an approximation since the coverage of different genomic regions is not identical using either sequence method. These structures arise at every stage of subculturing and are in low abundance, possibly due to high rates of TRP/INV processing. The fact that TRP/INVs are unstable suggests that rearrangements of TRP/INVs are themselves sources of genome instability.

Such TRP/INV instability may arise via distinct mechanisms of homologous and non-homologous recombination. Recombination between the direct repeats would result in excision of a TRP/INV extrachromosomal circle, restoring the original structure. On the other hand, recombination between inverted repeats would generate a novel pattern of direct repeats and inversions. Indeed, the high frequency of direct and inverted repeats that we have found in centromeric and subtelomeric regions is among the expected products of TRP/INV recombination. In contrast, unequal recombination between sister chromatids would result in both the loss and duplication of these regions.

The instability of palindrome DNA that leads to cruciform and hairpin extrusion and promotes double-strand break formation has been well documented in both prokaryotes and eukaryotes (4). Several pathways may be responsible for the resolution of the double-stranded breaks, including exonucleolytic degradation coupled with microhomology-dependent non-homologous end-joining (MMEJ) (47, 48) or alternatively, Ku-dependent non-homologous end-joining (NHEJ) (48). MMEJ is indeed found in yeast and filamentous fungi, although by mechanisms independent from the pol θ-dependent process in multicellular organisms, and is likely to involve, in part, subunit 3 (Schizosaccharomyces pombe cdc27 and Saccharomyces cerevisiae pol32) of polymerase ∂ (47, 49). The frequent processing of palindromic DNA in AT-rich sequences coupled with MMEJ or NHEJ is also expected to generate small deletions within these regions. Indeed, resolution is likely to account for characteristics of asymmetric class 2 and 3 TRP/INVs.

It is also possible that the transient presence of TRP/INV could be due to a strong selection against the rearranged sequences leading to diminished growth rate or cell death. This explanation seems unlikely, however, given the multinucleate nature of mycelial growth in A. flavus, which would permit the retention of deleterious rearrangements (29). If such a selection is present, however, it is more likely to be mediated through TRP/INV disruption of coding regions.

Our characterization of palindromic sequences associated with TRP/INVs suggests a testable model for spontaneous palindrome formation in Aspergillus, akin to those previously described under conditions of gene amplification (8) in Salmonella (Fig. 8). We propose that the structural constraints of palindromes result in fork stalling, followed by subsequent palindrome resolution, forming a foldback that primes the replication of the inverted repeat. Processing of the second palindrome is then responsible for the synthesis of the direct repeat. Replication stalling has indeed been observed in experimentally manipulated constructs in both prokaryotes and eukaryotes (12, 13). Following palindrome-mediated template switching, microhomology-mediated pathways, such as MMEJ or “micro” single-strand annealing (SSA) processes, would prime the synthesis of the direct repeat (50). This model is appealing, since we can experimentally test the genetic dependence on proteins involved in homologous recombination, MMEJ, and micro-SSA, as well as the inhibition of these processes by Ku binding and protection of damaged DNA. The presence of palindromes formed in the TRP/INV product can also give rise to multiple cycles of inversion and duplication, leading to the observed complex products (data not shown). Our ongoing work is testing this model by genetic analysis and is probing for the presence of these structures in other species, such as budding yeast.

Fig 8.

A mechanism for palindromic and direct repeat strand exchange in TRP/INV formation. A four-step model modified from Roth and colleagues based on their extensive studies in Salmonella is presented (see text): 1, palindrome extrusion and processing in replicating DNA and resolution by nucleolytic activities; 2, formation of hairpin foldback that primes synthesis of the inverted repeat; 3, DNA synthesis reaches the second palindrome (green); 4, Ku-independent (or inhibited) mediated strand transfer and the priming of the direct repeat synthesis. Created in BioRender. Smith, A. (2026) https://BioRender.com/f4060u4.

One of the exciting aspects of this study is the accumulation of TRP/INV in AT-rich subtelomeric and centromeric sequences, which is due, in part, to the critical role of palindrome abundance in TRP/INV formation. Indeed, this study is the first characterization of short repeat-mediated recombination within centromeric and subtelomeric regions. This finding may be related to our previous observation that AT-rich regions at the centromeres of non-homologs (and in many cases, even homologs) are highly divergent. We also found that specific subtelomeric ATE sequences are present at different sites within and among strains. This observation may be related to the role of subtelomeres in the elimination or the evolution of secondary metabolite gene clusters and their participation in other selected rearrangements in fungi (25). Both results support an equilibrium model whereby deleterious functions such as perturbation of kinetochore function or rampant subtelomeric rearrangement may counteract evolutionary advantageous roles for centromeric divergence and subtelomeric sequence shuffling. We propose that mechanisms that regulate such an equilibrium must be present at centromeres and subtelomeres, possibly modulated by recombination activities. In addition, heterochromatin formation may be regulated. The fungal centromeric (and probably subtelomeric) AT-rich sequences are heterochromatic. In yeast, as in most eukaryotes, histone modification and chromatin condensation lead to reduced levels of transcription, recombination, and DNA repair in heterochromatin (51–53). For example, the histone deacetylase protein Sir2 is responsible for telomeric heterochromatin condensation and position effects (54). The use of A. flavus sir2 (sirE) mutants with wide-ranging phenotypes (55) is an attractive possibility to test this hypothesis. Such regulation may be the underlying basis for the selection against larger sizes of centromeric TRP/INVs. In addition, subtelomeric genes that are bounded by heterochromatic regions may serve to regulate transcription from these regions.

Our findings suggest that centromeres and subtelomeres can be more recombinationally active than anticipated. ATEs are maintained in all filamentous fungi and are indeed sites of evolutionary polymorphism (20). However, any region that is high in AT or GC content has a statistically higher probability of palindrome formation and may be subject to a similar process, a possibility that requires further exploration. The TRP/INVs reported here are yet another example of the reported association of AT-rich sequences with genetic instability (56–58), which may form hot spots for rearrangement in the eukaryotic genome in normal and diseased cells. The exploration of spontaneous TRP/INVs in other genetically manipulatable organisms, using the methods described above, will be a valuable future direction to test the generality of this finding.

TRP/INVs are associated with genetic diseases. Brewer et al. (14) conducted an extensive study that demonstrated that many human disease-related loci had structures consistent with the involvement of TRP/INVs, and these loci were often flanked by direct repeats (14). Other laboratories have also demonstrated that these structures are among the most prevalent mutational signatures in cancer and are a common source of disease-related genomic instability (59, 60). Understanding the fundamental mechanisms of TRP/INV formation in the absence of selection will be critical to understanding the underlying basis of genetic instability that precedes the chromosomal amplifications found in these disease states.

MATERIALS AND METHODS

Strains, media, and growth conditions

G0 spore stocks of Aspergillus flavus NRRL 3357 were generated following nine-point inoculations of a diluted 1 × 10³/mL spore stock onto the surface of double-strength 5/2 agar (per liter: 50 mL V8 juice, 40 g agar, pH 5.2 [61, 62]) and incubation at 30°C in the light for 5 days, a condition that promotes conidiation. NRRL 3357 conidia were collected from plates by gentle scraping of the agar surface with 10 mL sterile 0.01% Triton X-100, then transferred to a sterile 15 mL screwcap polypropylene tube, and stored at 4°C. For forced vegetative growth studies, addition of 1 mL (about 1 × 10⁸) of the spore suspension into 100 mL of A&M medium [per liter: sucrose, 50 g; (NH₄)₂SO₄, 3 g; KH₂PO₄, 10 g; MgSO₄·7H₂O, 2 g; Na₂B₄O₇·10H₂O, 0.7 mg; (NH₄)₆Mo₇O₂₄·4H₂O, 0.5 mg; Fe₂(SO₄)₃·6H₂O, 10 mg; CuSO₄·5H₂O, 0.3 mg; MnSO₄·H₂O, 0.11 mg; and ZnSO₄·7H₂0, 17.6 mg; pH 6.5] and incubation at 28°C at 200 rpm in the dark for 7 days led to growth cycle 1 of the experiment (G1). After 7 days, 20 mL of the G1 culture was removed from the 100 mL A&M shake culture and placed in a sterile 50 mL screw cap, polypropylene tube, then macerated for 30 s using the Tekmar Tissuemizer (Tekmar Co., Cincinnati, OH). One milliliter of macerated G1 mycelia was transferred into fresh 100 mL A&M broth in a sterile 250 mL Erlenmeyer flask and incubated for another 7 days. This culture represented the second “growth cycle” (G2) of vegetative growth. This process was continued for 15 cycles (15 weeks, 105 days). We diluted a 10 µL aliquot of macerated mycelia in 990 µL sterile water, spread 10 and 100 µL aliquots onto the surface of double-strength 5/2 medium, and incubated for three days to observe colony phenotype. Mycelia for genomic DNA preparation from 5 (G5), 10 (G10), and 15-week (G15) cultures of NRRL 3357 mycelia were collected as follows: at the end of the desired growth period, 50 mL of potato dextrose broth (PDB, Difco) was inoculated with 0.5 mL of macerated mycelia and incubated for 24 h at 28°C with agitation (200 rpm). For NRRL 3357 G0, genomic DNA was prepared from a 24 h 50 mL PDB culture that had been inoculated with 0.5 mL of a 1 × 10⁸ spores/mL NRRL 3357 stock. Mycelia were collected from the 50 mL PDB shake cultures by vacuum filtration through sterile Miracloth (EMD Millipore, Temecula, CA). The remaining liquid was squeezed from the collected mycelial mat by pressing it between paper towels. Mycelia were lyophilized for at least 24 h and stored at −80°C until needed.

Preparation of genomic DNA

Genomic DNA was prepared from lyophilized mycelia using the Qiagen Plant Mini Kit (Qiagen, Beverly, MA) following the manufacturer’s protocol. Lyophilized mycelia were ground to a fine powder in liquid nitrogen and stored at −80°C until needed. Briefly, 15 mg of each ground sample was processed in duplicate. After a spin column, the DNA was eluted with two 100 µL elutions using Qiagen AE buffer. The DNA suspension was further enriched by the addition of an equal volume of 25:24:1 phenol:chloroform:isoamyl alcohol (IAA). The upper layer was collected and extracted with an equal volume of 24:1 chloroform:IAA, and then precipitated by the addition of one-tenth volume of 3 M sodium acetate pH 5.2 solution and two volumes of cold 95% ethanol followed by centrifuging at 10 K rpm for 10 min. The DNA pellet was washed with 70% ethanol, dried in a speed vac, and resuspended in 50 µL TE. DNA quantity and quality were determined using a NanoDrop OneC (Thermo Scientific). The quality of genomic DNA was also determined by electrophoresis on a 0.7% agarose gel.

Sequencing

CD Genomics (Shirley, NY) carried out all sequencing. Both sequencing methods gave an average of 500- to 1,500-fold coverage, although not equally distributed over all parts of the genome.

Nanopore sequencing

The library was prepared using the SQK-LSK112 ligation kit under standard conditions. The purified library was loaded onto primed R10.4 Spot-On Flow Cells and sequenced using a PromethION sequencer (Oxford Nanopore Technologies, Oxford, UK) with 48-h runs (35). Base calling of raw data used the Oxford Nanopore Dorado (v0.7.3) (https://github.com/nanoporetech/dorado/releases).

PacBio HiFi sequencing

The SMRTbell library was constructed using the SMRTbell Express Template Prep Kit 2.0 (Pacific Biosciences). Briefly, ~2 µg of the genomic DNA was carried into the first enzymatic reaction to remove single-stranded overhangs followed by treatment with repair enzymes to repair any damage that may be present on the DNA backbone. After DNA damage repair, ends of the double-stranded fragments were polished and subsequently tailed with an A-overhang. Ligation with T-overhang SMRTbell adapters was performed at 20°C for 60 min. Following ligation, the SMRTbell library was purified with 1× AMPure PB beads. The size distribution and concentration of the library were assessed using the FEMTO Pulse Automated Pulsed-Field Capillary Electrophoresis (Agilent Technologies, Wilmington, DE) and the Qubit 3.0 Fluorometer (Life Technologies, Carlsbad, CA). Following library characterization, 3 μg was subjected to a size selection step using the BluePippin System (Sage Science, Beverly, MA) to remove SMRTbells ≤15 kb. After size selection, the library was purified with 1× AMPure PB beads. Library size and quantity were assessed using the FEMTO Pulse and the Qubit dsDNA HS Reagent Assay kits. The sequencing primer and Sequel II DNA Polymerase were annealed and bound, respectively, to the final SMRTbell library. The library was sequenced using 8M SMRT Cell on the Sequel II System with Sequel II Sequencing Kit (39).

Quality control analyses

QC analysis was carried out using the MultiQC (63) of the ONT G0, G5, G10, and G15 DNA PacBio sequenced G15 DNA.

Assemblies

We bootstrapped G0 and G15 assemblies using both A. flavus strain NRRL-3357 (GCA_009017415.1_ASM901741v1) (23) and mitochondrial (A. flavus SRRC1009) (64) sequences that were concatenated to prepare a genome fasta file. This file served as input for G0 ONT or G15 PacBio reads to Rebaler (https://github.com/rrwick/Rebaler) (65).

Structural variant determination

Structural variants were identified relative to the G0 assembly using Sniffles 2 (v2.2) (37). Alignments used minimap 2 (38) (v2.26-r1175) with a gap open and gap extension set at 16 and 2, respectively. This version was determined empirically to be only ~70% efficient for low-abundance sequences in simple sequence regions, leading to both false positives and negatives. We therefore often coupled the algorithm with visual confirmation using IGV [v2(10.0)] (66, 67).

For qualitative analyses, structural variants were identified as ≥300 bp insertions with variant read numbers (DV) of 3 and confirmed by IGV analysis. Individual reads from these nanopore sequenced DNA were then used as BLAST queries to the G15 or G0 assemblies using SequenceServer (London, UK) (67). BLAST alignments in high A + T-rich DNA are difficult when the simple sequence filter is turned on. We therefore turn off this filter for the analysis of ATE sequences.

For quantitative analysis, we selected all reads that had a DV of 1; individual reads from these nanopore sequenced DNAs were then analyzed by BLAST as noted above. A similar analysis was carried out for PacBio insertions of ≥200 bp. For all studies, the variant frequency (VAF) calculated as {variant [read number (DV)] / [reference read number (DR)] × 100}. VAF values ranged from 0.1% to 0.2%, consistent with a high coverage.

The most frequent products identified by BLAST analysis among large “insertions” were duplications and tandem inverted duplications. The TRP/INV species were categorized based on the extent of deletion at each junction between duplicated and inverted sequences. We established 50 bp as the maximum error at a junction, which is likely due to sequencing and alignment defects.

The repeat finder function of the UGENE graphical interface program (30) was also used to characterize the direct and inverted repeats present within a subset of individual reads. For this analysis, we used windows of 100 bp, homology of 85%, minimum distance between repeats of 10 bp, and maximum distance between repeats of 100 bp.

Palindrome analysis

Palindromes that initiate TRPV/INV formation were identified first by the detection of inverted sequence junctions homologous to both genomic strands after BLAST analysis of single reads to the G15 assembly. We determined the predicted relative thermodynamic stability (T_m, ∆G, ∆H, and ∆S), and the secondary structure predictions of the palindromes by the application of UNAfold mfold 4.2 program (41–43), using our default settings of 25°C, 3 mM MgCl₂, and 50 mM NaCl. The stability of palindrome structures was routinely quantified in the presence of short flanking single-strand regions for uniform alignment, although the listed coordinates exclude these tails (Tables 2 to 4). As a control, we characterized the palindromes at the junctions of inverted repeats formed in a yeast model system (7). The UNAFold software (RPI 2017-042) is under license 80-00840 from Rensselaer Polytechnic Institute. We also analyzed the potential palindromes in the 200 bp of genomic sequence surrounding the incipient site of recombination in TRP/INVs exhibiting deletions of ≤40 bp.

The UNAfold algorithm most often generated simple hairpin predictions. However, three types of other pairing patterns were also observed. Structures in which the hairpin was closely adjacent to more extensive base pairing were termed extended forms, while predicted patterns containing two or more adjacent hairpins were termed multipalindromic. Structures that were paired but appeared to contain some form of distortion were termed complex forms. Examples of these forms are shown in Fig S3.

Overall palindrome predictions for centromeric and non-centromeric regions were determined by Palindrome Analyzer (34), using parameters exhibited by the majority of palindromes associated with the TRP/INVs from centromeric or dispersed domains of chromosomes. The same parameters were then used for the entire genome. We first determined the most frequent classes of TRP/INV-associated palindromes in the terminal portions of centromere 3 . This class had stem sizes of 12–100 bp and loop sizes of 4–8 bp, with 0–2 stem mismatches. We then identified these partial structures in the output of the Palindrome Analyzer (34) and analyzed them for the presence of part or all of the input palindrome structure. These ∆G_cf–∆G_lin values generally ranged from 10 to 16 kcal/mol. Since the distribution was similar for the observed and predicted palindromes, we determined the number of predicted palindromes as the sum of all candidates within this range. However, this number was corrected for those structures that had stabilities lower than the study threshold. For each centromere, we added candidates in the 10–16 kcal/mol range and adjusted these for the expected number of false positives.

We could not apply this method directly to non-centromeric sequences due to the different characteristics of TRP/INV-associated palindromes in palindromes in AT-rich and dispersed regions. Examination of non-centromeric palindromes from chromosomes 1, 3, and 4 revealed stem lengths ranging from 5 to 12 bp and loops ranging from 4 to 19 bp, with stem mismatches from 0 to 1. We empirically converted ∆G into ∆G_cf–∆G_lin values by first remeasuring ∆G at 37°C, 50 mM NaCl, and 1 mM MgCl₂._2. The values, representing the stability of a hairpin, were then doubled to generate ∆G_cf–∆G_lin, a measure of cruciform stability. ∆G_cf–∆G_lin values ranged from 3 to 12 kcal/mol, distributed as follows: 15 at 3–6 kcal/mol (class 1), 10 at 6.1–9 kcal/mol (class 2), and 14 at >9 kcal/mol (class 3). Since the predicted output from Palindrome Analyzer for a 60 kb region of centromere 3 had a different distribution (24 in class 1, 105 in class 2, and 507 in class 3) than the TRP/INV-associated palindromes, we normalized the output values using a weighted eligibility model (45, 46). Specifically, to estimate how many palindrome sites in a genomic interval are competent to initiate recombination, we compared the observed distribution across palindrome classes with the relative abundance of candidates. For each class $i$, we calculated a per-candidate usage rate $r_i=U_i/n_i$, where $U_i$ is the number of observed TRP/INV-associated palindromes and $n_i$ is the number of candidate palindromes in the interval. To account for the unequal usage across classes, we normalized by the maximum observed rate, defining a relative eligibility fraction $h_i=r_i/{\mathrm{m}\mathrm{a}\mathrm{x}}_j{r}_j$. The estimated pool of competent candidates in the interval is described by

$${\displaystyle {\displaystyle {\displaystyle A=\sum _i{h}_i{n}_i}}}$$

Sequence structure in centromeric and subtelomeric ATEs

To analyze the overall frequency of repeats within centromeric regions, we analyzed each centromeric region and two scrambled derivatives of each centromere (using Sequence Suite [31]) that maintained both length and base composition. We then employed several techniques to identify repeats. First, we used the repeat finder function of UGENE (30) to find both direct and inverted repeats (window size = 15 bp, homology = 90%, minimum space between windows = 10 bp, and maximum space between windows = 200). Second, to analyze the frequency of short palindromes, we utilized the Palindrome Analyzer (34) (repeat length range of 12–100 bp, a loop size of 4–19 bp, and a mismatch setting of 0). For an independent measure of the frequency of inverted repeats (32) and tandem repeats, we employed Inverted Repeat Finder and EMBOSS tandem repeat finders, respectively, with the parameters given in the Table S2 legend (33).

In addition, to ascertain the frequency of short repeats within ATEs, we identified 250 bp DNAs from centromeric and subtelomeric ATEs having a 90% A + T content and assayed for self-homology by BLAST analysis to identify direct, inverted, and palindromic repeats. We rated ungapped regions of homology as a function of the BLAST e-value. We compared this distribution with 250 bp sequences derived from an Excel random sequence generator (Microsoft) set at 90% A + T DNA. Both perfect and mismatched repeats at each value were combined for the analysis of direct and inverted repeats. To quantify the statistical significance of the deviation from non-random behavior, we evaluated the number of sequences containing at least one event present at e-values of <10⁻⁴ for genomic and random data sets by the Fisher exact test. Similarly, perfectly matched palindromes were analyzed at BLAST e-values of <10⁻⁸ and evaluated by the Fisher exact test.

DATA AVAILABILITY

All data presented herein are available to the research community. Author will also grant access to Tables 1–5, and Tables S1, S2, S3, S4, S5A-C as Excel spreadsheets at

https://wavetulane-my.sharepoint.com/:x:/r/personal/alustig_tulane_edu/_layouts/15/Doc.aspx?sourcedoc=%7BC844C84D-7D50-452B-A4F8-219C943B9AAD%7D&file=Cary%20et%20al%20Exel%20Srepadsheets.xlsx&action=default&mobileredirect=true

The raw data from nanopore and PacBio sequencing are publicly available at NCBI SRA listed as BioProject PRJNA1193636 with the following BioSample accession numbers: SRX26956798 (G0 Nanopore), SRX26956799 (G5 Nanopore), SRX26956800 (G10 Nanopore), SRX25956801 (G15 ), SRX25956802 (G15 PacBio-1), and SRX25956803 (G15 PacBio-2). The sequencing assemblies are listed as BioProject PRJNA1399503 with the following accession numbers. The SAMN45150368: SAMN45150363 (G0 Nanopore) SAMN45150367, SAMN45150364 (G5 Nanopore), SAMN45150365 (G10 Nanopore) and SAMN45150363, SAMN45150366 (G15 Nanopore), and SAMN45150366 (G15 PacBio).

SUPPLEMENTAL MATERIAL

The following material is available online at https://doi.org/10.1128/spectrum.03473-25.

Supplemental materials. spectrum.03473-25-s0001.docx.

Supplemental text, Figures S1 to S5, and Tables S1 to S5.

ASM does not own the copyrights to Supplemental Material that may be linked to, or accessed through, an article. The authors have granted ASM a non-exclusive, world-wide license to publish the Supplemental Material files. Please contact the corresponding author directly for reuse.