Recombination-independent recognition of DNA homology for meiotic silencing in Neurospora crassa

Significance

Homologous chromosomes (or selected chromosomal regions) can engage in sequence-specific pairing as intact entities. The mechanism behind this process remains mysterious, in part because prevailing evidence suggests that DNA molecules must be broken before being evaluated for homology. In this work, we study a phenomenon that detects dissimilar DNA fragments present at the allelic positions on a pair of homologous chromosomes during meiosis. We find that the recognition of such fragments is incompatible with canonical homology-sensing mechanisms that operate on broken DNAs. Instead, it is consistent with the direct pairing of homologous DNA duplexes through a series of short, interspersed quadruplexes. This pairing process may represent a general property of DNA and underlie a large repertoire of homology-dependent phenomena.

Abstract

The pairing of homologous chromosomes represents a critical step of meiosis in nearly all sexually reproducing species. In many organisms, pairing involves chromosomes that remain apparently intact. The mechanistic nature of homology recognition at the basis of such pairing is unknown. Using “meiotic silencing by unpaired DNA” (MSUD) as a model process, we demonstrate the existence of a cardinally different approach to DNA homology recognition in meiosis. The main advantage of MSUD over other experimental systems lies in its ability to identify any relatively short DNA fragment lacking a homologous allelic partner. Here, we show that MSUD does not rely on the canonical mechanism of meiotic recombination, yet it is promoted by REC8, a conserved component of the meiotic cohesion complex. We also show that certain patterns of interspersed homology are recognized as pairable during MSUD. Such patterns need to be colinear and must contain short tracts of sequence identity spaced apart at 21 or 22 base pairs. By using these periodicity values as a guiding parameter in all-atom molecular modeling, we discover that homologous DNA molecules can pair by forming quadruplex-based contacts with an interval of 2.5 helical turns. This process requires right-handed plectonemic coiling and additional conformational changes in the intervening double-helical segments. Our results 1) reconcile genetic and biophysical evidence for the existence of direct homologous double-stranded DNA (dsDNA)–dsDNA pairing, 2) identify a role for this process in initiating RNA interference, and 3) suggest that chromosomes can be cross-matched by a precise mechanism that operates on intact dsDNA molecules.

Correct pairing of homologous chromosomes in meiosis is essential for ensuring fertility and preventing birth defects. While in some organisms this process depends on programmed DNA breakage and recombination, in many others, it engages apparently intact chromosomes (1). Such recombination-independent pairing was proposed to rely on indirect sequence readouts via locus-specific proteins or RNAs while also being guided by landmark chromosomal features, such as transcriptional hubs, heterochromatin domains, centromeres, and telomeres (2 –6). Alternatively, a possibility exists that recombination-independent pairing can be specified at the DNA level (7, 8), but a molecular model, consistent with in vivo observations, has been lacking (9).

A process known as “meiotic silencing by unpaired DNA” (MSUD) induces transient RNA interference against dissimilar DNA sequences present at the allelic positions on a pair of homologous chromosomes (10). In the fungus Neurospora crassa, the capacity of MSUD to detect heterologous DNA segments as short as 1 kbp suggests that it involves an efficient recognition mechanism (11). Because meiosis in N. crassa and related fungi occurs immediately after karyogamy, this mechanism cannot rely on preexisting (premeiotic) pairing. More than two dozen MSUD factors with roles in transcription, messenger RNA processing, RNA interference, nuclear dynamics, and chromatin remodeling were identified, yet the basis of DNA homology recognition for MSUD has remained elusive (11, 12).

In addition to MSUD, N. crassa has another homology-dependent process known as “repeat-induced point mutation” (RIP), which takes place exclusively in premeiotic haploid nuclei (13, 14). During RIP, gene-sized duplications of genomic DNA become mutated by numerous cytosine-to-thymine transitions along their entire lengths (13, 14). The capacity of RIP to detect DNA repeats irrespective of their particular sequence, coding potential, and genomic positions suggests that it requires an exhaustive genome-wide homology search. Yet, RIP functions normally in the absence of MEI-3 (the only RecA-like recombinase in Neurospora) and SPO11 (a topoisomerase II–like protein that creates DNA breaks to initiate meiotic recombination) (15). By analyzing the ability of synthetic repeats with programmed patterns of homology to induce mutations, it was discovered that RIP detected the presence of matching trinucleotides (triplets), but only if those triples were interspersed with a periodicity of 11 or 12 bp along the two participating DNA segments (15). These results uncovered the existence of a cardinally different mechanism of DNA homology recognition that likely matches intact double-stranded DNAs (dsDNAs) directly, without involving single-stranded intermediates (15).

Recognition of interspersed homology for RIP was accounted for by a recently proposed all-atom model of direct dsDNA–dsDNA pairing (16). This model is based on the physical property of standard Watson–Crick (WC) base pairs to display distinct yet self-complementary electrostatic patterns along their major groove edges. This property allows homologous double-stranded stacks to pair without breaking WC bonds (16). According to the model, a sequence-specific contact between two DNA double helices represents a short stack of three to four planar quartets formed by identical (homologous) WC base pairs. The favorable energy of this reaction includes a large nonspecific contribution of ionic interactions and a sequence-specific hydrogen bonding term. The model has defined the biophysical properties of the quadruplex-based contact, yet it did not explore the role of the intervening double-helical segments in promoting overall pairing (16).

We now report that MSUD does not depend on the canonical mechanism of meiotic recombination. MSUD also proceeds normally in the absence of heterochromatin formation involving H3K9me3 and H3K27me3. At the same time, MSUD is promoted by REC8, thus revealing another recombination-independent role for this central meiotic factor. We show that MSUD can discriminate between the physical absence of DNA and the absence of DNA homology at the allelic position. Strikingly, we find that interspersed homologies containing short tracts of sequence identity arrayed with a periodicity of 21 or 22 base pairs are readily recognized as pairable during MSUD. Using this parameter in all-atom molecular modeling, we discover that homologous dsDNAs can associate by establishing quadruplex-based contacts with an interval of 2.5 helical turns. This process requires right-handed plectonemic coiling and additional conformational changes in the intervening double-helical segments. Taken together, our results 1) reconcile genetic and biophysical evidence for the existence of the direct homologous dsDNA–dsDNA pairing, 2) identify a role for this process in initiating posttranscriptional silencing by RNA interference, and 3) suggest that chromosomes can be cross-matched by a precise mechanism that operates on intact dsDNA molecules.

Results

In N. crassa and related fungi, the meiocyte (ascus) starts to develop shortly after karyogamy (Fig. 1A ). The ascus enlarges dramatically during prophase I, which is also when the pairing of homologous chromosomes takes place (Fig. 1A ). Each fruiting body (perithecium) contains 100 to 200 asci at different stages of meiosis. When dissected from the fruiting body, the asci form a rosette (Fig. 1B ).

Fig. 1.

MSUD does not depend on the canonical mechanism of meiotic recombination. (A) In N. crassa, meiosis occurs immediately after karyogamy. Meiotic cells (asci) elongate dramatically and become readily recognized during prophase I. Meiosis is followed by one round of mitosis, generating eight maturing ascospores. (B) A cartoon representation of the dissected rosette of asci. (C) Overall experimental procedure. Strain “GFP_Reporter1” (hH1-gfp::mei-3Δ, spo11Δ, sad-2+) ( SI Appendix, Table S2) is used as a standard male parent. (D) The reporter construct contains hH1-gfp (green) and a hygromycin resistance marker hygR (orange). The wild-type mei-3+ allele is replaced by the construct to generate the hH1-gfp::mei-3Δ allele. All DNA segments are drawn to scale. (E) Recombination-proficient, MSUD-proficient condition (Cross X1, SI Appendix, Table S3). (F) Recombination-proficient, MSUD-deficient condition (Cross X2, SI Appendix, Table S3). (G) Recombination-deficient, MSUD-proficient condition (Cross X3, SI Appendix, Table S3). (H) Recombination-deficient, MSUD-deficient condition (Cross X4, SI Appendix, Table S3). Test (female) genotypes are indicated. Representative meiotic nuclei are shown with white arrows. All panels displaying GFP expression are magnified twofold, as indicated.

MSUD Does Not Depend On the Canonical Mechanism of Meiotic Recombination.

The canonical meiotic pairing program requires SPO11 to produce dsDNA breaks and the eukaryotic RecA-like recombinases DMC1 and RAD51 to mediate a subsequent homology search (17). N. crassa has only one RecA protein, MEI-3, which is dispensable during vegetative growth but is essential during meiosis (18). Its lethal phenotype, however, can be suppressed by removing SPO11 (15). The ability of N. crassa to complete meiosis in the absence of MEI-3 and SPO11 allowed us to test the role of meiotic recombination in MSUD.

A microscopy-based approach was chosen to analyze MSUD by assaying expression of a histone H1–GFP (hH1-GFP) fusion protein (Fig. 1D ). Normally expressed hH1-GFP features readily detectable nuclear localization, but it rapidly and completely disappears from meiotic nuclei in heterozygous hH1-gfp crosses due to MSUD (19). Furthermore, when the hH1-gfp gene is provided solely by the male parent, hH1-GFP remains restricted to premeiotic, meiotic, and postmeiotic tissues, permitting its straightforward observation in dissected rosettes.

In the first reporter strain (GFP_Reporter1), hH1-gfp was inserted as a replacement of the mei-3+ allele (Fig. 1D ). GFP_Reporter1 had also the spo11+ allele deleted. By the overall design, this strain was always used as a standard male parent, while meiotic genotypes were manipulated by using different female strains (Fig. 1C ). Depending on which female strain was chosen, crosses could differ with respect to both recombination and MSUD proficiency. The cross could be recombination proficient (mei-3+/Δ, spo11+/Δ) or recombination deficient (mei-3Δ/Δ, spo11Δ/Δ). Additionally, the cross could be either MSUD proficient (sad-2 ^+/+ ) or MSUD deficient (sad-2Δ/+). The MSUD deficiency of sad-2Δ/+ crosses was caused by an effect known as “silencing the silencer,” in which an unpaired MSUD gene (in this case, sad-2+) was subjected to MSUD itself (11).

We found that hH1-GFP was completely and specifically silenced during meiosis in MSUD-proficient crosses (Fig. 1 E and G ). Critically, the levels of silencing appeared indistinguishable between the recombination-proficient and the recombination-deficient conditions (Fig. 1 E and G ). Under both regimes, the expression of hH1-GFP was restored by suppressing MSUD (Fig. 1 F and H and SI Appendix, Fig. S1 B and C ). The meiotic expression of hH1-GFP remained strong in the recombination-deficient cross with active MSUD if two identical hH1-gfp alleles were provided, in other words, if hH1-gfp could be paired ( SI Appendix, Fig. S1A ). Together, these results suggest that MSUD does not depend on the canonical mechanism of homologous meiotic recombination involving SPO11 and MEI-3.

MSUD Is Not Coupled to Noncanonical Genetic Recombination.

While SPO11 is required for chromosome synapsis in Neurospora meiosis (20), residual SPO11-independent meiotic recombination was also described in this organism (20). Those observations raised a possibility that the wild-type efficiency of MSUD in our spo11Δ/Δ crosses could be attributed to SPO11-independent breaks. To investigate this possibility, we measured genetic recombination over a large interval spanning more than half of chromosome I ( SI Appendix, Fig. S2A ). This interval included the his-3 locus, which is characterized by high levels of SPO11-independent recombination in crosses between certain Neurospora strains from a different genetic background (20, 21).

In N. crassa, SPO11-deficient crosses produce substantial numbers of aneuploid (partially diploid) progeny (20). These aneuploid progeny may undergo mitotic recombination and chromosome loss during vegetative growth, thus obscuring meiotic outcomes. To minimize the chance of encountering such progeny, csr-1 was used as one of two reference loci. More specifically, this gene was chosen because 1) only loss-of-function csr-1 alleles confer resistance to cyclosporin A, 2) csr-1–based resistance to cyclosporin A is recessive, and 3) csr-1+ itself is stable during the sexual phase (22). Overall, a recombination-deficient csr-1+/Δ cross can produce three csr-1 genotypes: haploid csr-1+ and csr-1Δ as well as partially diploid csr-1+/Δ; however, only csr-1Δ progeny should be retained on cyclosporin A. These isolates can be genotyped at the mat locus using a PCR-based assay designed to detect a potential mixture of mat A and mat a ( SI Appendix, Fig. S2 B and C ). The presence of mat A alleles among the csr-1Δ progeny would indicate a genetic exchange had occurred between csr-1 and mat loci. Preliminary analysis confirmed that the overall strategy was effective ( SI Appendix, Fig. S2D ).

We analyzed progeny from two of the crosses that were assayed for hH1-gfp silencing (Fig. 1 E and G ). In these crosses, GFP_Reporter1 provided the csr-1Δ allele, while the test strains provided the wild-type csr-1+ allele. The recombination-proficient cross (Fig. 1E ) yielded many recombinants: 26% of the csr-1Δ progeny carried the mat A allele ( SI Appendix, Fig. S2E , “mei-3+/Δ, spo11+/Δ”). In contrast, all 594 csr-1Δ isolates from the recombination-deficient cross (Fig. 1G and SI Appendix, Fig. S1B ) carried the mat a allele ( SI Appendix, Fig. S2E , “mei-3Δ/Δ, spo11Δ/Δ”). Interestingly, the occurrence of recombination in the heterozygous mei-3+/Δ, spo11+/Δ cross suggests that MEI-3 and SPO11 remained at least partially expressed in the presence of active MSUD.

We also considered the possibility that SPO11-independent recombinants were excluded from the offspring of the recombination-deficient cross because they required MEI-3 to complete DNA repair. To this end, we analyzed one additional cross between GFP_Reporter1 and another spo11Δ strain that carried the wild-type mei-3+ allele. From this cross, one recombinant isolate was recovered among the 594 csr-1Δ progeny ( SI Appendix, Fig. S2E , “mei-3+/Δ, spo11Δ/Δ”), suggesting that the level of residual SPO11-independent recombination was low. Taken together, these results show that MSUD is not coupled to genetic recombination.

MSUD Is Promoted by REC8.

MSUD is expected to rely on the preferential interactions between homologous chromosomes rather than sister chromatids. This phenomenon (generally known as “homolog bias”) requires the conserved meiotic kleisin REC8 (reviewed in ref. 23). N. crassa encodes a single REC8 ortholog (NCU03190) with a canonical structure ( SI Appendix, Fig. S3A ). Transcription of rec8+ is strongly up-regulated together with other meiotic genes (24).

To evaluate the role of REC8 in MSUD, we created another reporter (GFP_Reporter2) by integrating hH1-gfp as the replacement of csr-1+ in a strain that had both rec8+ and spo11+ deleted. The absence of homology between hH1-gfp and csr-1+ (provided by female parents) was expected to silence hH1-gfp ( SI Appendix, Fig. S3B ). One REC8-deficient cross (rec8Δ/Δ, spo11Δ/Δ) and one REC8-proficient cross (rec8+/Δ, spo11+/Δ) were tested. In both crosses, GFP_Reporter2 was used as a male parent. SPO11 was eliminated in the REC8-deficient cross to avoid potential complications associated with aberrant repair of SPO11-induced DNA breaks.

We found that the REC8-proficient cross contained a variable but small number of asci expressing hH1-GFP ( SI Appendix, Fig. S3D ). This result can be explained by either the “silencing the silencer” phenomenon or a less efficient recognition of the unpaired hH1-gfp::csr-1Δ allele by MSUD. Strikingly, many more asci expressed hH1-GFP in the REC8-deficient cross, indicating a silencing defect ( SI Appendix, Fig. S3C ). This defect was not absolute, and hH1-GFP was lost in some neighboring asci ( SI Appendix, Fig. S3C ). Thus, MSUD appears to be promoted by REC8, but it also occurs, albeit less efficiently, in the absence of this critical meiotic factor.

MSUD Does Not Depend on the Canonical Mechanisms of Heterochromatin Formation.

A process called “meiotic silencing of unsynapsed chromatin” induces repressive chromatin marks (such as di- and trimethylated histone H3 lysine 9, H3K9me2/3) on unpaired chromosomes in animals (25). In Caenorhabditis elegans, meiotic silencing of unsynapsed chromatin requires several RNA interference factors, including the RNA-dependent RNA polymerase EGO-1 (26). Thus, we explored if MSUD, reciprocally, could have a heterochromatin-related component.

In N. crassa, constitutive heterochromatin requires the lysine methyltransferase DIM-5, whereas facultative heterochromatin requires the lysine methyltransferase SET-7 ( SI Appendix, Fig. S4A ; ref. 27). DIM-5 and SET-7 mediate all H3K9me3 and H3K27me3 in N. crassa, respectively (27). Although dim-5Δ strains are largely infertile as females, this defect is suppressed by deleting set-7+ (28). We thus used the ability of N. crassa to complete its sexual phase in the absence of DIM-5 and SET-7 to assay the role of heterochromatin in MSUD. For this test, a third reporter strain (GFP_Reporter3) was engineered by replacing the csr-1+ allele with hH1-gfp in a heterochromatin-deficient (dim-5Δ, set-7Δ) background.

We analyzed two heterochromatin-deficient crosses, one of which was MSUD proficient ( SI Appendix, Fig. S4B ), while the other was MSUD deficient ( SI Appendix, Fig. S4C ). Rosettes were frequently aberrant, and very few elongating asci could be found in both conditions ( SI Appendix, Fig. S4 B and C ). Nevertheless, hH1-GFP was still expressed aptly in premeiotic and early meiotic nuclei ( SI Appendix, Fig. S4 B and C ). Importantly, hH1-GFP became silenced specifically in the elongating asci of the MSUD-proficient cross ( SI Appendix, Fig. S4 B and C ). These results suggest that meiotic silencing in N. crassa does not require heterochromatin formation.

Developing a Quantitative Approach to Analyzing the Homology Requirements of MSUD.

Thus far, our results show that MSUD parallels RIP in being independent from meiotic recombination. Earlier, we suggested that RIP involves a cardinally different homology recognition mechanism that matches dsDNAs directly without using single-stranded intermediates (15). This idea stemmed from the fact that RIP detects certain patterns of interspersed homology with an overall identity of only 36%, substantially below the limit of sequence recognition processes that rely on the annealing of complementary strands (15). Thus, it was important to test if MSUD could also recognize such interspersed homologies as pairable.

To understand the homology requirements of MSUD, a quantitative approach was developed (Fig. 2 A–C ). This approach relies on a classical MSUD assay, in which meiotic silencing of the Roundspore+ (Rsp+) gene results in the production of oval-shaped (“silenced”) rather than spindle-shaped (“wild-type”) ascospores (10). Two technical advances included 1) a sensitive reporter system to test engineered homologies (below), and 2) an image-processing algorithm to automatically segment and classify large numbers of ascospores ( SI Appendix, Materials and Methods ).

Fig. 2.

MSUD recognizes interspersed DNA homology as pairable. (A) A genetic reporter system to measure interactions between “Rsp1500” and “Test DNA” sequences. (B) An image processing algorithm to segment and classify ascospores (ejected meiotic products). (C) Interactions between Rsp1500 and Test DNA are assayed by quantifying the eccentricity of ascospores. (D) The eccentricity threshold of 0.81 (determined empirically) is used to classify ascospores as “wild-type” or “silenced”. “S” denotes the percentage of “silenced” ascospores (below the threshold). “N” corresponds to the total number of ascospores analyzed for each condition. Eccentricity distributions are shown as histograms. The following crosses are analyzed: X11 (“Identical alleles”) and X12 (“Deletion”) ( SI Appendix, Table S3). (E) Synthetic Test DNA sequences that form patterns of interspersed homology with Rsp1500 are designed based on the homology requirements of RIP (15). Corresponding overall sequence identities are indicated. (F) Eccentricity distributions are displayed as in (D). The following crosses are analyzed: X13 (“Perfect”), X14 (“7H-4N”), X15 (“6H-5N”), X16 (“5H-6N”), X17 (“4H-7N”), X18 (“3H-8N”), X19 (“2H-1N”), X20 (“Lambda”), X21 (“Ecoli1”), X22 (“Ecoli2”), X23 (“Sad-2”), and X24 (“Insertion”) ( SI Appendix, Table S3).

The reporter system is based on the genetic interaction between “Rsp1500” (an ectopic segment of Rsp+) and “Test DNA.” Both segments have the same length (1,500 bp) and occupy allelic positions near the his-3 gene (Fig. 2A ). While the sequence of Rsp1500 is fixed, the sequence of Test DNA can be varied as desired. Previously, we used a similar approach to dissect the homology requirements of RIP (15).

By overall design, if Rsp1500 and Test DNA are pairable, MSUD will not be induced, the endogenous Rsp+ alleles will be expressed normally, and spindle-shaped (more eccentric) ascospores will be produced (Fig. 2C ). On the other hand, if Rsp1500 and Test DNA are not pairable, MSUD will be activated, the endogenous Rsp+ alleles will be silenced in trans, and oval-shaped (less eccentric) ascospores will be produced (Fig. 2C ). To improve the sensitivity of the assay, an adjacent 2,500-bp region of heterology was engineered with two foreign sequences, “ECS” (Escherichia coli spacer DNA) and “LPS” (Lambda phage spacer DNA) (Fig. 2A ). In this situation, if Rsp1500 and Test DNA are not pairable, the total length of unpaired DNA containing the Rsp1500 segment will be 4 kbp instead of 1.5 kbp, a condition expected to enhance MSUD (29).

To extract ascospore features from brightfield images, an image-processing algorithm was developed (Fig. 2B ). Briefly, ascospores are detected using rough localization followed by local segmentation. Segmented objects are filtered based on their convexity and area values, and ascospores that are damaged, aberrant, or clumped together are excluded. The roundness of ascospores is determined using an eccentricity parameter ranging between zero (perfect circle) and one (straight segment). This approach benefited from the fact that a large number of ascospores could be obtained in a nearly pure form (Fig. 1B ).

A reporter strain carrying Rsp1500 adjacent to ECS (RSP_Reporter1) was always used as a female parent (Fig. 2A ). Test DNAs were provided by otherwise isogenic strains of the opposite mating type used as male parents. To calculate the percentage of silenced ascospores, an eccentricity threshold of 0.81 was applied (wild-type ascospores ≥ 0.81). For a given condition, three crosses were analyzed, each represented by 3 to 15 × 10⁴ ascospores, thus bringing the total number of assayed ascospores to 1 to 4 × 10⁵.

As a proof of principle, we analyzed two conditions, one featuring two identical reporter alleles and the other featuring a reporter allele and a deletion spanning the entire reporter construct (Fig. 2D ). The first condition produced 98.5% wild-type ascospores, while the second condition produced 95.9% silenced ascospores (Fig. 2D ). The corresponding distributions of eccentricity values were overlapping but separable. These results show that 1) ascospores can be classified as wild-type or silenced based on their eccentricity values, and 2) the reporter system has a broad dynamic range, permitting a systematic analysis of homology requirements for MSUD.

MSUD Recognizes Weak Interspersed DNA Homology as Pairable.

We engineered a series of synthetic Test DNA sequences, each related to Rsp1500 by a particular pattern of interspersed homology (Fig. 2E ). These patterns were designed with an “XH-YN” format, in which homology tracts of X bp are separated by nonhomology tracts of Y bp (Fig. 2E ). The following patterns were analyzed: 7H-4N, 6H-5N, 5H-6N, 4H-7N, 3H-8N, and 2H-1N (Fig. 2E ). In addition, three instances of random homology were created using 1,500-bp sequences chosen arbitrarily from E. coli and lambda phage. These sequences are unrelated to the ECS and LPS segments.

To ensure that our Test DNAs did not contain unintended cryptic homology, we plotted all possible 4-bp and 6-bp matches between each Test DNA (both in forward and reverse-complement orientation) and Rsp1500 [ SI Appendix, Fig. S5; all high-resolution plots are provided in SI Data File 1 (30) along with the scripts used to create them]. Some of the plots reveal linear patterns of variable density along the two principal diagonals, corresponding to the programmed homologies (direct or inverted). The lack of other visible linear patterns argues against the existence of cryptic homology in the analyzed constructs.

The presence of sensitizing heterology did not compromise the expected pairing of the identical Rsp1500 segments (compare Fig. 2D , “Identical alleles” and Fig. 2F , “Perfect”). At the same time, all three instances of random homology produced 74 to 77% silenced ascospores (Fig. 2F , “Lambda,” “Ecoli1,” and “Ecoli2”), substantially less than a deletion of the same length (Fig. 2D ). This result suggests that the Rsp1500 segment on the first homologous chromosome becomes partially protected from MSUD by the presence of heterologous DNA at the allelic position on the second homologous chromosome. If the “Lambda” DNA is extended by 3.5 kbp, the proportion of silenced ascospores increases to 87.7%, which corresponds to an intermediate level between random homology and the deletion (Fig. 2F , “Insertion”). This result suggests that the protection provided to Rsp1500 by heterologous allelic DNA is maximal when their lengths are matching.

Strikingly, interspersed homologies 7H-4N, 6H-5N, and 5H-6N failed to induce MSUD, while 4H-7N yielded only a small (8.1%) fraction of silenced ascospores (Fig. 2F ). Much stronger MSUD was triggered by 3H-8N (Fig. 2F ). Finally, 2H-1N induced silencing at the level indistinguishable from that of random homology (Fig. 2F ). To confirm that these effects were due to MSUD, we tried to suppress it by providing an ectopic 1,500-bp fragment of sad-2+ as the Test DNA. While this heterologous fragment should have activated strong MSUD, it was also expected to down-regulate MSUD due to the “silencing the silencer” effect. Indeed, only 13.4% of ascospores possessed the silenced phenotype in this case (while ∼75% were expected), suggesting that the observed effects were indeed caused by MSUD (Fig. 2F ).

Taken together, our results show that weak interspersed homologies (characterized by an overall sequence identity below 50%, e.g., patterns 4H-7N and 5H-6N) can be recognized by MSUD as pairable. These results also suggest that heterologous allelic sequences can escape MSUD, albeit less efficiently than homologous ones, with the maximal effect probably observed for matching sequence lengths.

Recombination-Independent Recognition of DNA Homology Directs the Expression of Small RNAs.

Previous studies identified a population of small RNAs (termed "MSUD-associated small interfering RNAs" or "masiRNAs") produced specifically from the unpaired DNA regions during MSUD (31). We investigated if the expression of these small RNAs was regulated by the recognition of interspersed homology. It was also important to test if the expression of masiRNAs is entirely controlled by the process of DNA homology recognition rather than by some unforeseen secondary effects, such as the disruption of local chromatin structure by transformed DNA.

We started by asking if our Rsp1500 reporter system is suitable for pursuing this question. We analyzed two conditions expected to yield robust small RNA expression profiles. The first one featured the deletion, while the second one featured “Lambda” DNA as an instance of random homology (Fig. 3A ). Each of the unpaired segments produced masiRNAs that mapped to both strands, and no spurious peaks were observed within the vicinity of the assayed regions (Fig. 3A and SI Appendix, Fig. S6B ). In general, the data demonstrate a good agreement between the genetic (ascospore eccentricity) and the molecular (masiRNA expression) readouts of MSUD using the Rsp1500 reporter system ( SI Appendix, Fig. S6 A and C ).

Fig. 3.

Recombination-independent recognition of DNA homology directs small RNA expression. (A) Small RNA profiles for two test conditions. The following crosses are analyzed: X12 (“Deletion”) and X20 (“Lambda”) ( SI Appendix, Table S3). The identifiable species of small RNAs are limited to those carrying 5′ phosphate and 3′ hydroxyl groups. Neighboring genes are indicated (light gray). (B) The six combinations of the allelic 1,500-bp segments. Female strains: RSP_Reporter1 (“Rsp1500”) and RSP_Reporter2 (“2H-1N”). Male strains: RSP_Test3 (“Rsp1500”), RSP_Test10 (“6H-5N”), and RSP_Test25 (“2H-1N”) ( SI Appendix, Table S2). (C) Small RNA profiles corresponding to the six combinations in B. The following crosses are analyzed: X13, X15, and X19 (Top, left to right) and X25, X26, and X27 (Bottom, left to right) ( SI Appendix, Table S3). (D) Eccentricity distribution (displayed as in Fig. 2D ) corresponding to the inverted perfect homology. Cross X49 ( SI Appendix, Table S3). (E) Eccentricity distribution (displayed as in Fig. 2D ) corresponding to inverted homology 6H-5N. Cross X50 ( SI Appendix, Table S3). (F) Small RNA profile (displayed as in Fig. 3A ) corresponding to the eccentricity distribution in E.

To perform a stringent test, one additional reporter strain (RSP_Reporter2) was created. Instead of Rsp1500, this strain carried 2H-1N (also adjacent to ECS). RSP_Reporter1 and RSP_Reporter2 were crossed to three test strains carrying either Rsp1500, 6H-5N, or 2H-1N. Thus, six combinations of the 1,500-bp segments were assayed (Fig. 3B ), including two instances of perfect homology (Rsp1500/Rsp1500 and 2H-1N/2H-1N), three instances of “no homology” (Rsp1500/2H-1N, 2H-1N/Rsp1500, and 2H-1N/6H-5N), and one instance of interspersed homology (Rsp1500/6H-5N). The term “no homology” stems from the fact that Rsp1500/2H-1N induces Rsp silencing at the level of random homology (e.g., “Rsp1500/Lambda”) (Fig. 2F ).

Abundant masiRNAs were expressed from all segments comprising the instances of “no homology” (Fig. 3C ). However, the same DNA constructs failed to express small RNAs above background in the context of perfect or interspersed homology (Fig. 3C ). These results demonstrate that the capacity to produce small RNAs is associated with the combined characteristics of a given pair of sequences rather than any individual sequence alone. Taken together, these results strongly suggest that masiRNA expression during MSUD is mediated entirely by DNA homology recognition.

Recognition of Homology between Rsp1500 and Test DNA Occurs within a Larger Genomic Context.

As reported previously, two identical DNA segments could be protected from MSUD even if they occupied somewhat different (non-colinear) positions on a pair of homologous chromosomes (32). It was important to determine if this concept also applies when interspersed homologies are involved. To this end, we tested if partially homologous DNA segments remain protected from MSUD when they are inverted with respect to each other. Pattern 6H-5N was chosen for this analysis because it is perceived as completely pairable with Rsp1500 in the direct (colinear) orientation (Figs. 2F and 3C ). As a control, we also examined the effect of inverting one of two Rsp1500 segments comprising perfect homology.

Strains carrying the inverted 6H-5N or Rsp1500 segments as Test DNAs were crossed to RSP_Reporter1. The non-colinear Rsp1500 segments are still recognized as fully pairable (Fig. 3D ), in agreement with the previous study (32). Surprisingly, inverting 6H-5N yields a large fraction of silenced ascospores (Fig. 3E ). This effect is supported by small RNA-seq data (Fig. 3F ). Nevertheless, the level of silencing induced by inverted 6H-5N is much lower than that induced by random homologies (53.6% versus∼75%, respectively), suggesting that the detection of this pattern as pairable is strongly impeded but not completely abolished in the non-co-linear orientation. Taken together, these results show that recognition of homology between the allelic 1,500-bp segments can be affected by their orientation with respect to each other and the surrounding genomic regions.

Interspersed Homologies with Periods of 21 or 22 bp Are Recognized as Pairable.

Thus far, our results show that interspersed homologies with the 11-bp periodicity are detected by MSUD as pairable (Fig. 2F ). To test if longer periodicities may also promote homology recognition, we created a series of 15 Test DNA sequences, each containing 6-bp units of homology arrayed with a fixed periodicity ranging from 17 to 31 bp (Fig. 4A , see pattern 6H-15N as an example).

Fig. 4.

Interspersed homologies with periods of 21 or 22 base pairs are recognized as pairable. (A) Interspersed homology patterns may differ with respect to the starting position of their homology frames. (B) Small RNA profiles (displayed as in Fig. 3A ) for the following crosses: X29 and X30 (Top, left to right) and X32 and X43 (Bottom, left to right) ( SI Appendix, Table S3). (C) Eccentricity distributions (displayed as in Fig. 2D ) for the following crosses: X29, X30, X31, X32, X33, and X34 (left to right) ( SI Appendix, Table S3). (D) Eccentricity distributions (displayed as in Fig. 2D ) for the following crosses: X43, X44, and X45 (left to right) ( SI Appendix, Table S3). (E) Eccentricity distributions (displayed as in Fig. 2D ) for the following crosses: X46, X47, and X48 (left to right) ( SI Appendix, Table S3).

We found that pattern 6H-15N (Period = 21) yields a relatively small (12.2%) number of silenced ascospores, implying that it was often detected as paired. Patterns 6H-13N (Period = 19), 6H-14N (Period = 20), and 6H-16N (Period = 22) induce much stronger silencing (62 to 68%) and are therefore only marginally better than random homology at evading MSUD. Patterns 6H-11N (Period = 17) and 6H-12N (Period = 18) produce ∼95% silenced ascospores (Fig. 4C and SI Appendix, Fig. S6A ). This outcome was unexpected given that such strong response is typical of the deletion rather than any of our previous homology tests, yet these results are fully supported by the corresponding small RNA profiles (Fig. 4B ). Lastly, patterns with period lengths equal to or greater than 23 (6H-17N–6H-25N) induced silencing at a level similar to random homology ( SI Appendix, Fig. S6A ).

To probe the ability of 6H-15N to escape MSUD, we engineered additional variants of 6H-15N, 6H-16N, and 6H-17N by changing the starting position of the 6-bp homology frames from 1 (default) to 10 (Fig. 4A ). These patterns were named 6H-15N_10, 6H-16N_10, and 6H-17N_10 (Fig. 4E ). We found that pattern 6H-15N_10 is perceived as less pairable than 6H-15N, whereas pattern 6H-16N_10 is perceived as more pairable than 6H-16N (Fig. 4 C and E ). Interestingly, in each series, the two less optimal patterns surrounding the optimal one (6H-14N versus 6H-16N or 6H-15N_10 versus 6H-17N_10) tend to produce similar responses (Fig. 4 C and E ).

We also explored if some additional periods, expected to promote sequence-specific plectonemic pairing of relaxed B-form DNAs (16), could promote homology recognition for MSUD. These patterns contained 13-bp units of homology arrayed with the periodicity of 37, 47.5 (alternating 47-bp and 48-bp heterology tracts), or 58 bp (Fig. 4D ). At first approximation, all of them are recognized as unpaired (Fig. 4D ).

Taken together, these results reveal a distinct ability of interspersed homologies with 21- to 22-bp periodicities to escape MSUD. However, this ability appears to be sequence or context dependent, potentially reflecting the dynamic nature of the underlying homology detection process.

An All-Atom Model of the Direct Homologous dsDNA–dsDNA Pairing for MSUD.

Our results show that MSUD is based on an efficient recombination-independent mechanism that can detect interspersed homology with an overall sequence identity below 40%. Thus, these results argue against the annealing of single complementary strands as the basis for homology recognition during MSUD and, instead, suggest that MSUD may involve direct pairing of dsDNAs.

According to the recently proposed model of direct homologous dsDNA–dsDNA pairing, one interaction unit contains three to four stacked quartets in the form of a short quadruplex (16). Each quartet is formed by two identical base pairs. Because of helical rotation constraints and concomitant structural deformations, two consecutive quadruplexes must be separated by more than one helical turn. For paranemic pairing, the spacing must be an integral number of helical turns, and it cannot be less than five (16). For right-handed plectonemic pairing, however, the minimal spacing distance 1) becomes shorter, 2) depends upon the writhe of the plectoneme, and 3) does not need to be an integer or half integer in terms of free B-DNA helical turns (16). Therefore, we investigated if the observed optimal periodicity (21 to 22 bp) could be compatible with such plectonemic pairing. To do so, we took a molecular dynamics (MD) approach, as explained below.

In principle, plectonemic structures were created as illustrated in SI Appendix, Fig. S7A . Specifically, two parallel B-DNA double helices were attached to a virtual frame that consisted of a scaffolding pole and several crossbeams. This frame had only a few degrees of freedom corresponding to rotation of the crossbeams around the pole. It was built from phantom atoms not involved in any interactions except for harmonic links to some backbone groups. The frame was used in MD simulations for applying appropriate twisting torques that coiled the two straight helices into a plectoneme ( SI Appendix, Fig. S7A ). The system was placed in a small water shell and neutralized by potassium ions (in a vacuum). The MD simulations were carried out using established methods (33 –35) that were previously applied to studying transitions between the A and B forms of DNA under low hydration (36 –38).

The actual model was built in several steps ( SI Appendix, Fig. S7B ). The initial conformation of the quadruplex structure was taken from one of the earlier MD trajectories (16). A complex with four stacked quartets was chosen and extended by canonical B-DNAs to the total length of 24 bp. Two such structures are shown in SI Appendix, Fig. S7B , Left, with orientations corresponding to the helical symmetry of a right-handed plectoneme. The lengths of the extended tails were 11 and 9 bp. In the final model, the shorter tails were ligated to obtain the spacing of 18 bp. This separation corresponds to interspersed homologies 3H-18N (Period = 21) or 4H-18N (Period = 22).

Next, the upper part was rotated and the red helices were ligated to create a structure shown in the second left panel ( SI Appendix, Fig. S7B ). This structure was then attached to a virtual frame and covered by a hydration shell ( SI Appendix, Fig. S7A ). Starting from this state, MD simulations were used to return the upper part back to its initial position. During this run, the red helix gradually assumed the shape corresponding to that in a right-handed plectoneme (second right panel in SI Appendix, Fig. S7B ). Once this step was complete, the free 8-bp ends of the blue helix were replaced by a continuous 16-bp fragment copied from the red helix. The accompanying structural perturbations and the rearranged hydration shell were re-equilibrated during several nanoseconds of MD simulations to produce the final structure ( SI Appendix, Fig. S7B , Right).

Starting from the equilibrated structure ( SI Appendix, Fig. S7B , Right), MD simulations were continued for several nanoseconds without any restraints, except having the DNA ends attached to the virtual frame to simulate the condition of long DNAs. The coordinates obtained during the last nanosecond were used for averaging (Fig. 5A ). During this control run, DNA conformations fluctuated around the original state without significant local perturbations or general structural tendencies, suggesting that the paired structure is stable (Fig. 5B ). The movie corresponding to the complete control run is provided as SI Data File 9 (39).

Fig. 5.

Direct dsDNA–dsDNA pairing as the basis for DNA homology recognition during MSUD. (A) Predicted structure of the paired complex with two DNA double helices of 48 bp. Two orthogonal views of an average conformation sampled during 1-ns all-atom MD run with a shell of water and potassium ions. The distance between the 4-bp quadruplexes corresponds to a period of 22 bp. The bonding of base pairs in quadruplexes is illustrated for one quartet (not in scale). (B) Time traces of two geometric parameters that characterize the overall stability of the complex during the 1-ns MD run. (C) The principal helical parameters of the two dsDNAs in the paired complex. MD averages are plotted, with error bars showing time fluctuations. The available experimental data (based on X-ray fiber diffraction) are shown by horizontal lines. The regions of quadruplex pairing are shaded in light gray. (D) During MSUD, unpaired DNAs may be recognized as being surrounded by the transiently paired regions.

Importantly, some helical parameters of the paired complex differ from the canonical B-DNA values (Fig. 5C ). The most critical is the structural shift of the unpaired double-stranded segments toward the highly twisted C-form DNA found in some crystal fibers (40, 41). The uncovered B- to C-form transition is likely induced by the plectonemic coiling in combination with locally reduced hydration and increased ion concentrations. Overall, the properties of the all-atom model (Fig. 5) indicate that the quadruplex-based pairing of dsDNAs is, in principle, compatible with the observed optimal periodicities of 21 to 22 bp.

Discussion

Standard mechanisms of DNA homology recognition, including recombination (42), CRISPR/Cas9 (43), and DNA interference (44), involve single-stranded DNA or RNA intermediates and probe dsDNA targets based on the WC bonding principle. Thus, if recombination-mediated homology recognition needs to take place between two dsDNA molecules, one of them must be broken to create such an intermediate (45). Yet, paradoxically, homologous chromosomes can still pair in the absence of DNA breakage in many organisms, including Drosophila melanogaster (46, 47), C. elegans (48), mice (49, 50), and fungi (2, 51). In all of these situations, the specificity basis remains largely unknown.

The strongest evidence for a mechanism that directly matches intact dsDNAs has been provided by RIP in N. crassa, which can detect partial homologies with an overall sequence similarity of only 25 to 27%, provided that such homologies contain short units of sequence identity interspersed with a periodicity of 11 or 12 bp (15). RIP occurs normally in the absence of MEI-3, the only RecA-like recombinase in N. crassa (15), setting itself further apart from the canonical recombination-mediated processes.

In this study, we investigated the molecular requirements of MSUD in N. crassa. During MSUD, any gene-sized chromosomal segment lacking a homologous allelic partner induces transient RNA interference in early meiosis (10, 11). We began by assaying the dependence of MSUD on the canonical mechanism of meiotic recombination, which requires SPO11 to cut dsDNAs and MEI-3 to catalyze the homology search (17). We found that both factors are completely dispensable for MSUD. We also found that SPO11-independent recombination (20) is not a relevant factor in our experimental system. This result was unexpected because in N. crassa, SPO11 is needed for robust chromosome synapsis (20). On the other hand, our findings reinforce the idea that even when homolog synapsis depends on SPO11 (e.g., in budding yeast), chromosomes may interact transiently by a recombination-independent mechanism prior to that (51). Here, our findings specifically suggest that such transient interactions involve direct dsDNA–dsDNA pairing.

MSUD is expected to rely on interactions between homologous chromosomes rather than sister chromatids. We find that REC8, a universally conserved meiotic kleisin that promotes interchromosomal recombination, enhances MSUD. REC8 is normally loaded during premeiotic DNA replication (23), which takes place before karyogamy in N. crassa and related fungi (52). MSUD becomes somewhat retarded in the absence of REC8, but it eventually occurs in many meiotic nuclei. The recombination-independent role for REC8 in MSUD may be related to its role in promoting interchromosomal interactions during meiosis. Of note, a role of REC8 in enhancing recombination-independent meiotic pairing in Schizosaccharomyces pombe was reported previously (reviewed in ref. 23).

A possibility of MSUD being independent from meiotic recombination was discussed earlier (29), yet its basis remained mysterious, in part because the previously revealed properties of MSUD appeared compatible with a strand-annealing mechanism (29). Specifically, it was found that allelic sequences with only 6% divergence already triggered MSUD (29). Given the homology requirements of RIP (15), we investigated if MSUD could also recognize interspersed homologies with sequence divergence greater than 50% as pairable. We started by creating a sensitive genetic assay to accurately measure the response of MSUD to interactions between two 1,500-bp allelic segments. The sequence of one (reporter) segment was always the same, whereas the other (test) segment was varied as desired. A similar approach was used earlier in our studies of RIP (15).

Our current results indicate that MSUD and RIP likely share the same basis of DNA homology search and recognition. Both processes detect short tracts of sequence identity, interspersed with the 11-bp periodicity, as pairable. Like in RIP, the capacity of MSUD to sense such interspersed homologies can be affected by the adjacent regions of perfect homology. Some pairable homologies in our assays had an overall identity of only 36.4%, thus arguing against the role of single-strand annealing as the basis of sequence recognition. On the other hand, the assayed interspersed homologies did not fully mimic perfect homology (e.g., in the colinearity test, Fig. 3), pointing to the existence of additional requirements. For example, homology search may proceed in several steps (e.g., coarse pairing followed by fine pairing) that are fully accommodated by perfect homology but not by the tested interspersed homologies.

By systematically exploring interspersed homologies with periods ranging between 17 and 31, we found that periods of 21 to 22 bp were especially effective at evading MSUD. In our system, the optimal periodicity value depends on which frame of homologous units is chosen. We used the 22-bp periodicity as a constraint in all-atom modeling of plectonemic DNA structures linked by interspersed quadruplexes (16). Our effort yielded a paired dsDNA complex that is consistent with standard chemical geometry and that does not interfere with the existing intrahelical WC bonding (according to empirical potentials currently used in MD simulations of DNA). The complex contains two 4-bp quadruplex nodes connected by plectonemically coiled dsDNA segments (Fig. 5A ). The conformation of the “free” segments is similar to that of C-DNA. The strongly twisted double helix of C-DNA derives from B-DNA by increasing the fraction of phosphate groups switched from the canonical BI to an alternative BII backbone conformation (53). The minor groove of C-DNA is deep and narrow, and the major groove of C-DNA is wide and relatively flat. Thus, the quadruplex-based pairing of homologous C-DNAs can be very fast, potentially satisfying the speed requirements of both RIP and MSUD. The B- to C-DNA transition likely depends on ion concentrations and molecular crowding in the immediate vicinity of the participating DNAs.

Our results suggest that homologous chromosomes engage in transient recombination-independent pairing in early prophase I. Such pairing may feature sequence-specific quadruplex-based contacts along with nonspecific interactions. The ability of the heterologous allelic segments to escape MSUD at an appreciable rate suggests that nonspecific interactions may play an important role. Overall, if the homologous chromosomal regions do not contain a large (<1 kbp) section of heterology, they will associate more or less continuously; however, if such heterology is present, it will remain largely unpaired and will be recognized as such in the context of the paired flanks (Fig. 5D ).

The proposed pairing mechanism likely relies on many additional factors, notably those involved in chromatin remodeling. One such factor, a RAD54-like helicase SAD-6, was already implicated in MSUD (32). In yeast, a large fraction of histones can be removed by a proteasome-mediated pathway in response to DNA damage (54), and proteasome activities are also required for meiotic homologous pairing and recombination (55, 56). It is possible that processes normally associated with DNA repair and recombination may also be involved in regulating recombination-independent homologous dsDNA–dsDNA pairing.

Materials and Methods

Interactions between allelic DNA segments in early prophase I of meiosis in N. crassa were assayed by three approaches. The first approach relied on observing the expression of fluorescent reporters in meiotic nuclei. The second approach required measuring the eccentricity of ascospores. The third approach was to detect (by high-throughput sequencing) masiRNAs produced from the analyzed DNA regions. These results were combined with MD simulations to build a consistent all-atom model of homologous dsDNA–dsDNA pairing. A full description of the materials and methods is provided in SI Appendix .

Data Availability

All data files are available as TAR/GZIP archives at https://figshare.com/s/6669fb33f3b249a41b25 (30, 39). All other study data are included in the article and/or supporting information.