CENP-A binding domains and recombination patterns in horse spermatocytes

Centromeres exert an inhibitory effect on meiotic recombination, but the possible contribution of satellite DNA to this “centromere effect” is under debate. In the horse, satellite DNA is present at all centromeres with the exception of the one from chromosome 11. This organization of centromeres allowed us to investigate the role of satellite DNA on recombination suppression in horse spermatocytes at the stage of pachytene. To this aim we analysed the distribution of the MLH1 protein, marker of recombination foci, relative to CENP-A, marker of centromeric function. We demonstrated that the satellite-less centromere of chromosome 11 causes crossover suppression, similarly to satellite-based centromeres. These results suggest that the centromere effect does not depend on satellite DNA. During this analysis, we observed a peculiar phenomenon: while, as expected, the centromere of the majority of meiotic bivalent chromosomes was labelled with a single immunofluorescence centromeric signal, double-spotted or extended signals were also detected. Their number varied from 0 to 7 in different cells. This observation can be explained by positional variation of the centromeric domain on the two homologs and/or misalignment of pericentromeric satellite DNA arrays during homolog pairing confirming the great plasticity of equine centromeres.

may disrupt pericentric sister chromatid cohesion, strongly affecting kinetochore functionality 15,24 . Since centromeres are known to reside in a heterochromatic environment, a contribution of heterochromatin in this crossover suppression has been proposed 15,24 . Mammalian centromeres are embedded in wide heterochromatic domains which are typically associated to satellite DNA, but it is still unclear whether the presence of repetitive DNA contributes to the "centromere effect" 24 .
To answer this question, we took advantage of the Equus caballus (domestic horse) model system 25 . In this species, the centromere of chromosome 11 (ECA 11) is devoid of satellite sequences while all the other centromeres are satellite-based 26,27 . This peculiar centromere, similarly to several satellite-less centromeres in other species of the genus Equus, arose recently during evolution as a result of repositioning [28][29][30] which is the movement of centromeric function along the chromosome without marker order variation 31 . Besides centromere movement, the genomes of equids are characterized by an exceptional plasticity regarding retrotransposition 32 and insertion of nuclear mitochondrial DNA fragments 33 .
We previously isolated and characterized the major horse centromeric satellite DNA family, 37cen 34,35 . The satellite-less centromere of ECA 11 was the first one to be identified and described at the molecular level 26 . Recent studies showed that the position of ECA 11 centromere is not fixed but, in different individuals, it slides within a 500 kb window giving rise to positional alleles 36,37 . These "epialleles" are inherited as Mendelian traits, but their position can slide in one generation being stable during mitotic propagation of cultured cells 37 . The "centromere sliding" phenomenon thus confirms the epigenetic nature of centromeres and proves that centromeric domains are characterized by positional instability 36 .
In the present work we mapped recombination foci along horse chromosome 11 and found that, although its centromere is devoid of satellite repeats, it exerts an inhibitory effect on recombination. We then discovered that double well separated CENP-A binding domains were present at several pachytene bivalents and proposed a model to explain their formation.

Results
Distribution of MLH1 foci in male horse meiosis. The distribution of MLH1 foci, a marker of meiotic crossovers, was investigated at pachytene stage by immunofluorescence. We used an anti-MLH1 antibody to label recombination sites, an anti-SCP3 antibody, for the immunostaining of the synaptonemal complex (SC), and a CREST serum, to label all centromeres. Pachytene cells from three different horses were analysed. In Table 1 the mean number of MLH1 foci per cell and the mean number of foci per bivalent are reported. The horse karyotype is characterized by a diploid number of 31 autosomal bivalents, 13 of which metacentric. The total number of autosomal MLH1 foci per cell ranged from 36 to 54 in horse TE, from 38 to 51 in horse PV and from 42 to 63 in horse MP, with an overall mean frequency of autosomal MLH1 foci of 45.30 ± 4.90 per cell. In the 13 metacentric bivalents we detected an average number of 24.12 ± 3.04 foci per cell, with 1.86 ± 0.23 foci per chromosome, while in the 18 acrocentric chromosomes the mean frequency of MLH1 foci per cell was 21.18 ± 2.78, with an average of 1.18 ± 0.15 foci per bivalent (Table 1). These results are in agreement with the previously described frequency of MLH1 foci in horse male meiosis 38 highlighting the requirement of an "obligatory CO" per bivalent to ensure chromosomal disjunction. MLH1 foci on the XY body were not taken into account, due to their peculiar organization and meiotic behaviour.
To evaluate whether a satellite-less centromere suppresses meiotic recombination at the same level as a satellite-based one, we compared the distance between each MLH1 focus and the centromere on ECA 11 and on all other chromosomes. To this end, we combined immunostaining to localize MLH1 foci and centromeres with FISH using the 37cen satellite DNA as probe. We previously demonstrated that this probe labels all horse centromeres except the ones of ECA 11 and ECA 2 27 . In Fig. 1a, MLH1 foci (green) and centromeres (red) in a pachytene spread are shown and the identification of ECA 11 in the same spread is shown in Fig. 1b. This experimental workflow allowed us to easily identify ECA 11 as the smaller chromosome negative to 37cen hybridization (Fig. 1b). It is important to point out that the centromere of ECA 11 is the only one totally lacking any satellite repeats while, at the centromere of chromosome 2, satellite repeats other that 37cen are present 26,27 . The distance between recombination foci on ECA 11 and the centromere was measured in 42 cells of horse TE, where a total number of 69 MLH1 signals were detected (red line in Fig. 1c). Since the distribution of crossovers is influenced by total chromosomal length 38 , in each one of the 42 spreads, metacentric bivalents with a meiotic length comparable to ECA 11 (length of ECA 11 ± 0.5 cm) were identified as control for a total of 67 synaptonemal complexes and 104 MLH1 foci (blue line in Fig. 1c). This analysis allowed us to compare the distribution of MLH1 foci on ECA 11 and on chromosomes with comparable meiotic length but with satellite-based centromeres. The distance between MLH1 foci and the centromere was also measured on ECA 2, that is the longer chromosome negative to 37cen hybridization www.nature.com/scientificreports www.nature.com/scientificreports/ (grey line in Fig. 1c). The same measurement was performed on all other metacentrics (orange line in Fig. 1c) and on acrocentric chromosomes (cyan line in Fig. 1c). MLH1 foci on the p and q arms of metacentrics were measured separately.
No MLH1 foci were detected around the centromere region of all bivalents. In particular, the distance between the centromere and MLH1 foci on all metacentrics, including ECA 11, follows a bimodal distribution with a marked depression around the centromere. The distribution of MLH1 foci on the acrocentrics is similar to the one on the q arms of metacentrics with a drop in the frequency of crossovers near the centromere. As expected, the maximum distance covered by the ECA 11 curve (red) and by the curve of the small metacentric chromosomes used as control (blue) is shorter than the one covered by the curves obtained from all other chromosomes. These results show that the centromere effect affects all horse chromosomes independently of the presence of satellite DNA at their centromeres. www.nature.com/scientificreports www.nature.com/scientificreports/

Identification of double and stretched CENP-A signals in horse bivalents.
During the analysis of recombination foci in spermatocytes at the pachytene phase of meiosis we observed a peculiar phenomenon: while, as expected, the centromere of the majority of bivalents was labelled with a single CREST signal, double-spotted centromeres were also detected on several bivalents (Fig. 2a). These signals were never observed on XY bodies.
Besides double signals composed by two well separated dots and canonical single-dotted signals, we could observe "stretched" signals likely deriving from two spots too close to be resolved separately (Fig. 2a). From now on, double and stretched signals will be called DSS (Double and Stretched Signals).
Since the CREST serum may recognize centromeric proteins other than CENP-A, we tested whether DSS are due to the presence of two separated CENP-A domains or to other peri-centromeric proteins that may not colocalize with CENP-A. To this goal we produced an anti-CENP-A serum in sheep immunized with the horse protein. The specificity of this new antiserum was tested by western blotting and immunofluorescence (see Supplementary Fig. S1). We then performed immunofluorescence experiments on horse pachytene spreads from horse TE either with a CREST serum or with the anti-CENP-A serum. Bivalents with DSS were identified in both CREST (Fig. 2a) and CENP-A experiments (Fig. 2b) with the same frequency (Fig. 3a), indicating that they are CENP-A binding domains. In addition, we performed a triple immunofluorescence experiment with the anti-SCP3 antibody, the CREST and the anti CENP-A sera on pachytene spreads. As shown in Supplementary  Fig. S2, CREST and CENP-A signals colocalize on all centromeres, including those with DSS signals. These results demonstrate that DSSs correspond to CENP-A binding domains.
The pattern of DSS signals suggests that they correspond to centromeric domains located in different positions on the two homologs. If this hypothesis was correct, the intensity of each one of the double spots should be about half the intensity of single spots. In particular, the intensity of each spot in double-spotted signals should www.nature.com/scientificreports www.nature.com/scientificreports/ be comparable to signal intensities on chromosomes X and Y, where the two centromeric regions are not paired. We quantified the intensity of all CENP-A signals on 20 pachytene spreads (Fig. 3b). Indeed, the intensity of each dot in double-spotted signals is comparable to the one on X and Y and about half the one on single-spotted centromeres (Fig. 3b). The intensity of "stretched" signals is similar to the one of canonical single-spotted centromeres (Fig. 3b).
Interestingly, no DSSs were observed on ECA 11. To characterize the centromeric domains of ECA 11 in horse TE, a ChIP-seq experiment was carried out with the CREST serum and two adjacent peaks were detected (Fig. 3c). As previously demonstrated 36,37 the two peaks correspond to the centromeric regions on the two homologs. The proximity of the two peaks in horse TE strongly suggests that they are too close to be resolved as DSSs on meiotic bivalents. The same ChIP-seq characterization of horse centromeres other than the one of ECA 11 is not possible since they contain satellite DNA sequences, which are not assembled in the horse reference genome due to their repetitive nature 35,37 .
DSSs were not observed on ECA 2, whose centromere lacks 37cen repeats but contains other satellite repeat families 27 . To further investigate intra-and inter-individual variability of the DSS centromeres, we performed the same immunofluorescence experiment using the anti-CENP-A antibody on two additional horses, scoring the number of DSS centromeres per cell. From each individual 25 cells were evaluated. Intra-individual variability was high in all the three individuals, with the number of DSS centromeres ranging from 0 to 7 (Fig. 3d) whereas the fraction of cells with at least one DSS did not vary among individuals. The average number of DSSs per cell was 2.0 ± 1.5 in horse TE, 2.4 ± 1.8 in horse LL and 2.0 ± 1.4 in horse KA. Using the Kruskal-Wallis test, we showed that inter-individual variability was not statistically significant (p value 0.8146).

Discussion
In this work, we describe the chromosomal distribution of recombination events in male horse meiosis. A great inter-cellular variability in the number of crossovers, which ranged from 36 to 63, was observed, in agreement with previous findings in other mammalian species [39][40][41][42][43] .
We took advantage of the horse model system to test whether the centromere effect on recombination suppression is related to the presence of satellite DNA. To this goal we mapped recombination foci on ECA 11 through the cytogenetic localization of the MLH1 protein along the synaptonemal complex and detected a crossover suppression around the satellite-less ECA 11 centromere, demonstrating that the centromere effect depends on the centromeric function itself rather than on satellite DNA sequences and supporting the hypothesis that recombination suppression at pericentromeres is not related to DNA sequence but rather to the epigenetic environment. It has been proposed that the occurrence of crossovers near centromeres is selectively disadvantageous because it may cause premature sister chromatid separation leading to non-disjunction events at the second meiotic division 24 . It is interesting to mention our recent observation that also mitotic segregation is not affected by the absence of centromeric satellite repeats on horse chromosome 11 44 .
The formation of chromosome bivalents at pachytene stage is accompanied by the pairing of centromeres, as reported for all eukaryotic species studied so far 45,46 . During the analysis of horse meiosis, we observed a peculiar phenomenon: centromeres with double or extended CENP-A signals (DSS) were frequently observed. The morphology of these unusual signals ranged from well separated double spots to long "stretched" signals, which were interpreted as double CENP-A binding domains too close to be resolved separately.
Previous reports on mammalian species, such as the common shrew and the dwarf hamster, detected the presence of double-spotted centromeres 47,48 , although their presence and frequency remained unexplained. In our horse system we identified DSS centromeres at a surprising high frequency on different chromosome bivalents, showing a high inter-cellular variability. In the light of our results, we hypothesized that, at DSS centromeres, the CENP-A binding domains of the two homologous chromosomes are localized on different regions, giving rise to double spots or to stretched signals, depending on the distance between them. This interpretation is strongly supported by signal intensity comparisons in single versus DSS centromeres and the XY body.
The number of DSS centromeres per cell was highly variable (from one to seven) also among cells of the same individual. No statistically significant inter-individual variation could be observed.
The observation of these peculiar centromeres raises the question whether they have any biological meaning. Taking advantage of the satellite-less centromeres of the genus Equus, we previously demonstrated that the position of the centromere is not fixed but slides, giving rise to different positional alleles, defined "epialleles", which are inherited as Mendelian traits 36,37 . Similar polymorphism regarding the position of the CENP-A binding domains was reported also in some human satellite-based centromere, such as the one of HSA 17 49 . Indeed, in HSA 17, the centromere can assemble on different alpha satellite arrays and individuals with heterozygosity in the position of the centromere were reported. A similar characterization is not possible for the horse satellite-based centromeres since their sequences are lacking chromosomal assignment in the horse genome.
It is known that repeat copy number of satellite DNA arrays can vary, even between homologous chromosomes [50][51][52] . Therefore, although satellite-less centromeres can only display epigenetic positional variation due to centromere sliding, satellite-based centromeres can also display DNA sequence variation due to variable numbers of tandem repeats. We recently demonstrated that the repeated units of the major horse centromeric satellite DNA, 37cen, are homogeneous at the sequence level 35 . We now propose that the presence of numerous conserved monomers may lead to homology-based pairing between different regions of the extensive array, thus, causing a "staggered" alignment, here called "misalignment". The presence of polymorphism among homologous chromosomes regarding the position of CENP-A binding domains and the number of tandem repeats suggests a possible interpretation for the occurrence of DSS centromeres. As depicted in the model presented in Fig. 4, we propose that DSSs are the result of two independent components: epiallelism for the position of the CENP-A binding domains and misalignment events between centromeric and pericentromeric satellite DNA arrays of the homologous chromosomes during pairing. This misalignment may increase the physical distance between centromeric domains which may be already in different positions on the two homologs due to CENP-A binding domain sliding. In this scenario, our DSS centromeres would become visible when the distance between the centromeric domains on the two homologous chromosomes is long enough to be resolved by our method (Fig. 4b,c). While epialleles are conserved from cell to cell and inherited as Mendelian traits, the degree of misalignment varies from meiosis to meiosis, resulting in the great variability of the number and distribution of DSSs. Therefore, since we observed that the number of DSSs is variable among pachytene spreads of the same individual and misalignment may affect all satellite-based centromeres, it is likely that all chromosomes could potentially carry a DSS.
Interestingly, DSSs were not identified on ECA 2, whose centromere is the only satellite-based one lacking the 37cen satellite. Although more data would be necessary to test this hypothesis, we propose that the 37cen satellite may be prone to misalignment because its repeated arrays are particularly conserved 35 . The model is also supported by the lack of DSSs on ECA 11 bivalents in horse TE. In this individual, the centromeric domains on the two homologs are very close (Fig. 3c) and no satellite repeat misalignment can obviously occur on this satellite-less chromosome. We cannot exclude that, in individuals in which the two ECA 11 centromeric domains are located on sufficiently distant regions, DSSs may be observed also on satellite-less centromeres.
The proposed model may also explain the observation of a double-spotted centromere in the common shrew and in the dwarf hamster 47,48 . In our system we identified DSS centromeres at a surprising high frequency on different chromosome bivalents, showing high inter-cellular variability. Overall, our findings suggest that the combination of centromere sliding and misalignment of satellite arrays may occur at high frequency in the horse, in agreement with the exceptional centromere plasticity of the Equus species.

Methods
Testis collection and treatment. Testicular samples from five horses (TE, MP, PV, LL and KA) were obtained by certified veterinarians following castration procedures under general anaesthesia. The castrations were not carried out for our research but were performed as routine management of riding horses. Testicular samples from the five horses were given to us instead of being discarded. All methods were carried out in accordance with relevant guidelines and regulations.
Testes were cut in small pieces (about 1 cm 3 ) using sterile scalpel blades and frozen at −80 °C until use.
Anti-CENP-A serum preparation. For antibody preparation, an E. coli codon optimized version of horse CENP-A (ENSECAP00000013849) was synthesized (Eurofins Genomics) and cloned into pDEST17 for expression of an N-terminally 6-his tagged CENP-A protein in E. coli BL21-AI. Inclusion bodies were purified by differential centrifugation, solubilized in 7 M guanidinium-HCl and protein was purified by affinity chromatography on Ni-NTA agarose in 7 M Urea (ThermoFisher). Purified protein was dialyzed against phosphate-buffered saline (PBS) and used as immunogen to raise an antibody in sheep.
Pachytene spread preparation and immunofluorescence. Pachytene spreads were prepared from frozen testis samples as previously described 53,54 with minor modifications to adapt the protocol to this horse tissue. Immunofluorescence experiments were performed with the following antibodies: anti-SCP3 antibody (Abcam ab15093), anti-CENP-A sheep serum, CREST serum (kindly provided by Dr. Claudia Alpini, Fondazione I.R.C.C.S. Policlinico San Matteo, Pavia, Italy) and anti-MLH1 antibody (BD Pharmingen, 551091). Fixation with 4% paraformaldehyde (pH 10) in 1x PBS, 0.015% TritonX-100 was used for the preparation of slides for immunofluorescence with the CREST serum. Fixation with 1% formaldehyde, 0.015% TritonX-100 (pH 9.8) was used for the preparation of slides for immunofluorescence with the anti-CENP-A antibody and for sequential immunofluorescence with the anti-CENP-A and CREST sera. The sequential protocol is not optimal for both CREST and anti-CENP-A sera. This is the reason of the sub-optimal immunostaining of centromeres obtained with the combined immunofluorescence. Slides were permeabilized in 0.05% Tween-20 in PBS. Rhodamine anti-rabbit, Alexa488 anti-sheep, Alexa488 or Alexa647 anti-human and Alexa488 anti-mouse secondary antibodies were used. Pachytene chromosomes were counterstained with DAPI (0.2 μg/ml) and mounted with Fluorescence Mounting Medium (Dako). Image acquisition, measurement and statistical analysis. Digital images from fluorescence signals were acquired with a fluorescence microscope (Zeiss Axioplan) equipped with a cooled CCD camera (Photometrics). Pseudo-colouring and merging of images were performed using the IPLab Imaging Software.