Islands of retroelements are the major components of Drosophila centromeres

Centromeres are essential chromosomal regions that mediate kinetochore assembly and spindle attachments during cell division. Despite their functional conservation, centromeres are amongst the most rapidly evolving genomic regions and can shape karyotype evolution and speciation across taxa. Although significant progress has been made in identifying centromere-associated proteins, the highly repetitive centromeres of metazoans have been refractory to DNA sequencing and assembly, leaving large gaps in our understanding of their functional organization and evolution. Here, we identify the sequence composition and organization of the centromeres of Drosophila melanogaster by combining long-read sequencing, chromatin immunoprecipitation for the centromeric histone CENP-A, and high-resolution chromatin fiber imaging. Contrary to previous models that heralded satellite repeats as the major functional components, we demonstrate that functional centromeres form on islands of complex DNA sequences enriched in retroelements that are flanked by large arrays of satellite repeats. Each centromere displays distinct size and arrangement of its DNA elements but is similar in composition overall. We discover that a specific retroelement, G2/Jockey-3, is the most highly enriched sequence in CENP-A chromatin and is the only element shared among all centromeres. G2/Jockey-3 is also associated with CENP-A in the sister species Drosophila simulans, revealing an unexpected conservation despite the reported turnover of centromeric satellite DNA. Our work reveals the DNA sequence identity of the active centromeres of a premier model organism and implicates retroelements as conserved features of centromeric DNA.


Introduction
Centromeres are marked by the histone H3 variant, CENP-A (also called Cid in Drosophila), which is necessary and sufficient for kinetochore activity [1,2]. Although epigenetic mechanisms play a major role in centromere identity and propagation [3], centromeric DNA sequences can initiate centromere assembly in fission yeast [4] and humans [5], and centromeric transcripts play a role in centromere propagation in human cells [6], suggesting that centromeric DNAencoded properties may contribute to centromere specification [7]. However, our current understanding of most centromeres remains at the cytological level, as metazoan centromeres are embedded in highly repetitive, satellite-rich pericentric heterochromatin and thus are largely missing from even the most complete genome assemblies. Only recently, long-read single molecule sequencing technologies made it possible to obtain linear assemblies of highly repetitive parts of multicellular genomes such as the human Y chromosome centromere [8] and maize centromere 10 [9].
Drosophila melanogaster provides an ideal model to investigate centromere genomic organization as it has a relatively small genome (~180 Mb), organized in just three autosomes (Chr2, Chr3, and Chr4) and two sex chromosomes (X and Y) [10]. The estimated centromere sizes in Drosophila cultured cells range between ~200-500 kb [11] and map to regions within large blocks of tandem repeats [12][13][14][15]. While CENP-A associates with simple satellites in ChIPseq data [16], it may bind to additional undiscovered sequences. The linear organization at the sequence level of any of the centromeres is unknown in this species. Early efforts to determine the structural organization of centromeres in D. melanogaster combined deletion analyses and sequencing of an X-derived minichromosome, Dp1187. These studies mapped the minimal DNA sequences sufficient for centromere function to a 420-kb region containing the AAGAG and AATAT satellites interspersed with "islands" of complex sequences [14,15]. However, it is unclear which parts of this minimal region comprise the active centromere, whether or not it corresponds the native X chromosome centromere, and if other centromeres have a similar organization. By and large, satellites have been regarded as the major structural elements of Drosophila, humans, and mouse centromeres [2,3,17].
In this study, we reveal the detailed organization of all functional centromeres in D. melanogaster. By mapping CENP-A on single chromatin fibers at high-resolution, we discover that CENP-A primarily occupies islands of complex DNA enriched in retroelements, which are flanked by large blocks of simple satellites. Our genomic analyses show that all centromeres have a unique sequence organization, even though many of the centromeric elements are shared among them. In particular, all centromeres are enriched for a non-LTR retroelement in the Jockey family, G2/Jockey-3. While none of these elements are specific to centromeres, they are significantly enriched within these regions. We also find G2/Jockey-3 enriched at the centromeres of D. simulans, which has centromeric satellite arrays highly divergent from those of D. melanogaster [16]. Collectively, these data are consistent with the model that retroelements may have a conserved role in centromere specification and function, as proposed for other species (for review see [18]).

Identification of candidate centromeres by long-read sequencing and ChIP-seq
To identify the centromeric DNA sequences of D. melanogaster, we combined a long-read genome assembly approach [19] with four replicate CENP-A chromatin immunoprecipitations (ChIP) on chromatin from Drosophila embryos, followed by paired-end Illumina sequencing (ChIP-seq). We also performed ChIP-seq in Drosophila melanogaster Schneider (S2) cells, a widely used model for cell division studies. We took four complementary approaches to discover regions of the genome enriched for CENP-A: 1) identifying simple repeats enriched for CENP-A based on kmers; 2) mapping reads to a comprehensive repeat library to summarize enriched transposable elements (TEs) and complex repeats; 3) using de novo assembly methods to assemble contigs from the ChIP reads and calculating enrichment relative to input post hoc; and 4) mapping reads to a heterochromatin-enriched assembly and calling ChIP peaks (Fig. 1A).
In our ChIP experiments, CENP-A pulls down simple satellites, consistent with a previous study [16]. Among the kmers most enriched in CENP-A ChIP relative to input are the dodeca satellite and its variants, and complex kmers that include tandem (AATAG) n and (AATAT) n repeats ( Fig.  1B; Fig. S1; Table S1). Prodsat (also known as the 10bp satellite) is enriched in the CENP-A ChIP, but not relative to input (Fig. 1B). In addition to satellites, we found that CENP-A is also strongly associated with retroelements, particularly non-LTR LINE-like elements in the Jockey family and with the Ribosomal Intergenic Spacer (IGS). Among the Jockey elements the most highly enriched in CENP-A ChIPs are annotated as G2 and Jockey-3 ( Fig. 1C; Table S2). However, our phylogenetic analysis suggests that these repeats correspond to the same type of element, as genomic copies of G2 and Jockey-3 are interleaved across the tree and not monophyletic (Fig. S2). Thus, we hereafter collectively refer to these elements as G2/Jockey-3.
To detect CENP-A-enriched sequences independently of known repeats in repeat libraries or of genome assemblies, we de novo assembled CENP-A ChIP reads into contigs (i.e. ChIPtigs [20]) and calculated their CENP-A enrichments. The resulting CENP-A enriched ChIPtigs primarily contained fragments of TEs and other complex repeats, and some simple satellite repeats (Table S3).
To determine the genomic location of CENP-A enriched sequences, we mapped ChIP reads to a new reference genome assembly that we generated using a heterochromatin-enriched assembly method resulting in greater representation of heterochromatin-associated regions [19] ( Table S4, and supplemental results). Five contigs were consistently the most CENP-A enriched in the assembly, with highly reproducible ChIP peaks across technical and biological replicates (IDR < 0.05; Table S5; Fig. S3). These CENP-A-enriched contigs have a similar organization: they contain islands of complex DNA (e.g. TEs) flanked by simple tandem satellite repeats with known centromeric locations ( Fig. 2; Fig. S4; Table 1). The candidate centromeric contig for the X chromosome (Contig79) is 70-kb and contains a 44-kb island of complex DNA (called Maupiti [15]), flanked by a short stretch of AAGAT satellite on one side and embedded in AAGAG satellite ( Fig. 2A). This region has an organization that is nearly identical to that of the Dp1187 minichromosome putative centromere [14,15], suggesting that this contig may contain at least part of the endogenous X centromere. The candidate centromeric contig for chromosome 4 (Contig119) contains a 42.8-kb island (we named Lampedusa) flanked by the AAGAT satellite (Fig. 2B). A recent study mapped the AAGAT satellite cytologically to centromeres of chromosome 4 and a B chromosome derived from this chromosome [21], consistent with Lampedusa being a candidate for centromere 4. The candidate centromeric contig for chromosome Y (Y_Contig26) consists of a 138-kb island (we named Lipari; Fig. 2C). The candidate centromeric contig for chromosome 3 (Contig 3R_5) contains a 68.5-kb island (we named Giglio) flanked by Prodsat and the dodeca satellite, which map to this centromere cytologically [12,22,23] (Fig. 2D). Finally, the candidate contig for chromosome 2 (tig00057289) contains a small 1.8-kb complex island (we named Capri), flanked by the AATAG and AAGAG satellites (Fig. 2E). The majority of the top enriched de novo ChIPtigs (88/100 for R1; 19/30 for R2; 26/30 for R3; and 82/100 for R4) map uniquely to these five contigs (Table  S3), providing independent support for the assembly and further substantiating our hypothesis that these contigs correspond to the centromeres.

Genomic distribution of CENP-A in embryos and S2 cells
Our ChIP-seq experiments and their analyses provide evidence that CENP-A is specifically associated with the island DNA sequences for Contig79 (X Maupiti ), Contig119 (4 Lampedusa ), Y_Contig26 (Y Lipari ), 3R_5 (3 Giglio ), and with a single interspersed G2/Jockey-3 fragment within tig00057289 (2 Capri ; Fig. 2; Fig. S4). A previous study that used a D. melanogaster native ChIPseq dataset (using anti-GFP antibodies and CENP-A-GFP expressing embryos) focused exclusively on the quantification of simple repeats and did not identify any complex DNA associated with CENP-A [16]. However, our re-analysis of this dataset showed association of CENP-A-GFP with the centromere islands ( Fig. S3B; Tables S4-S6). We validated individual elements for which we could design contig-specific qPCR primers in additional independent CENP-A ChIP experiments and confirmed that the CENP-A peaks in these regions are not a result of library amplification bias [24] (Fig. S5; Table S7).
Having shown that CENP-A is associated with the complex islands, we next analyzed if the centromere extends to the surrounding satellite DNA. Simple sequences flanking the islands appear among the kmers enriched in the CENP-A ChIP ( Fig. 1B; Table S1; Fig. S1). However, it is difficult to quantify the enrichment of CENP-A on simple satellite repeats for several reasons: 1) simple satellite sequences may be over or underrepresented as an artifact of library preparation [24], particularly for ChIPseq experiments that rely on PCR amplification to construct libraries; 2) satellites are abundant genomic sequences that are largely missing from whole genome assemblies [10], making it difficult to precisely quantitate how much of these sequences exist in genomes (and therefore how much to expect in the input); 3) highly abundant repeats are expected to have a low signal-to-noise ratio if a relatively small fraction of a simple repeat is enriched in CENP-A relative to the overall abundance of this satellite in the genome; and 4) simple satellite repeats present a challenge for even long-read based genome assembly methods [25]. While we may be confident in large-scale structural features of assemblies involving highly repetitive sequences-we see even PacBio read depth in islands, but uneven depth on simple satellites (Fig. S6). Due to these limitations, we caution against using strictly assembly-based approaches in regions with simple repeats. Nonetheless, we report the ChIP peaks on simple satellites (shaded in gray in Fig. 2). To confirm satellite localization near each centromere, we employed IF with anti-CENP-C (an inner kinetochore protein that co-localizes with CENP-A), followed by FISH with probes for the satellites dodeca, AAGAG, AATAT, AAGAT, AATAG, and Prodsat on metaphase chromosome spreads from 3 rd instar larval brains (Fig. S7); a summary of the co-localization data is shown in Table 1.
Although CENP-A localizes exclusively to the centromeres at the cytological level, it is possible that low-levels of CENP-A occupy non-centromeric DNA. We found a low, but consistent CENP-A enrichment at genomic regions outside of the centromere islands, including some telomereassociated elements (e.g. TART-A), rDNA genes from the rDNA clusters, and the LINE-like retroelements DMRT1B and R2 ( Fig. 1C; Table S8 and supplemental results). Many of these associations likely represent non-specific peaks [26], as they were not highly enriched by CENP-A ChIP-qPCR (Fig. S5). However, previous studies found evidence for an association of some centromeric proteins with the nucleolus [27], perhaps relating to the possible association that we detect for rDNA and rDNA-associated retroelements (e.g. R2). We also noted that noncentromeric copies of G2/Jockey-3 were not consistently enriched in CENP-A (Table S8).
CENP-A ChIP-seq reads from S2 cells showed a similar enrichment profile of sequences represented in the embryo ChIP-seq data (e.g. IGS and G2/Jockey-3) but were much more enriched for additional retroelements that were not represented within our centromere contigs (e.g. LTR elements Dm1731, HMSBeagle, and Max-I; Table S2). We also observed a similar pattern of CENP-A enrichment on simple satellite repeats in S2 cells (AATAT, AATAG, AAGAG, Prodsat, and dodeca; Table S1) and we confirmed that these satellites are also near centromeres cytologically using IF/FISH in S2 cells (Fig. S8). However, complex satellites that are pericentric in embryos, including complex satellites in the 1.688 family and Rsp (Table S2), are CENP-A-enriched in S2 cells. This suggests that the centromeres of S2 cells may have expanded into regions that are pericentromeric in flies; the additional retroelements enriched in CENP-A may be pericentric or they may represent new retroelements insertions occurred in this cell line. Our findings are consistent with the extensive structural rearrangements and polyploidy reported for these cells [28].

Centromeres are unique, but are composed of similar non-LTR retrotransposons
Although each island has a distinct arrangement of AT-rich sequences, repeats, and TEs, their compositions are overall similar. In particular, non-LTR retroelements in the Jockey family such as G2/Jockey-3, Doc, and Doc-2 are especially abundant within CenX, Cen4, and CenY ( Fig. 2;  Fig. 3A). Strikingly, G2/Jockey-3 is the only element present in all five of our centromere contigs, suggesting a potential role in centromere function or specification. In our phylogenetic analysis of genomic G2/Jockey-3 repeats in D. melanogaster, we cannot distinguish G2/Jockey-3 elements at centromeres from those across the genome, suggesting that centromeric TEs do not have a single origin ( Fig. 3B and supplemental results). Although G2/Jockey-3 is not unique to centromeres, and thus cannot be sufficient for centromere identity, it is significantly enriched at centromeres: ~63% of all genomic copies of G2/Jockey-3 are found within our candidate centromere contigs ( Fig. 4 and Table S9). G2/Jockey-3 elements show signs of recent or ongoing activity based on their insertion polymorphism [29], pattern of 5' truncation (see supplemental results), and expression ( Fig. S9A). At least some of this expression comes from the centromeres: we analyzed total embryo RNA extracts by reverse-transcription qPCR (RT-qPCR) using primers targeting centromere-associated copies and found evidence for low-levels of G2/Jockey-3 transcription from copies in the X, 4 and 3 centromeres, with no or negligible expression from the Y and 2 (Fig. S9B). In addition to G2/Jockey-3, the 3 Giglio island has 240 copies of a centromere-enriched variant of the ribosomal IGS (supplemental results; Fig. S10). Among the islands, 2 Capri differs the most, being the smallest and harboring only a single fragment of G2/Jockey-3 (Fig. 2E). Importantly, as was previously reported for the X-derived Dp1187 centromere [14,15], none of the sequences contained within these islands are exclusive to centromeres. However, several of these elements are enriched in these regions compared to the genome in addition to G2/Jockey-3. For example, DOC2, G, and Jockey-1 elements are non-LTR retroelements enriched in CENP-A with a genomic distribution biased toward centromeres (Fig. 3A, columns labeled "genome"; Fig. S11; Table S9).

Validation of centromeric contigs
To verify the association of our contigs with the centromeres, we performed IF with anti-CENP-C antibodies, followed by FISH with satellite probes and custom-designed Oligopaints libraries [30](see methods) for X Maupiti , 4 Lampedusa , Y Lipari , and 3 Giglio (Fig. 5 and Fig. S12). The X Maupiti Oligopaints hybridized to the X as well as the Y centromeres on 3 rd instar male larval brain metaphase spreads ( Fig. 5A and Fig. S12A). Similarly, the Oligopaints for 4 Lampedusa hybridized to the 4 th as well as to the Y centromere ( Fig. 5B and S12B), suggesting that Oligopaints for X Maupiti and 4 Lampedusa have homology to sequences at or near the Y centromere. In contrast, the Oligopaints for Y Lipari (Fig. 5C and Fig. S12C) and 3 Giglio were specific for their respective centromeres ( Fig. 5D and Fig. S12D). We could not use Oligopaints to validate 2 Capri because of its small size, but its organization, with AATAG and AAGAG satellites flanking a small CENP-A enriched island (Fig. 2E), is consistent with our FISH analyses (Fig. 5E). In line with the CENP-A ChIP-seq data, we observed significant differences between S2 and embryo centromeres by Oligopaint FISH. With the exception of 3 Giglio , centromeric island organization in S2 cells is dramatically different from larval brain metaphase spreads ( Fig. S13 and supplemental results), in contrast to the conservation of the centromeric distribution of simple satellites ( Fig. S8; Table  S1).
Drosophila centromeres tend to cluster in the nucleus cytologically [31,32]. We found independent support for the complex islands being centromeric by analyzing previously published Hi-C data from D. melanogaster embryos. Island-island interactions were among the most frequent inter-chromosome interactions, followed by interactions between islands and their own proximal pericentric heterochromatin, and lastly by interactions between islands and distal pericentric heterochromatin or euchromatin ( Fig. S14; supplemental results). This analysis also shows that indeed native centromeres interact with one another physically in the 3D nucleus.

Analysis of extended chromatin fibers reveals that CENP-A primarily occupies the islands
Based on the enrichment of CENP-A with island-associated repeats, we hypothesized that the TE-enriched islands are the major centromere components in D. melanogaster. To test this, we investigated CENP-A occupancy, a direct reflection of centromere activity, and estimated the size of each centromere by visualizing extended chromatin fibers [11,33]. This method has two major advantages: it does not rely on mapping low complexity ChIP reads, thus providing more information that can be inferred by ChIP, and it affords single-chromosome, rather than population, information on CENP-A localization. We carried out IF with CENP-A antibodies and FISH with Oligopaint and satellite probes on cells from 3 rd instar larval brains, selecting females to ensure specificity for our X Maupiti and 4 Lampedusa Oligopaints (Fig. 5). First, we calibrated our fiber stretching using three FISH probes spanning 100kb: two heterochromatic (one for the Rsp locus, Table S11 [34], and one Oligopaint targeting the pericentromere of chromosome 3L, see methods for coordinates) and one euchromatic (an Oligopaint targeting a region ~600kb from the telomere of chromosome 3L, see methods). The estimated stretching for these fibers is ~10 kb/µm for all three locations, with no significant difference among them (P=0.085; Fig. S15). We next determined the sizes of the CENP-A domain and corresponding island of each centromere ( Fig. 5 and Figs. S16-21). The size of the CENP-A domain varies between centromeres, ranging in mean size between 101-172-kb (~11µm-17µm), smaller than previous estimates that relied on the measuring of a mixture of centromeres in Drosophila Kc and S2 cells [11]. This is consistent with our ChIP-seq analysis suggesting that S2 cells may have expanded centromeres. X, Y and 2 are the largest centromeres, while 3 and 4 are the smallest (Fig. 5F-G). Importantly, CENP-A primarily occupies the centromeric islands X Maupiti , 4 Lampedusa , Y Lipari and 3 Giglio (~70% of the CENP-A domain overlaps with the Oligopaint FISH signal; Fig. 5G and Figs. S16-21). In some fibers, the X Maupiti Oligopaint FISH signal showed interspersion with FISH signal for the AAGAG satellite (Fig. S16), this could be due to non-specific binding of the AAGAG probe during FISH, which is optimized for Oligopaint specificity, or because AAGAG repeats are collapsed in our assembly, including within Maupiti. We also noticed that the estimated length of the Oligopaint-stained region was larger than the size of Maupiti in our CenX contig (101.3 ±48.3-kb versus 44-kb; Fig. 2A and Fig. 5G), a discrepancy that we attribute to variability in Maupiti Oligopaint probe hybridization. Alternatively, there could be additional sequences with similarity to Maupiti interspersed in the flanking satellites nearby the contig (and not in our assembly). Analysis of centromere 4 shows that the CENP-A domain overlaps primarily with 4 Lampedusa , and partially with the flanking AAGAT satellite (Fig. 5B, F and Fig. S17). The Oligopaints for Y Lipari target only the part of the island with the highest enrichment of CENP-A (Fig. 5C). Fibers for this centromere show a continuous CENP-A domain that extends past the FISH signal, likely representing the remainder of the Y Lipari island ( Fig. 5C and Fig. S18). Fibers for 3 Giglio show co-localization between CENP-A and the island as well as a short, variable region of co-localization with flanking dodeca satellite ( Figure 5D, Fig. S19-20). We did not observe CENP-A signal on the opposite side of Giglio, where Prodsat is located according to our assembly (Fig. 5D). The centromere 3 satellite dodeca co-localizes with CENP-A on fibers in S2 cells [12] and is highly enriched in our CENP-A ChIP-seq ( Fig. 1B; Fig. S1). When we tracked longer fibers from 3 Giglio along dodeca, we observed a second CENP-A domain where dodeca is interrupted by short fragments of Oligopaint FISH signal (Fig. S20), suggesting the existence of DNA sequences with homology to Giglio interspersed within dodeca that are not included in our assembly. A previous study identified sequences with homology to IGS within the dodeca satellite in one BAC [12]. It is possible that the dodeca-associated Oligopaint FISH signal in our extended fibers corresponds to these additional IGS sequences. These data indicate that centromere 3 has two CENP-A domains, a major one on 3 Giglio and one minor one on dodeca, although these appear as a single domain in standard metaphase spread IF. Unlike centromere 3, all other centromeres display a single CENP-A domain by fiber analysis (see Fig.  S21 for centromere 2 and data not shown). Our conclusions differ from the Talbert et al. study [16], which concluded that dodeca was not associated with CENP-A. As recognized by the authors, it is possible that different chromatin preparations, such as the MNase digestion, may introduce biases, leading to an underrepresentation of sequences like dodeca in ChIPs [16]. Lastly, we analyzed the organization of 2 Capri using FISH with a satellite combination unique to this chromosome: AATAG, AAGAG, and Prodsat and found that the CENP-A domain overlapped with all three satellites ( Fig. 5E and Fig. S21). Thus, we speculate that the Prodsat sequences pulled down by CENP-A as seen in our kmer analysis (Fig. 1B) and reported previously [16] are coming from the centromere 2, not 3. We therefore conclude that D. melanogaster CENP-A is primarily associated with the centromeric islands of chromosomes X, 4, Y, and 3, and less predominantly with the flanking satellites (Fig. 5G).

G2/Jockey-3 is centromere-associated in Drosophila simulans
The G2/Jockey-3 retroelement is a recently active transposon [29] shared amongst all D. melanogaster centromeres ( Fig. 3A; Table S2). To determine if G2/Jockey-3 is enriched at the centromeres outside of D. melanogaster, we investigated its centromeric distribution in the sister species, D. simulans, which diverged from D. melanogaster only ~2 million years ago [35] and yet displays major differences in satellite composition and distribution [36,37]. These differences are especially apparent in centromeric regions, where D. melanogaster displays simple satellite repeats while D. simulans contains complex satellite repeats with larger repeat units [16]. We reanalyzed published D. simulans cell line CENP-A ChIP-seq data [16] (see supplemental results) and found that G2/Jockey-3 elements are also highly enriched in CENP-A in this species, similar to D. melanogaster. The pileup of CENP-A ChIP reads on G2/Jockey-3 show that CENP-A is associated with the entire length of the retroelement in both D. simulans and D. melanogaster, with no apparent affinity for any particular sequence ( Fig. 6A-B).
To validate the association of G2/Jockey-3 with D. simulans centromeres, we designed a FISH probe that targets ~1.6 kb at the 3' of the D. melanogaster G2/Jockey-3 consensus sequence (see methods; ~94% identical to D. simulans G2/Jockey-3 consensus sequence) and performed IF/FISH on male larval brain metaphase spreads with anti-CENP-C antibodies, which recognize CENP-A in both species [38]. We observed co-localization between CENP-C and G2/Jockey-3 at all D. simulans centromeres ( Fig. 6C; note that chromosome 2 and 3 of D. simulans cannot be distinguished morphologically [37]). The same probe showed co-localization of CENP-C and G2/Jockey-3 at all D. melanogaster centromeres, except at centromere 2, which is consistent with our model for this centromere showing only one copy of G2/Jockey-3 ( Fig. 6D and Fig. 2E). Based on these observations, we propose that G2/Jockey-3 is a conserved centromereassociated retroelement in these species.

Discussion
Our study shows that combining long-read sequencing with ChIP-seq and chromatin fiber FISH is a powerful approach to discover centromeric DNA sequences and their organization. We reveal that, for all but one chromosome (chromosome 2, which has a single G2/Jockey-3 element), ~70% of the functional centromeric DNA of D. melanogaster is composed of complex DNA islands. The islands are rich in non-LTR retroelements and are buried within large blocks of tandem repeats (Fig. 7A). They likely went undetected in previous studies of centromere organization (e.g. [12]) because three of the five islands are either missing or are incomplete in the published reference D. melanogaster genome [10]. A recent study reported that satellite DNA repeats comprise the majority of centromeric DNA in D. melanogaster embryos and S2 cells, by counting the relative number of motifs matching simple repeats in CENP-A ChIP relative to input [16]. Our re-analysis of those data showed CENP-A enrichment on the islands, suggesting that having an improved reference genome assembly [19] is crucial for identifying centromeric DNA. To our knowledge, this is the first detailed report on the linear sequence of all centromeres in a multicellular organism. Our overall strategy therefore provides a blueprint for determining the composition and organization of centromeric DNA in other species.
To date, satellite DNAs have been regarded as the main sequence components of the centromeres of primary animal model systems-humans, mice, and Drosophila [2,3,17]. However, retroelements are also abundant and widespread at the centromeres of plants such as maize [39] and rice [40,41]. Retroelements are also found at the centromeres of fungi [42], humans [43], marsupials [44], bats [45], and gibbons [46], suggesting that they are pervasive centromeric features (Fig. 7B). Our study shows that retroelements, particularly G2/Jockey-3, are not merely present near centromeres but are components of the active centromere cores through their association with CENP-A. Our BLAST search for G2/Jockey-3 retroelements suggests that they are restricted to the melanogaster subgroup, therefore we hypothesize that different non-LTR retroelements may be present at the centromeres of other Drosophila species. Why retroelements are such ubiquitous components of centromeres and if they play an active role in centromere function remain open questions. In maize, centromeric retroelements invade neocentromeres following their inception [47], suggesting a preference for DNA sequences associated with CENP-A chromatin for retroelement insertion [18]. On the other hand, a LINE-element was found to be an integral component of a human neocentromere [48,49], raising the possibility that it is CENP-A that may bind preferentially to retroelementassociated genomic regions [18]. Other models have proposed that retroelements could produce non-coding RNAs that affect centromere specification [18,48], and that retroelement activity could help maintain centromere size through retrotransposition or by giving rise to tandem repeats via recombination-mediated mechanisms (e.g. [50,51]; reviewed in [52]). Centromeric transcription contributes to centromere homeostasis in several organisms, including fission yeast [53,54], wallaby [55], human [6,56], and Drosophila cells [57,58]. Our reanalysis of publicly available total RNA-seq reads was inconclusive about steady state transcription levels coming from centromeric contigs (data not shown) due to extremely low expression levels and insufficient read length, however our preliminary analysis with quantitative RT-PCR with centromere-specific G2/Jockey-3 primer sets shows some evidence for a low level of centromere expression.
In addition to retroelements, the centromeres of D. melanogaster display a diverse assortment of repeats, none of which are exclusive to centromeres, with the exception of IGS, for which we identified a centromere-enriched variant. The identification of the IGS tandem repeat within 3 Giglio is intriguing, as IGS sequences are dynamic in the potato [59], where they are located near the centromere, as well as in the tobacco [60], the tomato [61], and the common bean [62], where they show a dispersed pattern over several chromosomes. The origin of novel tandem repeats is still elusive, but one way it has been proposed to occur for the IGS repeat in plants is through the initial insertion of a retroelement within rDNA, followed by IGS duplication, amplification, and transposition to a new locus [61].
Defining the span of the CENP-A domain is important to understand precisely which sequences are associated with centromere activity and which are part of pericentric heterochromatin. Although we are able to confidently map our ChIP-seq reads to the islands to determine CENP-A occupancy, the same cannot be done for simple satellites due to the limitations of mapping to highly repetitive DNA. We therefore infer the organization of the centromere from analyzing extended chromatin fibers by IF/FISH. Blocks of simple satellite sequences flank the islands on each of our contigs, with the exception of the Y centromere contig, however, these regions represent only a fraction of the estimated abundance of those repeats in the genome. For example, dodeca satellite occupies ~1Mb of the genome [63] and we have ~570 kb total dodeca sequence in the assembly, however only ~35 kb of dodeca is on the centromere 3 contig. Therefore, for many satellite sequences, inferences based on read mapping, even uniquely mapped reads, are confounded because satellites are underrepresented in the assembly. Our analysis of chromatin fibers suggests that CENP-A spans beyond the islands into the simple satellites, although the precise boundaries remain elusive (Fig. 7A).
The finding that CENP-A can bind to several different sequences that are not uniquely associated with centromere regions is consistent with the epigenetic model of centromere specification, which proposes that specific sequences alone do not govern centromere activity [3]. Yet it is possible that the diverse sequence arrangements observed at each centromere somehow contribute to centromere activity or specification [18,39]. Possible mechanisms include the promotion of unusual types of transcription, as reported for fission yeast [64], or the formation of non-B DNA structures (e.g. stem loops, hairpins, and triplexes) that may promote CENP-A deposition [7,12,65]. Knowing the identity of D. melanogaster centromeric DNA will enable the functional interrogation of these elements in this powerhouse model organism.

Table 1: Location of centromeric and centromere-proximal satellites in Drosophila melanogaster.
Locations of satellite on chromosomes X, Y, 2, 3, and 4 according to previous reports as well our observations in this report by IF-FISH in the D. melanogaster sequenced strain iso-1. Each satellite location is characterized as being centromeric (Cen, overlaps with CENP-C), pericentric (Peri, juxtaposed to CENP-C) or heterochromatic (Het, more distal than pericentric). Note that the dodeca satellite includes its variant and Prodsat is also known as the 10bp satellite. *Indicates a small block not easily detected by FISH.

Author contributions
AML and BGM conceived the project, designed experiments, and wrote the manuscript. AC, JP, BS, and BGM performed wet lab experiments and analyzed data; C-HC, XW, and AML performed computational experiments and analyzed data; NMCM and BJB designed Oligopaints; JE and C-TW provided resources; all authors contributed to editing the manuscript and writing the supplemental results and methods.

Acknowledgments
We are grateful to Gary Karpen for the anti-CENP-A antibody and for comments on the manuscript. We also thank Guy Nir (Oligopaint design), Kevin Wei (    Organization of each CENP-A enriched island corresponding to centromere candidates: Different repeat families are color coded (see legend; note that Jockey elements are shown in one color even though they are distinct elements). Shown are the normalized CENP-A enrichment over input (plotted on a log scale) from one replicate (Replicate 2, other replicates are in Fig. S4) colored in gray for simple repeats and black for complex island sequences. While the mapping quality scores are high in simple repeat regions, we do not use these data to make inferences about CENP-A distribution (see text for details). The coordinates of the significantly CENP-A-enriched ChIPtigs mapped to these contigs (black) and the predicted ChIP peaks (orange) are shown below each plot.      For at least centromeres X, 4 and 3 (since the sequences flanking the Y centromere are not in our assembly), the bulk of CENP-A chromatin is associated with the centromere islands, while the remaining CENP-A is on the flanking satellites. Although the complexity of island DNA allowed us to identify centromere contigs by long-read sequencing, the flanking satellites remain largely missing from our genome assembly, due to their highly repetitive nature. The approximate satellite size estimates are based on Jagannathan et al. [37]. B) Phylogenetic tree showing that centromere-associated retroelements are common across highly diverged lineages: Gossypium hirsutum (cotton) [71], Zea mays mays (maize) [9,39], Oryza sativa (rice) [40,41,72], Triticum boeoticum (wild wheat) [73], Cryptococcus [42], Phyllostomid (bat) [45], Hoolock leuconedys (gibbon) [46], Homo sapiens (human) [43] and a human neocentromere [49], Macropus eugenii (tammar wallaby) [74][75][76], Phascolarctos cinereus (koala) [77], Drosophila melanogaster (this paper). The Phylogeny was constructed using TimeTree [78]. Indicated are the retroelement type and the clade that the element belongs to with element types as follows: LTR, Long Terminal Repeat retroelement; Non-LTR, Non-Long Terminal Repeat retroelement. The circles indicate the experimental evidence for centromere-association of retroelements: FISH, CENP-A ChIP-seq (ChIP), and genome or BAC sequencing (Seq).

Chromatin immunoprecipitations sequencing (ChIP-seq)
CENP-A ChIPs were performed using an affinity purified rabbit anti-CENP-A antibody (gift of Gary Karpen) that we previously verified works well for ChIP using S2 cells that contain LacI/lacO inducible ectopic centromeres and showing that CENP-A ChIP pulled-down lacO sequences [79] (Fig. S1A).

ChIP in S2 cells
Chromatin from 10 6 fixed Drosophila Schneider (S2) cells (approx. 90µg) were used for each IP and chromatin was sheared to 100-300bp using a Covaris sonicator. ChIPs were performed using the MAGnify kit (Thermo Fisher). 1µl of the anti-CENP-A antibody was coupled to 10µl of beads for 2 hours followed by incubation with chromatin overnight at 4 o C. DNA was eluted in 50µl of elution buffer. Libraries were generated using the TruSeq kit (Illumina) and paired-end sequenced using the Reagent kit v.3. (Illumina) on the NextSeq platform.

ChIP in Embryos
Embryo (OregonR) collection, fixation, and chromatin isolation were performed as described in [80]. We carried out four ChIP replicates as follows. From one embryo collection, we generated chromatin used in ChIP1, from a second independent embryo collection, we generated chromatin used for replicates, ChIP2-4. We used formaldehyde-crosslinked overnight collections of Oregon R embryos (~1.5g per collection). Chromatin was sheared to 200-500bp using a Bioruptor sonicator (Diagenode), aliquoted, and flash frozen. The first biological replicate (ChIP1) was performed following the protocol in [80] using 165µg of chromatin in 500µl volume and 30µl of protein A agarose beads, and 2µl of anti CENP-A antibody. For ChIP2, 3, and 4 we used the MAGnify kit, with 15µl of dynabeads, approximately 60µg of chromatin in 200µl volume, and 3µl of anti-CENP-A antibody. Libraries were made from eluted DNA using the TruSeq ChIP kit (Illumina) for ChIP1 and 4, while the Accel-NGS 2S Plus DNA Library (Swift Biosciences) was used for ChIP2, 3. Note that ChIP2-3 where performed in parallel and were sequenced the same way and are thus technical replicates. The libraries were sequenced by paired-end on the NextSeq platform using Reagent v.2. Chromatin extracted from the second embryo collection was also used for ChIP-qPCR experiments.
For both chromatin preparations, the quality of the chromatin was confirmed by control ChIPs with 15µg of chromatin in 200µl volume and 2µl of rabbit anti-H3K27Ac (Thermo Fisher). The eluted DNA was analyzed by qPCR confirming enrichment of the Rpl32 promoter (F-TTGTTGTGTCCTTCCAGCTTCA and R-TTGTTGTGTCCTTCCAGCTTCA), and lack of enrichment of Rpl32 5' region (F-GGCACGGCGCCAAAATTAATCA and R-CCGATGCCACTGCCTCTTTGGT [81,82].

ChIP quality control analyses
We estimated read quality of each replicate ChIP experiment using two metrics estimated in phantompeakqualtools [83]: the Normalized Strand Coefficient (NSC) and the relative strand correlation (RSC) ( Table S5). These statistics report the cross-correlation between Watson and Crick strands, as ChIP reads from a true positive are expected to be highly clustered and accumulate on either side of the binding site on both strands, with a shift between the peaks on the Watson and Crick strands that is determined by read length and fragment length distribution [84]. This shift should not occur in the input. NSC is the fragment-length cross-correlation peak divided by the background cross-correlation and RSC is the fragment-length cross-correlation peak divided by the read-length peak [83].

Analysis of repeat enrichment in ChIPs
To determine CENP-A enrichment in simple tandem repeats, we summarized repeat composition in the trimmed reads and identified overrepresented kmers using kseek (https://github.com/weikevinhc/k-seek; [24]). The CENP-A/Input ratio is normalized by the number of mapped reads to the genome assembly to remove possible read contamination. We consider a class of repeats to be enriched for CENP-A if the minimum number of kmers in the input is ≥10 in each replicate and the median normalized CENP-A/Input ratio is > 1 across all four replicate ChIP experiments (Fig. S1). Simple tandem repeats maybe over or underrepresented due to Illumina library preparation and the effects of PCR amplification on sequence library complexity. To determine CENP-A enrichment on complex repeats, we used a mapping approach. We annotated repeats in our assembly [19] using a custom Drosophilaspecific consensus repeat library [34] modified from RepBase to include complex satellite DNAs (Repbase version 20150807; [85]; Supplemental File 1). Using these RepeatMasker annotations, we generated a comprehensive library of all individual repetitive elements in the genome to capture sequence variation among repeats. We mapped ChIP and input reads to this comprehensive repeat library using bowtie2 (default parameters) and summarized read counts for each type of complex repeat (e.g. transposable elements, complex satellite DNAs with repeat units >100 bp) using custom python scripts. The CENP-A/Input ratio is normalized by the number of mapped reads to the genome assembly. We consider a class of repeats to be enriched for CENP-A if it is in top 20 th percentile of normalized CENP-A/Input in all four replicate ChIP experiments.
To address if any motif in G2/Jockey-3 is particularly enriched for CENP-A, we constructed a consensus sequence of G2/Jockey-3 in D. melanogaster and D. simulans. We mapped ChIP and input reads to this comprehensive repeat library with only one version of G2/Jockey-3 (either D. melanogaster or D. simulans) using bwa (default parameters). We then called the depth of reads with samtools depth (v1.7) using "-Q 10 (mapping quality ≥ 10)" and calculated ChIP/Input ratio across each site after normalization by the number of mapped reads to the genome assembly.

Identification of candidate centromeric contigs
We identified candidate centromeric contigs in the new iso-1 assembly [19] based on organization: we looked for contigs containing complex DNA flanked by satellites with known centromeric and pericentric locations. We first generated the assembly with long-read sequence data, including PacBio [89] and nanopore reads (Table S4; [90]). We filtered nanopore reads using Porechop and Filtlong (--min_length 500) to remove adaptors and short reads (https://github.com/rrwick/Porechop and https://github.com/rrwick/Filtlong). We assembled the nanopore and PacBio reads into a hybrid assembly using Canu v1.7 with default settings [91]. Our new hybrid PacBio-Nanopore assembly is less contiguous than our previous PacBio-only assembly despite using more reads (Supplemental File 7; Assembly size = 162,798,260 bp in 798 contigs; N50=5,104,646 bp). We thus decided to use our PacBio-only assembly [19], which has a greater representation of heterochromatin compared to previously published assemblies (see details in [19]). To ensure that we were not missing putative centromeric contigs, we looked for sequences with CENP-A-enriched repeats (based on our repeat analysis; Fig. 1B;  Fig. S1) in the error-corrected PacBio and nanopore reads, and the hybrid assembly, that were missing from our PacBio-only assembly. We were particularly interested in contigs containing repeat sequences that we identified as enriched in our ChIP data. To annotate contigs and unassembled corrected-reads, we used RepeatMasker 4.06 [92] with Repbase 20150807 and parameters "-species drosophila -s" to annotate interspersed repeats (described above) and Tandem Repeat Finder (TRF v4.09; [93]) to annotate tandem repeats. We extracted 19 nonredundant sequences from our new hybrid assembly and error-corrected reads with candidate centromeric repeats, including dodeca, Prodsat, AATAG, IGS 3cen (as determined by phylogenetic analysis below), and G2/Jockey-3 sequences. We added these 19 candidates to our new PacBio-only assembly [19] (Supplemental File 8) to create the final version of our assembly. We polished the final assembly 10 times using Pilon (v1.22 [94]; with Illumina [95,96] and long synthetic reads [97] (Table S6; with parameters "--mindepth 3 --minmq 10 --fix bases"). We annotated the finished assembly using our customized repeat library (-lib library.fasta -s) and RepeatMasker 4.06 [92] (Supplemental File 9). Additionally, we transferred gene annotations from Flybase r6.20 to our genome using BLAT [98] and CrossMap v0.2.5 [99] (Supplemental File 10).

ChIP qPCR
Quantitative real-time PCR (qPCR) was performed using SYBR-green (Bio-Rad) on a CFX96 Real-Time System (Bio-Rad). 1µl of Input or ChIP eluted DNA was used in each qPCR reaction. Melting curves were analyzed to ensure primer specificity. Only primers with reaction efficiencies within a linear dynamic range were used. The fold-enrichment of centromeric DNA after immunoprecipitation of CENP-A chromatin compared to its level in the bulk input chromatin was calculated with the equation: 100*E (Ctinput-Ctip), where E is the efficiency of the primer set. Enrichment values were normalized by the enrichment value of RpL32 as a noncentromeric control. qPCR primer sets are listed in Table S7.

Transcription of centromeric G2/Jockey-3 elements
Total RNA was extracted from three independent overnight collection of embryos (iso-1). Briefly, embryos were scooped from the apple juice plates and rinsed with water in a mesh basket, dechorionated in 50% bleach for 3.5 min with gentle shaking, rinsed thoroughly with water, moved to a 1.5ml microfuge tube, and resuspended in 300µl of Trizol reagent (Sigma Aldrich). Embryos were homogenized using a motorized pestle until the solution became clear (30-40 seconds). The homogenized solution was centrifuged at 13000rpm for 10 min at 4°C and the clear supernatant was transferred to a new RNAse free tube. RNA was isolated using the Direct-Zol RNA miniprep plus kit (Zymo Research) according to the manufacturers' protocol. The RNA was eluted in RNAse free water and quantified with a Nanodrop. A total of three consecutive Turbo DNAase (Invitrogen) followed by RNAeasy Clean up (Qiagen) treatments were performed to remove DNA contamination. Reverse-transcription was performed using the iScript Select cDNA Synthesis kit (Bio-Rad) according to the manufacturer's instructions. Briefly, 75ng of total embryo RNA were used to make cDNA libraries using random priming in a 30µl reaction. For the no-RT control, the Reverse Transcriptase was omitted from the reaction. qPCR was performed as described for ChIP qPCR using 1µl of cDNA in each reaction and primers sets targeting G2/Jockey-3 copies from each centromere (X-G2, 4-G2, Y-G2, 3-G2, 2-G2). Primers for Actin5C were used as a positive control for a highly expressed gene, while a primers for the testis-specific gene, Mst84Da, was used as a control for a non-expressed gene. The no-RT samples produced Ct values comparable to the negative non-expressed control showing successful removal DNA. Gene expression was analyzed as in Schmittgen et al. [102] by determining the mean 2 -ΔCt , where ΔCt is (Ct G2/Jockey-3 -Ct Mts84Da ), from three biological replicates. Primer sets are listed in Table S7.

Primer
Centromere Forward Reverse Table S7. List of qPCR Primers. List of primers used for qPCR in this study. The centromere contig that each target is associated with (X,4,Y,3,2) is designated in the "Centromere" column. Note that in silico PCR for the 3_G2 primers predicted 3 specific products from centromere 3, as well as two products on contig tig00022795 and additional non-specific products from the X chromosome when 3 or more mismatches are allowed all of the same 145bp size.

S2 mitotic chromosome preparation
Preparation of mitotic chromosomes from Drosophila S2 cells was performed as described in [79]. 2x10 5 cells were treated with 0.5µg/mL demecolcine solution (Sigma-Aldrich) and incubated at 25 o C for 1 hour to induce a mitotic arrest. Cells were pelleted (600g/5 min) and resuspended in 250µL 0.5% (w/v) sodium citrate for 8 min. Cells were loaded into cytofunnels and spun onto Superfrost ® Plus slides (VWR) at 1200rpm for 5 min using a Shandon Cytospin 4 (Thermo Fisher Scientific). Cells were fixed for 10 min with 3.7% formaldehyde in PBS, 0.1% Triton-X-100 (PBS-T). Slides were washed three times in PBS-T for 5 min and stored at 4 o C until ready for use.

D. melanogaster and D. simulans mitotic chromosomes preparation
Preparation of mitotic spreads was carried out from iso-1 D. melanogaster flies (Bloomington Drosophila Stock Center stock n. 2057: y 1 ; Gr22b iso-1 Gr22d iso-1 cn 1 CG33964 iso-1 bw 1 sp 1 ; MstProx iso-1 GstD5 iso-1 Rh6 1 ) and D. simulans (w501, gift of Andy Clark) in larvae following the method in [103] with minor modifications. Male 3 rd instar larval brains were dissected in PBS and immersed in 0.5% (w/v) sodium citrate for 8 min. Individual brains were fixed for 6 min in 6µL of 45% acetic acid, 2% formalin on siliconized coverslips. Whole brains were applied to clean poly-L-lysine slides (Thermo Fisher Scientific) and were manually squashed between coverslip and slide by pressing with the thumb. Slides were immersed in liquid nitrogen. Once bubbling stopped, the slides were removed from liquid nitrogen and the coverslip was immediately removed using a razor blade. Slides were immediately immersed in PBS and were either washed for 5 min before proceeding to IF or stored at 4 o C in PBS until ready for use.

Immunofluorescence staining
For IF, slides were washed in PBS-T for 5 min. S2 cell slides were blocked in 5% milk in PBS-T for 30 min. Larval squashes were blocked in 1% BSA, PBS, 0.02% sodium azide for 30 min. Primary antibodies anti-CENP-A (larval brain slides: rabbit, 1:500, Active motif; S2 cell slides: chicken, 1:1000, [70] and anti-CENP-C (larval brain slides: guinea pig, 1:500 [27]) were diluted in blocking solution and incubated on slides overnight at 4 o C. Slides were washed three times for 5 min in PBS-T and incubated with secondary antibodies (Life Technologies Alexa-488, 546, or 647 conjugated, 1:500) diluted in blocking solution and incubated at room temperature for 1 hour or overnight at 4 o C. Slides were washed three times for 5 min in PBS-T.

Satellite FISH
Satellite FISH was performed following the protocol described in [104] with a few modifications. Slides were post-fixed in 3.7% formaldehyde, PBS for 10 min, followed by a rinse in PBS and two 5 min washes in 2xSSC, 0.1% Tween-20 (2xSSC-T). Slides were washed once for 5 min in 50% formamide, 2xSSC-T at room temperature, once for 20 min in 50% formamide, 2xSSC-T at 60 o C, and then cooled to room temperature. For FISH, 25µL of hybridization mix containing 40pmol of each probe, 2xSSC-T, 10% dextran sulfate, 50% formamide, and 1µL of Rnase Cocktail (ThermoFisher) was applied to a 22 x 22mm hybrislip (Electron Microscopy Sciences), mounted on the slide and sealed with paper cement. Slides were denatured at 92 o C for 2.5 min and then incubated overnight at 37 o C. Slides were washed in 2xSSC-T at 60 o C for 20 min, followed by two 5 min washes in 2xSSC-T at room temperature, and one 5 min wash in PBS. Slides were mounted in Slowfade ® Gold Reagent (Invitrogen) containing 1µg/mL DAPI and sealed with nail polish.

Oligopaint FISH
Oligopaint FISH was performed as described above with the following modifications. 25µL of hybridization mix containing 10pmol of Oligopaint, 2xSSC-T, 10% dextran sulfate, 60-68% formamide, and 1µL RNase cocktail was applied to a 22 x 22mm hybrislip (Electron Microscopy Sciences), mounted on the slide and sealed with paper cement. Slides were denatured at 92 o C for 2.5 min in a thermocycler (Eppendorf) and incubated overnight at either 37 o C or 42 o C (see Table S10 for % formamide and hybridization temperatures used). For fluorescence detection, 10pmol of Alexa-488 labeled secondary oligo were applied either during the overnight hybridization or following post-hybridization washes, in which 25µL of 2xSSC, 30% formamide, 10pmol of probe was applied to each slide and incubated at room temperature for 30 min. Slides were washed twice in 2xSSC, 40% formamide for 20 min, once in 2xSSC-T for 15 min, and once in PBS for 5 min. Slides were mounted as described above and successful hybridization was checked under fluorescent microscope. Satellite probes were added after imaging by removing the coverslip with a razor blade, washing slides three times in 2xSSC-T for 5min, applying 25µL of 2xSSC, 30% formamide, 40pmol of satellite probe to each slide and incubating at 37 o C for 1 hour. Slides were washed once in 2xSSC-T at 60 o C for 20 min, twice in 2xSSC-T for 15 min, once in PBS for 5 min, and mounted as described above.

G2/Jockey-3 FISH
FISH for G2/Jockey-3 was performed as described in Dimitri et al. [105]. Slides were dehydrated in an ethanol row (successive 3 min washes in 70%, 90%, and 100% ethanol) and allowed to air dry completely. 20µL of probe mix containing 2xSSC, 50% formamide, 10% dextran sulfate, 1µL RNase cocktail, and 100ng of DIG-labeled G2 probe was boiled at 80 o C for 8 min, incubated on ice for 5 min, and then applied to slides, covered with a glass coverslip and sealed with paper cement. Sealed slides were denatured on a slide thermocycler for 5 min at  Alexa 488 IDT [108] Table S13. Secondary Oligo Probes. Sequence and fluors of secondary oligo probes used for fluorescence detection of Oligopaints and unlabeled satellite probes.

Preparation of extended chromatin fibers and IF-FISH
Extended chromatin fibers were prepared as described in Sullivan [33], with a few modifications. 3-4 brain imaginal discs from third-instar iso-1 wandering larvae (females were selected to avoid cross-centromere hybridization of our X Maupiti and 4 Lampedusa Oligopaints with the Y centromere while males were used for centromere Y) were dissected in 0.7% NaCl and dissociated in 250µl 0.5% (w/v) sodium citrate containing 40µg collagenase/dispase (Sigma-Aldrich) by incubating at 37 o C for 10 min. This mixture was briefly vortexed and spun, and loaded into a single chamber Shandon cytofunnel for centrifugation in a Shandon Cytospin 4 at 1200rpm for 5 min onto a clean polysine slide (Thermo Fisher Scientific). After centrifugation, the slides were immediately immersed in a glass coplin jar containing lysis buffer (500mM NaCl, 250mM Urea, 25mM Tris-HCl pH7.4, 1% Triton X-100) for 13-15 min following which the slides were gently removed at a steady state of ~25-30 seconds per slide. Fibers were fixed in 4% formaldehyde solution and washed in PBS for 5 min. After washing, the slides were processed for IF-FISH.
Fibers were extracted in PBS-T for 10 min then incubated in a 1.5% BSA, PBS blocking solution for 30 min. Slides were incubated with an anti-CENP-A antibody (rabbit, 1:100, Active Motif) diluted in blocking solution overnight at 4 o C in a humidified chamber. Slides were washed three times in PBS for 5 min and then incubated for 45 min with secondary antibodies (Cy5conjugated donkey anti-rabbit, 1:500, Life Technologies) diluted in blocking buffer at room temperature, followed by three 5 min washes in PBS. Slides were post-fixed in 3.7% formaldehyde, PBS for 10 min followed by one quick rinse and two 5 min washes in PBS. FISH was performed as described for 3D-FISH [104] with a few modifications. Slides were washed twice in 2xSSC-T at room temperature for 5 min, followed by denaturation in 50% formamide, 2xSSC-T at room temperature for 5 min, transferred at 60 o C for 20 min, then cooled to room temperature. 10pmol of primary Oligopaint probes (except X Maupiti which was 25pmol) and 40pmol of satellite DNA probes were each added to the slides in 25µL of hybridization solution: 2xSSCT, 60% or 68% (v/v) formamide (see TableS10), 1µL RNase cocktail (Invitrogen), 10% dextran sulfate (Millipore) and sealed with a 22x22 hybrislip (Electron Microscopy Sciences) using rubber cement. Slides were then denatured at 92 o C for 3 min on a slide thermo-cycler and allowed to hybridize overnight at 37 o C or 42 o C (See Table S10) in a humidified chamber. Slides were washed once in 2xSSC-T for 15 min at 60 o C, once in 2xSSC-T for 10 min at room temperature, and once in 0.2xSSC for 10 min at room temperature. Following washes, 25µL of hybridization mix containing 2xSSC, 30% formamide, 40pmol fluor-labeled secondary oligo probes (see Table S13) was added on to each slide and incubated for 45 min at room temperature except for the X Maupiti slides where the satellite probe was also added with the secondary Oligopaint probe. The slides were then washed once in 2xSSC-T at 60 o C for 15 min, followed by one wash in 2xSSC-T and 0.2xSSC for 10 min at room temperature and mounted as described above.
For FISH with only satellite probes, post-hybridization washes consisted of one wash in 2xSSC-T for 20 min at 60 o C, followed by one wash with 2xSSC-T at room temperature for 10 min and two 5 min washes in 0.2xSSC at room temperature. Slides were mounted in as described above.
For fiber measurement calibration, FISH using the 61C7 and 80C4 probes was performed using the conditions for Oligopaint FISH (see Table S10 for % formamide and hybridization temperatures used), while FISH using the Rsp probe was performed using the satellite FISH protocol.

Microscopy and image analysis
Image acquisition was done at 25 o C using an Inverted Deltavision RT restoration Imaging System (Applied Precision) equipped with a Cool Snap HQ 2 camera (Photometrics) and 100x/1.40 NA oil immersion lens (Olympus). Image acquisition and processing was performed using softWoRx software (Applied Precision). For mitotic chromosomes, 20 z-stacks were taken per image at 0.2µm per slice. For fibers, 12-15 z-stacks were taken per image at 0.15µm per slice. Images were deconvolved using the conservative method for 5 cycles. Maximum intensity projections were made using 3-5 z-stacks. Images were saved as Photoshop files and were scaled using Adobe Photoshop. Figure assembly was done using Adobe Illustrator.
Maximum intensity projections of individual fibers were analyzed to measure the signal length of various signals on fibers using the "measure distances" tool in Softworks (GE). Three calibration probes of known length (100kb; see Tables S10 and 15 for 80C4 and 61C7 Oligopaints; see Supplemental File 11 for Rsp probe) were used to determine the degree of stretching in our experiments. At least 20 fibers for each probe were measured in all cases. Length measurements were visualized by scatter plot using Prism. These lengths were then used to determine the average stretching in kb/µm and student t-test was used for statistical analyses.
We noticed that the variation in the measurements for the island Oligopaints was greater than what we observed for the probes used for calibration. We attribute this higher variation to the lower density of island Oligopaint probes (some of the island sequences were not targeted by probes to increase specificity), which causes the signal to be weaker and less consistent than in standard Oligopaint FISH. It is also important to note that we analyzed fibers from a mixed population at different stages of the cell cycle, which could display differences in CENP-A density. It is also possible that the stretching of the chromatin at the centromere is more variable than at non-centromeric regions.

Design
Oligopaint libraries were designed using the OligoMiner pipeline [30,109] with some variations. The genomic regions that showed significant enrichments of CENP-A via MACS and enriched ChIPtigs were targeted for Oligopaint design. The blockparse.py script (v1.3) using overlap mode was used to identify as many candidate probes as possible, with genome-targeting regions 35-41bp long, and a desired Tm of 42-47 o C. Unlike standard Oligopaints design, the candidate probes were not aligned to the genome using Bowtie2 [110] or filtered with OutputClean.py, so that probes which align multiple times would not be discarded. Candidate probes with partial alignments of 18bp-long kmers were filtered out using kmerfilter.py (v1.3) and Jellyfish [111], excluding any that matched 6 or more times to the genome. Probes were filtered further for least secondary structures using StructureCheck.py (v1.3) and NUPACK [112]. Finally, coverage and density of probes across the regions of interest and presence of densely clustered off-target alignments were manually checked by Bowtie2 alignment, filtering for different levels of mismatch to assess the effects of hybridization stringency.

Hi-C analysis
We used a publicly available Hi-C dataset from embryos (Gene Expression Omnibus accession number GSE103625) to provide additional support our candidate centromeric contigs [113]. We mapped Hi-C sequence reads to our assembly and processed the output with the HiC-Pro pipeline [114] to obtain informative valid interaction pairs (default parameters). We used a customized python script to count interactions between regions of interest and then normalized to the size of the regions (per 100 kb). To count interactions between different sized windows, we used BEDTools [115] to create windows of specified sizes across the assembly. We established the euchromatin-heterochromatin boundaries in our assembly based on previous studies. For chromosome 2, 3, X and Y, we transferred the euchromatin-heterochromatin boundary coordinates previously reported for D. melanogaster [116] to our assembly. For chromosome 4, we assigned the ~70 kb closest to the centromere in the assembled chromosome 4 as heterochromatin based one previously reported [117], and the rest of it as euchromatin. We then binned the genome into different regions based on their sequence content: centromere, centromere proximal heterochromatin, centromere distal heterochromatin, and euchromatin (Table S16). We then classified interactions between centromeric contigs and the different categories based on their genomic region (e.g. centromere to proximal heterochromatin, centromere to distal heterochromatin etc.). We reported the median count for each category and conducted data visualization and statistics in R. We calculated the significance between different categories using a Kruskal-Wallis test by ranks with Dunn's test for post-hoc analysis, and the pairwise Wilcoxon rank sum test with false discovery rate (FDR) correction [118] of type I error rates for multiple comparisons. We deem a result to be significant only if both tests agree.

G2/Jockey-3 activity
We asked if G2/Jockey-3 non-LTR retroelements have evidence for recent activity based on insertion polymorphism and expression. We examined RNAseq reads from testes for evidence of G2/Jockey-3 because of the enrichment of these elements on the Y chromosome. We mapped poly-A [122] and total RNA [123] (Table S6) transcriptome data to our Repeat library using HISAT 2.1.0 [124] and estimated read depth of uniquely mapped read using samtools (depth -Q10; v1.7 [125]).

Data availability
All supporting data associated with the manuscript is in the supplemental results. Supplemental files containing raw data and analyses are submitted to the dryad digital repository (doi:10.5061/dryad.rb1bt3j), Github (https://github.com/LarracuenteLab/Dmel.centromeres), and NCBI's SRA under the bioproject PRJNA482653 and accessions SRR7588743 -SRR7588752.