We searched Flybase to identify all genes in D. melanogaster that encode a ZAD domain (PF07776; Pfam database, Pfam.org). We found 91 ZAD-ZNF genes distributed across Chromosomes 2, 3, and X. 37 ZAD-ZNF genes occur in 13 gene clusters, containing two or more tandemly-arrayed ZAD-ZNF genes. Of these, many ZAD-ZNF genes share intron/exon structures with their neighbors and likely arose via segmental duplication. In contrast, seven ZAD-ZNF genes (CG3032, CG4318, CG9215, CG44002, CG17361, CG7963, CG17359) lack introns found in their closest relatives; we infer that these genes were likely born via retrotransposition.

Although ZAD-ZNF proteins are defined as having both a ZAD and a ZNF domain, further analysis of ZAD-containing proteins using NCBI’s Conserved Domain Database revealed significant variation in the ZNF domains. We found that ZAD-containing proteins have an average of six C2H2 domains. However, some ZAD-containing proteins have no C2H2 domains at all (dlip, dbr, CG15435, CG31109, CG31457), whereas others contain up to 23 C2H2 domains (CG11902). In addition to the ZAD in the N-terminus, 13 ZAD-ZNF proteins possess another N-terminal domain, such as a Sodium-Calcium exchanger domain (CG12391) or ASF1 histone chaperone domain (CG10321). The functional significance of these additional domains is unknown.

Next, we surveyed ZAD-ZNF genes in all 12 previously sequenced and annotated Drosophila genomes. These species represent a range of evolutionary divergence from D. melanogaster, from just a few million years (e.g., D. simulans), to more than 40 million years (e.g., D. virilis) (Drosophila 12 Genomes Consortium, 2007). Our analysis reveals a surprisingly wide range in number of ZAD-ZNF genes across different Drosophila species genomes (Figure 1). For example, we found that the D. persimilis genome encodes 130 ZAD-ZNF genes whereas the D. willistoni genome encodes only 75 ZAD-ZNF genes (Supplementary file 1). Such analyses are dependent on the state of completion and annotation of individual Drosophila species’ genomes. Thus, this number may be an underestimate. To complement these analyses, we assessed the apparent age of each D. melanogaster ZAD-ZNF gene by examining the presence of orthologs in the 12 annotated genomes of Drosophila species (flybase.org). We found that 74 of 91 D. melanogaster ZAD-ZNF genes arose in or prior to the common ancestor of all Drosophila species, 40 million years ago. Of these 74 ZAD-ZNF genes, 61 have been preserved over Drosophila evolution, whereas 13 have been lost in at least one lineage or species. We estimate that 17 ZAD-ZNF genes arose via gene duplication after the last common ancestor of D. melanogaster and D. virilis. At least 3 ZAD-ZNF genes found in D. melanogaster (CG4318, CG17612, neu2) originated via gene duplication less than 10 million years ago. Our findings complement previous large-scale surveys that identified rapid changes in ZAD-ZNF gene repertoires within insect genomes (Chung et al., 2002; Lespinet et al., 2002; Chung et al., 2007).

These rapid changes in ZAD-ZNF gene repertoires suggested that selection might favor their genetic innovation. We investigated whether evolutionary retention is a predictor of essentiality. We took advantage of the fact that knockdown or knockout phenotypes have been characterized for almost all D. melanogaster genes. Indeed, 85 of 91 D. melanogaster ZAD-ZNF genes have associated phenotypic outcomes (summarized in Supplementary file 1). Of these, knockdown or knockout of 22 ZAD-ZNF genes showed complete lethality or sterility in D. melanogaster, whereas the other 63 were determined not to be essential.

Of the 61 ZAD-ZNF genes with orthologs retained in all 12 annotated Drosophila species’ genomes, we found that 14 are essential for either fertility or viability in D. melanogaster, whereas 42 genes are not (five genes have no phenotypic data available). In comparison, we found that 8 of the 30 genes not universally conserved in Drosophila species are essential, whereas 21 are not (one has no phenotypic data available). Thus, somewhat surprisingly, we find that genes not globally retained over Drosophila evolution are just as likely to encode a necessary function in D. melanogaster as genes that have been strictly maintained over 40 million years of Drosophila evolution (8:21 versus 14:42, p=0.8, two-tailed Fisher’s exact test) (Table 1). These findings further support the idea that gene families subject to rapid evolutionary turnover may become involved in essential functions.

Gene duplication and loss is only one form of genetic innovation. We also analyzed the D. melanogaster ZAD-ZNF genes for signatures of recent positive selection. We performed McDonald-Kreitman tests of ZAD-ZNF orthologs found in both D. melanogaster and the closely-related species, D. simulans. The McDonald-Kreitman test assesses sequence diversity within a species versus divergence between species by comparing the ratio of non-synonymous (amino-acid altering, or replacement) to synonymous substitutions fixed during the divergence of the two species (Dn: Ds), to that of non-synonymous to synonymous polymorphisms within a species (Pn: Ps) (McDonald and Kreitman, 1991). The Pn: Ps ratio is a proxy for functional constraint acting on a gene within species and is expected to be similar to Dn: Ds between species under the null hypothesis. However, a higher than expected number of fixed non-synonymous changes would indicate the action of adaptive evolution during species divergence (McDonald and Kreitman, 1991).

We took advantage of previous efforts that sequenced the genomes of hundreds of D. melanogaster strains and a reference D. simulans strain (Lack et al., 2016; Langley et al., 2012) to perform the McDonald-Kreitman test using the Popfly server (popfly.uab.cat) (Hervas et al., 2017). Of the 91 D. melanogaster ZAD-ZNF genes, only CG2202 is absent in D. simulans. We found that 12 out of the remaining 90 ZAD-ZNFs show evidence for recent adaptive evolution, i.e., have an excess of fixed non-synonymous changes (Dn) (Table 2). Using D. yakuba as an outgroup species, we also polarized fixed differences between D. melanogaster and D. simulans to assess whether the lineage leading to D. melanogaster showed evidence of positive selection in these 12 ZAD-ZNF genes. We found that 5 out of 12 genes showed evidence of positive selection with this polarized McDonald-Kreitman test (Table 2).

Subsequently, we used a domain-restricted analysis to define which of the three protein domains (ZAD, linker, or ZNF) were subject to positive selection. Although we found evidence of domain-specific positive selection in eight cases (two in the case of the polarized McDonald-Kreitman test), we were unable to define the domain subject to positive selection in four cases, due to small numbers of intra-species polymorphisms (Table 2). One gene (CG7386) showed signatures of adaptive evolution in its ZAD domain, whereas three genes (CG2712, CG7386, CG10321) showed evidence of adaptive evolution in the ZNF domain. Unexpectedly, we found that the linker domain has evolved under positive selection in six ZAD-ZNF genes. The biochemical function of these linker domains is largely unexplored as they lack predicted structural motifs (Chung et al., 2002; Chung et al., 2007), are highly variable in length and sequence even between orthologs, and are predicted to encode intrinsically disordered domains (Ishida and Kinoshita, 2007). Our findings implicate the poorly characterized linker region in mediating the adaptive potential of several ZAD-ZNF genes.

Using this signature of positive selection in a subset of ZAD-ZNF genes, we evaluated whether positively selected genes encode essential functions. Intriguingly, we found that 8 of 12 ZAD-ZNF genes that have evolved under positive selection are essential, whereas only 14 of the remaining 73 genes that have been phenotypically assayed are essential (six have no phenotypic data available). Thus, contrary to the dogma, we find that ZAD-ZNF genes that are subject to positive selection are more likely to be essential for viability or fertility (8:4 versus 14:59, two-tailed Fisher’s exact test, p=0.0016) (Table 1). Our findings not only imply that several essential ZAD-ZNF genes evolve rapidly, but also raise the possibility that rapid evolution of some ZAD-ZNF genes might be critical for organismal viability.

To further investigate the biological basis for the correlation between positive selection and gene essentiality, we decided to focus on one cluster of ZAD-ZNF genes on chromosome 3 of D. melanogaster (Figure 2A). This cluster of five genes contains two of the eight positively-selected, essential genes in D. melanogaster: Oddjob (Odj, CG7357) and CG17802; both also show evidence of positive selection in the polarized McDonald-Kreitman test (Table 2). Odj is the best-characterized gene in this cluster. The Oddjob protein interacts with Heterochromatin Protein 1a (HP1a), dynamically localizes to heterochromatin, and is required for heterochromatin-mediated gene silencing, or position-effect variegation (Swenson et al., 2016). Knockdown or over-expression of Odj is lethal in D. melanogaster, although the functional basis for this lethality is still unknown (Schertel et al., 2015). In keeping with the Oddjob nomenclature theme of ‘James Bond henchmen,’ and the mutant phenotype (described below), we renamed CG17802 as Nicknack, or Nnk. This cluster of ZAD-ZNF genes also includes three genes that do not evolve under positive selection (CG17801, CG17806, and CG17803) in D. melanogaster (Figure 2A). Although previous studies found that the knockdown of CG17801 in the germline led to de-repression of HetA and Blood transposable elements in the ovary, CG17801 knockdown did not significantly impair fertility or viability (Czech et al., 2013). The other two ZAD-ZNF genes in this cluster (CG17803, CG17806) have not been previously investigated.

We evaluated the functional roles of all five ZAD-ZNF genes in the Odj-Nnk cluster by generating knockdowns using an Actin5C-Gal4 driver and RNAi constructs specific to each gene (Figure 2B) to generate ubiquitous knockdown of individual genes in the resulting zygotes. Consistent with previous results (Schertel et al., 2015), we found that the ubiquitous knockdown of Odj or Nnk is lethal. In contrast, we found that knockdowns of CG17801, CG17803 and CG17806 are viable (Figure 2B). Based on these results, we conclude that the two positively-selected members of this ZAD-ZNF cluster – Odj and Nnk – are both essential, confirming the unexpected correlation we previously observed between positive selection and gene essentiality in the ZAD-ZNF gene family (Table 1B).

To gain deeper insight into the evolutionary dynamics of the Odj-Nnk cluster, we identified orthologs of these genes using reciprocal TBLASTN searches with each of the five D. melanogaster genes as queries. We searched both the originally-sequenced, well-annotated 12 Drosophila genomes (Drosophila 12 Genomes Consortium, 2007), as well as eight additional genomes that were subsequently sequenced to sample the melanogaster group within the Sophophora subgenus more densely (Chen et al., 2014). In several cases, we were not able to confidently assign the TBLASTN hits to orthologous groups because they matched closely to more than one D. melanogaster gene, or there were several putative hits within a single genome, or because the hit contained only the ZAD or ZNF domain.

To more rigorously identify orthologs, we conducted phylogenetic analyses (Figure 2—figure supplement 1) using a multiple alignment of the ZNF domain (Supplementary file 2). Our phylogenomic analyses reveal that Odj-Nnk cluster evolution was highly dynamic during the evolution of the Drosophila genus (summarized in Figure 2C). Although Oddjob orthologs are present throughout 40 million years of Drosophila evolution, no other member of the D. melanogaster Odj-Nnk cluster is universally present in Drosophila species. In addition to Odj, CG17801 and CG17806 also date back prior to the origin of Drosophila but unlike Odj, CG17801 and CG17806 have since been lost in some species (Figure 2C). While CG17801 has been lost in the obscura group,CG17806 underwent multiple independent duplication and loss events. CG17803 arose in the ancestor of D. melanogaster and D. ananassae and underwent two independent losses. Finally, Nnk appears to have arisen in the ancestor of D. melanogaster and D. pseudoobscura (~30 mya, Figure 1), and later experienced multiple independent duplications and losses (Figure 2C). We note that our estimates for the age of Nnk are higher than those reported previously (Chen et al., 2010), likely because the prior estimate was based on fewer sequenced species. Based on these analyses, we conclude that, despite its essential function in D. melanogaster, Nnk is not universally conserved in Drosophila species.

Nnk has a dramatic evolutionary history: young evolutionary age, differential retention, positive selection. Yet, it serves an essential function in D. melanogaster based on our and other previous analyses. However, this claim of essentiality has been challenged by two previous findings. First, the screen that originally identified Nnk as an essential gene used the KK RNA-interference (RNAi) collection from the Vienna Drosophila Stock Center (VDRC) (Figure 3A; Chen et al., 2010). Many lines in this ‘KK’ collection were later found to harbor a second-site mutation that caused lethality as a result of ectopic tiptop expression when crossed to GAL4-driver lines (Green et al., 2014; Vissers et al., 2016). As a result of its dependence on the KK lines, the claim of Nnk essentiality remained ambiguous. A second Nnk mutant allele, created by CRISPR-Cas9-mediated mutagenesis (hereafter referred to as CRISPR-null), had a four base pair deletion within the coding sequence of the gene that created a frameshift and a premature stop codon within the linker domain of Nnk (Kondo et al., 2017; Figure 3A). Although this deletion was homozygous lethal, it was unexpectedly viable when paired with a deficiency covering this ZAD-ZNF cluster (Kondo et al., 2017), again challenging the result that Nnk is an essential gene.

Given these ambiguous results and the importance of Nnk for our claims of ZAD-ZNF essentiality despite evolutionary innovation, we re-investigated whether Nnk is essential for viability in D. melanogaster. We found several lines of evidence to support the conclusion that it is indeed essential (Figure 3B–D). First, we began by validating the RNAi line used in the original study (Chen et al., 2010), which first identified Nnk as a young, essential gene. We found that the Nnk VDRC RNAi line does not have an insertion upstream of the tiptop gene, which is associated with lethality in other KK lines (Green et al., 2014; Vissers et al., 2016). Second, we were able to rescue this lethality via complementation with an intact Nnk-mel rescue transgene. This Nnk-mel rescue transgene was flanked by regulatory sequences from the endogenous locus (1 kb segments upstream and downstream of endogenous Nnk) and recoded via mutations in synonymous sites to make it resistant to RNAi knockdown (Figure 3B; Supplementary file 4). Endogenous levels of Nnk expression are too low for us to validate expression of the transgene. However, our ability to rescue Nnk knockdown-mediated inviability strongly imply that the recoded transgene is appropriately expressed (Figure 3B).

To rule out any indirect effects arising from RNAi knockdown, we also examined two previously generated mutant alleles of Nnk. The first is a piggyBac insertion two bp upstream of the start codon in the 5’ UTR of Nnk (Schuldiner et al., 2008; Figure 3C). This piggyBac insertion, which is marked by a DsRed reporter driven by an eye-specific promoter, is flanked by stop codons in all three reading frames, which prevents translation downstream of the insertion. We refer to this insertion as a piggyBac-null allele of Nnk. We found that this insertion allele is homozygous lethal but viability can be fully rescued upon mobilization of the piggyBac element, which repairs the intact 5’ UTR in a ‘scarless’ fashion to restore Nnk function (schematized in Figure 3C). Furthermore, piggyBac-null flies can also be rescued by the Nnk-rescue transgene (Figure 3D; Figure 3—source data 1). Second, we re-examined the previously-generated CRISPR-null Nnk allele (Kondo et al., 2017; Figure 3A). Contrary to previous results, we found that this allele is lethal when paired with a Nnk-spanning deficiency (BL9207, Figure 3D) and also fails to complement the piggyBac-null allele (Figure 3D; Figure 3—source data 1; Supplementary file 4). Moreover, the Nnk-rescue transgene can restore the viability of the CRISPR-null Nnk allele (Figure 3D; Figure 3—source data 1). Based on all these results, we conclude that Nnk is unambiguously an essential gene in D. melanogaster.

Next, we investigated the developmental stage at which Nnk-null progeny die. For this, we crossed Nnk-null heterozygotes to each other (Figure 4A). These flies contain the piggyBac insertion upstream of the Nnk gene (Figure 3C) on one chromosome, along with a balancer chromosome (TM3G) marked with GFP and carrying a wildtype Nnk allele. Progeny homozygous for the balancer chromosome TM3G die as early embryos. Consequently, all larvae lacking GFP expression are Nnk-null homozygotes whereas those that are heterozygous express the GFP encoded on the balancer. We conducted egg-lay experiments from crosses between heterozygote Nnk-null flies and tracked the developmental progression of GFP-negative, Nnk-null progeny relative to their GFP-expressing heterozygote siblings. We found that Nnk-null progeny progress through embryogenesis at the same rate as their heterozygote siblings and are morphologically indistinguishable from heterozygotes until the L1 larval stage. The Nnk-null larvae are able to move toward and consume yeast paste much like their heterozygote siblings. However, when heterozygote siblings molt into the L2 stage 48 hr after egg laying (AEL), Nnk-null larvae do not molt (Figure 4B). Instead, 48 hr AEL, Nnk-null larvae progressively become unable to move or eat, eventually dying by 60 hr AEL.

The Nnk-null larval arrest phenotype is reminiscent of previous findings with the ZAD-ZNF genes Séance, Ouija Board and Molting Defective, which encode proteins necessary for expression of the Spookier and Neverland genes required for ecdysone biosynthesis (Uryu et al., 2018; Neubueser et al., 2005; Komura-Kawa et al., 2015). Defects in any of the Séance, Ouija Board and Molting Defective ZAD-ZNF genes leads to arrest of larval development and death (Uryu et al., 2018; Neubueser et al., 2005; Komura-Kawa et al., 2015). However, this lethality can be rescued by supplementing the diet with ecdysone or by overexpression of ecdysone biosynthetic enzymes, bypassing the requirement for these three ZAD-ZNF genes. We therefore, tested whether dietary supplementation with an ecdysone precursor, cholesterol (7DH, or 7-Dehydrocholesterol) or ecdysone (20E, or 20-Hydroxyecdysone) could bypass or significantly delay the death of Nnk-null larvae (Figure 4C). We found that it could not. In contrast, the same dietary supplementation is able to significantly restore the viability of Npc1a larvae, which lack an essential transporter of cholesterol (Figure 4C). Based on these results, we conclude that Nnk plays an essential role in larval progression in D. melanogaster that is distinct from known steps of the ecdysone biosynthesis pathway.

To further investigate our developmental findings, we also performed RNA-seq analyses, comparing the transcriptomes of Nnk-null larvae to their age-matched heterozygote siblings. We compared the two genotypes at each of two important time-points. First, we compared transcriptomes 24 hr AEL, when the two genotypes are morphologically indistinguishable L1 larvae. At this timepoint, we find that 249 genes (2.4% of all expressed genes) are differentially expressed between Nnk-null larvae and control larvae, with 116 genes at least two-fold upregulated and 133 genes two-fold downregulated in Nnk-null larvae (Figure 4D). Among genes downregulated in Nnk-null L1 larvae, we find a significant over-representation of functional categories related to proteolysis and sterol transport (both important functions during larval development), as well as dopamine monooxygenase activity (Supplementary file 3). In contrast, we find that genes related to lysosome, cytochrome P450s (also important for larval molts), and eye-related functions are upregulated upon Nnk loss.

Second, we performed comparisons 48 hr AEL when the Nnk-null larvae are significantly smaller and appear developmentally arrested compared to their age-matched controls. At this timepoint, we find that 3027 (28.1% of all expressed genes) genes are differentially expressed, with 1301 genes at least two-fold upregulated and 1726 genes two-fold downregulated in the Nnk-null mutants compared to the age-matched controls (Figure 4E). Thus, there are significantly more genes affected by Nnk loss by 48 hr AEL. Intriguingly, clustering samples by the transcriptional profile of all genes shows that Nnk-null larvae at 48 hr AEL are transcriptionally more similar to control larvae at 24 hr AEL of either genotype than they are to age-matched control larvae (Figure 4F). The transcriptional status of Nnk-null larvae therefore mirrors the phenotypes we observe, displaying a severe developmentally arrested phenotype and transcriptional profile at 48 hr AEL (Figure 4B).

Most of the Odj-Nnk cluster of ZAD-ZNF genes are functionally uncharacterized. Since Odj encodes a protein that is highly enriched in pericentric heterochromatin (Swenson et al., 2016), we speculated that its close paralog, Nnk, might also encode a protein with heterochromatic localization in D. melanogaster cells. To test this possibility, we used transient transfections to introduce epitope-tagged Odj and Nnk genes into D. melanogaster Schneider 2 (S2) cells and induced their expression with a heat-shock promoter (Figure 5A). Upon induction, we confirm that Oddjob has a broad localization pattern within heterochromatin (marked by histone H3 lysine nine methylation, H3K9me3 and outlined with a dashed line, Figure 5A). We found that the Nnk-encoded protein also localizes to heterochromatin, but its localization is restricted to discrete foci, unlike Oddjob (Figure 5A). Since heat-shock induction can alter chromatin properties in cells, we also employed a complementary transient transfection strategy in which we expressed mCherry epitope-tagged Odj and Nnk genes under the control of a constitutive pCopia promoter in S2 cells. These analyses also revealed a broad heterochromatic localization of Odj contrasting with discrete foci within heterochromatin for Nnk protein (Figure 5B). These foci do not overlap with centromeres (identified by the centromeric histone Cid) or dual-strand piRNA clusters (marked by the piRNA-binding HP1 protein Rhino) (Figure 5—figure supplement 1).

Odj has a broad heterochromatic localization in D. melanogaster cells, which could result from direct interaction with HP1a (Swenson et al., 2016). Indeed, Odj has two potential PxVxL motifs, which are putative interaction sites for HP1a (Smothers and Henikoff, 2000), in the linker and ZNF domains (Figure 5—figure supplement 2A). Investigating Odj orthologs in other Drosophila species revealed that the PxVxL motif in the linker domain is well-conserved, but the one in the ZNF domain is not. We found that mutation of the putative HP1a-interaction site in the linker domain (V164A) converted Odj localization from a broad to a discrete pattern that at least partially overlapped with Nnk (Figure 5—figure supplement 2B). In contrast, mutation of the second putative PxVxL site (V321A) did not significantly affect Odj localization. Our results suggest that the putative PxVxL motif in the Odj linker region is a major contributor to Oddjob’s broad localization to heterochromatin potentially by mediating a direct interaction with HP1a. In the absence of this interaction, Oddjob and Nnk (which lacks a canonical PxVxL motif) localize similarly to discrete foci within heterochromatin (Figure 5B). Our findings suggest that altered protein–protein interactions or DNA-binding specificity via the linker domain may provide a means for functional diversification between closely-related ZAD-ZNF paralogs.

To gain deeper insight into the heterochromatic localization of Nnk and Odj, we performed chromatin immunoprecipitation and sequencing (ChIP-seq) using transient transfection of S2 cells with the epitope-tagged pCopia constructs for Odj and Nnk. Consistent with its cytological localization (Figure 5A and B), we found that Odj was highly enriched throughout the large chromosome regions previously defined as pericentric heterochromatin (Hoskins et al., 2015) whereas Nnk-bound regions are more narrowly distributed within heterochromatin (Figure 5C). Finer-scale analyses suggested that Nnk and Odj signals are enriched close to transcription start sites classified by modENCODE as residing in TSS-proximal chromatin in S2 cells, both in heterochromatin and euchromatin (Figure 5—figure supplement 3; Ho et al., 2014). However, many of these regions overlap with false positive ‘phantom peaks’ previously identified in ChIP-seq experiments in S2 cells (Jain et al., 2015; Figure 5—figure supplement 3). While some heterochromatic TSSs could still represent true binding, we are more confident that the broader occupancy we observe outside TSSs represents Odj and Nnk’s true localization. Moreover, we did not observe a significant effect of Nnk loss on heterochromatin-embedded gene expression at the larval L1 stage via RNA-seq analyses (Figure 4D, Figure 5—figure supplement 4). Instead, we found that some heterochromatin-embedded repetitive elements are derepressed upon Nnk loss (Figure 5—figure supplement 4). Based on these results, we hypothesize that rather than acting as a transcription regulator for specific heterochromatin-embedded genes like its paralogs Séance, Ouija Board and Molting Defective (Uryu et al., 2018), Nnk may instead repress expression from heterochromatic repeats.

Our analyses revealed Nnk is an evolutionarily young, differentially-retained gene that is essential for larval development in D. melanogaster. Nnk has also evolved under dramatic positive selection in just the 2.5 million-year divergence between D. melanogaster and D. simulans, having undergone 52 fixed non-synonymous differences in the 439 aa protein-coding region (Table 2). Given the intriguing correlation between essentiality and positive selection that we had observed in the ZAD-ZNF genes, we investigated whether positive selection of Nnk has affected its function.

We first assayed the subcellular localization of epitope-tagged Nnk orthologs from D. melanogaster and D. simulans in D. melanogaster S2 cells. We used transient transfections to introduce these genes into S2 cells and used heat shock to drive their expression. We found that both proteins similarly localize to foci within heterochromatin (Figure 6A–B; dashed line marks heterochromatin boundaries). Thus, the rapid evolution of Nnk during D. melanogaster- D. simulans divergence has not dramatically affected its gross subcellular localization.

Next, we examined the consequences of Nnk positive selection on viability in D. melanogaster. We created a D. simulans Nnk ‘rescue’ transgene (Figure 6C). This rescue transgene is similar to the D. melanogaster Nnk-rescue construct (Figure 3B), except that it contains the D. simulans Nnk coding sequence (codon-optimized to D. melanogaster) with 1 kb Nnk-flanking sequences from D. melanogaster (Supplementary file 4). We introduced this D. simulans Nnk-rescue transgene into the same attP site on D. melanogaster X chromosome as the D. melanogaster Nnk-rescue transgene via PhiC31-mediated transgenesis (Figure 6D). The attP-insertion rescue design allowed us to put the transgene in the same genetic location in the Nnk-mel and Nnk-sim rescue crosses, normalizing for variability in expression of transgenes. This allowed a near-isogenic comparison of the D. simulans and D. melanogaster Nnk transgenes’ ability to rescue inviability of Nnk-null D. melanogaster flies, despite their high level of sequence divergence. Unfortunately, low levels of endogenous Nnk expression did not allow us to assess whether the expression levels of both Nnk transgenes were equivalent.

We crossed heterozygous females either containing one or no copy of the (RNAi-resistant) Nnk-rescue transgene and the Nnk-RNAi allele to Act5C-GAL4/CyO-GFP males. We expected to find that progeny that did not inherit the rescue transgene would be inviable. In contrast, in the resulting progeny from rescue transgene-bearing females, we expect half of the progeny to inherit the Nnk-rescue transgene (see Materials and methods). We found that the Nnk-mel transgene significantly rescued Nnk knockdown compared to no-transgene controls; however, males were recovered at slightly lower levels than females (67 males: 101 females compared to expectation of 1:1 ratio; p=0.08, Fisher’s exact test). We similarly found that the D. simulans Nnk transgene can significantly rescue the lethality caused by knockdown of endogenous D. melanogaster Nnk in females (Figure 6C), although at a slightly lower level than Nnk-mel rescue. In contrast, rescue of male viability by D. simulans Nnk is extremely poor (2 males: 33 females compared to expectation of 1:1 ratio; p<0.0001, Fisher’s exact test) resulting in a severe sex-bias. Thus, Nnk-sim is much worse than Nnk-mel in rescuing male viability (67:101 versus 2:33, p<0.0001, Fisher’s exact test). Based on these findings, we infer that the D. simulans Nnk-rescue transgene specifically fails to rescue Nnk-knockout males, in the presence of the heterochromatin-rich Y chromosome. Our findings suggest that not only is Nnk a positively-selected, essential ZAD-ZNF gene, but also that its positive selection is required for optimal function in the D. melanogaster genome.

We used the Flybase database (http://flybase.org) to identify all ZAD-containing proteins (Pfam motif PF07776) in 12 sequenced and annotated Drosophila species (Drosophila 12 Genomes Consortium, 2007). Using NCBI’s Conserved Domains search, we identified other domains found in the ZAD-containing proteins (Marchler-Bauer et al., 2017). To estimate the evolutionary age of these ZAD-containing genes, we used OrthoDB to identify orthologs of the 91 ZAD-containing proteins across Drosophila (Zdobnov et al., 2017). We used the divergence between the two most distantly-related species that still encoded and ortholog to calculate the minimum age of each gene using timetree.org (Kumar et al., 2017). If orthologs were identified in basally-branching but not later-branching species by OrthoDB, we defined this as a loss of the ZAD-ZNF gene in the lineage. The repertoire of different ZAD-ZNF genes in different Drosophila species inferred from our analysis is summarized in Supplementary file 1.

We used FlyBase gene summaries (FlyBase.org) and published studies when available to define gene essentiality. Our criteria for essentiality were broad: if there was a lethal allele reported in any assay, we counted the gene as essential for viability in D. melanogaster. Supplementary file 1 summarizes the phenotypic data (and its source) that we used to classify the ZAD-ZNF genes into either essential or non-essential categories.

For the McDonald-Kreitman test, we extracted the gene of interest from D. melanogaster population genetic datasets available through Popfly (www.popfly.com) and removed low frequency (<0.05) variants from the dataset to minimize the effects of false positives and low-frequency variants that may not have been subject to selection (Hervas et al., 2017; Fay et al., 2002). We used a manually trimmed alignment of the D. melanogaster filtered dataset and the reference D. simulans sequence for the McDonald-Kreitman test (http://mkt.uab.es/mkt/; McDonald and Kreitman, 1991; Egea et al., 2008). We also carried out a polarized McDonald-Kreitman test in which we only analyzed non-synonymous and synonymous sites that were fixed along the lineage leading to D. melanogaster, after its divergence from D. simulans, inferred using D. yakuba as an outgroup species, except for the Trem gene where we used D. erecta outgroup instead because the D. yakuba ortholog aligned poorly to D. melanogaster and D. simulans.

We used an Act-GAL4/CyO-GFP driver for ubiquitous knockdown. The RNAi lines used to specifically target Oddjob cluster genes are the VDRC KK or GD lines: Oddjob (27971), CG17803 (38869), CG17801 (29501), CG17806 (40106) and Nicknack (102311). RNAi controls used for the experiment were Cid (43856) and HP1B (26097) for the ubiquitous knockdown. Ubiquitous knockdown of Cid produced no viable progeny. We crossed five virgin females carrying Act-GAL4/CyO-GFP to 3 males of each RNAi line. We allowed the females to lay eggs for 3 days and flipped the flies into fresh vials three times. Each cross was performed in triplicate. Progeny were counted 10–15 days after each cross was set up. A minimum of 20 CyO-GFP males and 20 CyO-GFP females (i.e., control genotype) were required for us to quantify the crosses. Each cross is represented by a point on the graph and shown as a ratio of observed (number of knockdown or rescue progeny counted in each cross) over expected (the number of expected knockdown or rescue progeny from Mendelian segregation inferred by inheritance of the balancer chromosome). If the knockdown has no effect on viability or the rescue is 100% effective, the observed/expected = 1. If the knockdown of the essential gene is 100% penetrant or the rescue is ineffective, the observed/expected = 0. Plotting and statistical analyses were conducted using Graphpad Prism 8 software.

Since Oddjob cluster genes experienced numerous independent segmental duplications, it was not possible to determine orthologs by synteny alone. Instead, we used TBLASTN (Gertz et al., 2006) to identify candidate orthologs of Oddjob cluster genes, using the genes in D. melanogaster as queries. We used a reciprocal blast search strategy to identify potential orthologs and further investigated these candidates by making a maximum likelihood phylogenetic tree (LG substitution model in PhyML with 100 bootstrap replicates) of a manually trimmed protein alignment (constructed using Clustal Omega program [Thompson et al., 2002] in the Geneious package, www.geneious.com) (Supplementary file 2, Figure 2—figure supplement 1). We assigned orthologs based on genes that formed a monophyletic clade with the each of the D. melanogaster Oddjob cluster genes. We mapped these Oddjob cluster orthologs back to each genome assembly to determine the composition of the Oddjob locus across Drosophila species. In cases where there were other ZAD-ZNFs present, we blasted them against the D. melanogaster genome to examine if there were any orthologs present in D. melanogaster. If the top hit was not a member of the Oddjob locus, we did not include it in the tree. In the case where there are partial ZAD-ZNFs (containing just the ZAD domain or the zinc-finger domains), we performed a blast search against the D. melanogaster genome. If the top hit was a member of the Oddjob cluster, we assigned orthology by making a phylogenetic tree with all other Oddjob cluster orthologs. Our phylogenomic inferences are dependent on the accuracy and state of completion of genomes from other Drosophila species, which can vary.

We designed a D. melanogaster recoded transgene comprising a 3.3 kb fragment containing the genomic region of Nicknack plus 1 kb upstream and downstream of the start and stop codons, based on the D. melanogaster reference assembly. We recoded the sequence targeted by the VDRC RNAi KK line (103211) by making synonymous changes at each codon. The synonymous changes made the gene resistant to RNAi knockdown but did not alter the amino-acid sequence of Nnk. The recoded region spans 270 base pairs that corresponds to the region targeted by the VDRC RNAi KK line (103211). The resulting sequence was synthesized by GENEWIZ Co. Ltd. (Suzhou, China) and cloned into a plasmid we generated that contains 3xP3-DsRed attP, which produces fluorescent red eyes in the adult to mark the presence of the transgene. To generate the D. simulans transgenic allele, we codon-optimized the D. simulans Nnk coding sequence for the D. melanogaster genome using IDT’s codon-optimization tool. The resulting sequence was synthesized by GENEWIZ Co. Ltd. (Suzhou, China) and swapped for the D. melanogaster Nnk coding sequence in the plasmid described above, using the NEBuilder kit (New England Biolabs). We submitted transgenic constructs to The BestGene Inc (Chino Hills, CA) for injection into the X-chromosome attP18 line (BL 32107) using PhiC31 site-specific integration (Groth et al., 2004).

For the transgene rescue cross, we crossed five virgin female flies bearing one copy of Nnk-rescue transgene (on the X chromosome) and the RNAi allele (on the second chromosome) to 3 Act5C-GAL4/CyO-GFP males. We allowed the females to lay eggs for 3 days and flipped the cross three times. We set up each cross in triplicate and progeny were counted after 10–15 days. The description of the observed/expected calculation can be found in ‘viability studies’ section. Each cross had at least three replicates. All flies were raised at 25°C.

We placed 50–75 flies heterozygous for the Nnk-null allele (Nnk pBac null/TM3G) or for the Nnk CRISPR allele (Nnk CRISPR/TM3G) into a small embryo collection cage containing a grape-juice plate with a thin strip of yeast paste and collected embryos for 3 hr at 25°C. We transferred the larvae to fresh grape-juice plates containing yeast paste daily and scored developmental stage by mouth hook morphology. We used fluorescence to distinguish between heterozygotes (GFP-positive larvae) and homozygotes (GFP-negative larvae). For the trans-heterozygote evaluation, we crossed 30–40 virgin female Nnk CRISPR/TM3G to 10 Nnk pBac null/TM3G males. Crosses were done in triplicate and at least 100 progeny were counted per cross.

We placed 50–75 flies heterozygous for the Nnk-null allele (Nnk pBac/TM3G) into a small embryo collection cage containing a grape-juice plate with a thin strip of yeast paste and collected embryos for 3 hr at 25°C. The first time point was collected 24 hr after egg laying (AEL) and the second 48 AEL. We transferred the larvae to fresh grape-juice plates containing yeast paste daily.

Whole larvae (~30 animals at 24 hr AEL and ~20 animals at 48 hr AEL for each sample; RNA from each time point and genotype was prepared in triplicate) were ground in a 1.5 mL Eppendorf tube containing 50 μL of TRIzol reagent using a DNase, RNase and DNA free 1.5 mL pestle. 450 μL of TRIzol reagent was added after grinding. Immediately, we added 500 μL of chloroform and the tube was inverted gently 2–3 times. We removed the aqueous phase into a fresh tube containing 1 mL of 200 proof EtOH and mixed by inversion. The mixture was then bound to a Zymo-spin column according to the manufacturer's instructions (Zymo Research). We followed the DNase extraction and purification protocol outlined in the RNA Clean and Concentrator kit (Zymo Research). We eluted the RNA in 15 μL of DNase/RNase-free water and immediately placed the samples at −80°C. We checked the quality of the samples with a 2200 Tapestation (Agilent Technologies) and selected samples that had an RNA integrity number >9.0 for library preparation. Library construction and Illumina 150 bp paired-end RNA-sequencing were conducted at Novogene Bioinformatics Technology Co., Ltd (Beijing, China).

We used Kallisto (Bray et al., 2016) to quantify abundances of the D. melanogaster reference transcriptome (refMrna.fa for dm6, obtained from UCSC Genome Browser Oct 16^th, 2018, which contains 34,114 transcripts). For each transcript, we acquired the gene name using R (org.Dm.eg.db). Kallisto counts were read into R using the tximport package (Soneson et al., 2015) and using the summarizeToGene function, we summarized alternative splice-form counts into a single count per gene. We used DESeq2 to identify differentially expressed genes with adjusted p-value<=0.05 and absolute log₂(fold change)>=1 (Anders and Huber, 2010), comparing Nnk-null larvae to controls for each timepoint separately. Before performing each comparison, we excluded genes with low expression (<100 counts total across all samples): this filtering yields 10,574 ‘expressed’ genes in 24 hr AEL larvae, and 10,758 ‘expressed’ genes in 48 hr AEL larvae. The RNA-seq data has been deposited in the SRA database with the accession number PRJNA643855.

We performed Gene Ontology (GO) enrichment analysis on each of four gene lists: 116 over-expressed genes and 133 under-expressed genes in 24 hr AEL larvae, and 1301 over-expressed genes and 1726 under-expressed genes in 48 hr AEL larvae. We used the Bioconductor GOstats package to perform conditional hypergeometric enrichment tests for each of three ontologies (biological process, molecular function and cellular component) (Huber et al., 2015; Falcon and Gentleman, 2007). For the ‘universe’ of all genes examined (i.e., background) we used the corresponding list of ‘expressed’ genes at each developmental timepoint. We report only over-represented categories with p<0.001, and use annotations found in the org.Dm.eg.db Bioconductor package.

To estimate expression levels aggregated across all instances of each repeat type, we took an approach similar to that of Day et al., 2010. Here, rather than mapping reads to the typical reference genome assembly, we constructed a ‘repeat assembly’, where we used RepeatMasker annotations for dm6 (obtained from UCSC) to extract and concatenate all instances of each repeat type, adding 75 bp flanking sequences (half the length of each read) and inserting 150 N bases between each instance. The repeat assembly therefore consists of a single ‘chromosome’ for each repeat type. We then used BWA-mem to map sequences as single reads (not paired-ends) to the repeat assembly (Li, 2013), and samtools to count reads mapping to each repeat type (‘chromosome’). We combined counts for pairs of simple repeats that represent the reverse-complement of one another. We used DESeq2 to perform statistical analyses on raw counts, using the total number of sequenced fragments as size factors (Love et al., 2014). We normalized counts by dividing each by the number of reads sequenced for that sample in millions.

To evaluate the ability of dietary sterols to rescue Nnk-null phenotype, we carried out sterol supplementation as previously outlined (Uryu et al., 2018). Briefly, we mixed together 20 mg of dry yeast in 38 µL of water. To this yeast paste, we added either 2 µL of EtOH (negative control) or 2 µL of EtOH plus cholesterol (Sigma), 7-dehydrocholestrol (Sigma) or 20-hydroxyecdysone. Stocks used for these experiments were balanced over GFP-balancers (Npc1a57/CyO-GFP and Nnk-null/TM3G). The control used for these experiments was the w1118 stock. Eggs were laid on yeasted grape-juice plates for 3 hr at 25°C. 24 hr AEL, GFP-negative larvae were transferred onto grape plates containing fresh yeast paste at 25°C. For 72 hr AEL, larvae were transferred to fresh grape-juice plates containing yeast paste daily and scored for their developmental stage based on the morphology of their mouth hooks. Npc1a57/CyO-GFP stocks were a kind gift from Leo Pallanck (Fluegel et al., 2006).

Oddjob cluster genes were amplified from genomic DNA from D. melanogaster and D. simulans and directionally cloned into pENTER/D-TOPO (ThermoFisher) according to the manufacturer’s instructions. We verified that clones had the appropriate insertions by sequencing. We used LR clonase II (ThermoFisher) to get each Oddjob cluster gene into the Drosophila Gateway Vector destination vector to express the gene of interest with an N-terminal Venus tag under the control of the D. melanogaster Hsp70 promoter (pHVW). For Rhino localization in S2 cells, the Rhino gene was cloned into the Drosophila Gateway Vector destination vector to enable expression of Rhino with an N-terminal 3xFlag tag under the control of the D. melanogaster Hsp70 promoter (pHFW). Nnk and Odj were also cloned into pCopia vectors containing N-terminal and C-terminal mCherry epitope tags, respectively, and HP1a was previously tagged with mCerulean at the N-terminus.

Schneider 2 cells were obtained from the Drosophila Genomics Resource Center (Bloomington, IN, USA) and grown at 25°C in M3+BPYE+10%FCS. For the transfections, one million cells were seeded, and one day later 2 micrograms of plasmid DNA was transfected into cells using Xtremegene HP transfection reagent (Roche) according to the manufacturer’s specifications. For the Rhino localization experiment, cells were co-transfected with 1 µg of GFP-tagged ZAD-ZNF vector and 1 µg of Flag-tagged Rhino vector. Cells were allowed to recover for 24 hr post transfection, heat shocked for 1 hr at 37°C and recovered for 3 hr at 25°C prior to fixation. For the pCopia vectors, transient transfections were conducted on S2 cells using the TransIT‐2020 reagent (Mirus), and live imaging was performed 72 hr later, using an Applied Precision Deltavision microscope and analyzed using SoftworX software.

For the heatshock vector transfections, cells were transferred to coverslips for 30–45 min prior to starting the immunohistochemistry protocol. 0.5% sodium citrate hypotonic solution was added to the coverslip for 10 min to swell cells, which was then spun at 1900 rpm for 1 min in a Cytospin III to remove the cytoplasm from the cells. The sodium citrate was immediately removed and cells were subsequently fixed. For fixation, 4% PFA + PBST (PBS + 0.1% Triton) was added to the cells for 10 min. Coverslips were then washed in PBST and blocked for 30 min in PBST + 3% BSA. Cells were incubated with primary antibody overnight at 4°C in a humid chamber. For immunolocalization, the following dilutions of primary antibodies were used: GFP (Abcam AB13970) 1:1000, H3K9me3 (Abcam AB8898) 1:500 M2 FLAG (Sigma-Aldrich F4042) 1:1000. After washing with PBST three times for 10 min per wash, the following fluorescent secondary antibodies were used at 1:1000 dilution: goat anti-chicken (Invitrogen Alexa Fluor 488, A-11039), goat anti-rabbit (Invitrogen Alexa Fluor 568, A-11011) and goat anti-mouse (Invitrogen Alexa Fluor 568, A-11031). The cells were incubated with 1x DAPI in the final wash and mounted in SlowFade Gold Mounting Medium (ThermoFisher). We imaged cells on a Leica TCS SP5 II confocal microscope with LASAF software and images were processed using ImageJ and were representative of the cell population.

S2 cells transfected with either mCherry-Nnk or Odj-mCherry for 72 hr were fixed with 1% paraformaldehyde for 10 min and sheared with Bioruptor sonicator (Diagenode) to obtain chromatin. Chromatin fragments were confirmed to contain DNA in the 200–500 bp size range using a Bioanalyzer (Agilent). Each immunoprecipitation was performed on chromatin from 2 × 10⁷ S2 cells by overnight incubation with Protein-G Dynabeads and 5 µg of anti-mCherry antibody (Novus). Library construction from immunoprecipitated DNA was conducted using TruSeq sample preparation kits (Illumina). 150 bp paired-end sequences were generated by the Vincent J. Coates Genomics Sequencing Laboratory at UC Berkeley.

We used BWA-mem to map paired reads to version 6 of the D. melanogaster genome assembly (dm6) from which we had removed unplaced scaffolds (Li, 2013). We used deepTools’ bamCompare (with the ‘--binsize 1 --extendReads’ options) to obtain log2 ratios of fragment coverage for matched ChIP and input samples, and visualized those ratios in IGV (Robinson et al., 2011). We further visualized ChIP-seq signal around TSSs using deepTools’ computeMatrix and plotHeatmap tools, using TSS annotations obtained from the TxDb.Dmelanogaster.UCSC.dm6.ensGene BioConductor package, taking the most upstream TSS for genes with alternative start sites. We split TSS annotations according to whether they were within ‘TSS-proximal’ (active) chromatin according to modENCODE’s nine state annotation for S2 cells, obtained from http://intermine.modencode.org and converted from dm3 to dm6 coordinates using UCSC’s liftOver tool (Hinrichs et al., 2006). We further split the ‘active’ TSSs according to whether they are within cytogenetic heterochromatin, using coordinates from Supplementary file 2 of Hoskins et al., 2015 as well as the whole of chromosomes four and Y. The ChIP-seq data has been deposited in the SRA database with the accession number PRJNA644950.