Targeted substrate degradation by Kelch controls the actin cytoskeleton during ring canal expansion

Drosophila oogenesis provides a powerful model for investigating the specialized features of gamete development (Bastock and St Johnston, 2008). Female fruit flies produce up to 60 eggs per day, each ∼0.5 mm in length with a volume more than 1000-fold greater than that of a typical somatic cell. Oogenesis initiates at the anterior end of an ovariole, called the germarium, where a daughter of a germline stem cell undergoes four rounds of mitosis to generate a 16-cell cyst. One of these cells becomes the oocyte, while the remaining 15 differentiate as nurse cells. Cytokinesis does not complete during the mitotic divisions, leaving all 16 cells connected by the arrested cleavage furrows. The arrested cleavage furrows become stable intercellular bridges called ring canals through the recruitment of additional proteins (Haglund et al., 2011). Over the course of several days, the nurse cells become highly polyploid and synthesize mRNA, protein and organelles that support oocyte growth. These components move to the oocyte through the ring canals, which recruit a robust F-actin cytoskeleton and expand during oogenesis to accommodate the flux of materials.

An important mechanism for regulating developmental events during gametogenesis is the ubiquitin-proteasome system (UPS). Covalent attachment of ubiquitin to a target protein can alter the fate of the protein in a number of ways (Komander and Rape, 2012). Mono-ubiquitylation, the attachment of a single ubiquitin, can serve as a signal for endocytic sorting. When lysine 48-linked chains of ubiquitin are attached, the ubiquitylated protein is targeted for destruction by the proteasome. Protein ubiquitylation is carried out by three enzymes acting in sequence, termed E1, E2 and E3. Substrate specificity is conferred by the E3 enzyme, and several classes of E3 enzymes have been characterized. Genes encoding E3 enzymes have undergone significant expansion in higher eukaryotes such that 3% of human genes (>600 out of 20,000) encode substrate-specific E3 enzymes (Li et al., 2008).

BTB-BACK-Kelch (BBK) proteins function as substrate adaptors for a class of ubiquitin E3 ligases called Cullin 3-RING ubiquitin ligases (CRL3s) (Fig. 1A) (Xu et al., 2003). The elongated Cullin 3 (Cul3) protein binds a BTB-domain protein at its N terminus and a RING domain protein at its C terminus. The RING domain recruits an E2 ubiquitin-conjugating enzyme, whereas the BTB-domain protein recruits specific substrates for ubiquitylation through its C-terminal Kelch-repeat domain (KREP), which adopts a β-propeller structure (Adams et al., 2000; Li et al., 2004). The BBK gene family has undergone significant expansion during animal evolution; the Drosophila genome encodes ∼10 BBK proteins, whereas nearly 50 distinct BBK genes are present in humans (Dhanoa et al., 2013; Prag and Adams, 2003). Mutations in human BBK genes are associated with a number of diseases, including cancer, neurodegenerative disease, infertility and congenital skeletal myopathies, underscoring their clinical importance (Bomont et al., 2000; Gupta and Beggs, 2014; Padmanabhan et al., 2006; Yatsenko et al., 2006). A common thread suggested by these disease associations is BBK-mediated regulation of metazoan-specific cytoskeletal structures. Studying the targets of BBK/CRL3 ubiquitylation therefore requires access to experimental systems amenable to analysis of cytoskeletal regulation.

We have previously demonstrated that Drosophila Kelch functions with Cullin 3 to coordinate the growth of the ovarian ring canal cytoskeleton during oogenesis (Hudson and Cooley, 2010; Hudson et al., 2015). Here, we present evidence that HtsRC, one product of the hts gene, is the key substrate of CRL3^Kelch required for ordered ring canal expansion during egg chamber growth. We identified HtsRC as a potential substrate of CRL3^Kelch based on its association with the KREP domain, and we identified a candidate Kelch-binding site in HtsRC responsible for targeting by CRL3^Kelch. Disruption of this motif by gene editing resulted in a ring canal phenotype identical to kelch. We demonstrated that HtsRC can be ubiquitylated and degraded by the proteasome in cultured cells and in ovaries. Finally, we carried out a series of genetic interaction experiments that further support the conclusion that HtsRC is the substrate of CRL3^Kelch in ring canals.

Egg chambers from Drosophila females mutant for kelch have ring canals with a highly disorganized F-actin cytoskeleton that impedes the flow of cytoplasm to growing oocytes (Xue and Cooley, 1993). We have reported that Kelch functions as the substrate-targeting component of a Cullin-RING E3 ubiquitin ligase (CRL3^Kelch) and that its CRL activity is essential for ring canal cytoskeletal organization (Hudson and Cooley, 2010; Hudson et al., 2015). To understand the role of Kelch in ring canal morphogenesis, we sought to identify substrates of CRL3^Kelch. We have previously shown that overexpression of the KREP domain results in a kelch-like dominant-negative phenotype (Fig. 1C), marked by the accumulation of KREP at the ring canal (Hudson and Cooley, 2010). This phenotype suggested that the KREP domain might be bound to its substrate in a stable complex lacking Cul3 that cannot promote substrate ubiquitylation. We therefore used affinity purification of the KREP domain to identify potential substrates by mass spectrometry.

We expressed TAP-tagged mCherry::KREP (Fig. 1B) and a TAP-tagged mCherry control in ovarian germ cells. However, initial experiments were hindered by the poor solubility of ring canal components lysed under conditions that would preserve protein-protein interactions. We therefore performed purifications in a mutant background to liberate ring canal components from the insoluble cytoskeleton. cheerio (cher) encodes Drosophila Filamin (Sokol and Cooley, 1999). In cher mutants, ring canals lack the robust F-actin cytoskeleton found in wild type (Fig. 1C,D) and Kelch fails to properly localize despite being present in ovarian lysates (Robinson et al., 1997). We performed parallel purifications of mCherry or mCherry::KREP from cher mutant females (Fig. 1E) and identified co-purifying proteins by mass spectroscopy. From nearly 900 proteins identified in the KREP purification, we focused on 141 candidate substrates based on specificity of interaction and relative abundance (approximated by total spectral counts; see Table S1).

Sequence analysis of the candidate substrates revealed that they were enriched for proteins containing a short linear sequence motif consisting of several acidic residues followed by a consensus PEAEQ sequence (15 out of 141; P=8×10⁻⁸, binomial test) (Fig. 1F). A similar motif is also present in three human WNK kinases targeted by KLHL3, a human ortholog of Kelch (Boyden et al., 2012; Shibata et al., 2013). These results suggest that the PEAEQ motif is a binding site for Kelch and closely related BBK proteins. In further support of this idea, we identified four PEAEQ-containing proteins in a yeast two-hybrid screen for KREP domain interactors, and all cDNAs isolated contained the regions encoding the PEAEQ motif (Fig. S1).

Several PEAEQ proteins identified by mass spectrometry have known functions in F-actin assembly and organization, including Hts (Fig. 1F). The hts locus encodes multiple proteins, including conserved and ubiquitously expressed isoforms of the cytoskeletal protein Adducin (Fig. 1G) (Yue and Spradling, 1992). In the ovary, a germline-specific transcript designated ovhts encodes a polyprotein that is cleaved to produce two distinct products: an Adducin isoform that localizes to the fusome within mitotic cells (referred to as HtsF) and the novel Hts ring canal (HtsRC) polypeptide that localizes to germline ring canals (Fig. 1G) (Petrella et al., 2007). All of the Hts peptides identified by mass spectrometry mapped to the HtsRC part of Ovhts (Fig. 1H), suggesting that the ring canal protein, and not an Adducin isoform, was associated with the KREP domain. Sequence analysis of other Drosophila species revealed that the HtsRC PEAEQ sequence is invariant, whereas adjacent sequences show less conservation, suggesting a functional role for this sequence (Fig. S2). We therefore sought to determine whether HtsRC is a substrate of CRL3^Kelch.

To assess HtsRC ubiquitylation, we expressed HtsRC in cultured S2 cells by cloning exon 12 (see Fig. 1G) into an expression vector with HA epitope tags at both the N and C termini (Fig. 2A). We reasoned that transfected HtsRC should be regulated by CRL3^Kelch as all components of the CRL3^Kelch ubiquitin ligase machinery are present. S2 cells produced both 3xHA::HtsRC::HA precursor protein and the HtsRC::HA cleavage product with no N-terminal tag (Fig. 2A). Cells were treated with DMSO (control) or 1 µM bortezomib, a proteasome inhibitor, for 3 and 6 h. Whole-cell lysates analyzed by western blotting showed that proteasome inhibition resulted in a higher molecular weight, presumably ubiquitylated, HtsRC species (Fig. 2B). Proteasome inhibition also resulted in a twofold increase of HtsRC protein compared with DMSO control (Fig. 2C,D).

We tested HtsRC stability upon proteasome inhibition in the context of inhibiting new protein synthesis with cycloheximide (Fig. 2E,F). In cycloheximide-treated S2 cells, HtsRC levels declined over 6 h. With addition of proteasome inhibition, HtsRC declined more slowly (Fig. 2E,F). These data suggest that HtsRC is degraded by the proteasome in cultured Drosophila cells.

To determine whether endogenous HtsRC is ubiquitylated, we developed an assay to monitor protein ubiquitylation in ovarian germ cells. We used the germline-specific oskGal4 driver to express His-tagged Ubiquitin in combination with an shRNA targeting the proteasome subunit Rpn9. Knockdown of Rpn9 results in a moderate reduction of proteasome activity and was included to enrich for ubiquitylated proteins. We purified His-tagged proteins from wild type and oskGal4>His::Ub, Rpn9^KD ovaries, and observed a clear enrichment of ubiquitylated proteins in the HisUb samples (Fig. 2G), indicating that we could efficiently isolate ubiquitylated proteins. HtsRC was detected as a high molecular weight series of bands in purifications from His::Ub-expressing flies, indicating that HtsRC is a ubiquitylated protein in ovarian germ cells (Fig. 2H). We note that Rpn9 knockdown does not affect the processing of the Ovhts polyprotein (Fig. S3), eliminating the possibility that some of these bands could be partially processed forms of the Ovhts polyprotein. To determine whether ubiquitylation of HtsRC was dependent on the presence of Kelch, we performed a parallel experiment using flies lacking Kelch. To our surprise, we also detected HtsRC among the ubiquitylated proteins purified from kelch mutant lysates, with no apparent reduction in intensity relative to wild type (Fig. 2H). These results demonstrate that HtsRC is ubiquitylated in ovarian germ cells, but also reveal that the activity of CRL3^Kelch is not the only pathway leading to HtsRC ubiquitylation in these cells. We therefore turned to genetic experiments to determine whether CRL3^Kelch targets HtsRC specifically at the ring canal.

We performed a series of genetic enhancement and suppression experiments to determine how HtsRC protein expression levels affect the kelch-like ring canal phenotype caused by germline-specific proteasome inhibition (Fig. 3A). Ring canal F-actin thickness was assessed using the full width at half max (FWHM) measure of F-actin intensity plots. We analyzed ring canals from stage 6 egg chambers for this analysis because proteasome inhibition caused egg chamber arrest and degeneration beyond this stage (Hudson et al., 2015). Germline-specific expression of shRNAs targeting ProsBeta5, a catalytic subunit of the proteasome, led to kelch-like ring canals (Fig. 3A, red), marked by significantly thicker ring canal F-actin compared with wild-type ring canals (Fig. 3A, black). Increased HtsRC::GFP expression, achieved by expressing a UASH-ovhts::GFP construct (Petrella et al., 2007), significantly enhanced the kelch-like ring canal phenotype observed upon proteasome inhibition (Fig. 3A, green). Additionally, removal of one copy of kelch in the context of proteasome inhibition and HtsRC::GFP expression further increased the ring canal F-actin thickness (Fig. 3A, yellow). Conversely, removal of one copy of htsRC (achieved via CRISPR-mediated mutagenesis of the HtsRC-encoding exon; see Fig. 4A,F-G‴) was sufficient to suppress the kelch-like ring canal phenotype caused by proteasome inhibition (Fig. 3A, blue). These results reveal that increased or decreased HtsRC expression can significantly enhance or suppress, respectively, the kelch-like ring canal F-actin phenotype caused by proteasome inhibition, and that HtsRC, Kelch and the proteasome genetically interact.

We also tested the ability of HtsRC expression to affect the ring canal F-actin organization in stage 9 control egg chambers or egg chambers experiencing germline-specific knockdown of Cul3 (Fig. 3B). As shown previously (Hudson et al., 2015), knockdown of Cul3 leads to kelch-like ring canals (Fig. 3B, red). Removal of one copy of htsRC was sufficient to fully rescue this phenotype (Fig. 3B, blue). These data show that alteration of HtsRC protein levels can enhance and suppress the kelch-like ring canal F-actin phenotype in multiple contexts.

Functional analyses of HtsRC have been limited due to the multiple requirements for hts during oogenesis, including both germline mitotic divisions and oocyte specification (Yue and Spradling, 1992; Petrella et al., 2007). To determine the consequence of specifically disrupting HtsRC expression, we used CRISPR/Cas9 non-homologous end-joining (NHEJ) mutagenesis to introduce frameshift mutations in the exon encoding HtsRC (Figs 1G and 4A, green). We isolated three frameshift alleles that result in premature truncation of HtsRC (Fig. 4A). These alleles exhibited similar phenotypes whether homozygous or in trans to a deficiency, and none produced detectable HtsRC by immunofluorescence (Fig. S4E,E′) or immunoblot (Fig. S5A) analyses, indicating that they are strong loss-of-function alleles. Ovaries from these mutant females exhibited none of the germline mitotic defects associated with alleles affecting the Adducin isoforms. The fusome was present with Adducin localization indistinguishable from wild type (Fig. S4). Instead, the htsRC alleles displayed a specific ring canal phenotype, in which the robust inner rim of F-actin was absent (Fig. S4 and Fig. 4F,G′). We note that the htsRC-specific actin phenotype could be rescued by expression of Ovhts::GFP (Fig. S4). Consistent with previous work (Sokol and Cooley, 1999), Filamin was detectable at these ring canals (Fig. 4G‴), confirming that Filamin localization to ring canals is independent of HtsRC.

The loss-of-function phenotypes of kelch and htsRC were opposite of one another: in htsRC-specific mutants, ring canals lacked F-actin (compare Fig. 4C′ with 4G′; only peripheral subcortical F-actin is present in G′), whereas in kelch mutant females the F-actin cytoskeleton accumulated aberrantly in the ring canal lumens (Fig. 4E′). We examined double mutant combinations of loss-of-function alleles kel^DE1 and hts^R913fs. The double mutant was identical to the hts^R913fs single mutant with regard to the ring canal F-actin (Fig. 4F-I) and oocyte size phenotype (Fig. 4J), demonstrating that hts^R913fs is epistatic to kelch. This is consistent with HtsRC being an important substrate of CRL3^Kelch: when HtsRC is absent, loss of Kelch has no obvious consequence or additional phenotype.

If HtsRC levels at ring canals are regulated by CRL3^Kelch-mediated ubiquitylation, partial loss-of-function mutations in hts may be suppressed by mutations in kelch. We reduced the germline expression of HtsRC using an shRNA driven by matGal4 (Yan et al., 2014) (Fig. 5A,B). When hts shRNA was expressed in a kelch mutant background, HtsRC levels at ring canals were increased compared with expression of hts shRNA in wild type (Fig. 5C).

We also tested whether HtsRC protein levels are dependent on Kelch. Using western analysis, we have been unable to detect a significant change in endogenous HtsRC levels in a Kelch-dependent manner. We suspect that this is because western analysis does not distinguish between the cytoplasmic pool of HtsRC and ring canal-localized HtsRC. To overcome this challenge, we used a transgenic GFP-tagged HtsRC construct driven by matGal4 to assess HtsRC levels by GFP fluorescence (Fig. 5D,E) and western blotting (Fig. S5). Our goal was to increase our signal-to-noise ratio of ring canal-localized HtsRC by increasing its expression levels, so that any alteration of its levels by Kelch would be more easily detectable and quantifiable. HtsRC::GFP fluorescent protein levels were drastically reduced in egg chambers co-expressing mCherry::Kelch compared with mCherry control (Fig. 5D,E). Furthermore, western analysis revealed that total HtsRC protein species decreased significantly upon co-expression of mCherry::Kelch compared with mCherry control (Fig. S5). Of note, the levels of HtsRC::GFP species were most dramatically decreased upon Kelch overexpression, whereas endogenous HtsRC remained mostly unchanged (Fig. S5).

Finally, we examined the levels at ring canals of HtsRC::Venus fluorescent fusion protein expressed from a BAC transgene in the context of varying kelch gene copies (Fig. 5F,G). The pattern and levels of Venus fluorescence recapitulated localization patterns observed with the HtsRC monoclonal antibody. Furthermore, the expression of HtsRC::Venus was sufficient to rescue the loss of ring canal F-actin in htsRC-specific mutants, similar to otu-ovhts::GFP in Fig. S4. HtsRC::Venus fluorescent protein levels progressively increased as kelch gene copy number decreased (Fig. 5F). This effect was quantified by calculating the maximum fluorescence intensity of HtsRC::Venus for individual ring canals analyzed in four genotypes (Fig. 5G). The mean maximum fluorescence intensity of HtsRC::Venus progressively increased as kelch gene copy was reduced (compare black, green and blue data). These data show that Kelch protein levels directly affect ring canal-localized HtsRC protein levels.

Kelch contains an N-terminal region (NTR) of 120 amino acids that is rich in low-complexity sequence (Fig. 6A). We previously analyzed a transgene encoding a Kelch protein lacking the NTR expressed at low levels throughout oogenesis using the otu promoter (Robinson and Cooley, 1997). otu-driven expression of Kelch^ΔNTR resulted in a dominant-negative, female-sterile phenotype due to the destabilization of germ cell membranes. This phenotype was attributed to the premature localization of Kelch^ΔNTR to ring canals (Robinson and Cooley, 1997). Given our results indicating that Kelch regulates HtsRC levels through the UPS, we re-examined the phenotype caused by Kelch^ΔNTR expression by creating a UASp-kelch^ΔNTR transgene to express the Kelch^ΔNTR protein with greater control over the timing and level of expression. When we expressed wild-type Kelch at high levels using matGal4, the females were fertile with normal localization of HtsRC, consistent with our previous results (Fig. 6C; see also Fig. S5). In contrast, females expressing Kelch^ΔNTR were sterile, similar to previous results with otu-kelch^ΔNTR constructs. HtsRC was nearly undetectable within the expression domain of matGal4 (Fig. 6D,D′), and mat>kelch^ΔNTR egg chambers failed to complete oogenesis.

Expression of Kelch^ΔNTR at low levels using the weak otu-Gal4::VP16 driver (otuGal4) resulted in a similar but less severe phenotype, and western blotting revealed reduced levels of HtsRC (Fig. 6E,F). When we compared the expression levels of Kelch^ΔNTR with wild-type Kelch driven by otuGal4, we found that Kelch^ΔNTR was expressed at much higher levels than wild-type Kelch (Fig. 6E,G), suggesting that loss of the NTR resulted in a stabilized protein. This raised the possibility that elevated CRL3^Kelch activity toward HtsRC may have resulted from the increased quantities of a stabilized Kelch^ΔNTR. This would be similar to human mutations in KLHL24 that result in the production of a BBK protein that lacks its N-terminal region, leading to increased levels of KLHL24 and elevated levels of CRL3^KLHL24 activity (Lin et al., 2016). To address the effect of differing levels of Kelch, we compared the effects of Kelch^ΔNTR driven by otuGal4 with wild-type Kelch driven by matGal4. Wild-type Kelch driven by matGal4 was present at twice the level of Kelch^ΔNTR driven by otuGal4 (Fig. 6E,G); however, HtsRC levels were only reduced upon expression of Kelch^ΔNTR. This suggests that altered regulation of Kelch^ΔNTR, rather than increased quantity, is responsible for the apparent gain-of-function effects seen with Kelch^ΔNTR expression. Removal of the NTR appears to result in a hyper-active form of Kelch.

Identification of the conserved PEAEQ sequence in HtsRC and other KREP-interacting proteins suggested this motif is a binding site for the Kelch KREP domain. If this is the case, and if HtsRC ubiquitylation is essential for ring canal cytoskeletal organization, then mutations in the hts PEAEQ motif that inhibit Kelch binding should result in a kelch-like phenotype. To test this, we used CRISPR/Cas9 NHEJ mutagenesis to create mutations in the PEAEQ sequence (Fig. 7A).

We recovered in-frame deletions of both E922 and Q925 in the PEAEQ motif, and in marked contrast to truncation mutations that lack a robust ring canal F-actin cytoskeleton (Fig. 4 and Fig. S4), the in-frame deletions resulted in phenotypes similar to kelch loss-of-function mutants. The stronger allele, a deletion of the E922 codon (hts^ΔE922), resulted in a phenotype indistinguishable from the kel^DE1 null allele (Fig. 7B-D): F-actin and the mutant HtsRC protein accumulated in ring canals, and the flies were sterile. The levels of HtsRC at ring canals were twofold higher in hts^ΔE922 and kel^DE1 mutants compared with wild type (Fig. 7E). In addition, we found that, in hts^ΔE922 mutants, Kelch no longer localized to ring canals (Fig. 7G-I), despite Kelch protein being present (Fig. 7F). Curiously, although increased levels of HtsRC at ring canals in kelch or htsRC PEAEQ mutants were obvious by immunofluorescence, we detected no difference in their steady-state accumulation by western blot (Fig. 7H). The hts^ΔQ925 allele appeared to be weaker, and levels of HtsRC measured at ring canals were intermediate between wild type and hts^ΔE922 mutants (Fig. S6A-C,G). Consistent with this result, Kelch was detected at hts^ΔQ925 ring canals, but at reduced levels compared with wild type (Fig. S6D-F,H). Together, these results provide compelling in vivo evidence that the PEAEQ sequence is a binding site for Kelch.

In this study, we provide evidence that CRL3^Kelch targets HtsRC at ring canals for ubiquitylation and proteasome-mediated destruction, providing a mechanism for how Kelch shapes the cytoskeleton during ring canal growth. Four lines of evidence support the idea that HtsRC is the key substrate of CRL3^Kelch: first, we detected HtsRC in a complex with the substrate-binding KREP domain of Kelch; second, we showed that HtsRC is ubiquitylated in cultured cells and in ovarian cells; third, we showed that levels of HtsRC are sensitive to the dose of Kelch; and, finally, we identified a putative binding motif in HtsRC that, when mutated, recapitulates the kelch phenotype and results in a loss of Kelch localization to ring canals. Together, these results provide strong evidence that HtsRC levels at ring canals are controlled by CRL3^Kelch-mediated targeted destruction to promote the coordinated expansion of the ring canal cytoskeleton. We were unable to demonstrate that all HtsRC ubiquitylation is Kelch dependent. It is therefore possible that CRL3^Kelch removes HtsRC specifically from the ring canal lumen through a UPS-independent mechanism.

A striking feature of many proteins identified by mass spectrometry in complex with the KREP domain was the presence of a linear PEAEQ peptide motif. Two human BBK proteins, KLHL2 and KLHL3, are most closely related to Drosophila Kelch, and both have been shown to bind a similar EPEEPEADQ motif in WNK4, a substrate of KLHL3 (Schumacher et al., 2014). A structure of a KLHL3-WNK4 peptide complex revealed that the EPEEPEADQ peptide binds on the top surface of the KLHL3 KREP β-propeller, making extensive contacts with inter-strand surface loops derived from blades 2-4. We propose that the HtsRC PEAEQ motif binds Drosophila Kelch in a similar manner, and the prevalence of PEAEQ proteins we identified by both pulldown and yeast two-hybrid screening suggests that proteins with PEAEQ-like linear motifs are potential targets of Kelch and closely related BBK proteins.

In contrast, Keap1, a more distantly related BBK protein, targets its substrate Nrf2 for ubiquitylation through a high-affinity interaction with a short DxETGE sequence (Lo et al., 2006). A structure of the Keap1–DxETGE peptide complex revealed that Keap1 also engages its target peptide through contacts with surface loops extending from the top of the Keap1 β-propeller structure, similar to the KLHL3–PEADQ interaction. Unlike KLHL2 and KLHL3, however, the Nrf2 DxETGE peptide was found to be closely associated with surface loops derived from blades 1, 5 and 6; it therefore appears that the Keap1 KREP domain binds the DxETGE peptide in a manner distinct from the KLHL3-PEADQ interaction. These observations suggest that subfamilies of BBK proteins engage substrates through distinct short peptide motifs. A more thorough understanding of BBK substrate interaction interfaces will be important for the development of pharmacological inhibitors or biological therapies, in order to target disease-relevant BBK proteins or their substrates.

Interactions between ubiquitin ligases and their substrates are typically regulated such that substrate ubiquitylation is restricted to the appropriate time and place. The Germ cell-less protein (GCL) is a component of a CRL3 that degrades the Torso receptor tyrosine kinase in order to maintain primordial germ cell fate (Pae et al., 2017). GCL is located at the nuclear envelope during interphase, and is only free to act following nuclear envelope breakdown during mitosis. Mutants that disrupt this mode of regulation result in the apparent ubiquitylation of inappropriate substrates and a severe germ cell defect (Pae et al., 2017). KLHL10, a BBK CRL3 substrate adaptor required for spermatid differentiation, is regulated by Scotti, a protein that acts as a pseudosubstrate inhibitor of CRL3^KLHL10 to regulate its activity in a graded fashion during spermatid elongation (Arama et al., 2007; Kaplan et al., 2010). We found that the Kelch NTR is important in regulating CRL3^Kelch activity. We have previously observed that transgenic constructs driving low-level germline expression of Kelch^ΔNTR resulted in a dominant female-sterile phenotype (Robinson and Cooley, 1997). Our observation that Kelch^ΔNTR expression depletes HtsRC from germline cells suggests that Kelch^ΔNTR is a hyperactive protein that targets HtsRC in an unregulated manner. We note that the phenotype of Kelch^ΔNTR is similar to, but more severe than, loss-of-function mutations in htsRC, suggesting that the Kelch^ΔNTR may target proteins that are not normally substrates of CRL3^Kelch during oogenesis.

The Drosophila Kelch NTR is a 120 amino acid extension rich in low-complexity sequence. It is conserved in Kelch orthologs in insects, but not in KLHL2 and KLHL3. However, 13 human BBK proteins have N-terminal extensions greater than 60 amino acids, and many of these are rich in low-complexity sequence. Interestingly, gain-of-function mutations in one of these genes, KLHL24, produce a mutant protein lacking part of its NTR that accumulates to high levels, resulting in unregulated ubiquitylation and degradation of its substrate, keratin 14, which causes skin fragility (Lin et al., 2016). Stabilization of mutant KLHL24 results from impaired auto-catalytic ubiquitylation and degradation of KLHL24, a commonly observed feature of CRLs (de Bie and Ciechanover, 2011). Although Kelch also undergoes Cul3-dependent autocatalytic degradation (Hudson and Cooley, 2010; Hudson et al., 2015), we do not think the hyperactive phenotype of Kelch^ΔNTR is simply a consequence of increased protein levels as elevated levels of wild-type Kelch do not have a similar phenotype.

Our results support a model in which CRL3^Kelch targets HtsRC located at ring canals for ubiquitylation, leading to its degradation by the proteasome (Fig. 8). Consistent with this model, we found that, in cultured cells, HtsRC is stabilized upon proteasome inhibition, and that its stabilization is accompanied by the presence of high molecular weight ubiquitylated forms. In addition, we present direct evidence that HtsRC is ubiquitylated in ovarian extracts. We did not find HtsRC ubiquitylation was strictly dependent on Kelch, however; nor did we observe an increase in HtsRC levels by western blotting. It is possible that these discrepancies arise from the existence of two pools of HtsRC: an insoluble pool associated with the ring canal cytoskeleton; and a second, more soluble, cytoplasmic pool of HtsRC. We suspect that a distinct ubiquitylation pathway targets HtsRC in the cytosol and that the ubiquitylated HtsRC we observe in kelch mutants is a result of this alternative pathway. Our previous work supports the idea of a Kelch-independent degradation pathway acting on HtsRC present in the cytosol. HtsRC is selectively maintained in egg chambers; during the germline mitotic divisions in the germarium, both HtsF and HtsRC are produced, but only HtsF is observed whereas HtsRC appears to be actively degraded (Petrella et al., 2007). The degradation of HtsRC in the germline mitotic region is independent of Kelch, as HtsRC is also undetectable during mitosis in kelch mutants. HtsRC has also been suggested to function in other F-actin-rich structures in egg chambers, including nurse cell actin cables (Huelsmann et al., 2013) and the oocyte cortical cytoskeleton (Pokrywka et al., 2014). An alternative ubiquitylation pathway may regulate these non-ring canal functions of HtsRC. In addition, if cytoplasmic HtsRC is more easily extracted than the ring canal-associated HtsRC, it could mask differences in levels of the ring canal-HtsRC on western blots. Nevertheless, the findings that HtsRC is poly-ubiquitylated in ovaries and that its levels at the ring canal are regulated by Kelch strongly support a model for ring canal-specific HtsRC degradation mediated by CRL3^Kelch.

Based on our results, we propose that HtsRC is the key driver of ring canal growth (Fig. 8A). HtsRC is essential for the dynamic and robust F-actin cytoskeleton that lines the lumen of the ring canal; hts mutant egg chambers display multiple phenotypes, including the failure of F-actin to accumulate at ring canals (Petrella et al., 2007; Yue and Spradling, 1992), and our new exon 12 mutations affect only ring canals. The molecular mechanism by which HtsRC recruits F-actin to ring canals is not known; sequence analysis of HtsRC reveals no recognizable motifs aside from a predicted coiled-coil structure at its C terminus. One notable biochemical feature of HtsRC is its poor solubility. Ovarian lysates contain little soluble HtsRC, and our attempts to produce soluble recombinant protein in its native state for in vitro studies have had limited success (J.A.G., unpublished). We speculate that the HtsRC at the ring canal may function by forming an insoluble matrix on which the dynamic ring canal F-actin cytoskeleton is assembled, driving the expansion of the ring canal diameter. However, to allow for the coordinated expansion of the ring canal lumen, HtsRC and associated proteins must be disassembled at the luminal surface, and this requires targeting by CRL3^Kelch and the UPS. This represents an unusual mechanism for cytoskeletal remodeling, which generally relies on non-destructive mechanisms to create and dismantle cytoskeletal assemblies. However, removal by the UPS may be the only effective way to control the potent activity of HtsRC. Regulation of cytoskeletal structures by the ubiquitin-proteasome system is an emerging regulatory mechanism (Deng and Huang, 2014). However, most previous studies have used yeast or cultured cells (Wang et al., 2003; Chen et al., 2009; Razinia et al., 2011, 2013; Juanes and Piatti, 2016; Girouard et al., 2016; Wu et al., 2018). Our work is one of few examples to date detailing the mechanism by which the UPS organizes and remodels a cytoskeletal structure in a metazoan developmental model system.

Drosophila were maintained at 25°C on standard fly food medium. Prior to ovary dissections, females were fattened on wet yeast paste overnight at 25°C. See Table S2 for a detailed list of fly stocks used in this study.

Drosophila S2 cells (DGRC) were cultured at 27°C in Schneider's Drosophila Media (Gibco) supplemented with 10% fetal bovine serum and 1× antibiotic-antimycotic (Gibco). We confirmed that the cells were not contaminated.

The cher¹ allele (89E) was recombined with the UASp-TAPmCherry (Neelakanta et al., 2012) and UASp-TAPmCherry::KREP (Hudson and Cooley, 2010) constructs integrated at the attP2 site (68A4), and also recombined with matGal4::VP16 (mapped to 91D4 by inverse PCR). Recombinant chromosomes were balanced with a TM3, hs-hid balancer (Bloomington #1558) to allow for heat-shock-induced selection of desired cross progeny. Approximately 500 ovaries from cher¹ mutant females expressing either TAPmCherry or TAPmCherry::KREP were homogenized in SBP lysis buffer [50 mM HEPES (pH 7.5), 150 mM NaCl, 2 mM EDTA, 0.5%Triton X-100, 10% glycerol, 1 mM PMSF and 5 µg/ml each of chymostatin, leupeptin, antipain and pepstatin] in a glass Duall homogenizer with a teflon pestle driven by motorized spinning. Lysates were clarified by centrifugation for 30 min at 20,000 g at 4°C, and approximately 40 mg of clarified supernatants were incubated with 75 µl streptavidin beads (Thermo Fisher Scientific, 53114). Beads were washed 2× in SBP lysis buffer followed by two additional washes in His-tag buffer [50 mM NaPi (pH 8), 300 mM NaCl, 5 mM imidazole, 0.2% Triton X-100]. Proteins bound to streptavidin beads were eluted in 1 ml 10 mM biotin in His-tag buffer for 20 min at 4°C. Eluted proteins were incubated with 75 µl 50% Ni⁺⁺-NTA slurry for 2 h, washed three times in His-tag buffer, and eluted twice in 35 µl SDS-PAGE sample buffer containing 200 mM DTT and 200 mM imidazole. Each sample (25 µl) was loaded on a 4-12% gradient NuPAGE gel and stained with EZBlue colloidal Coomassie (Sigma-Aldrich). A second 25 µl aliquot was analyzed at MS Bioworks by running the samples on a 10% Bis-Tris NuPAGE gel, slicing each lane into 10 equal fragments, and subjecting each fragment to in-gel trypsin digestion followed by LC-MS/MS on a Thermo Fisher Q Exactive instrument. Mass spectrometry data were searched against the Uniprot Drosophila proteome at MS Bioworks using Mascot (Matrix Science) and the results were parsed into Scaffold (Proteome Software) for further analysis.

To make pAHW-HtsRC::HA for expression of HtsRC in S2 cells, exon 12 of ovhts with a C-terminal 1xHA tag was PCR amplified to contain flanking attB1 and attB2 sites. The PCR product was recombined into the Gateway Donor vector pDONR201 in a Gateway BP Clonase II reaction (Thermo Fisher Scientific). The entry clone was further recombined into the Gateway Expression vector pAHW (Drosophila Gateway Vector Collection; Drosophila Genomics Resource Center, 1095) in an LR Clonase II reaction (Thermo Fisher Scientific). The resulting pAHW-HtsRC::HA expression plasmid allows for expression of HtsRC with a 3xHA N-terminal tag and 1xHA C-terminal tag, driven by the Actin5C promoter. pUASp-kelch^ΔNTR was generated by amplifying a region of the kelch CDS encoding amino acids Q121-M689, the same range used in earlier Kelch^ΔNTR constructs (Robinson and Cooley, 1997). PCR primers included flanking attB1 and attB2 Gateway recombination sites and the product was recombined first into pDONR201 and then into pPW-attB (a gift from Mike Buszczak, UT Southwestern Medical Center, TX, USA), an untagged UASp vector from the Drosophila Gateway collection modified to include a phiC-31 attB recombination target. The pUASp-kelch^ΔNTR was injected into a strain carrying the attP2 phiC-31 landing site on Chromosome 3L at Rainbow transgenics. To construct pUASp-7xHis::Ubiquitin, a human HA-tagged ubiquitin cDNA was excised as an EcoRI-XhoI fragment from pUASt-HA::Ubiquitin (a gift from Tian Xu, Westlake University, China) and cloned into pBluescript II (Stratagene). The HA tag in this plasmid was replaced with oligonucleotides encoding a 7xHis tag by ligation into the NcoI and BglII sites, and the 7xHis::Ubiquitin fragment was then transferred to pMTV5 His (Invitrogen) as an EcoRI-XhoI fragment. 7xHis Ubiquitin from the pMTV5 plasmid was then cloned into pUASp as a KpnI-Xba fragment. The plasmid was transformed into the w¹¹¹⁸ strain using standard procedures.

A cDNA encoding the Kelch KREP domain was cloned into pEG202, the bait vector for the LexA yeast two-hybrid system (Golemis et al., 2011). For the interaction screen, 6×10⁶ yeast transformants from the Ovo II ovarian 2-Hybrid cDNA library (Grosshans et al., 1999) were screened for interaction by β-galactosidase production on X-gal plates and by complementation of leu2 auxotrophy. Two-hundred of the fastest growing colonies were selected for further analysis.

Oligos encoding sgRNAs described in Table S2 were cloned into the BbsI site of the pBFvU6.2 vector (Kondo and Ueda, 2013) and integrated at the attP2 or attP40 phiC31 integration site (Groth et al., 2004). Mutagenesis was carried out in males expressing Cas9 from a germline promoter (nanos-Cas9; Port et al., 2014; Bloomington, 54591) in combination with ubiquitous sgRNA expression from the snRNA:U6:96Ab promoter. Single male progeny bearing CRISPR-mutagenized chromosomes were crossed to a deficiency for hts [Df(2R)BSC135/CyO, Bloomington, 9423], and hemizygous progeny were screened by sequencing and/or fertility tests. Balanced stocks were established for mutations of interest.

BAC clone CH321-84O22 (Venken et al., 2009), which contains all of the hts locus on a 91 kb genomic fragment (chr2R:19,368,835…19,460,115, FlyBase release 6), was used as a template to introduce fluorescent protein-coding sequences. HA::Venus::FLAG sequence was inserted before the stop codon of exon 12, the exon encoding HtsRC (codon starting position chr2R:19,3399,860), using 2-step BAC recombineering. Briefly, a RpsL-neo selection cassette (Wang et al., 2009) was inserted using Kanamycin selection, and the RpsL-neo cassette was subsequently replaced with HA::Venus::FLAG using streptomycin counterselection. The same approach was then used to insert HA::TagRFP-T::FLAG before the first codon of ovhts exon 8 (Trp 473 of Ovhts polypeptide, codon starting position chr2R:19,409,361). Two additional rounds of recombineering were used to reduce the size of the BAC; the trimmed BAC contained a 44 kb genomic fragment from chr2R:19,390,835 (in CG11257) to chr2R:19,434,833 (in Fak). The BAC was injected into the attP2 landing site on chr3L at GenetiVision. TagRFP-T fluorescence was not observed in transformed lines, suggesting that the introduction of TagRFP-T into the Adducin portion of Ovhts was detrimental to its folding. However, Venus fluorescence recapitulated the expression and localization patterns that are observed with the HtsRC monoclonal antibody, indicating that the cleaved ring canal protein tagged with Venus was produced normally. Consistent with this, the HtsRC protein produced by this construct was sufficient to rescue the loss of ring canal F-actin in htsRC-specific mutants, similar to otu-ovhts::GFP in Fig. S4.

Ovaries were fixed in 4% paraformaldehyde (Electron Microscopy Sciences) in PBS with or without 0.1% Triton X-100 for 10 min, washed in PBT (PBS, 0.1% Triton X-100, 0.5% BSA) and incubated in primary antibody in PBT at 4°C overnight. Antibodies and other fluorescent reagents are listed in Table S2. Samples were washed four times in PBT, and incubated with secondary antibodies and fluorescent phalloidin (if used) in PBT for 2 h at room temperature. Following secondary antibody incubation, samples were washed four times in PBT and mounted on slides in ProLong Gold or Diamond antifade reagents (Thermo Fisher Scientific). Samples were imaged on one of three microscopes: a Leica SP8 scanning confocal system using a 40× Plan Apo 1.30 NA objective; a Nikon TiE inverted microscope with a Yokogawa CSU-W1 spinning disc system, an Andor iXon Ultra888 1024×1024 EMCCD and 40× Plan Fluor 1.30 oil immersion objective; or a Zeiss Axiovert 200m inverted microscope with a CrEST X-light spinning disc system, Photometrics CoolSNAP HQ2 camera and either a 20× Plan Apo 0.8 NA objective or a 40× C-Apo 1.2 NA water-immersion objective. Images were processed and analyzed with ImageJ/FIJI or Imaris 9.0 (Bitplane).

S2 cells at 50-80% confluency were transfected with 1 µg of pAHW-HtsRC::HA DNA following the Effectene Transfection Reagent protocol (Qiagen). To assess HtsRC ubiquitylation and stability in response to drug/inhibitor treatments, cells were treated with DMSO control (Sigma-Aldrich), 1 µM bortezomib (Cell Signaling) and/or 100 µg/ml cycloheximide (Sigma-Aldrich) beginning at 24 or 36 h post-transfection, as indicated in Fig. 2 legend. Cells were harvested at the indicated time points (0, 3 and 6 h following inhibitor treatment), lysed in SDS-PAGE sample buffer and analyzed by western analysis to visualize HtsRC and Tubulin protein levels. HtsRC protein levels were quantified using ImageJ by measuring the integrated density of HtsRC protein bands relative to Tubulin protein bands to control for protein loading. GraphPad Prism was used to perform a Student's t-test using data from three independent experiments.

For His::Ubiquitin purifications, flies of genotype UASp-7xHis::Ub/+; oskGal4/+; UASp-7xHis::Ub/shRNA-Rpn9^HMS01007 were used. For purifications in a kelch mutant background, flies of genotype UASp-7xHis::Ub/+; kel^DE1, oskGal4/ kel^DE1; UASp-7xHis::Ub/shRNA-Rpn9^HMS01007 were used. Control purifications were performed using w¹¹¹⁸. Isolation of ubiquitylated proteins was based on established procedures (Laney and Hochstrasser, 2002). Briefly, ovaries from ∼200 flies of each genotype were lysed in a glass Duall homogenizer in His/Guanidine buffer: 6 M guanidine, 50 mM NaPi/10 mM Tris (pH 8), 300 mM NaCl, 20 mM imidazole, 0.1% Triton X-100, 10 mM N-ethylmaleimide, 1 mM PMSF and 5 µg/ml each of chymostatin, leupeptin, antipain and pepstatin. Lysates were clarified by centrifugation for 30 min at 20,000 g at 4°C and supernatants were incubated with 50 µl Ni⁺⁺-NTA resin (Qiagen) in His/Guanidine buffer for 4 h at 4°C. Beads were washed twice in His/Guanidine buffer, once in Urea wash buffer at pH 8 [8 M Urea, 50 mM NaPi/10 mM Tris (pH 8), 300 mM NaCl, 20 mM imidazole, 0.1% Triton X-100] and twice in Urea wash buffer (pH 6.0). Proteins were eluted in SDS-PAGE sample buffer containing 200 mM DTT and 200 mM imidazole and analyzed by immunoblotting.

Egg chambers were labeled with HtsRC 6A (Robinson et al., 1994) or Kel 1B monoclonal antibodies, both of which result in specific labeling of ring canals with little plasma membrane signal. Image stacks of wild-type and mutant egg chambers that included all ring canals in selected stage 10 egg chambers were collected on a spinning disc confocal with constant settings for exposure time, illumination power and camera gain. Image stacks were analyzed in Imaris using automatic 3D thresholding of HtsRC or Kelch signal to identify ring canal volumes, and the summed fluorescence intensity was calculated for each ring canal. Raw fluorescent intensities were normalized to the mean fluorescence intensity of the control sample, which was defined as 1.

To measure ring canal F-actin thickness, egg chambers were stained with fluorescent phalloidin. Maximum intensity projections of imaged egg chambers were rendered using ImageJ, and ring canals that provided a full longitudinal cross-section with a top-down view were analyzed (see top panel of Fig. 3 for representative ring canals). A line was drawn through the center of measurable ring canals, and a F-actin fluorescence intensity plot spanning the line was generated. The full width at half maximum (FWHM) function of F-actin fluorescence intensity was calculated and used to approximate the thickness of the ring canal F-actin (see Hudson et al., 2015 for more details). Ring canals from stage 6 egg chambers were analyzed for Fig. 3A because that was the oldest stage the egg chambers progress to before undergoing degeneration (Hudson et al., 2015). For Fig. 3B, we analyzed ring canals from stage 9 egg chambers because those egg chambers were abundant, easily identifiable and the ring canals were larger, which was more advantageous for quantifying the F-actin thickness. For Fig. 3 graphs, n refers to the number of ring canals analyzed, the black line denotes the mean F-actin thickness, and error bars represent the 95% confidence interval. One-way ANOVA with Tukey's multiple comparison correction was used in GraphPad Prism to detect for statistically significant differences in the mean ring canal F-actin thickness between samples.

To quantify how HtsRC fluorescent fusion protein levels were dependent on Kelch (Fig. 5), egg chambers of all stages were imaged on a Zeiss spinning disc microscope using the same exposure time (500 ms). Maximum intensity projections of imaged egg chambers were rendered in ImageJ, and the maximum fluorescence value for each ring canal was calculated. The mean maximum fluorescence intensity of ovhts::Venus x2 sample (Fig. 5B, black) was normalized to 100 relative fluorescence units. Each plotted point represents the maximum fluorescence intensity of the HtsRC fluorescent fusion protein (HtsRC::Venus for Fig. 5B, HtsRC::GFP for Fig. 5D). n equals the number of ring canals analyzed and the black line represents the mean maximum fluorescence intensity while error bars represent 95% confidence interval. Statistical analyses (one-way ANOVA for Fig. 5B and Student's t-test for Fig. 5D) were performed in GraphPad Prism to test for differences in the average maximum fluorescence intensity between the various genotypes, as indicated in the figure legends. Data are compiled from two independent experiments.

To analyze HtsRC protein levels by western analysis, ovary lysates were generated by homogenizing dissected ovaries in SDS-PAGE sample buffer. One ovary equivalent was loaded per lane and separated on a 8.5 or 9% polyacrylamide gel. Proteins were transferred to nitrocellulose membrane, stained with amido black to visualize total protein, blocked in 5% milk in TBST (Tris-buffered saline with 0.1% Tween-20) and probed with the following antibodies: anti-HtsRC, anti-Kelch and anti-Tubulin. Blots were incubated with HRP-conjugated secondary antibodies followed by ECL development (Bio-Rad) and imaged on a CCD camera (Protein Simple). For Fig. 6, levels of HtsRC, Kelch and β-Tubulin were measured using FIJI/ImageJ, with β-Tubulin serving as a loading control. Levels of HtsRC and Kelch are expressed relative to wild-type levels in w¹¹¹⁸, defined as 1. Data are from four independent experiments. For Fig. S5, total HtsRC protein levels were quantified using ImageJ by measuring the integrated density of HtsRC::GFP and HtsRC protein bands relative to amido black integrated density to control for protein levels loaded. HtsRC levels for mCherry-expressing control lysates were normalized to 1. HtsRC and HtsRC::GFP levels were parsed apart for graphical visualization. Error bars represent s.e.m. Student's t-test was performed using GraphPad Prism. Data are from three independent experiments.

Data visualization and statistical analysis was performed in Prism (GraphPad). For quantification of protein levels by western blotting, error bars represent the mean±s.e.m. For all other quantification experiments, individual data points are plotted and the error bars represent the mean±95% confidence interval (CI), as indicated in the figure legend. Binomial test for enrichment of PEAEQ-containing proteins compared observed (15 PEAEQ proteins out of 141 KREP-interacting proteins) to expected [∼3/141, based on the Drosophila genome encoding 262 PEAEQ proteins from 13,931 protein coding genes (Table S1)]. Other statistical tests used (one-way ANOVA or Student's t-test), P values and n are also listed in the figure legends.

We thank Jovan Williams and Sowjanya Kallakuri for their help in recombineering the Ovhts BAC transgene. We thank Tian Xu for providing access to the Leica SP8 confocal microscope. Stocks obtained from the Bloomington Drosophila Stock Center (NIH P40OD018537) were used in this study. We thank the TRiP at Harvard Medical School for providing transgenic RNAi fly stocks used in this study. We thank the Drosophila Genomics Resource Center, supported by NIH grant 2P40OD010949, for the S2 cells used in this study. J.A.G. was funded in part by a Gruber Science Fellowship from Yale University.