Whole-organism developmental expression profiling reveals a cilium-specific gene expression pattern

To provide a complementary approach to ciliary gene discovery, we took advantage of the temporally-invariant birth of all C. elegans cells and tissues, including ciliated neurons, during development [48] (Fig 1A). Nearly all of the 60 ciliated neuronal cell types in C. elegans hermaphrodites are born within a discrete embryonic time period 300–450 minutes post-fertilisation, with cilium formation occurring very shortly thereafter (Fig 1A). We hypothesised that ciliogenesis genes are highly expressed during this time period and therefore distinguishable from genes required for general neuronal formation and development, which are expressed during a broader time window (Fig 1A). Using an available whole-organism developmental series of RNA-seq libraries from C. elegans [45–47], we confirmed this hypothesis: many well-characterised cilia genes are highly expressed in the early embryo, display peak expression in the late embryo and first larval stage, and show greatly reduced expression during subsequent developmental stages (Fig 1B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Whole-organism expression profiling identifies a signature pattern enriched for known ciliary genes.

(A) Timeline (in minutes) of C. elegans cell births according to tissues. Small shapes in each row represent the birth of a cell of the corresponding tissue type. Embryonic development, hatching and post-embryonic development are highlighted above the x-axis with the worm stages illustrated below. Most ciliated neurons are born in late embryogenesis with ciliogenesis occurring shortly after these cells are born. Worm illustrations modified from WormAtlas (www.wormatlas.org). Q/G/K/P/H/V/T/M refer to different embryonic cell lineages (http://www.wormatlas.org/celllineages.html). (B) Normalised expression levels for all ‘bait’ cilium-associated genes (red) and the top 276 target genes (black, p-value < 5e-7) over the developmental series of analysed RNA-Seq libraries. Expression peaks for all genes during late embryo or larval stage 1, which matches closely the birth of the ciliated neurons shown in (A). This cilia-specific gene expression pattern shows much lower levels of expression from larval stage 2 to young adult. (C) Box and whisker plots of the normalised expression across the whole-organism developmental RNA-Seq libraries for the chemosensory and ciliary gene sets. Expression of the ciliary set immediately precedes that of the chemosensory set. Peak expression for ciliary genes is at the late embryo-L1 larval stage, whereas expression peaks at L1-L2 larval stages for chemosensory genes. EE; early embryo. LE; late embryo. L1; larval stage 1. L2; larval stage 2. L3; larval stage 3. L4; larval stage 4. YA; young adult.

https://doi.org/10.1371/journal.pgen.1006469.g001

Next we sought to identify novel ciliary/ciliogenic genes displaying a similar cilia-related expression pattern. Using a set of 41 well-characterised ciliary component genes as “baits”, representing the cilia-related gene expression profile during development (sheet 1 in S1 Table), we queried our ciliary transcriptome to identify other gene “preys” with similar expression profiles across the RNA-Seq libraries (Fig 1B and sheet 2 in S1 Table). Hierarchical clustering of genes based on temporal expression reveals a tight cluster of 34 of the bait genes with 151 prey genes (cluster 1), many of which are uncharacterised (S1 Fig and sheet 3 in S1 Table). We also filtered the gene list to only include those with human orthologues [49]. To validate our dataset, we determined if filtered cluster 1 is enriched for genes with >12 hits in the ciliary database Cildb [23,24] (sheets 3 and 4 in S1 Table). We found that conserved Cildb-represented genes are significantly enriched (84.5 fold) in filtered cluster 1 compared to the entire genome (p<0.0001, chi-squared test; sheet 4 in S1 Table). We further compared the cluster 1 gene set to two previous expression studies [25,50] and see significant overlap between lists (p<0.0001; sheet 4 in S1 Table).

Because our genes were identified by shared temporal gene expression, we reasoned that they may share common promoter regulatory elements beyond the previously identified X-box motif [27,28] and so we performed a motif discovery analysis on the conserved genes in cluster 1 (sheet 3 in S1 Table). Not surprisingly, we find several motifs that are significantly enriched, including matches to the GAGA factor binding site (S2 Fig). As might be expected, the 14-bp X-box was not identified by the motif elucidation program despite its enrichment in our dataset (sheet 4 in S1 Table), likely due to the number of degenerate bases in its consensus sequence and because the program identifies motifs only up to 8-bp long [26–28,51]. None of the elements we identified have been previously associated with cilia gene expression.

Post-ciliogenesis chemoreceptor co-expression profiling

We observed that chemoreceptors (G protein-coupled receptors or GPCRs) [52], many of which are known to be cilium-localised, are largely absent from our ciliary dataset (sheet 3 in S1 Table). We hypothesised that such genes might be transcribed post-ciliogenesis. Using a statistical predictive modelling strategy similar to that used above, we identified 80 genes sharing the expression profile of srg-36, a cilium-localised dauer pheromone receptor required for entry into the dauer larval diapause life stage (Fig 1C and sheet 5 in S1 Table) [53]. Of the 80 genes, 27 are serpentine transmembrane receptors, representing a significant 7-fold enrichment over all such genes in the genome (p<0.0001, chi-squared test). Analysis of the upstream regulatory regions revealed that 25 of the 28 chemoreceptor genes (including srg-36) possess an E-box regulatory element (S3 Fig) [54]. This suggests that their similar temporal expression profiles stem from regulation by the same, or similar, transcription factors such as the E-box-binding Basic Helix-Loop-Helix bHLH transcription factor [54]. Interestingly, 4 of the predicted serpentine receptors, including srg-36, are known to be expressed in ciliated sensory neurons [53,55,56]. We note that the rise and fall of chemosensory/GPCR gene expression trails that of the cilia-related genes (Fig 1C), indicating that these two expression profiles of related developmental processes are temporally discrete.

Identification of previously unstudied proteins from C. elegans that are cilium-localised: FAM-161, CCDC-104, CCDC-149, RPI-1, MAPH-9, YAP-1 and RAB-28

To confirm that our predictive expression profiling model identifies novel ciliary proteins in C. elegans, we used GFP reporters to investigate the cell and tissue expression patterns, and subcellular protein localisations, of seven previously uncharacterised C. elegans proteins from filtered cluster 1, namely FAM-161 (Y38H6C.14), CCDC-104 (Y108G3Al.3), RPI-1 (W07G1.5), RAB-28 (Y11D7A.4), CCDC-149 (F29G6.2), MAPH-9 (C34D4.1) and YAP-1 (F13E6.4). These were chosen because at the onset of this study there was little or no published evidence of ciliary associations for most of these proteins in any system. Specifically, we made transcriptional ‘promoter fusion’ reporters (endogenous gene promoter fused to GFP) for CCDC-104, CCDC-149 and RAB-28, and translational ‘protein fusion’ reporters (endogenous gene promoter + genomic exon/intron or cDNA sequence fused to GFP) for all 7 candidates. The only exception was the YAP-1 translational reporter, where the bbs-8 gene promoter active only in ciliated cells [57] was used because of the widespread expression of YAP-1 in multiple tissues and cell types [58]. We also made a second RAB-28 translational reporter driven by the bbs-8 gene promoter.

All reporters employing the endogenous gene promoter show enriched or almost exclusive expression within ciliated cells (S4A Fig), thus validating the predictions from our coexpression profiling. Specifically, maph-9, rpi-1, ccdc-104, rab-28, and fam-161 are exclusively expressed in ciliated neurons, with ccdc-149 expressed in most ciliated neurons as well as pharyngeal neurons, touch receptor neurons, and motor neurons as previously reported (Figs 2 and S4A) [59]. Furthermore, the translational reporters reveal that all 7 proteins localise at the base of, or within, cilia (Fig 2 and Fig 3A). FAM-161::GFP signals are found at the ciliary base and within the proximal part of the ciliary axoneme, including the transition zone (TZ) (Fig 2). This localisation is similar to that of the mammalian protein implicated in retinitis pigmentosa, and is consistent with the suggested role of FAM161b in cargo delivery to photoreceptor outer segments (cilia) [60,61]. Both GFP reporters for RAB-28 (driven by bbs-8 gene or endogenous promoter), the orthologue of the small ciliary GTPase RAB28 linked to human autosomal-recessive cone-rod dystrophy [62–64], localise along the entire cilium. (Fig 2 and Fig 3A). Ciliary axoneme localisations are also observed for the GFP-tagged RPI-1, MAPH-9, YAP-1 and CCDC-104 translational reporters; in contrast, CCDC-149::GFP is absent from the ciliary axoneme, although a pool of signal is evident at the ciliary base (Fig 2). The ciliary localisation of CCDC-104::GFP is consistent with the mammalian ciliary base and axonemal localisations reported for the ARL3 interacting protein, CCDC104/BARTL1 [65]. We also found that RPI-1::GFP appears to localise to ciliary and cytoplasmic microtubules, similar to its mammalian orthologue (RP1) implicated in retinitis pigmentosa (Fig 2) [66]. Our finding of ciliary localisations for YAP-1, MAPH-9 and CCDC-149 are the first in any cellular system or organism, and is consistent with a known ciliogenesis role for mammalian YAP, and cilia-related phenotypes in MAP9 (MAPH9 orthologue)-disrupted zebrafish and dachshunds [67–70]. YAP is regulated by the ciliary protein NPHP4, and MAP9 is phosphorylated by the cilia-disassembly/centrosomal kinase PLK1 [71,72].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Candidate cilium-associated proteins localise to ciliary structures Representative images of amphid (head) and phasmid (tail) cilia from N2 wild type worms co-expressing GFP-tagged ‘translational’ reporters for candidate ciliary genes and tdTomato-tagged XBX-1 (IFT protein that localises to the ciliary base and axoneme).

All reporters driven by the endogenous gene promoter, except for YAP-1 and RAB-28, where a bbs-8 gene promoter sequence was used (see Fig 3A for a GFP::RAB-28 reporter driven by the endogenous promoter). GFP tags are on the N- or C-terminus, as indicated above the panels. PCMC (periciliary membrane compartment); distal dendrite (den) swelling enriched for endocytosis-associated proteins and vesicles that regulate ciliary membrane homeostasis [19]. The transition zone (TZ) extends from basal body (BB) and functions as a ciliary gate that regulates protein entry and exit to and from cilia [6]. Middle (MS) and distal (DS) segments; characterised by a circular array of 9 doublet (A and B tubules) and 9 singlet (A tubules) microtubules, respectively (not shown). Note that for GFP::CCDC-149, signals are observed at the amphid/phasmid basal bodies (arrows) and as punctae (asterisks) along the dendrite of the OLQ ciliated neuron running parallel to the amphid neurons. Scale bars; 5 μm.

https://doi.org/10.1371/journal.pgen.1006469.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. RAB-28 undergoes IFT.

(A, B) Representative images of phasmid cilia (left) from N2 wild type and che-11(e1810) worms expressing GFP::RAB-28, together with corresponding kymographs and kymograph schematics derived from time-lapse imaging. Distribution plot (and mean values) in B shows kymography-determined anterograde and retrograde GFP::RAB-28 velocities from wild type worms. MS; middle segment, DS; distal segment, PC; periciliary membrane compartment, DD; distal dendrite. Scale bars; 3 μm (phasmid image; and horizontal bar on kymographs); 3 seconds (vertical bar on kymographs). (C) Fluorescence recovery after photobleaching (FRAP) plots for GFP::RAB-28 in the phasmid neurons of N2 and che-11(e1810) mutant worms. GFP::RAB-28 also undergoes free diffusion in wild-type (N2) and che-11 IFT mutant animals. Ciliary GFP signals bleached at time 0. Intensity measurements normalised to pre-bleach levels. Curves derived from 3 separate FRAP experiments. Error bars; SEM. Images taken from a representative FRAP experiment in N2 wild type worms. s; seconds. Scale bar; 2 μm.

https://doi.org/10.1371/journal.pgen.1006469.g003

We also examined genes excluded from filtered cluster 1 due to a reported lack of predicted human orthologues (according to the Ortholist database; [49]). Highly ranked in unfiltered cluster 1 is an uncharacterised gene with an X-box promoter element [26], tza-3, which we demonstrate is expressed exclusively in ciliated neurons, and encodes a ciliary TZ protein (S4B Fig). Whilst TZA-3 is not conserved, the TZ localisation of this predicted transmembrane protein depends on the highly conserved core TZ scaffolding protein MKS-5 and “MKS module” components MKSR-1 and MKSR-2, but not the “NPHP module” protein NPHP-4 (S4B Fig). These findings are consistent with TZA-3 associating with the MKS module, which plays a role in ciliary gating [6–8]. Thus, our expression profiling approach identifies non-conserved ciliary proteins in addition to conserved ciliary proteins.

Of the novel C. elegans ciliary proteins we identified, we sought to characterise RAB-28 further for the following reasons: (1) human RAB28 is associated with a possible ciliopathy (cone-rod dystrophy [62,63]), (2) many small GTPases play essential roles in ciliogenesis and ciliary membrane trafficking pathways [10], and (3) aside from reports of RAB28 localising at the basal body [62] or the cilium [64], there is a complete absence of molecular studies on this evolutionarily conserved protein that support a functional and mechanistic link to cilia.

RAB-28 undergoes intraflagellar transport (IFT)

We further investigated the ciliary localisation of GFP::RAB-28 using time-lapse imaging and found that this GTPase undergoes continuous, bidirectional, IFT-like movement along ciliary axonemes (S1 Movie). In contrast, the additional non-ciliary GFP::RAB-28 signals present throughout the neurons are diffuse, with no processive movement detected. In C. elegans amphid and phasmid cilia, anterograde IFT is driven by two kinesin-2 motors (kinesin-II and OSM-3) to yield distinct anterograde rates along the middle and distal segments [14,73]. Kymograph analysis of phasmid (tail) cilia confirm that ciliary GFP::RAB-28 moves at IFT-associated velocities, displaying characteristic average anterograde rates of ~0.7 μm/sec (middle segment; proximal part of axoneme) and ~1.2 μm/sec (distal segment; distal part of axoneme), and retrograde rates of ~1.4 μm/sec (along entire cilium length) (Fig 3A and 3B and S1 Movie) [73]. To examine if this ciliary trafficking is truly associated with IFT, we examined GFP::RAB-28 in worms with disrupted CHE-11 (IFT140), which is a component of IFT-A essential for IFT [28,74,75]. Although GFP::RAB-28 is observed within the truncated cilia of che-11 mutants, we could not detect processive movement of the GFP signals, indicating that the trafficking behaviour of RAB-28 within cilia is bona fide IFT (Fig 3A and S1 Movie). In contrast to observations for other IFT-associated proteins in IFT deficient-worms [28,74,75], GFP::RAB-28 does not abnormally accumulate at the ciliary base or tip, or along the axoneme of these animals (Fig 3A). This lack of accumulation, despite the active transport defect, suggests that our GFP::RAB-28 reporter also freely diffuses within (and between) the ciliary and dendritic compartments, in addition to undergoing IFT. We confirmed this hypothesis using a fluorescence recovery after photobleaching assay in wild type and che-11 mutant worms, which show that GFP::RAB-28 is highly mobile, displaying very rapid exchange kinetics between ciliary and dendritic pools (Fig 3C). Thus, in addition to active transport via IFT, GFP::RAB-28 undergoes IFT-independent free diffusion, which explains why this reporter does not abnormally accumulate within che-11 mutant cilia.

RAB-28 is an IFT cargo

To determine if RAB-28 is required for IFT, we examined ciliary structure, function and protein transport in worms containing a deletion (gk1040) in rab-28. The gk1040 mutation removes the GTP-binding switch II domain and the farnesylated C-terminal CAAX motif (S5A Fig), both of which are critical for RAB protein function [76]; thus, gk1040 is likely a severe loss-of-function or null allele. We outcrossed the gk1040 background at least 3 times with wild type worms to remove background mutations unlinked to gk1040. We find that ciliated amphid and phasmid sensory neurons in rab-28 mutant worms display a normal dye filling response, which suggests that ciliary structures are intact (short cilia usually abrogate dye uptake due to lack of environmental exposure) [77] (S5B Fig). Localisation of an IFT protein, OSM-6 (IFT52 orthologue), throughout normal-length cilia confirms this finding (S5C Fig). In addition, transmission electron microscopy (TEM) analyses reveal that ciliary ultrastructures appear normal in rab-28 mutants (S5D Fig). Furthermore, these worms are normal for cilia-related sensory behaviours (S5E Fig), as well as cilium-dependent carbon dioxide avoidance and development (body size) (S5F Fig) [78–80]. Finally, the localisation (and movement) of BBSome (BBS-5), IFT-A/B and various ciliary membrane proteins is also grossly normal in rab-28 mutants (S5G Fig). Thus, unlike the disruption of IFT proteins, loss of RAB-28 does not affect ciliary structure, transport, or function, at least for those proteins and cilia that were analysed. We conclude, therefore, that RAB-28 behaves more like an IFT cargo, or peripherally-associated component of IFT complexes, rather than a central component of the machinery required for bidirectional movement (IFT).

RAB-28 association with the periciliary membrane and IFT is dependent on GTP-binding and the BBSome subunit orthologue, BBS-8

Next, we assessed the requirements of GDP and GTP nucleotide binding for RAB-28 localisation and transport. We made GFP-tagged constructs containing T49N or Q95L mutations in RAB-28, which are predicted to trap the GTPase in GDP (inactive) or GTP (active)-locked states, respectively [81]. These constructs were injected into wild type worms at the same concentration (5 ng/μl) as the GFP::RAB-28(WT) construct described above (Fig 3), and all appear to be expressed at similar levels. Like the RAB-28(WT) reporter, RAB-28(T49N) (hereafter termed RAB-28(GDP)) displays diffuse localisation throughout the entire neuron, including the cilium and its associated periciliary membrane compartment (PCMC) (Fig 4A). However, unlike wild type RAB-28, RAB-28(GDP) fails to undergo detectable IFT, indicating a highly reduced ability to associate with IFT trains (Fig 4A and S1 Movie). In contrast, the localisation of the RAB-28(Q95L) (hereafter termed RAB-28(GTP)) is highly enriched at the PCMC compared with the rest of the cell, forming a ‘tunnel-like’ localisation indicative of association with the periciliary membrane (Fig 4A). RAB-28(GTP) also undergoes clearly detectable IFT, albeit at an apparently reduced frequency compared to the wild type reporter (Fig 4A and S1 Movie). Notably, the localisation and IFT behaviour of all three GFP::RAB-28 reporters (WT, GDP-locked, GTP-locked) is not altered when expressed in the rab-28(gk1040) null background (S6 Fig), indicating that endogenous RAB-28 does not affect the distribution and transport behaviour of the RAB-28 markers. Furthermore, transgenic worms expressing GFP::RAB-28(GTP) at a very low level (injected at 0.5 ng/μl; oqEx304) is exclusively localised at the periciliary membrane and undergoes IFT (S7 Fig and S2 Movie), thus confirming our observations with the higher expressing marker, and indicating that the periciliary membrane is the primary site of activated RAB-28 function in nematode sensory neurons (Fig 4A). Taken together, these data show that GTP binding targets RAB-28 to the periciliary membrane and facilitates its association with IFT trains.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. BBSome-dependent recruitment of activated RAB-28 to the periciliary membrane.

(A, B) Representative images of amphid and phasmid cilia from N2 wild type (A) and bbs-8(nx77) mutant (B) worms expressing GFP-tagged RAB-28(WT), RAB-28(GDP) and RAB-28(GTP) reporters. All reporters are driven by the endogenous rab-28 gene promoter. Kymographs and kymograph schematics derived from time-lapse imaging of GFP signals in phasmid cilia. Phenotypes summarised in cartoons. Large phasmid images are placed on black backgrounds. m; middle segment, d; distal segment. pcmc; periciliary membrane compartment (also denoted by arrow). Scale bars; 2 μm and 5 μm (low magnification phasmid images).

https://doi.org/10.1371/journal.pgen.1006469.g004

These observations for C. elegans RAB-28 are reminiscent of those for mammalian RAB8, whose ciliary targeting also depends on GDP-GTP exchange, via a mechanism involving RABIN8 (RAB8 guanine nucleotide exchange factor) and BBSome functions at the base of cilia [82–84]. We therefore examined the localisation and transport behaviour of all three GFP-tagged RAB-28 reporters (WT, GDP-locked and GTP-locked) in a likely null allele (nx77) of a C. elegans BBSome gene orthologue, bbs-8. The bbs-8(nx77) mutation does not alter the ciliary targeting and distributions of RAB-28(WT) and RAB-28(GDP) compared to a wild type background (Fig 4B). In contrast, BBS-8 loss has a striking effect on both of our RAB-28(GTP) reporters, preventing their accumulation at periciliary membranes (Fig 4B and S7 Fig). Furthermore, the IFT movement of RAB-28(WT) and RAB-28(GTP) is abolished in bbs-8 mutant cilia (Fig 4B and S1 Movie). From these data, we conclude that the IFT-associated BBSome is required for targeting activated RAB-28 (GTP-locked) to periciliary membranes and loading onto IFT trains.

Overexpression of constitutively active or inactive RAB-28 induces distinct cell autonomous and cell-non autonomous sensory pore defects

The differences in localisation and IFT behaviour between the constitutively inactive (GDP-locked) and active (GTP-locked) forms of RAB-28 prompted us to investigate cilium structure and function in worms overexpressing these small GTPase variants. Assessment of dye-filling reveals reduced levels of dye uptake in the amphid neurons of wild type worms expressing RAB-28(GTP), compared to normal dye-filling for wild type worms expressing RAB-28(WT) or RAB-28(GDP) (Fig 5A). Similar dye-filling results were obtained when these variants were expressed in rab-28(gk1040) worms, although worms expressing RAB-28(GDP) may display a very weakly penetrant phenotypic defect (S8A Fig). Despite the distinct dye-filling phenotypes, worms expressing RAB-28(GTP) or RAB-28(GDP) both show a reduced roaming behaviour (Fig 5B), suggestive of cilia-related sensory abnormalities [78].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Worms overexpressing RAB-28(GTP) or RAB-28(GDP) display overlapping and distinct defects in sensory pore formation and function.

(A) Dye filling (DiI) of amphid neurons in non-transgenic (control) or transgenic wild type (N2) worms expressing GFP::RAB-28(WT), GFP::RAB-28(GTP) or GFP::RAB-28(GDP). All constructs driven by the endogenous rab-28 gene promoter and expressed to similar levels in transgenic animals (all injected at 5 ng/ul). For each amphid pore, the number of dye-filling neurons was scored. Each dataset represents mean ± standard deviation (error bars) from 3 independent experiments (at least 40 amphid pores scored for each strain per experiment). (B) Roaming scores for non-transgenic (control) and transgenic wild type (N2) worms expressing GFP-tagged RAB-28 variants. Scores normalised to non-transgenic N2 worms. For each strain, >45 worms were scored. *p<0.01 (unpaired Student t-test vs non-transgenic N2 controls). (C) Transmission electron microscopy images of the amphid pore from serial cross-section of wild type (N2) and N2 worms expressing GFP-tagged RAB-28(GTP) or RAB-28(GDP). Low magnification images (top rows) show the enlarged amphid channel (yellow outline and asterisk) in RAB-28(GTP)-expressing worms and dense matrix filled vesicles (mfv) in the amphid sheath cell of RAB-28(GDP) expressing worms (white arrows). High magnification images (bottom rows) display close ups of the amphid pore and ciliary axonemes, showing the distal segments (DS), middle segments (MS), transition zones (TZ) and periciliary membrane compartments (PCMC). Note the misplaced (disconnected) channel axoneme (white arrowhead) that fails to enter the channel of worms expressing RAB-28(GDP). Images representative of at least 4 analysed pores for each strain. Cartoons show the amphid channel in cross section and longitudinal orientations (only 3 of the 10 axonemes shown for simplicity in longitudinal cartoon), and indicate observed phenotypes. Numbers above images indicate the position of the section relative to the most anterior section (at ‘0’); section positions also indicated in cartoon. Scale bars; 1 μm (top rows); 200 nm (large images in bottom rows); 100 nm (small images in bottom rows).

https://doi.org/10.1371/journal.pgen.1006469.g005

To investigate the structural basis of these defects, we examined amphid pore ultrastructure in wild type (N2) worms expressing GFP-tagged RAB-28(GTP) or RAB-28(GDP). Somewhat surprisingly, despite the dye uptake phenotype, RAB-28(GTP)-expressing amphid channel cilia are of normal length and morphology, with intact middle and distal segments, as well as TZ compartments (Fig 5C). However, an increased number of ciliary axonemes possess B-tubule seam breaks (‘unzipped’ microtubules), indicative of incomplete protofilament formation (S8B Fig). A more striking phenotype is that the proximal region of the sheath cell-defined amphid channel is greatly enlarged, and the ciliary axonemes are less tightly bundled (Fig 5C). Also, the channels of RAB-28(GTP) expressing worms display reduced electron densities, potentially indicating abnormally low concentrations of matrix material, which is thought to be secreted by the sheath cell (Fig 5C) [43]. In contrast, although the amphid channel in RAB-28(GDP)-expressing worms is not expanded, these worms exhibit a number of abnormally large dense matrix-filled vesicles (MFVs) in the sheath, in that portion of the cell surrounding the channel (Fig 5C). Such electron dense vesicles are not found in the channel region of the amphid sheath cell of wild type (N2), rab-8(gk1040) or RAB-28(GTP) overexpressing worms (Fig 5C), indicating this phenotype is specific to RAB-28(GDP) overexpression. A second phenotype unique to RAB-28(GDP)-expressing worms is that one channel axoneme is missing from the amphid pore, and instead lies adjacent to the channel, embedded in the sheath (Fig 5C).

Together, these findings indicate that overexpression of constitutively active or inactive RAB-28 cause variant-specific defects in the amphid sheath (enlarged channel; dense MFV accumulation) and ciliated cells (dye-filling defect; misplaced amphid channel axoneme), although shared phenotypes are also observed (roaming defect). The sheath cell phenotypes likely represent a cell non-autonomous function because our transcriptional reporter for RAB-28 (GFP under the control of the rab-28 promoter) is not expressed in the sheath cell (S9 Fig). Thus, we conclude from these overexpression data that RAB-28 serves ciliated cell autonomous and sheath cell non-autonomous roles in sensory pore formation and function.

RAB-28 cell non-autonomous function

Another revealing observation from our work is that overexpression of constitutively active (GTP-locked) or inactive (GDP-locked) RAB-28 cause variant-specific defects in the non-ciliated amphid sheath cell that supports the ciliated endings of amphid neurons. Specifically, RAB-28(GTP) overexpression causes an enlargement of the sheath cell-defined portion of the amphid sensory pore, whereas RAB-28(GDP) overexpression causes the accumulation of abnormally dense matrix-filled vesicles (MFVs) in the sheath cell. The RAB-28 variant constructs, as well as the RAB-28(WT) construct, are expressed at similar levels, using the same promoter sequence; thus, the observed variant-specific phenotypes are not likely due to differences in expression level, but instead are linked to the nucleotide bound state of RAB-28. Most interestingly, the abnormal sheath cell phenotypes appear to reflect a cell-non autonomous effect because RAB-28 is expressed in ciliated sensory neurons, and not in the amphid sheath cell (S9 Fig).

In C. elegans, the amphid channel is fashioned from glial cell (sheath and socket) processes which extend to the nose tip in close proximity to the dendritic processes of the ciliated sensory neurons. The proximal part of the channel forms from a hole in the sheath process which the channel cilia fully penetrate, whereas the distal part of the channel derives from a doughnut-shaped ending of the socket process that fuses with the cuticle and the underlying sheath (Fig 5C). The formation and function of neurons and glia in the amphid pore appear tightly linked, with channel morphogenesis thought to depend on signals from both cell types [43,44,90]. It is also suggested that amphid channel size is regulated via a balance of ‘exocytic’ membrane delivery via MFVs (increases channel size) and ‘endocytic’ membrane retrieval (reduces channel size) at the sheath cell plasma membrane [91]. By integrating this model with our findings, we hypothesise that constitutively active RAB-28 may enhance MFV delivery to the sheath cell membrane, leading to an enlarged channel, whereas constitutively inactive RAB-28 blocks MFV delivery and results in abnormally dense MFVs in the sheath. One prediction of this model is that expression of constitutively inactive RAB-28 would reduce amphid channel size, and consistent with this idea, one channel axoneme fails to enter the amphid pore of RAB-28(GDP) expressing worms.

How RAB-28 expressed in ciliated amphid neurons might control morphogenesis events in the supporting non-ciliated sheath cell is unclear. One scenario is that RAB-28 regulates extracellular release of neuronal factors purported to signal to the sheath cell during channel morphogenesis in the embryo [42–44] (Fig 6). In support of this notion, activated RAB-28 is almost exclusively concentrated at the periciliary membrane, which is the site of ectosome release in certain nematode sensory neurons; also, RAB-28 expression is elevated in nematode ectosome-releasing cells [92,93]. Future work will be required to investigate possible roles for RAB-28 in cilia-related ectosome release, and to identify sheath cell morphogenic factors released by the ciliated neurons.

A surprising observation is that sensory pore structure and function appears grossly normal in the rab-28 deletion mutant, which is likely a null allele. This indicates that other genes or pathways can compensate for rab-28 loss-of-function, but that the phenotypic effects of the overexpressed dominant active and inactive RAB-28 variants are more severe and cannot be as easily compensated. One possibility is that the cilium structure, and cilium-dependent behavioural phenotypes in worms overexpressing RAB-28(GTP) and/or RAB-28(GDP) derive from excessive activation of downstream effectors or possibly aberrant sequestration of GTPase activating proteins (GAPs) or GDP-GTP exchange proteins (GEFs) common to RAB-28 and other ciliary GTPases. To fully decipher the mechanistic basis of the RAB-28 gain vs loss of function phenotypes, and the physiological relevance of our findings, future efforts should include an examination of worms expressing RAB-28 variants expressed at endogenous or near-endogenous levels and the identification of currently unknown regulators of RAB-28 (GEF and GAP proteins).

Possible functional relationship between RAB28 and IFT27?

Given the close sequence similarity between RAB28 and IFT27, and some of their common functional properties discussed below, we suggest that these two RAB GTPases may have similar or partially overlapping ciliary functions. RAB28 is conserved across metazoans and many ciliated protists (S10 Fig), and inferred to be one of 23 founding RAB family members present in the last eukaryotic common ancestor (LECA) [94,95]. Phylogenetically, RAB28 is most closely related to IFT27 (RABL4), which likely also existed in LECA and is dispersed throughout the eukaryotic lineage [94,95] (S10 Fig). Like the majority of ciliary and IFT components, both RAB28 and IFT27 are restricted to organisms that have cilia (S10 Fig). Interestingly, although RAB28 and IFT27 are present in most metazoans and many protists (e.g., Chlamydomonas and Trypanosoma have both), neither protein appears to be present in Drosophila and IFT27 is missing from C. elegans (S10 Fig).

RAB28 and IFT27 appear to share overlapping functional properties. As we show for nematode RAB-28, Trypanosome and mammalian IFT27 associate with IFT in a GTP-dependent manner [87,88,96]. Also, cilium formation is unaffected in mammalian cells with null mutations in IFT27 (similar to C. elegans rab-28 mutants), although disruption of IFT27 in protists results in short flagella [74,96]. In addition, mutations in IFT27 cause BBS, displaying a broad spectrum of ciliopathy phenotypes, including the cardinal retinal degeneration phenotype observed in RAB28 patients [97,98]. Furthermore, both small GTPases are functionally associated with the BBSome, albeit in different ways. The localisation and IFT motility of nematode RAB-28 is BBSome-dependent, but not vice versa (i.e., BBS-5 localisation is normal in rab-28 mutants), whereas mammalian IFT27 regulates BBSome assembly and removal from the cilium. When taken together, the above observations could suggest that in vertebrates RAB28 and IFT27 function at distinct steps of a common pathway, potentially ‘upstream’ (IFT27) and ‘downstream’ (RAB-28) of the BBSome [87,88]. Since IFT27 and the BBSome remove proteins from cilia [85,87,88,99], RAB28 may perform a similar function by facilitating the BBSome-mediated removal of a more-restricted set of cargo.

In further support of a common, yet tissue-specific, transport role for related ciliary GTPases such as IFT27 and RAB28, it is noteworthy that disruption of the short ARL6/BBS3 isoform in mice and zebrafish results in broad spectrum BBS-related phenotypes, whereas disruption of the long isoform causes a retina-specific phenotype [100]. Transition zone-localised RPGRIP1 and RPGRIP1L represent additional examples of very closely related ciliopathy proteins linked to a broad tissue pathogenesis (RPGRIP1L), versus a retinal-specific disease (RPGRIP1) [101–104]. Finally, whereas IFT27 is known to regulate the ciliary localisation of sonic hedgehog proteins [87,88], the identity of RAB28-associated cargoes is currently unknown. Interestingly, RAB28 itself may be a cargo of the prenyl-binding protein PDE6D (cGMP phosphodiesterase delta subunit) which immunoprecipitates RAB28 [105]; in agreement with this possibility, loss of PDE6D in the mouse phenocopies the photoreceptor degeneration of RAB28 patients [106]. In light of our worm data, it is possible that PDE6D delta functions with the BBSome, either together or at distinct parts of a transport pathway, to target RAB28 to ciliary membranes.

Concluding remarks

Our study describes a novel method for identifying genes with ciliary functions that complements other studies (bioinformatics, genomics, proteomics) aimed at deciphering a complete ‘ciliome’. From our list of candidates, we uncover several novel, evolutionarily-conserved cilium-associated proteins. Our in-depth analysis of RAB-28 in C. elegans reveals it to be a novel IFT- and BBSome-associated protein with cell autonomous and non-autonomous roles in maintaining the structural and functional integrity of cilia, and of the sensory neuron-supporting cells. Further analyses of mammalian RAB28 are required to reveal its functional similarities and differences with IFT27, and possible functional associations with PDE6D and ARL6, to help explain its seemingly specific function in ciliary photoreceptors, and thus specific association with cone-rod dystrophy.

Identifying genes with expression correlated with known ciliary genes

We curated a list of 41 C. elegans genes with established roles in ciliary function and used these to search for genes with correlated expression. These “bait” genes included: R148.1, F38G1.1, Y75B8A.12, F20D12.3, Y41G9A.1, R01H10.6, C54G7.4, C27A7.4, R31.3, C48B6.8, F32A6.2, C27H5.7, T28F3.6, Y37E3.5, F59C6.7, F19H8.3, ZC84.2, F35D2.4, H01G02.2, B0240.3, ZK520.3, C02H7.1, ZK418.3, K08D12.2, F23B2.4, Y105E8A.5, F46F6.4, K07G5.3, F33H1.1, F02D8.3, K03E6.4, C38D4.8, T26C12.4, T27B1.1, Y110A7A.20, F53A9.4, M04C9.5, C30B5.9, C09G5.8, R13H4.1 and F54C1.5. We used expression data derived from RNA-seq for seven developmental stages (EE = early embryo, LE = late embryo, L1 = mid-L1 larvae, L2 = mid-L2 larvae, L3 = mid-L3 larvae, L4 = mid-L4 larvae and YA = young adult) obtained from [45–47]. As the expression values were relative to individual isoforms, we first computed an average RPKM (reads per kilobase per million) value for each gene across all isoforms. Using these 41 genes, we next computed the pairwise Pearson correlation of each gene against all 20,363 C. elegans genes in the data set across all libraries and estimated a P value for each test. As we were interested in capturing genes with expression correlates to any of these known ciliary genes, we retained the smallest P value from these 41 tests for subsequent calculations. To avoid circularity, we retained the second-smallest P value for correlation tests involving one of the bait genes. In addition to direct correlation, we sought to identify genes exhibiting a developmental expression pattern similar to known ciliary genes. We observed that all ciliary genes exhibit at least a 10-fold reduction in expression level (relative to their peak expression level) in the latter three developmental stages. A total of 5960 genes showed this expression trend.

Following the identification of those genes exhibiting strongly correlated expression to any of the known ciliary genes, this set was further clustered to search for sub-groupings of genes with strongly coordinated expression. The distance between individual genes was calculated using the ‘dist’ function in R and hierarchical clustering was performed using single, average and complete linkage. S1 Fig shows a heat map representation of the expression for each of the candidate ciliary genes with the dendrogram derived from clustering using complete linkage and cluster 1 at the top.

Chemosensory expression pattern and promoter element prediction

Using the method described above, we used srg-36 as a bait to identify co-expressed genes. We found 80 genes (p < 0.001), including 27 other predicted chemoreceptors [52]. The 1000 bp upstream from the start codon of all 28 chemoreceptors (including the bait srg-36) were entered into the program MEME [107,108], which found a consensus that matches the published E-box sequence [54]. Separately, the 1000 bp upstream of all the conserved cluster 1 were entered into the MEME suite program DREME [107,108], which found several potential transcription factor binding sites. We used the MEME suite program TomTom to compare identified sites to known sites [108]. Identified promoter elements were compared to consensus splice and trans-splice sites and any that matched were subsequently removed.

C. elegans maintenance and crossing

All nematode strains were maintained and cultured at 20°C on nematode growth medium (NGM) plates seeded with OP50 Escherichia coli using standard techniques. Standard genetic crossing techniques were used to introduce transgenes into genetic backgrounds. The rab-28(gk1040) mksr-1(ok2092), mksr-2(tm2452), mks-5(tm3100), mks-3(tm2547), mks-6(gk674), and nphp-4(tm925) mutations were followed using genotyping PCR (primer sequences available upon request).

Strains

N2 (Bristol)

rab-28(gk1040)

che-11(e1810)

bbs-8(nx77)

N2;oqEx300[Prab-28::gfp + Punc-122::gfp]

N2;oqEx301[Prab-28::gfp::rab-28 + Punc-122::gfp]

N2;oqEx302[Prab-28::gfp::rab-28(T49N) + Punc-122::gfp]

N2;oqEx303[Prab-28::gfp::rab-28(Q95L) + Punc-122::gfp]

N2;oqEx304[Prab-28::gfp::rab-28(Q95L) + Punc-122::gfp]

bbs-8(nx77);oqEx301[Prab-28::gfp::rab-28 + Punc-122::gfp]

bbs-8(nx77);oqEx302[Prab-28::gfp::rab-28(T49N) + Punc-122::gfp]

bbs-8(nx77);oqEx303[Prab-28::gfp::rab-28(Q95L) + Punc-122::gfp]

bbs-8(nx77);oqEx304[Prab-28::gfp::rab-28(Q95L) + Punc-122::gfp]

che-11(e1810);oqEx301[Prab-28::gfp::rab-28 + Punc-122::gfp]

rab-28(gk1040);oqEx58[arl-13::gfp + rol-6(su1006)]

rab-28(gk1040);Is[osm-6::gfp]

rab-28(gk1040);myIs[pkd-2::gfp]

rab-28(gk1040);nxEx289[rpi-2::gfp + xbx-1::tdTomato + rol-6(su1006)]

rab-28(gk1040);nxEx475[bbs-5::gfp + pCeh361]

N2;nxEx869[fam-161::gfp + xbx-1::tdTomato + rol-6(su1006)]

N2;nxEx1157[Pbbs-8::gfp::rab-28 + xbx-1::tdTomato + rol-6(su1006)]

N2;nxEx250[rpi-1::gfp + xbx-1::tdTomato + rol-6(su1006)]

N2;nxEx1223[maph-9::gfp + xbx-1::tdTomato + rol-6(su1006)]

N2;nxEx1230[Pbbs-8::gfp::yap-1 + xbx-1::tdTomato + rol-6(su1006)]

N2;nxEx2619[Pccdc-149::gfp::ccdc-149 + xbx-1::tdTomato + rol-6(su1006)]

N2;nxEx2623[Pccdc-104::gfp::ccdc-104cDNA + xbx-1::tdTomato + rol-6(su1006)]

N2;nxEx2626[Pccdc-149::gfp + xbx-1::tdTomato + rol-6(su1006)]

N2;nxEx2628[Pccdc-104::gfp + xbx-1::tdTomato + rol-6(su1006)]

N2;nxEx654[tza-3::gfp + xbx-1::tdTomato + rol-6(su1006)]

mks-6(gk674);nxEx654[tza-3::gfp + xbx-1::tdTomato + rol-6(su1006)]

nphp-4(tm925);nxEx654[tza-3::gfp + xbx-1::tdTomato + rol-6(su1006)]

mks-3(tm2547);nxEx654[tza-3::gfp + xbx-1::tdTomato + rol-6(su1006)]

mks-5(tm3100);nxEx654[tza-3::gfp + xbx-1::tdTomato + rol-6(su1006)]

mksr-1(ok2092);nxEx654[tza-3::gfp + xbx-1::tdTomato + rol-6(su1006)]

mksr-2(tm2452);nxEx654[tza-3::gfp + xbx-1::tdTomato + rol-6(su1006)]

Dye-filling and sensory behavioural assays

For the dye-filling assay [109], worms were incubated in 200 μl of DiI solution (Invitrogen; diluted 1:200 with M9 buffer) for 30 min. After incubation, worms were recovered on seeded nematode growth medium plates for a further 30 min and then mounted on slides. Epifluorescence wide-field imaging under the red filter was used to image DiI uptake into the ciliated amphid and phasmid cells. For the roaming (foraging) assay, single worms were placed for 18 hours onto seeded NGM plates and track coverage assessed using a grid reference [109]. For the osmotic avoidance assay, 5–6 worms were placed within a ring of 8M glycerol (Sigma) supplemented with Bromophenol Blue (Alfa Aesar) on unseeded NGM plates and observed for 10 min. Worms that crossed the barrier were removed from the assay. Chemotaxis attraction assays (towards isoamyl alcohol) were performed as previously reported [109], with a chemotaxis index calculated at 30 min and 60 min.

Fluorescent reporter construction and transgenic strain generation

All constructs were generated by PCR fusion [110]. For the transcriptional (promoter) Prab-28::gfp construct, gfp amplified from pPD95.67 was fused to an 432-bp fragment of the 5’UTR of rab-28 that included the first 14 bp of exon 1 (start codon thymine mutated to guanine). For the transcriptional reporters for Pccdc-104 and Pccdc-149, gfp amplified from pPD95.77 was fused to 1997 and 395 bp, respectively, of the 5’UTR for each gene. For the translational (promoter + protein) fam-161, rpi-1, maph-9, ccdc-149 and tza-3 reporters, the entire genomic sequence (exons and introns) were fused to gfp including promoter sequences of 1373 bp, 2317 bp, 973 bp, 395 bp and 457 bp, respectively. For the translational reporter for rab-28 and yap-1 in Fig 2, 341 bp of the bbs-8 promoter followed by gfp were fused to the genomic sequence including 3’ UTR for each gene. For the ccdc-104 translational reporter, 1997 bp of the bbs-8 promoter followed by gfp were fused to the ccdc-104 cDNA and 3’ UTR. For the translational Prab-28::gfp::rab-28 reporter (Figs 3 and 4), the entire genomic sequence and 970-bp of the 3’UTR of rab-28 was first fused 5’ to a gfp fragment, amplified from pPD95.77. The resulting gfp::rab-28 amplicon was subsequently fused to a 422-bp fragment consisting of the 5’UTR (promoter) sequence of rab-28. To make the Prab-28::gfp::rab-28(GDP) and Prab-28::gfp::rab-28(GTP) constructs, primers incorporating the corresponding T49N and Q95L mutations were first used to amplify 5’ and 3’ fragments of the rab-28 genomic sequence, and these were fused by PCR to establish the rab-28(T49N) and rab-28(Q95L) amplicons. These amplicons were subsequently fused to a Prab-28::gfp fragment, generated by fusing the rab-28 promoter sequence (see above) to gfp amplified from pPD95.77. Transgenic worms expressing the above constructs were generated by gonadal transformation of N2 hermaphrodites via microinjection and subsequent screening for transgenic progeny. rab-28 constructs were injected at a concentration of either 5 ng/μl (all translational constructs except oqEx304 which was injected at 0.5 ng/μl) or 50 ng/μl (transcriptional constructs), together with a coelomocyte (Punc-122::gfp) or pRF4 (rol-6(su1006)) co-injection marker at 50–100 ng/μl.

C. elegans fluorescence microscopy, IFT motility and FRAP assays

Strains in Fig 2 and S4B Fig were anaesthetised with 10 mM levamisole in M9 buffer, mounted on slides with 8% agarose pads, and observed by epifluorescence or spinning-disc (WaveFX; Quorum Technologies) confocal microscopy performed on an Zeiss Axio Observer Z1 with a Hamamatsu 9100 EMCCD camera. Image capture and visualisation were performed on Volocity (PerkinElmer). All other worms were immobilised with 40mM tetramisole (Sigma no. L9756) or microbeads (Polysciences no. 00876–15) and mounted on 4% or 10% agarose pads. Epifluorescence images were taken on an upright Leica DM5000B and confocal images on an inverted Nikon Eclipse Ti microscope with a Yokogawa spinning-disc unit (Andor Revolution). Images were acquired using a charge-coupled device camera (iXon+EM-CCD, Andor Technology) and analysed using Image J software. For IFT assays, time-lapse (multi tiff) movies of IFT along phasmid cilia were taken at 200 ms exposure and 4 fps. Separated anterograde and retrograde kymographs were generated from multi tiff files using Icy image analysis software (http://icy.bioimageanalysis.org/) and rates determined using ImageJ [111]. Fluorescence recovery after photobleaching (FRAP) assays were performed using the above confocal microscope with an attached FRAPPA unit (Andor Technology). Samples were bleached using a single pulse of the 488nm laser at 100% with a dwell time of 100 μs. Images were recorded immediately post-bleach and continuously thereafter. Fluorescence intensities measured with image J software following a recently described protocol [109]. Values were normalised to pre-bleach values and corrected for signal intensity loss during image acquisition.

Transmission Electron Microscopy

Young adult worms were fixed, sectioned and imaged as previously reported [109].