When we commenced our bioinformatic analyses of zebrafish pkd genes, the reference sequence XM_009294890 aligned in Ensembl Zv9 with a 12843 bp transcript, ENSDART00000039911 (associated with gene ENSDARG00000030417 on chromosome 1) called pkd1, that lacked both start and stop codons (Figure 1A; Table 1). However, this annotation changed in the current genome assembly, GRCz10, which contains a revised 15472 bp ENSDART00000039911 transcript that lacks the first 2607 nucleotides present in the older transcript. Supporting the older 5′ sequence, Coxam et al. (2014) showed enriched expression in zebrafish trunk at late embryonic stages using an in situ hybridization riboprobe designed against nucleotides 1307–1838 of the older transcript. We have also amplified this region from zebrafish cDNA and, in our hands, a riboprobe designed against this region is strongly expressed in embryonic pronephros, consistent with pkd1 expression in other animals (see expression analyses below). This expression is identical, although stronger, to the expression that we see when we use an alternative riboprobe designed against a more 3′ region included in the newer Ensembl transcript (see Section Materials and Methods and Figure 1A). In addition, Mangos et al. (2010) describe a splice-blocking pkd1 morpholino aligned with the older transcript (Figure 1A), that induced kidney cysts in some animals, consistent with Pkd1 function in other animals. Taken together, these data suggest that the current Ensembl annotation is incorrect and that at least some pkd1 transcripts include parts of the older upstream sequence.

Therefore, we used inverse PCR and overlapping PCR amplicons, to identify/map the correct pkd1 sequence (see Section Materials and Methods, Figure 1A and Table 1). All but 173 bases of the older ENSDART00000039911 transcript (Zv9) are present in our newly mapped pkd1 transcript as are all but the first 45 nucleotides of the current ENSDART00000039911 transcript (GRCz10). However, compared to our new transcript, there are some gaps in the Zv9 transcript, and we have also identified novel 5′ sequence that is not present in either transcript nor the 5′ genomic sequence in GRCz10, suggesting that there may be sequence missing from chromosome 1 in Ensembl. Taken together, our data suggest that a transcript of at least 18401 bp exists that encodes a 4608 amino acid protein and we have deposited this sequence in NCBI (accession number KY074550). This sequence lacks a start methionine. There is a methionine codon in-frame at 527–529 nucleotides (Figure 1A), but the region upstream of this methionine encodes a leucine-rich-repeat domain which is conserved in mouse, human and stickleback PKD1. Therefore, we think that this methionine is unlikely to be the start codon. We consider that the start codon is more likely to be a putative in-frame methionine 54 nucleotides upstream of our present transcript. We are confident that this gene is pkd1, given our synteny and phylogeny analyses discussed below (see also Table 1).

In contrast to pkd1, since the inception of this study the annotation of ENSDARG00000033029, the gene called pkd1b in our initial bioinformatic searches, has remained unchanged within Ensembl. The current longest pkd1b transcript, ENSDART00000153412.2, encodes a 3817 amino acid protein (Table 1, Figure 2). This Ensembl sequence is strongly supported by the reference sequence XM_017358624.1, which has the Gene ID 565697 (https://www.ncbi.nlm.nih.gov/gene/), with the exception that this reference sequence encodes an additional 73 amino acids at the amino terminus. Therefore, we cannot rule out the possibility that the pkd1b transcript might be longer than that currently shown in Ensembl. However, this would not affect the predicted domain structure of Pkd1b protein, as the additional 73 amino acids do not contain any additional predicted protein domains (Figure 2). We are confident that this gene is pkd1b based on the protein domains that it encodes (Figure 2) and our phylogeny analyses discussed below (see also Table 1). However, it is worth noting that, despite its name and the absence of this gene in mammals, we do not think that this gene is a teleost duplicate of pkd1 (see Section Discussion below).

The reference sequence XM_009297371, called pkd1l1 in our initial analyses, aligns with the forward strand of chromosome 24. In Ensembl Zv9 this region contained a small ORF of 225 amino acids, called pkd1l3 (2 of 4). In GRCz10, this region of homology is now called BX005392.1 (gene ENSDARG00000099162). The longest transcript at this locus is ENSDART00000169516.1, which encodes a protein of 2153 amino acids. Whilst there is no stop codon in either ENSDART00000169516.1 or an additional shorter transcript associated with BX005392.1, there is a putative stop codon in-frame 27 bp downstream in the 3′ flanking sequence of ENSDART00000169516.1. To assess whether this locus might encode a pkd gene, we generated two alternative riboprobes, one designed against exons 39–43 and the other against exons 46–49. As described below, both of these riboprobes labeled putative taste receptors and we also saw expression in dorsal forerunner cells / Kupffer's vesicle similar to pkd1l1 expression in medaka (Kamura et al., 2011). The structure of this protein, including its shorter polycystin-cation-channel domain and fewer PKD domains is consistent with that of PKD1L1 in other vertebrates (Figure 2). In addition, our synteny and phylogenetic analyses described below also suggest that this gene is pkd1l1.

The XM_002662913 (used to be called pkd1l3) reference sequence aligns with the reverse strand of chromosome 7. In an early release of Zv9, this region contained two adjacent novel genes, ENSDARG00000074116 and ENSDARG00000090210, encoding proteins of 488 and 294 amino acids respectively (Figure 1B). Temporarily these genes received the annotations pkd1l3 (1 of 4) and pkd1l3 (3 of 4), before being retired from the final release of Zv9. Given the proximity of these genes to one another on the same chromosome and their consecutive alignment with the XM_002662913 reference sequence, we hypothesized that these genes might constitute different parts of one longer gene. Consistent with this, riboprobes generated against each gene produced identical expression patterns in zebrafish ventral spinal cord and putative taste receptors (Figure 1B; and see expression analyses below).

In GRCz10 this locus is now called pkd1l2a. A single 17-exon transcript, ENSDART00000173234.1, is predicted to encode an 1124 amino acid protein. However, the 5′ sequence of this transcript only partially overlaps the 3′ sequence of ENSDARG00000090210 and is, therefore, likely to be incomplete (Figure 1B). It is also likely to contain inaccuracies because sequence present in our ENSDARG00000074116 riboprobe is annotated as being intronic in ENSDART00000173234.1.

To identify the correct gene sequence, we used the XM_002662913 sequence, which aligns with both of the retired pkd1l3 genes as well as sequence upstream of them, to map the mRNA transcript using RT-PCR. Since a preliminary protein domain search of the reference sequence suggested we might be missing amino-terminus sequence, we used the RefSeq GFF3 annotation import track in GRCz10 to identify a putative locus, LOC101884812, immediately upstream (Figure 1B). A protein domain search of this sequence using the Pfam protein database identified lectin and galactose-binding-lectin domains typically found in the amino-terminus of PKD1L2 proteins (Figure 2). We confirmed that these regions were transcribed using RT-PCR (Figure 1B). We also performed inverse PCR to identify any 5′ coding sequence that might be further upstream (Figure 1B). Using these approaches we identified an 8526 bp mRNA transcript, comprising 41 exons, both 5′ and 3′ UTR sequences and encoding a protein of 2485 amino acids that is identical in domain structure and length to mouse PKD1L2 (Figure 2, Table 1). This transcript encompasses both LOC101884812 and XM_002662913 and contains additional exons not present in either of these sequences (Figure 1B). There is also considerable overlap between our transcript and the current ENSDART00000173234.1 transcript (Figure 1B). The structure of the protein encoded by our mapped transcript, including its lack of a PKD domain and the presence of a galactose-binding-lectin domain, is consistent with PKD1L2 proteins in other animals (Figure 2). Our synteny and phylogenetic analyses (see below) also suggest that this is a pkd1l2 gene. Our pkd1l2a transcript has been deposited at NCBI (accession number KY074551).

In Zv9 the XM_009303604 reference sequence partially aligned with pkd1l2 (ENSDARG00000088121) on the forward strand of chromosome 7 (1–2985 bp) and to a non-coding region on the forward strand of Zv9_Scaffold3511 (1782–6344 bp), suggesting that the pkd1l2 annotation on chromosome 7 might be incomplete (Figure 1C). Consistent with this, the 5′ coding sequence of the longest ENSDARG00000088121 transcript, ENSDART00000156286, lacked a start codon and only encoded a protein of 859 amino acids. This gene was subsequently renamed pkdrej (2 of 2) in a later Zv9 release. Interestingly, in that same genome release an additional gene, pkdrej (1 of 2), encoding a protein of 124 amino acids, was reported immediately downstream of pkdrej (2 of 2). Our PFAM protein domain analysis revealed that these “Pkdrej” proteins contained classic features of Pkd proteins (Lectin C-type domain, galactose-binding-lectin domain, and GPS motif [Pkdrej (2 of 2)] and PLAT/LH2 domain [Pkdrej (1 of 2)]. However, neither protein contained the REJ domain, present in amniote PKDREJ proteins (Figure 2), nor the polycystin-cation-channel domain present in all other Pkd proteins. By the first release of GRCz10, pkdrej (2 of 2) had become si:ch211-168k15.4 and parts of pkdrej (1 of 2) had been included in the largest transcript of a new gene, ENSDARG00000101214, called pkd1l3, although in the most recent release of GRCz10 (version 86.10), ENSDARG00000101214 is named pkd1l2. These two genes overlap each other (Figure 1C). Exons 10–16 and the first 231 bases of exon 17 of si:ch211-168k1.5.4 are identical to exons 2–9 of ENSDARG00000101214. However, the si:ch211-168k15.4 transcript utilizes a stop codon present in intron 9–10 of ENSDARG00000101214, and the transcript for ENSDARG00000101214 (ENSDART00000124969.2) utilizes a start codon present in exon 10 of si:ch211-168k15.4. Since the protein encoded by ENSDART00000124969.2 is 1442 amino acids long and contains a polycystin-cation-channel domain, we tested whether this transcript and si:ch211-168k15.4 might actually be part of a larger pkd gene using RT-PCR. Since the 5′ coding sequence is incomplete in si:ch211-168k15.4 we also performed inverse PCR to identify any 5′ coding sequence that might be further upstream (Table 1). These analyses identified a longer, combined 7898 bp transcript that contains both the si:ch211-168k15.4 and ENSDARG00000101214 sequences, utilizing a start codon 4 bases upstream of exon 1 in the current si:ch211-168k15.4 annotation, and transitioning between exon 16 of si:ch211-168k15.4 immediately into exon 9 of ENSDART00000124969.2. Whilst the identification of the start codon is unique to this study and we have also identified novel 3′ UTR sequence, nucleotides 683–7026 of our new 7898 bp mRNA transcript align perfectly with the original XM_009303604 6344 bp reference sequence. The resulting 1902 amino acid protein has very similar domains to Pkd1l2a (Table 1, Figure 2) and its polycystin-cation-channel domain is most similar to that of Pkd1l2a, with which it has >60% identity, more than 30% higher than with any other zebrafish Pkd protein (Table 2). The lack of a PKD domain and the presence of a galactose-binding-lectin domain are also consistent with PKD1L2 proteins in other animals, with the exception that, unlike zebrafish Pkd1l2a and amniote PKD1L2, this protein is missing a REJ domain. Our synteny and phylogenetic analyses, discussed below, also suggest that this is a pkd1l2 gene. Therefore, we are confident that this gene is pkd1l2b and we have deposited the transcript sequence at NCBI (accession number KY074552).

During this study the annotation of the gene called pkd2 remained unchanged within Ensembl. The pkd2 mRNA reference sequence identified at the start of this study, DQ175629.1, aligns with the 14 exons present in the current Ensembl pkd2 transcript, ENSDART00000020412.7, although the latter contains additional UTR sequence. This suggests that pkd2 (ENSDARG00000014098) encodes a 904 amino acid protein. Our synteny and phylogeny analyses (see below) confirm that this gene is pkd2. Similar to PKD2 proteins in other vertebrates, the main conserved domain in the encoded protein is the polycystin-cation-channel domain (Figure 2, Table 1).

The annotation of the gene called pkd2l1 also remained unchanged within Ensembl during this study. The pkd2l1 mRNA reference sequence XM_690312 aligns with 100% homology to exons 1–9, 11–12, and 14–15 of the current pkd2l1 transcript ENSDART00000145948.1, which contains the additional exons 10 and 13. This suggests that pkd2l1 generates a 790 amino acid protein and is encoded by ENSDARG00000022503. Consistent with this, our in situ hybridization riboprobe (see expression analyses below), which was designed against the 3′ coding and UTR sequence present in ENSDART00000145948.1 (nucleotides 2062–2614), generated data similar to expression reported using a riboprobe designed against more upstream sequence (nucleotides 1148–2022; Djenoune et al., 2014). As for zebrafish Pkd2, and PKD2 family proteins in other vertebrates, the main conserved domain in this protein is the polycystin-cation-channel domain (Figure 2, Table 1). Our synteny and phylogeny analyses (see below) also suggest that this gene is pkd2l1.

We also performed additional bioinformatics searches using a newer version of the zebrafish genome released during our study, GRCz10, to test if there were any additional potential pkd genes. For this, we identified peptide sequences for the polycystin-cation-channel domain in each of the zebrafish pkd genes and performed Tblastn analyses with each of these sequences against this newer version of the genome (see Section Materials and Methods). We used the polycystin-cation-channel domain because this is the domain that defines Pkd proteins (Figure 2). When we did this, several of the domains identified other already-identified pkd genes (Table 1), but no new pkd genes were identified.

Since the zebrafish pkd genes that we had identified included only one set of potential teleost-duplicates or ohnologs, (just pkd1l2a/pkd1l2b as we do not think that pkd1 and pkd1b are teleost duplicates because these genes exist in both cartilaginous and holostei fish as described below), we further investigated whether teleost ohnologs of any additional pkd genes might exist by searching for pkd genes in medaka, stickleback and green spotted pufferfish. We performed a textual search for pkd genes, and also blasted the polycystin-cation-channel domain for all seven zebrafish Pkd proteins, against each of these genomes. We identified the same complement of seven genes in both stickleback and green spotted pufferfish and six genes in medaka (pkd1b was missing; Table 1). For each zebrafish Pkd protein, the polycystin-cation-channel domain had greatest homology with the gene that our phylogeny and synteny analyses suggest is its closest ortholog in each of the other teleost genomes (Table 1). Whilst Pkd1l2b in green spotted pufferfish and stickleback have slightly higher sequence homology with the polycystin-cation-channel domain of zebrafish Pkd1l2a than Pkd1l2b, Pkd1l2a in both of these teleosts shows even higher sequence homology with zebrafish Pkd1l2a. Therefore, these data, together with our phylogeny and synteny analyses, suggest that we have correctly classified the genes that encode these proteins (Table 1). We found no evidence in any of the teleosts examined for additional duplicate (ohnolog) pkd genes.

Interestingly, our Tblastn analyses with full-length zebrafish Pkd1b identified a small region of homology in each of the teleost genomes with the PLAT/LH2 region of Pkd1b (data not shown). Visual inspection of each of these loci revealed conserved synteny with the region surrounding amniote PKD1L3 genes (Figure 4E). To assess whether these teleost genomes might contain pkd1l3 orthologs, we performed Tblastn analyses with full-length mouse PKD1L3 (Supplementary Table 4). These identified the same loci. The putative pkd1l3 locus is not annotated in either zebrafish or medaka genomes, although a transcript is present in a previous zebrafish genome assembly (ENSDARG00000091803 in Zv9, Table 1). In the green spotted pufferfish and stickleback genomes, this locus contains a novel gene. Consistent with our Tblastn analyses with polycystin-cation-channel sequences, we have not detected these sequences encoding this domain within any of these loci. Given that the polycystin-cation-channel domain is the one domain that is present in all PKD proteins, we do not consider these sequences bona-fide pkd genes and therefore have not analyzed them further in this study.

Given that we found sequences with homology to part of the pkd1l3 gene, to confirm that pkd212 and pkdrej are absent in teleosts we performed Tblastn with full-length mouse PKD2L2 and PKDREJ against all of the teleost genomes discussed above. Both of these analyses only produced alignments with already identified Pkd proteins. Therefore, we are confident that there are no pkd212 or pkdrej genes in these teleost genomes.

To investigate potential relationships between the zebrafish pkd genes we aligned the polycystin-cation-channel domains for each of the Pkd proteins and determined the percentage identity of this domain between each of them. Pkd1l2a and Pkd1l2b have the highest identity (62%; Table 2), which is consistent with them being recently duplicated genes. Pkd2l1 and Pkd2 also have a high degree of identity at 57%. However, all of the other pair-wise comparisons have <30% identity at the amino acid level.

To investigate the evolution of zebrafish pkd genes we identified all of the PKD genes in both spotted gar (Lepisosteus oculatus; a holostei fish) and elephant shark (Callorhinchus milii; a cartilaginous fish). In both of these species we found 8 pkd genes: pkd1, pkd1b, pkd1l1, pkd1l2, pkdrej, pkd2, pkd2l1, and pkd2l2 (Figure 3, Supplementary Table 5). Unlike mammals, spotted gar, and elephant shark both have a pkd1b gene. In contrast, similar to mammals and unlike teleosts they only have one pkd1l2 gene (Supplementary Table 5). However, like teleosts, spotted gar only has a partial pkd1l3 sequence that lacks the polycystin-cation-channel domain but is located in a region with conserved synteny with other pkd1l3 regions (data not shown). It is currently less clear whether a pkd1l3 gene exists in elephant shark. In the mammalian, teleost and holostei genomes that we have examined, PKD1L3 is always located close to a gene called DHODH (Figure 4E). dhodh is located on Scaffold_12 of the elephant shark current genome, but we did not find any evidence for a pkd gene nearby (data not shown), although we cannot rule out the possibility that this gene exists elsewhere in the genome.

To confirm orthologous relationships between teleost and mammalian PKD1-family and PKD2-family genes we performed phylogenetic analyses of human (Homo sapiens), mouse (Mus musculus), spotted gar (Lepisosteus oculatus), elephant shark (Callorhinchus milii), zebrafish (Danio rerio), medaka (Oryzias latipes), green spotted pufferfish (Tetraodon nigroviridis), and stickleback (Gasterosteus aculeatus) PKD1-like and PKD2-like proteins. We used the region of the polycystin-cation-channel domain that was present in all of the proteins and both Neighbor-Joining (NJ) and maximum likelihood methods (see Section Materials and Methods, Supplementary Figures 1–2, Figure 3). In the resulting phylogenetic trees (Figure 3; data not shown), all of the zebrafish proteins cluster with the expected proteins from other species, suggesting that their annotations are correct. Consistent with Pkd1l2a and Pkd1l2b being teleost duplicates of PKD1L2 proteins in other vertebrates, all of the PKD1L2 proteins cluster together. Interestingly, in the maximum likelihood tree, mammalian PKD1L3 genes are also contained in this cluster, although this was not the case in the NJ analysis (Figure 3 and data not shown). However, we are confident that none of the teleost pkd1l2 genes are pkd1l3 genes as we have also found partial pkd1l3 genes in teleosts as discussed above.

To further test whether we had correctly identified orthologous relationships, we also examined the genomic regions around each of the zebrafish pkd genes and their proposed orthologs in other vertebrates for conserved syntenic relationships with other neighboring genes. We found that the pkd1 genomic locus contains both a NTHL1 and TSC2 gene in humans, mouse and all four teleost species (Figure 4A). In contrast, whilst the genomic regions around green spotted pufferfish and stickleback pkd1b share synteny with each other, none of the genes in this region are found near zebrafish pkd1b (Figure 4B). The PKD1L1 locus has considerable shared synteny between human and mouse and between teleosts, but the teleost loci don't have any obvious shared synteny with the amniote loci (Figure 4C). In contrast, PKD1L2 is located near a GCSH and a BCO1 gene in both mammals and at least one teleost and there are other genes found in common near most of the teleost pkd1l2a genes (Figure 4D).

The PKD2 locus, like the PKD1 locus, also has some conserved synteny between different vertebrates (Figure 5A). This genomic region contains an ABCG2 and PPM1K gene in humans, zebrafish, green spotted pufferfish, medaka, and stickleback. While the mouse Pkd2 locus doesn't seem to contain these genes—it shares three other genes with the human locus. The Pkd2l1 locus also has considerable conserved synteny between different teleost genomes and between mouse and human. In addition, a couple of genes are present in both mammals and at least one teleost: CHUK (present in mammals and zebrafish) and a SEC gene—SEC31B (present in mammals) and sec23lp (present in zebrafish and stickleback; Figure 5B).

Based on all of these analyses, we are convinced that the genes that we have identified are indeed pkd1, pkd1b, pkd1l1, pkd1l2a, pkd1l2b, pkd2, and pkd2l1 and that these are the only bona-fide pkd genes in zebrafish and the other teleosts that we have examined.

As previous studies have established that pkd2 is expressed in Kupffer's vesicle (KV) during early zebrafish embryogenesis (Bisgrove et al., 2005; Schottenfeld et al., 2007; Roxo-Rosa et al., 2015), we investigated expression of all seven zebrafish pkd genes in dorsal forerunner cells/KV at 8.3, 10, and 12 h (Figure 6). We found no expression of pkd1, pkd1b, pkd1l2a, pkd1l2b, and pkd2l1 at any of these stages (Figures 6J–S), with the exception that there is some spinal cord expression of pkd1b at 12 h (arrows in Figure 6M). In contrast, pkd1l1 was expressed in a cluster of dorsal forerunner cells at 8.3 h (Figure 6D), that later condense to form the KV at 10 and 12 h (Figures 6E–F). pkd2 expression was not detected at 8.3 h (Figure 6G), but was present in a group of cells in the KV region at 10 h, resolving in to a ring of cells surrounding the KV by 12 h (Figures 6H–I).

It has already been reported that pkd2 expression is enriched in developing zebrafish pronephros (Bisgrove et al., 2005; Schottenfeld et al., 2007), and Pkd2 protein is present in zebrafish kidney epithelial cells (Obara et al., 2006). However, before this study, it was unknown if any other pkd genes are expressed in the developing zebrafish pronephros. Therefore, we examined expression in WT embryos (8.3, 10, 12, 24, 27, 30, 36, and 48 h, 3, 4, and 5 dpf). We found no expression of pkd1b, pkd1l1, pkd1l2a, pkd1l2b, or pkd2l1 in the developing pronephros at any of these stages (Figures 6–9 and data not shown). In contrast, while pkd1 and pkd2 are not expressed in presumptive pronephric mesoderm at 8.3, 10, or 12 h (Figures 6G–K), they are both expressed in the pronephros by 24 h, although the expression of pkd1 is much stronger than that of pkd2 (Figures 7A,K, 9O,P,A',B'). The expression of pkd1 in the pronephros persists until 3 dpf. We also found that weak pkd2 expression persists in the pronephros during the pharyngula period (arrows in 27 and 36 h) but it was not enriched above the otherwise ubiquitous expression of pkd2 by 2 dpf (Figures 10A–D and data not shown).

Zebrafish pkd2l1 was recently shown to be expressed in a unique population of spinal cord cells, called Kolmer Agduhr (KA) cells or cerebrospinal fluid-contacting neurons (CSF-cNs; Djenoune et al., 2014). In addition, pkd1b has been reported as being expressed broadly in the spinal cord at late somitogenesis stages but restricted to medial floor plate ependymal cells by 3.5 dpf (Mangos et al., 2010) and low level expression of pkd2 has been observed in the floor plate during somitogenesis stages (Bisgrove et al., 2005; Schottenfeld et al., 2007). Therefore, we were very interested in investigating expression of pkd genes in spinal cord cells, particularly to see if we could identify a potential partner for Pkd2l1 in KA cells and/or additional Pkd proteins expressed in the floor plate.

We examined spinal cord expression of all 7 pkd genes in WT embryos at 12, 24, 27, 30, 36, and 48 h, and 3, 4, and 5 dpf. In addition, we examined expression in mindbomb mutants at 24 h. mindbomb encodes an E3-ubiquitin-ligase that is required for efficient Notch signaling. In mindbomb mutants Notch signaling is lost and, as a result, the vast majority of spinal cord progenitor cells precociously differentiate as early-forming populations of spinal cord neurons at the expense of later forming neurons and glia (Jiang et al., 1996; Schier et al., 1996; Itoh et al., 2003; Park and Appel, 2003; Batista et al., 2008). Therefore, comparing expression of genes in the spinal cords of mindbomb mutants and WT embryos enables us to distinguish between progenitor domain expression (which should be lost) and post-mitotic expression (which is often, although not always, expanded). In addition, if a gene is expressed very weakly in post-mitotic spinal cord cells, its expression is usually easier to observe in mindbomb mutants, where the expression is often expanded and stronger (Batista et al., 2008). Therefore, examining expression in mindbomb mutants also helps us to be more confident about whether a gene is expressed in the spinal cord or not.

Both WT and mindbomb mutant expression analyses suggest that pkd1, pkd1l1, and pkd1l2b are not expressed in zebrafish spinal cord (Figures 7A,B,E,F,I,J, 8S–A', 9M–X and Supplementary Figures 3G–U). It is likely that pkd2 is also not expressed in spinal cord, other than very weakly in floor plate (Figures 7K,L, 8B'–D', 9Y–B' and Supplementary Figures 3V–Z). In contrast, pkd1b was expressed broadly in the spinal cord and pkd1l2a and pkd2l1 were both expressed in post-mitotic cells in ventral spinal cord (Figures 7–9, 11, see more detailed descriptions below).

pkd1b is expressed very broadly throughout the dorsal/ventral extent of the spinal cord at 24 h in what appears to be mainly progenitor cells (Figures 8A, 9A,C). By 27 h, expression in the most dorsal part of the spinal cord has reduced. By 30 h the expression has resolved into two broad domains in the ventral spinal cord, one just above the notochord and one in the middle of the dorsal/ventral axis. This continues until 5 dpf, at which point expression in the more dorsal domain is much weaker (Figures 8B–F). Consistent with pkd1b being expressed by progenitor cells, most of its spinal expression is lost in mindbomb mutants, with the exception of the floor plate expression, which remains. This suggests that pkd1b may be expressed in floor plate in addition to being broadly expressed in spinal cord progenitor cells (Figures 9B,D). Consistent with this, we observe pkd1b expression in the floor plate of the hindbrain from 24 h until at least 5 dpf (insets in Figures 12 E'–G').

pkd1l2a and pkd2l1 have very similar spinal expression patterns. Both genes are already expressed in two rows of ventral cells by 24 h (Figures 8G,M) as well as being expressed more weakly in occasional more dorsal cells (Figures 8, 11). This expression also extends into caudal hindbrain (insets in Figures 12M–O, Y–A'). Unusually for post-mitotic spinal cord cells, but consistent with KA cells, these cells are located medially in ventral spinal cord, in positions where they can contact the CSF-containing central canal (Figures 11A,B, 9E,G,I,K). The expression of both of these genes continues throughout all of the stages that we examined, although the two rows become less distinct at later stages and pkd2l1 expression seems to become weaker in the more ventral row of cells by 4 dpf (Figures 8H–L,N–R). To confirm that these genes are both expressed by KA cells, we performed double labels with Tg(−8.1gata1:gata1-EGFP) zebrafish which express GFP in both KA and V2b cells in the spinal cord (Kobayashi et al., 2001; Batista et al., 2008). Consistent with the previous report (Djenoune et al., 2014), we find that pkd2l1 is expressed in all ventral KA″ cells and more dorsal KA′ cells. In addition, we show for the first time that zebrafish pkd1l2a is also expressed in both of these cell types (Figures 11C–I').

We also detected expression of two pkd genes in specific territories of the ear at 4–5 dpf (Figures 10E–H). We observed pkd1l1 and pkd2l1 expression in the ectoderm of the inner ear that supports the posterior canal and posterior crista at 4 dpf. pkd1l1 is also weakly expressed in the utricular otolith. By 5 dpf, pkd1l1 expression persists in the utricular otolith and the underlying utricular macula. It is also expressed in the neighboring ectoderm flanking the lateral canal and lateral crista (Figures 10E,F). In contrast, the expression of pkd2l1 persists in the tissue surrounding the posterior canal and posterior crista (Figures 10G,H).

Interestingly, we also detected expression of pkd1, and only pkd1, in the neuromasts (asterisks, Figures 10I,J,L,M) and lateral line primordium (white dotted line, Figures 10I,K). This expression was first apparent at 36 h and it persists in the neuromasts until 3 dpf, but is lost by 4 dpf.

Similarly, out of all 7 pkd genes, we only detected expression of pkd1 in the pectoral fin buds. pkd1 is expressed in these structures at 36, 48 h, and 3 dpf, but we could not detect expression above background by 4 dpf (arrows in Figures 10N,O and data not shown). At 36 and 48 h, pkd1 expression is ubiquitous throughout these structures but as reported previously (Mangos et al., 2010) expression is restricted to the base of the fin by 3 dpf.

We also only detected expression of one pkd gene in somites. pkd2 is first expressed in the ventral half of each rostral somite at 4 dpf and this expression persists at 5 dpf (Figure 7L, arrows in Figures 10P–R).

We detected transient expression of five different pkd genes in the retina. pkd1b, pkd1l1, pkd1l2a, pkd1l2b, and pkd2l1 are expressed in the ganglion cell layer (adjacent to the lens, single cross) and the amacrine cells (outer cell layer immediately adjacent to the ganglion cell layer, double cross) of the eye at 4 dpf. The expression of pkd1b and pkd1l2b is weak, but the expression of pkd1l1, pkd1l2a, and pkd2l1 is stronger (Figures 10S–W). However, only pkd1l2b expression persists at 5 dpf (data not shown).

In other vertebrates, PKD1L3 and PKD2L1 have been proposed to function as sour taste receptors (Huang et al., 2006; Ishimaru et al., 2006, although also see Nelson et al., 2010; Hofherr and Kottgen and references therein). In zebrafish, the taste buds first form at about 3 dpf on the lips/jaw and pharyngeal arches and then at 4 dpf they also form on the mouth and oro-pharyngeal cavity (Hansen et al., 2002; Kapsimali et al., 2011). Consistent with possible expression in the taste buds, we observe dynamic expression of all seven zebrafish pkd genes in the pharyngeal and/or jaw regions at these stages.

Mangos et al. (2010) reported expression of pkd1 in the pharyngeal arches and jaw forming regions at both 52 and 72 h and our data agrees with this. We see strong expression in the pharyngeal arches by 48 h (data not shown) and this expression remains strong at 3 dpf on both the ceratobranchial arches (the more caudal parts of the pharyngeal arches, posterior to the eyes, arrowheads Figures 12D, 13A), and also more anteriorly on the hyosymplectic cartilage (medio-lateral and just posterior to the eyes, derived from the second pharyngeal arch, arrowheads Figures 12D, 13A). Expression is decreased at 4 and 5 dpf (arrowheads Figures 12A–F, 13A–C). Consistent with an earlier report (Bisgrove et al., 2005), we also find pkd2 expression in the pharyngeal arches at 3 dpf (arrowheads Figures 12G,J). In addition, we demonstrate here that strong expression of pkd2 persists in the pharyngeal arches at 4 dpf (arrowheads Figures 12H,K). By 5 dpf, there is also some expression on the pharyngeal walls and the expression in the pharyngeal cartilage has reduced caudally and remains in only the rostral-most pharyngeal cartilage (arrowheads Figures 12I,L, 13D–F).

pkd1l2a and pkd1l2b are also expressed in putative taste buds, although their expression patterns differ from one another and from those of pkd1 and pkd2. We observed pkd1l2a expression on the pharyngeal walls and cartilage at all stages examined. In contrast to pkd1, which is expressed not only across the entire surface of the ceratobranchial arches but also on the medio-lateral hyosymplectic cartilage at 3 dpf, pkd1l2a is expressed only laterally on the ceratobranchial arches and in very few cells medially, in the pharyngeal cavity between the eyes. By 4 dpf, pkd1l2a expression is strongest on the ceratohyal cartilage, which is the most rostral and medial component of the pharyngeal cartilage and this expression persists at 5 dpf (arrowheads Figures 12M–R, 13G–I). In contrast to pkd1l2a, we did not detect expression of pkd1l2b on the pharyngeal walls. Instead it is expressed in a few medial cells of the ceratobranchial arches (caudal-most pharyngeal arches) at 3 dpf. This expression spreads laterally by 4 dpf, and is expanded further by 5 dpf such that it is adjacent and caudal to that of pkd1l2a at these stages (arrowheads Figures 12S–X, 13J–L). At 3 dpf the expression of pkd2l1 partially overlaps that of pkd1l2a and pkd1l2b across the ceratobranchial arches, although, like pkd1l2b, pkd2l1 is not expressed on the pharyngeal walls, it is only expressed on the pharyngeal cartilage. This expression of pkd2l1 increases and expands laterally across the ceratobranchial arches over the next 2 days (arrowheads Figures 12Y,Z,A'–D', 13M–O).

We also see weak expression of pkd1b in the pharynx, probably on the pharyngeal walls at 2, 3, 4, and 5 dpf. By 4 dpf, and persisting at 5 dpf, this gene is also expressed in a few locations on the pharyngeal cartilage (arrowheads Figures 12E'–J', 13P–R). pkd1l1 is only expressed very transiently in these regions. There is no obvious expression on either the pharyngeal cartilage or the pharyngeal walls at either 3 or 5 dpf. However, there is a transient pulse of expression on both the pharyngeal walls and parts of the pharyngeal cartilage at 4 dpf (arrowheads Figures 12K'–P', 13S–U).