Hmx3a has essential functions in zebrafish spinal cord, ear and lateral line development

Transcription factors that contain a homeodomain DNA-binding domain have crucial functions in most aspects of cellular function and embryonic development in both animals and plants. Hmx proteins are a sub-family of NK homeodomain-containing proteins that have fundamental roles in development of sensory structures such as the eye and the ear. However, Hmx functions in spinal cord development have not been analyzed. Here we show that zebrafish (Danio rerio) hmx2 and hmx3a are co-expressed in spinal dI2 and V1 interneurons, whereas hmx3b, hmx1 and hmx4 are not expressed in spinal cord. Using mutational analyses, we demonstrate that, in addition to its previously reported role in ear development, hmx3a is required for correct specification of a subset of spinal interneuron neurotransmitter phenotypes, as well as correct lateral line progression and survival to adulthood. Surprisingly, despite similar expression patterns of hmx2 and hmx3a during embryonic development, zebrafish hmx2 mutants are viable and have no obviously abnormal phenotypes in sensory structures or neurons that require hmx3a. In addition, embryos homozygous for deletions of both hmx2 and hmx3a have identical phenotypes to severe hmx3a single mutants. However, mutating hmx2 in hypomorphic hmx3a mutants that usually develop normally, results in abnormal ear and lateral line phenotypes. This suggests that while hmx2 cannot compensate for loss of hmx3a, it does function in these developmental processes, although to a much lesser extent than hmx3a. More surprisingly, our mutational analyses suggest that Hmx3a may not require its homeodomain DNA-binding domain for its roles in viability or embryonic development.


Introduction
Homeobox-containing genes and the Homeodomain-containing transcription factors that they encode, have crucial functions in most aspects of cellular function and embryonic development in both animals and plants (Burglin and Affolter 2016). They were also some of the first examples discovered of invertebrate developmental genes that are highly conserved in vertebrates (Carrasco et al. 1984;Gehring 1985). One important subclass of homeodomain proteins are NK proteins. NK genes are evolutionarily ancient and are part of the ANTP megacluster, which also includes Hox and ParaHox genes. NK proteins have fundamental roles in the development of mesoderm, endoderm, the nervous system and the heart in all bilaterian animals examined so far (Wotton et al. 2010;Holland 2013;Treffkorn et al. 2018) and they are also found in sponges, one of the most basal animals still alive, and potentially the sister group to all other animals (Larroux et al. 2007;Fortunato et al. 2014;Pisani et al. 2015;Simion et al. 2017).
Hmx proteins (H6 Family Homeodomain proteins, previously called Nk5 or Nkx5 proteins, see Table   S1) are a key sub-family of NK proteins. In vertebrates there are usually three or four different Hmx genes as Hmx4 is only found in some species (Wotton et al. 2010). Interestingly, Hmx2 and Hmx3 are usually located adjacent to each other on the same chromosome and this is also the case for Hmx1 and Hmx4, suggesting that both pairs of genes arose from tandem duplication events rather than the two rounds of whole genome duplication that occurred at the base of the vertebrates (Wotton et al. 2010). In teleosts, there are occasionally extra duplicates of one or more of these genes as the result of the additional genome duplication in this lineage, although interestingly, the retained genes are not consistent between different teleost species (Wotton et al. 2010). In zebrafish there are two hmx3 genes, hmx3a and hmx3b, but only one hmx1, hmx2 and hmx4 gene.
Previous research has shown that Hmx2 and Hmx3 have crucial functions in ear development in mouse and our recent work shows that this is also the case for Hmx3a in zebrafish (Wang et al. 1998;Wang et al. 2001;Wang et al. 2004;Wang and Lufkin 2005;Hartwell et al. 2019). In mouse, both Hmx2 and Hmx3 mutants have ear defects and these are more severe in double mutants (Wang et al. 2001;Wang et al. 2004). Hmx2 and Hmx3 are also required for correct specification of the mouse hypothalamus (Wang et al. 2004) and morpholino knock-down experiments have suggested that they are required for correct lateral line development in zebrafish (Feng and Xu 2010). Hmx2 and Hmx3 are also expressed in two distinct domains in mouse spinal cord but the spinal cord functions of these genes are unknown (Bober et al. 1994;Wang et al. 2000;Wang et al. 2004).
Here we show that zebrafish hmx2 and hmx3a are co-expressed in spinal dI2 and V1 interneurons, whereas hmx3b, hmx1 and hmx4 are not expressed in spinal cord. Using knock-down and mutational analyses, we demonstrate that, in addition to its role in ear development, hmx3a is required for correct specification of a subset of spinal cord interneuron neurotransmitter phenotypes as well as lateral line progression and viability (survival to adulthood). Our data suggest that in the absence of functional Hmx3a protein, a subset of dI2 spinal interneurons switch their neurotransmitter phenotype from glutamatergic (excitatory) to GABAergic (inhibitory). This is important because currently very little is known about how dI2 spinal interneuron neurotransmitter phenotypes are specified, or indeed how spinal cord excitatory neurotransmitter phenotypes in general are specified, and if neurons do not acquire the correct neurotransmitter phenotypes, they cannot function appropriately in spinal cord circuitry.
Surprisingly, despite the fact that hmx2 and hmx3a have similar expression patterns during embryonic development and both genes are required for correct ear development in mouse, our mutational analyses did not uncover any requirement for hmx2, by itself, in viability, ear development, lateral line progression or specification of spinal cord interneuron neurotransmitter phenotypes in zebrafish. This is surprising, especially given that embryos injected with a hmx2 morpholino have reduced numbers of spinal cord glutamatergic neurons and a corresponding increase in the number of spinal cord GABAergic neurons and that embryos injected with both hmx2 and hmx3a morpholinos have more severe spinal cord phenotypes than single knock-down embryos. Zebrafish hmx2 mutants are viable and have no obviously abnormal phenotypes in these sensory structures and neurons that require hmx3a, even when almost all of the hmx2 locus is deleted. (In our most severe mutant allele, hmx2 SU39 , only 84 nucleotides of 5' and 60 nucleotides of 3' coding sequence remain). In addition, zebrafish embryos homozygous for deletions of both hmx2 and hmx3a have identical phenotypes to severe hmx3a single mutants. However, mutating hmx2 in hypomorphic hmx3a mutants, that usually develop normally, results in abnormal ear and lateral line progression phenotypes, suggesting that while hmx2 cannot compensate for mutations in hmx3a, it does function in these developmental processes, although to a much lesser extent than hmx3a. Our analyses of homozygous mutant phenotypes for several different hmx3a mutant alleles also suggest that Hmx3a may not require its homeodomain for its roles in viability or embryonic development. This is surprising, as homeodomain proteins usually function by binding DNA through their homeodomain and regulating gene expression. In contrast, our mutational analyses suggest that Hmx3a may only require its N terminal-domain for its vital functions in viability and sensory organ and spinal cord interneuron development. 6

Ethics statement
All zebrafish experiments in this research were carried out in accordance with the recommendations and approval of either the UK Home Office or the Syracuse University IACUC committee.

CRISPR mutagenesis and screening
The hmx3a SU3 allele was described previously (Hartwell et al. 2019). With the exception of hmx3a sa23054 (generated in the Zebrafish Mutation Project and obtained from ZIRC), we created all of the other hmx2, hmx3a and hmx2/3a double deletion mutants described in this paper using CRISPR mutagenesis. For all alleles, other than the hmx2 MENTHU allele, we designed and synthesized single guide RNA (sgRNA) and Cas9 mRNA as in (Hartwell et al. 2019). For the hmx2 MENTHU allele, we designed the crRNA using the Microhomology-mediated End joining kNockout Target Heuristic Utility (MENTHU) tool (version 2.1.2), in the Gene Sculpt Suite (Ata et al. 2018;Mann et al. 2019). The MENTHU allele crRNA design was verified with CHOPCHOP (version 3.0.0) (Montague et al. 2014;Labun et al. 2016;Labun et al. 2019) and the CRISPR-Cas9 gRNA design checker tool (Integrated DNA Technologies). The hmx2 MENTHU crRNA was purchased together with a universal 67mer tracrRNA (1072533) and Alt-R S.p.
We identified founder fish for hmx2 SU35 , hmx2 SU36 , hmx2 SU37 and hmx3a SU42 alleles using high resolution melt analysis (HRMA), and the supermix and amplification programs described in (Hartwell et al. 2019). For the PCRs described below, we used Phusion High-Fidelity DNA Polymerase (M0530L, NEB) unless otherwise stated. HRMA primers and PCR primers for sequencing are provided in Table   S2.
We used the following PCR conditions to identify hmx2 SU38  We then diluted the nested 1 PCR product 1:5 in sterile distilled water and performed the Nested PCR 2 reaction using the following conditions: 98.0 o C for 30 seconds, 35 cycles of: 98.0 o C for 10 seconds, 66.0 o C for 20 seconds, 72.0 o C for 60 seconds, followed by a final extension at 72.0 o C for 5 minutes. The mutant allele was distinguished by gel electrophoresis on a 1% TAE agarose gel (110V for 30 minutes).
The WT allele generated a 1758 bp product compared with a 322 bp mutant allele product.
We identified hmx2 MENTHU F0 embryos by PCR, followed by sequencing with the FW primer that generated the amplicon (Table S2). The PCR was performed on DNA extracted from individual embryos using the following conditions: 98.0 o C for 30 seconds, 35 cycles of: 98.0 o C for 10 seconds, 64.0 o C for 20 seconds, 72.0 o C for 15 seconds, followed by a final extension at 72.0 o C for 5 minutes. We assayed that the PCR was successful by gel electrophoresis on a 2.5% TBE agarose gel (100V for 40 minutes).
The PCR generates a 155 bp product. The PCR product was purified using EZ-10 Spin Column PCR Products Purification Kit (BS664, Bio Basic) and eluted in 30 µl sterile water prior to sequencing.
We used either assessment of germline transmission, as described above, or PCR to identify hmx3a SU43 founder fish. PCR conditions were: 98.0 o C for 30 seconds, 35 cycles of: 98.0 o C for 10 seconds, 69.0 o C for 20 seconds, 72.0 o C for 60 seconds, followed by a final extension at 72.0 o C for 5 minutes. The mutant allele was distinguished by gel electrophoresis on a 1% TAE agarose gel (110V for 30 minutes). A 1471 bp PCR product was only generated by fish heterozygous for the allele. It was not produced from WT animals since the reverse primer only recognizes the inserted donor DNA sequence.
We identified hmx2;hmx3a SU44 and hmx2;hmx3a SU45 founder fish by nested PCR, using the following conditions: Nested PCR 1: 98.0 o C for 30 seconds, 35 cycles of: 98.0 o C for 10 seconds, 67.0 o C for 20 seconds, 72.0 o C for 30 seconds, followed by a final extension at 72.0 o C for 5 minutes. The mutant allele was distinguished by gel electrophoresis on a 1% TAE agarose gel (110V for 30 minutes). The WT product was too large to be generated by these PCR conditions, so only heterozygous animals are detected by the presence of a 514 bp product on the gel. We then diluted the nested 1 PCR 1:5 in sterile distilled water and performed the Nested PCR 2 reaction using the following conditions: 98.0 o C for 30 seconds, 35 cycles of: 98.0 o C for 10 seconds, 66.0 o C for 20 seconds, 72.0 o C for 30 seconds, followed by a final extension at 72.0 o C for 5 minutes. The mutant allele was distinguished by gel electrophoresis on a 1% TAE agarose gel (110V for 30 minutes). Again, the WT product was too large to be generated by these PCR conditions, so only heterozygous animals were detected by the presence of a 445 bp product.
Once stable lines were established, we identified hmx2 SU35 fish by PCR, followed by sequencing (the mutation introduces a 1 bp insertion that cannot be resolved by restriction digestion, and we cannot distinguish heterozygotes from homozygotes using HRMA). We performed this PCR using Taq DNA Polymerase (M0320S, NEB) and the following conditions: 95.0 o C for 30 seconds, 35 cycles of: 95.0 o C for 20 seconds, 52.0 o C for 30 seconds, 68.0 o C for 45 seconds, followed by a final extension at 68.0 o C for 5 minutes. We used HRMA and the conditions described above to identify hmx2 SU36 stable mutants.
Homozygous mutants segregate from heterozygous animals by the scale of their deflection in the HRMA plot. We identified hmx2 SU37 mutants by performing the PCR used to sequence hmx2 SU35 stable mutants (see above). When we analyzed the products on a 1% TAE gel (110V for 30 minutes), the WT allele generated a 580 bp product, compared with a 528 bp mutant product. We identified hmx2 SU38 stable mutants using the same PCR conditions initially used to identify founders (see above). We used the same nested PCR conditions to identify hmx2 SU39 mutants. However, the WT product was not always visible on the gel. Therefore, we also performed a WT amplicon PCR identical to that described above for identifying stable hmx2 SU35 fish, as this genomic region is only present in WT and heterozygous animals.
We identified stable hmx3a SU42 mutants by PCR, using Taq DNA Polymerase (M0320S, NEB) and the following conditions: 94.0 o C for 2 minutes, 35 cycles of: 94.0 o C for 30 seconds, 64.9 o C for 30 seconds, 72.0 o C for 30 seconds, followed by a final extension at 72.0 o C for 2 minutes. Whilst the mutant PCR product (321 bp) could sometimes be distinguished from the WT product (331 bp) by running on a 2% TBE gel (100V for 55 minutes), the mutation also deletes a BanI restriction site. Following digestion with BanI (R0118S, NEB), the products were run on a 2% TBE gel (100V for 40 minutes). The WT amplicon digested to completion, producing 120 bp + 211 bp bands, whereas the mutant product did not cut. We identified stable hmx3a SU3 mutants by running the same PCR used to identify hmx3a SU42 mutants. The insertion in hmx3a SU3 was easily visualized on a 2% TBE gel. The WT product was 331 bp, compared to a mutant product of 400 bp. Since the PCR used to detect hmx3a SU43 mutants was specific to the inserted donor DNA, and the WT amplicon in hmx2;hmx3a SU44 and hmx2;hmx3a SU45 mutants was too large to detect by standard PCR, for these alleles we also performed a WT amplicon PCR to distinguish WTs from hets. The WT amplicon PCR was identical to that performed prior to BanI digestion on hmx3a SU42 mutants (see above). For hmx3a SU43 , the WT amplicon PCR results were compared to the PCR results (identical PCR to that first used to identify founders, see above), and for hmx2;hmx3a SU44 and hmx2;hmx3a SU45 , the WT amplicon PCR results were compared to the Nested 2 PCR results (identical Nested 2 PCR to that first used to identify founders -see above).
In all cases, stable F1 heterozygous fish were confirmed by sequencing. See Fig. 4 for details of how individual mutant alleles differ from one another. To further confirm the mutant sequences of hmx2 SU39 and hmx3a SU42 , we extracted total RNA from embryos produced by incrosses of homozygous viable adults using TRIzol Reagent (15596018, ThermoFisher Scientific) and the RNEasy Mini Kit (74104, Qiagen). Total RNA was converted to cDNA using the iScript cDNA synthesis kit (1708891, Bio-Rad).
We performed transcript-specific PCRs using the following primers and conditions: hmx2-FW: followed by final extension at 72.0 o C for 5 minutes). We then confirmed these homozygous mutant transcript sequences by sequencing.
To circumvent this, they were incubated at 32 o C from 9 hours post fertilization (h) onwards. This ensured that control and injected embryos reached the desired developmental stage of 27 h at approximately the same time. The lateral line primordium does not migrate in DKD animals, so this could not be used to stage injected embryos. Instead, these embryos were visually inspected and fixed when they displayed the same head-trunk angle, head size and eye size as prim-staged uninjected control embryos (Kimmel 1995). Migration of the lateral line primordium is unaffected in SKD embryos, so these were prim-staged prior to fixing for experiments (Kimmel 1995). Morpholino injections always produce a spectrum of phenotypes, since it is hard to ensure that every cell receives the same dose. Therefore, prior to fixing at 27 h, we removed any embryos with severely abnormal morphology (stunted length and/or severely developmentally delayed, likely caused by receiving too much morpholino). Embryos injected with hmx2/3a morpholinos (SKD and DKD) display a slight curled-tail-down morphology. Embryos that lacked this morphology (and may therefore not have received any or sufficient morpholino) were also removed before fixing.
For the mRNA + morpholino rescue experiments, we co-injected either each individual or both translation-blocking hmx morpholinos (at the same volume and dose described above), together with a total dose of up to 500 pg of morpholino resistant (MOR) full-length hmx2 or hmx3a mRNA. Both hmx mRNAs had 7 nucleotides altered in the morpholino recognition sequence. Each change was in the third nucleotide of a codon. This codon wobble was used so that the same amino acid was encoded in each case, but the mRNA would not be recognized by the morpholino. The protein encoded by the injected RNA is therefore the same as either endogenous Hmx2 or Hmx3a.
Heat-inactivation of the DNase was performed for 10 minutes at 65 o C. 20 µl RT-PCRs were performed as per the manufacturer's instructions using the Qiagen One-Step RT-PCR kit (210210, Qiagen)  RT-PCR E1-2 primers generate either a 1204 bp genomic (unspliced) or 426 bp spliced product. The hmx3a RT-PCR E1-2 primers generate either a 779 bp genomic (unspliced) or 393 bp spliced product.
To assess whether genetic compensation occurs in either hmx2 SU39 or hmx3a SU42 mutants, which lack obvious phenotypes, or hmx3a SU3 mutants, which have milder spinal cord phenotypes than hmx2/3a DKD embryos, we injected the same dose of either hmx2 + p53 MOs (hmx2 SU39 ) or hmx3a + p53 MOs (hmx3a SU42 , hmx3a SU3 ) as described above, into the single-cell of one-cell stage embryos generated from incrosses of heterozygous hmx2 SU39 , hmx3a SU42 or hmx3a SU3 parents respectively. If genetic compensation is occurring, the upregulated compensating gene(s) will not be knocked-down by the hmx morpholino and the phenotype of homozygous mutants should be unchanged. In contrast, WT and heterozygous animals, which contain at least one WT copy of the respective hmx gene will be susceptible to the hmx morpholino and should exhibit stronger, morphant-like phenotypes. For these experiments, while we removed any embryos with severely abnormal morphology, we did not remove embryos that lacked the curled-tail-down morphology, incase these were morpholino-resistant mutant embryos. After fixing, we performed an in situ hybridization for the glutamatergic marker, slc17a6a/b. We visually inspected the embryos on a dissecting microscope and categorized them as either the stronger, morphantlike phenotype (large reduction in the number of slc17a6a/b-expressing cells) or a more subtle phenotype (WT-like in the case of hmx2 SU39 and hmx3a SU42 or a smaller reduction in the number of slc17a6a/bexpressing cells in the case of hmx3a SU3 embryos). Embryos within each class were then genotyped as described in the CRISPR mutagenesis section above.

Genotyping
We isolated DNA for genotyping from both anesthetized adult fish and fixed embryos via fin biopsy or head dissections, respectively. For assaying ear phenotypes, we dissected tail tips instead. We genotyped the hmx CRISPR mutants as described above. Scientific) and 2 volumes of ice-cold RNase-free ethanol. Samples were precipitated at -20 o C overnight.
Genomic DNA was recovered by centrifugation at +4 o C, followed by washing with 70% RNase-free ethanol and further centrifugation at +4 o C. After carefully removing the ethanolic supernatant, the pellets were air dried for 5-10 minutes at room temperature before resuspending in 15 µl of sterile distilled water.
in situ hybridization and immunohistochemistry We fixed embryos in 4% paraformaldehyde / phosphate-buffered saline (PBS) and performed single and double in situ hybridizations and immunohistochemistry plus in situ hybridization double labeling experiments as previously described (Concordet et al. 1996;. Sources of in situ hybridization probes are provided in Table S1. To amplify in situ hybridization probe templates for hmx1 and hmx3b, we created zebrafish cDNA from 27 h WT zebrafish embryos. We extracted total RNA by homogenizing 50-100 mg of embryos in 1 mL of TRIzol reagent (Ambion, 15596-026). We confirmed RNA integrity (2:1 ratio of 28S:18S rRNA bands) and quality (A260/A280 ratio of ~2.0) using agarose gel electrophoresis and spectrophotometry respectively. We synthesized cDNA using Bio-Rad iScript  (Ku et al. 2004;Lauter et al. 2011).
In cases where we did not detect expression of a particular gene in the spinal cord, we checked for low levels of expression by exposing embryos to prolonged staining. In some cases, this produced higher background (diffuse, non-specific staining), especially in the hindbrain, where ventricles can sometimes trap anti-sense riboprobes.
To determine neurotransmitter phenotypes we used probes for genes that encode proteins that transport or synthesize specific neurotransmitters as these are some of the most specific molecular markers of these cell fates (Higashijima et al. 2004b;Higashijima et al. 2004c and references therein). A mixture of probes to slc17a6a and slc17a6b (previously called vglut), which encode glutamate transporters, was used to label glutamatergic neurons (Higashijima et al. 2004b;Higashijima et al. 2004c). GABAergic neurons were labeled using probes to gad1b (probes previously called gad67a and gad67b) (Higashijima et al. 2004b;Higashijima et al. 2004c). The gad1b gene encodes for a glutamic acid decarboxylase, which is necessary for the synthesis of GABA from glutamate. A mixture of probes (glyt2a and glyt2b) for slc6a5 (previously called glyt2) was used to label glycinergic cells (Higashijima et al. 2004b;Higashijima et al. 2004c). slc6a5 encodes for a glycine transporter necessary for glycine reuptake and transport across the plasma membrane.
The antibodies that we used for fluorescent in situ hybridization were mouse anti-Dig (200-002-156, Jackson ImmunoResearch, 1:5000) and rabbit anti-Flu (A889, Invitrogen, 1:2500). These were detected using secondary antibodies: goat anti-rabbit-HRP (G-21234, ThermoFisher Scientific, 1:750) and goat anti-mouse-HRP (G-21040, ThermoFisher Scientific, 1:750) and Tyramide SuperBoost Kits B40922 and B40915 (ThermoFisher Scientific). Phalloidin staining 4 d old embryos generated from incrosses of heterozygous hmx2 SU39 or hmx2;hmx3a SU44 parents were fixed and processed for phalloidin staining as described in (Hartwell et al. 2019 The hmx2 and hmx3a primers were generated in this study. The mob4 primers were generated by Hu and colleagues (Hu et al. 2016). They demonstrated that mob4 is a more effective reference gene than actb2 across a broad range of zebrafish developmental stages, including early stages where only maternal RNAs should be present (Hu et al. 2016). To generate amplification data the program used was: 95.

Screening lateral line and otolith phenotypes
We examined whether any of the hmx mutants generated in this study had lateral line and/or fused otolith phenotypes, as reported for hmx2;hmx3a double-knockdown embryos (Feng and Xu 2010 Axio Imager M1 compound microscope, we located the tip of the lateral line primordium and counted the somite number adjacent to this position. We also used this method routinely to determine the developmental stage of embryos prior to fixing for in situ hybridization. To assay lateral line phenotypes in fixed embryos, we performed in situ hybridizations for hmx3a or krt15 (both of which label the migrating primordium and neuromasts) and then determined the lateral line position as in live embryos.
To examine live otolith phenotypes, embryos were raised until 3 dpf, before anaesthetizing (as for assessing live lateral line phenotypes) and examining the spatial location of otoliths in both ears. WT embryos have two otoliths in each ear: one smaller, anterior (utricular) otolith, and one larger, posterior (saccular) otolith. These are separate from each other and spatially distinct. We classified otoliths as fused if only one large, amalgamated otolith was visible in a mid-ventral position within the otic vesicle (see Fig. 5U).

Imaging
Embryos were mounted in 70% glycerol:30% distilled water and Differential Interference Contrast (DIC) pictures were taken using an AxioCam MRc5 camera mounted on a Zeiss Axio Imager M1 compound microscope. Fluorescent images were taken on a Zeiss LSM 710 confocal microscope.
Images were processed using Adobe Photoshop software (Adobe, Inc) and Image J software (Abramoff

Cell counts and statistics
In all cases except where noted to the contrary, cell counts are for both sides of a five-somite length of spinal cord adjacent to somites 6-10. Embryos were mounted laterally with the somite boundaries on each side of the embryo exactly aligned and the apex of the somite over the middle of the notochord.
This ensures that the spinal cord is straight along its dorsal-ventral axis and that cells in the same dorsal/ventral position on opposite sides of the spinal cord will be directly above and below each other.
Embryos from mutant crosses were counted blind to genotype. Labeled cells in embryos analyzed by DIC were counted while examining embryos on a Zeiss Axio Imager M1 compound microscope. We In some cases, cell count data were pooled from different experiments. Prior to pooling, all pairwise combinations of data sets were tested to determine if there were any statistically significant differences between them as described below. Data were only pooled if none of the pairwise comparisons were statistically significantly different from each other. In addition, as in situ hybridization staining can vary slightly between experiments, we only compared different mutant results when the counts from their corresponding WT sibling embryos were not statistically significantly different from each other.
To determine whether differences in values are statistically significant, data were first analyzed for normality using the Shapiro-Wilk test. Data sets with non-normal distributions were subsequently analyzed using the Wilcoxon-Mann-Whitney test (also called the Mann Whitney U test). For data sets with normal distributions, the F-test for equal variances was performed, prior to conducting either a type 2 (for equal variances) or type 3 (for non-equal variances) student's t-test. P values generated by Wilcoxon-Mann-Whitney, type 2 student's t-test and type 3 student's t-test are indicated by ^, + and § respectively. To control for type I errors, when comparing three or more experimental conditions, a oneway analysis of variance (ANOVA) test was performed. Prior to conducting ANOVA tests, data were first analysed for normality using the Shapiro-Wilk test, as described above. All data sets for ANOVA analysis had normal distributions and so were subsequently assessed for homogeneity of variances using Bartlett's test. All of the data sets also had homogeneous (

Microarray expression profiling experiments
These experiments are described in detail in . P values were corrected for multiple testing (Benjamini and Hochberg 1995;Gentleman et al. 2004;Tarraga et al. 2008). These data have been deposited in the NCBI Gene Expression Omnibus with accession number GSE145916.

Data and Reagent Availability
Plasmids and zebrafish strains are available upon request. Two supplemental figures and two supplemental tables are available at FigShare. One figure contains RT-PCR and cell-count data demonstrating the efficacy of hmx2;hmx3a DKD with splice-blocking morpholinos and the other an alignment of mouse and zebrafish Hmx2 and Hmx3a protein sequences. Table S1 includes gene names, ZFIN identifiers and references for in situ hybridization probes and Table S2 contains the sgRNA and primer sequences used for hmx2, hmx3a and hmx2;hmx3a CRISPR mutagenesis and genotyping.
Microarray data have been previously deposited in the NCBI Gene Expression Omnibus with accession number GSE145916.

Results
hmx2 and hmx3a are the only hmx genes expressed in spinal cord.
While the expression and functions of zebrafish hmx genes have been analyzed during the development of sensory structures such as the eye and the ear, the expression of hmx1, hmx2, hmx3a and hmx4 in the developing spinal cord has not been investigated and no expression data has previously been reported for hmx3b, which only appeared in more recent versions of the zebrafish genome sequence (Zv9 and above). Therefore, to determine which of the hmx genes are expressed in the spinal cord we performed in situ hybridizations for hmx1, hmx2, hmx3a, hmx3b and hmx4 at different developmental stages ( Fig.   1). At all of these stages, we observed no spinal cord expression of hmx1, hmx3b or hmx4 ( Fig. 1).
However, consistent with previous reports, both hmx1 and hmx4 were expressed in the developing eye, As hmx3a is expressed in the spinal cord, and teleost duplicate genes often have similar expression patterns, we wanted to further test whether there was any spinal cord expression of hmx3b. Therefore, we performed in situ hybridization on mindbomb1 (mib1 ta52b ) mutants at 24 h. mib1 encodes an E3ubiquitin protein ligase required for efficient Notch signaling. Consequently, Notch signaling is lost in mib1 ta52b mutants and this causes most spinal progenitor cells to precociously differentiate into earlyforming classes of spinal neurons at the expense of later-forming classes of neurons and glia (Jiang et al. 1996;Schier et al. 1996;Itoh et al. 2003;Park and Appel 2003;. As a result, weak expression in spinal neurons is often expanded and stronger and hence easier to observe in mib1 ta52b mutants at 24 h England et al. 2017). However, even in mib1 ta52b mutants we detected no expression of hmx3b (Fig. 1I). We also analyzed the expression of the other hmx genes in mib1  hmx2 and hmx3a are expressed in V1 and dI2 interneurons in the spinal cord.
To identify the spinal cord neurons that express hmx2 and hmx3a, we performed several different doublelabeling experiments. Double in situ hybridization with hmx2 and hmx3a confirmed that these genes are co-expressed in the exact same cells in the spinal cord ( Fig. 2A). Approximately half of these hmx2 and hmx3a (hmx2/3a) co-expressing spinal cells also co-express slc32a1, which is only expressed by inhibitory (glycinergic and GABAergic) interneurons (Jellali et al. 2002), and approximately half co- determine neurotransmitter phenotypes and additional references). In addition, the inhibitory hmx2/3aexpressing cells are generally more ventral than the excitatory double-labelled cells. Our previous expression-profiling of V1 interneurons, suggested that these cells might be the ventral inhibitory neurons that express hmx3a (Fig. 2G, for a description of these experiments see )).
Results from our lab and others, have established that V1 interneurons are the only spinal cord cells that express engrailed1b (en1b) (Higashijima et al. 2004a;. Therefore, to confirm that V1 interneurons also express hmx3a we performed double in situ hybridizations for hmx3a and en1b. These experiments showed that all of the en1b-expressing spinal cells co-express hmx3a, and that approximately half of the hmx2/3a-expressing spinal cells co-express en1b (Fig. 2C). Taken together, these data clearly identify the inhibitory hmx2/3a-expressing spinal cells as V1 interneurons.
As mentioned above, the glutamatergic hmx2/3a-expressing cells are generally located more dorsal to the inhibitory hmx2/3a-expressing cells. Therefore, these excitatory cells could be V0v, dI5, dI3, dI2 or dI1 interneurons ( Fig. 2a; Cheng et al. 2005;Grossmann et al. 2010;Satou et al. 2012;Talpalar et al. 2013). The zebrafish embryonic spinal cord is relatively small. For example, at 27h, the dorsal-ventral axis is only about 10 cells high. As a result, the different neuronal populations are often intermingled, rather than clearly separated as they are in amniotes (e.g. see England et al. 2011).
In addition, studies in amniotes suggest that many dorsal neurons migrate dorsally or ventrally soon after they are born (e.g. see Gross et al. 2002;Müller et al. 2002). Taken together, this means that it is hard to accurately identify dorsal spinal cord cell types by position alone. Therefore, to identify the excitatory Knock-down experiments suggest that hmx2 and hmx3a may be redundantly required for correct specification of a subset of spinal interneuron glutamatergic phenotypes.
As an initial step to try and identify the function(s) of hmx2 and hmx3a in spinal cord development we performed morpholino knock-down experiments. As previous analyses suggested that these genes have redundant roles in ear and lateral line development (Feng and Xu 2010), we designed and injected translation blocking morpholinos for both of these genes (see materials and methods). In embryos coinjected with the two morpholinos (hmx2;hmx3a double knock-down (DKD) animals), we observed stalled lateral line progression and fused otoliths in the ears (Fig. 3A-D). Normally there is an anterior (utricular) and a posterior (saccular) otolith in each ear, but in DKD embryos there was just one fused otolith in a medio-ventral region of each ear (Fig 3D). When we analyzed spinal cord phenotypes, we detected no change in the number of hmx3a-or en1b-expressing cells in DKD embryos, suggesting that dI2 and V1 interneurons still form in normal numbers ( Fig. 3E-J, Table 1A). However, when we examined markers of neurotransmitter phenotypes, we observed a reduction in the number of spinal excitatory (glutamatergic, slc17a6-expressing) cells and a corresponding increase in inhibitory (slc32a1-expressing) cells ( Fig. 3K-P, Table 1A). As hmx2 and hmx3a are only expressed by dI2 neurons (which are glutamatergic) and V1 neurons (which are inhibitory) in the spinal cord, this suggested that at least some dI2 interneurons had switched their neurotransmitter phenotype from glutamatergic to inhibitory.
Consistent with the idea that the two genes act redundantly, both of these spinal cord phenotypes were less severe in SKD embryos ( Fig. 3M, P, Table 1A). Interestingly, we also did not see abnormal lateral line progression or ear phenotypes in SKD embryos.
To confirm the specificity of these morpholino knock-down results, we first tested whether we saw a similar spinal cord phenotype if we injected splice-blocking morpholinos against hmx2 and hmx3a (see materials and methods). In these experiments we obtained a partial reduction in correct splicing of these genes ( Table 1A), consistent with the fact that we only obtained a partial knock-down of each gene using the splice-blocking morpholinos. We then tested whether co-injecting a morpholino-resistant hmx2 or hmx3a mRNA with the morpholinos could rescue the reduction in the number of spinal glutamatergic cells that occurs in translation blocking SKD and DKD embryos (see materials and methods for the design of the mRNAs). We found that both hmx3a and hmx2 morpholino-resistant mRNA could completely rescue this phenotype in hmx2 SKD embryos (Fig. 3T-V and Table 1A) and hmx3a could completely rescue and hmx2 could partially rescue the phenotype in hmx3 SKD embryos. In addition, either hmx2 or hmx3a morpholino-resistant mRNA was able to partially rescue the number of glutamatergic spinal neurons in DKD embryos (Fig. 3Q-S and Table 1A). Injections of higher amounts of mRNA or of both mRNAs at the same time led to embryo death, probably because of the toxic effects of injecting considerable amounts of both mRNA and morpholinos into the embryos during early development.
Mutational analyses suggest that hmx2 is not, by itself, required, and that Hmx3a protein may not require its DNA-binding homeodomain, for viability, correct migration of lateral line primordium, or correct development of ear otoliths or a subset of spinal cord interneurons.
To further and more robustly test the hypothesis that hmx2 and hmx3a are required for the correct specification of a subset of spinal cord interneuron neurotransmitter phenotypes we created CRISPR mutants in each of these genes, targeting a region upstream of the homeobox (see materials and methods; Fig. 4). We also obtained a hmx3a sa23054 allele from the Sanger zebrafish mutation project (Kettleborough et al. 2013) that introduces a stop codon upstream of the homeobox.
Our analyses of homozygous mutant embryos demonstrate that hmx3a SU3 and hmx3a SU43 mutants have fused otoliths, stalled lateral line progression and are homozygous lethal (Fig. 4, Fig. 5O, U, Table 2 and data not shown, also see (Hartwell et al. 2019) for a detailed description of the hmx3a SU3 ear phenotype).
When we examined the spinal cords of these mutants, we observed a statistically significant reduction in the number of glutamatergic cells, but in both cases the reduction was smaller than we had previously observed for morpholino-injected DKD embryos ( Fig. 5D & I, Table 1B). There was also an increase in the number of inhibitory spinal cord interneurons, although, again, the increase was less than in the DKD morpholino-injected embryos (Fig. 6C, H, M, Table 1B). However, similar to the DKD embryos, there was no change in the number of spinal hmx3a-or en1b-expressing cells in hmx3a SU3 mutants, suggesting that dI2 and V1 interneurons are forming in normal numbers and not dying or changing into different classes of interneurons (Fig. 6A, B, F, G, K, L, Table 1B).
In contrast, embryos homozygous for hmx3a SU42 do not have fused otoliths or stalled lateral line progression (Fig. 4, Fig. 5M, Table 2) and unlike hmx3a SU3 mutants, they have normal expression of  Table   1B) and are homozygous viable (Fig. 4, Table 2). Interestingly, embryos homozygous for hmx3a sa23054 have variable, incompletely penetrant otolith fusion phenotypes that range from no fusion, through incomplete fusion (Fig. 5T), to complete fusion, despite the fact that any protein encoded by this allele should retain more WT sequence than that encoded by hmx3a SU42 (Fig. 4). However, hmx3a sa23054 mutants have normal lateral line progression, no reduction in the number of spinal cord glutamatergic neurons and they are also viable (Fig. 4, Fig. 5C, H & N, Tables 1B & 2).
Surprisingly, we also found that all four of the different hmx2 alleles we created in these experiments To test whether our hmx2 mutants had no obvious phenotypes because they had retained some Hmx2 function, we created a large deletion allele, hmx2 SU39 , that deletes most of the hmx2 genomic sequence.
Only 84 nucleotides of 5' and 60 nucleotides of the most 3' coding sequence remain (Fig. 4). However, hmx2 SU39 homozygous mutants also lack fused otoliths and are homozygous viable. They also have normal expression of hmx3a and pax5 in the anterior otic epithelium, normal expression of hmx3a in the adjacent anterior neuroblasts, the normal complement of three distinct cristae and two distinct maculae in the ear, and normal lateral line progression (Fig. 4, Fig. 5Q Table 1B).
There are also no obvious differences in the otolith fusion or lateral line progression phenotypes between these single mutants and embryos homozygous for the double deletion alleles ( To determine whether the increase in the number of spinal inhibitory interneurons reflects an increase in glycinergic or GABAergic neurons, we examined the expression of genes expressed exclusively by cells with these inhibitory neurotransmitter phenotypes (slc6a5 for glycinergic and gad1b for GABAergic, see materials and methods). While we found no statistically significant difference in the number of spinal cord glycinergic cells in hmx3a SU3 mutants or hmx2;hmx3a SU44 deletion mutants, there was a statistically significant increase in the number of GABAergic cells in hmx3a SU3 and hmx2;hmx3a SU44 mutants compared to WT sibling embryos (Fig. 6E, J, O, P, Q, V, W, AB, AC, Table 1B).
As hmx2 spinal expression is initially weaker than hmx3a expression (Fig.1)  Given that approximately a quarter of these embryos should lack almost all of the coding sequence for both alleles of hmx2 (hmx2;hmx3a SU44 lacks all but the last 66 bp and hmx2 SU39 lacks all but the first 84 and last 60 bp of hmx2a coding sequence, see Fig. 4) as well as all of the coding sequence for one allele of hmx3a, this demonstrates that one WT allele of hmx3a is sufficient for normal otolith development and lateral line progression.
As expected, as both alleles produce homozygous mutant phenotypes, when we crossed fish heterozygous for hmx3a SU3 with fish heterozygous for hmx3a SU43 we observed mendelian ratios of embryos with fused otoliths and stalled lateral line progression (Table 3). Similarly, we obtained mendelian ratios of embryos with fused otoliths and stalled lateral line progression when we crossed fish heterozygous for hmx3a SU3 with fish heterozygous for hmx2;hmx3a SU44 , or fish heterozygous for hmx3a SU43 with fish heterozygous for hmx2;hmx3a SU44 .
More interestingly, when we crossed fish heterozygous for hmx3a sa23054 with fish heterozygous for either hmx2;hmx3a SU44 or hmx3a SU3 , we also obtained mendelian ratios of embryos with fully penetrant otolith fusion phenotypes and stalled lateral line progression, as opposed to the variable otolith fusion phenotypes and normal lateral line progression that occurs in hmx3a sa23054 homozygous mutants (cf Table  phenotype (Table 3, similar to Fig. 6AA). This suggests that while two hmx3a sa23054 alleles provide sufficient Hmx3a activity for normal lateral line progression and, in some cases, normal otolith development, this is not the case for either the combination of one hmx3a sa23054 and one hmx3a SU3 allele or the combination of one hmx3a sa23054 allele over a hmx3a deletion.
Even more surprisingly, when we crossed fish heterozygous for hmx3a SU42 with fish heterozygous for either hmx2;hmx3a SU44 or hmx3a SU3 we also obtained embryos with fused otoliths and stalled lateral line progression, although most of the embryos had the slightly weaker lateral line phenotype mentioned above ( Fig. 6 AA & AG, Table 3). For the combination of hmx3a SU42 and hmx2;hmx3a SU44 , these phenotypes occurred in mendelian ratios. However, for the combination of hmx3a SU3 and hmx3a SU42 , while we observed a mendelian ratio of embryos with stalled lateral line progression, only 17% of embryos had abnormal otolith phenotypes and in most of these cases the otoliths were either adjacent but not fused in both ears or there was an abnormal otolith phenotype in only one ear (Table 3). When we genotyped a subset of these embryos, we found that all of the embryos with abnormal otolith phenotypes (n=18, 6 embryos each with either fused otoliths in both ears, adjacent otoliths in both ears, or a fused or adjacent otolith phenotype in only one ear) were hmx3a SU3/+ ;hmx3a SU42/+ trans-hets.
Interestingly, when we genotyped embryos with WT otolith phenotypes (two normal otoliths per ear, We injected CRISPR reagents to mutate hmx2 into embryos from an incross of fish that were heterozygous for hmx3a SU42 . We used the Microhomology-mediated End joining kNockout Target Heuristic Utility (MENTHU) tool to identify a sgRNA target site that should predominantly result in the same 5 bp deletion frame-shift allele (Fig. 4, Fig. 7C), being generated through microhomology- When we did this, we found that at ~3.5 dpf, 28.25% (n=807) of hmx2 CRISPRinjected embryos had an abnormal otolith phenotype (21.31% had fused otoliths in both ears and 6.94% had fused or adjacent otoliths in only one ear; Fig.7B, cf to uninjected control, Fig.7A). In comparison, only 0.6% (n=670) of uninjected embryos and 2% (n=347) of embryos injected with a CRISPR crRNA ribonucleoprotein complex that we have used successfully to make mutations in an unrelated gene, had abnormal otolith phenotypes. These control experiments were performed at the same time as the hmx2 CRISPR injections, using embryos obtained from the same heterozygous hmx3a SU42 parent fish.
We examined 40 of the hmx2 CRISPR-injected embryos at ~30 h for lateral line progression phenotypes and then let these embryos develop to ~3.5 dpf, so we could correlate lateral line and otolith phenotypes.
25% of the embryos had strong or medium stalled lateral line progression phenotypes and all of these embryos also developed fused otoliths in both ears (Table 4A). A few additional embryos had a weaker lateral line progression defect (migration of the primordium was only delayed by 2-3 somites compared to stage-matched injected siblings) and two of these also developed fused otoliths in both ears. When we genotyped these embryos for hmx3a SU42 , we found that all of the embryos with fused otoliths were homozygous for hmx3a SU42 (Table 4B).
We also genotyped 72 additional hmx2 CRISPR-injected embryos, just over half of which had otolith defects. The vast majority of the embryos with otolith phenotypes were homozygous for hmx3a SU42 (Fig,7B, Table 4C, one embryo with fused otoliths in both ears and one embryo with an otolith defect in one ear only were heterozygous). In contrast, all except one of the 35 embryos that did not have obvious defects in otolith development were heterozygous for hmx3a SU42 or WT (Table 4C).
Taken together, these results suggest that we obtained a high efficiency of hmx2 mutations in our injected embryos and that CRISPR-mediated knock-down of Hmx2 causes hmx3a SU42 mutants to have defects in otolith development and lateral line progression. To test this, we sequenced the hmx2 allele from 23 of the embryos that we had genotyped for hmx3a SU42 that had different otolith phenotypes (Table 4D). We found that all of these embryos had a substantial frequency of hmx2 nonsense alleles. As predicted by the MENTHU algorithm, the mutated alleles all contained a 5 bp deletion, although some also had additional mismatches in the three bases immediately prior to the deletion and the location of the deletion differed by 1 bp in a few cases. In all cases, we estimate that at least 60% of the amplified hmx2 sequences were mutant (Table 4D; Fig.7C, D). In one of the WT embryos that lacked a phenotype, approximately 90% of the amplified hmx2 sequences were mutant, suggesting that, consistent with the lack of abnormal phenotypes in hmx2 SU39 mutants, CRISPR mutagenesis of hmx2 is not sufficient for abnormal otolith development (Fig.7C, D).
hmx2 and hmx3a are not expressed maternally.
One possible explanation for why the spinal cord phenotype is less severe in hmx2/3a deletion mutants than in morpholino-injected DKD embryos would be if hmx2 and/or hmx3a are maternally expressed, as in this case the morpholinos might knock-down both maternal and zygotic function whereas the mutants would only remove zygotic function. In addition, maternal expression of hmx2 might explain the lack of any obvious abnormal phenotypes in hmx2 single mutants. To test this, we performed in situ hybridization for hmx2 and hmx3a at the 16-cell stage. However, we did not detect any maternal expression of hmx2, hmx3a or any of the other hmx genes (Fig. 8A-E). We also performed qRT-PCR for hmx2 and hmx3a on whole embryos at different developmental stages. We did not observe expression of either gene at either the 16-cell stage or at 6 h, suggesting that neither hmx2 nor hmx3a are maternally expressed (Fig. 8F). At 14 h, shortly after when both of these genes start to be expressed in the ear and spinal cord, we observed low levels of expression and, for both genes, as expected, this became more abundant at 27 h and 48 h (Fig. 8F). Finally, we also generated embryos from adults that were homozygous mutant for hmx2 SU38 , hmx2 SU39 and hmx3a sa23054 . However, even though half of the embryos in each of these crosses should have been maternal zygotic mutants, we did not observe any embryos with fused otoliths (Fig. 4, Table 2).
Even though hmx3b, hmx1 and hmx4 are not normally expressed in the spinal cord (Fig.1), it was theoretically possible that they are upregulated in response to the absence, or reduced levels, of either Hmx2 and/or Hmx3a protein function, in which case they could partially substitute for the loss of hmx2 and/or hmx3a. To test this, we performed in situ hybridization for these genes in hmx3a SU3 and hmx2;hmx3a SU44 mutants at 27 h. In both cases, we did not observe any spinal cord expression of these genes in either genotyped mutants or their sibling embryos, although, as observed previously in WT embryos (Fig. 1) hmx1 and hmx4 were expressed in the eye, ear and anterior lateral line neuromasts in both mutants and WT sibling embryos (Fig. 8G, J, K and data not shown). As expected, given the deletion of the entire hmx3a coding sequence and all but the last 66 bp of hmx2 coding sequence in hmx2;hmx3a SU44 mutants (Fig, 4), we did not detect any hmx2 or hmx3a transcripts in these mutants ( Fig.   8H-I). However, this seems highly unlikely given that the hmx2 gene is almost completely deleted in this allele ( Fig. 4) and concordantly we do not detect any hmx2 transcripts by in situ hybridization (Fig. 8N).
We initially examined the injected embryos down a stereomicroscope and divided them into two groups: those that had an obvious reduction in glutamatergic cells and others that either lacked a phenotype or had a more subtle phenotype. When we genotyped these embryos, we found both homozygous mutant and WT embryos at frequencies that were not statistically significantly different from Mendelian ratios in each group (Table 5). In addition, when we compared the average number of glutamatergic cells in morpholino-injected WT and mutant embryos, there was no statistically significant difference between them, regardless of whether we compared all of the morpholino-injected embryos of each genotype or just compared embryos within the same phenotypic group (Table 5). This suggests that the lack of an abnormal spinal cord phenotype in hmx2 SU39 homozygous mutant embryos is not due to genetic compensation and that the differences that we observed between the two groups of injected embryos instead probably reflect exposure to different levels of morpholino (see materials and methods).
hmx3a SU42 mutant alleles do not lack a spinal cord phenotype because of genetic compensation.
We also tested whether the spinal cord phenotype of hmx3a SU42 mutants is less severe than embryos injected with a hmx3a morpholino because of genetic compensation. As above, we injected the hmx3a morpholino into embryos from a cross of fish heterozygous for hmx3a SU42 and performed in situ hybridization for slc17a6a/b. When we examined the injected embryos down a stereomicroscope, we were able to separate them into one group that had an obvious reduction in spinal glutamatergic cells and another group that either lacked, or had a more subtle, phenotype. However, when we genotyped the embryos in these two groups, we found similar numbers of homozygous mutant and WT embryos in both groups and the frequencies of different genotypes in each group were not statistically significantly different from Mendelian ratios (Table 6). In addition, when we compared the average number of glutamatergic cells in morpholino-injected WT and mutant embryos, there was no statistically significant difference between them, regardless of whether we compared all of the morpholino-injected embryos of each genotype or just compared embryos within the same phenotypic group (Table 6). These data suggest that the lack of an abnormal spinal cord phenotype in hmx3a SU42 homozygous mutant embryos is not due to genetic compensation and that the differences that we observed between the two groups of injected embryos just reflect exposure to different levels of morpholino (see materials and methods). Consistent with this, we also do not observe any nonsense-mediated decay (NMD) of hmx3a mRNA in hmx3a SU42 mutants ( Fig. 8O; NMD has been suggested to play a key role in at least some instances of genetic compensation; El-Brolosy et al. 2019).
hmx3a SU3 mutants also do not have genetic compensation.
We also tested whether hmx3a SU3 mutants have genetic compensation. As above, we injected the hmx3a morpholino into embryos from a cross of fish heterozygous for hmx3a SU3 and performed in situ hybridization for slc17a6a/b. When we examined these embryos down a stereomicroscope, some of the embryos appeared to have a severe reduction in glutamatergic cells that resembled the morpholino knock-down phenotype, whereas in the other embryos any reduction was more subtle. However, when we genotyped these embryos, we again found similar numbers of homozygous mutant and WT embryos in both groups and the frequencies of the different genotypes in each group were not statistically significantly different from Mendelian ratios (Table 7). In addition, when we compared the average number of glutamatergic cells in morpholino-injected WT and mutant embryos, there was no statistically significant difference between them, regardless of whether we compared all of the morpholino-injected embryos, or just compared the injected embryos within a particular phenotypic group (Table 7). For the "weaker" phenotypic group, the difference between WT and mutant embryos approached statistical significance. However, this is probably because some of the WT embryos in this category had almost no reduction in the number of glutamatergic cells, whereas all of the mutants in this category had at least their normal mutant phenotypes. These results suggest that the differences that we observed between the two groups of injected-embryos probably just reflect exposure to different levels of morpholino (see materials and methods), and that embryos in the "weaker" phenotypic group did not receive sufficient morpholino to effectively knock-down hmx3a mRNA or cause the more severe morpholino phenotype.
Taken together, these data suggest that the spinal cord phenotype in hmx3a SU3 homozygous mutant embryos is not less severe than the morpholino knock-down phenotype because of genetic compensation.
Consistent with this, we also do not detect any NMD of hmx3a mRNA in hmx3a SU3 mutants (Fig. 8Q).
Discussion hmx3a is required for correct neurotransmitter phenotypes of a subset of spinal cord interneurons.
In this paper, we identify for the first time, a requirement for hmx3a in spinal cord interneuron development. We demonstrate that hmx2 and hmx3a are expressed by V1 and dI2 interneurons, which is consistent with very recent scRNA-Seq data from mouse spinal cord (Delile et al. 2019). Of these cell types, only dI2 interneurons are glutamatergic. Therefore, the most likely explanation for the reduction in the number of glutamatergic spinal cord cells in hmx3a mutants is that dI2 interneurons are losing their glutamatergic phenotypes. Given that we also detect a corresponding increase in GABAergic spinal cord cells, but the number of V1 cells (indicated by en1b expression) does not change, it is likely that the dI2 interneurons that are losing their glutamatergic phenotypes are becoming GABAergic instead.
Unfortunately, the respective in situ hybridization probes are not strong enough to formally confirm with double-labeling experiments that dI2 interneurons switch their neurotransmitter phenotype from glutamatergic to GABAergic. However, unless Hmx3a is acting in a cell non-autonomous manner, which we think is unlikely as this protein has a nuclear localization sequence and no obvious signal peptide, this is the most likely explanation of our data.
hmx3a is required for progression of the posterior lateral line primordium.
Feng and Xu previously reported that the number of posterior lateral line primordium neuromasts was either severely reduced or completely lost at 3 dpf in hmx2/3a DKD animals (Feng & Xu, 2010).
Intriguingly, the few neuromasts that sometimes persisted were located very rostrally in the embryo, close to the earliest forming somites. Our analyses demonstrate that at 27 h, when the posterior lateral line primordium has migrated to somite 10 in WT embryos, in hmx3a SU3 , hmx3a SU43 , hmx2;hmx3a SU44 and hmx2;hmx3a SU45 mutants the primordium is stalled adjacent to somites 1-4. This suggests that the previously reported loss of neuromasts at 3 dpf is probably caused by the posterior lateral line primordium failing to migrate and deposit neuromasts. Feng & Xu also described reduced cell proliferation (at 15 h) and reduced hmx3a expression (at 24 h) in the posterior lateral line primordium of hmx2/3a DKD animals. Whilst we cannot rule out the possibility that the lateral line primordium fails to migrate because it has not formed correctly, we observe persistent expression of both hmx3a and krt15 in the stalled primordium of our hmx3a SU3 , hmx3a SU43 , hmx2;hmx3a SU44 and hmx2;hmx3a SU45 mutants (data not shown), suggesting that some other mechanism, possibly chemosensory, might underlie the stalled migration.
Hmx3a protein may not require its homeodomain for its functions in viability and otolith, lateral line and spinal cord interneuron development.
Our results also suggest that Hmx3a protein may not require its homeodomain for either its role in viability or its essential functions in otolith development, lateral line progression or correct specification of a subset of spinal cord neurotransmitter phenotypes. This is surprising because most homeodomain proteins act as transcription factors and use their homeodomain to bind DNA and regulate gene expression. Instead, our data suggest that there may be at least one other, as yet undiscovered, crucial functional domain in the N-terminal region of Hmx3a, that is required for its functions in embryo development and viability, as embryos homozygous for hmx3a SU42 are viable and have no obvious abnormal phenotypes and embryos homozygous for hmx3a sa23054 are also viable, have normal lateral line progression and spinal cord interneuron neurotransmitter phenotypes and produce viable progeny. It is highly unlikely that the lack of obvious abnormal phenotypes in these two different mutants is due to an alternative translation start site creating a truncated Hmx3a protein that contains the homeodomain, as the only downstream methionine in hmx3a is more than a third of the way through the homeodomain, and also, in this case we would expect the hmx3a SU3 and hmx3a SU43 alleles to also make this truncated protein. The lack of obviously abnormal phenotypes in hmx3a SU42 and hmx3a sa23054 homozygous mutants can also not be explained by alternative splicing, as these mutations are in the second of two coding exons and also, when we sequenced cDNA made from homozygous hmx3a SU42 mutants we obtained the sequence that we expected (see materials and methods). It is still theoretically possible that there is translational read-through in these two alleles and not in the other hmx3a mutant alleles that have obvious abnormal phenotypes. While this seems unlikely given how similar these different alleles are, we cannot rule out this possibility as we have not been able to identify an antibody that is specific to Hmx3a and we could not detect any Hmx3a peptides in SWATH analysis (see discussion below). However, the most parsimonious explanation of our data so far is that Hmx3a does not need its homeodomain for its functions in viability and otolith, lateral line and spinal cord interneuron development. In this case, while Hmx3a may still bind to other DNA-binding proteins and function in transcriptional complexes, unless the N-terminal of Hmx3a contains a novel DNA-binding domain, Hmx3a is not acting as a classic transcription factor (defined in the strict sense as a protein that binds DNA and regulates transcription) during these developmental processes. Nevertheless, as the homeodomain in highly conserved, suggesting that it is still under evolutionary pressure to be maintained, it is possible that Hmx3a has additional functions that do require this domain, either in adult fish or in aspects of development that we did not assay. However, if this is the case, it is still striking that these functions are not required for such fundamental processes as viability or reproduction.
There are a few other examples of homeodomain proteins that can function in some contexts without their homeodomain. For example, protein interaction and over-expression experiments suggest that Lbx2 does not require its homeodomain to enhance Wnt signaling during gastrulation in zebrafish embryos (Lu et al. 2014). Instead it sequesters TLE/Groucho, preventing this protein from binding to TCF7L1 and reducing TLE/TCF co-repressor activity. In addition, homothorax (hth) does not require its homeobox for its functions during Drosophila head development and proximo-distal patterning of the appendages, although the homeodomain is required for antennal development (Noro et al. 2006). hth has 16 exons and three alternative splice forms. Only one of these isoforms contains the homeobox, but all three contain a protein interaction domain, called the HM domain, that binds to, and can induce the nuclear localization of, Extradenticle (Noro et al. 2006). However, in contrast to hth, zebrafish hmx3a has only two exons and one splice form. While relatively rare, there also examples of transcription factors from other families that only need to bind DNA for some of their functions. For example, Scl/Tal1 has both DNA-binding dependent and DNA-binding independent functions in hematopoietic and vascular development (Porcher et al. 1999;Ravet et al. 2004 Our experiments also suggest that the lack of any obvious abnormal phenotypes in hmx2 SU39 single mutants is not due to genetic compensation or maternal expression of hmx genes. One possible explanation for why zebrafish Hmx2 might have a diminished role in development compared to mouse Hmx2 or zebrafish Hmx3a could be if zebrafish Hmx2 has evolved to be less conserved with mouse Hmx2 and Hmx3 than zebrafish Hmx3a. However, a comparison of all four proteins only reveals six residues that are shared between mouse Hmx2 and Hmx3 and zebrafish Hmx3a but not zebrafish Hmx2, and four of these residues are upstream of the hmx3a SU3 mutation (Supp. Data Fig. 2). Our prior research identified Hmx2 and Hmx3 in all of the different vertebrates that we analyzed, including five different teleost species, and our phylogenetic analyses of these proteins did not suggest that Hmx2 has evolved any faster in zebrafish than in other species, or that Hmx2 has evolved faster than Hmx3 (Wotton et al. 2010). Taken together, these observations suggest that there is still considerable evolutionary pressure to maintain zebrafish Hmx2, which in turn suggests that it should have an important role(s) in zebrafish survival and/or reproduction. Therefore, it is surprising that we did not detect more severe consequences from loss of Hmx2. It is possible that Hmx2 has important functions later in development and/or in aspects of development that we did not assay. However, if this is the case, these functions are not required for viability or reproduction as even hmx2 SU39 homozygous mutants survive to adulthood and produce viable progeny. It is also possible that hmx2 has important function(s) in adult fish, as our assays would not have detected this. It would be interesting to investigate these possibilities in future studies.
Very similar hmx3a mutant alleles have different homozygous mutant phenotypes.
It is currently unclear why hmx3a SU42 retains more WT function than hmx3a SU43 when both should encode proteins with only 107 WT amino acids. As discussed above, we are confident that this is not due to alternative splicing or exon skipping. We also do not observe any NMD of hmx3a mRNA for any of our hmx3a single mutant alleles ( Fig. 8O-R; the double deletion mutants lack all hmx3a coding sequence, so there is no mRNA to assess, Fig. 8I ), so the difference in allelic strength is not due to some of the mutant mRNAs being degraded. However, it is possible that different mutant alleles result in different amounts of truncated protein due to differences in translation efficiency or protein stability. Unfortunately, we were not able to test this as there are currently no antibodies that uniquely recognize Hmx3a and we would require an antibody that recognizes the N-terminal region of Hmx3a that should be conserved in our single mutant Hmx3a proteins. In addition, we were unable to detect any Hmx3a peptides in a SWATH analysis (data not shown; Hmx3a has also not been detected in other SWATH analyses Blattmann et al. 2019;Lin et al. 2019), presumably because, like many transcription factors, it is expressed in either two few cells and/or at too low a level.
It is also possible that the overall length of the mutant protein is important for retaining function and that the additional abnormal amino acids after the frameshift but before the premature stop codon in hmx3a SU42 help this allele retain more WT function. A longer protein sequence might facilitate a required protein conformation and/or binding with other proteins or molecules. If this is the case, then it could also explain why Hmx3a SU42 protein (which is predicted to contain 107 WT + 42 abnormal amino acids, Fig.4) appears to retain more WT function than Hmx3a sa23054 protein (which should contain only 118 WT amino acids; Fig.4). Currently, there are no known binding partners of Hmx3a. However, if future analyses identify any it would be interesting to test if they can bind to Hmx3a SU42 and Hmx3a SU3 .
Another related possibility is that the different stretches of abnormal amino acids after the frameshift in hmx3a SU3 , hmx3a SU42 and hmx3a SU43 might introduce sequences that influence protein stability, degradation or function. Using a variety of online analysis tools, we did not detect any sumoylation, or PEST sequences in any of the predicted mutant protein sequences and the only ubiquitination motifs that we identified are located in the WT sequence present in all four mutant proteins (Rice et al. 2000;Sarachu and Colet 2005;Brameier et al. 2007;Radivojac et al. 2010;Zhao et al. 2014). However, the eukaryotic linear motif prediction tool identified a monopartite variant of a classic, basically-charged nuclear localization signal in Hmx3a SU42 protein that is not present in any of the other predicted mutant protein sequences (Via et al. 2009;Gould et al. 2010;Kumar et al. 2020), although this domain was not detected using default parameters with cNLS Mapper or NucPred (Brameier et al. 2007;Kosugi et al. 2009). This is potentially very interesting as WT Hmx3a has a nuclear localization signal located between amino acids 167-177, overlapping the start of the homeodomain at amino acid 171, which is downstream of the mutations in all of these alleles.
In addition, online tools that identify disordered versus ordered protein structure suggest that both the WT amino acids in Hmx3a sa23054 that are not found in the other predicted Hmx3a mutant proteins and the non-WT amino acids in the predicted protein product of hmx3a SU42 may provide longer stretches of disordered sequence than are present at the end of the predicted protein products of hmx3a SU3 or hmx3a SU43 (Linding et al. 2003;Dosztanyi et al. 2005;Ishida and Kinoshita 2007;Dosztanyi 2018;Meszaros et al. 2018;Erdos and Dosztanyi 2020). These findings raise the intriguing possibility that hmx3a SU42 might retain Hmx3a function because it can still localize to the nucleus and/or that hmx3a sa23054 and hmx3a SU42 might retain some WT activity because of the disordered sequences at the end of their predicted proteins. As disordered protein regions can switch between disordered and ordered states in the presence of a binding partner, it is tempting to speculate that these disordered stretches at the C-termini of Hmx3a SU42 and Hmx3a sa23054 proteins may still be able to bind proteins essential for Hmx3a function that the other alleles cannot (Meszaros et al. 2018). Investigation of these possibilities is outside the scope of the current study but would be interesting to address in future work.
hmx2 and hmx3a morpholino injections produce more severe spinal interneuron phenotypes than hmx2 and hmx3a mutants.
Our original morpholino data suggested that all dI2 interneurons might be switching their neurotransmitter phenotypes as the increase in the number of spinal inhibitory cells and the reduction in the number of excitatory spinal cells in DKD embryos were both roughly equal to the number of dI2 interneurons (glutamatergic hmx3a-expressing cells). However, even in our hmx2;hmx3a deletion mutants, the number of cells changing their neurotransmitter phenotypes is lower than this. The differences between hmx3a morpholino-injected embryos and mutant embryos can be seen clearly in our experiment to test whether genetic compensation occurs in hmx3a SU3 homozygous mutants. These data directly compare uninjected mutants with morpholino-injected mutants and WT siblings from the same experiment. While there were a range of morpholino-injected phenotypes, overall the hmx3a morpholino-injected WT and hmx3a SU3 mutant embryos had a more severe reduction of glutamatergic cells than uninjected mutants (Table 7). While it is possible that hmx3a SU3 mutants may be slightly hypomorphic, their spinal cord phenotype is the same as hmx2;hmx3a SU44 mutants, in which the hmx3a coding sequence is completely deleted. Therefore, the more severe phenotypes in some of the hmx3a SU3 mutants injected with hmx3a morpholino cannot be explained by the morpholino removing any residual Hmx3a function. This experiment also suggests that the less severe phenotype in uninjected hmx3a SU3 mutants is not caused by genetic compensation. Consistent with this, we have also shown that this less severe mutant phenotype is not due to other hmx genes being upregulated in these mutants or maternal expression of hmx3a or any other hmx genes. These results are very puzzling. There are several reasons to suggest that the more severe spinal cord phenotype in DKD embryos is not due to non-specific effects of either the hmx3a or hmx2 morpholino. First, we were able to rescue more glutamatergic spinal neurons in our DKD embryos, with co-injection of either hmx2 or hmx3a morpholino-resistant mRNA than are lost in any of our mutants (hmx2;hmx3a deletion mutants have a reduction of ~ 14 glutamatergic neurons in the region of the spinal cord that we assayed, whereas both mRNA and morpholino co-injection experiments "rescued" about 20 glutamatergic neurons in DKD embryos (Tables 1A & B)). Second, we were able to fully rescue the hmx3a SKD phenotype by co-injecting a morpholino resistant hmx3a mRNA, and the hmx2 SKD phenotype by co-injecting either a morpholino resistant hmx2 mRNA or a morpholino resistant hmx3a mRNA. Third, it is unlikely that the more severe phenotype is due to cell death or a delay in embryo development (which are a common non-specific side effects of morpholino injections), as there was no change in the number of hmx3a-or en1b-expressing spinal cord cells in DKD embryos and there was an increase in the number of inhibitory spinal cord interneurons equivalent to the reduction in glutamatergic neurons. Finally, it is also unclear, why a non-specific effect of a morpholino would exacerbate the real loss of function phenotype, causing additional spinal cord interneurons to lose their glutamatergic phenotypes and instead become inhibitory. This suggests that if the more severe morpholino-injection phenotypes are due to non-specific effects of the morpholinos, these non-specific effects produce an identical phenotype to the specific knock-down effect, namely a switch in neurotransmitter phenotype.
It is also puzzling why hmx2 morpholino-injected SKD embryos have reduced numbers of spinal cord glutamatergic cells and an increase in the number of inhibitory spinal interneurons, while hmx2 SU39 mutants do not, and our experiments suggest that this is also not due to genetic compensation. In addition, co-injection of a morpholino resistant hmx2 mRNA rescues DKD embryos as well as hmx3a injection, even though WT hmx2 is not sufficient for normal development in hmx3a SU3 or hmx3a SU43 mutants. The latter result could be explained by RNA injection providing higher levels of Hmx2 function than is normally found endogenously. However, the first result is harder to explain. As discussed above, our data suggest that it is unlikely that the hmx2 morpholino has non-specific effects on neurotransmitter phenotypes in the spinal cord. These results are also not due to cross-reactivity of the hmx2 translationblocking morpholino with hmx3a. The hmx3a and hmx2 morpholino sequences are completely different from each other and there is no homology between the hmx2 morpholino and hmx3a upstream or coding sequence. When we BLAST the hmx2 morpholino sequence against the zebrafish genome the only homology is with hmx2 (25/25 residues) and with intronic sequence for a gene si:dkey-73p2.3 on chromosome 3 (18/25 residues), which is predicted to encode a protein with GTP-binding activity.
Similarly, when we blast the hmx3a morpholino sequence against the zebrafish genome the only homology is with hmx3a (25/25 residues).
One intriguing alternative possibility that could explain the apparent specificity of the additional phenotype in the morpholino-injected embryos could be that Hmx3a and Hmx2 are acting as fate guarantors, to make the normal neurotransmitter phenotype more robust, as has previously been described for some transcription factors (Topalidou et al. 2011;Zheng et al. 2015;Zheng and Chalfie 2016). In these cases, mutating the gene usually caused a partially penetrant phenotype in ideal conditions but a more severe phenotype in stressed conditions. In this case, the injection of morpholinos could be such a stressed condition, and this could account for the more severe phenotypes that we see in morpholino knock-down experiments compared to mutational analyses. Future analyses could investigate this possibility by testing if other stressed conditions increase the severity of hmx3a SU3 or hmx2 SU39 single mutants or hmx2;hmx3a deletion mutant phenotypes. Even if this is not the case, our results suggest that something other than just non-specific effects from the morpholinos may be occurring. Therefore, we felt that it was crucial to report this morpholino injection data, as an intriguing and hopefully thought-provoking contribution to the continuing discussion about the pros and cons of using morpholinos to investigate gene function.
hmx3b has very limited expression during early zebrafish embryogenesis.
The data in this paper also provide the first characterization of zebrafish hmx3b expression. Surprisingly, For example, in previous analyses we identified three highly conserved non-coding regions in the vicinity of hmx2 and hmx3a (two upstream of hmx3a and one in between hmx3a and hmx2) that are conserved in mammals, frog and teleosts, (Wotton et al. 2010) but none of these regions are present near hmx3b (data not shown).
In conclusion, in this paper we provide the first description of zebrafish hmx3b expression. Our results also identify the spinal cord cells that express hmx2 and hmx3a (dI2 and V1 interneurons) and uncover novel functions for hmx3a in correct specification of a subset of spinal cord neurotransmitter phenotypes and in lateral line progression. Our data suggest that while hmx3a is required for viability, correct otolith development, lateral line progression and specification of a subset of spinal neurotransmitter phenotypes, hmx2 is not, by itself, required for any of these developmental processes, although it can act partially redundantly with hmx3a in situations where hmx3a function is significantly reduced, but not completely eliminated. Finally, our results also suggest that Hmx3a protein may not require its homeodomain for its roles in viability or embryonic development. Taken together, these findings significantly enhance our understanding of spinal cord, ear and lateral line development and suggest that, intriguingly, more homeobox proteins may not require their homeobox domain for many of their essential functions.

Acknowledgments
We thank ZFIN for providing information on nomenclature and other essential zebrafish resources and and NSF IOS 1755354.

55
SJE created all the hmx mutants described in this study except hmx3a sa23054 , performed most of the  Table Legends   Table 1A.      Table 3: Trans-heterozygous mutant analyses identify an allelic series of hmx3a mutants.

Fig. Comparison Gene
Parents heterozygous for different mutant alleles were mated and otolith and lateral line progression phenotypes were assayed in their progeny. Otoliths were assayed at 3 dpf by visual inspection down a stereomicroscope. Lateral line progression was assayed by in situ hybridization for krt15 at 27 h. Column one indicates the allele combination tested. Column two shows the percentage of embryos with otolith phenotypes, column three indicates the total number of embryos analyzed. In contrast to incrosses of hmx3a SU3 or hmx3a SU43 , where mutant embryos have fused otoliths in both ears ("severe" phenotype), in some trans-heterozygous crosses we observed two additional types of ear phenotypes. Phenotypes were classified as "moderate" if otoliths were adjacent but not fused in both ears, and as "weak" if there was an otolith fusion or adjacent otolith phenotype in only one ear. The percentage of embryos with each of these phenotypes is provided in columns five (severe), six (moderate) and seven (weak). Column eight indicates the percentage of embryos with stalled lateral line progression and column nine shows the total number of embryos analyzed. In contrast to incrosses of hmx3a SU3 or hmx3a SU43 , where mutant embryos lack any migration of the lateral line primordium along the trunk ("strong" phenotype), in some trans-heterozygous crosses, we observed embryos where the lateral line primordium had migrated slightly more caudally ("weaker" phenotype). Column 11 shows the percentage of embryos with the strong phenotype, column 12 shows the percentage of embryos with the slightly weaker phenotype. Chi-squared tests were performed to test if the frequency of embryos with otolith fusion or lateral line phenotypes was Mendelian and the P values for these tests are provided in columns 4 or 10 respectively. Statistically significant values are indicated in bold. We also performed a binomial distribution test, using the cumulative distribution function, to test whether the number of embryos from the SU39 x SU44 cross that had fused otoliths was statistically significantly different from zero. P=0.264 for this test.    Table 4A) and otolith phenotype (fused in both ears, two normal otoliths in both ears, as in Table 4A). Columns 2-4 show hmx3a SU42 genotypes. Column 5 = total number of embryos with each combination of lateral line primordium and otolith fusion phenotypes. Row 7 = total number of embryos with each hmx3a SU42 genotype.           Table   1A). Statistically significant (p < 0.001) comparisons are indicated with brackets and three asterisks. All data were first analyzed for normality using the Shapiro-Wilk test. Data set in G is non-normally that the lateral line primordium (LLP) has migrated to its expected position over somite 10 (S10 + black arrow) (A). In contrast, at 27 h in hmx2;hmx3a DKD embryos, the LLP is stalled beside somites 1-4 (S1, S4, black arrows, B). This is identical to the stalled LLP phenotype observed in hmx3a SU3 and hmx2;hmx3a SU44 mutants (see Fig. 5O  Left-hand side: schematics of 11 mutant alleles and one double mutant analyzed. Top row = genomic locus; lower rows indicate predicted protein products. There is a 6454 bp gap between hmx2 and hmx3a. Vertical black bars on genomic locus indicate locations of sgRNA sequences, A-F, used to generate the mutants shown. These sequences and the combinations of sgRNAs used to generate the mutants shown here, are listed in Table S2. For each mutant allele, the genomic location plus the nature of the mutation or indel size is shown in brown text at the right-hand side of each mutant protein schematic. Coding bases refer to the translated sequence, e.g. coding bases 1-3 correspond to the bases encoding the start methionine. Right-hand side: Column 1: allele number. Column 2: indicates whether embryos with fused otoliths were observed in incrosses of heterozygous (z) or homozygous (mz) parents. (Also see Table 2 and Fig. 5). Column 3 indicates whether a reduction in the number of spinal excitatory cells was observed at 27 h in homozygous mutants, as assayed by in situ hybridization for slc17a6a/b (Fig. 5 and Table 1B). All counts are an average of at least 5 embryos. Statistically significant (p < 0.001) comparisons are indicated with brackets and three asterisks. White circles indicate WT data and black circles the appropriate mutant data as indicated in key under panel A. All data were first analyzed for normality using the Shapiro-Wilk test. Data in G is not normally distributed and so a Wilcoxon-Mann-Whitney test was performed. Data sets in H-K are normally distributed and so the F test for equal variances was performed, followed by a type 2 student's t-test (for equal variances). P-values are provided in Table 1B    control: ± 1.9, hmx2;hmx3a DKD: ± 1.7). All counts are an average of 10 embryos. The statistically significant (p < 0.001) comparison is indicated with brackets and three asterisks. All data were first analyzed for normality using the Shapiro-Wilk test. The data are normally distributed and so the F test for equal variances was performed. Since the variances were equal, a type 2 student's t-test was performed (p = <0.0001). These data show that following co-injection of hmx2 and hmx3a spliceblocking morpholinos, there is a statistically significant reduction in the number of excitatory