nkx3.2 mutant zebrafish accommodate jaw joint loss through a phenocopy of the head shapes of Paleozoic jawless fish

The jaw is a functionally versatile innovation that facilitated explosive diversification of gnathostomes (a clade containing jawed vertebrates), but its basic structure is surprisingly simple and highly conserved (Miyashita, 2016). A jaw consists of ‘a hinge and caps’: upper and lower skeletal levers hinged at a jaw joint (Depew and Simpson, 2006). As the joint enables biting motions, its origin is considered the critical step in the evolutionary assembly of the vertebrate jaw (Cerny et al., 2010; Kuratani, 2012; Miyashita, 2016). Across jawed vertebrates, the presumptive jaw joint is marked by the expression of nkx3.2, an NK2 class homeobox gene (a.k.a. bapx), at the mid-height of the embryonic mandibular arch (Gillis et al., 2013; Lukas and Olsson, 2018a; Miller et al., 2003; Tucker et al., 2004). Chondrogenesis dorsal to this expression domain gives rise to a palatoquadrate (upper jaw), whereas chondrogenesis ventral to it forms Meckel's cartilage (lower jaw) (Medeiros and Crump, 2012). This basic pattern remains conserved among jawed vertebrates, but later development varies. Marginal bones arise intramembranously around the often endo-/peri-chondrally ossified jaw cartilages, except in chondrichthyans (sharks, rays and skates) (Hall, 2015). In mammals, the jaw joint forms between two such intramembranous bones (temporal and dentary), whereas the proximal jaw joint becomes the malleus–incus interface that is, in mice, no longer affected by Nkx3.2 knockout (Tucker et al., 2004). Despite these variations after pharyngeal chondrogenesis, no gnathostome lineage has secondarily lost functional jaws.

Our aim here was to find out whether – and how – jaw function affects vertebrate skull shape during development. Clinically documented agnathia in humans typically accompanies severe congenital disorders such as holoprosencephaly (forebrain defect with a failure to form cerebral hemispheres) and otocephaly (lethal malformation of the first pharyngeal arch), but jaw loss is clearly a secondary effect and not a cause in these cases (Bixler et al., 1985; Brown and Marsh, 1990; Gekas et al., 2010; Schiffer et al., 2002). Instances of temporomandibular joint ankylosis (stiffening of the mammalian jaw joint due to bone fusion) may result from trauma or infection, or may be congenital (Adekeye, 1983; Chidzonga, 1999; Manganello-Souza and Mariani, 2003). If untreated, the ankylosis can lead to the ‘bird face’ deformity in humans (El-Sheikh et al., 1996). However, these cases do not fully document the effects of functional jaw loss. In mammalian models, various jaw/skull deformations have been induced by surgical resection, detachment or repositioning of the jaw muscles and/or bones (Bayram et al., 2010; Gomes et al., 2012; Horowitz and Shapiro, 1955; Lifshitz, 1976; Miyazaki et al., 2016; Rodrigues et al., 2009; Sarnat, 1970; Sarnat and Muchnic, 1971; Toledo et al., 2014). These manipulations occurred well after formation of the jaw skeleton and muscles, and the jaws remained partially functional because operations were unilateral or because surgeries only affected a component of the jaw system. The defects and deformities reported in these studies imply: (a) jaw movements are potentially an important factor in shaping the skull; and (b) any allele disrupting jaw movements would be generally maladaptive. Nevertheless, these implications from mammalian systems are difficult to explore without an accessible experimental model.

To fill this gap, we engineered two distinct null alleles of nkx3.2 in zebrafish. Previously, transient knockdown of nkx3.2 during early development (using morpholinos) resulted in fusion of the nascent jaw cartilages in both zebrafish and frogs (Lukas and Olsson, 2018a; Miller et al., 2003). We confirmed in zebrafish that the mutants recapitulate this phenotype. Surprisingly, mutant zebrafish are viable – despite loss of the jaw joints – and grow through to adulthood. Functionally jawless as a result, nkx3.2^−/− zebrafish develop dramatically altered skull shape: the mouth is fixed open, the snout reduced and the branchial region expanded. The late onset and broad extent of morphological changes indicate that this open-mouth phenotype is induced by functional agnathia, secondarily to nkx3.2 loss of function. Although previously unknown in zebrafish or any other jawed vertebrates, similar head shapes also occurred in two extinct lineages of ≥400 million year old jawless vertebrates, anaspids and thelodonts. In the absence of any homology in individual facial bones, this convergence implies that the adult mutant phenotype is an adaptive response to functional requirements of feeding without mobile jaws. Thus, nkx3.2 mutant zebrafish provide a unique model for both skeletal remodeling and joint diseases, and to re-evaluate evolutionary implications of mutant phenocopies in general.

Zebrafish [Danio rerio (F. Hamilton 1822)] maintenance and experiments were approved as protocol number AUP00000077 by the Animal Care and Use Committee: Biosciences at the University of Alberta, as dictated by the Canadian Council on Animal Care. Other zebrafish work was approved by the University of Southern California Institutional Animal Care and Use Committee.

Embryos were incubated at 28°C, and treated with 0.003% PTU (1-phenyl-2-thiourea) in 10% Hank's saline starting at 24 h post-fertilization (hpf). Larvae were introduced into the nursery at 1 week to 10 days post-fertilization (dpf). Genomic DNA was extracted from clipped fins of 3–5 dpf larvae or from adults.

Nkx3.2 protein is a transcription factor with a homeobox DNA binding domain that is 100% conserved in amino acid sequence among zebrafish, mouse and human homologs. In zebrafish, a single nkx3.2 gene is apparent in the genome, and its homology to mammalian NKX3.2 is strongly supported by gene synteny, e.g. the neighbor genes flanking nkx3.2 on zebrafish Chromosome 14 (wdr1 and bod1l1) are positioned coordinately in mouse, human and spotted gar.

Two disparate regions of the gene were targeted by CRISPR or TALEN, producing two disparate null alleles that produced similar phenotypes (Fig. 1A). One allele, nkx3.2^ua5011, was engineered with CRISPR/Cas9 (Gagnon et al., 2014) targeted at the beginning of the homeodomain, and harbors a 20 bp deletion resulting in a frameshift (Fig. 1A; Dryad data supplement 1, https://doi.org/10.5061/dryad.8w9ghx3j9). The disrupted translation of codons is predicted to abrogate production of the critical homeobox domain, and instead produce random amino acids. This is predicted to produce a non-functional Nkx3.2 protein and a null allele. A disparate allele (el802) was generated using TALEN (Barske et al., 2016) targeted at the start of the gene. This produced a stably inherited gene with 14 bp deletion, removing A of the ATG translation start codon (Fig. 1A). The allele nkx3.2^el802 is predicted to not produce Nkx3.2 protein. Morphologically, these two alleles are not readily distinguishable from each other (see Results).

To generate nkx3.2^ua5011, we designed sgRNA to five different targets near or within the nkx3.2 homeodomain, which were all injected: GGCGGCCATCTGACGTCGCT; GGCTGACGCCAGCAGATCGG; AAGCAGCGGAAGAAGCGCTC; GAGCGCTTCTTCCGCTGCTT; and GGCCGCGTTCTCCCACGCGC. These targets were selected using the web resource CHOPCHOP (Labun et al., 2016; Montague et al., 2014). Following the protocol developed by Gagnon and colleagues (2014), we prepared a template for each sgRNA using two different oligonucleotides: one containing a target sequence led by the SP6 promoter (ATTTAGGTGACACTATA) and followed by the overlapping region (GTTTTAGAGCTAGAAATAGCAAG) of the reverse oligonucleotide, and the reverse containing the constant, Cas9-binding domain of sgRNA. These oliogonucleotides were annealed after 5 min incubation at 95°C, through graded cooling (−2°C s⁻¹ to 85°C; −0.1°C s⁻¹ to 25°C), and filled in for the non-overlapping regions using T4 DNA polymerase (NEB: M0203S). To synthesize sgRNAs using these templates, MegaScriptTM SP6 Transcription Kit (Ambion: AM1330) was used. The RNAs were precipitated in ammonium acetate solution, resuspended in UltraPure™ H₂O and stored in 2–3 μl aliquots at −80°C. For injection, sgRNAs were diluted to 400–600 ng μl⁻¹, and 1 μl was mixed with 1 μl aliquot of Cas9 nuclease from Streptococcus pyogenes (NEB: M0646) at 1 μg ml⁻¹. This solution was mixed with 3 μl of 0.2 mol l⁻¹ KCl, 0.2% Phenol Red and ddH₂O. The final injection volume per embryo was approximately 5 nl, containing 400–600 pg sgRNA and 1 ng Cas9 nuclease. For control, GFP 5′GA was used in P₀ (ubi:switch/RH+AB was crossed at this stage), which can be phenotyped by reduction of ubiquitous GFP in the progenies with dsRed expression in the heart.

We generated nkx3.2^ua5011 against an AB background. Cas 9 and sgRNAs targeted for nkx3.2 were co-injected with sgRNA that disrupts GFP GA5′ (CTCGGGCTGAAGCTCGGCG), at stages between fertilization and first cleavage, into fertilized eggs collected from the crossing of ua3140 ubi:switch/AB+RH and the background AB line. At 3 dpf, injected larvae were sorted for reduced expression of ubiquitous GFP and the presence of dsRed fluorescence in the heart. These larvae provided the P₀ population. Sequencing of genomic DNA extracted from fin clips of the P₀ adults identified a female with a 20 bp deletion in the homeodomain-coding region of nkx3.2 (ua5011). The P₀ female carrying this mutation was crossed to the sox10:GFP transgenic line, and progenies were sorted at 3 dpf for the presence of sox10:GFP expression and the absence of the two markers (ubiquitous GFP expression and fluorescent red heart). ua5011 heterozygotes were identified by both sequencing of extracted gDNA and restriction fragment length polymorphism (RFLP) analysis using one XmaI (NEB: R01805) restriction site within the deleted region (primers for genotyping: 5′-GGACGAGACGGATCAGGAATC-3′; 5′-CACTCGGCGTGTTCGGTAAA-3′). These F₁ heterozygotes were incrossed for F₂ embryos, which were genotyped by RFLP analysis and phenotyped at 4 dpf by identifying sox10:GFP-positive chondrocytes and staining cartilages using Alcian Blue. Homozygotes were reared with a strictly small-grained diet to the adult stage. In this study, nkx3.2^ua5011 fish represent the F₂ generation derived from the P₀ mutant female and a wild-type male (AB; sox10:GFP), whereas the comparative wild-type (AB; sox10:GFP) fish come from incrossing of half-siblings of the P₀ male.

One- and two-month-old nkx3.2^+/+ and nkx3.2^−/− zebrafish were fixed in 4% paraformaldehyde (PFA) for 24 h. Zebrafish were eviscerated prior to decalcification with 0.5 mol l⁻¹ EDTA for 4 weeks. Samples were dehydrated post-decalcification in a series of graded ethanol and embedded in paraffin. Tissue blocks were embedded in a sagittal orientation and sections were cut at 7 μm using a 820 Spencer microtome. Hematoxylin and eosin staining was performed on zebrafish sections by firstly placing sections in an oven at 60°C for 10 min. The deparaffinized sections were rehydrated using xylene and graded ethanol (100%, 95%, 70%), followed by staining with hematoxylin and eosin. The slides were then dehydrated and mounted using Permount.

Alcian Blue staining of cartilages partly followed the protocol developed by Michael Shapiro (University of Utah). Specimens fixed in 4% PFA were rinsed with ddH₂O and transferred to 70% ethanol. Once equilibrated, larvae were immersed in Alcian Blue solution (0.167 mg ml⁻¹ Alcian Blue, 15% acetic acid, 70% ethanol), rinsed through an ethanol/ddH₂O series, and washed in a saturated sodium borate solution. Specimens were immersed in trypsin solution (0.125% trypsin, 30% sodium borate) overnight, washed in 1% KOH solution, bleached in 0.15% H₂O₂, 0.1% KOH, 25% glycerol solution, and immersed through a 1% KOH/glycerol graded series into 100% glycerol for storage. Specimens older than 21 dpf were immersed in 0.005% Alizarin Red solution in 1% KOH overnight, after the first 1% KOH wash and before bleaching in 0.15% H₂O₂, 0.1% KOH. We selected intact skeletal preparations for linear morphometrics (wild-type: 4 dpf, n=7; 14 dpf, n=7; 21 dpf, n=27; 30 dpf, n=60; 60 dpf, n=21; nkx3.2^ua5011: 4 dpf, n=5; 14 dpf, n=9; 21 dpf, n=22; 30 dpf, n=42; 60 dpf, n=18).

Wild-type (AB; sox10:GFP) and nkx3.2^ua5011 (one male and female for each genotype) were filmed at 2 months post-fertilization (mpf) to show feeding (Movie 1). A fish was placed in a 1.4 l tank with a dark background and given brine shrimp larvae. Feeding was filmed using Canon EOS 760D at 30 frames s⁻¹ in dimensions of 1280×720 pixels. The films were cropped and assembled using iMovie (v. 10.1.9, © 2001–2018 Apple Inc.) and slowed to 1/10 original speed.

Wild-type (AB; sox10:GFP) and nkx3.2^ua5011 zebrafish were scanned using MILabs μCT at the School of Dentistry, University of Alberta. For µCT scanning, 2-month-old individuals were selected from two different F₂ populations for each genotype (wild-type: n=10; nkx3.2^ua5011: n=11), fixed in 4% PFA for 24 h, and dehydrated in a graded ethanol series. The following parameters were applied for scanning: voxel size 10 µm, voltage 50 kV, current 0.24 mA and exposure time 75 ms. Skeletons were reconstructed by selecting and delineating regions of high tissue density at a voxel size of 25 μm using AVIZO (Milabs, Utrecht, The Netherlands). The µCT-based skull reconstructions were used for anatomical comparison and geometric morphometrics.

Fluorescent images were acquired on a Zeiss Axio Observer.Z1 with an LSM 700 confocal microscope via ZEN 2010 software (version 6.0, Carl Zeiss MicroImaging, Oberkochen, Germany). Brightfield imaging of stained preparations was performed on a Leica MZ16F dissection microscope (Concord, ON, Canada) with a 12.8 megapixel digital camera (DP72, Olympus, Richmond Hill, ON, Canada). We selected lateral views captured of sox10:GFP positive fishes at 4 dpf for linear morphometrics (n=15 for wild-type; n=13 for nkx3.2^ua5011 zebrafish).

The purpose of this comparison was to quantitatively test the qualitative phenotypic similarities and differences among nkx3.2^−/− mutants, wild-type zebrafish and anaspids (and thelodonts for gape angles). Within zebrafish, the skulls of nkx3.2^−/− mutants morphologically depart from wild-type at adult stage (Fig. 2). To describe these morphological differences quantitatively, we compared skeletal growth between nkx3.2^−/− and wild-type zebrafish. We also assessed similarities between the nkx3.2^−/− phenotype and the general head configuration in anaspids and furcacaudiid thelodonts. Zebrafish and anaspids are phylogenetically distant: the former is a cypriniform nested deep within gnathostomes, whereas anaspids represent either a stem gnathostome or even a stem cyclostome lineage (Donoghue et al., 2000; Janvier, 2007; 1996; Keating and Donoghue, 2016; Miyashita et al., 2019). Furcacaudiids form a small group within thelodonts, a jawless stem gnathostome lineage (Märss et al., 2007; Wilson and Caldwell, 1993). The skull roof and facial bones of a zebrafish are macromeric. In contrast, the skull of an anaspid is largely micromeric with scales of acellular bone (Blom et al., 2001), and that of a thelodont also consists of micromeric scales with dentinous crowns and bases of acellular bone (Märss et al., 2007). There is no morphological correspondence in individual elements of the skull roof between zebrafish and anaspids. The parabranchial cavity is closed by the operculum in zebrafish, whereas each branchial pouch had its own outlet in the series of external pores in anaspids and thelodonts (Blom et al., 2001; Janvier, 1996). A single gene mutation in nkx3.2 did not reverse these, and other morphological differences accumulated after the last common ancestor of zebrafish and anaspids. For morphometric comparison, we chose metric traits that could be identified in both zebrafish and anaspids.

We compared the gape among zebrafish, anaspids and thelodonts. In zebrafish, the angle was taken by extrapolating the axis of the premaxilla until it meets the axis of the dentary. At stages younger than the onset of dermal ossification (4 and 14 dpf), the angle was measured between the axes of the palatal process (palatoquadrate) and Meckel's cartilage. In anaspids, the upper and lower lips are demarcated by a series of plates or relatively larger scales, which allowed delineation of the gape angle (measured where the extrapolated upper and lower lip margins meet). The angle was measured similarly in thelodonts, except that the lip margins were identified along the series of small marginal scales. Gape angle exhibits a roughly normal distribution within each age class of both nkx3.2^−/− and wild-type zebrafish and among anaspids (for original measurements, see Dryad data supplement 2, https://doi.org/10.5061/dryad.8w9ghx3j9). In addition to gape angles, we measured the length of the skull and lower jaw in zebrafish, and orbit diameter in anaspids and thelodonts.

Eight landmarks were assigned to both zebrafish (2 mpf) and anaspid samples for geometric morphometric comparison. These landmarks capture the general configuration of the head (1: anterior tip of upper lip; 2: junction between upper and lower lips; 3: anterior tip of lower lip; 4: nostril, or nasohypophyseal opening; 5: anterior extremity of orbit; 6: posterior extremity of orbit; 7: trunk–head boundary at dorsal outline; 8: ventral point of hypobranchial region). They describe structures that are homologous across vertebrates (landmarks 4, 5, 6) or geometrically determined comparable positions. They are free of morphological discontinuity between anaspids and gnathostomes (none of the landmarks represent anaspid- or gnathostome-specific morphology). TpsDig (Rohlf, 2018) was used to place landmarks on 2D images of anaspids and zebrafish. The digitized file was entered into MorphoJ software (Klingenberg, 2011) and all images were aligned using the anterior and posterior extremity of the orbits. These landmark data were transformed using the Procrustes superimposition method (Rohlf, 1999), and the resulting coordinates were compared using principal component analysis (PCA). In PCA, PC scores from anapsids, nkx3.2^+/+ and nkx3.2^−/− zebrafish were grouped by equal frequency ellipses with a P-value of 0.95.

From the pool of nearly a thousand cataloged specimens of anaspids, we selected a total of 70 specimens that show lateral compression during the fossilization process, with the least taphonomic distortion, to reflect the lateral view of the head. Furcacaudiid thelodonts were excluded from geometric morphometrics because only a handful of exceptionally preserved specimens are available; the sample size is small for this group (n<5), and all such specimens were collected from a single locality, making it difficult to identify (and thus control for) taphonomic artifacts. In addition, some landmarks could not be assigned confidently in this group. The transition from head to trunk is ambiguous because there is no apparent change in scale morphology (Wilson and Caldwell, 1993, 1998).

There are many other lineages of jawless vertebrates, including living cyclostomes, heterostracans, galeaspids and osteostracans (Janvier, 1996, 2007). Living cyclostomes are difficult to compare as they have an anguilliform profile and lack ossified skeletons, or cartilages unambiguously comparable to the jaw/lip elements of the zebrafish skull (Miyashita, 2012, 2016; Miyashita et al., 2019). Other jawless vertebrates generally have a depressiform profile (Janvier, 1996). For this reason, it is difficult to directly compare these agnathans with the compressiform zebrafish and anaspids. Some qualitative generalization is still possible. Furcacaudiid thelodonts have a lateromedially compressed body profile, and have a lip morphology consistent with that of anaspids (Wilson and Caldwell, 1998, 1993). These similarities suggest that depressed lower lips are a general condition in both cyclostome and gnathostome stems (Janvier, 1996), as reconstructed conventionally across these lineages.

At 4 dpf, nkx3.2^−/− zebrafish replicated the nkx3.2 morpholino knockdown phenotype: they lacked a jaw joint (Miller et al., 2003). The palatoquadrate and Meckel's cartilage fused together, and the retroarticular process was absent (Fig. 1D,G,I). Apart from joint ankylosis, no marked differences were apparent in overall morphology or survival rates between nkx3.2^−/− mutants and wild-type at this stage (some minor differences in skull size and lower jaw proportions are discussed in the next section). Heterozygotes were morphologically indistinguishable from wild-type (Fig. 1C,F), and the nkx3.2 alleles displayed recessive Mendelian inheritance (F₂ genotypes followed Mendelian ratio; heterozygotes developed wild-type morphology). Remarkably, these functionally jawless homozygous mutants survived beyond the early larval stages.

Despite the maladaptive nature of jaw dysfunctions in general, nkx3.2^−/− zebrafish continued to grow without a jaw joint. Marked phenotypic differences from wild-type began to emerge between the 2nd and 3rd weeks post-fertilization (Fig. 2). The lower jaw became downturned in nkx3.2^−/− fish, resulting in a fixed open mouth, whereas both upper and lower jaws were upturned in wild-type zebrafish (Fig. 2A,B). This timing coincides with the onset of ossification and the period of active feeding in normal juveniles (Cubbage and Mabee, 1996; Kimmel et al., 1995). Although the palatoquadrate and Meckel's cartilage remained fused in nkx3.2^−/− mutants, skeletal staining revealed that the upper and lower jaw elements ossified independently of each other – still without a ball-and-socket joint structure (Fig. 3A,B). All skull elements in nkx3.2^−/− mutants ossified without apparent delay.

The lower jaws were increasingly turned downward in nkx3.2^−/− by the end of the first month, resulting in a greater gape (Fig. 2D,E). The jaw joint was still absent in nkx3.2^−/− mutants: a sheet of perichondrium lay between the ossifying quadrate and articular – so the two bones remained distinct elements – but this interface had no other essential components of synovial diarthrosis (Fig. 3D,E). Many additional osteological differences emerged by this stage. Normally, the premaxilla and the maxilla swing forward to sit nearly vertical (Hernandez, 2000; Hernandez et al., 2007) (Fig. 2C). In nkx3.2^−/− mutants, however, the premaxilla and the maxilla became oriented posteroventrally and abutted against the antorbital wall (Fig. 2D). The downturned lower jaws of the nkx3.2^−/− mutants were relatively shorter than the normal lower jaws of wild-types. The basihyal protruded anteroventrally as much as the lower jaw, implying that the muscle connecting those two elements (m. intermandibularis posterior) (Schilling and Kimmel, 1997) may be responsible for lower jaw orientation. As a result of these modifications, nkx3.2^−/− mutants had a shorter snout, a fixed open gape, and a dorsoventrally tall profile.

Linear morphometrics corroborated the departure from normal morphology in nkx3.2^−/− mutants in the second half of the first month (14 dpf onward). For absolute size, no significant difference (P>0.05) in skull length emerged between wild-types and nkx3.2^−/− mutants except at 4 dpf (t=2.202; P=0.338) (Fig. 4B). Also at this stage, the lower jaws appeared to be shorter relative to skull length in the mutants than in the wild-types (t=2.809; P=0.0078) (Fig. 4C). These minor but statistically significant differences at 4 dpf may have been due primarily to the fusion of jaw cartilages. By the second week, however, differences between the mutants and wild-types became non-significant in these metric traits. The lower jaw depression (gape angle >45 deg) in the mutants was pronounced at 21 dpf (t=−11.834; P<<0.01) (Fig. 4A), but proportional changes to lower jaw lengths in the same mutants were only expressed in significant magnitude at 1 month (21 dpf: t=1.7225; P=0.091559; 30 dpf: t=10.56; P<<0.01) (Fig. 4C). This lag indicates that the rate of skeletal growth shifted in the jaws after their orientations changed. Morphological variations within a cohort of nkx3.2^−/− mutants were greater at this stage than in any other (Fig. 4A,C).

At 2 months of age and approaching sexual maturity, nearly all nkx3.2^−/− mutants had a gape angle greater than 90 deg (Fig. 4A). These nkx3.2^−/− adults continued to be characterized by the morphological differences identified at 1 mpf. The skulls appeared to be shorter, wider and more highly ossified in nkx3.2^−/− mutants than in wild-types (Figs 2 and 5; Movie 1). The short profile of the mutant skulls was primarily due to reduced relative size of the snout. With the premaxilla and maxilla held rigid, the kinethmoid – a sesamoid that acts as a hinge between upper jaw and skull roof (k in Fig. 5A,C,D; Hernandez et al., 2007) – was absent in the fully ossified nkx3.2^−/− adults. The nostrils sat between the eyes in the mutants, whereas they were level with the anterior margin of the orbit in wild-types [compare nasals (na) with respect to lateral ethmoids (le) in Fig. 5A–F]. The skull width was apparently greater in the mutants than in the wild-type zebrafish across the skull roof and between the opercula (Fig. 5C–J). The circumorbital bones including supraorbital (sr) and jugal (j) became massive and flared laterally in the mutants (Fig. 5B). In comparison to wild-type zebrafish, the nkx3.2^−/− mutants developed substantially larger basihyals (bh), laterally oriented ceratobranchials (cb), and a larger mass of bones and cartilages around the interface of the quadrate (q) and the articular (aa) (Fig. 5B,F,I,J). The occiput had a larger surface area for the epaxial muscles to attach to (Fig. 5B). All homozygous mutants showed the descriptive skeletal traits identified here. Morphologically, the phenotypic effects of the two alleles (nkx3.2^ua5011 and nkx3.2^el802) were virtually indistinguishable from each other at respective ages (Fig. 2G,J), given variations within a strain (for examples, see variation in gape angle, relative length of lower jaws, or sexual dimorphism in nkx3.2^ua5011: Figs 2M,N and 4A,C).

With their modified skulls, adult nkx3.2^−/− mutants exhibited feeding markedly different from wild-type. Wild-type zebrafish fed by suction in a sequence consistent with general teleost feeding mechanics (Alexander, 1969, 1970; Lauder, 1979, 1980; Westneat, 2004, 2005): rapid lower jaw depression; a forward swing of the mobile premaxilla–maxilla complex and head elevation; maximum expansion of the parabranchial cavity; and closing of the operculum, while holding themselves steady with backstrokes of the pectoral fins (Fig. 6A; Movie 1). In contrast, adult nkx3.2^−/− mutants were constrained by the fixed upper jaw unit and open gape. They instead showed ram feeding (swimming through food) (Fig. 6B; Movie 1). The jaws remained fixed, and no significant movement was observed in the lips. The mutants elevated the head and the basihyal before ingesting food, shortly followed by maximum expansion of the parabranchial cavity (Fig. 6B). These externally visible movements may provide weak suction to assist the ram feeding. Whereas in wild-type one complete cycle of the jaw opening and closing took approximately 80 ms (and approximately one-tenth of a second to close the operculum), nkx3.2^−/− mutants required double that time from changing direction of swimming toward the food (0 s) to doing so again away from the food (0.2 s) in the particular feeding episode in Fig. 6B.

This ram feeding was correlated with skull remodeling in nkx3.2^−/− mutants. Until passing 1 mpf, the fusion between jaw cartilages was not completely ossified in these mutants, potentially allowing plastic deformation (Fig. 2E,L). These observations suggest that phenotypic changes (open mouth and ram feeding) accommodate functional jawlessness in nkx3.2^−/− zebrafish.

Through this dramatic transformation, nkx3.2^−/− zebrafish assumed a head shape reminiscent of two lineages of extinct jawless vertebrates that also have laterally compressed body profiles: (a) birkeniiform anaspids (Fig. 7C), stem cyclostomes known mostly from the Silurian period (Blom et al., 2001; Miyashita et al., 2019); and (b) furcacaudiid thelodonts (Fig. 7D), much more elusive stem gnathostomes known from the Silurian and Devonian periods (Märss et al., 2007; Wilson and Caldwell, 1998). Qualitatively, the resemblance is particularly striking in functional and proportional features, including: fixed gape (depressed lower lips), shortened snout, interorbital position of the nostril, relatively large branchial region and massive occiput.

In linear morphometrics, the gape angle revealed that nkx3.2^−/− mutants closely resemble the anaspid- and thelodont-like conditions late in ontogeny (Fig. 4A). The trait in nkx3.2^−/− mutants departed from the wild-type condition after the onset of skull ossification and active feeding (14–21 dpf), and overlapped with the range occupied by anaspids and thelodonts in later stages (1 and 2 mpf). When the jaws were at rest, the gape angle in wild-type zebrafish was consistently around 30 deg regardless of ontogenetic stage. This was also true for nkx3.2^−/− mutants at 4 and 14 dpf when they still rely on yolk as the main or partial intake. The gape angle increased steadily thereafter in the mutants. In anaspids, the gape angle appeared to vary independently of body size. No correlation was detected between gape angle and orbit diameter in this clade (r=−0.114; P=0.306).

In landmark-based geometric morphometrics, nkx3.2^−/− mutants aligned more closely with anaspids than wild-type zebrafish along the principal components (PCs) that explain the adult phenotype (PC1 and PC5) (Fig. 7E,F). Nearly one-third of the overall shape variation loaded on PC1 and primarily concerned anteroposterior length and dorsoventral height of the entire head. Compared with wild-type zebrafish, nkx3.2^−/− mutants and most anaspids had a much shorter snout (with the lips and the nostrils shifting posteriorly toward the eye) and dorsoventrally deeper lower lips. PC 5 explained 7.0% of overall shape variation, and the traits that varied along it were orientation of lower lip and relative height of nostrils, which clearly set wild-type zebrafish apart from nkx3.2^−/− mutants and anaspids. On the plot of PC1 and PC5, nkx3.2^−/− mutants nearly entirely overlapped with anaspids in morphospace occupation and apart from wild-type zebrafish (Fig. 7F). PC2–4 (not shown in Fig. 7) largely explained variation among anaspids or within wild-type/mutant zebrafish samples and were therefore uninformative for comparison between the groups (see Dryad data supplement 3, https://doi.org/10.5061/dryad.8w9ghx3j9).

Our nkx3.2^−/− zebrafish reinforce the morphant-based insight that this transcription factor is essential to the development of the jaw joint in non-mammalian vertebrates (Miller et al., 2003, 2004). This dependence is likely widespread among jawed vertebrates. Nkx3.2 knockdown also results in fusion between the palatoquadrate and Meckel's cartilage in amphibians (Lukas and Olsson, 2018a), and a similar phenotype is observed in chicks with ectopic expression of BMP4 and FGF8 that repressed Nkx3.2 (Wilson and Tucker, 2004). In mammals, the palatoquadrate (malleus) and Meckel's cartilage (incus) migrate to form middle ear ossicles. The malleus–incus joint is not affected in Nkx3.2^−/− mice, but Nkx3.2 plays a role in specification of the gonium and the anterior tympanic ring and maintains the malleus size (Tucker et al., 2004). Aside from its function in the head, Nkx3.2 plays a role in various structures in mice and chicks, including the axial column (Herbrand et al., 2002; Lettice et al., 2001; Murtaugh et al., 2001) and visceral lateralities of the spleen and pancreas (Hecksher-Sørensen et al., 2004; Schneider et al., 1999). Comparing these phenotypes identified in amniote models with nkx3.2^−/− zebrafish requires further analysis to determine the effect of nkx3.2 loss of function on trunk morphology.

Thus, nkx3.2 loss of function accounts for jaw joint ankylosis in the skulls of our mutant zebrafish. The bones forming the jaw joint develop from cartilages, and this corroborates previous findings about the regulatory role of Nkx3.2 in chondrogenesis. During development of the jaw joint, nkx3.2 is thought to specify the joint interzone by inhibiting maturation or hypertrophy of the chondrocytes (Miller et al., 2003; Smeeton et al., 2016). Similar functions have been ascribed to irx7 and irx5a in the hyoid joint (Askary et al., 2015). This is consistent with the predicted regulatory function of Nkx3.2 in vertebral development, intervertebral and interphalangeal joint formation, or somatic chondrogenesis in general (Herbrand et al., 2002; Lettice et al., 2001; Murtaugh et al., 2001), via repression of Runx2, by upregulating Col2α1 and/or through a positive feedback loop with Sox9 (Smeeton et al., 2016). These insights are based on: (a) experimental results using amniote embryos or somitic mesodermal cell cultures (Cairns et al., 2008; Kawato et al., 2012; Lengner et al., 2005; Murtaugh et al., 2001; Provot et al., 2006; Yamashita et al., 2009; Zeng et al., 2002); and (b) clinical and genetic studies of human pathologies, including osteoarthritis and spondylo-megaepiphyseal-metaphyseal dysplasia (including pseudoepiphyses) (Caron et al., 2015; Hellemans et al., 2009).

Although the fused jaw cartilages clearly result from nkx3.2 mutation, nkx3.2 loss of function, by itself, seems unlikely to yield the remaining phenotypic effects on skull form described here. Instead, the late onset and topology of changes to the skull morphology suggests that these modifications arose secondarily to jaw joint ankylosis. Other than fused jaw cartilages, nkx3.2^−/− zebrafish appear normal until metamorphosis (14–21 dpf) (Fig. 1D,G). The open gape then becomes conspicuous, which likely began with m. intermandibularis posterior abducting the elastic Meckel's cartilage toward the basihyal.

As the skull ossifies, the observed phenotype becomes increasingly prominent (Fig. 2B,D,G,J,M,N). The skull bones arise endochondrally (quadrate, anguloarticular, basihyal) or intramembranously (premaxilla, maxilla, dentary, jugal, opercular, preopercular) (Cubbage and Mabee, 1996; Schilling and Kimmel, 1997). Much of the skeletal remodeling observed in adult nkx3.2^−/− mutants occurred in the intramembranous bones, even though the known function of nkx3.2 in skeletal development is exclusively related to cartilaginous structures (Chatterjee et al., 2011; Crotwell and Mabee, 2007; Miller et al., 2003; Smeeton et al., 2016). Except for the palatoquadrate and Meckel's cartilage, the highly modified endochondral skull bones in adult nkx3.2^−/− mutants ossify spatially and temporarily well outside the known expression domain of nkx3.2 (16 hpf to pre-metamorphosis; mandibular arch, sclerotomes and their derivatives, and midline fins) (Askary et al., 2017; Barske et al., 2018; Chatterjee et al., 2011; Crotwell and Mabee, 2007; Miller et al., 2003; Nair et al., 2007; Nichols et al., 2013).

Therefore, this dramatic modification of the skull comes second to the direct result of nkx3.2 loss of function. To support this further, a genome-wide analysis suggests that nkx3.2 patterning effects in the skull are restricted to the mid-portion of the mandibular arch (Askary et al., 2017). The hypertrophied mandibular cartilage (Fig. 5I) may result from nkx3.2 loss of function in regulating local chondrogenesis (Smeeton et al., 2016). Still, further investigation is warranted for this hypertrophy, because in amphibians an ectopic cartilage formed in the mandibular arch with Nkx3.2 overexpression, not repression (Lukas and Olsson, 2018b).

Functional jaw loss resulting from nkx3.2 null mutations allowed the mutants to depart markedly in morphology from their wild-type cousins, despite the nearly identical genetic backgrounds. This departure implies that movements at the jaw joint can limit skull morphology. The specific combination of traits observed in adult nkx3.2^−/− zebrafish – for example, nostrils in interorbital position, premaxilla and maxilla abutted against antorbital wall, basihyal protrusion – is likely maladaptive when the jaws function properly. In this regard, the absence of a kinethmoid in nkx3.2^−/− mutants is a classic example. The sesamoid bone functions as a hinge in wild-type fish as the upper jaw swings forward against the skull roof (Hernandez et al., 2007). With the premaxilla and maxilla fixed in the mutants, however, the kinethmoid never formed as a distinct element (Fig. 5E). Our nkx3.2^−/− zebrafish provide empirical evidence that jaw movements are an important factor in the development – and thus morphological diversity – of jawed vertebrate skulls.

Simultaneously, the absence of a jaw joint (or jaw apparatus altogether) also limits functionally viable forms. This interpretation is bolstered by the superficial convergence in head shapes between nkx3.2^−/− zebrafish and two Paleozoic agnathan lineages (anaspids and furcacaudiform thelodonts). The nkx3.2^−/− zebrafish and these agnathans share a functional condition – the absence of a hinge joint between upper and lower lips – and transversely compressed body profile. No evidence suggests any more similarities in the otherwise wild-type-like young mutants (Fig. 1D,G) until the lower jaw skeletons begin rotating posteroventrally after 14 dpf (Fig. 2B,D). Therefore, we interpret the anaspid/thelodont-like traits in the adult nkx3.2^−/− zebrafish not as recapitulations of a conserved, genetically hard-wired ancestral development (atavism) but rather as parallel developmental responses to a shared functional constraint (convergence). This convergence implies that the vertebrate bauplan allows a limited repertoire of functionally viable morphological patterns in the skull.

A survey across vertebrates reveals many examples of jaw movements affecting bone morphology. Discrete variation in cichlid jaw morphology has an epigenetic basis in behaviorally mediated skeletal remodeling, where gaping frequencies in juveniles correlate with dimensions of the retroarticular process (Hu and Albertson, 2017). Additionally, dietary differences explain some taxonomically significant variations in the jaw morphology of multiple cichlid species (Galis, 1993; Greenwood, 1965; Liem and Osse, 1975; Meyer, 1987; Wimberger, 1992, 1991). Compared with cases in which jaw functions were perturbed, the magnitude of morphological changes in the adult nkx3.2^−/− phenotype seems consistent with a series of surgical experiments in mammalian jaw skeletons (Bayram et al., 2010; Gomes et al., 2012; Horowitz and Shapiro, 1955; Lifshitz, 1976; Miyazaki et al., 2016; Rodrigues et al., 2009; Sarnat, 1970; Sarnat and Muchnic, 1971; Toledo et al., 2014) or with the ‘bird face’ deformity observed in clinical cases of the temporomandibular (=mammalian jaw) joint ankylosis in humans (El-Sheikh et al., 1996). Collectively, these studies imply a strong link between movements of the jaw skeletons and latent developmental potentials that shape the skulls.

Jaw joint loss (nkx3.2 loss-of-function phenotype) released nkx3.2^−/− zebrafish from having to facilitate jaw movement, and exposed them to a different functional requirement: feeding without mobile jaws. The nkx3.2^−/− mutants clearly altered their feeding. The mutants ram fed through the water column, whereas wild-type zebrafish typically maintained their position with pectoral backstrokes and used suction in a much faster feeding cycle (Fig. 6). Teleost fishes generally use head and basihyal elevation to assist suction (Westneat, 2004, 2005), but the nkx3.2^−/− mutants retain these movements in a different functional context where there is no jaw motion. A fixed open gape accompanied by reduction of the snout is compatible with ram feeding and ventilating in the absence of the oral pump. The expanded branchial space facilitated by the greater skull width is also consistent with this. The massive occiput and basibranchial elements are likely linked to head elevation and basihyal retraction, respectively (Fig. 6B). At the onset of the open-mouth phenotype (2–3 weeks post-fertilization), cartilages are the main skeletal support of the jaws. The phenotypic effects become more pronounced as ossification proceeds around the deformed cartilaginous template (Fig. 2D). But all these correlated changes eventually lead to the well-constrained morphospace that nkx3.2^−/− adults occupy (Fig. 7F). These results clearly point to developmentally plastic responses that coordinated changes in feeding and skull morphology and facilitated convergence with anaspids and thelodonts.

Remarkable convergence arises via developmental plasticity under similar functional requirements – even in the absence of homology in individual skeletal elements – between taxa widely separated chronologically and phylogenetically. Bichirs (Polypterus) routinely trained under terrestrial conditions develop modified pectoral fin skeletons, where induced, adaptive morphological changes parallel the traits observed among stem tetrapods (Standen et al., 2014). West-Eberhard (2005b) reviewed a goat born with congenital paralysis of the forelimbs, which performed bipedal locomotion. This behavioral accommodation led to highly modified musculoskeletal anatomy of the axial column, pelvis and hindlimbs (Slijper, 1942a,b). Variation resulting from developmental plasticities – whether induced by environmental cues, developmental perturbation or mutation – are often non-random and adaptive (Palmer, 2012; West-Eberhard, 2003, 2005a,b). Therefore, as shown in nkx3.2^−/− zebrafish, phenotypic accommodation of a defect resulting from a single null allele can initiate a set of correlated responses to form a complex trait independently acquired by a distant lineage, even after development has laid out lineage-specific patterns (such as homologies of individual bones).

Plasticity, whether at the level of whole-animal behavior or gene transcription, remains an elusive component of skeletal development that is challenging to test experimentally. Here, nkx3.2^−/− zebrafish provide a unique experimental system. Many questions remain about these fish, including: what genetic mechanisms regulate the development of anomalous shapes of endochondral and intramembranous ossifications?; how are developmental pathways coordinated to respond to mechanical cues resulting from jaw joint dysfunction?; do mechanical interactions – especially those between developmentally independent, but functionally connected units such as the lower jaw and m. intermandibularis posterior – accurately predict morphological changes?; and finally, what is the cost of the nkx3.2^−/− phenotype to fitness, and how does selection, in general, evaluate developmental plasticity?

The non-atavistic nature of the nkx3.2^−/− phenotype potentially helps in re-evaluating evolutionarily inspired interpretations of phenocopies in developmental studies. There are no evolutionary relationships in the similarities between nkx3.2^−/− zebrafish and anaspids or thelodonts. Anaspids are a stem cyclostome lineage (Miyashita et al., 2019) (Fig. 7G), and thus a poor surrogate for an ancestral state. The lineages between anaspids and thelodonts, or those between thelodonts and the gnathostome crown, do not share a similar combination of morphological traits, but instead are dorsoventrally depressed forms (Janvier, 1996; Miyashita, 2016; Miyashita et al., 2019). Finally, head similarities between the mutants and the stem taxa do not extend beyond overall configuration. No skull elements are lost or replaced in nkx3.2^−/− zebrafish to achieve the anaspid/thelodont-like micromery (Blom et al., 2001; Janvier, 1996; Märss et al., 2007). Therefore, the adult nkx3.2^−/− phenotype clearly does not represent a reversal to an ancestral state.

Experimental phenocopies (defective phenotypes resembling outgroup conditions) resulting from gene knockdowns and knockouts are often interpreted as atavistic (ancestral reversal), but alternative interpretations are seldom tested (Smith and Schneider, 1998). Epigenetic responses can correct or accommodate defective phenotypes resulting from perturbations to development (West-Eberhard, 2003). For example, normalization of craniofacial defects occurs via thyroid hormone-independent tissue remodeling in tadpoles of Xenopus laevis treated with Sox6 morpholino, pharmacological inhibitors and ethanol (Pinet et al., 2019; Vandenberg et al., 2012). Here, our nkx3.2^−/− zebrafish deployed a set of responses (modifications to the skull form and feeding) to accommodate a loss-of-function phenotype beyond correction (jaw joint ankylosis), while phenotypically resembling distant outgroups (anaspids and thelodonts) that lacked mobile jaws. Developmental plasticity is therefore a testable alternative to atavistic reversal to explain a phenocopy. By showing that a gnathostome can survive without a jaw, the anaspid/thelodont-like nkx3.2^−/− zebrafish also offer a comparative model to make inferences about the functional morphology of these long-extinct agnathans.

We thank B. F. Eames (University of Saskatchewan) for facilitating our collaboration; M. E. Bronner (California Institute of Technology), S. J. Childs (University of Calgary), M. I. Coates, R. K. Ho, V. E. Prince and M. W. Westneat (University of Chicago) for discussion; and members of the Allison and Prince labs for maintenance of the nkx3.2^ua5011 line.