Chondroitin/dermatan sulfate glycosyltransferase genes are essential for craniofacial development

Chondroitin/dermatan sulfate (CS/DS) proteoglycans are indispensable for animal development and homeostasis but the large number of enzymes involved in their biosynthesis have made CS/DS function a challenging problem to study genetically. In our study, we generated loss-of-function alleles in zebrafish genes encoding CS/DS biosynthetic enzymes and characterized the effect on development in single and double mutants. Homozygous mutants in chsy1, csgalnact1a, csgalnat2, chpfa, ust and chst7, respectively, develop to adults. However, csgalnact1a-/- fish develop distinct craniofacial defects while the chsy1-/- skeletal phenotype is milder and the remaining mutants display no gross morphological abnormalities. These results suggest a high redundancy for the CS/DS biosynthetic enzymes and to further reduce CS/DS biosynthesis we combined mutant alleles. The craniofacial phenotype is further enhanced in csgalnact1a-/-;chsy1-/- adults and csgalnact1a-/-;csgalnact2-/- larvae. While csgalnact1a-/-;csgalnact2-/- was the most affected allele combination in our study, CS/DS is still not completely abolished. Transcriptome analysis of chsy1-/-, csgalnact1a-/- and csgalnact1a-/-;csgalnact2-/- larvae revealed that the expression had changed in a similar way in the three mutant lines but no differential expression was found in any of fifty GAG biosynthesis enzymes identified. Thus, zebrafish larvae do not increase transcription of GAG biosynthesis genes as a consequence of decreased CS/DS biosynthesis. The new zebrafish lines develop phenotypes similar to clinical characteristics of several human congenital disorders making the mutants potentially useful to study disease mechanisms and treatment.


Introduction
Heparan sulfate (HS) and chondroitin sulfate/dermatan sulfate (CS/DS) proteoglycans are heavily glycosylated proteins, crucial for animal development and homeostasis.They consist of long, unbranched, sulfated glycosaminoglycans (GAGs), covalently attached to core proteins via serine residues.GAGs are composed of repeating disaccharide units of an amino sugar and an hexuronic acid or galactose [1].The GAG chains are synthesized in the Golgi apparatus, and the mature proteoglycans are secreted into the extracellular matrix, stored in secretory granules or are incorporated into the plasma membrane.In animals, GAGs are highly abundant and produced by most cells, where a single core protein may contain more than one type of GAG chain [1].
CS/DS proteogylcans are essential components of cartilage and bone tissues with important roles in development and function of the skeleton [2,3].The pharyngeal cartilage is derived from migrating neural crest cells and produces an extracellular matrix rich in proteoglycans [4].Most of these cartilage structures undergo endochondral ossification [5,6].Mutations in the human genes encoding CS/DS biosynthesis enzymes cause a number of genetic disorders, often causing skeletal dysmorphism and growth retardation [7].
A complex biosynthetic machinery with a variety of enzymes is required to polymerize and modify CS/DS chains in vertebrates ( Fig 1).Initiation of the CS/DS chain, on the tetrasaccharide linkage region which is shared between HS and CS/DS, is carried out by Csgalnact1 and Csgalncat2, while further polymerization is performed by Chsy and Chpf enzymes (Fig 1).Then the CS/DS chain is modified by epimerization and sulfation resulting in generation of binding sites for a vast number of proteins, including cytokines and chemokines, growth factors and morphogens, fibrous proteins like collagens, signaling receptors and cell adhesion proteins [8][9][10].
Zebrafish (Danio rerio) belonging to the teleost lineage, is a well-established animal model broadly used in biomedical research because of its genetic tractability.Teleosts underwent an additional round of whole genome duplication, compared to mammals and while most of the duplication was eventually lost through "rediplodization", roughly 20% of zebrafish genes still have two copies in the genome.Zebrafish orthologues of mammalian HS and CS/DS biosynthetic enzymes have been previously identified (Fig 1) [11,12].In some cases, two gene copies of GAG biosynthetic enzymes are retained (indicated by blue bars in Fig 1).In zebrafish the CS/DS biosynthetic enzymes are spatially and temporally regulated during embryonic development, indicating that tissue specific GAG structures exist [13][14][15][16][17][18][19][20][21].
The introduction of random mutations in the zebrafish genome has for a long time been used for forward genetic screens, where the mutagens can introduce loss-of-function mutations.A number of genetic mutants with defective GAG biosynthesis have been identified using this approach.Loss-of-function mutations in ughd, uxs1, xylt1, fam20b, b4galt7, b3gat3, genes for precursor transport and synthesis of the shared linkage structure of HS and CS/DS GAGs (Fig 1, indicated by #) typically develop abnormal jaw and pharyngeal cartilage structures [22,23], see also recent review [12]), highlighting the importance for these genes in skeletal development.
Zebrafish lines with mutations in HS glycosyltransferases, which reduce HS but not CS/DS accumulation, also affect craniofacial cartilage formation and in addition pectoral fin development and axon sorting [11].No genetic knockout of CS/DS biosynthesis enzymes or core proteins has yet been reported in zebrafish.In mouse Csgalnact1 -/-and Chsy -/-animals have been reported to show skeletal phenotypes [24,25].Morpholino knockdown in zebrafish is a transient alternative to genetic knockouts, but the outcome can differ significantly [26], either because of morpholino off target effects [27] or due to genetic compensation [28].This highlights the need of genetic mutants to complement knockdown experiments [29].In this study we investigate the role of CS/DS in zebrafish development and the importance of individual biosynthesis enzymes.We hypothesized that knockout of one or several glycosyltransferase genes would decrease CD/DS production and affect zebrafish development, allowing us to study the specific role of CS/DS in zebrafish development.For this purpose, we have generated loss-of-function alleles in genes encoding for zebrafish CS/DS biosynthetic enzymes.The generated genetic mutants revealed a functional redundancy for enzymes in CS/DS biosynthesis and by combining different null-alleles, we were able to create a set of lines differing in CS/DS production and displaying varying phenotypes.

Mutagenesis of key CS/DS biosynthetic enzymes
To reduce CS/DS biosynthesis in zebrafish we generated loss-of-function alleles in a number of genes encoding enzymes with key roles in CS/DS biosynthesis.We selected a subset of CS/ DS modifying enzymes and glycosyltransferases for targeted mutation where we from previous studies expected a key function in CS/DS biosynthesis [12] and where suitable CRISPR targets could be found.We have previously developed protocols for high-throughput CRISPR/Cas9 modification of the zebrafish genome [30,31].CRISPR targets were designed in the early part of the coding region of the CS/DS glycosyltransferases csgalnact1a, csgalnact2, chsy1, chpfa and the CS/DS sulfotransferases ust, chst3a, and chst7.Using this method, we isolated 20 zebrafish alleles with frame shifts in the coding sequence (Table 1) in seven different genes related to CS/DS biosynthesis (Table 1).

Some GAG biosynthetic enzymes are not critical for zebrafish development
To find genes with key roles in CS/DS biosynthesis we screened for morphological defects in homozygous mutants in larvae at 6 days post fertilization (dpf) and in adults (Table 2).With the exception of csgalnact1a -/-(discussed below), they all displayed overall normal morphology (Table 2).From these results we conclude that defective gene function of single GAG biosynthetic enzymes typically allows for normal zebrafish development.Given the many enzymes that can both polymerize and modify the CS/DS molecule (Fig 1 ), this finding suggests widespread redundancy among CS/DS biosynthesis genes and might explain why no loss-of-function alleles in genes encoding for CS/DS biosynthesis enzymes have ever been identified in forward genetic screens, particularly since most screens focused on larval phenotypes.csgalnact1a -/-adults develop craniofacial malformations HS and CS/DS are synthesized on a common link structure and the committing step of CS/DS synthesis is the addition of a GalNAc residue by Csgalnact enzymes (Fig 1).Adult csgal-nact1a -/-zebrafish developed a fully penetrant head phenotype, compared to control animals, showing malformations of the head skeleton (Fig 2A -2F).To study head bone structures in detail we used micro-CT scanning to generate high resolution images of bone structures (Fig 2G -2L).Micro-CT scanning followed by 3D image reconstruction gave information on the relative position of adult cranial skeletal elements and made it possible to determine the size and shape of individual structures (S1 Movie and Fig 2G , 2I and 2K).The head of csgalnact1a mutants (Fig 2H , 2J and 2L) was wider and shorter than that of control adults, compressed along the anterior/posterior axis.The lower jaw, in particular the mandible element, was PLOS GENETICS Table 1.Genomic and amino acid sequence for identified loss-of-function alleles.CRISPR target sequences are underlined and the PAM site is colored blue.Aberrant protein sequence is colored in red.Stop codon is indicated by a star ( � ).§ (position of amino acid sequence interruption)/(total number of amino acids in protein).

Gene/Allele
Position of mutation § Position of stop codon relative to conserved functional motif upstream of the conserved B4GT domain including the conserved W(G/V)GED motif at 453-457 aa specific for the β4-glycosyltransferase family.[20,32] 93/801aa upstream of the conserved B3GT domain including the FMRADD motif at 164-172 aa specific for the β3-glycosyltransferase family.[20,32] upstream of the conserved B4GT domain including the conserved W(G/V)GED motif at 493-499 aa specific for the β4-glycosyltransferase family.[20,32] 6_3; -4bp upstream of all conserved motifs in the CHPF family after the transmembrane region, very early in the protein.[20,32]  As a comparison, Csgalnact1 -/-mice showed impaired intramembranous ossification of the skull resulting in a shorter face, higher and broader calvaria, reminiscent of the zebrafish csgalnact1a -/-phenotype [24].In contrast to the zebrafish phenotype, the lower jaw in mice were largely unaffected.Human patients with mutations in CSGALNACT1 have been described in several studies with mild skeletal dysplasia [35][36][37].
Our zebrafish csgalnact1a -/-line may thus serve as a model to study this rare human syndrome (OMIM 618870).

Gene Allele Major morphological abnormalities 6 dpf adult
No No that chsy1 is critical for viability.In contrast, mouse Chsy mutants are slightly smaller than controls and viable albeit with joint patterning defects [25].In several patients a mutation in CHSY1 was found to be the cause for Temtamy Preaxial Brachydactyly syndrome (OMIM 605282) [39][40][41][42].These patients show limb malformations, have a short stature and hearing loss.
In contrast to the strong head phenotype in csgalnact1a -/-adult zebrafish (Fig 3A, 3B and 3F), we found that chsy1 -/-zebrafish did not develop severe morphological defects (Fig 3D ), colors.Compared to control adults (A,C,E,G,I,K), csgalnact1a -/-adults (B,D,F,H,J,L) present a shortened jaw (arrow B) resulting in an open mouth phenotype (arrow D).In dorsal view the protruding eyes of csgalnact1a -/-adults are apparent (F).https://doi.org/10.1371/journal.pgen.1010067.g002Fig 3 .Severe malformation in the adult head skeleton.Alterations in the morphology of the ceratohyal and urohyal were magnified in all four genotypes (for spatial position of elements in the zebrafish head, see S1 movie).Adult csgalnact1a -/-; chsy1 -/-zebrafish displayed gross malformations and reduced body size compared to control and csgalnact1a -/-adults (A).3D reconstructions of all genotypes showed that no bone structures were missing (B,D,F,H and S1-S4 Figs).However, shorter and misshaped elements were common in csgalnact1a -/-(F) and csgalnact1a -/-; chsy1 -/-fish (H), while the chsy1 -/-skeletal phenotype (D) was milder, compared to the control (B).The differences in morphological integrity of skeletal elements was evident from magnifications, for example the boundary between the anterior and the posterior part (arrow) of the ceratohyal and posterior edge ( � ) and of the urohyal (C,E,G,I and S5-12).https://doi.org/10.1371/journal.pgen.1010067.g003

PLOS GENETICS
more similar to the corresponding mouse knockouts than to the previously published morpholino studies [38,39].To test if depletion of multiple CS/DS glycosyltransferases would enhance the phenotype compared to the single homozygous mutants, we then generated csgal-nact1a;chsy1 and csgalnact1a;csgalnact2 double mutants.The csgalnact1a;csgalnact2 double mutants were not adult viable, further discussed below, while csgalnact1a -/-;chsy1 -/-adults resulted in growth retardation and reduced body size compared to both control and csgal-nact1a -/-fish (Fig 3A and 3H).In addition we observed a disturbed swimming behavior and reduced buoyancy in csgalnact1a;chsy1 double mutants.csgalnact1a -/-;chsy1 -/-adults developed a more severe skeletal phenotype compared to the csgalnact1a -/-adults with a smaller head including the anterior parts of the skull (Fig 3H and 3F).Smooth, finished bone surface morphology in the ceratohyal as well as the urohyal and well-defined contours were characteristics of control adults (Fig 3C The posterior edge of the urohyal ( � ) was severely affected, which probably disturbs muscular attachment and therefore results in the constantly opened mouth.Notably, despite general gross malformations in the mutant lines, all cranial skeletal elements were present and morphologically distinct.This suggests that developmental patterning was generally intact but bone growth or formation was impacted.We conclude that the lack of both Csgalnact1a and Chsy1 has a more severe effect on adult mutant skeletal phenotypes than lack of either enzyme alone (Fig 3).

csgalnact1a -/-and chsy1 -/-develop malformations in the craniofacial cartilage elements already at larval stages
The early skeletal structures in the zebrafish larvae are mainly cartilage.In order to visualize the skeleton already at developmental stages we performed alcian blue staining on csgal-nact1a -/-, chsy1 -/-and their respective control larvae.Data from optical tomography where then used to generate maximum projections of average patterns resulting in an average 3D representation from 8-10 individuals of each genotype (Fig 4A -4F).These representations show that pharyngeal cartilage structures of mutant larvae at 9 dpf develop all major elements but that in particular csgalnact1a -/-develop a malformed morphology (Fig 4C and 4D).By combining the 3D average structures of mutants with the controls, a rigid alignment showed the altered morphology more clearly as an overlay of mutant (magenta) and control (green) larvae was generated (Fig 4G -4J).The chsy1 -/-larvae developed a milder phenotype (Fig 4I -4J) compared to csgalnact1a -/-larvae (Fig 4G -4H).The pharyngeal skeleton in chsy1 -/-larvae was smaller compared to control larvae and even further reduced in csgalnact1a -/-larvae (Fig 4K).The mild skeletal phenotype of chsy1 -/-larvae was further analyzed in 40 dpf juveniles.The total body length and the head of chsy1 -/-juveniles were reduced compared to control fish, reminiscent of the facial dysmorphism and short stature in human patients [39][40][41][42].

PLOS GENETICS
stages csgalnact1a and csgalnact2 showed strong expression in the craniofacial cartilage [20].We investigated larvae in the transgenic Tg(col2a1a:mEGFP) background where the membranes of chondrocytes are labeled.At 6 dpf all craniofacial cartilage elements were formed in all larvae (Fig 5A -5H) and intercalation of chondrocytes occurred (Fig 5C , 5F and 5I).However, several elements were shorter and thicker in the csgalnact1a -/-larvae (Fig 5D and 5E) compared to control larvae (Fig 5A and 5B).The ceratohyal was 15% shorter in csgalnact1a -/- larvae compared to control larvae (p<0,005).The angle between the ceratohyal elements were widened in csgalnact1a -/-larvae compared to controls.The craniofacial malformations were enhanced in csgalnact1a -/-;csgalnact2 -/-larvae (Fig 5G and 5H).csgalnact1a -/-;csgalnact2 -/-larvae also died before 10 dpf.Whether some CS/DS synthesis still occurred in these double mutants, and to what extent HS could be produced on core proteins normally modified by CS/ DS, as suggested by previous experiments [15], remains to be investigated.

PLOS GENETICS
and CS/DS reduction (Fig 1).This similarity may indicate a general property of CS/DS biosynthesis to shift from 6S to 4S when CS/DS polymerization is reduced.Alternatively, the reduction of CS/DS polymerization may be stronger in tissues which produce large amounts of CS/ DS with higher 6S content-thereby making the overall CS/DS composition in mutant larvae resemble the 4S that dominated CS/DS of the earlier embryo [21].The effect on CS/DS biosynthesis in 9 dpf chsy1 -/-was similar to the effect in csgalnact1a -/-larvae (Fig 7B ), with the notable exception that the proportion of 6S disaccharides in CS/DS was not changed compared to the control.The difference between effects in chsy1 -/-and csgalnact1a -/-larvae may reflect different roles in the CS/DS biosynthesis or may be a consequence of differential expression of Chsy1 and Csgalnact1a during development [21].An analysis of CS/DS in 40 dpf chsy1 -/-juveniles did not reveal statistically significant differences compared to control juveniles, suggesting that differences in CS/DS accumulation does not increase with age (Fig 7C).One possibility may be that older animals decrease CS/DS degradation in the extracellular matrix to compensate for reduced CS/DS biosynthesis.In addition, low amounts of non-sulfated, 4S, 6S and 4S2S disaccharides were detected in 6 dpf csgalnact1a -/-;csgalnact2 -/-and csgalnact1a -/-;chsy1 -/-larvae respectively.We conclude that none of the mutant lines presented in this study are completely devoid of CS/DS which emphasizes the redundant function of CS/DS biosynthesis enzymes.This confirms previous cell-based studies showing that while removal of one or two CS/DS glycosyltransferases reduce biosynthesis, the remaining enzymes can still to some extent polymerize CS/DS [44].

Mutations in different CS/DS glycosyltransferases result in similar effects on transcriptome composition
We next investigated how a reduction in CS/DS biosynthesis affected the transcriptome in zebrafish larvae.We chose 6 dpf old larvae depleted of Chsy1, larvae depleted of Csgalnact1a and larvae depleted of Csgalnact1a and Csgalnact2 to include phenotypes ranging from mild to severe (Figs 4 and 5) for the transcriptome analysis.
In this study we defined differentially expression of a gene as a two-fold decrease or increase in gene expression (p<0.05).With this definition we detected 98 differentially expressed genes

PLOS GENETICS
in larvae depleted of Chsy1 and 72 genes in Csgalnact1a depleted larvae.Approximately half of the genes were the same in the two groups (Fig 8A).Although both Chsy1 or Csgalnact1a are involved in CS/DS synthesis, we conclude that effects on the transcriptome of the larva in the respective mutants were not identical.The differences might be due to differential expression in the developing embryo since csgalnact1a expression is restricted mainly to developing cartilage structures while chsy1 is more broadly expressed [38] and this difference is also manifested in the differences in skeletal phenotype (Figs 3 and 4).In larvae depleted of both Csgalnact1a and Csgalnact2, 173 genes were differently expressed compared to the control which was in line with the stronger and lethal phenotype (Fig 5).18 genes were differently expressed in all three groups.In total 263 genes were differently expressed in at least one of the three mutant lines and the similarity is illustrated by heatmap analysis (Fig 8B).We grouped and colored these genes according to increase or decrease in transcription which revealed an overall striking similarity in the trend of change in expression between the mutant lines (i.e. if the expression of a gene is significantly changed in one mutant line, it is almost always changed in the same direction in the two other mutant lines) (Fig 8B).We conclude that the changes in transcriptome composition was similar but not identical in the three investigated cohorts of larvae.

Mutations in csgalnact1a, csgalnact2 or chsy1 genes do not activate compensatory transcription of CS/DS glycosyltransferase genes
The fully viable and fertile adult chsy1 -/-phenotype was unexpected given the severe embryonic phenotype reported for chsy1 morpholino knockdowns and distinct skeletal phenotype in patients with mutations the human ortholog CHSY1 [38,39].The common observation of stronger morpholino knockdown phenotypes as compared to genetic knockout phenotypes has been discussed in recent years.One explanation may be that many reported morpholino phenotypes are at least in part a result of off-target effects [26].Another possibility may be genetic compensation in mutants triggered by mutant mRNA degradation (i.e.nonsense-mediated decay) which induces compensatory transcription of genes with sequence similarities (i.e.paralogues) [28,45].Candidates for genetic compensation in chsy1 -/-mutants would be other genes in the GAG biosynthesis machinery with high sequence similarity (Fig 1).However, all GAG glycosyltransferase genes identified on the microarray (chsy1, chpf2, chpfa, chsy3, csgalnact1a and csgalnact2) were expressed similarly in chsy1 -/-and control larvae, indicating that the induced frameshift mutations do not increase the degradation of glycosyltransferase mRNA (S2 Table ).The same is true for csgalnact1a -/-and csgalnact1a -/-;csgalnact2 -/-mutants (S2 Table ).We further note that the reduced CS/DS biosynthesis in chsy1 -/-, csgalnact1a -/-and csgal-nact1a -/-;csgalnact2 -/-larvae does not affect transcription of any GAG sulfotransferases, epimerases, sugar transporters and degrading enzymes.Thus, in the three mutant lines investigated, we found no evidence for genetic compensation due to mRNA decay.Our data further indicates that no mechanism exists to increase transcription of GAG biosynthesis genes as a consequence of missing CS/DS glycosyltransferases or decreased CS/DS accumulation.

Aggrecan and Grhl3 are dysregulated in mutants
CS/DS proteoglycans are modified by CS/DS GAG chains which are critical for function.Aggrecan is a major structural component of the extracellular matrix and heavily modified by CS.Our transcriptome data from the chsy1 -/-, csgalnact1a -/-and csgalnact1a -/-;csgalnact2 -/- mutants revealed a small (�2-fold) but significant upregulation of both acana and acanb (S2 Table ), potentially representing a mechanism to compensate reduced CS/DS in the extracellular matrix.Our dataset revealed only small effects on the expression of 12 additional CS/DS proteoglycans, indicating that compensatory transcription of proteoglycan core proteins is not PLOS GENETICS The area proportional venn diagram (A) shows differentially expressed genes in csgalnact1a -/-� , csgalnact1a -/-; csgalnact2 -/-and chsy1 -/-larvae compared to control larvae.The heat map (B) shows all genes with a two-fold increase (red) or decrease (blue) in the csgalnact1a -/-, csgalnact1a -/-;csgalnact2 -/-or chsy1 -/-larvae.� 50% of the csgalnact1a -/- individuals lack one functional allele of chsy1.https://doi.org/10.1371/journal.pgen.1010067.g008

PLOS GENETICS
a general effect to meet reduced CS/DS polymerization.Another gene that stood out in our dataset was grainyhead-like transcription factor 3 which was 2-3 fold downregulated in chsy1 -/-, csgalnact1a -/-and csgalnact1a -/-;csgalnact2 -/-larvae.This gene encodes a transcription factor which in humans has been linked to cleft lip and palate syndromes (OMIM 606713, [46]).In zebrafish, grhl3 is expressed throughout the pharyngeal cartilage and morpholino knockdown of grhl3 results in underdeveloped lower jaws [47].Our results suggest that CS/DS is involved in activating grhl3 expression but the mechanism remains to be investigated.One possibility is a reduced activity of upstream grhl3 activators such as irf6 and p63, which are slightly less expressed in CS/DS depleted larvae (S2 Table ).Expression of edn1, hand2 and dlx3b, genes in the downstream signaling pathway of grhl3, were not altered in the mutants analyzed (S2 Table ).

Conclusions
Despite having many important functions during animal development, mutants for CS/DS biosynthesis enzymes have never been isolated in forward genetic phenotypic screens in zebrafish.In this study, we have applied a CRISPR/Cas9-based reverse genetics approach to generate mutant alleles in a number of CS/DS biosynthesis genes.Our data show that elimination of single enzymes rarely causes major phenotypic changes, indicating a high degree of functional redundancy of CS/DS biosynthesis enzymes, which likely explains why mutant alleles in these genes have not been picked up in forward genetics phenotypic screens.To study the combinatorial effects of depleted CS/DS biosynthesis on zebrafish development we then combined multiple loss-of-function alleles of CS/DS enzymes which resulted in variable phenotypes ranging from subtle skeletal deformations in viable and fertile adults to severe morphological defects and lethality.Mutant fish that survived into adulthood predominantly developed malformations in the craniofacial skeleton.Transcriptome analysis of mutant zebrafish larvae with craniofacial phenotypes ranging from subtle to severe, reveal that expression of GAG biosynthesis genes is not affected which suggests that depleted CS/DS does not infer compensatory expression of other GAG biosynthesis genes.The analysis also showed that genes with an altered expression in one mutant typically were altered also in the other mutants, indicating that these genes are important for similar functions.We conclude that mutant alleles in CS/DS glycosyltransferase genes, alone and in combination, result in distinct effects on craniofacial skeleton and transcriptome composition and that the mutant lines will provide valuable tools for modeling diseases with a CS/DS component.

Ethics statement
This study was approved by Animal Care and Use Committee of the NHGRI with the protocol number G-01-3 and by Uppsala Djurfo ¨rso ¨ksetiska na ¨mnd, Uppsala, Sweden (Permit number C161/14 and 5.8.18-11830/2019).

Target design
Using the CRISPR targets TrackHub in the USCS Genome Browser, two target sequences for each gene of interest, all 20 bp long, flanked by NGG, the protospacer adjacent motif (PAM), needed for target site recognition and cleavage by Cas9, were selected.Preferably, targets starting with GG, crucial for the T7 polymerase action, were preferred.In cases where suitable sgRNA recognition motifs (5'GG-N18-NGG3´) were not found, one or two Gs were incorporated at the beginning of the sgRNA, essentially as previously described [30].We typically targeted sequences within the first or second transcribed exon and avoided target sites overlapping with regions of known DNA sequence variability [48].Sequences of all targets are presented in Table 2.

Preparation of sgRNAs and Cas9 RNA
Preparation of sgRNAs was carried out as previously described [49].In short, annealing of a fragment containing the T7 promoter, the target specific sequence and a DNA stretch overlapping with the guide core sequence, with a second fragment containing the guide core sequence, was performed.The product was then used as a template for RNA in vitro transcription (HiScribe T7 High Yield RNA Synthesis Kit, NEB) and the generated RNA was purified prior to injection.To prepare Cas9 mRNA, pT3Ts-nCas9 plasmid (Addgene 46757, kindly provided by Dr. Wenbio Chen) was digested with Xba1 (NEB), purified and used for T3 driven in vitro transcription according to the manual provided by the manufacturer (mMESSAGE m MACHINE T3 Kit, Life Technologies).Integrity of the mRNA was confirmed on a denaturating gel and concentration was measured by Nanodrop.

Injections
Fertilized zebrafish (TAB5) eggs were obtained in natural crosses, injected at the one-cell stage with 150ng/μl of Cas9 mRNA and 25ng/μl sgRNA per target in RNAse free H 2 O, as previously described [30].

Genotyping by fluorescent PCR and sequencing
Injected founder fish were raised and outcrossed with wild-type zebrafish (TAB-5).Adult F1 fish were fin-clipped and PCR products were analyzed by fragment separation using capillary electrophoresis (or fragment length analysis), as previously described [51].DNA extraction from either fin clip of adults or whole embryos was performed by dissolving tissue in 30μl 50mM NaOH for 20 min in 95˚C, adding 60μl 50 mM Tris-HCl, and diluting 1:100.PCR (all primers listed in S1 Table ) was performed, adding a third M13 forward primer, fluorescently labeled with 6-FAM.The length of the denaturated PCR products was measured on a Genetic Analyzer 3130xl using POP-7 polymer.The results were analyzed by Gene Mapper (Life Technologies) or Peak Scanner Software (Life Technologies), as previously described [31].The length of the fragment/fragments was compared to the WT product length, predicted by in silico PCR using the USCS Genome browser.Heterozygous fish with two fragments, where deletions or insertions would predict a frame shift in the protein sequence (not a multiple of 3), were selected.These PCR products were cleaned up with ExoSAP-IT (Affymetrix), which degrades residual single-stranded primers and hydrolyzes remaining dNTPs.Due to addition of the M13 forward primer to the PCR reaction, M13 could now be used for sequencing all products.Sequences obtained for the heterozygous fish were aligned to the reference sequences with help of the web-based tool Poly Peak Parser [52].Heterozygous carriers were incrossed after sequence confirmation.Their embryos were observed during early development, and raised.

Imaging-lightsheet microscopy
Larvae were imaged by lightsheet microscopy in a Zeiss Z.1 microscope at 6 dpf.Tg(col2a1a: mEGFP) positive larvae were selected, anesthetized in 0,3% Tricaine solution and mounted in low melting agarose (1%) in FEB tubes.Z-stacks from different angles were taken and maximum intensity projections of lateral and ventral views were exported.

Alcian blue
Larvae fixed in fresh 4% paraformaldehyde at 4˚C overnight, were permeabilized using increasing concentrations of methanol and stored at -20˚C.After washing (PBST), bleaching (30% hydrogen peroxide, 2 hours), and washing again (PBST) larvae were transferred into an Alcian blue solution (1% concentrated hydrochloric acid, 70% ethanol, 0.1% Alcian blue).Overnight incubation at room temperature was followed by rinsing with acidic ethanol (5% concentrated hydrochloric acid, 70% ethanol).Rehydration was performed in acidic ethanol of decreasing concentrations and finally samples were cleared in glycerol in PBST and imaged.

Optical Projection Tomography (OPT)
For imaging 9 dpf fixed and alcian blue stained larvae were transferred in 99% Glycerol.OPT data was generated using the zOPT system described in [55] and generation of the average patterns are done using a similar method as described in [56].Each group of 8-10 larvae was aligned to an average pattern using an Iterative Shape Averaging (ISA) algorithm [57].All samples withing a group is first coarsely aligned to each other using a rigid alignment.From all the aligned images an average image is created.This average is then iteratively improved using an affine transformation and in the last iteration step a non-linear transformation.All average images are then positioned in the same view to make it easier to compare the group patterns.

Gene expression analysis
Gene expression analysis was performed on samples of RNA isolated from 10-20 6 dpf larvae from clutches generated by separate crossings of adult zebrafish.Incross of chsy1 -/-produced chsy1 deficient offspring.Offspring from an incross of csgalnact1a +/-;csgalnact2 -/-) produced offspring where the 25% of individuals deficient in both csgalnact1a and csgalnact2 developed a distinct phenotype (Fig 4E and 4F) and were isolated for analysis.Outcross of csgalnact1a -/-; chsy1 +/-with csgalnact1a -/-;chsy1 +/+ produced offspring deficient in csgalnact1a (50% of offspring also lack one copy of chsy1).Three RNA samples of each group were isolated, in total 12 independent samples.10-20 larvae were placed in 0,7 ml QIAzol (Qiagen), and total RNA was purified with the miRNeasy Mini kit (Qiagen) following the manufacturer's instructions.RNA quality and quantity were ensured using Bioanalyzer (Agilent Technologies, Inc, Santa Clara, CA) and NanoDrop (Thermo Scientific, Wilmington, DE), respectively.Microarray analysis was performed at the NHGRI Microarray Core following the recommended Affymetrix protocol.Fragmented and labeled cDNA was hybridized onto GeneChip Human Gene 1.0 ST Arrays (Affymetrix, Santa Clara, CA).Staining of biotinylated cDNA and scanning of arrays were performed according to the manufacturer's recommendations.The heatmap was created in R (version 3.5.1)using the heatmap.2command (gplots package, version 3.0.1),with standard parameters.The dendrogram was computed using the complete linkage method to find similar clusters based on Euclidian distance.

Fig 1 .
Fig 1. Heparan sulfate (HS) and chondroitin/dermatan sulfate (CS/DS) biosynthesis in zebrafish.HS and CS/DS chains are attached to serine (Ser) residues of the core protein.The first four monosaccharides form the linkage region of both HS and CS/DS.Extl3 initiates HS polymerization, while Csgalnact1a and Csgalnact2 perform this function in CS/DS polymerization.Elongation of HS is carried out by Ext enzymes, while Chsy and Chpf enzymes polymerize the CS/DS chain.The first modification of HS is carried out by Ndst enzymes, replacing an N-acetyl groups of GlcNAc residues with an N-sulfate group.Glcea and Glceb epimerize GlcA into IdoA in HS, while Dse, Dsela and Dselb are responsible for this modification in DS.Hs2st enzymes add a sulfate groups to the IdoA C-2 position of HS and Hs6st and Hs3st enzymes add sulfate groups to the GlcNAc or GlcNS residues.Ust adds a sulfate group to the hexuronic acid C-2 position of CS/DS while Chst11, Chst12a, Chst12b, Chst13 and Chst14 are GalNAc 4-O-sulfotransferases while Chst3a, Chst3b, Chst7 and Chst15 are GalNAc 6-O-sulfostransferases.Blue bars indicate two or three zebrafish genes orthologous to a single mammalian gene.Ndst3 in zebrafish is a single gene orthologous to two mammalian genes (NDST3 and NDST4).# Indicates previously published mutant zebrafish alleles while � indicates the targeted mutations reported in this study.Xyl: xylose, Gal: galactose, GlcNAc: N-acetylglucosamine, IdoA: iduronic acid, GlcA: glucuronic acid, GalNAc: N-acetylgalactosamine.https://doi.org/10.1371/journal.pgen.1010067.g001

-Fig 2 .
Fig 2. Csgalnact1a -/-adults display obvious craniofacial malformations.Bright field images of zebrafish heads (A-F) and corresponding 3D reconstructions of microCT scans (G-H).Different skeletal elements are highlighted with ).In contrast, bone elements in mutant zebrafish were differently shaped and distorted and display irregular contours indicating a less dense bone volume (Fig 3E, 3G and 3I).The distinct boundary between the anterior and the posterior part of the ceratohyal (arrows) seen in controls (Fig 3C) was uneven in both chsy1 (Fig 3E) and csgalnact1a (Fig 3G) mutants and almost absent in csgalnact1a;chsy1 double mutants (Fig 3I).The midline segments, such as the urohyal, were compressed and showed an asymmetry in the midline.

Fig 4 .
Fig 4. Malformations in the craniofacial cartilage structures.Maximum projections of average patterns generated from control, csgalnact1a -/-and chsy1 -/-alcian blue stained larvae at 9 dpf show the pharyngeal cartilage structures in a ventral view (A-F).Maximum projections for each mutant (magenta) aligned to the control (green) are displayed using a rigid transformation (G-J).Measurements of the length and width of the head skeleton of 9 dpf larvae were performed on maximum projection images as shown in the image to the right (K) and plotted as a factor of control larvae (n = 8 for all genotype groups) (K).csgalnact1a -/-and chsy1 -/-larval head skeleton is significantly smaller compared to control (K).Measurements of the standard body length (1) and different other measurements of the head (2-5) are indicated on images of 40 dpf old juvenile fish (L).chsy1 -/-juveniles are significantly shorter and have a smaller head compared to control larvae (chsy1 -/-n = 17, control n = 27) (L).Statistical significance is indicated by � for p-values <0.05 and �� for p-values <0.005.https://doi.org/10.1371/journal.pgen.1010067.g004