An FDA-Approved Drug Screen for Compounds Influencing Craniofacial Skeletal Development and Craniosynostosis

Neural crest stem/progenitor cells (NCSCs) populate a variety of tissues, and their dysregulation is implicated in several human diseases including craniosynostosis and neuroblastoma. We hypothesised that small molecules that inhibit NCSC induction or differentiation may represent potential therapeutically relevant drugs in these disorders. We screened 640 FDA-approved compounds currently in clinical use for other conditions to identify those which disrupt development of NCSC-derived skeletal elements that form the zebrafish jaw. In the primary screen, we used heterozygous transgenic sox10:gfp zebrafish to directly visualise NCSC-derived jaw cartilage. We noted partial toxicity of this transgene in relation to jaw patterning, suggesting that our primary screen was sensitised for NCSC defects, and we confirmed 10 novel, 4 previously reported, and 2 functional analogue drug hits in wild-type embryos. Of these drugs, 9/14 and 7/14, respectively, are known to target pathways implicated in osteoarthritis pathogenesis or to cause reduced bone mineral density/increased fracture risk as side effects in patients treated for other conditions, suggesting that our screen enriched for pathways targeting skeletal tissue homeostasis. We selected one drug that inhibited NCSC induction and one drug that inhibits bone mineralisation for further detailed analyses which reflect our initial hypotheses. These drugs were leflunomide and cyclosporin A, respectively, and their functional analogues, teriflunomide and FK506 (tacrolimus). We identified their critical developmental windows of activity, showing that the severity of defects observed related to the timing, duration, and dose of treatment. While leflunomide has previously been shown to inhibit NCSC induction, we demonstrate additional later roles in cartilage remodelling. Both drugs altered expression of extracellular matrix metalloproteinases. As proof-of-concept, we also tested drug treatment of disease-relevant mammalian cells. While leflunomide treatment inhibited the viability of several human NCSC-derived neuroblastoma cell lines coincident with altered expression of genes involved in ribosome biogenesis and transcription, FK506 enhanced murine calvarial osteoblast differentiation and prevented fusion of the coronal suture in calvarial explants taken from Crouzon syndrome mice.

Craniosynostosis, premature fusion of sutures within the skull at birth, is a relatively common and serious condition affecting approximately 1 in 2,100 children [Hehr and Muenke, 1999;Wilkie et al., 2010]. A major focus of research into craniosynostosis is to identify drug treatments to improve outcomes following corrective surgery for this condition. The calvarium (or skull vault) forms through the process of intramembranous ossification [Twigg and Wilkie, 2015]. The cranial sutures are the major site of new bone formation within the skull, where osteoblasts form bone directly within the mesenchyme without the need for a cartilage anlagen. In simple terms, the formation of new bone and maintenance of suture patency is driven by the balance between osteoblast proliferation and differentiation. As such, osteoblasts have long been considered an obvious target for chemical therapies for craniosynostosis. However, cultured chick and rodent calvarial mesenchyme expresses chondrocytespecific collagen isoforms and can differentiate into both osteoblasts and chondrocytes, and small islands of cartilage can form within the cranial sutures [Aberg et al., 2005]. Therefore, calvarial mesenchyme has been proposed to have dual osteoblast and chondrogenic identity. The finding of mutations in interleukin genes in syndromes featuring immunodeficiency, failed molar tooth eruption, and craniosynostosis also suggests the involvement of bone-resorbing osteoclasts in maintenance of suture patency [Nieminen et al., 2011;Keupp et al., 2013;Schwerd et al., 2017]. Therefore, several key bone cell types are potentially relevant targets for treatment of craniosynostosis.
Neural crest stem/progenitor cells (NCSCs) are a transient population of embryonic multipotent stem cells that are induced at the lateral margins of the neural plate and migrate extensively to populate a variety of tissues. Several lines of evidence suggest that NCSC patterning defects cause craniosynostosis. Within the skull, NCSCs have been shown to make a significant contribution to the frontal bones of the skull and the sagittal suture, but they make no contribution whatsoever to the parietal bone or coronal suture [Jiang et al., 2002]. Twist mutant mice exhibit coronal craniosynostosis with mislocalisation of NCSCs within the coronal sutures and parietal bone [Merrill et al., 2006]. Knockout of the Msx2 gene, which drives osteoblast differentiation within the skull, rescued these defects, leading to proposal of a model whereby ectopic localisation of NCSCs to the coronal suture and parietal bone leads to enhanced osteoblast differentiation. Ephrin signalling is known to regulate NCSC migration, and thus heterozygosity for EFNB1 mutations in craniofrontonasal syndrome have been suggested to cause craniosynostosis secondary to cellular interference and NCSC migration defects in female foetuses (it should be noted, however, that EFNB1 has also been shown to regulate osteoblast differentiation directly) [Twigg et al., 2004[Twigg et al., , 2006[Twigg et al., , 2013Babbs et al., 2011;Cheng et al., 2013]. Engrailed 1 ( En1 ) has also been identified as a necessary regulator of cell movement and neural crest/mesoderm lineage boundary positioning during coronal suture formation [Deckelbaum et al., 2012]. Given these findings, we hypothesised that drugs which inhibit NCSC identities or their subsequent differentiation/remodelling could be therapeutically relevant in craniosynostosis.
The zebrafish has emerged as a powerful model organism for in vivo drug screening [Wiley et al., 2017]. As a vertebrate model, it presents several key advantages including large clutch size and fecundity, external embryonic development, visual transparency, and genetic tractability (although recent results from reverse genetics suggest that the zebrafish has become redundant as a genetic model owing to compensatory mechanisms and gene redundancy). Oxygen also diffuses freely into zebrafish embryos, which is a key advantage, allowing embryos to survive drug treatments that can frequently disrupt cardiovascular system formation. While zebrafish embryos develop rapidly, their subsequent growth and maturation is relatively much slower. In particular, the membranous bones of the skull vault do not ossify over the brain until the second month of life, thereby precluding drug screens based on calvarial development in this organism [Mork and Crump, 2015].
An important consideration relates to drug screen design, which generally falls into 1 of 2 camps. One option is a "modifier" screen, whereby a faithful (typically genetic) model of disease is treated with drugs in an attempt to ameliorate a disease phenotype. A second option that we will refer to as a "perturbation" screen involves the testing of drugs for their ability to generate a phenotype in otherwise wild-type/normal animals. This approach requires careful selection of a cell type that is relevant to a disease and analysis of a specific marker of these cells in order to identify drugs that are therapeutic. Perturbation Seda et al. Mol Syndromol 2019;10:98-114 DOI: 10.1159/000491567 100 screens can also allow investigation of drug teratogenicity. For example, many drugs are known to disrupt craniofacial/skeletal development and tissue homeostasis as side effects, e.g., association between valproate ingestion during pregnancy and craniosynostosis [Twigg and Wilkie, 2015]. Clearly, perturbation screens would require subsequent validation/follow-up in a (mammalian) disease-relevant model. Given that perturbation screens are disruptive in their design, they have the advantage of being less sensitive to genetic threshold effects and buffering mechanisms within a network than modifier screens, and may therefore be more transferrable across species.
While perturbation screens may seem counterintuitive in terms of therapy, this approach has proved to be highly productive, whereas the published evidence has demonstrated the modifier approach to be far less successful. Approximately 30 unbiased in vivo drug screens using zebrafish have been reported to date to our knowledge, and all but 4 of these were based on the perturbation design [Peterson et al., 2004;Stern et al., 2005;North et al., 2007;Kitambi et al., 2009;Ishizaki et al., 2010;Paik et al., 2010;Rihel et al., 2010;Ni et al., 2011;Peal et al., 2011;Rovira et al., 2011;Namdaran et al., 2012;Ridges et al., 2012;Gut et al., 2013;Le et al., 2013;Kong et al., 2014;Mork and Crump, 2015]. This suggests a reporting bias such that most of the many modifier screens that we know to have been conducted in the field have not yielded any hits. Of 4 published modifier screens, an average confirmed drug hit rate of 0.086% (range 0.17-0.06%) was reported [Peterson et al., 2004;Stern et al., 2005;Paik et al., 2010;Peal et al., 2011], whereas for perturbation screens that reported their results in full, an average confirmed hit rate of 2.50% (range 8.2-0.08%) was achieved [North et al., 2007;Kitambi et al., 2009;Ishizaki et al., 2010;Rihel et al., 2010;Ni et al., 2011;Rovira et al., 2011;Namdaran et al., 2012;Ridges et al., 2012;Gut et al., 2013;Le et al., 2013;Kong et al., 2014;Mork and Crump, 2015]. Thus a typical screen of 1,000 small molecules, which is achievable in most labs using nonautomated zebrafish screens, is expected to produce 25 hits in a perturbation screen but none in a modifier screen. Clearly, there is also likely to be a significant rate of attrition in transferring suitable drug hits to mammalian systems for treatment of disease, and so a number of hits are likely to be required for a realistic chance of translation.
The zebrafish viscerocranium is derived from NCSCs that contribute to the 7 pharyngeal arch derivatives, including the mandible/Meckel cartilages and the pterygoid/ethmoid plate complex, which constitute the most anterior aspects of the lower and upper jaw, respectively [Schilling and Kimmel, 1997;Mork and Crump, 2015]. The lower jaw consists of 7 paired and bilaterally symmetric cartilage elements. In anteroposterior sequence, these include Meckel cartilage, the ceratohyoid cartilage and arches 3-7 which consist of ceratobranchials attached to hypobranchials. These elements are arranged in a stereotypical pattern and are composed of chondrocytes that first emerge during the third day post-fertilisation (dpf). These cells are derived from cranial NCSCs that are first induced in the neural plate border at the earliest somite stages and subsequently migrate through the branchial arches. Failed NCSC induction prevents formation of this cartilage, and many zebrafish models of abnormal patterning/migration of NCSCs have been shown to exhibit quantitative increases in the angle between the paired cartilage elements and particularly the ceratohyoid cartilages. Once formed, these cartilage elements are remodelled through convergent extension cell movements of chondrocytes, enzymatic regulation of extracellular matrix glycosaminoglycans, and cellular hypertrophy. Both osteoblasts and osteoclasts localise throughout the zebrafish jaw at these stages or shortly afterwards [Eames et al., 2013;Sharif et al., 2014;Mork and Crump, 2015]. Anatomically, this manifests as a progressive reduction in the angle between the ceratohyoids and the other branchial arch cartilages between 3-5 dpf.
Here, we report a drug screen of FDA-approved compounds that disrupt ceratohyoid cartilage formation in a perturbation screen. We report a hit rate consistent with published perturbation screens, and we enrich for novel "hits" that target skeletal tissue homeostasis. Our results demonstrate the importance of sustained drug treatment, which might be necessary to overcome compensatory mechanisms and could have relevance in future clinical translation. As proof-of-concept, we demonstrate possible efficacy of 2 compounds in mammalian ex vivo models of disease, including treatment of craniosynostosis.

Materials and Methods
Zebrafish Zebrafish ( Danio rerio ) embryos were obtained from a wildtype strain and raised at 28.5 ° C in accordance with Home Office licence PPL 70/7892. Heterozygous sox10:gfp zebrafish were crossed to wild types to produce heterozygous embryos. Embryos were cultured in 24-well plates (3 per well) and treated with one of 640 FDA-approved drugs from the Enzo BML-2842 (v1.5) library, from mid-to late-gastrulation (8 hours post-fertilisation; hpf) until 5 dpf, at which stage NCC-derived craniofacial cartilage was imaged. Drugs stocks were stored at -80 ° C at 10 m M in DMSO. For drug treatments, small molecules were defrosted at 37 ° C and diluted to 10 μ M in E3 medium. In the secondary screen, putative drug hits were re-tested on >10 sox10:gfp heterozygotes, and in the tertiary screen, hits were tested on wild-type zebrafish stained using Alcian blue. In situ hybridisation was performed by standard techniques, using RNA probes labelled with digoxigenin (Roche) and detected using NBT/BCIP (Sigma). We note that, in order to aid visualisation of cartilage elements in Figure 4 only, PTU treatment was used in this particular experiment. However, throughout our drug screen and in subsequent analyses, PTU was not used, as this has been shown to sensitise zebrafish to jaw patterning defects through dysregulated retinoic acid signalling. Findings in this experiment were confirmed in parallel in embryos not treated with PTU.

Neuroblastoma Cell Culture
Neuroblastoma cell lines SK-N-AS, SK-N-SH, IMR32, and LAN5 are derived from the neural crest lineage within the adrenal gland. Cells were cultured in RPMI medium and 10% fetal calf serum, except for SK-N-SH which used MEME medium plus 10% fetal calf serum. Cells have been STR genotyped by ATCC or by the group of Frank Speleman, University of Ghent. For cell proliferation assays with leflunomide or teriflunomide, between 10,000 and 20,000 cells were plated per well in 96-well plates, and these were grown in the presence or absence of chemicals for 3 days, before being processed with Cell Cycle Kit-8 as per manufacturer's protocol (Sigma Aldrich). Absorbance values were measured at OD450nm and background (no cells) values subtracted before normalisation against untreated samples.

RNAseq and Analysis
The stranded total RNA libraries were prepared in accordance to the Illumina TruSeq Stranded mRNA Sample Preparation (October 2013 Rev E.) for Illumina Paired-End Multiplexed Sequencing. Poly-A mRNA in the tRNA samples were first purified using Illumina poly-T oligo-attached magnetic beads and 2 rounds of purification. During the second elution of the poly-A RNA, the mRNA were also fragmented and primed with random hexamers for cDNA synthesis. Cleaved RNA fragments that were primed with random hexamers were reverse transcribed into first strand cDNA using reverse transcriptase and random primers followed by second strand cDNA synthesis. A single "A" nucleotide is added to the 3 ′ end of the blunt fragments to prevent ligating to one another during adapter ligation reaction. The adapters and indexes are ligated to the end of the double-stranded cDNA to prepare the sample for hybridisation onto a flow cell. The sample is then PCR amplified through 15 cycles of PCR to select samples that have the adapter molecules present on both ends of the DNA. The libraries were validated on the Agilent BioAnalyser 2100 to check the size distribution of the libraries and on the Qubit High Sensitivity to check the concentration of the samples. The pool was then loaded at a concentration of 8 p M onto 1 lane of an Illumina HiSeq2000 flow cell v3. Samples were sequenced using 100 bp SE runs. KEGG pathway gene ontology terms statistically significantly enriched within the list of differentially expressed genes (log 2fold change >2; p < 0.05) while identified according to DAVID software analysis (https://david.ncifcrf.gov/).

Calvarial Explant Culture
We tested FK506 for rescue of craniosynostosis in a calvarial explant system that we and others have developed using Crouzon mice [Eswarakumar et al., 2006]. Heterozygous stud males carrying the Fgfr2 C342Y mutation were crossed with wild-type females on a CD-1 genetic background. Typically 8-12 pups were produced, with an expected 50: 50 ratio of heterozygotes and wild types. Pups were sacrificed at embryonic day (E) 18.5. Skin was removed from the scalp, the skull base was dissected and this and the brain were discarded. No attempt to remove dura or other tissue was made. Explants were cultured in one well of a 12-well plate in 2 mL of DMEM F12 medium (Gibco) supplemented with 10% fetal bovine serum. Explants were cultured for 14 days. Media was changed every 2 days and included 100 n M FK506 dissolved in DMSO, or DMSO alone, as a control. Differentiation of primary osteoblasts derived from wild-type CD-1 calvarial from 4-day-old mice and grown for 14 days.

Primary Osteoblast Culture
Primary mouse osteoblastic cells were obtained by sequential enzyme digestion of excised calvarial bones from 4-day-old CD1 mice using a 3-step process (1% trypsin in PBS for 10 min, 0.2% collagenase type II in Hanks balanced salt solution (HBSS) for 30 min, and 0.2% collagenase type II in HBSS for 60 min). The first 2 digests were discarded and the cells resuspended in MEM supple-

Fig. 1.
Overview of zebrafish drug screen employed in this study. Three embryos from a wild-type x sox10:gfp heterozygote cross per well were treated with one of 640 FDA-approved drugs in a 24-well plate format from 8 hours post-fertilisation (hpf) and screened for quantitative increases in the angle of the ceratohyoid cartilage or gross malformations at 5 days post-fertilisation (dpf).

A Sensitised Screen for FDA-Approved Drugs Targeting Craniofacial Neural Crest
Given that abnormal location of NCSCs within the coronal suture or parietal bone has been proposed to drive ectopic osteoblast differentiation and cranial suture fusion, we hypothesised that small molecules that inhibit NCSC induction or their subsequent differentiation could be therapeutically relevant in craniosynostosis. Following the rationale outlined above, we designed a perturbation screen using sox10:gfp transgenic zebrafish in which green fluorescent protein (GFP) expression is driven by 4.9 kb of the zebrafish sox10 promoter/upstream sequence [Dutton et al., 2008]. In this line, neural crestderived craniofacial cartilage making up the jaw is labelled with GFP, allowing direct visualisation of these structures in live embryos ( Fig. 1 ). During the course of our initial studies, we noticed that one quarter of fry from an incross of sox10:gfp heterozygotes demonstrated severe abnormalities of the jaw, including a strong reduction in Meckel cartilage and severe widening of the ceratohyoids and ceratobranchials ( Fig. 2 A-C), suggesting that the transgene is toxic to NCSC patterning when homozygous. We therefore included only heterozygous transgenic zebrafish in our screen by crossing heterozygous females with wild-type males. We included 3 fish per well in a 24-well plate format. On average, 50% of fry were heterozygous and could therefore be analysed in our screen.
In the primary screen, the effects of 640 FDA-approved drugs were tested by treating embryos from 8 hpf [equating to approximately 50% epiboly (E50)] to avoid defects in early cell division or gastrulation, until 5 dpf when ev- ery embryo was manually photographed ( Fig. 1 ). Embryos were treated with a standard 10 μ M dose of each of 640 compounds from the Screen-Well ® FDA-approved drug library (Enzo, BML-2842 version 1.5) in E3 medium. This is a dose at which most bioactive compounds have been shown to be effective [Rennekamp and Peterson, 2015]. Measurement of the ceratohyoid jaw angle in the 871 heterozygous transgenic animals that were included in our primary screen revealed an average angle of 92.17°, whereas wild-type embryos stained for Alcian blue had an average angle of 60.11° ( Fig. 2 D). Not a single transgenic animal exhibited an angle as low as the average value for wild types. This suggests that the sox10:gfp transgene is therefore partially toxic in heterozygotes, and that our screen was sensitised for NCSC patterning defects.
In the primary screen, we identified 3 categories of candidate hits ( Fig. 3 A). This included drugs that were associated with: (i) a quantitatively increased angle be-tween the ceratohyoid cartilages as a purely objective measure, (ii) gross morphological defects, or (iii) no GFP expression. Given the breeding scheme used, the latter would be expected in 12.5% of embryo trios by chance alone. In total, 156 putative hits were identified in the primary screen, including 30, 88, and 38 of categories i, ii, and iii, respectively ( Fig. 3 B). In a secondary screen, these compounds were each re-tested on approximately 10 transgenic animals, and reproducible hits here were tested in a tertiary screen for their effects on wild-type zebrafish that were stained for Alcian blue. Of the 640 compounds screened, 14 (2.2%) had reproducible effects on craniofacial cartilage in all 3 screens ( Fig. 3 , 4 , and data not shown). We also re-ordered each of the 14 drugs that caused consistent phenotypes across all 3 screens from independent sources and tested them to exclude artefacts relating to chemical batch. Ten of these compounds were novel hits, having not previously been reported to affect craniofacial development to our knowledge. Four of the hits we describe (pimozide, retinoic acid, nisoldipine, and leflunomide) were previously reported in a recent screen of zebrafish craniofacial development with a similar design to ours [Kong et al., 2014]. Notably, of the 38 drugs associated with no GFP expression in the primary screen, only 1 drug (leflunomide) was a confirmed hit, and we showed that this compound completely prevented formation of craniofacial cartilage ( Fig. 4 A).

Temporal and Functional Assessment of Selected Compounds
Representative images of the effects of 4 drug hits on craniofacial cartilage in wild-type embryos are shown in Figure 4 . Treatment from mid-to-late epiboly (i.e., "timezero" in our drug screen) with cyclosporin A, climbazole, or escitalopram caused widening of the ceratohyoids and ceratobranchials and an overall reduction in the lower jaw cartilage, when observed ventrally at 5 dpf ( Fig. 4 A).
In lateral view, the ceratohyoids and Meckel cartilage can be seen to be smaller than in control-treated embryos and downturned ( Fig. 4 A'). In all 3 cases, the ethmoid carti-lage of the upper jaw is present and appeared normal. For each of these drugs, treatment from 29 hpf led to much less severe defects ( Fig. 4 B, B'). Quantitative increases in the angle between the ceratohyoids in embryos treated from this later time point were most apparent for cyclosporin A. Treatment of embryos with leflunomide from E50 prevented the formation of any cartilage ( Fig. 4 A), as reported previously [White et al., 2011]. Remarkably, we found that all embryos treated from 29 hpf, after the critical window of NCSC specification and migration [Luo et al., 2001;Carney et al., 2006], were able to form both upper and lower jaw cartilage, although these structures were grossly abnormal ( Fig. 4 B). This phenotype has not been reported previously and suggests that leflunomide treatment has independent dual effects, inhibiting early NCSC induction and influencing later pharyngeal arch patterning and/or cartilage remodelling. We confirmed that early treatment with leflunomide completely inhibited cranial NCSCs, as demonstrated by in situ hybridisation for sox10 (see below). Collectively, these results hint at the importance of the timing of drug treatments in relation to formation of NCSC-derived jaw cartilage. Leflunomide is an inhibitor of dihydroorotate dehydrogenase (DHODH), an enzyme that is essential for the production of uracil monophosphate, a constituent base that is integral to ribosomal RNA synthesis and ribosome biogenesis [Trainor and Merrill, 2014]. Cyclosporin A is an inhibitor of calcineurin, which is a calcium-dependent serine/threonine protein phosphatase [Kaminuma, 2008]. In response to increased intracellular calcium, calcineurin dephosphorylates the nuclear factor of T-cells (NFAT) transcription factors, allowing their entry into the cell nucleus where they drive transcription. Nfatc1deficient mice exhibit osteopetrosis within the long bones owing to reduced numbers of bone-resorbing osteoclasts. Importantly in the context of craniosynostosis, they also have fewer osteoblasts and exhibit reduced formation of mineralised bone within the frontal bones of the skull resulting in wider cranial sutures [Winslow et al., 2006]. Given our initial hypothesis that compounds that inhibited either NCSC identities or their differentiation could be candidates for treatment of craniosynostosis, we selected leflunomide and cyclosporin A as representative drugs of each of these categories for further temporal and functional investigations. We next undertook a full quantitative temporal analysis of the effects of 10 μ M doses of these 2 drugs using wild-type embryos stained with Alcian blue. In treatment regimens 1-5, embryos were treated from progressively delayed 24-h time points and visualised at 5 dpf, while in treatment regimens 6-10, embryos were treated during 24-h intervals using drug washout ( Fig. 5 ). The angle of the ceratohyoid cartilage was quantified for approximately 10 zebrafish for each regimen for control-and drugtreated embryos. We noted some variability in this angle in control-treated animals in different clutches of fish, and so we have included this value for matched clutchmate controls for direct comparisons to be made. While leflunomide prevented cartilage formation when applied from E50 ( Fig. 4 ), treatments from either 24 or 48 hpf (regimens 2 and 3) were compatible with jaw cartilage formation but severely enhanced the angle of the ceratohyoid ( Fig. 5 ). By contrast, treatment after 48 hpf had no effect. Interestingly, treatment during day 1 or 2 only (regimens 6 and 7) was compatible with cartilage formation but caused patterning defects of similar magnitude to regimens 2 and 3. Treatment during day 3 caused only subtle patterning defects. Interestingly, we also found that treatment with leflunomide for 5 days did permit formation of cartilage when the dose was diluted by 20-fold, showing that lower doses are consistent with at least partial NCSC induction ( Fig. 6 ). We conclude that leflunomide treatment inhibits cartilage formation during day 1 and disrupts cartilage patterning/differentiation during day 2, but that in both cases, sustained treatment is required to exert these effects. For cyclosporin A, we found that sustained treatment between 2-5 dpf was sufficient to disrupt cartilage, identifying day 2 as the critical effective window, although transient 24-h treatments at any time point were not sufficient to disrupt the lower jaw ( Fig. 5 ). For both drugs, these results identify day 2 as a critical window when the targets of these drugs regulate cartilage patterning/remodelling and suggest that compensatory mechanisms can restore normal cartilage formation if drug treatment is attenuated subsequently.
Finally, to begin to investigate potential cellular and mechanistic targets of these drugs, we performed qRT-PCR on 4-dpf zebrafish treated with 10 μ M teriflunomide or FK506 from 29 hpf to monitor markers of key skeletal cell types and cartilage matrix regulatory enzymes using procedues and primers previously reported [Pashay Ahi et al., 2016] ( Fig. 7 ). Teriflunomide treatment reduced expression of the osteoblast marker spp1 and increased expression of ctsk which is expressed in osteoclasts, suggestive of reduced osteoblast and expanded osteoclast cells. Notably, expression of rankl which promotes osteoclast formation through paracrine signalling was reduced, which could reflect a feedback mechanism. Teriflunomide also reduced expression of the chondrocyte marker, col2a1 . Expression of cartilage matrix degrading metalloproteinase enzymes mmp9 and mmp13 was increased, while the inhibitor of these enzymes, timp2a , was reduced. These findings are consistent with reduction in the jaw cartilage observed following leflunomide/teriflunomide treatment. The same general patterns were observed following FK506 treatment, although the effects were less pronounced. Collectively, these results suggest that drug treatments influence several key skeletogenic cell types and reduce cartilage extracellular matrix.

Functional Validation of Leflunomide through Inhibition of NCSC-Derived Neuroblastoma Cells
Perturbation drug screens rely on careful selection of a clinically relevant cell type(s), and because of their disruptive nature, they require subsequent testing in mammalian disease models. To ascertain whether our drug screen could have any relevance to NCSC-related diseases, we began to assess the clinical potential of leflunomide and cyclosporin A in radically different biological settings to the zebrafish jaw.
A first consideration is potential off-target effects of a drug. For this, we tested for phenocopy of the very specific cartilage-loss phenotype associated with leflunomide treatment in zebrafish using a functional analogue. Treatment of embryos with a 10 μ M dose of teriflunomide, which is a metabolic derivative of leflunomide, identically reproduced the absence of cartilage formation in all embryos tested ( Fig. 8 A). Given that this phenotype is consistent with inhibition of NCSC induction, we assessed sox10 by in situ hybridisation as a marker of endogenous cranial NCSCs. Indeed, leflunomide treatment completely blocked formation of sox10+ NCSC formation, although expression within the otic vesicles was preserved, where sox10 is known to play independent roles in formation of the ear primordium ( Fig. 8 A). Collectively, these results confirm that DHODH inhibition inhibits NCSC induction.
We next chose to investigate the effects of leflunomide/teriflunomide treatment on human NCSC-derived neuroblastoma cells, as inhibition of NCSC identities in these cells has shown to be of therapeutic value. Specifically, persistent Sox10 expression has been shown to drive neuroblastoma formation, and depletion of Sox10 can in-hibit NCSC-derived tumour formation [Nagashimada et al., 2012;Shakhova et al., 2012;Kaufman et al., 2016]. We therefore hypothesised that chemical inhibition of DHODH may be therapeutically relevant in neuroblastoma. Treatment of several neuroblastoma cell lines with either leflunomide or teriflunomide caused a dose-dependent reduction in viability ( Fig. 8 B). By contrast, SK-N-AS cells, which are known to exhibit generalised drug resistance mechanisms [Cohen et al., 1995;Gatsinzi et al., 2012], were more resistant to drug treatment, thereby serving as a control for non-specific toxic effects of the drug. Subsequent to this work, a separate study confirmed our findings, reporting that leflunomide treatment represses neuroblastoma tumour growth in vivo, by inhibiting proliferation, enhancing apoptosis, and inducing Sphase arrest [Zhu et al., 2013]. Therefore, leflunomide is a potential new lead for development of therapeutics in neuroblastoma. Current chemical treatment options for neuroblastoma include application of retinoic acid which induces cellular differentiation and production of dendrites, leading to cellular senescence. In contrast, we found that leflunomide treatment did not cause dendrite production, indicating that this drug works through a distinct mechanism ( Fig. 8 B).
To gain mechanistic insight into the effects of DHODH inhibition in NCSCs, we used RNAseq to identify global gene expression changes following teriflunomide treatment of SK-N-SH neuroblastoma cells ( Fig. 8 C) -full dataset given in online supplementary Table 1 (see www. karger.com/doi/10.1159/000491567 for online suppl. material). In total, 133 genes exhibited at >2-fold change in expression levels (adjusted p value <0.05) following 24 h of teriflunomide treatment and were strongly enriched Nuclear-transcribed mRNA catabolic process, nonsense-mediated decay Fig. 8. Functional evaluation of DHODH inhibition. A Phenocopy of cartilage-loss phenotype following treatment with 10 μ M teriflunomide between 50% epiboly to 5 days post-fertilization, and inhibition of sox10 expression at 24 hours post-fertilization following teriflunomide treatment, shown by in situ hybridisation. B Dose-dependent inhibition of cell numbers in SK-N-SH ( n = 6), LAN5 ( n = 3), and SK-N-AS ( n = 5) cells after 3 days treatment with leflunomide. Data are from CCK8 assays with untreated controls given an arbitrary value of 1. All 100 μ M samples are significantly different from untreated samples ( p < 0.05). Furthermore, at 100 μ M both SK-N-SH and LAN5 are significantly more sensitive than SK-N-AS (LAN5:SK-N-AS, p = 0.025; SK-N-SH:SK-N-AS, p = 0.002; ANOVA with post-hoc Scheffé test). Images show morphology of cells treated with 100 μ M leflunomide for 3 days (note no increase in neurite formation, a differentiation marker). C Gene ontology terms statistically significantly enriched within the list of differentially expressed genes (log 2 -fold change >2, p < 0.05) according to DAVID software analysis. Number of terms within each category is listed. * * * * p < 5 × 10 -5 ; * * * p < 10 -4 ; * * p < 10 -3 ; * p < 10 -2 ; no asterisk, p < 0.05. for genes involved in transcription, RNA processing and translation. Consistent with known roles of DHODH in ribosomal RNA synthesis and ribosome biogenesis [Trainor and Merrill, 2014], of 67 downregulated genes with an adjusted p value <0.05, seven (10.4%) encoded ribosomal proteins (of 307 downregulated genes at the less significant p value <0.05 cut-off, 15 (4.9%) encoded ribosomal proteins). In contrast, no ribosomal genes were upregulated, thereby suggesting that leflunomide/teriflunomide inhibit neuroblastoma cell viability through blockade of ribosome biogenesis.

Functional Validation of Tacrolimus as a Potential
Therapeutic Compound in Craniosynostosis FK506 (tacrolimus) is a more potent functional analogue of cyclosporin A. We therefore chose to focus on this drug to evaluate the therapeutic relevance of targeting calcineurin signalling in craniosynostosis. Treatment of zebrafish with 10 μ M FK506 from E50 until 5 dpf caused identical jaw cartilage patterning defects that we demonstrated using cyclopsorin A (data not shown), thereby excluding off-target effects as causes of this phenotype.
Given the importance of osteoblasts in craniosynostosis pathogenesis, we began by testing the effect of tacrolimus treatment on primary osteoblasts. Culture of these cells for 14 days led to expression of osteoblast progenitor, pre-osteoblast and mature osteoblast markers, including Col1a1 , ALP , Runx2 , and Osx ( Fig. 9 A), indicating that this protocol enriched for actively differentiating osteoblasts. These cells were treated with a 10-fold dilution series of FK506 within the range 0-10 μ M , which are doses that have previously been shown to correlate with the effects of genetic mutations in Nfat genes and are thus likely to be physiologically relevant [Winslow et al., 2006]. Generally, at higher FK506 concentrations, expression of osteoblast differentiation markers was increased compared to control, with the maximal effect generally being achieved at 0.1 μ M . We next tested FK506 for rescue of craniosynostosis in a calvarial explant system that we and others have developed using Crouzon mice [Eswarakumar et al., 2006]. Whereas the coronal sutures fused bilaterally over the 2 weeks of culture in all untreated mutant explants, which manifested as a loss of overlap of the frontal and parietal bones and an obliterated suture, none of the FK506-treated mutant explants showed overt signs of fusion after 2 weeks of culture ( p < 0.0001; Fisher exact test) ( Fig. 9 B). Quantification of the area of overlap between the frontal and parietal bones for both the left and right coronal sutures demonstrated complete obliteration of this suture in Crouzon mutant calvaria treated with DMSO alone, while there was no dif-ference between Crouzon mutant calvaria treated with FK506 and either treated or untreated wild-type calvaria ( p = 0.530 and 0.690, respectively; t test) ( Fig. 9 B). These data provide initial evidence for the therapeutic potential of FK506 in craniosynostosis and supports the need for further in vivo testing.

Discussion
In this study, we undertook a perturbation-based screen of FDA-approved compounds using sox10:gfp transgenic zebrafish. This transgene allowed direct visualisation of NCSC-derived craniofacial cartilage that makes up the jaw, without the need for additional staining procedures. We noted partial toxicity of this transgene which manifested as a quantitative increase in the angle between the ceratohyoid cartilages. Thus, it is likely that our screen was sensitised for defects in NCSC-derived craniofacial cartilage. All 14 hits in our screen were reproducible in wild-type/non-transgenic zebrafish confirming that they are not artefacts. Our confirmed drug hit rate was 2.2% of the tested compounds. This is consistent with previously reported perturbation screens [Peterson et al., 2004;Stern et al., 2005;North et al., 2007;Kitambi et al., 2009;Ishizaki et al., 2010;Paik et al., 2010;Rihel et al., 2010;Ni et al., 2011;Peal et al., 2011;Rovira et al., 2011;Namdaran et al., 2012;Ridges et al., 2012;Gut et al., 2013;Le et al., 2013;Kong et al., 2014;Mork and Crump, 2015] suggesting that this screen design had good sensitivity. As discussed in the introduction, while published perturbation screens have an average 29-fold increased hit rate as compared to modifier screens, their success in potential clinical translation is likely to depend on selection of a specific cell type assessed in the screen as well as further testing in mammalian disease models. Repurposing screens using FDA-approved drugs have the advantage that drugs have already passed safety testing in humans. This is important because AstraZeneca reported that data linking a biological target to disease in human subjects and samples were available for 73-82% of the projects that remained active in Phase II, whereas 57-60% of the projects failed on the basis of efficacy where human data was not previously available [Cook et al., 2014].
Our screen focused on NCSC derivatives within the craniofacial cartilage and could therefore identify compounds that regulate NCSCs or remodelling of cartilage/ chondrocytes. Since osteoblasts and osteoclasts are also located throughout this cartilage around the time points that we analysed [Eames et al., 2013;Sharif et al., 2014; Mork and Crump, 2015], it is conceivable that our screen could also identify compounds that influence these cell types, although we note that no mineralised bone is present at these stages. Indeed, our temporal analyses of leflunomide treatment suggested biphasic effects in repressing early NCSC induction and later independent effects on cartilage remodelling. By contrast, our temporal analyses of other drug hits, including cyclosporin A, supported roles limited to cartilage remodelling. For both drugs, our data suggested that sustained drug treatments are also important to reproduce jaw patterning defects, which raises the possibility that compensatory mechanisms can restore normal formation of the jaw. In future, it will be interesting to characterise the exact cellular targets of these drugs as well as drug treatment timing and duration when considering possible clinical application (e.g., craniosynostosis). For both drugs, we were also able to generate initial ex vivo data suggesting their potential relevance for treatment of neuroblastoma or craniosynostosis in mammalian systems. While these data provide some proof-of-concept support for our drug screen, it will be necessary to undertake further pre-clinical testing in precise in vivo models of disease treatment, and this will form the subject of future reports. Of our drug hits, 50-64% are known to target pathways implicated in osteoarthritis (OA) pathogenesis or regulation of bone mineral density ( Table 1 ) [Mazziotti et al., 2010], and at least 1 drug disrupted early NCSC development. This likely reflects the design of our screen which visualised patterning of NCSC-derived skeletal elements. Thus, the hits that we report could be relevant to skeletal tissue homeostasis and disease in a number of different contexts and emphasise the transferability of the perturbation screen design when focused on a specific cell/tissue type of medical relevance. Chondrocytes represent a key cellular component of articular cartilage which is affected in all joints with OA [Buckwalter et al., 2005]. Changes within the microenvironment of the joint can induce an inflammatory response, whereby chondrocytes release cytokines (e.g., IL1β) and matrixdegrading enzymes. Joint destabilisation can lead to increased expression of metalloproteinases, such as AD-  [Burleigh et al., 2012]. Several known human OA susceptibility genes, including gdf5 and mcf2l , are expressed in zebra fish craniofacial cartilage elements and joints from just a few dpf [Mitchell et al., 2013]. Analysis of zebrafish cartilage is therefore directly relevant to OA of the temporomandibular joint which affects approximately 17% of the OA patients [Gremillion et al., 1993]. The drug hits fall into several functional categories, including calcium, retinoic acid receptor, oestrogen, and neuropeptide signalling pathways ( Table 1 ). Leflunomide was the only compound for which we demonstrate clear evidence for direct effects on NCSCs, and this drug has previously been shown to inhibit BRAF V600E -driven melanoma cell growth, which has been proposed to reflect suppression of NCSC identities [White et al., 2011]. Our RNAseq data in neuroblastoma cells suggested that DHODH inhibition particularly affected ribosomal gene expression and transcription. NCSC-derived neuroblastoma cells are known to activate expression of ribosomal genes such that protein synthesis can keep pace with the demands of increased proliferation. This is driven by the protooncogene, MYCN [Boon et al., 2001], which is commonly activated in neuroblastoma.
Cyclosporin A and FK506 are inhibitors of calcineurin, which is a calcium-dependent serine/threonine protein phosphatase [Kaminuma, 2008]. In response to increased intracellular calcium, calcineurin dephosphorylates the NFAT transcription factors, allowing their entry into the cell nucleus where they drive transcription. Nfatc1 -deficient mice exhibit osteopetrosis within the long bones owing to reduced numbers of bone-resorbing osteoclasts. They also have fewer osteoblasts and exhibit reduced formation of mineralized bone within the frontal bones of the skull resulting in wider cranial sutures [Winslow et al., 2006]. Similar effects on osteoblast and osteoclast cell lineages have been observed in calcineurin β1 mutant mice. Calcineurin β1 [Cnb1, also known as Ppp3r1, protein phosphatase 3, regulatory subunit B, alpha isoform (calcineurin B, type I)] is a Ca 2+ -binding regulatory subunit of heterodimeric calcineurin. Loss of the Cnb1 subunit results in a lack of calcineurin enzymatic activity in non-germline cell types. Conditionally targeted mutant mice whereby the Cnb1 gene was inactivated in osteoblasts increased bone mineral density, associated with increased differentiation and mineralisation of primary calvarial osteoblast cultures, and failure of these osteoblasts to support osteoclast growth and survival [Winslow et al., 2006]. These observations led us to hypothesise that FK506 treatment could prevent cranial suture fusion in Crouzon syndrome mice, and we generated initial support for this idea using a calvarial explant system. To take this observation further in future, it will be necessary to test the drug in vivo, evaluate systems for localised drug delivery to avoid unwanted side effects, and to characterise the key skeletal cell type(s) that are influenced within the calvarium since calcineurin signalling can influence osteoblast and osteoclast cells and precursors. Of note, this compound is approved for long-term immunosuppression of very young babies who undergo organ transplantation, suggesting that it may also be efficacious for treatment of craniosynostosis over the first few years of life.
In summary, the identification of new chemical structures that influence NCSC-derived skeletal structures provides new insight into global mechanisms of skeletogenesis and provides a basis to begin to explore possible treatments for conditions such as craniosynostosis. We add to the growing literature concerning use of zebrafish as an in vivo system for chemical screening, which is an important application of this model organism. In future, it will be necessary to characterise these and other drug hits that regulate craniofacial development, including the particular target cell lineages and the utility of these compounds in mammalian disease models.