|Home | About | Journals | Submit | Contact Us | Français|
Mammals and birds have common embryological facial structures, and appear to employ the same molecular genetic developmental toolkit. We utilized natural variation found in bird beaks to investigate what genes drive vertebrate facial morphogenesis. We employed cross-species microarrays to describe the molecular genetic signatures, developmental signaling pathways and the spectrum of transcription factor (TF) gene expression changes that differ between cranial neural crest cells in the developing beaks of ducks, quails and chickens. Surprisingly, we observed that the neural crest cells established a species-specific TF gene expression profile that predates morphological differences between the species. A total of 232 genes were differentially expressed between the three species. Twenty-two of these genes, including Fgfr2, Jagged2, Msx2, Satb2 and Tgfb3, have been previously implicated in a variety of mammalian craniofacial defects. Seventy-two of the differentially expressed genes overlap with un-cloned loci for human craniofacial disorders, suggesting that our data will provide a valuable candidate gene resource for human craniofacial genetics. The most dramatic changes between species were in the Wnt signaling pathway, including a 20-fold up-regulation of Dkk2, Fzd1 and Wnt1 in the duck compared with the other two species. We functionally validated these changes by demonstrating that spatial domains of Wnt activity differ in avian beaks, and that Wnt signals regulate Bmp pathway activity and promote regional growth in facial prominences. This study is the first of its kind, extending on previous work in Darwin's finches and provides the first large-scale insights into cross-species facial morphogenesis.
Craniofacial abnormalities are among the most common birth defects, accounting for approximately one third of congenital abnormalities (1). Targeted mutagenesis in animal models such as the mouse have provided important information on the effects of single genes in defects such as cleft lip and/or palate and craniofacial development in general. However, we still lack a comprehensive description of the spectrum of molecular genetic players in vertebrate facial development. We also have only a very rudimentary description of the genes and pathways that underlie species-specific variation in facial structures. Herein, we present the first large-scale comparative genomic study to exploit the natural variation found in avian facial structures as a novel source of candidate genes for studies of facial morphological variation and mammalian craniofacial development.
Avian and mammalian facial development appear to use the same molecular genetic toolkit. For example, Bmp4 was first correlated with beak width in Darwin's finches (2), was subsequently found to cause cleft lip when mutated in mice (3) and variants in this genes were then identified as being associated with human cleft lip cases (4). Wnt signaling has also been extensively studied in neural crest (NC) generation and migration (5), as well as the development and regionalization of the face in many species (6); however, the role of this pathway in modifying different facial shapes has yet to be evaluated. Recent work (7,8) suggests that evolution of species-specific facial morphology probably involves subtle quantitative, spatial and/or temporal alterations in gene expression. It appears unlikely that the observed variation is due to the exploitation of entirely novel genes or genetic networks. Thus, the genes involved in specifying a bird face probably play similar roles in the development of a mammalian face. Identifying these genes may yield a valuable new source of candidate genes for the study of human craniofacial abnormalities.
Subtle alterations in growth of the frontonasal prominence (FNP) in birds lead to dramatic differences in adult beaks, allowing optimal exploitation of feeding niches. We previously used cell grafting assays to demonstrate that cranial NC cells transplanted from a quail donor to a duck host (the ‘quck’) or from duck donor to a quail host (the ‘duail’) developed facial features reminiscent of the donor species (9). These experiments revealed a cellular origin for species-specific facial morphology by showing that NCs provided patterning information for the face. What these studies left unanswered was how that patterning information was actually encoded. Subsequent studies in birds (2,10–12) and fish (13) identified two genes involved in regulating growth of facial prominences. Bmp4 is up-regulated in Darwin's finches with broad beaks (2,12), although Calmodulin1 is up-regulated in finches that exhibit an elongated beak morphology (12). However, those studies were conducted after morphological variation is already evident and did not resolve what upstream regulators might be responsible for regulating differential Bmp4 and Calmodulin1 expression. In the current study, we sought to identify the set of transcription factors (TFs) and signaling molecules that differ during embryonic development of the avian face in three bird species (chicken, quail and duck) that exhibit very different facial adaptations. Our objective was to identify early mediators of species-specific craniofacial morphology, at stages prior to morphological variation and differential Bmp4 and Calmodulin1 expression, and to identify a set of genes that may play major roles in driving vertebrate facial development and evolution.
The first step in our analysis was to identify the stage at which avian embryos still retain a maximum degree of facial similarity prior to the onset of overt morphological differences. Chick, quail and duck embryos were collected between stages (Sts.) 13 and 28 and examined for species-specific phenotypic variation in their facial prominences, both visually and quantitatively via multivariate analysis. These experiments indicate that at Hamburger–Hamilton developmental stage 20 (St. 20) embryos exhibited a maximum degree of similarity. Soon afterwards, differences in growth trajectories give rise to species-specific morphologies (e.g. narrower beaks in chick and quail embryos versus broader bills in duck embryos) that were readily apparent by St. 25 (48 h later) (Fig. 1, Supplementary Material, Fig. S1). Therefore, we confined our gene expression studies to the embryonic period preceding (St. 20) and following (St. 25) morphological variation, and to the relevant cellular population (the FNP NC).
To identify the differences in gene expression that may be responsible for the morphological variation between species, we utilized a cross-species microarray platform (14,15) to measure gene expression changes in TF genes plus a large number of genes involved in signaling pathways. Gene expression changes of the FNP cranial NC were measured both between species at St. 20 and St. 25, and between developmental stages within one species.
We detected 232 differentially expressed genes in any two-species comparison at either developmental stage (>2-fold change and P-values < 0.05; see Fig. 2 for examples and Supplementary Material, Table S1 for a complete listing). This number is almost certainly an underestimate since it includes only those genes that have clearly identifiable orthologs in the Gallus gallus genome (16). Unfortunately, it appears that ~10% of chicken orthologs are still missing from the published genomic DNA sequence. For example, our Wnt10b microarray probe, designed from the mouse Wnt10b gene, reported a >20-fold increase in transcript levels in duck versus chicken or duck versus quail, NC cells. However, we could not computationally identify a clear Wnt10b ortholog in the chicken genome and thus filtered out data of that type. Probes with unclear orthologs in the G. gallus genome are omitted from the remainder of this discussion (Supplementary Material, Table S2 for a listing that includes these genes). Nevertheless, our analysis revealed dramatic variations in 13 canonical Wnt signaling molecules between the three species, including a 20-fold elevation in Wnt1, Dkk2 and Frizzled-1 expression in duck NCs compared with either chicken or quail, at either St. 20 or St. 25 (Fig. 2, Supplementary Material, Table S1). Additionally, the Wnt antagonist Apc and receptor Lrp5 were elevated 4–5-fold, and the expression levels of genes with known Wnt interactions including C3ip1 (17), Pfdn5 (18) and Tbx20 (19,20) were elevated by at least 9-fold in duck NCs compared with the other species. A recent report extensively analyzed the in situ expression pattern of Wnt signaling components during chicken facial development (21). In contrast to that study, we observed robust expression of Wnt1 throughout the developing face by in situ (Fig. 3I) and microarray analysis (Fig. 2, Supplementary Material, Table S1). This agrees with previous studies that identified Wnt1 as a marker of NC cells (22). Although Wnt signaling has previously been implicated in various aspects of facial development, this is the first description of changes in Wnt signaling components in the faces of different species.
In contrast, nine members of the transforming growth factor beta/bone morphogenetic protein (TGFbeta/BMP) signaling pathway are differentially expressed and many of these are up-regulated in chick and quail. For example, Bmp10, Tgfb2 and Tgfb3 were up-regulated in both quail and chick by 2–4-fold relative to the duck. Furthermore, eight components of fibroblast growth factor (Fgf) signaling and seven components of Notch signaling varied across the comparisons, including a 2–6-fold up-regulation of Fgf10, Fgf13, Fgf16, Hes1 and Lfng in duck.
Finally, we observed remarkably large gene expression changes in specific TFs. Given the fact that our microarray platform interrogates mostly TF gene expression changes, it is not surprising that the majority of differences (180 out of 232) are in this class of genes. Many of the observed TF differences, however, were remarkably large, particularly between duck and the other two species. For example, the homeobox gene Irx2 was up-regulated by ~8-fold in the duck NC cells relative to the chicken or quail (Fig. 2, Supplementary Material, Table S1). Numerous studies have indicated that TF gene expression changes as low as 1.5-fold can have biological relevance, since small changes in these regulators can have large effects on downstream targets (14,23,24).
Despite their similar beak morphologies, comparisons between chicken and quail embryo FNP NCs revealed a few large differences in gene expression. Among these were the Calmodulin1 and Calmodulin2 genes, which were expressed 2-fold less in duck than in chicken and 2-fold less in chicken than in quail. Comparisons between duck and quail showed similar results, with expression of Calmodulin1 and Calmodulin2 being 4-fold less in duck than in quail. Up-regulation of calmodulin gene expression in quail was coupled with changes in the Bmp signaling network, in agreement with previous observations in the beaks of Darwin's finches (2,12). Quail NCs reported a 2–4-fold up-regulation of Bmp pathway members Bmp2, Bmp9, and the Bmp antagonist Noggin, along with a 5-fold down-regulation of Madh1. In contrast, chicken NCs exhibited up-regulated Bmpr1a, Jagged2, Madh2, Osr2, Pax9, Pitx2 and Satb2 (Fig. 2, Supplementary Material, Table S1). It is interesting to note that knock-out of any of these genes in mice results in a variety of craniofacial defects (reviewed in 25). In addition to the genes previously implicated in facial development, 72 differentially expressed genes from the total of 232 fall within known genetic intervals for various craniofacial abnormalities (as defined by Online Mendelian Inheritance in Man, http://www.ncbi.nlm.nih.gov/omim). See the Discussion section for more on this, Table 1 for examples and Supplementary Material, Table S4 for a complete listing. These genes represent a new set of candidate genes for these human disorders.
In addition to comparing between species, we also compared St. 20 with St. 25 samples within each species to measure temporal differences in TF expression between stages that exhibit substantial morphological changes (Fig. 1). In general, we found only minimal changes in gene expression between Sts. 20 and 25 for all three species. Ten genes were differentially expressed by >1.5-fold, but only one of these genes (the transcriptional coactivator Psip1) had more than a 1.7-fold change between the two developmental stages (Supplementary Material, Table S3). These data indicate that, at least for the ~2400 genes measured on our array, the species-specific genetic program for frontonasal mesenchyme was established by St. 20, prior to visible morphological variations. This genetic program is then largely maintained through St. 25, when morphological variations are evident. Thus, frontonasal mesenchymal cells show dramatic, species-specific changes in gene expression and importantly, these changes predate any species-specific variation in facial morphology.
In this study we used three species of birds that diverged ~90–101.7 million years ago (26,27). Although the oligonucleotides in our microarray have been shown to accurately report in the chicken (14,23), it is possible that evolutionary divergence between the species could contribute to false positives in the data set due to varying degrees of sequence homology with the probes on the microarray. This might be expected to manifest itself as higher chicken signals relative to duck, due to better base pair matching. However, by sequencing selected DNA segments from three genes (Osr1, Satb2 and Tgfb2), we found >98% identity to chicken sequences in regions that align with the 50-mer oligonucleotides (data not shown). Therefore, it appears that sequence divergence is most probably not a major source of error in our microarray data.
In order to qualitatively confirm our microarray data we conducted whole mount RNA in situ hybridizations on St. 25 chicken and duck embryos (Fig. 3, Supplementary Material, Fig. S2). All in situs confirmed the trends observed in the microarray data and revealed spatial variations in gene expression (see figure legend for details on specific genes). Additionally, RT–PCRs confirmed that Fzd1, Irx2, Madh2, Pfdn5 and Tbx20 are differentially expressed between duck and chick FNPs from Sts. 17 to 27, at levels that are consistent with the microarray data (data not shown).
We further evaluated one of our microarray observations, the dramatic up-regulation of multiple components of Wnt signaling in the duck NC relative to chicken and quail, by examining the functional consequences of Wnt mis-expression in the developing face. We first determined that ectopic Wnt signal provided by an implanted bead soaked in WNT3A protein could directly increase cell proliferation within the developing FNP (Supplementary Material, Fig. S3). We then utilized a retrovirus expressing a Wnt ligand to test whether excessive Wnt signaling in the chick FNP was sufficient to increase the size of the facial prominences. Unilateral injection of RCAS-Wnt3a into the St. 20 chick face (Fig. 4A) resulted in changes in morphology after 24 h. On the control (uninjected) side, the FNP is close to the lateral nasal and maxillary prominences but has yet to fuse with them to create the nasal pit (Fig. 4B). In contrast, the Wnt infected side demonstrated dramatic enlargement of the FNP both grossly and by tissue sections (Fig. 4C–F).
To place our findings in context with previously published reports (2,10,11) regarding molecular mechanisms involved in beak morphology, we cultured St. 20 FNP NCs with WNT3A protein, and assayed for changes in Bmp expression using qRT-PCR. Within 24 h of exposure to a Wnt signal, FNP cells showed an up-regulation of both Bmp2 and Bmp4 transcripts (Fig. 5A and B). This was confirmed in ovo; St. 13 FNPs infected with RCAS-Wnt2b (Fig. 5C) showed increased expression of Bmp4 after 24 h (Fig. 5D–I).
Since we observed dramatic outgrowth of a chick FNP with excessive Wnt signaling, we hypothesized that different spatial patterning of Wnt signaling could account for shape differences in the chicken beak and duck bill. To test regions of Wnt responsiveness in the developing beak, we developed a Wnt reporter construct in which enhanced green fluorescent protein (eGFP) is under the control of seven Tcf binding sites (6). We examined Wnt responsiveness in both chick and duck in ovo by infecting embryos at St. 13 with the GFP reporter, and examining them after and 96 h, at St. 28 (Fig. 6). We chose St.13 because the NC cells that contribute to the upper beak have populated the FNP but growth has yet to ensue; consequently, injections at this stage produce widespread infection by St. 20 (6). The St. 28 chicks displayed a robust region of reporter activity in a midline stripe down the FNP, keeping with the dramatically elongated V-shaped FNP in chicks (Fig. 6E). In contrast, duck embryos showed prominent GFP expression in two lateral domains of the FNP (Fig. 6F); this expression corresponds to outgrowth of the U-shaped bill of ducks. These experiments suggest that differential regulation and location of Wnt signaling may contribute to species-specific beak morphology through alteration of the growth trajectories of the FNP and regulation of Bmp signaling, and functionally validate the dramatic changes in Wnt signaling identified by our microarray analysis of chick, duck and quail FNPs.
We employed a novel comparative genomic approach exploiting natural variation in bird beak shape as a tool to discover new candidate genes that regulate mammalian craniofacial development. Of the 232 genes we identified as being differentially expressed between the developing beak of the chicken, quail and duck, 22 genes, including Fgfr2, Jagged2, Msx2, Satb2 and Tgfb3, have been previously implicated in a variety of mammalian craniofacial defects. An additional 72 genes reside in genomic intervals associated with various craniofacial abnormalities (Table 1, Supplementary Material, Table S4) and are a new source of candidate genes for these disorders. For example, deletions and duplications of a 3 Mb region at 22q11.2 have been associated with DiGeorge, Opitz GBBB and velocardiofacial syndromes, and result in craniofacial dysmorphisms such as broad nasal root, midface hypoplasia and cleft palate (reviewed in 28). Although Tbx1 has been proposed as a candidate gene for these disorders, mutations in this gene do not explain all cases, suggesting additional genes may have roles in disease pathogenesis. Within the 3 Mb region, we observe differential expression of the TFs Lztr1 and Pcqap (Med15), suggesting they may also serve as candidates. Finally, this data set has identified genes previously not known to be expressed in the developing face, such as the TFs Phf16 and Tbx20 (Figs 2 and and3).3). Phf16 is a TF of as yet unknown function. Tbx20 has been investigated in heart development, but it has never before been implicated in facial development. Tbx20 is particularly interesting, since it has been shown to negatively regulate the Wnt signaling pathway during Drosophila segmentation (19) and positively regulate non-canonical Wnt signaling during facial neuron development (20). Additionally, this gene is highly expressed in the FNP of the duck, whereas only detectable by extensive RT–PCR in the developing face of the chicken and quail. It should be noted that of those genes verified by RT–PCR and in situ hybridization, none were uniquely expressed in one species, but not in the others.
The greater then 200 genes we identified using an unbiased genome-scale screen of TFs and members of developmental signaling pathways can also be used to further study evolution of the face. Our work is complimentary to, and extends upon, previous studies of morphological variation in avian beaks (2,10–12) that identified two genes influential in beak outgrowth. Together those studies indicated that modulations in Bmp4 and Calmodulin1 activity can alter beak morphology, but they did not clarify whether these genes are initiating morphological changes or whether their expression is simply changing in response to an upstream mediator. Studies in cichlid fish strongly support this latter hypothesis: Bmp4 expression in the cichlid face is associated with changes in jaw morphology, but Bmp expression itself is controlled by transcriptional regulators that are themselves most likely responsible for facial variation (13,29).
In keeping with this latter role for Bmp activity in patterning the face, our microarray analyses did not detect significant variations in Bmp4 expression levels between billed and beaked embryos. One explanation for this finding is that Bmp4 expression begins to gradually switch from epithelia to mesenchyme (which we analyzed in this study) at St. 24 (30) and it may be at this later stage of embryonic development when Bmp signaling becomes most critical for the growth of the facial prominences. However, our studies did identify other members of the Bmp signaling network, as well as the previously described Calmodulin pathway (12).
We demonstrate that a Wnt signal is capable of inducing Bmp expression in the FNP cranial NC both in vivo and in vitro. Taken together with the fact that our studies were completed at stages prior to identifiable species-specific facial morphologies (Fig. 1), this suggests that the TFs and signaling pathways we identified may function upstream of, and likely in conjunction with, the Bmp and Calmodulin pathways to influence species-specific facial morphology.
Chick, quail and duck embryos were obtained through AA Farms (Westminster, CA, USA). Eggs were set in a rocking incubator at 37°C and windowed at 48 h to determine developmental stage. Staging was done according to Hamburger–Hamilton criteria (31). Embryos were collected at appropriate stages, fixed in Bouin's solution and photographed.
FNPs were isolated in cold PBS from St. 20 and St. 25 embryos, staged according to Hamburger–Hamilton criteria (31). FNPs were placed in 1.26 U dispase in 1× PBS for 15 min at room temperature, then into DMEM with 10% FBS. Surface ectoderm and forebrain neuroectoderm were removed using sharpened tungsten needles. Isolated samples of frontonasal mesenchyme from 40 embryos were pooled in TRIZOL (Invitrogen), and total RNA extracted per manufacturer's instructions. cDNA was linearly amplified as previously described (14,23). On average, 25 µg of purified, sense strand polyadenylated RNA was generated for use in microarray hybridizations.
Oligonucleotide probes on the array are 50–70mers designed to coding regions of genes, thus allowing cross-species comparisons. The probes on the array interrogate ~2000 TFs genes (15), as well as ~400 growth factors and morphogens implicated in craniofacial development. These morphogens included nearly all components of the Fgf, Wnt, TGFbeta/BMP, Pax-Eya-Six-Dach, Notch and Hedgehog signaling networks. We have previously shown that these probes accurately report in the chicken when used under appropriate hybridization conditions (14,23). Oligonucleotides were spotted in duplicate on glass slides.
One microgram of run-off RNAs from each sample were used as templates in cDNA synthesis reaction using an oligo dT17-primer tagged with either Cy3- or Cy5-specific oligonucleotide sequence (3DNA Array 50 kit, Genisphere). Cy3- and Cy5-labeled cDNA populations were hybridized to microarray slides at 42°C for 12 h. This temperature was calculated assuming ~70–75% sequence identity between chicken (target) and human or mouse (probe) sequences (16). Hybridized arrays were washed and processed with Cy3 and Cy5 dyes per manufacturer's instructions.
For each species, the early (St. 20) was compared with the later (St. 25) developmental stage. Stage-matched comparisons were also performed for each pair of bird species at both St. 20 and St. 25. A minimum of four separate microarray hybridizations (two replicates and two dye switches) was carried out for each comparison. Given that the NC samples were a mixed pool containing at least 40 embryos from various hatchings, replicate biological samples were not necessary. A total of 55 array comparisons were conducted for this study. All data, array designs and analysis parameters are available through http://www.ncbi.nlm.nih.gov/geo/ under accession number GSE11099 and comply with the ‘minimum information about a microarray experiment’ requirements.
Microarray image intensity levels were quantitatively measured using confocal laser scanning (GMS 418 Scanner, Affymetrix) and the resulting data were analyzed with the BioDiscovery Imagene 6.0. Data from raw intensity values were normalized by locally weighted linear regression (LOWESS), which compensates for non-linear intensity variations. After normalization, fold changes from replicate oligonucleotide probes was averaged. Unsupervised hierarchical clustering was then performed using the dChip software package to determine the quality of replicate microarray experiments. We implemented an arbitrary cut-off for background intensity values for each microarray chip based on the intensity of control oligonucleotides, and low intensity filtering was performed to exclude genes with intensities lower than this specified threshold. We next selected genes that followed the same trend in at least 80% of the replicated hybridizations; genes that did not follow the same trend over replicate hybridizations were excluded from further analysis. P-values were calculated using a one-sample t-test on fold change data from replicate hybridizations.
The genes identified by the above methods did not necessarily meet all criteria for all comparisons. For instance, a gene may be significant in duck relative to the other species, but not between chicken and quail comparisons. Therefore, we manually extracted the data for ‘missing’ comparisons, allowing us to analyze the patterns of expression across all comparisons.
St. 25 Chicken (Gallus domesticus, Ideal Poultry, Cameron, TX, USA) and duck (Anas platyrhynchos, Metzer Farms, Gonzales, CA, USA) embryo heads were dissected in cold PBS and fixed in 4% paraformaldehyde in PBS overnight at 4°C. Embryos were serially dehydrated to 100% methanol for storage, and rehydrated in PBS before in situ hybridization. PCR primers were used to amplify 500–600 bp products from highly conserved regions of chicken; all products were sequence verified. A T7 viral promoter was added to the 5′ end of either the sense (negative control) or antisense (experimental) strand using PCR. DIG-labeled RNA probes were generated from the cDNA fragments using Ambion T7 transcription kits and DIG-UTP (Roche). Whole mount in situ hybridization were then performed as previously described (6) on stage-matched embyros.
Affi-Gel Blue beads (100–200 mesh Bio-rad; Hercules, CA, USA) were soaked in recombinant WNT3A protein (R&D Systems, Minneapolis, MN, USA) (400 ng/ml in PBS) or PBS at 37°C for 1 h. WNT3A-soaked beads were placed under the facial ectoderm within the FNP mesenchyme at St. 23. Cell proliferation was assessed 48 h later by BrdU immunostaining of tissue sections.
Replication competent retrovirus (RCAS) was produced, concentrated and titered in DF-1 cells (American Type Culture Collection; Manassus, VA, USA) as described (32). For virus delivery, 2 µl of viral supernatant (109 pfu/ml) was injected to the facial tissue of developing embryos. Overexpression of Wnt was achieved by injection of RCAS encoding Wnt2b. Transgene expression was assayed 48 h after injection by in situ hybridization with probes against Wnt2b or the RCAS virus (Rsch).
NC cells from St. 20 chicks were isolated and cultured in 12-well plates until confluent. Media was supplemented with either PBS or 200 ng/ml recombinant WNT3A protein (R&D Systems, Minneapolis, MN, USA) in PBS for 24 h. RNA isolation was performed with the RNeasy Mini Kit (Qiagen Sciences, Maryland) as previously described (33). After DNase treatment, reverse transcription was performed with Taqman Reverse Transcription Reagents (Applied Biosystems, Foster City, CA, USA). Quantitative real-time polymerase chain reaction was carried out using the Applied Biosystems Prism 7900HT Sequence Detection System and Power Sybr Green Master Mix (Applied Biosystems). The levels of gene expression were determined by normalizing to the values of Gapdh. All reactions were performed in triplicate.
We constructed a Wnt reporter virus where eGFP expression was driven by 7xTcf binding sites (34). This DNA construct was cloned into the RCAN vector and virions were produced by combining the 7xTcf-eGFP construct with the VSV-G envelope plasmid and the packaging plasmid (35) then transiently transfecting 293T cells. The DNA was introduced into cells via calcium phosphate precipitation. Media was collected, pooled, filtered and concentrated by ultra-centrifugation. The resulting viral pellet was re-suspended in PBS then tittered on chick embryonic fibroblasts. A control retrovirus expressing eGFP under a strong constitutive reporter was employed for all assays (6). St. 13 embryos were infected with 1.0 µl of a 106 virions/μl solution and incubated until they reached St. 25 or St. 28. The pattern of eGFP activity was determined by examination of whole embryos under fluorescent light, and by immunostaining of tissue sections.
This work was supported by the National Institutes of Health (NRSA F32DE017499-01 S.A.B., RO1-DE012462-06A1 J.A.H., P50DE16215-05 and R01NS039818-09S1 M.L.).
We would like to thank members of the Helms lab for helpful discussion and Dr Anne Bowcock for critical reading of this manuscript.
Conflict of Interest statement. None declared.