Fgf4 is critical for maintaining Hes7 levels and Notch oscillations in the somite segmentation clock

During vertebrate development, the presomitic mesoderm (PSM) is periodically segmented into somites, which will form the segmented vertebral column and associated muscle, connective tissue, and dermis. The periodicity of somitogenesis is regulated by a segmentation clock of oscillating Notch activity. Here, we examined mouse mutants lacking only Fgf4 or Fgf8, which we previously demonstrated act redundantly to prevent PSM differentiation. Fgf8 is not required for somitogenesis, but Fgf4 mutants display a range of vertebral defects. We analyzed Fgf4 mutants by quantifying mRNAs fluorescently labeled by hybridization chain reaction within Imaris-based volumetric tissue subsets. These data indicate that FGF4 controls Notch pathway oscillations through the transcriptional repressor, HES7. This hypothesis is supported by demonstrating a genetic synergy between Hes7 and Fgf4, but not with Fgf8. Thus, Fgf4 is an essential Notch oscillation regulator and potentially important in a spectrum of human Segmentation Defects of the Vertebrae caused by defective Notch oscillations.

Introduction 1 2 A common developmental mode employed by many embryos is a segmentation clock that 3 oscillates within a posterior growth zone; with each cycle, a segment forms. This stratagem has 4 evolved independently within the three major bilaterian clades (annelids, arthropods, and 5 chordates) as well as in plants [1][2][3][4][5] . In chordates, a segmentation clock oscillates in the presomitic 6 mesoderm (PSM); with each cycle, a pair of somites form flanking the neural tube 6,7 . Somites 7 differentiate into dermis, skeletal muscle, tendons, as well as the vertebral column, which retains 8 the segmented attribute of the somites 8 . 9 10 In vertebrates, genes with an oscillatory pattern within the PSM include those encoding 11 components or targets of the FGF, WNT and Notch signaling pathways 7,9,10 . However, which 12 individual genes oscillate differs among species, with the exception of the Notch-responsive 13 HES/HER transcription factors, suggesting that Notch signaling is at the core of the 14 somitogenesis clock 9 . Supporting this idea, genetic and pharmacological manipulations 15 demonstrate that Notch pathway oscillation in the mouse embryo is essential for somitogenesis 16 11 . In the mouse, oscillatory waves of the Notch1 receptor and Delta-like 1 (Dll1) ligand 17 expression, as well as the activated Notch receptor (cleaved Notch intracellular domain, NICD) 18 sweep from the posterior to anterior, where oscillations arrest 12,13 . These oscillations are 19 established through negative-feedback loops of Notch components such as the transcriptional 20 repressor, HES7 14 , and the glycosyltransferase, LFNG 15 , both of which are encoded by 21 oscillating genes under Notch pathway regulation 14 . Notch oscillations arrest in the anterior 22 PSM, where NICD cooperates with TBX6 to periodically activate Mesp2 expression 16 . MESP2 23 then regulates formation of the nascent somite as well as its rostro-caudal patterning, which is 1 essential for normal patterning of the subsequent vertebral column 17,18 . 2   3 Mutations in nearly all of the aforementioned Notch pathway genes have been identified in 4 human patients where defective somitogenesis is thought to be the cause of Segmentation 5 Defects of the Vertebrae (SDV) 19,20 . For example, frequently recessive mutations in DLL3, 6 HES7, MESP2 and LFNG and a dominant mutation in TBX6 21,22 have been identified in patients 7 with spondylocostal dysostosis (SCDO), which is characterized by severe vertebral 8 malformations that include hemivertebrae, vertebral loss and fusion along the length of the axis 9 23 . Whereas SCDO is relatively rare, congenital scoliosis (CS), defined as a lateral curvature of 10 the spine exceeding 10%, is much more common, with a frequency of 1:1000, which is suspected 11 to be an underestimation because asymptomatic individuals do not seek medical care 24 . Mutated 12 alleles of HES7, LFNG, MESP2, and TBX6 are all associated with CS 22-25 . 13 14 The position in the anterior PSM where Notch oscillations arrest and Mesp2 expression 15 establishes the future somite boundary is called the determination front 15,18,26 . This region is also 16 the anterior limit of the wavefront in the classical clock-and-wavefront, a theoretical model 17 proposed over 40 years ago to explain the precipitous and periodic formation of segments in the 18 PSM 27 . In this model, wavefront activity prevents the PSM from responding to the segmentation 19 clock; hence somites form only anteriorly, where the wavefront ends. In chick or zebrafish 20 embryos, exogenously added FGF protein or pharmacological inhibition of FGF signaling will 21 shift the determination front rostrally or caudally, respectively. In the mouse, Cre-mediated 22 inactivation of Fgf4 and Fgf8 specifically in the PSM results in an initial expansion of Mesp2 23 expression, followed by the premature expression of somite markers throughout the PSM 28 . 1 Canonical WNT signaling is also a wavefront candidate, mostly because ectopic activation of 2 bcatenin in the PSM results in an expansion of this tissue and a rostral shift of Mesp2 expression 3 29,30 . However, this expansion observed in bcatenin "gain-of-function" mutants requires FGF4 4 and FGF8 activity, suggesting that these FGF signals are synonymous with the wavefront 28 . 5 6 In addition to its role in preventing somite differentiation, FGF signaling is also implicated in the 7 regulation of key Notch components of the segmentation clock. Several studies have 8 demonstrated that pharmacological inhibition of FGF disrupts Notch oscillations 31,32 . In mouse 9 embryos with PSM-specific loss of FGF receptor 1 or both Fgf4 and Fgf8, Notch pathway genes 10 such as Hes7, are downregulated, although this analysis can be complicated due to the loss of 11 PSM tissue in these mutants 28,32 . Here, we focus on FGF-Notch interactions by analyzing mouse 12 mutants that lack only one of the wavefront Fgf genes, Fgf8 or Fgf4. An indispensable technique 13 in our analysis is whole mount in situ hybridization chain reaction (HCR), which allows us to 14 multiplex different gene expression domains and quantify mRNA levels within specific 15 embryonic tissues [33][34][35] . We demonstrate that, while Fgf8 is not required for somitogenesis, Fgf4 16 is required for normal Notch oscillations and patterning of the vertebral column. To analyze the role of Fgf4 or Fgf8 expression in the primitive streak and PSM ( Figure 1A -B), 5 we inactivated each gene specifically within these tissues using TCre transgenic activity 36 , thus 6 generating "Fgf4 mutants" (TCre; Fgf4 flox/D ; "D"= "deleted" or null) or "Fgf8 mutants" (TCre; 7 Fgf8 flox/D ; see Table 1). Whereas Fgf8 mutants do not survive much beyond birth due to kidney 8 agenesis 36 , Fgf4 mutants are viable and found at Mendelian ratios at weaning (n = 18 controls 9 and n = 21 mutants). Skeleton preparations of Fgf8 mutant embryos at E18.5 (n = 22) revealed 10 all vertebral bodies were present and normally patterned. Fgf8 mutants also presented with minor 11 cervical and/or lumbar homeotic transformations, each with incomplete penetrance and 12 expressivity: small ribs were sometimes present in the most posterior cervical vertebra (8/22, 4 13 bilateral and 4 unilateral) or on the most anterior lumbar vertebra (8/22, 4 bilateral and 4 14 unilateral). On the other hand, Fgf4 mutants display a variety of segmentation defects in the 15 cervical and thoracic vertebrae with 100% penetrance (Compare Figure 1D, D' with C). An 16 average of 7.9 defects occurred per mutant and consisted of hemivertebrae, and misshapen and 17 deleted vertebrae ( Figure 1E). 18

19
We examined gross somite patterning in Fgf4 mutants by staining embryos at various stages for 20 Uncx4.1 mRNA, which marks the posterior somite compartment 37,38 . Analysis of 39 Fgf4 21 mutants revealed a frequency of irregular Uncx4.1 expression specifically in future cervical and 22 thoracic somites with full penetrance ( Figure 1F-J). 23 To determine if these malformed somites were due to an error in segmentation, we examined 1 Mesp2 expression, which occurs in the anterior presomitic mesoderm (PSM) and is required for 2 normal somite segmentation and rostral-caudal somite identity 18 . At the stages when irregular 3 somites are emerging from Fgf4 mutant PSM, we detected aberrant Mesp2 expression, whether 4 we analyzed gene expression by traditional wholemount in situ hybridization (WISH) staining 5 ( Figure 1K  we examined the anterior-posterior length and expression levels of genes responsive to this 4 activity at early somite stages when aberrant Mesp2 expression occurs. Both Msgn1 and Tbx6 are 5 required to specify the paraxial mesoderm 39-41 , and are silenced in the absence of wavefront 6 activity in Fgf4/Fgf8 double mutants 28 . Neither the length nor level of expression of Tbx6 and 7 Msgn1 were significantly altered in Fgf4 mutants ( Figure 2C, C", D, D", E, F). Consistent with 8 this observation, the paraxial differentiation marker, Meox1, was not expanded into the PSM 9 (Figure 2 C' and D'), as occurs in Fgf4/Fgf8 double mutants 28 . 10 11 Therefore, we conclude that wavefront activity is normal in Fgf4 mutants, an insight consistent 12 with the observation that normal axis extension occurs in these mutants, with no loss of caudal 13 vertebrae 42 . We surmise that this normal wavefront position and signaling in Fgf4 mutants are 14 likely maintained by Fgf8, which is expressed at normal levels ( Figure 3A-C) and is sufficient 15 for maintaining normal expression levels of the FGF target genes, Spry2, Etv4, and Spry4 ( Figure  16 3D-F'''); these FGF targets are silenced in Fgf4/Fgf8 double mutants 28 . Therefore, neither a 17 change in wavefront activity nor a change in determination front position explains the aberrant 18 pattern of Mesp2 expression in Fgf4 mutants. 19 20 Oscillation of Notch family components is altered in the PSM of Fgf4 mutants. 21 We then examined the pattern of activated Notch in Fgf4 mutants by immunostaining for Notch 22 intracellular domain (NICD), which is an obligate factor in the transcriptional complex that 23 activates Mesp2 17 . Overall NICD levels were unchanged between Fgf4 mutants and littermate 1 controls at the embryonic stages when somitogenesis was abnormal ( Figure 4A). At these stages, 2 we always observed two to three distinct stripes of NICD along the anterior-posterior axis in 3 both mutant and littermate control PSM ( Figure 4B-C), demonstrating the oscillatory nature of 4 Notch activation 12,15,43 . However, mutant oscillatory stripes of NICD were always less distinct, 5 compared to controls. This was more evident when the relative intensities of NICD 6 immunostaining signals were modeled using Imaris software. In controls, cells medial-lateral to 7 each other had similar levels of activated NICD, whereas in mutants, this coordination was less 8 distinct ( Figure 4B'-C'). Importantly, this blurred pattern occurred in the anterior PSM, where 9 NICD activates Mesp2 expression at the determination front ( Figure 4B'-C', brackets). We proceeded to analyze the expression of Hes7 in greater detail in Fgf4 mutants. Hes7 is within 22 the Hes/Her class of transcriptional repressors that are the only oscillating clock genes conserved 23 amongst mouse, chicken, and zebrafish 9 . In the mouse, mutations that accelerate HES7 1 production will accelerate the tempo of the segmentation clock 44 . Hence, Hes7 is considered to 2 be a fundamental pacemaker of the segmentation clock that controls somitogenesis 45  by classifying static expression patterns into three phases 46 . This approach has been used to 10 characterize Hes7 expression at E9.5-E10.5, when one to two stripes of expression are observed 11 47,48 . However, in the E8.5 PSM, when aberrant Notch signaling occurs in Fgf4 mutants, there are 12 two to three Hes7 expression stripes in both controls and Fgf4 mutants, indicating a faster 13 somitogenesis clock at this stage ( Figure 4, 5). Therefore, we generated criteria for classification 14 of phases at E8.5 as follows. Phase I: three stripes with the most posterior stripe limited to the 15 posterior midline and the most anterior stripe having reached the anterior boundary of the PSM 16 ( Figure 5A). Phase II: two stripes with the posterior stripe having expanded laterally compared to 17 phase I and the anterior stripe having not reached the anterior limit of the PSM ( Figure 5B). 18 Phase III: two distinct stripes and a third stripe initiating at the posterior midline ( Figure 5C). 19 Control embryos are distributed nearly equally between each phase ( Figure 5G) as is the case for 20 Hes7 expression in older embryos 47,48 . It is challenging to place Fgf4 mutants in phases, 21 emphasizing the aberrant Hes7 expression pattern. However, in our assessment, we allocated 22 Fgf4 mutants within all three phases ( Figure 5D-F), with most found in oscillation phase II 1 ( Figure 5G). 2 3 Hes7 is expressed in both the PSM and neural ectoderm. To quantify expression only in the 4 PSM, we used Imaris software to create a volumetric model of the PSM, based on Tbx6 5 expression, which is limited to the PSM at these stages 49 . By measuring the intensity of the Hes7 6 HCR signal within this volume, we obtained a PSM-specific Hes7 quantification for each 7 embryo (see  To correlate Hes7 expression levels with pattern, we created Imaris-generated spot models of the 18 HCR signal, which correspond to a single mRNA molecule or clusters of mRNA molecules 19 within a subcellular-volume (see Materials and Methods). We then colored-coded individual 20 spots to reflect the intensity of the Hes7 HCR signal, using a linear series of intensity-based 21 cutoffs every 20% to generate five colors (quintiles) ( Figure 6A-B). In controls, the lowest 22 intensity quintile (blue) contains the most spots (40%) and each higher intensity group contains 23 10% fewer spots ( Figure 6A, white bars in C). The distribution of spots in the Fgf4 mutant, 1 compared to controls, is significantly skewed towards the lowest quintile at the cost of higher-2 level expression spots ( Figure 6B, black bars in C). This modeling provides an effective 3 illustration of the pattern of Hes7 expression. 4 5 In control embryos, each stripe of Hes7 expression contains a concentric gradient of signal 6 intensity with highest expression at the center (white spots) and lowest expression at the outside 7 (blue spots) ( Figure 6A, D). Fgf4 mutants maintain some of this concentric organization, but the 8 boundaries between groups is less clear (Figures 6B,E); in particular the trough between peaks 9 of Hes7 expression is more shallow, and contains more higher-intensity spots (mostly second 10 quintile, green) than littermate controls (Figures 6D' and E'). To determine if this indicated a 11 failure to repress Hes7 transcription in these regions, we performed HCR using a probe that 12 hybridized to the Hes7 intronic sequences, and therefore specific to newly transcribed pre-13 mRNA. This analysis revealed that transcription of Hes7 is reduced in the posterior PSM and is 14 more widespread, extending into the trough between Hes7 mRNA peaks in the Fgf4 mutant 15 (white brackets in Figure 6 F', G'). Therefore, we hypothesized that FGF4 is required to 16 maintain Hes7 transcription in the PSM above a threshold required for normal Notch oscillation. 17 The reduced level of HES7 in Fgf4 mutants is insufficient to fully autorepress its own 18 expression, resulting in an aberrant pattern of expression. 19 20 21 22 A synergistic defect in Fgf4/Hes7 mutants reveals that Fgf4 is required to maintain Hes7 1 above a critical threshold 2 To test our hypothesis that a reduction of Hes7 expression causes the Fgf4 mutant vertebral 3 defects, we asked if these defects worsen if we further reduce Hes7 expression by removing one 4 gene copy. We compared such mutant to littermate controls that were simple Fgf4 mutants (with 5 two wildtype Hes7 alleles) or Hes7 heterozygotes. These Hes7 heterozygotes also carried TCre 6 and one floxed Fgf4 allele (see Table 1, and Figure 7) resulting in Fgf4 heterozygosity in the 7 TCre expression domain. However, Fgf4 heterozygosity had no effect on the phenotype because 8 the defects we observed were similar to the Hes7 heterozygous defects reported by the 9 Dunwoodie lab 25 . About 50% of our Hes7 heterozygotes had defective lower thoracic vertebrae 10 with a frequency of 2.8 defects per animal. In littermate Fgf4 mutants, vertebral defects were 11 completely penetrant, with an average of 7.7 defects per animal, a frequency similar to that of 12 progeny in our original genetic cross ( Figure 1). However, compound Fgf4/Hes7 mutants 13 displayed a large set of defects with 25 defects per animal ( Figure 7C-E), a frequency 14 significantly greater than littermate Fgf4 mutants (3-fold greater, p < 0.005) or Hes7 15 heterozygotes (9-fold greater, p < 0.001). These data indicate a synergistic, as opposed to 16 additive, effect of loss of one Hes7 allele in the Fgf4/Hes7 compound mutant. 17

18
We performed a similar genetic analysis where we determined if Hes7 heterozygosity likewise 19 affects Fgf8 mutant defects. We examined littermates of the last cross described in Table 1  in the posterior cervical vertebra (3/5, 2 bilateral and 1 unilateral) or anterior lumbar vertebra 23 (3/5, 1 bilateral and 2 unilateral). Hes7 heterozygotes (also heterozygous for a floxed Fgf8 allele) 1 displayed a nearly identical frequency (2.7 defects per animal) as observed in 2 the Fgf4/Hes7 experiment ( Figure 7A). Vertebral defects in compound Fgf8/Hes7 mutants were 3 not synergistic (4 defects per animal) but were clearly the Hes7 heterozygous defects added to 4 the relatively mild Fgf8 defects (1.8 defects per animal, Figure 7 -figure supplement 1). 5 Therefore, we demonstrate that the relationship between Hes7 and Fgf4 is unique in that there is 6 no genetic interaction between Fgf8 and Hes7. 7 8 We then examined levels and spatial patterns of Hes7 mRNA expression in each genotype from 9 the Fgf4/Hes7 mutant littermates, specifically within a volumetric model of the Tbx6 expression 10 domain, as we had in Fgf4 mutants ( Figure 5, 6). We observed a reduction in Hes7 expression 11 that correlated with the severity of vertebral defects within each genotype ( Figure 7E). Hes7 null 12 heterozygotes had only a 19% reduction in expression, presumably because a loss of HES7 auto-13 repression results in enhanced transcription from the remaining wildtype allele. Such 14 compensatory upregulation fails in compound Fgf4/Hes7 mutants because we observed an 80% 15 Hes7 reduction to occur in these embryos ( Figure 7E). Analysis of intensity-colored spot models 16 of PSMs of these genotypes suggests a threshold effect of Hes7 levels on oscillatory gene 17 expression ( Figure 7F-I). Oscillations appear relatively normal if Hes7 levels are reduced 19% 18 (in Hes7 heterozygotes, Figure 7G) and disordered with a 33% reduction (in Fgf4 mutants, 19 Figure 7H). The synergistic 80% reduction of Hes7 that occurs in compound Fgf4/Hes7 mutants 20 causes a severe dampening of oscillatory gene expression ( Figure 7I). The resulting vertebral 21 defects, though relatively severe ( Figure 7A-D), are not as extreme as occurs in Hes7 null 22 homozygotes 47 , indicating that this reduced level of Hes7 supports limited patterning during 23 segmentation. Together, our data support our hypothesis that FGF4 acts to maintain Hes7 above 1 a necessary threshold for normal somitogenesis. In Fgf4 mutants, a reduction in Hes7 levels, 2 generates uncoordinated, asynchronous Notch oscillations. Disordered Notch oscillations initiate 3 Mesp2 unevenly in the anterior PSM leading to improperly shaped somites and subsequently, 4 malformed vertebrae. 5 6 Discussion 1 2 Here we describe mutants with only Fgf4 or Fgf8 inactivated specifically in the PSM. We found 3 that Fgf4 mutants display a range of cervical and thoracic vertebral defects that are caused by 4 defective Notch oscillations during somitogenesis. In contrast, the vertebral columns of Fgf8 5 mutants are normally segmented, with about 30% displaying minor homeotic transformations; 6 such alternations in vertebral identity are not likely due to defects in somitogenesis per se. 7 Previously, a role for FGF signaling in wavefront activity in chick and zebrafish embryos was 8 supported by pharmacological manipulation 26,50 and in mouse mutants where both Fgf4 and 9 Fgf8 are simultaneously inactivated using the same TCre activity that we use in this study 28 . 10 These double Fgf4/Fgf8 mutants as well as mutants with tissue-specific inactivation of Fgfr1 11 support a role for FGF signaling in clock oscillations, but these conclusions were complicated by 12 a loss of PSM tissue due to potential wavefront defects 16,28,31,32 . Compared to these Fgf pathway 13 mutants, the phenotype of Fgf4 mutants is relatively subtle, with no wavefront defect, as 14 indicated by no loss of caudal vertebrae 42 and no quantitative change in wavefront gene markers 15 (Fig 2). This lack of any overt axis extension defect allows us to unambiguously identify Fgf4 as 16 an Fgf ligand gene required for a normal segmentation clock. This insight is supported by 17 recently published in vitro work on the human segmentation clock 51 . 18 19 Our mutants model human Segmentation Defects of the Vertebrae (SDV); Fgf4 mutants 20 resemble CS and the synergistic phenotype resulting from the additional removal of one Hes7 21 copy in these mutants resembles human SCDO. Both of these diseases are caused by mutations 22 in Notch signaling components that we find are misexpressed in Fgf4 mutants, such as HES7,23 MESP2,and LFNG 19,20 . With regard to the human FGF pathway and SDV, a straightforward 1 gene-disorder relationship has not been uncovered, probably because of the pleiotropic 2 phenotypes of such putative mutations would preclude embryonic survival. Sparrow et al found 3 that gestational hypoxia in mice results in an increase in the severity and penetrance of CS in 4 Notch1, Mesp2, and Hes7 heterozygotes 25 . Intriguingly, CS is more severe in children living at 5 high altitudes 52 , suggesting a similar environmental effect may affect human development. In 6 mice, hypoxia was found to diminish FGF signaling components, but expression of Fgf8 was 7 unchanged, and Fgf4 was unexamined 25 . Our data suggest that FGF4 may be part of the system 8 that is responsive to this environmental insult. However, if this is the case, reduced FGF4 activity 9 cannot be the only response to reduced oxygen levels because gestational hypoxia reduces 10 expression of the FGF target gene Spry4 in the PSM, whereas we found no change in expression 11 in this or other such canonical FGF target genes in Fgf4 mutants. 12 13 Rather, as is the case for hypoxia-treated embryos 25 , Hes7 expression is reduced in Fgf4 mutants 14 and we propose that this leads to aberrant segmentation. An indispensable tool to this insight was 15 the use of multiplex HCR imaging [33][34][35] . The simultaneous fluorescent imaging of multiple 16 mRNA domains with this technique is particularly useful in a complex embryonic process, such 17 as somitogenesis, where many complex and dynamic gene expression patterns need to be 18 analyzed. HCR is a relatively new tool for mutant embryo analysis, just beginning to be used to 19 analyze mutant embryos 53-56 . We used HCR to examine the expression of multiple genes in a 20 single tissue by combining it with Imaris image analysis software. This combination of HCR and 21 Imaris modeling is broadly applicable for in situ quantification of gene expression within any 22 complex embryonic and clinical sample. With distinct molecular markers, one can use this 23 approach to quantify gene expression in any cell population with a precision that heretofore was 1 not possible and still retain the intact whole mount embryo or tissue. Here, we generated and 2 analyzed Imaris-based volumetric models of the PSM, based on Tbx6 expression, and determined 3 that Hes7 levels are reduced by about 40-50% specifically in the PSM of Fgf4mutants during 4 early somite stages when the progenitors of aberrant vertebrae are segmenting. 5 6 We propose that this reduction in Hes7 expression is the primary defect in Fgf4 mutants, and 7 subsequently causes an irregular activated Notch (NICD) pattern, misexpression of Mesp2, and 8 ultimately leads to vertebral defects. Given the molecular feedbacks during genetic oscillations 9 in the PSM, we considered other possible models, but they do not fit our data. For example, we 10 observed aberrant expression of Lfng, which encodes a glycosyltransferase that modulates Notch 11 signaling 57 . Although embryos completely lacking Lfng display aberrant Hes7 oscillations 32,58 , 12 overall Hes7 mRNA levels are not reduced 11 , unlike the case in our Fgf4 mutants. Another 13 factor required for Hes7 transcription is encoded by Tbx6 59,60 , but we found no significant 14 difference in expression of this gene. We support our proposal that the observed 50% reduction 15 in Hes7 is the cause of the Fgf4 mutant with the synergistic worsening of the vertebral 16 segmentation defect that occurs when we further reduce Hes7 levels by removal of one wildtype 17 allele. From another perspective, the mild CS caused by loss of one Hes7 allele in a wildtype 18 background 25 is worsened 9-fold by removing Fgf4 (Figure 7). Therefore, Fgf4 is a robustness-19 conferring gene that buffers somitogenesis against a perturbation in Hes7 gene dosage 61 . 20

21
We observed that in Fgf4 mutants Hes7 mRNA levels are reduced during rostral somitogenesis 22 at E8.5 (5-6 somite stages), but are unaffected at about E9.5 (24-26 somite stages) when the 23 embryo is beginning to generate correctly patterned somites that will differentiate into normally 1 patterned vertebrae. Tam showed that between these two stages of mouse development, somites 2 more than double in size 62 . Such an increase can be due to a slower segmentation clock and/or 3 faster regression of the determination front 63 . Gomez et al. found that the caudal movement of 4 the wavefront does not significantly change at these stages; therefore the clock must slow 5 between E8.5 and E9.5 64 . Consistent with a slowing clock, we observed at least 2 and frequently 6 3 bands of Hes7 expression at the Fgf4-sensitive stages (5-6 somite stages ; Figure 4, 5), but only 7 1 to 2 bands when Fgf4 loss has no effect on Hes7 expression (24-26 somite stages; Figure 5 -8 figure supplement 5). Thus, it appears that FGF4 activity is required to maintain Hes7 mRNA 9 levels above a certain threshold when the segmentation clock is faster; when Notch oscillations 10 slow, they may become FGF-independent, or other FGFs may be at play. If this is the case in all 11 vertebrates, we might expect embryos with faster segmentation clocks, such as snakes 64  Cephalochordata, the most basal living chordates 68 . In both Cephalochordates and Urochordates, 23 the FGF essential for embryonic axis extension is an Fgf8 ortholog 69,70 , suggesting that this Fgf 1 ("Fgf8/17/18") may be the ancestral gene in this process. However, Fgf8/17/18 activity in these 2 invertebrate chordates does not control gene oscillations as these embryos apparently lack a 3 segmentation clock 69,71 . Therefore, we speculate that the recruitment of an Embryos were then washed 4x 10 minutes in TS-PBS then soaked overnight in DAPI, as 19 described above in the HCR section, and embedded and cleared as described below. WISH and 20 skeletal staining were performed as previously described 75 . 21 22

Embedding and Clearing 1
Embedding: Stained embryos where mounted in coverslip bottomed dishes suspended in ultra-2 low gelling temperature agarose (Sigma, A5030) that had been cooled to room temperature. 3 Once correct positioning of embryos was achieved the dishes were moved to ice to complete 4 gelling. Scientific Corp) is increased to match the refractive index (RI) of the mounting solution to that of 8 standard microscopy oils (nD = 1.515; not utilized in this study) and 1-thioglycerol is omitted to 9 reduce toxic compounds within the solution (thereby making the solution safer to handle) and 10 increase shelf life. To account for dilution by water from a sample (including agarose volume) 11 we used Ce3D++, in which the iohexol concentration is further increased. Ce3D++ was designed 12 to produce desired RI after two incubations -detailed protocol is available upon request. The 13 protocol for preparing these solutions is as follows: in a 50 mL tube, add iohexol powder (20 g 14 for Ce3D or 20.83 g for Ce3D++), then 10.5 g of 40% v/v solution of N-Methylacetamide 15 (M26305, Sigma) in 1X PBS, and then 22.5 mg Triton-X-100 (T8787, Sigma) for a final 16 concentration of 0.1%. The solution is mixed overnight at 37 C on an orbital rocker with 17 intermittent vortexing then stored at room temperature. Embedded embryos were cleared using 2 18 changes of Ce3D++ solution while rocking at room temperature for a twenty-four-hour period. 19 20 Imaging 21 All images were obtained on an Olympus FV1000 confocal microscope, with an image size of 22 1620 x 1200 pixels and with a Kalman averaging of 3 frames. To capture the entire tissue a 10x 23 UPlanApo objective (NA= 0.4) was used achieving a pixel size of 0.9um x 0.9um. Tissues were 1 oriented in the same way with the anterior-posterior axis of the tissue oriented from left-right 2 within the center of the field. Microscope settings were kept consistent between imaging. 3 Intensity calibration and shading correction was performed as previously described 77 , however 4 fluorophore dye concentrations used for shading correction were 0.05mg/L for fluorescein 5 (Sigma 46960), 1mg/L for acid blue 9 (TCI CI42090), and 2mg/L for rose bengal (Sigma 6 198250). Dyes were placed in coverslip bottomed dishes. The Shading Correction plugin within 7 Fiji was then used with the median flat field images acquired using the dye solutions. 8 9 Image processing 10 Images within figures were processed using Fiji 78 and represent max projections of z-stacks. 11 Compared images are presented with identical intensity ranges for each channel. Orthogonal 12 projections were made using Imaris software (Imaris V9.2.1, Bitplane Inc). 13 14

Statistical analysis 15
For all analysis at least 3 embryos were used unless otherwise stated in the text or caption. 16 Significance was determined using a Students two-tailed t-test. 17 18

Imaris Fluorescence Quantification and Modeling 19
HCR data: Flat-fielded image stacks were imported from Fiji into Imaris (Imaris V9.2.1, 20 Bitplane Inc). A baseline subtraction was then performed for each probe, using a cutoff value 21 specific for each probe and fluorophore combination. For all probes, except Hes7, the baseline 22 cutoff was adjusted until signal from a tissue known not to express the gene was no longer 23 detectable (e.g. neural epithelium for Tbx6 40,49 . For Hes7, this cutoff was determined by using 1 embryos that were homozygous for a Hes7 null allele 47 that lacked sequences complementary to 2 the HCR probes; the cutoff was chosen that resulted in no signal in these mutants ( Figure 5 -3   figure supplement 5 ). 4 The Surface model tool was used to build surfaces for each expression domain to be quantified 5 except for Fgf8 and Hes7, which were quantified using a surface derived from the Tbx6 6 expression domain. Quantification of Spry2, Spry4 and Etv4 were determined by measurement of 7 fluorescence intensity per cubic micrometer within the volume the respective gene expression 8 domain. Surface models were generated using a surface detail value of 3uM, absolute intensity 9 setting, and an absolute intensity threshold cutoff that was set to exclude background signal in 10 tissues known not to express the gene being modeled. Once established for each probe, the same 11 intensity threshold cutoff values were used for generating surfaces for all embryos. For 12 generating Tbx6 surfaces, a range of absolute intensity threshold cutoffs were used to achieve a 13 final surface volume between 9.0e6 and 1.1e7 um3. Values for mRNA expression represent the 14 intensity sum within the volumetric model divided by the volume of the model. 15 Immunostaining data: In images where NICD was immunostained, the surface creation tool 16 was used. The mesoderm was selected, eliminating the ectodermal and endodermal tissue, on 17 alternating z-planes through the entirety of the z-stack. The resulting pattern was interpolated for 18 intervening planes. Values for NICD expression represent the intensity sum within the 19 volumetric model divided by the volume of the model. 20 Spot modeling: Spot diameter was set to 4uM with a point spread function value of 8uM, local 21 background subtraction was used, and sum intensity of the channel being modeled was used to 22 threshold. Threshold cutoff values for spot models were set at 1,500 for all figures except Figure  23 6. In Figure 6 the thresholds values (a.u.) for the sum intensity are: low (blue), 1,000-6,000; low-1 medium (green), 6,000-11,000; medium (yellow), 11,000-16,000; medium-high (red), 16,000-2 21,000; high (white), 21,000 or greater. "Heat maps" in Figures 4 and 7 were generated within 3 the Imaris software using a linear color scale; the scale in Figure 4 ranged from intensity values 4 of 7,000 (blue) to 20,000 (red), the scale in Figure 7 ranged from 1,000 (blue) to 15,000 (red). 5 6 7 Acknowledgements We thank E. Kamiya and W. Heinz of the NCI Optical Microscopy and 8 Image Analysis Lab, for their assistance with clearing using Ce3D+ and Imaris, respectively. We 9 are grateful for technical assistance from T.  predominantly as a heterodimer with BMP2 or BMP4 during mammalian embryogenesis. 34 Elife 8  Meox1 (control, n = 4; mutant, n = 4), Msgn1 (control, n = 9; mutant, n = 8), and Tbx6 (control, n 6 = 10; mutant, n = 8). E) There is no significant difference in the anterior-posterior length of Tbx6 7 and Msgn1 domains in control (n = 9) and mutant embryos (n = 8; green and red bars in C and 8 C'). F) Quantification of Tbx6 and Msgn1 mRNA expression, determined by measurement of 9 fluorescence intensity per cubic micrometer within the volume of the Msgn1 or Tbx6 domain of 10 expression, respectively. There is no significant difference between control and mutant embryos. 11 In E and F, data are mean±s.e.m, significance determined by a Student's t-test. Msgn1: control 12 n=7; mutant n = 8. Tbx6: control, n = 9; mutant, n = 8. All images: MIP, dorsal view, anterior 13 left. Same embryo shown in C-C'' and D-D''. 14 1 Figure 3. images as in (B,C) but visualized using the Imaris spot modeling function where each fluorescent 7 signal is represented as a colored sphere according to pixel intensity (lowest value is purple and 8 highest value is red, as indicated). Note that the pattern in the Fgf4 mutant (C') is less distinct 9 and blurred compared to littermate control (B'); brackets indicate anterior PSM. D-E'') HCR 10 staining of representative 5-6 somite stage control (n = 10) and Fgf4 mutant (n = 8) embryos for 11 the indicated genes. Note Fgf4 mutant expression of Hes7 and Lfng in mutants is abnormal. 12 All of phases in control and Fgf4 mutants shown in A-F; controls: phase I, n = 9; phase II, n = 10; 5 phase III, n = 11. Fgf4 mutants: phase I, n = 6; phase II, n = 9; phase III, n = 2. H) Quantification 6 of Hes7 expression specifically within the PSM (all phases combined) in 5-6 somite stage control 7 (n=11) and Fgf4 mutant (n=8) embryos shows significantly decreased expression in the Fgf4 8 mutant. Quantification of Hes7 mRNA expression, determined by measurement of fluorescence 9 intensity per cubic micrometer within the volume the mesoderm-specific Tbx6 expression 10 domain. Data in H are mean±s.e.m, significance determined by a Student's t-test. All images are 11 MIPs, dorsal view, anterior to left. 12 13 Linked to Figure  anterior to the left. G) Quantification of Hes7 expression within the PSM, reveals no difference 7 between controls and Fgf4 mutants. Measurement of Hes7 fluorescence intensity per cubic 8 micrometer within the volume the mesoderm-specific Tbx6 expression domain; data are mean ± 9 s.e.m, significance determined by a Student's t-test. Controls: phase I, n = 3; phase II, n = 3; 10 phase III, n = 4. Mutants: phase I, n = 4; phase II, n = 4; phase III, n = 3. 11 12 1 Figure 6. Reduced Hes7 expression with less distinct peaks and troughs in the Fgf4 mutant 2 PSM. 3 A, B) Spot models based on HCR analysis of Hes7 expression, in 5-6 somite stage control and 4 Fgf4 mutant embryos within the PSM, as defined by the Tbx6 expression domain. Localized 5 Hes7 expression is colored by level of expression from low (blue) to high expression (white), as 6 indicated. Note there are less high expressors (yellow, red and white) and more low expressors 7 (blue). This insight is quantified in C), where the percentage of spots found in each expression-8 level group is graphed. Data are mean±s.e.m, significance determined by a Student's t-test; 9 control, n = 6; mutant, n = 5. D-E') Composite of spot models from A and B. Note that the Hes7 10 expression trough (boxed region, expanded in D' and E') between anterior and posterior 11 oscillatory peaks is less distinct in mutant (E') with more higher expressors (green) than in 12 control embryos (D'). F-G') MIP of HCR staining of 5-6 somite stage control and Fgf4 mutant 13 embryos using probes against Tbx6, Hes7, and Hes7intron which specifically labels active Hes7 14 transcription. Note the trough of Hes7intron signal between peaks (brackets) in mutants is less in the key in A. Note that the vertebral defects in the Fgf8/Hes7 mutants are additive, not 7 synergistic. 8