The level of protein in the maternal murine diet modulates the facial appearance of the offspring via mTORC1 signaling

The development of craniofacial skeletal structures is fascinatingly complex and elucidation of the underlying mechanisms will not only provide novel scientific insights, but also help develop more effective clinical approaches to the treatment and/or prevention of the numerous congenital craniofacial malformations. To this end, we performed a genome-wide analysis of RNA transcription from non-coding regulatory elements by CAGE-sequencing of the facial mesenchyme of human embryos and cross-checked the active enhancers thus identified against genes, identified by GWAS for the normal range human facial appearance. Among the identified active cis-enhancers, several belonged to the components of the PI3/AKT/mTORC1/autophagy pathway. To assess the functional role of this pathway, we manipulated it both genetically and pharmacologically in mice and zebrafish. These experiments revealed that mTORC1 signaling modulates craniofacial shaping at the stage of skeletal mesenchymal condensations, with subsequent fine-tuning during clonal intercalation. This ability of mTORC1 pathway to modulate facial shaping, along with its evolutionary conservation and ability to sense external stimuli, in particular dietary amino acids, indicate that the mTORC1 pathway may play a role in facial phenotypic plasticity. Indeed, the level of protein in the diet of pregnant female mice influenced the activity of mTORC1 in fetal craniofacial structures and altered the size of skeletogenic clones, thus exerting an impact on the local geometry and craniofacial shaping. Overall, our findings indicate that the mTORC1 signaling pathway is involved in the effect of environmental conditions on the shaping of craniofacial structures.


Introduction
A key aspect of most social communication between humans is facial recognition 1 .Accordingly, congenital craniofacial malformations, including cleft palate, craniosynostosis, and craniofacial skeletal hypoplasia, which together account for more than one-third of all congenital birth defects, can have a profound in uence on social interactions 2,3 .From an evolutionary perspective, the viscerocranium harbors vital structures such as the feeding apparatus and supports sensory organs.Its precise sculpting and inheritable reproducibility are of unequivocal importance for survival.At the same time, adaptability of viscerocranium and particularly of the feeding apparatus to the environmental cues allows adjusting to environmental changes, e.g., feeding sources, or even concurring novel ecological niches.The latter type of adaptation is less common in mammals than in several classes of Gnathostomata, e.g., Actinopterygii 4 .
The craniofacial skeleton, one of the most complex and sophisticated part of the skeletal system, is composed of many different parts and formed via the interplay of a variety of genetic, epigenetic, and environmental factors [5][6][7] .Studies involving families with monozygotic and dizygotic twins indicate that the genetic inheritability score for craniofacial morphology in humans varies widely among different facial features, e.g., from 0.8 for the distance between the inner corners of the eyes to approximately 0.5 for the position of the point midway between the nose and upper lip, as well as for nasal protrusion [8][9][10][11] .
One well-known environmental in uence on facial morphogenesis in humans is alcohol consumption during pregnancy 12 .
In all Gnathostomata, including zebra sh, mice, and humans, the viscerocranium develops from descendants of the neural crest cells (NCCs), a transient and multipotent population of embryonic progenitors 13 .Multiple subpopulations of the mesenchyme derived from NCCs condense and then differentiate further into chondrocytes and osteoblasts, the major types of skeletal cells in the craniofacial region [14][15][16] .Other developmental subpopulations derived from NCCs, such as Schwann cell precursors, also contribute to the formation of chondrocytes and osteoblasts, but to a relatively limited extent 17 .The shapes of the craniofacial skeletal elements are determined primarily by the shape of these mesenchymal chondrogenic condensations during development, with subsequent ne-tuning by localized intercalation of new chondrogenic clones arising from the surrounding mesenchyme 18,19 .
A complex interplay between NC cells, the facial ectoderm, placodes, endoderm and neuroepithelium orchestrates accurate sculpturing of the viscerocranium.Not surprisingly, this process involves continuous changes in the expression of thousands of different genes 20 .Genome-wide association studies (GWAS) have implicated more than 100 loci in the formation of facial morphology within the normal range and more than two hundred single nucleotide polymorphisms (SNPs) that exert a signi cant impact on this formation [21][22][23] .In association with abnormal morphology of the human facial skeleton, the HPO database (http://hpo.jax.org)lists 1165 genes 24 .Although many of these are associated primarily with other systems, such as hematopoiesis and neurogenesis, their large number re ects the complexity of facial morphogenesis.
At the same time, the proper migration and differentiation of NCCs, as well as their interaction with surrounding tissues during facial development involves limited number of developmentally and evolutionarily conserved signaling pathways.Among those are hedgehog (HH), broblast growth factors (FGF), bone morphogenic proteins (BMPs), WNT, retinoic acid (RA) and platelet-derived growth factor (PDGF) pathways 25 .These pathways and associated morphogens form a hardware system, genetically responsible for sculpting of viscerocranium.However, the signaling pathways, capable to sense environmental clues and integrate these signals into genetical hardware of facial morphogenesis are rather unknown.One proposed system is HH signaling pathway, which may sense mechanical forces via cilium 26 and modify craniofacial formation as it has been shown in bony sh 27 .
Clearly, a major factors underlying natural selection has been the availability of nutrition 28 and the feeding apparatus, part of the viscerocranium, is of particular importance in this context.Nutritional sensing by the Mechanistic Target of Rapamycin Complex 1 (mTORC1) signaling pathway has been highly conserved evolutionarily 29 .Budding yeasts sense the availability of amino acids via mTORC1 and, in response to this information, shift towards the synthesis of proteins or autophagy 29 .Although this pathway plays a similar role in multicellular organisms 30,31 , in this case levels of oxygen, energy and growth factors (primarily those transducing via P13 kinase and Akt 30 ) also exert an in uence.
On the systemic level, mTORC1 can be regarded as belonging to an endocrine network of both up-and downstream of insulin-like growth factors (IGFs) that regulates a variety of processes in response to the availability of nutrition 30,32−34 .Changes in the activity of mTORC1 can alter the shape of craniofacial structures 35,36 and, in addition, the mTORC1 pathway interacts with the HH, BMPs, and Wnts [37][38][39] pathways, which are strongly involved in sculpting of the viscerocranium.Accordingly, we hypothesize that the mTORC1 signaling pathway may play a role in mediating interactions between certain environmental factors and the inherited program of craniofacial morphogenesis.

Results
The PI3K/mTORC1 pathway is associated with facial appearance in humans To identify enhancers actively involved in human facial development, embryonal facial material was CAGE-sequenced (Fig. 1A) and all enhancers actively transcribed during human facial development between week 3 and 12 of gestation were identi ed (GEO #xxx, supplementary metadata le).All thus identi ed enhancers were further cross-checked and enriched against enhancers previously identi ed in ENCODE project 40 (see the Methods).The resulting pool of enhancer coordinates was overlapped with published GWAS hits 22 (see the Methods) identi ed to be associated with normal-range facial morphology (Fig. 1A).Among enhancers thus identi ed there was a clear enrichment in components of the PI3K/mTORC1/autophagy pathway (Fig. 1B, C, Extended data Fig. 1).Predictions based on the STRING database (Fig. 1D) and the speci c facial phenotype related to each individual polymorphism (Fig. 1E) indicated that among the major enhancers active during sculpting of the human face, the mTORC1 pathway was clearly enriched.
Thus, this approach identi ed PI3K/mTORC1 pathway as a potentially important player in human facial morphogenesis.To explore the mechanism(s) underlying involvement of this pathway in craniofacial shaping, we manipulated the mTORC1 pathway during facial development in experimental animals.mTORC1 modulates the shaping of chondrogenic condensations in the mouse First, we activated the mTORC1 pathway in neural crest cells (NCCs) by crossing Tsc1 oxed mice with the Sox10 CreERT2 strain 41 , in which a pulse of tamoxifen on embryonic day 8.5 (E8.5) causes recombination in NCCs 15 .Reconstruction of the developing craniofacial structures in the offspring utilizing 3D µ-CT images with enhanced contrasting of soft tissues revealed alterations in the thickness of skeletal elements, as well as minor developmental abnormalities already on E17.5 (Fig. 2A-D).Overlay of the reconstructed cartilage of Tsc1 cKO and control (Tsc1 heterozygous) embryos revealed enlargement of a variety of elements of the craniofacial skeleton, as well as enhanced thickness of all components of the nasal cartilage (Fig. 2E, F).These observations con rmed the involvement of the mTORC1 pathway in craniofacial shaping 35,36 and, in addition, showed that this pathway is involved during early development.
Previously, we demonstrated that in mice craniofacial shape is established at the time of mesenchymal condensation (E12.5-E13.5),with subsequent ne-tuning via intercalation of new clones 18,19 .To reveal the shape of these mesenchymal condensations, KO embryos were stained for Sox9 on E12.5 and, although the overall shape was preserved, their nasal prominence and nasal capsule compartments were thicker (Fig. 2G-I).We then bred in the R26R Confetti reporter transgene, which allows clonal behavior to be assessed.Analysis of Sox10 CreERT2 ;Tsc1 ;R26R Confetti embryos on E17.5 (with previous pulsing on E8.5) revealed that in the absence of the Tsc1 gene, the clones of nasal chondrocytes appeared as bulky large clusters (Fig. 2J-O), with more extensive dispersion and misalignment (Fig. 2P-S ), compared to the individual columns observed in the Sox10 CreERT2 ;Tsc1 + ;R26R Confetti heterozygotes (Fig. 2J-S).
Thus, activation of the mTORC1 pathway in murine NCCs modulated both chondrogenic condensation and clonal arrangement.
Modulation of the activity of the mTORC1 pathway at the stage of intercalation in uences shaping of the craniofacial skeleton to a relatively minor extent To explore the in uence of the mTORC1 pathway during intercalation of new clones into existing mesenchymal condensations 19 , we injected tamoxifen on E12.5, the stage at which Sox10 CreERT2 targets perichondrial cells surrounding cartilage elements and Schwann cell precursors 42 .Surprisingly, ablation of Tsc1 at this developmental stage did not change the shape of the craniofacial skeleton (not shown) and increased clonal size only slightly (Fig. 3A-J).To further verify this observation, we targeted chondroprogenitors involved in early mesenchymal condensation employing Col2 CreERT mice coupled with both the Tsc1 oxed and R26R Confetti strains and pulsed with tamoxifen on E12.5.In line with the previous observation, activation of mTORC1 signaling by Col2 CreERT at this developmental stage did not alter the structure of the craniofacial skeleton and increased clonal size slightly (Fig. 3K-O).
At the same time, ablation of mTORC1 signaling in chondro-progenitors by crossing Raptor oxed mice with the Col2 CreERT and R26R Confetti strains and pulsing with tamoxifen on E12.5 augmented facial length on E17.5 (as detected by µCT, see Fig. 3P-R) without affecting any other skeletal parameters (Extended data Fig. 2A-C).Ablation of Raptor under these same conditions lowered the number of large clones somewhat and enhanced the number of cells that were single-labeled (Fig. 3S-Y).Successful manipulation of mTORC1 activity in these various strains was con rmed by assessment of S6 phosphorylation (Extended data Fig. 2D-E and data not shown).
These observations indicate that in mice the mTORC1 pathway is involved in craniofacial shaping predominantly prior to and/or during the stage at which chondrogenic condensations occur.
Inhibition of mTORC1 immediately prior to chondrogenic condensations alters the formation of cartilaginous facial structures in both mice and zebra sh To establish the stage of craniofacial skeletogenesis during which the role of mTORC1 signaling is most important, we inhibited mTORC1 with a single injection of rapamycin into pregnant animals on E10.5, when migration of cranial NCCs has been completed, but chondrogenic condensation has not yet begun 15,17 .This resulted in a slightly elongated snout in the embryos on E17.5 in comparison to the controls injected with DMSO (Fig. 4A-C).Moreover, the thickness of chondrogenic mesenchymal condensations on E12.5 was reduced (Fig. 4D-F).Clonal lineage tracing of chondro-progenitors in these same embryos beginning on E12.5 revealed disorganized clones, with relatively fewer elongated clones containing more than three chondrocytes and a relatively higher number of labeled cells that had not divided (i.e., in which recombination had occurred, but which did not proliferate during the period of tracing) (Fig. 4G-M).
To determine whether this in uence of mTORC1 signaling on mesenchymal condensation is conserved among species, we also studied zebra sh, in which shaping of the craniofacial skeleton also occurs via chondrogenic condensation and intercalation of chondro-progenitors into the primary cartilage anlagen 43 .In these animals the craniofacial skeleton begins to develop between 48 and 72 hours postfertilization (hpf) 44 and the rst Sox9-and Col2-positive cells appear at 48 and 53 hpf, respectively 17 .Col2a1aBAC:mcherry zebra sh larvae were exposed to rapamycin at various time-points in their development, washed free of this compound, and then allowed to develop until 120 hpf.Exposure prior to (14-22 hpf, 24-32 hpf) or during (32-56 hpf) chondrogenic condensation did not affect the overall size of the facial skeleton, but led to narrowing of cartilaginous structures (Fig. 4N-S).Interestingly, signi cant elongation of the face occurred when the larvae were exposed to rapamycin at 14-22 hpf or 32-48 hpf (Fig. 4R).Furthermore, exposure prior to chondrogenic condensation resulted in slight curvature of the ethmoid plate (ETH) and re-positioning of several other elements of the cartilage (MC, PQ and CH) (Fig. 4P,Q).
Altogether, these ndings indicate that in both mice and zebra sh the mTORC1 pathway modulates the shape of craniofacial structures by regulating the recruitment and clonal expansion of mesenchymal derivatives of neural crest cells.Interestingly, even transient inhibition of mTORC1 activity early during development altered the clonal behavior of NCC progeny, thereby leading to subsequent modulation of the shape of the craniofacial skeleton.

Dietary interventions that modulate mTORC1 activity also alter the skeletal structure by impairing clonal dynamics
The evolutionary conservation and mild variability described above indicate that mTORC1-dependent modulation of craniofacial structures, and particularly those of the feeding apparatus, may be an important adaptive mechanism.As also mentioned above, the activity of the mTORC1 pathway is regulated by nutritional status and, in particular, by dietary levels of amino acids, which act both directly at the cellular level through receptors for arginine and leucine and systemically via pathways involving growth hormone and insulin growth factors (IGFs), which are themselves also controlled by amino acids levels 30,32 .Accordingly, we examined whether alteration of mTORC1 activity through feeding diets containing different levels of protein to pregnant dams might modulate craniofacial shaping in the offspring.For this purpose, starting on E6.5 pregnant C57BL/6J mice consumed isocaloric diets containing either 20% protein (a level similar to that in standard mouse chow = the control), 4% (low) protein or 40% (high) protein, with subsequent analysis of at least 4 different litters of embryos from each group.
As expected, mTORC1 activity (as re ected in the level of pS6) was lowest in the control embryos and most pronounced in the high group (Extended data Fig. 3A,B), with no differences in body weight (Extended data Fig. 3C).µCT scans of embryos on E17.5 utilizing phosphotungstic acid (PTA) to augment contrast (Fig. 5A) revealed that both the length and width of the nasal capsule (Fig. 5B,C), as well as the length of the Meckel's cartilage (Fig. 5D) were all in uenced by the level of protein in the maternal diet.Thus, comparison of 3D segments of the chondrocranium cartilage showed that both the nasal capsule and mandible were slightly smaller in the embryos whose dams received 4% dietary protein (Fig. 5E).In addition, the thickness of the cartilage of the nasal capsule was elevated by the higher level of dietary protein (Fig. 5F).To con rm these observations, the same experiment was performed, but utilizing Hexabrix 320 for contrast in connection with the µCT, and similar changes in craniofacial structures were observed (Extended data Fig. 3D-M).Lowered proliferation of cells within skeletal elements was observed only in embryos whose dams received the lowest level of dietary protein (Extended data Fig. 3N-T) and there were no differences between the groups with respect to the extent of cell death within cartilaginous elements (Extended data Fig. 3Q-U).
When the level of protein in the diets was manipulated in this same manner in pregnant Sox10 CreERT2 ;Tsc1 dams (pulsed with tamoxifen on E8.5 and, accordingly, having embryos with constitutively active mTORC1 in all their NCCs-derived cells) and the craniofacial structures of these embryos analyzed on E17.5, again by µCT scans utilizing PTA for contrasting, no differences in any of craniofacial parameters were detected (Fig. 5G-J).Power analysis showed that at least 320 pups need to be analyzed to detect signi cant difference in the length of nasal capsule with a power of 0.95 and 55 pups for its width, which is beyond the feasibility in 3D reconstructions of PTA-enhanced embryos.Thus, the lack of any alteration in the craniofacial parameters examined (Fig. 5H-J), together with the changes in pS6 activity observed above (Extended data Fig. 3A,B), indicate that the alterations in craniofacial structures in response to the different levels of dietary protein were mediated by mTORC1 signaling.
Next, when the level of protein was manipulated in this same manner in the diets of Col2 CreERT2 ;R26R Confetti mice and these animals injected with tamoxifen on E12.5 and E13.5, both the low and high levels of protein caused remarkable disorganization of the clones within developing cartilaginous elements (Fig. 5K-L).This nding is in agreement with the conclusion above that both elevation and attenuation of mTORC1 activity disturbs clonal organization within developing cartilage.
Finally, incorporation of the average values obtained in mice with the low and high protein diets to a mathematical model of human skulls for visualization purpose indicated slight, but clear alterations in multiple elements of the craniofacial skeletons (Extended data Fig. 4, see the Methods for further details).
Altogether, these ndings indicate that the level of protein in the maternal murine diet during pregnancy in uences embryonic shaping of craniofacial cartilage by altering the activity of mTORC1, which in turn changes the clonal dynamics of neural crest progeny.

Discussion
Here, we have revealed cellular mechanisms underlying mTORC1-dependent shaping of elements of the craniofacial skeleton and demonstrated in both zebra sh and mice that this shaping occurs predominantly in association with mesenchymal chondrogenic condensations, with subsequent netuning to a lesser degree via intercalation.In addition, we have demonstrated that mTORC1 activity in embryos of these species is modulated by the level of protein in the maternal diet, with associated effects on the chondro-cranium and ne-tuning of the shape of the craniofacial skeleton.
In greater detail, we show here that alterations in the behavior of progeny of NCCs in uence skeletal shaping, both at the stage when chondrogenic mesenchymal condensations occur and when the 3D morphology of the cranial skeleton is ne-tuned via clonal intercalation.The nding that the shape of mesenchymal condensations largely determines the subsequent shape of cartilaginous and, later, bony structures 19 provides a link between the expansion of ectomesenchyme derived from murine NCCs lacking Tsc1 and resulting changes in craniofacial shape 35 .It is noteworthy that mTORC1 activity in uences the shaping of different chondrogenic mesenchymal condensations to different extents, with imperfect preservation of the rough 3D geometry of the entire chondrocranium.For example, constitutively active mTORC1 increases the thickness of the condensations underlying the nasal prominence and nasal capsule, while changing the patterning of the nasal septum to a much more limited degree.
Therefore, speci c mechanisms or processes appear to be localized within distinct regions of the chondrocranium.The potential underlying mechanism(s) may involve the known interactions between the mTORC1 pathway and the major morphogens involved in the shaping of craniofacial structures, including HHs (hedgehogs), FGFs ( broblast growth factors), BMPs (bone morphogenetic proteins), WNTs (Wingless/Integrated family of morphogens), RA (retinoic acid) and PDGFs (platelet-derived growth factors) 25,39,45−49 .For instance, ablation of mTOR speci cally in NCCs reduces the activities of the canonical Wnt and BMP pathways 36 .SHH, which is secreted in localized regions by the neuroepithelium and the brain, participates in shaping the anterior chondrocranium in a highly speci c manner, e.g., by inducing or permitting formation of the nasal septum 18 .At the same time, S6K1, a kinase downstream of mTORC, augments HH signaling by phosphorylating GLl1 39 .Thus, the differential chondrogenic activity of SHH, in combination with its functional interactions with the mTORC1 pathway, may contribute to the difference in the consequences of chondrogenic condensations at different locations in the developing face.
Furthermore, our present ndings indicate that the mTORC1 pathway in uences facial skeletal shaping by modulating the clonal expansion of committed chondro-progenitors.Previously, we reported that the growth of facial skeletal elements depends on intercalation of chondrocyte clones originating from committed chondro-progenitors within the perichondrium surrounding these elements and oriented transversally into pre-formed cartilage 19 .This intercalation and subsequent expansion of chondrogenic clones plays a key role in controlling the nal thickness and geometry of cartilaginous elements.Here, we show that manipulation of mTORC1 activity prior to formation of the perichondrium and committed chondro-progenitors alters the formation of these oriented clones later in development.With attenuated mTORC1 activity, the intercalated clones in the nasal cartilage of embryos were smaller, whereas elevation of mTORC1 activity in chondro-progenitors via deletion of the Tsc1 gene resulted in intercalation of bulky clonal clusters rather than individual clonal columns.Intercalation of these aberrant clones likely underline the altered length and thickness of the nasal cartilage, which eventually in uenced the overall craniofacial shape.Thus, during mesenchymal condensation mTORC1 activity regulates the overall geometry of facial cartilage, whereas with respect to committed chondro-progenitors this activity in uences individual cartilaginous elements.Therefore, modulation of mTORC1 activity at different timepoints may result in a spectrum of somewhat different craniofacial shapes, perhaps thereby also contributing to the variety of defects in patterning observed.
It is worth pointing out that the mTORC1 pathway is also involved in chondrogenesis in the limbs, with ablation of Raptor in the limb bud mesenchyme resulting in growth impairment 50 .However, modulation of mTORC1 activity in mature chondrocytes does not in uence limb growth 51,52 .These observations indicate that the appropriate level of mTORC1 activity in chondro-progenitors, rather than in mature chondrocytes, is important for skeletogenesis, in line with our present results.Since mTORC1 is primarily involved in adjusting cellular responses to the nutrition available 53 , either being enhanced directly by amino acids 54 or via insulin and insulin-like growth factors (all of which are tightly regulated by nutritional levels 32,33,55 ), it is not surprising that we found that modulation of the level of protein in the maternal diet regulates mTORC1 activity resulting in subtle, but distinct changes in the craniofacial shape of the embryos.Availability of nutrition is a major factor in connection with natural selection and such a spectrum of closely related craniofacial shapes may re ect adaptive phenotypic plasticity, and, accordingly, allow various feeding strategies.Phenotypic plasticity in the feeding apparatus of teleost sh has been observed, both in the wild 56 and under experimental conditions 4 .Recently, it has been reported that the HH pathway mediates plasticity of the feeding apparatus in response to the mechanical properties of the foraging species 27 , with mechanical sensing being, at least in theory, mediated by cilia, a mechanical sensor that is a key component of the HH pathway 26,57,58 .Thus, interactions between nutritional sensing by the mTORC1 pathway and mechanical sensing by the HH pathway may mediate phenotypic plasticity of the feeding apparatus in response to external conditions.Interestingly, mTORC1 is also involved in regulating the phenotypic plasticity of skeletal muscles 59 , as well as in long-term synaptic plasticity 60 .
In humans, craniofacial plasticity has been described in response to the consistency of the diet and alcohol consumption by the mother during pregnancy, as well as to the climate 12,61,62 .Thus, plasticity of the feeding apparatus, as well as of the entire facial skeleton, may be an evolutionarily conserved characteristic of all gnathostomes, including humans.On the basis of the ndings of others and the data documented here, we propose that the mTORC1 pathway is a key part of the molecular machinery that adapts craniofacial structures to nutritional conditions.
In summary, we have demonstrated here that the mTORC1 pathway modulates the embryonic shaping of craniofacial skeletal elements at the stage of chondrogenic condensations, with subsequent ne-tuning during intercalation of chondro-progenitors.Furthermore, we provide evidence for an impact of maternal protein intake during pregnancy on the shaping of fetal craniofacial cartilage.These ndings provide novel and important insights into the mechanisms underlying craniofacial shaping and, potentially, the phenotypic plasticity of this process as well and, in addition, help elucidate the role of material dietary protein during pregnancy in this context.

Human embryos
Human fetal tissue collection was reviewed and approved by the local ethics committee of Institute of Fundamental Medicine and Biology of Kazan Federal University (No. 8, May 2018).Written informed consent was obtained from the patients subjected to medical abortion.
To identify enhancers actually transcribed in human embryonic faces, human facial material was collected between weeks 3 and 12 of development, time-window that potentially in uence human facial individuality.Next, we performed CAGE-sequencing on embryonic human facial material and compared the transcriptional start sites, proximal promoters and distal transcribed enhancers 63 thus identi ed to loci indicated as being involved in human facial variability by genome-wide sequencing 22 (http://portaldev.sph.umich.edu/docs/api/v1/#introduction).
For our CAGE-sequencing, total RNA (2-3 mg) was extracted from the facial portion of human embryos, preserved in RNAlate and stored at -80 o C using the RNeasy Fibrous Tissue Kit (Qiagen, Hilden, Germany) in accordance with the manufacturer's protocol.The concentration and purity of extracted RNA were determined on the basis of absorption employing the NanoDrop™ 8000 Spectrophotometer (ThermoFisher Scienti c, Waltham, MA, USA) and quality veri ed with the Agilent Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA, USA).
Libraries were then prepared utilizing the standard nAnT-iCAGE (non-Ampli ed non-Tagging Illumina Cap Analysis of Gene Expression) protocol 64 , employing 2.5-3 µg total RNA as a template for synthesis of the rst cDNA strand (nAnT-iCAGE Library Preparation kit DNA form, Yokohama, Japan and SuperScript III Reverse Transcriptase, Invitrogen, Waltham, MA, USA).This cDNA was subsequently biotinylated at its 5´end (nAnT-iCAGE Library Preparation kit, DNA form, Yokohama, Japan), which allowed selection of the 5´cap containing molecules with streptavidin beads (Dynabeads M-270 Streptavidin, ThermoFisher Scienti c, USA).In this manner, rRNA, as well as truncated or not fully transcribed RNA was eliminated.
For more complete removal of RNA, the cDNA was treated with RNase I and H (nAnT-iCAGE Library Preparation kit, DNA form, Japan) and then puri ed using RNACleanUP (Beckman Coulter, Brea, CA, USA).Next, linkers were ligated to the 5′and 3′ ends (nAnT-iCAGE Library Preparation kit, DNAform, Japan) of the cap-trapped cDNA.The 5′-linker employed contained recognition sites for the XmaJI restriction endonuclease (nAnT-iCAGE Library Preparation kit, DNA form, Japan) and the MmeI class II restriction enzyme (nAnT-iCAGE Library Preparation kit, DNA form, Japan), as well as a barcode for multiplexing.The 3′-linker contained a recognition site an XbaI restriction enzyme (nAnT-iCAGE Library Preparation kit, DNAform, Japan).Treatment with these restriction enzymes yielded short CAGE tags to which a sequencing primer was ligated.
In the nal stage, a second cDNA strand was synthesized from these short CAGE tags??? (nAnT-iCAGE Library Preparation kit, DNAform, Japan).The concentration of the resulting libraries was determined by the PicoGreen Assay in a GloMax® Multi Detection System (Promega, Madison, WI, USA) and their quality assessed using an Agilent Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA, USA).Finally, the libraries were validated using real-time PCR (KAPA Library Quanti cation Kits Illumina, KAPA Biosystems, Wilmington, MA, South Africa) and sequencing on a HiSeq 2500 platform (Illumina, San Diego, CA, USA) using the HiSeq v4 reagent kit (HiSeq SR Cluster Kit v4 cBot and HiSeq SBS Kit v4 50 cycles, Illumina, San Diego, CA, USA) in the 50-bp single-end mode.
Single-read sequences were analyzed for quality and over-represented adapter sequences identi ed with the FastQC tool.Quality ltering trimming was performed with the fastx_trimmer (FASTX Toolkit 0.0.13.2) and Trimmomatic-0.39 and RNAdust 1.06 utilized as adapters and for removal of rRNA removal.Read mapping on human genome hg38 and mouse genome mm10 was performed with BWA-0.7.10, with unmapped reads being realigned using Hisat2-2.2.1.Aggregation of CAGE tag start sites (CTSS) for each sample, with subsequent peak clustering, were carried out employing the PromoterPipeline script from the C1 CAGE protocol 63 .Bidirectional enhancers were identi ed using the pipeline described by Andersson and colleagues (2014) 65 .The statistical signi cance of the differential expression of CAGE peaks was calculated using the edgeR package for R.
Triple overlap of GWAS-derived data of coordinates of face shape-affecting loci with already preidenti ed and annotated enhancers (from genome annotation 40 ; https://genome.ucsc.edu/cgibin/hgTrackUi?db=mm10&g=encodeCcreCombined;ENCODE Project Consortium) and our human CAGEseq-derived coordinates of active facial human embryonic enhancers was done involving a speci c prior ltration step such as: we selected polymorphisms falling within 5 kilo-base pairs distance from the CDS in any direction for GWAS-identi ed genes.

Mice
All animal experiments were pre-approved by the Stockholm North Ethical committee and performed in accordance with the guidelines of the Swedish Animal Agency.The Sox10-CreERT2, Col2-CreERT2, R26Confetti, Tsc1 ox, and Raptor ox strains of mice employed have been described in detail previously 41,66−69 .Embryonic Cre recombination was induced by intraperitoneal (i.p.) injection of 1-3 mg tamoxifen (Sigma) into each pregnant dam.The day on which the plug was detected was de ned as embryonic day 0.5 (E0.5).Rapamycin (0.02 mg, LC Laboratories) was injected i.p. into each pregnant dam.

Zebra sh
The Col2a1aBAC:mcherry strain of zebra sh was kindly provided by Prof. Chrissy Hammond (University of Bristol, UK) and has been utilized as described in detail elsewhere [70][71][72] .Zebra sh larvae were exposed to 400 nM rapamycin at the time-points indicated.

Manipulation of the maternal murine diet
Pregnant dams received standard mouse chow containing 4% protein until E6.5 and thereafter an isocaloric diet containing 4%, 20% or 40% protein (TD.93032, TD. 91352 and TD.90018 from Envigo) until the day of sacri ce.
X-ray computed microtomography (µCT) with enhanced contrast achieved with phosphotungstic acid (PTA) On E17.5, the heads of mouse embryos were placed in a 1% PTA/methanol solution to enhance the contrast of soft structures, as described previously 73 .µCT scans were performed with the GE phoenix v|tome|x L 240 system equipped with a nanofocus X-ray tube (180 kV/15 W maximal power) and high at panel (dynamic 41|100 with 4000 × 4000 pixels, each 100 × 100 µm in size).Acquisition involved the use of a 0.2-mm aluminum lter to soften the beam; 60 kV and 200 µA; exposure for 600 ms; and averaging of 3 projections to reduce noise.1800 images were acquired over 360°, requiring a scanning time of one hour per sample.The isotropic voxel size was 6.2 µm in all cases.The tomographic reconstructions were performed in the GE phoenix datos|x 2.0 3D computed tomography software.Segmentation of craniofacial structures was performed manually using a combination of the Avizo (Thermo Fisher Scienti c, USA) and VG Studio MAX 3.2 software (Volume Graphics GmbH, Germany), as described elsewhere 73 .
High-resolution microfocus computed tomography using enhanced contrast with Hexabrix (CE-HRµCT) and subsequent image processing and 3D analysis Following xation, samples were stored in PBS at 4° C. Prior to scanning, these samples were transferred to Eppendorf tubes containing 1.5 ml 30% Hexabrix 320 in PBS (Guerbet Nederland B.V.); incubated for two weeks with continuous gentle shaking at 4° C; and then scanned while still inside the same tubes.Hexabrix 320, a negatively charged ioxaglate, is repelled by the anionic sulfated-glycosaminoglycan (sGAG), resulting in negative staining of cartilage, while still providing good contrast between mineralized tissues and the background.
For acquisition of all images, the NanoTom M (GE Measurement and Control Solutions, Germany) system in combination with a diamond-coated tungsten target was employed with the following conditions: a 0.2-mm aluminum lter to soften the beam; 60 kV and 300 µA; exposure for 500 ms; and averaging of each sample individually and a skip of 0 ('fast scan mode').2400 images were acquired over 360°, requiring a scanning time of 20 minutes per sample.In all cases the isotropic voxel size was 5 µm.
Reconstruction was performed using the Phoenix datos|x CT software, applying a correction of 5 for beam hardening and a Gaussian lter (radius 3) to reduce noise.
The transaxial, coronal and sagittal cross-sections of each sample were visualized with the DataViewer (Bruker MicroCT, Belgium); while 3D visualization of the cartilage and mineralized tissue and quanti cation of their volumes were performed with the Mimics Innovation suite (Materialise NV, Belgium).Brie y, two threshold values were selected manually to distinguish between non-mineralized cartilage and mineralized tissues (i.e., mineralized cartilage and subchondral bone) and these thresholds then ne-tuned with dynamic region-growing and multi-slice edit.Using these adjusted threshold values, 3D models based on marching-cubes were generated and the volumes of mineralized tissue versus nonmineralized cartilage and the ratio between these volumes calculated.In addition, the length and width of the nasal capsule and the Meckel cartilage were measured.
TUNEL staining 30-µm tissue sections were treated with 10 µg/ml proteinase K (Ambion) for 40 minutes at 37° C before applying the TUNEL reaction mix (Roche Inc.) for 90 minutes.The cell nuclei were then counterstained with DAPI.
Staining with haematoxylin and eosin 15-µm frozen sections were stained with haematoxylin for 30 seconds and 0.02% eosin for 2 minutes.

Microscopy and image analysis
Images were acquired with a LSM710 confocal microscope.3D visualization and all quanti cation were performed utilizing the IMARIS (Bitplane) and ImageJ software.

Mathematical modelling of human craniofacial anatomy
To examine how the morphological changes observed in mice might be manifested in humans, we transformed polygon data on human craniofacial anatomy extracted from full-body MRI (BodyParts3D, Database Center for Life Science, Japan, Tokyo).The skull was divided into 53 high-resolution images of teeth, bones and ligaments and transformation carried out in Mathematica 11.0 (Wolfram Research, USA, Illinois) using custom-written code.The non-linear 3D transform was designed as a three-dimensional 'magnifying glass' (adopted from 74 ) with a radius of 3 cm and centered on the nasal cavity.The magni cation was adjusted so that the width of the nasal cavity increased in the same manner as observed in mice.The algorithm allowed us to selectively magnify de ned anatomical regions of the skull, while maintaining the rest of the craniofacial anatomy unchanged.Inhibition of mTORC1 at various stages of embryonic development modulates the shape of craniofacial structures in both mice and zebra sh.
Pregnant C57BL/6J dams were exposed to a single dose of either DMSO or rapamycin on E10.5 and stained with hematoxylin-eosin on E17.5.Representative images of the general appearance (A) or sagittal sections of their heads (B-C) are shown.Embryos were treated as in A, but stained for SOX9 on E12.5 to reveal chondrogenic condensations (D,E), the thickness of which in the three major nasal compartments was quanti ed (F).The same treatment as in A was applied to Col2 CreERT ;R26R Confetti mice pulsed with tamoxifen on E12.5 and E13.

Supplementary Files
This is a list of les associated with this preprint.Click to download.
thank Olga Kharchenko for the artwork included and Ostap Dregval for technical assistance.This study was supported by the Swedish Research Council (Projects 2020-02298 to A.S.C., 2018-02713 to I.A., 2022-00611 to K.F. and 2021-01805 to M.X.), an ALF-agreement (ALFGBG-966178 to A.S.C.) and the NovoNordisk Foundation (NNF21OC0070314 to A.S.C.).M.X. was supported by a long-term postdoctoral fellowship from the European Molecular Biology Organization and by Stiftelsen Frimurare Barnhuset i Stockholm.E.I. was supported by a grant from the Russian Basic Science Foundation (#19-29-04115 to A.S.C).M.T., T.Z. and J.K. acknowledge nancial support in the form of project CEITEC 2020 (LQ1601) from the Ministry of Education, Youth and Sports of the Czech Republic under the National Sustainability Program II and help from the CzechNanoLab Research Infrastructure supported by MEYS CR (LM2018110).M.T. was the recipient of a Ph.D. Talent Scholarship from the Brno City Municipality.

Figures
Figures

Figure 4
Figure 4 5 and clonal appearance analyzed on E17.5 (G-L) and clonal size quanti ed (M).(N-S) Col2a1aBAC:mcherry zebra sh larvae were exposed to DMSO or rapamycin at various periods in development.Ventral (N-O) and lateral (P-Q) views of the larvae exposed to DMSO (N,P) or rapamycin (O,Q) 14-22 hours post-fertilization (hpf) and imaged at 120 hpf are shown.The facial length (R) and width (P) of 120-hpf-old Col2a1aBAC:mcherry zebra sh larvae exposed to rapamycin during the intervals of time indicated were quanti ed.Means ± SD are presented, with individual values also indicated.The unpaired t-test was employed to compare the values in F and M, and one-way ANOVA followed by Dunnett's multiple comparisons test in R and S.following parameters indicated in A were measured: the thickness (1) and length (2) of the nasal capsule, the thickness (3) and length (4) of the entire chondrocranium, and the length of the right (5) and left (6) portions of the Meckel cartilage.The thickness (B) and length (C) of the nasal capsule and the average length of the Meckel cartilage (D) are shown.Overlayed 3D reconstructions (E) and the thickness of craniofacial structures (F) from embryos whose mothers consumed diets containing different levels of protein are presented for direct comparison.(G-J) Pregnant Sox10 CreERT2 ;Tsc1 / dams consumed isocaloric diets containing either 4% or 40% protein from E6.5 of pregnancy and Tsc1 was ablated in the neural crest cells of their embryos by pulsing with tamoxifen on E8.5.The skeleton in the heads of their embryos was reconstructed on day E17.5 employing µCT with PTA enhancement (G).Quanti cation of the width (H) and length (I) of the nasal capsule and the average length of the Meckel cartilage (J) are depicted.The same diets as in A-F were administered to pregnant Col2 CreERT ;R26R Confetti dams pulsed with tamoxifen on E12.5 and E13.5.Representative images of clonal appearance (K) and quanti cation of clonal size (L) in the entire nasal cartilage are shown.Means ± SD are presented, with individual values also indicated.One-way ANOVA followed by Tukey's multiple comparisons test was employed to compare the values in B-D and L, and the unpaired t-test in H-J.The white dashed lines in K outline the cartilage. FigureS1.pdfFigureS2.pdfFigureS3.pdfFigureS4.pdf