Hox gene expression determines cell fate of adult periosteal stem/progenitor cells

Hox genes are evolutionarily conserved transcription factors that during embryonic development function as master regulators of positional identity. In postnatal life, the function of Hox proteins is less clear: Hox genes are expressed during tissue repair, but in this context their function(s) are largely unknown. Here we show that Hox genes are expressed in periosteal stem/progenitor cells in a distribution similar to that during embryonic development. Using unbiased sequencing, we established that periosteal stem/progenitor cells from distinct anatomic sites within the skeleton significantly differ in their transcriptome, and that Hox expression status best defines these differences. Lastly, we provide evidence that Hox gene expression is one potential mechanism that maintains periosteal stem/progenitor cells in a more primitive, tripotent state, while suppression of Hox genes leads to fate changes with loss of tripotency. Together, our data describe an adult role of Hox genes other than positional identity, and the modulatory role of Hox genes in fate decisions may offer potential druggable targets for the treatment of fractures, non-unions and bone defects.

www.nature.com/scientificreports www.nature.com/scientificreports/ using siRNA and antisense oligonucleotides (ASOs) against the long noncoding RNAs Hotairm1 and Hottip, suppress Hox expression in Hox-positive periosteal stem/progenitor cell populations, and this Hox-suppression led to a transcriptional and phenotypic change suggestive of a reversal of lineage commitment. These data demonstrate that Hox proteins are developmentally conserved master regulators of stem/progenitor cell fate during wound healing.

Results
embryonic Hox gene signature is maintained in adult skeletal stem cells. "General Purpose" control genes, such as the Hox gene cluster, control the body plan of the embryo along the anterior-posterior axis (reviewed in 12 ). During this process, patterns of Hox gene activity assign each anatomic body part a segmental identity, which culminates in the creation of a complex tissue, organ or organism. While such "Bauplan" is essential during development, it becomes less clear why these control genes would be necessary during adulthood. The most likely function may be found during regeneration of an injured tissue. Here, stem cells, once activated, organize within the regenerate to restore form and function of the injured body part, and it is in this scenario that a body plan gene cluster may provide vital regulatory function. We hypothesized that Hox genes continue to function as "general purpose" genes far into adulthood, and in order to test this conserved function of the Hox gene cluster, we made use of the skeleton, a contiguous organ, spanning the entire body from cranial to caudal. The skeleton is one of the few adult tissues that regenerates instead of repair/scar 13 and contains skeletal progenitor cells that are located within distinct anatomic sites of the skeleton, such as the periosteum 10,[14][15][16] . First, we had to confirm that indeed Hox gene expression is conserved and present in adulthood. Periosteal stem/progenitor cells were harvested from four anatomic locations 4 , spanning the entire body, and were subjected to transcriptional profiling. RNAseq analysis revealed that embryonically Hox-negative periosteal stem/progenitor cells maintained their Hox-negative status into adulthood (Fig. 1A), while embryonically Hox-positive periosteal stem/progenitor cells continued to express Hox genes (Fig. 1A). qRT-PCR analysis confirmed that anterior Hox genes continued to be expressed in the hyoid, while posterior Hox genes, such as Hoxa11 and Hoxa13, were expressed in periosteal stem/progenitor cells originating from the tibia (Fig. 1B). These findings were confirmed using in situ hybridization, demonstrating expression of the anterior Hox gene, Hoxa2, in the cambial layer of the hyoid periosteum, while the posterior Hox gene, Hoxa11 and Hoxa13, were expressed within the tibial periosteum (Fig. 1C). Transcriptome analysis reveals difference between Hox-negative and Hox-positive periosteal stem/progenitor cells. Besides Hox gene cluster expression, periosteal stem/progenitor cells present another unique identifying signature: distinctive embryonic origins 4 . Most of the craniofacial skeleton, except for the parietal bone 17 , are derived from the neural crest 18 , while the entire axial and appendicular skeleton are derived from the mesoderm 19 (Fig. 2A). These different embryonic origins, superimposed with a distinct Hox gene expression pattern, allowed us to define four unique periosteal stem/progenitor cell populations, different in embryonic origin and Hox expression ( Fig. 2A). If in fact Hox gene expression imprints a certain positional identity to these periosteal stem/progenitor cells, then transcriptional profiling using RNAseq should be able to identify which distinct feature (embryonic origin or Hox expression) better describes these stem/progenitor cell populations. Hierarchical cluster analysis of the transcriptome of the frontal, parietal, hyoid and tibial periosteal stem/progenitor cells was carried out. Clustering revealed that transcriptional similarities assigned the periosteal cells into two clusters defined by their Hox expression status: Periosteal stem/progenitor cells without Hox gene expression (frontal and parietal) homogenously clustered together and separately from periosteal (D) MA plot comparing differential gene expression between neural crest and mesoderm derived SSCs, and between (E) Hox-positive and Hox-negative SSCs. (F,G) Integration of both RNAseq and ATACseq data sets reveals genes that were differentially regulated (differential expression by RNAseq and differential open chromatin by ATACseq). Again, Hox + vs. Hox − best described their differences, as 6.5% of the genes were differentially regulated in this comparison (RNAseq: p < 0.05, ATACseq: p < 0.05).
www.nature.com/scientificreports www.nature.com/scientificreports/ stem/progenitor cells with Hox gene expression (hyoid and tibia) (Fig. 2B). To further dissect the discrepancies/ uniformities between the four stem/progenitor cell origins, we built a principal component analysis (PCA) that plotted the different anatomic origins apart from each other (Fig. 2C). The separation of each cluster confirms that each anatomic site gives rise to a periosteal stem/progenitor cell with a unique transcriptional signature; however, as shown in the hierarchical cluster analysis, ultimately the respective Hox-positive and Hox-negative periosteal stem/progenitor cells are transcriptionally more similar. MA plots, describing the logarithmic differences between two cell populations, were then built in order to provide a visual representation of the transcriptional similarities/differences either between periosteal stem/progenitor cells from neural crest-(frontal and hyoid) and mesoderm-derived (parietal and tibia) skeletal elements (Fig. 2D) or between periosteal stem/progenitor cells with or without endogenous Hox gene expression (Fig. 2E). The MA plot distinguishing periosteal stem/ progenitor cells according to their embryonic origin revealed that periosteal stem/progenitor cells (SSCs) from NC-derived and MD-derived bones demonstrated significant differences in FPKM reads of 216 genes (Fig. 2D). However, when plotted as Hox-expressing and Hox-negative SSCs, 5,390 out of 17,569 genes measured by RNAseq revealed statistically different expression levels (Fig. 2E, red dots), indicating that if there is a functional difference between periosteal stem/progenitor cells from the craniofacial and appendicular skeleton, then this difference can be best described by the cells' Hox gene expression status. To further understand this difference in transcriptional signature, we performed ATACseq to identify genes that were not only differentially expressed but also differ in their chromatin accessibility. Integration of the RNAseq and ATACseq data allowed us to identify genes that are differentially regulated between the comparisons. In line with the RNAseq data, the integrated analysis revealed that only 79 genes were differently expressed in cells from the neural crest versus mesoderm (Fig. 2F). However, in the comparison of Hox-positive and Hox-negative SSCs, 1135 genes exhibited differential regulation (Fig. 2G), again supporting our hypothesis that the Hox gene expression status best differentiates SSCs from distinct anatomic skeletal regions.
Hox-negative and Hox-positive periosteal stem/progenitor cells respond differently to injury. We have previously shown that skeletal stem/progenitor cells reside within the periosteum of the tibia (appendicular skeleton) and the mandible (craniofacial skeleton), and contribute to repair and regeneration of each specific skeletal element after injury 4 . RNAseq data have now revealed that Hox-expressing periosteal stem/progenitor cells are transcriptionally different than Hox-negative periosteal stem/progenitor cells. We next sought to examine these cells in vivo and investigate whether the transcriptional difference results in a morphological change in an uninjured and injured state. Histologic evaluation of uninjured periostea from Hox-negative and Hox-positive bones revealed an identical anatomic composition with its characteristic two-layer structure ( Fig. 3A-H). In response to a scratch injury, Hox-negative periostea responded with a pure osteogenic reaction within the layer immediately adjacent to the cortical bone (  10 . Using their gating strategy, we observed significantly greater numbers of periosteal stem/progenitor cells in the Hox-positive hyoid and tibia (Fig. 4A,B). In addition, we surveyed other general surface marker combinations that have been used in the past to describe stem/progenitor cells in the skeletal system. FACS analyses revealed that the Hox-positive periosteum was enriched in Sca1-and CD146-expressing cells ( Fig. 4C-F), while the Hox-negative periosteum contained more CD166-positive cell than the Hox-positve tissue (Fig. 4G,H). Both Sca1 and CD-146 are associated with the more primitive skeletal/periosteal progenitor cell 20 , while CD166 is labeling cells that have progressed on the lineage tree towards an osteoprogenitor cell 20 . These data support the observation that the Hox-positive periosteum is comprised of more primitive cells with both chondrogenic and osteogenic potential, while the Hox-negative calvarium is more differentiated and thus less plastic in response to injury.

Hox gene expression imparts skeletal multi-lineage differentiation potential on postnatal periosteal stem/progenitor cells.
We previously demonstrated that craniofacial periosteal stem/progenitor cells exhibit superior osteogenic differentiation compared to appendicular periosteal stem/progenitor cells in an in vitro mineralization assay 4 . Having now established that adult periosteal stem/progenitor cells come in two transcriptional flavors, Hox-positive and Hox-negative, we utilized functional in vitro differentiation assays to test whether this previously described transcriptional difference results in a measurable functional difference in osteogenic, chondrogenic and adipogenic differentiation. Periosteal stem/progenitor cells were subjected to tri-lineage differentiation. Osteogenic differentiation resulted in confluent mineralization of the frontal and parietal periosteal stem/progenitor cells assays, while hyoid and tibial periosteal stem/progenitor cells showed significantly less mineralization (Fig. 5A). Chondrogenic differentiation and adipogenic differentiation exhibited the exact opposite differentiation pattern with significantly more chondrogenic and adipogenic differentiation of periosteal stem/progenitor cells derived from the hyoid and tibia (Fig. 5B,C). These differentiation data reveal a striking similarity to the previously shown transcriptional separation of Hox-positive and Hox-negative periosteal stem/progenitor cells. Here, Hox-positive periosteal stem/progenitor cells display a more chondrogenic and adipogenic phenotype, while Hox-negative periosteal stem/progenitor cells are more osteogenic. (2019) 9:5043 | https://doi.org/10.1038/s41598-019-41639-7 www.nature.com/scientificreports www.nature.com/scientificreports/ In order to test whether cell fate decisions of periosteal stem/progenitor cells during these differentiation assays are Hox-dependent, we utilized a knock-down approach targeting lncRNAs that regulate Hox expression of hyoid and tibial OPCs (Fig. 6A). Hotairm1, a lncRNA located on the noncoding strand and situated at the 3′ end of the HoxA cluster (close to Hoxa1 and Hoxa2), has been shown to regulate Hox expression at this 3′ end (reviewed in 21 ). We utilized an siRNA approach to silence the anterior Hoxa cluster and then assessed the effect of Hox repression on osteogenic, chondrogenic and adipogenic differentiation. siHOTAIRM1 resulted in a significantly lower expression of Hotairm1 in hyoid periosteal stem/progenitor cells (Fig. 6B). This led to a significant reduction of Hoxa2 expression in these cells, confirming the successful knockdown (Fig. 6B). If our hypothesis that Hox expression leads to decreased osteogenic and increased chondrogenic and adipogenic differentiation, is correct, then we should observe a reversal after Hox gene knockdown. Indeed, we detected an increase in osteogenic differentiation of hyoid periosteal stem/progenitor cells after Hoxa2 knockdown, as shown by a significant increase in Collagen type 1 expression. In contrast, chondrogenic differentiation, measured by Sox9 and Collagen type 2 expression, and adipogenic differentiation, measured by Ppar-gamma and Fabp4 expression, was significantly decreased after Hoxa2 knockdown (Fig. 6B).
Next, we sought to suppress Hoxa cluster expression at the 5′ end, and here antisense oligonucleotides (ASO) were used to knockdown the lncRNA Hottip. Hottip is expressed in posterior anatomic locations and has been shown to be expressed in a conserved pattern from development to adulthood, where it regulates the expression www.nature.com/scientificreports www.nature.com/scientificreports/ of the distal 5′ Hox cluster 22 . Using an ASO approach, we successfully achieved Hottip knockdown in tibial periosteal stem/progenitor cells, and in response, we observed a significant decrease in Hoxa11 and Hoxa13 expression levels (Fig. 6C). While our approach was unable to detect a significant difference in osteogenic differentiation on a transcriptional level, there was a significant increase of mineralization detected in a functional osteogenic differentiation assay. Hoxa11 and Hoxa13 suppression using Hottip ASOs also resulted in a significant reduction of Sox9 and Collagen type 2 expression, suggesting a reduction in chondrogenic potential. Finally, Ppar-gamma and Fabp4 expression was significantly decreased after Hottip ASO treatment, and the function readout of adipogenesis using oil-red-O staining revealed a decrease in adipogenesis after Hottip knockdown (Fig. 6C), confirming our hypothesis that Hox gene expression is intimately involved in adult periosteal stem/progenitor cell differentiation.

RNAseq and ATACseq confirm stemness of Hox-positive periosteal cells. These data suggest that
Hox gene expression imparts tri-lineage potential on the periosteal stem/progenitor cell, while loss/suppression of Hox expression leads to progression on the lineage tree towards a more committed osteochondroprogenitor cell, osteoprogenitor cell, or osteoblast. We returned to the RNAseq data to confirm this observation. In particular, we employed a gene ontology (GO) analysis to determine the distinct biological functions assigned to either Hox + or Hox − skeletal cells. Gene ontology categories that were enriched in each group revealed that Hox + cells share common GO categories such as "embryonic skeletal system morphogenesis", "embryonic skeletal system development", and "mesenchyme migration", suggesting a less committed, more primitive stem-like cell population (Fig. 7A). This is in contrast to Hox − cells, which show enrichment in GO categories such as "bone morphogenesis", "cell fate commitment", and "cell differentiation", indicating a more committed progenitor population (Fig. 7A). Gene set enrichment analysis demonstrates biological processes modulated by Hox gene expression. GSEA analysis of GO terms revealed that Hox gene expression regulates gene sets associated with www.nature.com/scientificreports www.nature.com/scientificreports/  www.nature.com/scientificreports www.nature.com/scientificreports/ stemness (Fig. 7B). Finally, we utilized ATACseq to quantify the accessibility of chromatin in Hox-positive and Hox-negative SSCs. Increased accessibility of chromatin is associated with stemness 23 , and our analysis revealed that Hox-positive periosteal stem/progenitor cells exhibited a larger area of open chromatin near transcriptional start sites compared to Hox-negative periosteal stem/progenitor cells (Fig. 7C).
Taken together, the current study provides the first experimental evidence that Hox gene expression in adult periosteal stem/progenitor cells imparts lifelong genetic regulation of tri-lineage differentiation potential of these cells. Notably, we provide proof that periosteal stem/progenitor cells are not just one cell population, equal in function and character, but rather a diverse population with distinct differentiation potential, likely responsible for the unique regenerative response seen in different skeletal elements.

Discussion
In comparison to other tetrapod vertebrates, humans possess limited regenerative capacity. While amphibians can replace entire limb segments 24,25 including the most complex musculoskeletal structures -mammals, including humans, can only partially regenerate injured bone segments, let alone restore entire appendages. Previous work has identified maintenance of Hox gene expression in adult mouse skeletal stem cells 4  www.nature.com/scientificreports www.nature.com/scientificreports/ adulthood 6,7 , but no studies have yet identified a functional role of Hox gene expression in adult skeletal stem/ progenitor cells specifically in the periosteum. The current study presents first evidence that Hox gene expression provides periosteal skeletal stem/progenitor cells with an anatomic signature, and at the same time imparts differentiation cues to these stem cells, which is a prerequisite for successful skeletal element regeneration.
Our previous work has shown that Hox gene expression of adult tibial skeletal stem cells confers upon them a sense of positional identity, which is unchanged when cells are placed into a new environment 4 , similar to experiments performed on embryonic tissues 27 . While this positional memory of the skeletal stem cell may be important in transplantation experiments and surgical procedures, it is not clear why skeletal stem/progenitor cells possess and maintain this identity into adulthood. Should we consider bone regeneration that occurs after fracture in a human or mouse as complex as limb regeneration in an axolotl after amputation -one that requires positional identity to undergo proper morphogenesis of specific bone or limb segments? What if organization of the early hematoma after fracture requires a "Bauplan" similar to the one established by the Hox expression patterning during embryonic development or in a blastema after amputation? Rux et al. recently published compelling data that Hox expression is limited to adult progenitor-enriched mesenchymal stem/stromal cells and is essential for proper differentiation during repair 26 . In response to injury, it is this cell population that is essential for successful regeneration 28 , but their initial prevalence is miniscule compared to the other cell types present in the hematoma. Communication between these few cells is likely impossible due to the physical distance between them, therefore each skeletal stem cell has to be equipped with an architectural plan outlining the final regenerative product. Similar to the developing embryo, the establishment of cellular and tissue compartments within an early regenerate occurs in an environment, where gradients of transcription factors cannot be established over many cell diameters, nor can extracellular signaling proteins/morphogens modulate positional identity of this scarce subset of cells 29,30 . Activation and maintenance of Hox gene expression in the skeletal stem cell pool is auto-and cross-regulated by products of Hox genes themselves, as well as through a group of proteins that modulate transcription and chromatin conformation, the Polycomb and Trithorax group. lncRNAs, such as HOTTIP and HOTAIRM1 can bind to these complexes and can thus modulate Hox gene expression 31 . We took advantage of this regulatory mechanism and manipulated Hox gene expression using siRNA and antisense oligonucleotides against these lncRNAs. Using this strategy, we gathered crucial information on the functional role of Hox gene expression in adult periosteal stem/progenitor cells. First, our in vitro and RNAseq data provide convincing data that the presence or absence of Hox expression delineates two very distinct populations of periosteal stem/ progenitor cells, varying not only in a vast number of transcriptional products, but also demonstrating unique cell fate decisions. While Hox-negative periosteal stem/progenitor cells favor osteogenesis, Hox-positive periosteal stem/progenitor cells primarily differentiate into cartilage and fat. When Hox expression was suppressed using siRNA or ASOs, we observed a reduction of the pro-chondrogenic and pro-adipogenic phenotype in the Hox-positive periosteal stem/progenitor cells population towards a more osteogenic phenotype. While these data only provide evidence that Hox paralogs impart differential functional information on periosteal stem/progenitor cells when Hox genes are ON or OFF, it remains unknown whether different Hox genes convey specific and unique patterning, repair and morphology instructions to the regenerating skeletal element. This may add critical complexity to the scientific journey towards mammalian limb regeneration and may revolutionize the procedure of bone grafting in orthopaedic surgery and dentistry. If the positional identity of a skeletal stem cell is essential for its regenerative capacity, then bone grafting procedures should take this positional information into account when the graft donor site is chosen.

Materials and Methods
Mice. All procedures were approved by the New York University Committee on Animal Research and were performed in accordance with the institutional guidelines and regulations. The studies were conducted on 12-week-old C57BL/6J male mice purchased from Jackson Laboratories (Bar Harbor, ME). periosteum dissection and RNA isolation. The periosteum from four different skeletal sites (F-frontal bone, H-hyoid bone, P-parietal bone, T-tibia), each representing a unique signature of embryonic Hox code (positive/negative) and embryonic origin (neural crest (NC)/mesoderm) were analyzed. After the soft tissues were carefully removed, the periosteum was collected with a Gracey curette (Hu Friedy, Chicago, IL) and stored in RNAlater (Qiagen, Hilden, Germany). The tissues were homogenized in Qiazol (Qiagen) with a mortar and pestle complemented with Qiashredder columns (Qiagen) then purified with the miRNeasy kit (Qiagen) following the manufacturer's instructions. The RNA purity was verified with Nanodrop (Thermo Fisher Scientific, Waltham, MA) and the integrity was evaluated with Bioanalyzer (Agilent Technologies, Santa Clara, CA). Only samples with 260/230 ratios superior to 2.0 and RIN superior to 8 were further analyzed. RNA sequencing. RNA sequencing was performed utilizing the high output, paired-end reads with the Illumina HiSeq. 2500 System. Bioinformatic analyses were performed with the Tophat (version 2.0.9) alignment program for reads mapping with two mismatches allowed. Cufflinks (version 2.2.0) was used to calculate FPKM values, and Htseq (version 0.6.1.p.1) was used to find the read counts for annotated genomic features. For the differential gene statistical analysis, DESeq. 2 R/Bioconductor package in the R statistical programming environment was used. The contrast groups for the comparison were: Frontal (F) (2 replicates) versus hyoid (H) (2 replicates), parietal (P) (2 replicates) versus tibia (T) (2 replicates), neural crest (ND) (F + H, 4 replicates) versus mesoderm (MD) (P + T, 4 replicates) and Hox-negative (F + P, 4 replicates) versus Hox-positive (H + T, 4 replicates). Gene set enrichment analysis was performed using the GSEA software (http://www.broadinstitute.org/ gsea/index.jsp) on log2 expression data of periosteal cells from the four bones aforementioned and classified in the corresponding classes. Gene sets were taken from the Molecular Signatures Database (http://www.broadinstitute.org/gsea/msigdb/index.jsp). In particular, we investigated whether periosteum from each source was (2019) 9:5043 | https://doi.org/10.1038/s41598-019-41639-7 www.nature.com/scientificreports www.nature.com/scientificreports/ associated with over-or under-represented genes in pairwise comparisons between each class and the rest. Gene sets were permuted 1000 times, the normalized enrichment score (NES) was calculated for each gene set, the nominal P value was obtained and the false discovery rate (FDR) was calculated to correct the P value.
Gene ontology (GO) analysis was performed using the Database for Annotation, Visualization, and Integrated Discovery (DAVID; http://david.abcc.ncifcrf.gov/). Genes that were significantly enriched (p ≤ 0.05; ≥2. 33-fold) in Hox + vs. Hox − periosteal cells (1401 genes) or Hox − vs. Hox + periosteal cells (2163 genes) were used to generate GO terms corresponding to biological processes (with GOTERM_BP_DIRECT for functional annotation). Those significantly associated (p ≤ 0.05 or −log 10 (p) ≥ 1.3) with the gene list are ranked by p-value. AtACseq analysis. ATACseq differential chromatin detection analysis was performed for three lanes of a paired-end 50 Illumina HiSeq. 2500 run. Per-read per-sample FASTQ files were generated using the bcl2fastq Conversion software (v1.8.4) to convert per-cycle BCL base call files outputted by the sequencing instrument into the FASTQ format. The alignment program, Bowtie2 (v2.3.4.1), was used for mapping reads of 18 mouse samples to the mouse reference genome mm10 and the application Sambamba (v0.6.7) was utilized to remove duplicate reads. The algorithm, MACS (in Python v2.7.3), was used to call peaks of signal for annotated genomic features and, similarly, the Python package NucleoATAC was functioned to call nucleosome positions. The computeMatrix and plotProfile tools in the deepTools suite (v2.3.3) were utilized for generation of signal profile plots. For the differential peak statistical comparisons between six groups of samples with three replicates each, the DiffBind package (Bioconductor v3.3.0) in the R statistical programming environment was utilized. Venn diagrams of differentially detected annotated genomic features comparing the ATACseq sample datasets with the corresponding sample sets from a related RNAseq experiment were generated using the Venny web application (v2.1).
Flow cytometry. Tibial, hyoid, frontal, parietal bones were harvested and periosteal cells were isolated as Quantitative Rt-pCR. RNA was isolated from frontal, parietal, hyoid bones and the tibia as described and then reverse transcribed into cDNA with the Omniscript RT kit (Qiagen). The cDNA was amplified for specific targets using specific primers (listed in Table 1) and RT2 SYBR Green ROX PCR Master Mix in a QuantStudio3 www.nature.com/scientificreports www.nature.com/scientificreports/ Real-Time PCR System (Thermo Fisher Scientific). Results are presented as 2 −ΔΔCt values normalized to the expression of 18S or beta-actin and negative control samples. All reactions were performed in triplicate; means and standard deviations were calculated in GraphPad Prism 7 software. periosteal injury. In order to compare the periosteal reaction of upon injury amongst the different bones, we surgically performed a periosteal scratch on the frontal, parietal, hyoid bones and the tibia. After anesthesia was induced with Isoflurane inhalation (1-5%), 3 mm incisions through the skin were created on the head, neck and shin then the periosteum from the aforementioned bones was scratched with the tip of a 27-gauge syringe needle. The skin was closed and the bones were allowed to heal for 7 days. Non-injured bones were used as control.
Histology and immunofluorescence antibody staining. Frontal, parietal, hyoid bones and the tibia with and without periosteal injury were harvested and fixed in 4% paraformaldehyde (n = 5). For bright field microscopy images, the samples were decalcified in 19% ethylenediaminetetraacetic acid (EDTA) for 3 weeks, paraffin-embedded and 10-μm thick sections were stained with Movat's pentachrome 32 . The sections were examined and photographed using a Leica digital imaging system. For immunofluorescence, samples were decalcified in 19% EDTA for 48 hours then, after cryoprotection in 30% sucrose, embedded in Tissue-Tek OCT media (Sakura Finetek, Torrance, CA) and cryosectioned at 100 μm thickness. Sections were incubated overnight at 4 °C in OSX antibody (rabbit anti-mouse A-13, 1:100, Santa Cruz, Santa Cruz, CA) and SOX9 antibody (goat anti-human, 1:20, R&D). Next, sections were stained with DAPI (Thermo Fisher Scientific), coverslipped with Fluoromount (Thermo Fisher Scientific), examined and photographed using a Zeiss LSM 710 laser scanning confocal microscope.
In situ hybridization. In situ hybridization was performed using the RNAscope Probe-Mm-Hoxa2, RNAscope Probe-Mm-Hoxa11 and made-to-order RNAscope Probe-Mm-Hoxa13 (Advanced Cell Diagnostics, Newark, CA). Positive (cyclophylin B) and negative antisense controls were performed on additional samples. Hybridized probes were detected using a manual singleplex RNAscope 2.0 HD brown kit (Advanced Cell Diagnostics) following the manufacturer's instructions. The sections were examined and photographed using a Leica digital imaging system. periosteal cell isolation. Primary SSCs were obtained from frontal, parietal, hyoid bones and the tibia.
After careful dissection, bones with intact periosteum were submitted to 4 serial collagenase digestions in 0.2% collagenase type 2 (Thermo Fisher Scientific) in DMEM at 37 °C for 20 minutes with gentle rocking. After each of the first three digestions, bones were subjected to light centrifugation (1000 rpm) for 5 min and then transferred to a fresh tube of collagenase. After the last digestion, bones were centrifuged at 1400 rpm for 5 min and the pelleted cells were resuspended in growth media. Selective isolation of periosteal stem/progenitor cells was confirmed using FACS analysis.
In vitro differentiation assays. Periosteal progenitor cells from frontal, parietal, hyoid and tibia were isolated as described above and submitted to tri-lineage differentiation. For osteogenic differentiation, cells were cultured in DMEM containing 10% FBS, 100 μg/ml ascorbic acid 10 mM ß-glycerophosphate and 1% penicillin/ streptomycin, which was replaced every 3 days. After 21 days, cultures were stained with alizarin red and in vitro mineralization was quantified as previously described 33 . For chondrogenic differentiation, micro masses were cultured in chondrogenic differentiation media, supplemented with TGF β-3 (Lonza, Basel, Switzerland). After 14 days, the micro masses were fixed in 4% PFA, photographed under polarized light microscope (Leica) and paraffin embedded. Ten μm-thick sections were obtained and stained with alcian blue and hematoxylin. For adipogenic differentiation, SSCs were cultured in adipogenic differentiation media (Lonza). After 14 days, cells were stained with Oil Red-O and the number of positive cells per well was counted by an examiner blinded to the groups.

RNA interference. Primary hyoid SSCs were transfected with commercially available Lincode Mouse
Hotairm1 SMARTpool siRNAs targeting Hotairm1 with the target sequence UGGUUUACAUGACUAA (GE Dharmacon, Lafayette, CO) and primary tibial SSCs were transfected with Hottip antisense oligonucleotides (ASOs). The sequence for the Hottip ASOs was TAGTGCTTCTAAAACG. A non-targeting ASO control with the sequence AACACGTCATATACGC was used. RNAi Max lipofectamine (Thermo Fisher Scientific) was used as transfection reagent as per manufacturer's instructions. 24 hours after the transfection, the transfection media was replaced by regular growth media, osteogenic, chondrogenic or adipogenic differentiation media as described. Total RNA was harvested 48 h later using RNeasy Mini Kit (Qiagen). The knockdown of Hotairm1 in the hyoid SSCs, Hottip in tibial SSCs, as well as the downstream effect on the expression of specific Hox genes and tri-lineage differentiation markers were assessed by qPCR as described using specific primers (listed in Table 1). statistical analysis. A priori power analysis to obtain statistical significance (p = 0.05, power 80%) resulted in an n of 5 for each group, expecting a 25% difference between the two groups. Prism 7 (GraphPad Software, Inc., La Jolla, CA) was used for statistical computations. A Student's t test was used for all comparisons in which there were two groups; ANOVA analyses followed by the Holms-Sidak correction for post-hoc testing was applied for analyses in which there were two or more comparisons being made. Error bars represent standard deviation. An asterisk symbol (*) denotes a p value of less than 0.05.

Data Availability
The data that support the findings of this study are available from the corresponding author upon request.