1 Notch Effector Hes1 Marks an Early Perichondrial Population of Skeletal Progenitor Cells At

Home , DLX5

bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

1 Notch effector Hes1 marks an early perichondrial population of skeletal progenitor cells at

2 the onset of endochondral bone development

4 Yuki Matsushita1, Mizuki Nagata1, Joshua D. Welch2, Sunny Y. Wong3, Wanida Ono1 and

5 Noriaki Ono1

7 1. University of Michigan School of Dentistry, Ann Arbor, MI

8 2. University of Michigan Medical School, Computational Medicine, Ann Arbor, MI

9 3. University of Michigan Medical School, Department of Dermatology, Ann Arbor, MI

11 *Correspondence: Noriaki Ono, [email protected]

15 Summary: 150 words

16 Main text: 5,291 words

17 Methods: 2,046 words

18 Main Figures: 5

19 Extended Data Figures: 11

20 References: 42

21 Summary

23 The perichondrium, a fibrous tissue surrounding the fetal cartilage, is an essential

24 component of developing endochondral bones that provides a source of skeletal progenitor cells.

25 However, perichondrial cells remain poorly characterized due to lack of knowledge on their

26 cellular diversity and subset-specific mouse genetics tools. Single cell RNA-seq analyses reveal

27 a contiguous nature of the fetal chondrocyte-perichondrial cell lineage that shares an overlapping

28 set of marker genes. Subsequent cell-lineage analyses using multiple creER lines active in fetal

29 perichondrial cells – Hes1-creER, Dlx5-creER – and chondrocytes – Fgfr3-creER – illustrate

30 their distinctive contribution to endochondral bone development; postnatally, these cells

31 contribute to the functionally distinct bone marrow stromal compartments. Particularly, Notch

32 effector Hes1-creER marks an early skeletal progenitor cell population in the primordium that

33 robustly populates multiple skeletal compartments. These findings support the concept that

34 perichondrial cells participate in endochondral bone development through a distinct route, by

35 providing a complementary source of skeletal progenitor cells.

2 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

36 Introduction

38 The perichondrium is a poorly characterized fibrous tissue surrounding the fetal cartilage

39 despite its essential roles in endochondral bone development – a highly organized process in

40 which initial cartilage templates are replaced by bone and its marrow space. This process starts

41 when chondrocytes and their surrounding perichondrial cells are defined from undifferentiated

42 mesenchymal cells in condensations owing to actions of the transcription factor Sox91. As

43 chondrocytes organize the growth plate in their avascular environment, surrounding

44 perichondrial cells develop as distinct cell types in their vascular-rich environment. The

45 perichondrium provides two important functions; first, it provides cues for adjacent chondrocytes

46 in the cartilage template2, and second, it provides a putative source of progenitor cells supporting

47 skeletal development. Cellular heterogeneity and fates of perichondrial cells are not well defined.

48 Initially, the perichondrium is formed as elongated fibroblastic layers without differential

49 morphological characteristics. When chondrocytes within the cartilage template undergo

50 hypertrophy, the adjacent perichondrium becomes differentiated and forms the osteogenic

51 perichondrium, partly due to actions of indian hedgehog (Ihh) released from pre-hypertrophic

52 chondrocytes3-5. The osteogenic perichondrium is where the first osteoblast precursor cells

53 expressing osterix (Osx) are formed; these cells can translocate into the marrow space and

54 become osteoblasts6. However, these cells eventually disappear from the marrow space in the

55 later stage of development5,7. The perichondrium persists well into the postnatal stage as long as

56 the growth plate maintains its structure. Cathepsin K (Ctsk)-expressing cells in the perichondrial

57 groove of Ranvier provide a source for other perichondrial and periosteal cells, which include a

58 population of skeletal stem cells (SSCs) contributing to bone fracture healing8-10. In contrast, the

59 cartilage template provides a principal cellular source of bone marrow stromal cells (BMSCs)

60 including mesenchymal progenitor populations11; particularly, these precursor cells express

61 Sox91,12. However, how early perichondrial cells participate in endochondral bone development

62 remains largely unclear.

63 Analyses of chondrocytes in early skeletal development have been facilitated by a well-

64 defined set of tamoxifen-inducible mouse genetics tools, including Col2a1-creER13, Sox9-

65 creER14 and Acan-creER15; however, these lines can simultaneously mark perichondrial cells, at

66 least to some extent12. Studies of perichondrial cells have been largely hampered due to lack of

67 their subset-specific marker genes and mouse genetics tools; particularly, existing inducible

68 genetic tools that are highly active in perichondrial cells, such as Prrx1-creER16 and Osx-creER6,

69 simultaneously mark chondrocytes within the cartilage template. Therefore, a perichondrial cell-

70 specific mouse genetics tool would be essential to investigate the function of the perichondrium

71 in skeletal development and regeneration.

72 In this study, we aimed to achieve two major goals. First, we aimed to define the cellular

73 diversity of perichondrial cells, chondrocytes and their precursors of the pre-cartilaginous

74 condensation and the cartilage template by single cell RNA-seq analyses. Second, by learning

75 from novel marker genes and their expression patterns identified by single cell RNA-seq

76 analyses, we sought to characterize previously unreported tamoxifen-inducible creER lines to

77 identify potential cell-type specific inducible mouse genetics tools for fetal perichondrial cells

78 and chondrocytes. We successfully characterized three new creER lines – Hes1-creER, Dlx5-

79 creER and Fgfr3-creER – that might be instrumental for designing future studies. Our descriptive

80 findings illustrate that perichondrial cells and chondrocytes collaborate in endochondral bone

4 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

81 development, supporting the notion that these two types of cells are important contributors to the

82 postnatal bone marrow stromal compartment.

84 Results

86 1. Undifferentiated Hes1-creER+ mesenchymal cells provide precursors for skeletal

87 progenitor cells within the cartilage template and the perichondrium of femurs

89 Endochondral bone development starts when undifferentiated mesenchymal cells make

90 condensations, particularly around embryonic day (E) 10.5 in the mouse limb bud. These

91 condensing mesenchymal cells express Sox9 and activate the transcription machinery for

92 chondrocyte differentiation17. To define cellular heterogeneity of these undifferentiated

93 mesenchymal cells in the condensation, we first undertook a single cell RNA-seq analysis.

94 Mesenchymal cells contributing to the skeletal element of the limb are unanimously marked by a

95 cre recombinase driven by a 2.4kb Prrx1 promoter/enhancer18 (Fig.1a). We dissociated limb

96 mesenchymal cells from Prrx1-cre; R26RtdTomato mice at embryonic day 11.5 (E11.5), and

97 isolated tdTomato+ cells by fluorescence-activated cell sorting (FACS) (Supplementary Fig.1a).

98 We profiled 4,804 cells tdTomato+ cells using the 10X Chromium Single-Cell Gene Expression

99 Solution platform. A graph-based clustering analysis using Seurat19 revealed 9 clusters, including

100 two clusters of cells abundant in Sox9 (Cluster 2,7, dotted contour) and six clusters of their

101 surrounding cells (Cluster 0,1,3-6) (Fig.1b). The latter clusters were composed of mesenchymal

102 cells expressing unique homeobox proteins, such as Msx1 (Cluster 5,6), Lhx9 (Cluster 5), Irx3/5

103 (Cluster 3,7), Meox2 (Cluster 0) and Emx2 (Cluster 4) (Supplementary Fig.1b,c). In addition, Shh

104 and Fgf18 were expressed in cells in Cluster 6 and Cluster 0,1, respectively. Therefore, the limb

105 bud skeletal element is constituted by distinct groups of mesenchymal cells potentially serving as

106 precursors for chondrocytes and perichondrial cells.

107 In search for a cell-type specific marker for undifferentiated mesenchymal cells not yet

108 committed to the chondrogenic pathway, we noticed that a Notch effector gene Hes1 exhibited a

109 pattern overlapping but negatively correlated with Sox9, identified as cell-type specific markers

110 for Cluster 3,5,6 (Fig.1b, right panels). An RNAscope analysis revealed that Hes1 was expressed

111 in a manner surrounding the condensation expressing Sox9 (Supplementary Fig.1d). To further

112 validate Notch-responsive cells in situ, we took advantage of a Notch signaling reporter strain

113 expressing a histone 2B (H2B)-bound Venus protein under a C promoter binding factor 1

114 (CBF1) promoter (CBF1:H2B-Venus)20. Interestingly, Notch-responsive Venusbright cells were

115 mostly located outside the SOX9+ domain of the condensation (Fig.1c). The discrepancy

116 between Hes1 expression pattern and CBF1:H2B-Venus reporter activities might indicate that

117 Hes1+ cells might not be entirely Notch-responsive.

118 To study cell fates of these Hes1+ undifferentiated mesenchymal cells, we performed

119 lineage-tracing experiments using a Hes1-creER knock-in allele21 that activates an R26R-

120 tdTomato reporter in a tamoxifen-dependent manner. First, we pulsed Hes1-creER; R26RtdTomato

121 mice at E10.5 and analyzed these mice after 24 hours at E11.5, to define the characteristics of

122 Hes1-creER+ cells (Hes1CE-E10.5 cells). Hes1CE-E10.5 cells were located throughout the limb

123 bud except in SOX9+ pre-cartilaginous condensations, particularly in the area immediately

124 adjacent to SOX9+ cells with abundant CBF1:H2B-Venus reporter activities (Fig.1d,e). Hes1CE-

125 E10.5 cells expressed Hes1, but not Sox9, mRNAs at E11.5 (Supplementary Fig.2a). We also

126 noticed that Hes1-creER simultaneously marked other types of cells outside the skeletal element

6 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

127 of the limb bud, including MYH3+ skeletal muscle cells and EMCN+ endothelial cells

128 (Supplementary Fig.2b). Taken together, Hes1-creER can mark mesenchymal cells in the SOX9-

129 negative domain of the condensation upon tamoxifen injection.

130 Subsequently, we traced the fate of Hes1+ cells in vivo during endochondral bone

131 development. After 3 days of chase at E13.5, Hes1CE-E10.5 cells contributed to both

132 chondrocytes and perichondrial cells of the cartilage template; these cells became SOX9+

133 chondrocytes by translocating into the cartilage template, and a majority of perichondrial cells

134 including those expressing CBF1:H2B-Venus (Fig.1f). In fact, Hes1CE-E10.5 cells contributed to

135 all the layers of the perichondrium including OSX+ osteoblast precursors, in addition to MYH3+

136 skeletal muscle cells outside the perichondrium and EMCN+ endothelial cells (Fig.1g,

137 Supplementary Fig.2c). After 9 days of chase at P0, Hes1CE-E10.5 cells contributed to a number

138 of columnar chondrocytes of the growth plate, Col1a1(2.3kb)-GFP+ osteoblasts on the cortical

139 and trabecular bone and stromal cells throughout the marrow space (Supplementary Fig.2d).

140 Postnatally, Hes1CE-E10.5 cells contributed to the entire spectrum of the bone marrow stromal

141 compartment; these cells became growth plate chondrocytes, osteoblasts both in the cortical and

142 trabecular bone, and Cxcl12-GFPhigh stromal cells throughout the marrow space both in the

143 metaphysis and the diaphysis (Fig.1h, Supplementary Fig.2e,f). Skeletal muscle cells or

144 endothelial cells that were initially marked by Hes1-creER at E10.5 did not contribute to these

145 skeletal lineage cells, as cells marked by skeletal muscle-specific cre lines (Acta1-cre, Myl1-cre

146 and Mck-cre) or an endothelial cell-specific cre line (Tie2-cre) did not contribute to

147 chondrocytes, osteoblasts or bone marrow stromal cells at any time points (Supplementary

148 Fig.3a,b).

149 Therefore, Hes1-creER+ mesenchymal cells of the condensation can behave as primordial

150 skeletal progenitor cells, which robustly contribute to chondrocytes in the cartilage template,

151 osteoblasts of the primary ossification center and the bone collar, and marrow stromal cells

152 during endochondral bone development (Fig.1i).

153

154 2. Hes1-creER marks self-renewing osteo-stromal progenitor cells in the fetal femur

155

156 The perichondrium is defined when the fetal growth plate is established upon initial

157 formation of hypertrophic chondrocytes22. Perichondrial cells are composed of multiple layers of

158 mesenchymal cells with diverse functions; particularly, located on the innermost layer of the

159 perichondrium adjacent to the prehypertrophic and hypertrophic layer is the osteogenic

160 perichondrium, where cells become committed to the osteoblast lineage. First, we set out to

161 define cellular heterogeneity of the fetal perichondrium by a single cell RNA-seq analysis. Both

162 perichondrial cells and chondrocytes at the fetal stage are marked by a cre recombinase driven

163 by a Col2a1 promoter12 (Fig.2a). We dissociated limb mesenchymal cells of Col2a1-cre;

164 R26RtdTomato mice at E13.5, isolated tdTomato+ cells by FACS (Supplementary Fig.4a) and

165 profiled 7,932 tdTomato+ cells using the 10X Chromium Single-Cell Gene Expression Solution

166 platform. A graph-based clustering analysis revealed 11 clusters, including three clusters of

167 chondrocytes abundant in Sox9 (Cluster 0,1,3) and three clusters of perichondrial cells abundant

168 in Hes1 (Cluster 2,4,8) (Fig.2b). Hes1 is most abundantly expressed in the fetal perichondrium23.

169 In addition, Notch signaling is essential for maintaining skeletal progenitor cells by repressing

170 Runx2 transcription activities24. Among the chondrocyte clusters, cells in Cluster 1 were

171 enriched for Col2a1, Fgfr3, Acan, Ihh as well as Runx2 and Sp7, representing relatively more

172 mature chondrocytes (Fig.2b, Supplementary Fig.4b,c). Among the perichondrial cell clusters,

173 cells in Cluster 4 were enriched for Dlx5, Runx2, Prrx1 and Sp7, representing perichondrial cells

174 partially committed to the osteoblast lineage (Fig.2b, Supplementary Fig.4b,c). Other

175 perichondrial cells in Cluster 2 and 8 were enriched for Igf1 and Acta2, respectively

176 (Supplementary Fig.4b,c). Therefore, this analysis revealed the contiguous nature of the fetal

177 chondrocyte-perichondrial cell lineage that shares an overlapping set of marker genes.

178 We further constructed single-cell trajectories of tdTomato+ cells using Monocle 225,26.

179 This unsupervised analysis predicted a branched trajectory starting from Prrx1+ cells and ending

180 either at Ihh+ pre-hypertrophic chondrocytes or Acta2+ perichondrial cells, indicating that

181 chondrocytes and perichondrial cells represent discrete cell fates of Prrx1+ cells in the

182 primordium (Supplementary Fig.4d). We subsequently analyzed the branch point using branched

183 expression analysis modeling (BEAM) and heatmap-based visualization of representative

184 branch-dependent genes, to define genes involved with two different cell fate choices

185 (Supplementary Fig.4d). Hierarchical clustering revealed that a chondrocyte fate and a

186 perichondrial cell fate represent distinct transcriptomic paths; a chondrocyte fate was associated

187 with sequential upregulation of Bmp227, Sox928 and Runx229, whereas a perichondrial cell fate

188 was associated with upregulation of a Wnt-target gene, Axin230 (Supplementary Fig.4e).

189 Therefore, these trajectory analyses support the notion that perichondrial cells represent a

190 discrete state emanating as a cell fate alternative to chondrocytes.

191 We noticed that the expression pattern of Hes1 was negatively correlated with that of

192 Fgfr3. To reveal the relationship between Hes1 and Fgfr3, we performed a MAGIC imputation

193 analysis31. As expected, cells expressing Fgfr3 expressed Sox9 at a unanimously high level

194 (Supplementary Fig.4f). In contrast, Hes1 expression demonstrated an overall negative

195 correlation with Fgfr3 expression. In fact, an RNAscope analysis confirmed that Hes1 was

196 expressed in the perichondrium in a manner distinct from Fgfr3, which was predominantly

197 expressed within the cartilage template (Supplementary Fig.5a,b).

198 Further analysis with CBF1:H2B-Venus mice revealed that Notch-responsive Venusbright

199 cells were located outside SOX9+ cells in a pattern surrounding the fetal cartilage at E13.5

200 (Fig.2c). We also validated in situ expression patterns of Fgfr3 by taking advantage of an Fgfr3-

201 GFP bacterial artificial chromosome (BAC) transgenic strain. Fgfr3-GFP+ cells were found

202 within the SOX9+ domain of the fetal cartilage template, in a pattern distinct from CBF1:H2B-

203 Venusbright cells (Supplementary Fig.5b). Fgfr3-GFP+ cells expressed Fgfr3 mRNA

204 (Supplementary Fig.5b). Therefore, Notch effector gene Hes1 is expressed in a pattern largely

205 distinct from that of Fgfr3.

206

207 Subsequently, to delineate cell fates of perichondrial cell populations, we took advantage

208 of tamoxifen-inducible creER lines – Hes1-creER, Dlx5-creER32 and Osx-creER6 – whose

209 expression is regulated by cell-type specific marker genes identified by the above-mentioned

210 single cell RNA-seq analysis. We also took advantage of an Fgfr3-creER P1-derived artificial

211 chromosome (PAC) transgenic line33 to contrast their cell fates with those of cells within the

212 cartilage template. First, we pulsed these mice carrying R26RtdTomato and Col1a1(2.3kb)-GFP at

213 E12.5, and analyzed these mice after 24 hours at E13.5, to define the characteristics of Hes1-

214 creER+, Dlx5-creER+, Osx-creER+ and Fgfr3-creER+ cells at this stage (Hes1CE-E12.5, Dlx5CE-

215 E12.5, OsxCE-E12.5 and Fgfr3CE-E12.5 cells, respectively). Hes1CE-E12.5 cells were located

216 outside the SOX9+ domain of the cartilage template with minimal overlap with Col1a1(2.3kb)-

217 GFP+ osteoblasts in the perichondrium (Fig.2d). Some of these cells also expressed CBF1:H2B-

218 Venus (Supplementary Fig.5a). Hes1CE-E12.5 cells were also found outside the skeletal element

219 among MYH3+ skeletal muscle cells and EMCN+ endothelial cells (Supplementary Fig.5c).

220 Dlx5CE-E12.5 cells occupied the domain of the perichondrium equivalent to Hes1CE-E12.5 cells,

221 but in a more specific pattern without any overlap with MYH3+ skeletal muscle cells or EMCN+

222 endothelial cells (Fig.2d, Supplementary Fig.5d). In contrast, OsxCE-E12.5 cells occupied the

223 perichondrium in a pattern distinct from Dlx5CE-E12.5 cells, with the majority of these cells

224 either overlapping or adjoining with Col1a1(2.3kb)-GFP+ osteoblasts (Fig.2d). A substantial

225 number of OsxCE-E12.5 cells were also located within the cartilage as reported previously6.

226 Fgfr3CE-E12.5 cells were predominantly located within the cartilage with minimal overlap with

227 Col1a1(2.3kb)-GFP+ perichondrial osteoblasts; a small number of Fgfr3CE-E12.5 cells were also

228 found in the perichondrium (Fig.2d).

229 Quantification of tdTomato+ perichondrial cells in each model revealed that Hes1-creER

230 and Dlx5-creER could mark undifferentiated precursor populations of perichondrial cells. The

231 number of Hes1CE-E12.5 and Dlx5CE-E12.5 cells in the perichondrium was comparable at E13.5

232 (Fig.2e, top left panel). In addition, Hes1CE-E12.5 and Dlx5CE-E12.5 perichondrial cells were

233 distinct from perichondrial osteoblasts, as virtually no fraction of these cells was Col1a1(2.3kb)-

234 GFP+ (Fig.2e, bottom left panel). Importantly, Hes1+ descendants continued to express Hes1 in

235 the perichondrium, as the relative fraction of Hes1CE-E10.5 and Hes1CE-E12.5 cells in the

236 perichondrium remained comparable at E13.5 (Fig.2e, top right panel). Further quantification of

237 tdTomato+SOX9+ cells revealed that Hes1CE-E12.5 and Dlx5CE-E12.5 cells were distinct from

238 chondrocytes (Fig.2e, bottom right panel). Therefore, Hes1-creER and Dlx5-creER can mark

239 undifferentiated perichondrial cells that are largely distinct from chondrocytes and perichondrial

240 cells marked by Osx-creER or Fgfr3-creER at this stage.

241 CBF1 (encoded by Rbpj) and its transcriptional target Hes1 regulate differentiation and

242 hypertrophy of chondrocytes34,35. Analysis of CBF1:H2B-Venus mice at a later stage of E15.5

243 and P5 revealed that Venusbright cells were also present in proliferating and articular

244 chondrocytes; however, Hes1-creER did not mark any chondrocyte within the growth plate at

245 E15.5 and P5 upon a pulse at E14.5 and P3, respectively, (Supplementary Fig.5e,f), revealing the

246 dissociation between CBF1:H2B-Venus and Hes1-creER activities at a later stage.

247

248 Subsequently, we traced the fate of these distinct populations of perichondrial cells

249 during formation of the primary ossification center and the marrow space. After 3 days of chase

250 at E15.5, Hes1CE-E12.5 cells, Dlx5CE-E12.5 cells and OsxCE-E12.5 cells contributed to

251 mesenchymal cells within the primary ossification center as well as Col1a1(2.3kb)-GFP+

252 osteoblasts in the bone collar (Supplementary Fig.6a). Fgfr3CE-E12.5 cells expanded within the

253 cartilage template as SOX9+ chondrocytes, while exuberantly translocating into the nascent

254 primary ossification center (Supplementary Fig.6a). After 6 days of chase at E18.5, these cells –

255 Hes1CE-E12.5, Dlx5CE-E12.5, OsxCE-E12.5 and Fgfr3CE-E12.5 cells – contributed to

256 Col1a1(2.3kb)-GFP+ osteoblasts in the trabecular bone (Fig.2f, Supplementary Fig.6b) and

257 Cxcl12-GFPhigh marrow stromal cells in the marrow space (Supplementary Fig.6c). Importantly,

258 unlike those pulsed at an earlier stage, Hes1CE-E12.5 cells did not contribute to SOX9+

259 chondrocytes in the growth plate (Fig.2f, Supplementary Fig.6b). Therefore, mesenchymal cells

260 within the primary ossification center might have dual origins in the perichondrium and the

261 cartilage template.

262 Additionally, we noticed a notable difference with regard to how these cells contributed

263 to formation of the perichondrium, the periosteum and the bone collar. Fgfr3CE-E12.5 cells did

264 not contribute to the perichondrium, the bone collar or the cortical bone at the subsequent stages

265 (Fig.2f, Supplementary Fig.6b). OsxCE-E12.5 cells did not remain in the perichondrium as

266 reported previously5, therefore not robustly contributing to the bone collar or the cortical bone at

267 the subsequent stages (Fig.2f, Supplementary Fig.6b). In contrast, Dlx5CE-E12.5 cells maintained

268 themselves within the perichondrium including the groove of Ranvier (Fig.2f, Supplementary

269 Fig.6b, arrowheads), and robustly contributed to the bone collar and the cortical bone. Hes1CE-

270 E12.5 cells demonstrated a pattern similar to Dlx5CE-E12.5 cells without notable contribution to

271 growth plate chondrocytes (Fig.2f, Supplementary Fig.6b). Therefore, Hes1-creER and Dlx5-

272 creER, but not Osx-creER or Fgfr3-creER, can mark a self-renewing population of fetal

273 perichondrial cells with robust capability to contribute to cortical and trabecular bone

274 compartments. Although whether these cell populations are fully overlapping cannot be formally

275 demonstrated, it is important to note that Hes1-creER+ and Dlx5-creER+ cells behave in a similar

276 manner.

277 To clarify the uniqueness of the Fgfr3-creER line that predominantly marks chondrocyte-

278 like cells within the cartilage template, we also analyzed a well-characterized Col2a1-creER

279 line13. Analysis of Col2a1-creER; R26RtdTomato mice at E13.5 after 24 hours of pulse at E12.5

280 revealed that Col2a1CE-E12.5 cells were present not only in the cartilage template but also in the

281 perichondrium (Supplementary Fig.6d). After 2 weeks of chase at P7, Col2a1CE-E12.5 cells

282 contributed to essentially all perichondrial cells and growth plate chondrocytes, whereas Fgfr3CE-

283 E12.5 cells contributed to virtually no perichondrial cells and to a certain fraction of growth plate

284 chondrocytes (Supplementary Fig.6d). Therefore, Fgfr3-creER marks a specific subset of cells

285 that are marked by Col2a1-creER.

286

287 We further set out to define the functional significance of these cell populations in

288 endochondral bone development. For this purpose, we performed inducible cell ablation

289 experiments using a Rosa26-iDTA (inducible diphtheria toxin fragment A) allele, which was

290 crossed with aforementioned creER lines. This iDTA-mediated approach can only partially

291 ablate target cells of interest upon tamoxifen injection36,37. We pulsed Hes1-creER, Osx-creER or

292 Fgfr3-creER; Rosa26lsl-tdTomato/+ (Cont) and Hes1-creER, Osx-creER or Fgfr3-creER; Rosa26lsl-

293 tdTomato/iDTA (DTA) mice at E12.5, and analyzed these mice at 6 days later at E18.5. All of these

294 cells – Hes1CE-E12.5, OsxCE-E12.5 and Fgfr3CE-E12.5 cells – were functionally significant

295 contributors to bone development, as the femur length was significantly reduced in all groups

296 owing to iDTA-mediated cell ablation (Fig.2g, left panel, Supplementary Fig.6e). Notably,

297 ablation of Hes1+ cells selectively impaired the formation of bone marrow without affecting the

298 cartilage, as the bone marrow length, but not the cartilage length, was significantly reduced in the

299 Hes1CE-E12.5 DTA group (Fig.2g, middle and right panel).

300 Therefore, all of these cells – Hes1-creER+, Dlx5-creER+, Osx-creER+ and Fgfr3-creER+

301 cells – provide the functionally significant pathway to generate skeletal lineage cells within the

302 nascent marrow space (Fig.2h).

303

304 3. Hes1-creER+ and Dlx5-creER+ perichondrial cells substantially contribute to the

305 postnatal bone marrow stromal compartment of femurs

306

307 We further set out to define how Hes1-creER+ and Dlx5-creER+ cells contribute to the

308 postnatal bone marrow stromal compartment. For this purpose, we chased Hes1CE-E12.5,

309 Dlx5CE-E12.5, OsxCE-E12.5 and Fgfr3CE-E12.5 until postnatal day (P) 21. Hes1CE-E12.5 and

310 Dlx5CE-E12.5 cells predominantly contributed to the diaphyseal bone marrow stromal

311 compartment; these cells became Col1a1(2.3kb)-GFP+ osteoblasts both in the cortical and

312 trabecular bone and Cxcl12-GFPhigh stromal cells (Fig.3a,b,e,f, Supplementary Fig.7a,b). In

313 contrast, OsxCE-E12.5 cells contributed to osteoblasts and marrow stromal cells in a more limited

314 domain of the diaphyseal bone marrow, particularly on the proximal end of the marrow space

315 (Fig.3c,g, Supplementary Fig.7c, ref.5,7,38). Fgfr3CE-E12.5 cells became cortical and trabecular

316 osteoblasts as well as Cxcl12-GFPhigh stromal cells predominantly in the metaphyseal marrow

317 space, in a mutually exclusive manner to aforementioned cell types (Fig.3d,h, Supplementary

318 Fig.7d). Interestingly, Hes1CE-E14.5 cells contributed to only a small fraction of osteoblasts or

319 Cxcl12-GFPhigh stromal cells; instead, Hes1-creER marked cells of the endothelial lineage in the

320 nascent marrow space and contributed to a large number of sinusoidal endothelial cells in the

321 postnatal marrow space (Supplementary Fig.8).

322 To assess quantitatively how perichondrial cells and chondrocytes contribute

323 differentially to Cxcl12-GFPhigh stromal cells at P21, we performed flow cytometry analysis of

324 cells isolated from femurs of triple transgenic mice carrying a Cxcl12-GFP reporter (Fig.3i,

325 Supplementary Fig.9a). First, undifferentiated Hes1-creER+ mesenchymal cells in the

326 condensation demonstrated a robust potential as stromal progenitor cells, as Hes1CE-E10.5 cells

327 contributed to a substantial fraction of Cxcl12-GFPhigh stromal cells (34.0±4.0%). Second, Hes1-

328 creER and Dlx5-creER could mark comparable perichondrial progenitor cell populations at

329 E12.5, as Hes1CE-E12.5 and Dlx5CE-E12.5 cells contributed to an equivalent fraction of Cxcl12-

330 GFPhigh stromal cells (17.6±2.9% and 18.6±4.6%, respectively). Third, Hes1-creER+ and Dlx5-

331 creER+ cells, but not Osx-creER+ cells, represent a substantial source of bone marrow stromal

332 cells, as OsxCE-E12.5 cells contributed to only a small fraction of Cxcl12-GFPhigh stromal cells

333 (1.0±0.4%). Fourth, fetal Fgfr3-creER+ cells may provide a primary source of bone marrow

334 stromal cells and osteoblasts, as Fgfr3CE-E12.5 cells contributed to a higher fraction of Cxcl12-

335 GFPhigh stromal cells (26.4±3.7%) than Hes1CE-E12.5 or Dlx5CE-E12.5 cells did. Additional

336 analysis of transgenic mice carrying a Col1(2.3kb)-GFP reporter revealed that Fgfr3CE-E12.5

337 cells contributed to a significantly higher fraction of Col1(2.3kb)-GFP+ cells (19.6±3.0%) than

338 Hes1CE-E12.5 cells did (5.5±1.7%) (Supplementary Fig.9b). Lastly, Hes1-creER+ cells may

339 contribute to bone marrow stromal cells in a time-specific fashion, as Hes1CE-E14.5 cells

340 contributed to a much less fraction of Cxcl12-GFPhigh stromal cells than Hes1CE-E12.5 cells did

341 (5.8±0.2%).

342 To interrogate the characteristics of Hes1-creER+ cell-derived bone marrow stromal cells,

343 we further performed an integrative single cell RNA-seq analysis combining two datasets

344 (Cxcl12-GFP+||Hes1CEE12.5-tdTomato+ and Cxcl12-GFP+||Fgfr3CEE12.5-tdTomato+) at P21

345 using LIGER (Fig.3j). Mesenchymal GFP+||tdTomato+ cells clustered into several groups,

346 including four groups of Cxcl12+ reticular cells (Cluster 0,2,4,5), Alpl+ pre-osteoblasts (Cluster

347 9) and Ifitm5+ mature osteoblasts (Cluster 13) (Fig.3k, Supplementary Fig.10a-c). Interestingly,

348 cells in Cluster 9 also expressed Itgav (encoding CD51) and Cd200 (CD200), but not Thy1

349 (CD90) or Eng (CD105), indicating that these cells might coincide with previously reported

350 mouse skeletal stem cells (mSSCs)39. We further subclustered tdTomato+ cells for the following

351 analysis. Hes1CE-E12.5 and Fgfr3CE-E12.5 cells were distributed across all seven clusters

352 including Cluster 9 that was enriched for mSSC-like cells. Flow cytometry analysis revealed that

353 a majority of both Cxcl12-GFP+Hes1CE-E12.5 and Cxcl12-GFP+Fgfr3CE-E12.5 cells expressed

354 CD200, but not CD90 (Supplementary Fig.10e). We noticed that Fgfr3CE-E12.5 cells were

355 relatively overrepresented in a mature osteoblast cluster (Cluster 13, Ifitm5+Bglap+) (Fig.3l),

356 suggesting that Fgfr3+ chondrocyte-like cells might preferentially contribute to mature

357 osteoblasts within the stromal compartment. Furthermore, Fgfr3CE-E12.5 cells exhibited higher

358 expression of chondrocyte signature genes, including Col2a1, Col10a1, Acan, Matn3 as well as

359 osteoblast signature genes, including Prrx1 and Sp7, both in terms of the percentage of cells

360 expressing given genes and the expression level, compared with Hes1CE-E12.5 cells

361 (Supplementary Fig.10d). In contrast, Hes1CE-E12.5 cells exhibited higher expression of

362 adipocyte signature genes including Fabp4 and Cebpb (Supplementary Fig.10d). Therefore,

363 BMSCs originating from two different sources of the perichondrium and the cartilage template

364 might maintain at least some of the unique gene expression signatures after translocating into the

365 stromal compartment.

366 Taken together, these findings reveal that Hes1-creER+, Dlx5-creER+ and Fgfr3-creER+

367 cells, but not Osx-creER+ cells, in the fetal stage contribute to a substantial fraction of the bone

368 marrow stromal compartment during the postnatal stage.

369

370 4. Hes1-creER+ perichondrial cell-derived marrow stromal cells persist in the diaphysis

371 and participate locally in osteogenesis of femurs

372

373 Lastly, we set out to determine the functional significance of Hes1-creER+ cell-derived

374 cells in the bone marrow stromal compartment. To this end, we first performed CFU-F assays of

375 Hes1CE-E12.5, Fgfr3CE-E12.5 and Osx CE-E12.5 cells. At an early postnatal stage of P7, Hes1CE-

376 E12.5 cells contributed to a higher faction of CFU-Fs than Fgfr3CE-E12.5 cells did; importantly,

377 OsxCE-E12.5 cells contributed to no CFU-F (Hes1CE-E12.5: 33.8±7.2%, Fgfr3CE-E12.5:

378 13.2±7.0%, OsxCE-E12.5: 0.0±0.0%, Fig.4a,c). By contrast, Fgfr3CE-E12.5 cells contributed to a

379 higher faction of CFU-Fs than Hes1CE-E12.5 cells did at P21 (Hes1CE-E12.5: 13.7±7.2%,

380 Fgfr3CE-E12.5: 57.1±11.3%, Fig.4b,c). This inversed relationship was associated with a

381 significant decrease of Hes1CE-E12.5 CFU-Fs and a significant increase of Fgfr3CE-E12.5 CFU-

382 Fs during postnatal bone growth (Fig.4c). Isolation of individual tdTomato+ colonies and their

383 subsequent passaging revealed that a small fraction of both Hes1CE-E12.5 and Fgfr3CE-E12.5

384 CFU-Fs could survive at least for ten generations (Fig.4d). Therefore, Hes1-creER+ cell-derived

385 cells make a substantial contribution to a colony-forming fraction of BMSCs particularly during

386 the early postnatal stage. Interestingly, BMSCs derived from Fgfr3-creER+ cells make a

387 progressive contribution to a colony-forming fraction, and compose a majority of these

388 clonogenic cells in the later postnatal stage.

389 To reveal the unique molecular signature characterizing Hes1-creER+ cell-derived

390 BMSCs, we performed a comparative bulk RNA-seq analysis of Cxcl12-GFPhightdTomato+ cells

391 isolated from Cxcl12GFP/+; Hes1-creER; R26RtdTomato and Cxcl12GFP/+; Fgfr3-creER; R26RtdTomato

392 bones with tamoxifen injection at E12.5 (Supplementary Fig.11a). A principal component

393 analysis revealed that biological replicates of Hes1CE-E12.5 and Fgfr3CE-E12.5 Cxcl12-GFPhigh

394 cells grouped by lineage (Supplementary Fig.11b). Notably, 2,969 genes were differentially

395 expressed between the two groups (DEGs) (Supplementary Fig.11c,d). Gene Ontology (GO)

396 enrichment analysis of DEGs revealed significant enrichment of a number of GO terms related to

397 osteoblast differentiation and mineralization, including integrin-mediated signaling pathway

398 (GO:0007229), positive regulation of osteoblast differentiation (GO:0045669), extracellular

399 matrix organization (GO:0030198) and ossification (GO:0001503). Particularly, genes

400 associated with osteoblast differentiation were upregulated in Cxcl12-GFPhighFgfr3CE-E12.5-

401 tdTomato+ cells, including Pth1r, Spp1 and Mmp13 (Fig.4e), indicating that BMSCs derived

402 from Fgfr3-creER+ cells possess a state more biased for osteoblasts than BMSCs derived from

403 Hes1-creER+ cells.

404 Subsequently, we further set out to investigate the osteogenic potential of these two

405 classes of BMSCs. First, we tested the response of Hes1CE-E12.5 and Fgfr3CE-E12.5 BMSCs to

406 bone anabolic agents by giving once daily injections of PTH (1-34, 100µg/kg b.w.) to

407 Col1a1(2.3kb)-GFP; Hes1-creER or Fgfr3-creER; R26RtdTomato mice (pulsed at E12.5) for 3

408 weeks from P21 to P42 (Fig.4f). As expected, this anabolic regimen substantially increased the

409 trabecular bone mass particularly in the metaphysis (Fig.4g,h) (Col1-GFP+ cells: 372.5±29.6 and

410 362.0±65.0 cells/area for Hes1CE-E12.5 and Fgfr3CE-E12.5 femurs, respectively, compared to

411 114.8±11.6 cells/area in Vehicle-treated mice). Importantly, Fgfr3CE-E12.5 cells contributed to a

412 great majority of newly-formed osteoblasts, while virtually no Hes1CE-E12.5 cells contributed to

413 these cells in the metaphysis (Fig.4g,h) (GFP+tdTomato+ cells; 2.6±2.3 % and 55.1±5.0 % of

414 total Col1-GFP+ cells for Hes1CE-E12.5 and Fgfr3CE-E12.5 femurs, respectively), demonstrating

415 that Hes1-creER+ cell-derived BMSCs barely account for anabolic responses during the postnatal

416 stage. Second, we investigated an injury-responsive osteogenesis capability of Hes1CE-E12.5 and

417 Fgfr3CE-E12.5 cells using a femoral bone marrow ablation model, which induced direct

418 differentiation of BMSCs within the marrow cavity (Fig.4i)40. After 7 days of marrow ablation,

419 Hes1CE-E12.5 cells contributed to de novo bone formation exclusively in the diaphysis, while

420 Fgfr3CE-E12.5 cells contributed predominantly in the metaphysis (Fig.4j,k) (GFP+tdTomato+

421 cells; 4.7±3.4 % and 76.0±7.9 % of total Col1-GFP+ cells in metaphyseal bone marrow,

422 53.0±8.0 % and 9.0±3.0% of total Col1-GFP+ cells in diaphyseal bone marrow for Hes1CE-E12.5

423 and Fgfr3CE-E12.5 femurs, respectively). Therefore, Hes1-creER+ cell-derived BMSCs

424 contribute to osteogenesis specifically in the diaphyseal marrow space.

19 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

425 Discussion

426

427 Taken together, our study provides descriptive data on the identity, location and fate of

428 perichondrial cells, chondrocytes and their precursors in the fetal stage, based on a combination

429 of single cell transcriptomic analyses and in vivo cell-lineage analyses of identified cell

430 populations. Single cell RNA-seq analyses of the pre-cartilaginous condensation and the

431 cartilage template reveal a contiguous nature of the fetal chondrocyte-perichondrial cell lineage

432 that largely shares an overlapping set of marker genes. Subsequent studies identified three

433 previously unreported tamoxifen-inducible creER lines, namely Hes1-creER and Dlx5-creER

434 that are highly active in fetal perichondrial cells, and Fgfr3-creER that is highly specific to fetal

435 chondrocytes. These newly identified creER lines have advantages over existing lines as detailed

436 below, therefore might be useful for designing future studies on the fetal perichondrium.

437 First, we identified that Hes1-creER allows marking an early perichondrial population of

438 skeletal progenitor cells in a SOX9-negative domain of the condensation. Prrx1-cre and Prrx1-

439 creER are widely utilized tools that mark skeletal progenitor cells at the condensation stage;

440 however, the caveat of these transgenes is that they simultaneously mark SOX9+ cells within the

441 condensation. However, we acknowledge that Hes1-creER has limitations, in that it also marks

442 non-skeletogenic cell populations in the limb bud. It may be advantageous to utilize multiple

443 tools in tandem to compliment the strengths and weaknesses of each line.

444 Importantly, our findings from Hes1-creER-based lineage-tracing experiments may pose

445 a challenge the established concept that all osteo-chondroprogenitors are derived from Sox9+

446 (Sox9-cre+) cells. There are two scenarios explaining the difference. First, it is possible that

447 Hes1-creER marks the precursors for Sox9+ (Sox9-cre+) cells, supporting the notion that Hes1-

20 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

448 creER+ cells function as earlier perichondrial skeletal progenitor cells. Second, it is also possible

449 that Hes1-creER marks perichondrial cells after these cells have turned off Sox9 (Sox9-cre)

450 expression. Identifying the precise mechanism will be important to define the origin and the

451 function of early perichondrial cells in endochondral development.

452 Second, we identified that Dlx5-creER represents a highly perichondrial cell-specific

453 genetic tool, which provides substantial technological and conceptual advances in this study.

454 Dlx5-creER-based lineage-tracing experiments unambiguously demonstrate the existence of a

455 self-renewing population of skeletal progenitor cells in the fetal perichondrium, which provide a

456 continuous source of precursors for Osx-creER+ perichondrial cells.

457 Third, we identified that Fgfr3-creER is highly specific to chondrocytes, with a

458 significant advantage over previously reported Acan-creER and Col2a1-creER lines. Acan is

459 expressed in the perichondrium; as a result, Acan-creER marks a large number of perichondrial

460 cells, at least in perinatal mice12. In addition, we showed in this study that Col2a1-creER marks a

461 substantial number of fetal perichondrial cells. In contrast, Fgfr3-creER marks almost

462 exclusively chondrocytes within the cartilage template, marking little in the perichondrium.

463 Therefore, this highly specific Fgfr3-creER may be useful for analyzing chondrocytes in the fetal

464 stage.

465 Further, identification of Dlx5-creER and Fgfr3-creER as highly specific inducible

466 genetic tools for perichondrial cells and chondrocyte, respectively, allows us to demonstrate that

467 these two types of cells contribute to the functionally distinct bone marrow stromal

468 compartments in postnatal mice. However, we acknowledge that, as Fgfr3-creER can mark a

469 small number of perichondrial cells, it cannot be entirely excluded that bone marrow stromal

21 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

470 cells are also derived from Fgfr3-creER+ perichondrial cells, not only from Fgfr3-creER+

471 chondrocytes.

472 In addition, we acknowledge that the conclusions of this study may be valid for

473 developing femurs, but not for other skeletal elements that develop earlier or later than the femur

474 during the fetal life in mice. A different time point for tamoxifen injection would be necessary to

475 correctly analyze cell fates in another skeletal element.

476 A combination of multiple inducible mouse genetics tools in this study raises an

477 intriguing possibility that fetal perichondrial cells might provide a complementary source of

478 skeletal progenitor cells through a distinct route during formation of the marrow space (see the

479 proposed diagram in Fig.5). Fetal perichondrial cells might contribute particularly to important

480 stages of early endochondral bone development when the demand for cytogenesis is especially

481 high; first, when the mesenchymal condensation expands explosively, and second, when the

482 nascent marrow space is established within the cartilage template. In contrast, in later stages of

483 development, skeletal progenitor cells established within the cartilage template appear to be self-

484 sufficient to sustain cytogenesis for continual endochondral bone growth due to their robust self-

485 renewal capacity; these cells also give rise to PTHrP+ chondrocytes with skeletal stem cell

486 properties within the resting zone of the postnatal growth plate36.

487 The adult periosteum is derived from the fetal perichondrium. In fact, adult periosteal

488 cells exhibit greater clonogenic and regenerative potential than bone marrow stromal cells,

489 although they share the common embryonic origin and specific markers10. Adult periosteal cells

490 might maintain some of the key features of fetal perichondrial cells, such as the ability to

491 generate chondrocytes and skeletal progenitor cells to direct endochondral ossification. How

22 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

492 cells marked by creER lines identified in this study contribute to fracture healing in adult bones

493 need to be defined through further investigation.

494 Importantly, the fetal perichondrial route appears to be transient, rendering the bone

495 marrow compartment transitional in terms of the developmental origin during postnatal

496 development. While the colony-forming fraction of the early postnatal bone marrow

497 compartment is largely contributed by Hes1-creER+ cells, cells contributed by Fgfr3-creER+

498 cells become predominant in later stages in the newly formed portion of endochondral bones in

499 the metaphysis. These latter cells appear to be the major contributor of new bone formation in

500 response to bone anabolic agents. This unique partnership between the perichondrium and the

501 cartilage template may be one of the major driving forces for explosive mammalian

502 endochondral bone development.

503

23 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

504 Acknowledgement

505

506 We thank T. Nagasawa (Osaka University) for Cxcl12GFP/+ mice, C. Murtaugh for Hes1-

507 creER mice, and H.M. Kronenberg for Osx-creER mice. This research was supported by

508 National Institute of Health grants R01DE026666 (to N.O.), R03DE027421 (to W.O.),

509 R01AR065409 (to S.W.) and University of Michigan MCubed 2.0 Grant (to N.O., W.O. and

510 S.W.), MCubed 3.0 Grant (to N.O., W.O. and J.D.W.), JSPS Overseas Research Fellowship and

511 JSPS KAKENHI Grant JP19K19236 to Y.M. We thank M. Pihalja and K. Saiya-Cork of the

512 University of Michigan Flow Cytometry Core, Yusuke Ono of Kumamoto University, Masaki

513 Matsushita of Nagoya University and T. Lau for supporting this study.

514

515 Author contributions

516

517 Y.M. and N.O. conceived the project and wrote the manuscript. Y.M., M.N., W.O. and

518 N.O. performed the experiment. J.D.W. performed the bioinformatic analysis. S.Y.W. provided

519 the mice and critiqued the manuscript.

520

521 Declaration of interests

522

523 The authors declare no competing interests.

24 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

524 References

525

526 1 Akiyama, H. et al. Osteo-chondroprogenitor cells are derived from Sox9 expressing

527 precursors. Proc Natl Acad Sci U S A 102, 14665-14670, doi:10.1073/pnas.0504750102

528 (2005).

529 2 Ornitz, D. M. & Marie, P. J. Fibroblast growth factor signaling in skeletal development

530 and disease. Genes Dev 29, 1463-1486, doi:10.1101/gad.266551.115 (2015).

531 3 Bianco, P., Cancedda, F. D., Riminucci, M. & Cancedda, R. Bone formation via cartilage

532 models: the "borderline" chondrocyte. Matrix Biol 17, 185-192 (1998).

533 4 Joeng, K. S. & Long, F. Wnt7b can replace Ihh to induce hypertrophic cartilage

534 vascularization but not osteoblast differentiation during endochondral bone development.

535 Bone Res 2, 14004, doi:10.1038/boneres.2014.4 (2014).

536 5 Ono, N. et al. Vasculature-associated cells expressing nestin in developing bones

537 encompass early cells in the osteoblast and endothelial lineage. Dev Cell 29, 330-339,

538 doi:10.1016/j.devcel.2014.03.014 (2014).

539 6 Maes, C. et al. Osteoblast precursors, but not mature osteoblasts, move into developing

540 and fractured bones along with invading blood vessels. Dev Cell 19, 329-344,

541 doi:10.1016/j.devcel.2010.07.010 (2010).

542 7 Mizoguchi, T. et al. Osterix marks distinct waves of primitive and definitive stromal

543 progenitors during bone marrow development. Dev Cell 29, 340-349,

544 doi:10.1016/j.devcel.2014.03.013 (2014).

545 8 Yang, W. et al. Ptpn11 deletion in a novel progenitor causes metachondromatosis by

546 inducing hedgehog signalling. Nature 499, 491-495, doi:10.1038/nature12396 (2013).

25 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

547 9 Debnath, S. et al. Discovery of a periosteal stem cell mediating intramembranous bone

548 formation. Nature 562, 133-139, doi:10.1038/s41586-018-0554-8 (2018).

549 10 Duchamp de Lageneste, O. et al. Periosteum contains skeletal stem cells with high bone

550 regenerative potential controlled by Periostin. Nat Commun 9, 773, doi:10.1038/s41467-

551 018-03124-z (2018).

552 11 Ono, N. & Kronenberg, H. M. Bone repair and stem cells. Curr Opin Genet Dev 40, 103-

553 107, doi:10.1016/j.gde.2016.06.012 (2016).

554 12 Ono, N., Ono, W., Nagasawa, T. & Kronenberg, H. M. A subset of chondrogenic cells

555 provides early mesenchymal progenitors in growing bones. Nat Cell Biol 16, 1157-1167,

556 doi:10.1038/ncb3067 (2014).

557 13 Nakamura, E., Nguyen, M. T. & Mackem, S. Kinetics of tamoxifen-regulated Cre activity

558 in mice using a cartilage-specific CreER(T) to assay temporal activity windows along the

559 proximodistal limb skeleton. Dev Dyn 235, 2603-2612, doi:10.1002/dvdy.20892 (2006).

560 14 Soeda, T. et al. Sox9-expressing precursors are the cellular origin of the cruciate ligament

561 of the knee joint and the limb tendons. Genesis 48, 635-644, doi:10.1002/dvg.20667

562 (2010).

563 15 Henry, S. P. et al. Generation of aggrecan-CreERT2 knockin mice for inducible Cre

564 activity in adult cartilage. Genesis 47, 805-814, doi:10.1002/dvg.20564 (2009).

565 16 Kawanami, A., Matsushita, T., Chan, Y. Y. & Murakami, S. Mice expressing GFP and

566 CreER in osteochondro progenitor cells in the periosteum. Biochem Biophys Res

567 Commun 386, 477-482, doi:10.1016/j.bbrc.2009.06.059 (2009).

26 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

568 17 Ohba, S., He, X., Hojo, H. & McMahon, A. P. Distinct Transcriptional Programs

569 Underlie Sox9 Regulation of the Mammalian Chondrocyte. Cell Rep 12, 229-243,

570 doi:10.1016/j.celrep.2015.06.013 (2015).

571 18 Logan, M. et al. Expression of Cre Recombinase in the developing mouse limb bud

572 driven by a Prxl enhancer. Genesis 33, 77-80, doi:10.1002/gene.10092 (2002).

573 19 Butler, A., Hoffman, P., Smibert, P., Papalexi, E. & Satija, R. Integrating single-cell

574 transcriptomic data across different conditions, technologies, and species. Nat Biotechnol

575 36, 411-420, doi:10.1038/nbt.4096 (2018).

576 20 Nowotschin, S., Xenopoulos, P., Schrode, N. & Hadjantonakis, A. K. A bright single-cell

577 resolution live imaging reporter of Notch signaling in the mouse. BMC Dev Biol 13, 15,

578 doi:10.1186/1471-213X-13-15 (2013).

579 21 Kopinke, D. et al. Lineage tracing reveals the dynamic contribution of Hes1+ cells to the

580 developing and adult pancreas. Development 138, 431-441, doi:10.1242/dev.053843

581 (2011).

582 22 Kronenberg, H. M. The role of the perichondrium in fetal bone development. Ann N Y

583 Acad Sci 1116, 59-64, doi:10.1196/annals.1402.059 (2007).

584 23 Dong, Y. et al. RBPjkappa-dependent Notch signaling regulates mesenchymal progenitor

585 cell proliferation and differentiation during skeletal development. Development 137,

586 1461-1471, doi:10.1242/dev.042911 (2010).

587 24 Hilton, M. J. et al. Notch signaling maintains bone marrow mesenchymal progenitors by

588 suppressing osteoblast differentiation. Nat Med 14, 306-314, doi:10.1038/nm1716 (2008).

27 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

589 25 Trapnell, C. et al. The dynamics and regulators of cell fate decisions are revealed by

590 pseudotemporal ordering of single cells. Nat Biotechnol 32, 381-386,

591 doi:10.1038/nbt.2859 (2014).

592 26 Qiu, X. et al. Reversed graph embedding resolves complex single-cell trajectories. Nat

593 Methods 14, 979-982, doi:10.1038/nmeth.4402 (2017).

594 27 Shu, B. et al. BMP2, but not BMP4, is crucial for chondrocyte proliferation and

595 maturation during endochondral bone development. J Cell Sci 124, 3428-3440,

596 doi:10.1242/jcs.083659 (2011).

597 28 Akiyama, H., Chaboissier, M. C., Martin, J. F., Schedl, A. & de Crombrugghe, B. The

598 transcription factor Sox9 has essential roles in successive steps of the chondrocyte

599 differentiation pathway and is required for expression of Sox5 and Sox6. Genes Dev 16,

600 2813-2828, doi:10.1101/gad.1017802 (2002).

601 29 Chen, H. et al. Runx2 regulates endochondral ossification through control of chondrocyte

602 proliferation and differentiation. J Bone Miner Res 29, 2653-2665,

603 doi:10.1002/jbmr.2287 (2014).

604 30 Usami, Y. et al. Possible Contribution of Wnt-responsive Chondroprogenitors to The

605 Postnatal Murine Growth Plate. J Bone Miner Res, doi:10.1002/jbmr.3658 (2019).

606 31 van Dijk, D. et al. Recovering Gene Interactions from Single-Cell Data Using Data

607 Diffusion. Cell 174, 716-729.e727, doi:10.1016/j.cell.2018.05.061 (2018).

608 32 Taniguchi, H. et al. A resource of Cre driver lines for genetic targeting of GABAergic

609 neurons in cerebral cortex. Neuron 71, 995-1013, doi:10.1016/j.neuron.2011.07.026

610 (2011).

28 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

611 33 Rivers, L. E. et al. PDGFRA/NG2 glia generate myelinating oligodendrocytes and

612 piriform projection neurons in adult mice. Nat Neurosci 11, 1392-1401,

613 doi:10.1038/nn.2220 (2008).

614 34 Rutkowski, T. P. et al. HES factors regulate specific aspects of chondrogenesis and

615 chondrocyte hypertrophy during cartilage development. J Cell Sci 129, 2145-2155,

616 doi:10.1242/jcs.181271 (2016).

617 35 Hosaka, Y. et al. Notch signaling in chondrocytes modulates endochondral ossification

618 and osteoarthritis development. Proc Natl Acad Sci U S A 110, 1875-1880,

619 doi:10.1073/pnas.1207458110 (2013).

620 36 Mizuhashi, K. et al. Resting zone of the growth plate houses a unique class of skeletal

621 stem cells. Nature 563, 254-258, doi:10.1038/s41586-018-0662-5 (2018).

622 37 Zhao, H. et al. The suture provides a niche for mesenchymal stem cells of craniofacial

623 bones. Nat Cell Biol 17, 386-396, doi:10.1038/ncb3139 (2015).

624 38 Liu, Y. et al. Osterix-cre labeled progenitor cells contribute to the formation and

625 maintenance of the bone marrow stroma. PLoS One 8, e71318,

626 doi:10.1371/journal.pone.0071318 (2013).

627 39 Chan, C. K. et al. Identification and specification of the mouse skeletal stem cell. Cell

628 160, 285-298, doi:10.1016/j.cell.2014.12.002 (2015).

629 40 Ono, N. et al. Constitutively active PTH/PTHrP receptor specifically expressed in

630 osteoblasts enhances bone formation induced by bone marrow ablation. J Cell Physiol

631 227, 408-415, doi:10.1002/jcp.22986 (2012).

632 41 Ara, T. et al. Long-term hematopoietic stem cells require stromal cell-derived factor-1 for

633 colonizing bone marrow during ontogeny. Immunity 19, 257-267 (2003).

29 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

634 42 Welch, J. D. et al. Single-Cell Multi-omic Integration Compares and Contrasts Features

635 of Brain Cell Identity. Cell 177, 1873-1887.e1817, doi:10.1016/j.cell.2019.05.006 (2019).

636

30 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

637 Methods

638

639 Mouse strains.

640 Osx-creER6, Hes1-creER21 and Cxcl12GFP/+41 mice have been described previously.

641 Prrx1-cre (JAX005584), Col2a1-cre (JAX003554), Col2a1-creER (JAX006774), Fgfr3-creER

642 (JAX025809), Dlx5-creER (JAX010705), Acta1-cre (JAX006139), Myl-cre (JAX024713), Mck-

643 cre (JAX006475), Tie2-cre (JAX008863), Rosa26-CAG-loxP-stop-loxP-tdTomato (Ai14: R26R-

644 tdTomato, JAX007914), Rosa26-SA-loxP-GFP-stop-loxP-DTA (JAX006331), Col1a1(2.3kb)-

645 GFP (JAX013134), Osteocalcin-GFP (JAX017469) and CBF:H2B-Venus (JAX020942) mice

646 were acquired from the Jackson laboratory. Fgfr3-GFP (MMRRC:012012-UCD) mice were

647 acquired from the Mutant Mouse Resource and Research Centers. All procedures were

648 conducted in compliance with the Guidelines for the Care and Use of Laboratory Animals

649 approved by the University of Michigan’s Institutional Animal Care and Use Committee

650 (IACUC), protocol 7681 (Ono). All mice were housed in a specific pathogen-free condition, and

651 analyzed in a mixed background. Mice were identified by micro-tattooing or ear tags. Tail

652 biopsies of mice were lysed by a HotShot protocol (incubating the tail sample at 95oC for 30 min

653 in an alkaline lysis reagent followed by neutralization) and used for PCR-based genotyping

654 (GoTaq Green Master Mix, Promega, and Nexus X2, Eppendorf). Perinatal mice were also

655 genotyped fluorescently (BLS miner's lamp) whenever possible. Mice were euthanized by over-

656 dosage of carbon dioxide or decapitation under inhalation anesthesia in a drop jar (Fluriso,

657 Isoflurane USP, VetOne).

658

659 Tamoxifen and induction of cre-loxP recombination.

31 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

660 Tamoxifen (Sigma T5648) and 4-hydroxytamoxifen (4-OHT, Sigma H6278) were mixed

661 with 100% ethanol until completely dissolved. Subsequently, a proper volume of sunflower seed

662 oil (Sigma S5007) was added to the tamoxifen-ethanol mixture and rigorously mixed. The

663 tamoxifen-ethanol-oil mixture was incubated at 60oC in a chemical hood until the ethanol

664 evaporated completely. The tamoxifen-oil mixture was stored at room temperature until use.

665 Tamoxifen was injected intraperitoneally using a 26-1/2-gauge needle (BD309597).

666

667 Intermittent administration of parathyroid hormone (PTH).

668 Mice were given once daily subcutaneous injections of PTH (1-34, 100µg/kg b.w.,

669 4095855, Bachem) for 3 weeks from P21 to P42.

670

671 Bone marrow ablation surgery.

672 Mice were anesthetized through a nose cone, in which 1.5-2% isoflurane was constantly

673 provided with oxygen through a vaporizer. For bone marrow ablation surgery, right femurs were

674 operated, while left femurs were untreated and used as an internal control. After skin incision,

675 cruciate ligaments of the knee were carefully separated using a dental excavator, and a hole was

676 made in the intercondylar region of femurs using a 26-1/2-gauge needle (BD309597). The

677 endodontic instruments (K-file, #35, 40, 45, 50) (GC, Tokyo, Japan) were used in a stepwise

678 manner to remove a cylindrical area of the marrow space. The surgical field was irrigated with

679 saline, and the incision line was sutured.

680

681 Histology and immunohistochemistry.

32 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

682 Samples were dissected under a stereomicroscope (Nikon SMZ-800), and fixed in 4%

683 paraformaldehyde for a proper period, typically ranging from 3 hours to overnight at 4oC, then

684 decalcified in 15% EDTA for a proper period, typically ranging from 3 hours to 14 days.

685 Embryonic samples were not decalcified. Subsequently, samples were cryoprotected in 30%

686 sucrose/PBS solutions and then in 30% sucrose/PBS:OCT (1:1) solutions, each at least overnight

687 at 4oC. Samples were embedded in an OCT compound (Tissue-Tek, Sakura) under a

688 stereomicroscope and transferred on a sheet of dry ice to solidify the compound. Embedded

689 samples were cryosectioned at 14µm using a cryostat (Leica CM1850) and adhered to positively

690 charged glass slides (Fisherbrand ColorFrost Plus). Sections were postfixed in 4%

691 paraformaldehyde for 15 min at room temperature. For immunostaining, sections were

692 permeabilized with 0.25% TritonX/TBS for 30 min, blocked with 3% BSA/TBST for 30 min and

693 incubated with rabbit anti-Sox9 polyclonal antibody (1:500, EMD-Millipore, AB5535), rat anti-

694 endomucin (Emcn) monoclonal antibody (1:100, Santa Cruz Biotechnology, sc65495), rabbit

695 anti-Myh3 polyclonal antibody（1:500, Abcam, ab124205）or rabbit anti-Osx polyclonal

696 antibody (1:500, Abcam, ab22552) overnight at 4oC, and subsequently with Alexa Fluor 647-

697 conjugated donkey anti-rabbit IgG (A31573) or Alexa Fluor 633-conjugated goat anti-rat IgG

698 (A21049) (1:400, Invitrogen) for 3 hours at room temperature. Sections were further incubated

699 with DAPI (4’,6-diamidino-2-phenylindole, 5µg/ml, Invitrogen D1306) to stain nuclei prior to

700 imaging.

701

702 RNAscope in situ hybridization.

703 Samples were fixed in 4% paraformaldehyde overnight at 4oC, and then cryoprotected.

704 Frozen sections at 14µm were prepared on positively charged glass slides. In situ hybridization

33 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

705 was performed with RNAscope 2.5 HD Reagent kit Brown (Advanced Cell Diagnostics 322300)

706 using following probes: Hes1 (417701), Fgfr3 (440771) and Sox9 (custom-designed) according

707 to the manufacturer’s protocol.

708

709 Imaging and cell quantification.

710 Images were captured by an automated inverted fluorescence microscope with a

711 structured illumination system (Zeiss Axio Observer Z1 with ApoTome.2 system) and Zen 2

712 (blue edition) software. The filter settings used were: FL Filter Set 34 (Ex. 390/22, Em. 460/50

713 nm), Set 38 HE (Ex. 470/40, Em. 525/50 nm), Set 43 HE (Ex. 550/25, Em. 605/70 nm), Set 50

714 (Ex. 640/30, Em. 690/50 nm) and Set 63 HE (Ex. 572/25, Em. 629/62 nm). The objectives used

715 were: Fluar 2.5x/0.12, EC Plan-Neofluar 5x/0.16, Plan-Apochromat 10x/0.45, EC Plan-Neofluar

716 20x/0.50, EC Plan-Neofluar 40x/0.75, Plan-Apochromat 63x/1.40. Images were typically tile-

717 scanned with a motorized stage, Z-stacked and reconstructed by a maximum intensity projection

718 (MIP) function. Differential interference contrast (DIC) was used for objectives higher than 10x.

719 Representative images of at least three independent biological samples are shown in the figures.

720 Quantification of cells on sections was performed using NIH Image J software.

721

722 Cell preparation.

723 For embryonic samples, hind limbs were harvested and incubated with 2 Wunsch units of

724 Liberase TM (Sigma/Roche 5401127001) in 2ml Ca2+, Mg2+-free Hank’s Balanced Salt Solution

725 (HBSS, Sigma H6648) at 37oC for 15 min on a shaking incubator (ThermomixerR, Eppendorf).

726 For postnatal samples, soft tissues and epiphyses were carefully removed from dissected femurs.

727 After removing distal epiphyseal growth plates and cutting off proximal ends, femurs were cut

34 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

728 roughly and incubated with 2 Wunsch units of Liberase TM and 1mg of Pronase (Sigma/Roche

729 10165921001) in 2ml Ca2+, Mg2+-free HBSS at 37oC for 60 min on a shaking incubator. After

730 cell dissociation, cells were mechanically triturated using an 18-gauge needle with a 1ml Luer-

731 Lok syringe (BD) and a pestle with a mortar (Coors Tek), and subsequently filtered through a

732 70µm cell strainer (BD) into a 50ml tube on ice to prepare single cell suspension. These steps

733 were repeated for 5 times, and dissociated cells were collected in the same tube. Cells were

734 pelleted and resuspended in an appropriate medium for subsequent purposes. For cell culture

735 experiments, cells were resuspended in 10ml culture medium and counted on a hemocytometer.

736

737 Flow cytometry.

738 Dissociated cells were stained by standard protocols with the following antibodies (1:500,

739 eBioscience). eFluor450-conjugated CD31 (390, endothelial/platelet), CD45 (30F-11,

740 hematopoietic), Ter119 (TER-119, erythrocytes), allophycocyanin (APC)-conjugated CD31 (390,

741 endothelial/platelet), CD45 (30F-11, hematopoietic) and Ter119 (TER-119, erythrocytes). Flow

742 cytometry analysis was performed using a four-laser BD LSR Fortessa (Ex. 405/488/561/640

743 nm) and FACSDiva software. Acquired raw data were further analyzed on FlowJo software

744 (TreeStar). Representative plots of at least three independent biological samples are shown in the

745 figures.

746

747 Single cell RNA-seq analysis of fluorescence-activated cell sorting (FACS)-isolated cells.

748 Cell sorting was performed using a four-laser BD FACS Aria III

749 (Ex.407/488/561/640nm) high-speed cell sorter with a 100µm nozzle. Cells of interest

750 (tdTomato+ cells for the experiment described in Fig.1 and Supplementary Fig.1, Cxcl12-GFPhigh

35 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

751 and/or tdTomato+ [both single and double positive: GFPhigh, tdTomato+ and GFPhightdTomato+]

752 cells for the experiment of Fig.3 and Supplementary Fig.6) were directly sorted into ice-cold

753 DPBS/1% BSA, pelleted by centrifugation and resuspended in appropriate amount of DPBS/1%

754 BSA (1,000 cells/μl). Cell numbers were quantified by Countless II automated Cell Counter

755 (ThermoFisher) before loading onto the Chromium Single Cell 3’ v2 microfluidics chip (10x

756 Genomics Inc., Pleasanton, CA). cDNA libraries were sequenced by Illumina HiSeq 4000 using

757 two lanes and 50 cycle paired-end read, generating a total of ~ 770 million reads, or by Illumina

758 or NovaSeq 6000. The sequencing data was first pre-processed using the 10X Genomics

759 software Cell Ranger. For alignment purposes, we generated and used a custom genome fasta

760 and index file by including the sequences of eGFP and tdTomato-WPRE to the mouse genome

761 (mm10). Further downstream analysis steps were performed using the Seurat19 and Monocle25 R

762 package. We filtered out cells with less than 500 genes per cell and with more than 20%

763 mitochondrial read content. The downstream analysis steps include normalization, identification

764 of highly variable genes across the single cells, scaling based on number of UMI, dimensionality

765 reduction (PCA, CCA and t-SNE), unsupervised clustering, and the discovery of differentially

766 expressed cell-type specific markers. Differential gene expression to identify cell-type specific

767 genes was performed using the nonparametric Wilcoxon rank sum test.

768 To integrate Hes1 and Fgfr3 scRNA-seq data, we used LIGER42. This method performs

769 integrative nonnegative matrix factorization (iNMF) to identify shared and dataset-specific

770 metagene factors, which produces an aligned latent space. LIGER then uses this aligned space

771 for joint clustering and t-SNE visualization. For our analysis here, we used k = 25 factors and

772 default clustering parameters.

773

36 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

774 Code availability.

775 LIGER algorithm is fully available as described42.

776

777 Bulk RNA-seq analysis of FACS-isolated cells.

778 Dissociated bone marrow cells harvested from P21 littermate mice were pooled based on

779 the genotype [Cxcl12GFP/+; Hes1-creER; R26RtdTomato (pulsed at E12.5, Hes1-E12.5 mice) or

780 Cxcl12GFP/+; Fgfr3-creER; R26RtdTomato (pulsed at E12.5, Fgfr3-E12.5 mice)]. Cell sorting was

781 performed using a four-laser BD FACS Aria III (Ex.407/488/561/640nm) high-speed cell sorter

782 with a 100µm nozzle. Cxcl12-GFPhightdTomato+ cells isolated from Hes1-E12.5 or Fgfr3-E12.5

783 mice were directly sorted into ice-cold DPBS/10% FBS and pelleted by centrifugation. Total

784 RNA was isolated using PicoPure RNA Isolation Kit (KIT0204, ThermoFisher), followed by

785 DNA-free DNA removal kit (AM1906, ThermoFisher) to remove contaminating genomic DNA.

786 RNA Integrity Number (RIN) was assessed by Agilent 2100 Bioanalyzer RNA 6000 Pico Kit.

787 Samples with RIN>8.0 were used for subsequent analyses. Complementary DNAs were prepared

788 by SMART-Seq v4 Ultra Low Input RNA Kit for Sequencing (Clontech 634888). Post-

789 amplification quality control was performed by Agilent TapeStation DNA High Sensitivity

790 D1000 Screen Tape system. DNA libraries were prepared by Nextera XT DNA Library

791 Preparation Kit (Illumina) and submitted for NextGen sequencing (Illumina HiSeq 4000). DNA

792 libraries were sequenced using following conditions; six samples per lane, 50 cycle single-end

793 read. Reads files were downloaded and concatenated into a single .fastq file for each sample. The

794 quality of the raw reads data for each sample was checked using FastQC to identify quality

795 problems. Tuxedo Suite software package was subsequently used for alignment (using TopHat

796 and Bowtie2), differential expression analysis, and post-analysis diagnostics. FastQC was used

37 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

797 for a second round of quality control (post-alignment). HTSeq/DESeq2 was used for differential

798 expression analysis using UCSC mm10 as the reference genome sequence. Only non-

799 ambiguously mapped reads were counted. Meta-analysis of RNA-seq data was performed using

800 iPathwayGuide software (Advaita).

801

802 Colony-forming assay and subcloning.

803 Nucleated bone marrow cells were plated into tissue culture 6 well plates (BD Falcon) at

804 a density of <105 cells/cm2, and cultured in low-glucose DMEM with GlutaMAX supplement

805 (Gibco 10567022) and 10% mesenchymal stem cell-qualified FBS (Gibco 12662029) containing

806 penicillin-streptomycin (Sigma P0781) for 10~14 days. Cell cultures were maintained at 37oC in

807 a 5% CO2 incubator. Representative images of at least three independent biological samples are

808 shown in the figures. For CFU-Fs, cells were fixed with 70% Ethanol for 5 min and stained for

809 2% methylene blue.

810 Colonies marked by tdTomato were individually subcloned using a method described

811 previously36. Briefly, a tile-scanned virtual reference of tdTomato+ colonies and cloning

812 cylinders (Bel-Art) were used to isolate an individual colony. Single cell-derived clones of

+ 813 tdTomato cells were cultured in a basal medium described above at 37°C with 5% CO2 with

814 exchanges into fresh media every 3–4 days.

815

816 Statistical analysis.

817 Results are presented as mean values ± S.D. Statistical evaluation was conducted using

818 the Mann-Whitney's U-test or one-way ANOVA. A P value of <0.05 was considered significant.

819 No statistical method was used to predetermine sample size. Sample size was determined on the

38 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

820 basis of previous literature and our previous experience to give sufficient standard deviations of

821 the mean so as not to miss a biologically important difference between groups. The experiments

822 were not randomized. All of the available mice of the desired genotypes were used for

823 experiments. The investigators were not blinded during experiments and outcome assessment.

824 One femur from each mouse was arbitrarily chosen for histological analysis. Genotypes were not

825 particularly highlighted during quantification.

826

827 Data availability.

828 The datasets generated during and/or analyzed during the current study are available from

829 the corresponding author on reasonable request. The single cell and bulk RNA-seq datasets

830 presented herein have been deposited in the National Center for Biotechnology Information

831 (NCBI)’s Gene Expression Omnibus (GEO), and are accessible through GEO Series accession

832 numbers GSE126966 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE126966],

833 GSE126967 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE126967], GSE126968

834 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE126968], GSE126969

835 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE126969] and GSE144411

836 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE144411]. The source data underlying

837 all Figures and Supplementary Figures are provided as a Source Data file.

838

39 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Figure 1. Undifferentiated Hes1+Sox9neg mesenchymal cells provide precursors for skeletal progenitor cells within the cartilage template and the perichondrium.

(a,b) Single cell RNA-seq analysis of Prrx1-cre-marked limb bud mesenchymal cells at E11.5. (a): Prrx1-cre; R26RtdTomato femur stained for Sox9. Grey: DIC. Scale bar: 200µm. n=3 mice. (b): UMAP-based visualization of major classes of Prrx1cre-tdTomato+ cells (Cluster 0 – 8). Right panels: feature plots. Blue: high expression. Arrowheads: clusters in which a given gene is identified as a cell type-specific marker. Dotted contour: Sox9+ cells. n=4,804 cells. Pooled from n=5 mice. (c) Notch reporter CBF1:H2B-Venus femur immunostained for Sox9 at E11.5. Grey: DIC. Scale bar: 200µm. n=3 mice. (d-h) Cell-fate analysis of Hes1-creER+ mesenchymal cells of the condensation stage, pulsed at E10.5. Hes1-creER; R26RtdTomato femurs carrying CBF1:Venus (e,f) or Col1a1(2.3kb)-GFP (g,h) reporters. (d): Limb bud at E11.5. Grey: DIC. Scale bar: 200µm. (e): Magnified view of mesenchymal condensation at E11.5. Grey: DAPI. Scale bar: 20µm. (f): Cartilage template at E13.5. Right panel: magnified view of Area (1). Dotted line: cartilage-perichondrium border. Grey: DIC. Scale bar: 200µm. (g): Cartilage-perichondrium immunostaining for Sox9, osterix (Osx), Myh3 and endomucin (Emcn). Grey: DAPI. Scale bar: 20µm. (h): Distal femur at P21 with growth plates on top. Whole bone (left), growth plates (upper right), endocortical marrow space at metaphysis (lower center) and diaphysis (lower right). Grey: DIC. Scale bar: 500µm (left panel), 50µm (upper right panel), 100µm (lower center and right panels). n=4 mice per each time point. (i) Diagram of Hes1+ mesenchymal cell fates. Hes1+Sox9neg cells surrounding the condensation are bona fide skeletal progenitor cells during endochondral bone development, contributing to both chondrocytes and perichondrial cells, and subsequently to all limb skeletal cells.

1 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Figure 2. Hes1-creER marks self-renewing Dlx5+Osxneg progenitor cells in the fetal perichondrium.

(a,b) Single cell RNA-seq analysis of Col2a1-cre-marked chondrocytes and perichondrial cells at E13.5. (a): Col2a1-cre; R26RtdTomato femur. R: Round layer. F: Flat layer. PH: Prehypertrophic layer. PC: Perichondrium. Grey: DIC. Scale bar: 200µm. n=3 mice. (b): UMAP-based visualization of major classes of Col2a1cre-tdTomato+ cells (Cluster 0 – 10). Red contour: perichondrial cells, Blue contour: chondrocytes. Right panels: feature plots of genes enriched in each cluster. Cluster 0,1,3: Sox9+, cluster 2,4,8: Hes1+, cluster 1: Fgfr3+. Blue: high expression. Dotted contour: chondrocytes. n=7,932 cells. Pooled from n=5 mice. (c) Notch reporter CBF1:H2B-Venus femur immunostained for Sox9 at E13.5. Grey: DIC. Scale bar: 200µm. n=3 mice. (d-f) Cell-fate analysis of Hes1-creER+, Dlx5-creER+, Osx-creER+ or Fgfr3-creER+ cells, pulsed at E12.5, carrying Col1a1(2.3kb)-GFP reporters. (d): Cartilage template at E13.5. Left panels: Grey: DIC. Scale bar: 200µm. Right panels: magnified view of (1-4). Grey: DAPI. n=4 mice per each group. (e): Upper: percentage of total tdTomato+ outer perichondrial cells among total outer perichondrial cells. n=4 mice per each group. ***p<0.001, ****p<0.0001, Hes1-E12.5 versus Fgfr3-E12.5, mean difference = 66.00, 95% confidence interval (59.56, 72.44); Hes1-E12.5 versus Osx-E12.5, mean difference = 61.29, 95% confidence interval (54.84, 67.73); Hes1-E12.5 versus Dlx5-E12.5, mean difference = 12.48, 95% confidence interval (6.034, 18.92); Fgfr3- E12.5 versus Osx-E12.5, mean difference = -4.711, 95% confidence interval (-11.15, 1.731); Fgfr3-E12.5 versus Dlx5-E12.5, mean difference = -53.52, 95% confidence interval (-59.96, - 47.08); Osx-E12.5 versus Dlx5-E12.5, mean difference = -48.81, 95% confidence interval (- 55.25, -42.37). Two-tailed, One-way ANOVA followed by Tukey’s post-hoc test (left). Two- tailed, Mann-Whitney’s U-test (right). Data are presented as mean ± s.d. Lower left: percentage of Col1a1(2.3kb)-GFP+tdTomato+ cells among total Col1a1(2.3kb)-GFP+ perichondrial osteoblasts. n=4 mice per each group. ****p<0.0001, Hes1-E12.5 versus Fgfr3-E12.5, mean difference = -2.513, 95% confidence interval (-12.96, 7.933); Hes1-E12.5 versus Osx-E12.5, mean difference = -69.33, 95% confidence interval (-79.78, -58.88); Hes1-E12.5 versus Dlx5- E12.5, mean difference = -6.147, 95% confidence interval (-16.59, 4.299); Fgfr3-E12.5 versus Osx-E12.5, mean difference = -66.82, 95% confidence interval (-77.26, -56.37); Fgfr3-E12.5

versus Dlx5-E12.5, mean difference = -3.634, 95% confidence interval (-14.08, 6.813); Osx- E12.5 versus Dlx5-E12.5, mean difference = 63.18, 95% confidence interval (52.74, 73.63). Two-tailed, One-way ANOVA followed by Tukey’s post-hoc test. Data are presented as mean ± s.d. Lower right: Percentage of Sox9+tdTomato+ cells among total Sox9+ chondrocytes. n=4 (Hes1-E12.5, Dlx5-E12.5 and Fgfr3-E12.5), n=6 (Osx-E12.5). ****p<0.0001, Hes1-E12.5 versus Fgfr3-E12.5, mean difference = -36.38, 95% confidence interval (-41.46, -31.30); Hes1- E12.5 versus Osx-E12.5, mean difference = -17.72, 95% confidence interval (-22.36, -13.08); Hes1-E12.5 versus Dlx5-E12.5, mean difference = -1.910, 95% confidence interval (-6.992, 3.171); Fgfr3-E12.5 versus Osx-E12.5, mean difference = 18.66, 95% confidence interval (14.02, 23.30); Fgfr3-E12.5 versus Dlx5-E12.5, mean difference = 34.47, 95% confidence interval (29.39, 39.55); Osx-E12.5 versus Dlx5-E12.5, mean difference = 15.81, 95% confidence interval (11.17, 20.45). Two-tailed, One-way ANOVA followed by Tukey’s post-hoc test. Data are presented as mean ± s.d. (f): Whole femurs at E18.5. Arrowhead: groove of Ranvier. Grey: DIC. Scale bar: 500µm. n=4 mice per each group. (g) Inducible partial cell ablation of Hes1-creER+, Osx-creER+ or Fgfr3-creER+ cells using an inducible Diphtheria toxin fragment A (iDTA) allele. Quantification of femur length, bone marrow length and distal cartilage length. n=7 (Hes1-Cont), 6 (Hes1-DTA, Fgfr3-DTA), 5 (Osx- Cont), 4 (Osx-DTA, Fgfr3-Cont). *p<0.05, **p<0.01, ***p<0.001, two-tailed, Mann-Whitney’s U-test. Data are presented as mean ± s.d. (h) Diagram of Hes1-E12.5, Dlx5-E12.5, Osx-E12.5, or Fgfr3-E12.5, cell fates. Hes1-creER and Dlx5-creER, but not Osx-creER or Fgfr3-creER, can mark a self-renewing population of fetal perichondrial cells with robust capability to contribute to cortical and trabecular bone compartments.

3 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Figure 3. Hes1+Dlx5+Osxneg perichondrial cells substantially contribute to the postnatal bone marrow stromal compartment

(a-i) Contribution of fetal Hes1-creER+ cells, Dlx5-creER+ cells, Osx-creER+ cells or Fgfr3- creER+ cells (pulsed at E12.5) to Cxcl12-GFP+ bone marrow stromal cells at P21. Cxcl12GFP/+; Hes1-creER; R26RtdTomato (a,e), Cxcl12GFP/+; Dlx5-creER; R26RtdTomato (b,f), Cxcl12GFP/+; Osx- creER; R26RtdTomato (c,g) or Cxcl12GFP/+; Fgfr3-creER; R26RtdTomato (d,h) femurs with growth plates on top. (a-d): Whole bone, upper panels in (e-h): metaphyseal bone marrow, lower panels in (e-h): diaphyseal bone marrow. Grey: DIC (a-d), DAPI (e-h). Scale bar: 500µm (a-d), 20µm (e-h). n=4 mice per group. (i): Percentage of Cxcl12-GFPhightdTomato+ cells per total Cxcl12- GFPhigh cells. n=5 (Hes1-E12.5), n=4 (Dlx5-E12.5), n=5 (Osx-E12.5), n=10 (Fgfr3-E12.5), n=6 (Hes1-E10.5), n=4 (Hes1-E14.5) mice per group. Left: **p<0.01, ***p<0.001, ****p<0.0001, Hes1-E12.5 versus Dlx5-E12.5, mean difference = -1.050, 95% confidence interval (-6.686, 4.586); Hes1-E12.5 versus Osx-E12.5, mean difference = 16.65, 95% confidence interval (11.34, 21.96); Hes1-E12.5 versus Fgfr3-E12.5, mean difference = -8.840, 95% confidence interval (- 13.44, -4.238); Dlx5-E12.5 versus Osx-E12.5, mean difference = 17.70, 95% confidence interval (12.06, 23.34); Dlx5-E12.5 versus Fgfr3-E12.5, mean difference = -7.790, 95% confidence interval (-12.76, -2.819); Osx-E12.5 versus Fgfr3-E12.5, mean difference = -25.49, 95% confidence interval (-30.09, -20.89). Two-tailed, One-way ANOVA followed by Tukey’s post- hoc test. Data are presented as mean ± s.d. Right: ***p<0.001, ****p<0.0001, Hes1-E10.5 versus Hes1-E12.5, mean difference = 16.42, 95% confidence interval (11.40, 21.44); Hes1- E10.5 versus Hes1-E14.5, mean difference = 28.23, 95% confidence interval (22.88, 33.58); Hes1-E12.5 versus Hes1-E14.5, mean difference = 11.82, 95% confidence interval (6.258, 17.38). Two-tailed, One-way ANOVA followed by Tukey’s post-hoc test. Data are presented as mean ± s.d. (j-l) Combined lineage-tracing and single cell RNA-seq analyses of Cxcl12-GFP+ bone marrow stromal cells and tdTomato+ mesenchymal cells derived from fetal Hes1+ perichondrial cells and Fgfr3+ chondrocytes. (j): Both single and double positive cells (GFPhigh, tdTomato+ and GFPhightdTomato+) were FACS-isolated from Cxcl12GFP/+; Hes1-creER; R26RtdTomato and Cxcl12GFP/+; Fgfr3-creER; R26RtdTomato femurs at P21, pulsed at E12.5 (Hes1-E12.5 or Fgfr3- E12.5). Two single cell datasets were integrated by LIGER. (k): t-SNE-based visualization of

major classes of mesenchymal Cxcl12-GFPhigh and/or tdTomato+ (Hes1-E12.5 or Fgfr3-E12.5) cells (Cluster 0, 2, 4, 5, 9, 13, 15). Lower panels: feature plots. Blue: high expression. Blue contour: Cxcl12+ reticular cells. n=4,437 cells. Pooled from n=4 mice per group. (l): Cells colored by conditions (Hes1-E12.5, Fgfr3-E12.5), and percentage of cells in each cluster among total cells, demonstrated for each condition. n=1,628 cells (tdTomato+).

5 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Figure 4. Site-specific roles of fetal perichondrium-derived and chondrocyte-derived mesenchymal stromal cells in osteoblast differentiation.

(a-d) Colony-forming unit fibroblast (CFU-F) assay of E12.5-pulsed Hes1-creER; R26RtdTomato (left panels), Fgfr3-creER; R26RtdTomato (right panels) and Osx-creER; R26RtdTomato bone marrow cells, at P7 (a) and P21 (b). MB: Methylene blue staining. Scale bar: 5mm. (c): Percentage of tdTomato+ colonies among total CFU-Fs. Hes1-E12.5: n=4 (P7), n=5 (P21), Fgfr3-E12.5: n=5 (P7), n=7 (P21), Osx-E12.5: n=3 (P7) mice. Hes1-E12.5 versus Fgfr3-E12.5 versus Osx-E12.5 (Day7): *p<0.05, **p<0.01, ***p<0.001, Hes1-E12.5 versus Fgfr3-E12.5, mean difference = 20.51, 95% confidence interval (8.789, 32.23); Hes1-E12.5 versus Osx-E12.5, mean difference = 33.75, 95% confidence interval (20.40, 47.10); Fgfr3-E12.5 versus Osx-E12.5, mean difference = 13.24, 95% confidence interval (0.4796, 26.00). Two-tailed, One-way ANOVA followed by Tukey’s post-hoc test. Data are presented as mean ± s.d. Hes1-E12.5 versus Fgfr3-E12.5 versus Osx-E12.5 (Day7), Hes1-E12.5 (Day7 versus Day21), Fgfr3-E12.5 (Day7 versus Day21): *p<0.05, **p<0.01, two-tailed, Mann-Whitney’s U-test. Data are presented as mean ± s.d. (d): Survival curve of Hes1-E12.5 and Fgfr3-E12.5 individual tdTomato+ clones (P21) over 10 passages. n=50 (Hes1-E12.5), n=60 (Fgfr3-E12.5) clones per group. Gehan-Breslow-Wilcoxon test. (e) Comparative bulk RNA-seq analysis of Cxcl12-GFPhigh stromal cells derived from fetal Hes1+ perichondrial cells and Fgfr3 + chondrocytes. Heatmap of representative differentially expressed genes (DEGs) associated with osteoblast differentiation. Star: statistically significant difference between Hes1-E12.5 and Fgfr3-E12.5. n=2 biological replicates (Hes1-E12.5: pooled from n=4 mice, Fgfr3-E12.5: pooled from n=5 mice). (f-h) Differential responses of mesenchymal stromal cells derived from Hes1+ perichondrial cells and Fgfr3+ chondrocytes to intermittent administration of parathyroid hormone (PTH). (g): Col1a1(2.3kb)-GFP; Hes1-creER; R26RtdTomato (Hes1-E12.5, left), Col1a1(2.3kb)-GFP; Fgfr3- creER; R26RtdTomato (Fgfr3-E12.5, center), and Col1a1(2.3kb)-GFP; R26RtdTomato (Control, right) distal femurs, with iPTH or Vehicle for 3 weeks from P21 to P42. Grey: DIC. Scale bar: 500µm. (h): Quantification of Col1a1(2.3kb)-GFP+ cells (left) and Col1a1(2.3kb)-GFP+tdTomato+ cells (right) in the metaphysis. n=6 (Hes1-E12.5), n=5 (Fgfr3-E12.5), n=4 (Control) mice. ****p<0.0001, Hes1-E12.5 versus Fgfr3-E12.5, mean difference = 10.5, 95% confidence

interval (-58.2, 79.2); Hes1-E12.5 versus Vehicle, mean difference = 257.8, 95% confidence interval (184.5, 331); Fgfr3-E12.5 versus Vehicle, mean difference = 247.3, 95% confidence interval (171.1, 323.4). Two-tailed, One-way ANOVA followed by Tukey’s post-hoc test (left). **p<0.01, two-tailed, Mann-Whitney’s U-test (right). Data are presented as mean ± s.d. (i-k) Injury-responsive osteogenesis of mesenchymal stromal cells derived from fetal perichondrial cells (Hes1-E12.5) and chondrocytes (Fgfr3-E12.5) cells. (j): Bone marrow ablation for trabecular bone regeneration in Col1a1(2.3kb)-GFP; Hes1-creER; R26RtdTomato (left panels) and Fgfr3-creER; R26RtdTomato (right panels) femurs, pulsed at E12.5, undergone surgery at P21 and analyzed at P28. Metaphyseal (upper panels) and diaphyseal (lower panels) marrow space after one week of surgery. Grey: DIC. Scale bar: 500µm. (k): Quantification based on sections. Percentage of Col1(2.3kb)-GFP+tdTomato+ osteoblasts per total Col1(2.3kb)-GFP+ osteoblasts in the metaphysis (left) and diaphysis (right). n=4 per each group. *p<0.05, two- tailed, Mann-Whitney’s U-test. Data are presented as mean ± s.d.

7 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Figure 5. Notch effector Hes1 marks an early perichondrial population of skeletal progenitor cells in endochondral bone development

(a) Process of endochondral bone development and bone marrow formation. Skeletal primordium is formed by condensing mesenchymal cells. These mesenchymal cells differentiate into chondrocytes and their surrounding perichondrial cells. Both cartilage template and perichondrium contribute to endochondral bone development by providing skeletal progenitor cells. These cells in concert form the nascent marrow space along with invading endothelial and hematopoietic cells. (b) Early perichondrial population of skeletal progenitor cells in endochondral bone development. Undifferentiated mesenchymal cells surrounding condensation or cartilage template provide an important source during fetal endochondral bone development. During the condensation stage, Hes1+ undifferentiated mesenchymal cells take two distinct routes to participate in endochondral bone development. First, these cells translocate into the cartilage template and directly differentiate into Sox9+ chondrocytes, and then contribute to bone formation (chondrocyte- dependent pathway). Second, these cells contribute to the outer layer of the perichondrium and become Hes1+ and Dlx5+ perichondrial cells. These Hes1+ and Dlx5+ perichondrial cells translocate into the nascent marrow space and directly differentiate into marrow mesenchymal cells by bypassing a Sox9+ state (chondrocyte-independent pathway). These perichondrial skeletal progenitor cells keep providing osteoblasts and bone marrow stromal cells even during the postnatal stage, unlike Osx+ perichondrial osteoblast precursors with limited life span. As a result, postnatal bone narrow is characterized by transitional mosaicism composed of both chondrocyte-derived and perichondrium-derived stromal cells.

8 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Figure 1. Single cell RNA-seq analysis of limb bud mesenchymal cells.

(a-c) Single cell RNA-seq analysis of Prrx1-cre-marked limb bud mesenchymal cells at E11.5. (a): FACS-sorting strategy for Prrx1cre-tdTomato+ cells (red box), cells isolated from Prrx1-cre; R26RtdTomato limbs. Right panel: control cells. (b): UMAP-based visualization of major classes of Prrx1cre-tdTomato+ cells (left) and feature plots of representative markers (right). Blue: high expression. Dotted contour: Sox9+ cells. n=4,804 cells. Pooled from n=5 mice. (c): Feature plots of genes enriched in each cluster. Cluster 2,7: Sox9+, Cluster 6: Shh+Msx1+, Cluster 5: Lhx9+Msx1+, Cluster 3,7: Irx3+Irx5+, Cluster 0: Meox2+Osr1+, Cluster 4: Emx2+. Red contour: featured clusters. (d) RNAscope in situ hybridization analysis of E11.5 limb bud for Hes1 and Sox9 mRNA. Scale bar: 200µm. n=3 mice.

9 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Figure 2. Contribution of Hes1+Sox9neg mesenchymal cells to endochondral bone development.

Cell-fate analysis of Hes1-creER+ mesenchymal cells of the condensation stage, pulsed at E10.5. Hes1-creER; R26RtdTomato femurs carrying Col1a1(2.3kb)-GFP (c,d) or Cxcl12-GFP (e,f) reporters. (a): Short-chase analysis of Hes1-creER+ cells at E11.5. Mesenchymal condensation. RNAscope in situ hybridization analysis for Hes1 (center) and Sox9 (right) mRNA. Scale bar: 20µm. (b): Immunostaining for Myh3 (top) and Emcn (bottom). Arrow: Hes1-creER+Emcn+ cells. Grey: DAPI. Scale bar: 20µm. (c): Cartilage template at E13.5, after 3 days of chase. Imunostaining for Sox9, Osx, Myh3 and Emcn. Grey: DIC. Scale bar: 200µm. (d): Neonatal femur at P0, after 9 days of chase. Right panels: magnified views of (1: growth plate, 2: bone marrow). Grey: DIC. Scale bar: 200µm. (e): Whole femur at P21 with growth plates on top, after one month of chase. Grey: DIC. Scale bar: 500µm. (f): metaphyseal bone marrow (top) and diaphyseal bone marrow (bottom). Grey: DAPI. Scale bar: 20µm. n=4 mice per each time point.

10 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Figure 3. Skeletal muscle cells or endothelial cells do not contribute to chondrocytes, osteoblasts or bone marrow stromal cells.

(a) Fate mapping analysis of Acta1-cre, Myl1-cre or Mck-cre-marked skeletal muscle cells at P0 (top) and P21 (bottom). Acta1-cre; R26RtdTomato (left), Myl1-cre; R26RtdTomato (center) and Mck- cre; R26RtdTomato (right) femurs. Immunostaining for Myh3. Grey: DIC. Scale bar: 500µm. n=3 mice per each group. (b) Fate-mapping analysis of Tie2-cre-marked endothelial cells at P21. Col1a1(2.3kb)-GFP; Tie2-cre; R26RtdTomato and Cxcl12GFP/+; Tie2-cre; R26RtdTomato femurs. Left panel: whole bone. Grey: DIC. Scale bar: 500µm. Upper center panels: magnified views of boxed areas (1: trabecular bone, 2: endosteal marrow space). Lower center panel: magnified view of diaphyseal bone marrow of Cxcl12GFP/+; Tie2-cre; R26RtdTomato femur. Grey: DAPI. Scale bar: 20µm. n=3 mice per each group. Upper right panels: flow cytometry analysis of bone marrow cells. Right center panel: histogram showing GFP expression. Blue lines: control cells. n=5 (Cxcl12GFP/+), n=4 (Col1a1(2.3kb)-GFP) mice. Lower right panels: CFU-F assay of Tie2-cre; R26RtdTomato bone marrow cells. Left: tdTomato epifluorescence. Scale bar: 5mm. Right: F4/80 immunostaining. Scale bar: 50µm. n=3 mice.

11 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Figure 4. Single cell RNA-seq analysis of fetal chondrocytes and perichondrial cells.

Single cell RNA-seq analysis of Col2a1-cre-marked chondrocytes and perichondrial cells at E13.5. (a): FACS-sorting strategy for Col2a1cre-tdTomato+ cells (red box), cells isolated from Col2a1-cre; R26RtdTomato limbs. Right panel: control cells. (b): UMAP-based visualization of major classes of Col2a1cre-tdTomato+ cells. Red contour: perichondrial cells, Blue contour: chondrocytes. n=7,932 cells. Pooled from n=5 mice. (c): Feature plots of genes enriched in chondrocyte clusters and perichondrial cell clusters. Blue: high expression. Cluster 0, 1, 3: chondrocytes, cluster 2,4,8: perichondrium. Dotted contour: chondrocytes. (d): Pseudotime and single cell trajectory analysis by Monocle. Right panel: inferred lineage trajectory. (e): Heatmap of representative pseudotime-dependent differentially expressed genes. (f): MAGIC imputation analysis showing the Sox9-Fgfr3 and Hes1-Fgfr3 relationships.

12 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Figure 5. Hes1-creER marks self-renewing Dlx5+Osxneg progenitor cells in the fetal perichondrium.

(a-d) Cell-fate analysis of Hes1-creER+, Dlx5-creER+ or Fgfr3-creER+ cells, pulsed at E12.5. (a): CBF1:Venus; Hes1-creER; R26RtdTomato femurs at E13.5. Left: RNAscope in situ hybridization analysis of Hes1 mRNA. Grey: DIC. Scale bar: 200µm. (b): Fgfr3-GFP; Fgfr3- creER; R26RtdTomato femurs at E13.5. Left: RNAscope in situ hybridization analysis of Fgfr3 mRNA. Right: Fgfr3-GFP femur at E13.5 with Sox9 immunostaining. Grey: DIC. Scale bar: 200µm. (c): Col1a1(2.3kb)-GFP; Hes1-creER; R26RtdTomato femurs at E13.5. Left: Myh3 immunostaining. Grey: DIC. Scale bar: 200µm. Center: magnified views of boxed area (1). Grey: DAPI. Right: Emcn immunostaining. Grey: DAPI. Scale bar: 20µm. (d) Col1a1(2.3kb)- GFP; Dlx5-creER; R26RtdTomato femurs at E13.5. Left: Myh3 immunostaining. Grey: DIC. Scale bar: 200µm. Center: magnified views of boxed area (2). Grey: DAPI. Right: Emcn immunostaining. Grey: DAPI. Scale bar: 20µm. n=4 mice per each group. (e) CBF1:Venus; Hes1-creER; R26RtdTomato femurs at E15.5, pulsed at E14.5. Left: Grey: DIC. Scale bar: 200µm. Center, right: magnified views of boxed area (3: articular cartilage, 4: proliferating chondrocytes) with Sox9 immunostaining. Grey: DAPI. n=4 mice. (f) CBF1:Venus; Hes1-creER; R26RtdTomato femurs at P5, pulsed at P3. Grey: DIC. Scale bar: 200µm. n=4 mice.

13 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Figure 6. Hes1-creER marks self-renewing Dlx5+Osxneg progenitor cells in the fetal perichondrium.

(a-c) Cell-fate analysis of Hes1-creER+, Dlx5-creER+, Osx-creER+ or Fgfr3-creER+ cells, pulsed at E12.5, carrying Col1a1(2.3kb)-GFP (a,b) or Cxcl12-GFP (c) reporters. (a): Left panels: whole femurs at E15.5. Grey: DIC. Scale bar: 200µm. Right panels: magnified view of (1-4). Grey: DAPI. n=4 mice per each group. (b): Perichondrium, bone collar and growth plate at E18.5. Arrowhead: groove of Ranvier. Grey: DIC. Scale bar: 50µm. n=4 mice per each group. (c) Bone marrow at P0. Grey: DAPI. Scale bar: 20µm. n=3 mice per each group. (d) Cell-fate analysis of Col2a1-creER+ or Fgfr3-creER+ cells, pulsed at E12.5. Upper left: Col1a1(2.3kb)-GFP; Col2a1creER; R26RtdTomato femurs at E13.5. Grey: DIC. Scale bar: 200µm. n=3 mice. Right: Col1a1(2.3kb)-GFP; Col2a1-creER; R26RtdTomato or Col1a1(2.3kb)-GFP; Fgfr3-creER; R26RtdTomato femurs at P7. Scale bar: 500µm. Lower left: magnified view of (5,6: perichondrium, bone collar and growth plate). Grey: DIC. n=3 mice per each group. (e) Inducible partial cell ablation of Hes1-creER+, Osx-creER+ or Fgfr3-creER+ cells using an inducible Diphtheria toxin fragment A (iDTA) allele. Hes1-creER, Osx-creER or Fgfr3-creER; Rosa26lsl-tdTomato/+ (Cont) and Hes1-creER, Osx-creER or Fgfr3-creER; Rosa26lsl-tdTomato/iDTA (DTA) mice at E18.5, pulsed at E12.5. n=10 (Hes1-Cont), 7 (Hes1-DTA), 6 (Fgfr3-DTA), 5 (Osx-Cont), 4 (Osx-DTA, Fgfr3-Cont).

14 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Figure 7. Differential contribution of Hes1+, Dlx5+, Osx+ and Fgfr3+ cells to postnatal growth plate and bone marrow stromal compartment.

(a-d) Contribution of fetal Hes1-creER+, Dlx5-creER+, Osx-creER+ or Fgfr3-creER+ cells (pulsed at E12.5) to growth plate chondrocytes and osteoblasts at P21. Col1a1(2.3kb)-GFP; Hes1-creER; R26RtdTomato (a), Col1a1(2.3kb)-GFP; Dlx5-creER; R26RtdTomato (b), Osteocalcin (Ocn); Osx-creER; R26RtdTomato (c) or Col1a1(2.3kb)-GFP; Fgfr3-creER; R26RtdTomato femurs (d) with growth plates on top. Whole bone (left), growth plates (upper right), endocortical marrow space at metaphysis (lower center) and diaphysis (lower right). Grey: DIC (a-d). Scale bar: 500µm (left panels), 50µm (upper right panels), 100µm (lower center and right panels). n=3 mice per each group.

15 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Figure 8. Hes1-creER predominantly marks endothelial cells in late fetal stages.

Contribution of Hes1-creER+ cells (pulsed at E14.5) to Cxcl12-GFP+ bone marrow stromal cells, growth plate chondrocytes or osteoblasts at P21. Cxcl12GFP/+; Hes1-creER; R26RtdTomato (left, upper) femurs with growth plates on top. Left: Whole bone. Grey: DIC. Scale bar: 500µm. Upper: Metaphyseal and diaphyseal bone marrow. Grey: DAPI. Scale bar: 20µm. Col1a1(2.3kb)- GFP; Hes1-creER; R26RtdTomato (lower center, center right, lower right) femurs with growth plates on top. Whole bone (lower center), growth plates (center right), endocortical marrow space at metaphysis and diaphysis (lower right). Grey: DIC. Scale bar: 500µm (lower center), 50µm (center right), 100µm (lower right). n=3 mice per each group.

16 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Figure 9. Flow cytometry analysis of lineage-marked Cxcl12-GFP+ bone marrow stromal cells.

(a) Flow cytometry analysis of CD45/Ter119/ CD31neg cells at P21. Gating strategy for bone marrow cells isolated from Cxcl12GFP/+; Hes1-creER; R26RtdTomato femurs (pulsed at E10.5, E12.5 or E14.5), Cxcl12GFP/+; Dlx5-creER; R26RtdTomato femurs (pulsed at E12.5), Cxcl12GFP/+; Osx-creER; R26RtdTomato femurs (pulsed at E12.5) and Cxcl12GFP/+; Fgfr3-creER; R26RtdTomato femurs (pulsed at E12.5). n=5 (Hes1-E12.5), n=4 (Dlx5-E12.5), n=5 (Osx-E12.5), n=10 (Fgfr3- E12.5), n=6 (Hes1-E10.5), n=4 (Hes1-E14.5) mice per group. (b) Percentage of Col1(2.3kb)-GFP+tdTomato+ osteoblasts per total Col1(2.3kb)-GFP+ osteoblasts. n=3 (Hes1-E12.5), n=5 (Fgfr3-E12.5) mice. *p<0.05, two-tailed, Mann-Whitney’s U-test. Data are presented as mean ± s.d.

17 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Figure 10. Combined lineage-tracing and single cell RNA-seq analyses of bone marrow stromal cells derived from fetal perichondrial cells and chondrocytes.

(a-d) Combined lineage-tracing and single cell RNA-seq analyses of Cxcl12-GFP+ bone marrow stromal cells and tdTomato+ mesenchymal cells derived from fetal Hes1+ perichondrial cells and Fgfr3+ chondrocytes. Both single and double positive cells (GFPhigh, tdTomato+ and GFPhightdTomato+) were FACS-isolated from Cxcl12GFP/+; Hes1-creER; R26RtdTomato and Cxcl12GFP/+; Fgfr3-creER; R26RtdTomato femurs at P21, pulsed at E12.5 (Hes1-E12.5 or Fgfr3- E12.5). Two single cell datasets were integrated by LIGER. (a): t-SNE-based visualization of major classes of mesenchymal Cxcl12-GFPhigh and/or tdTomato+ (Hes1-E12.5 or Fgfr3-E12.5) cells (Cluster 0, 2, 4, 5, 9, 13, 15). n=4,437 cells. Pooled from n=4 mice per group. (b,c): Feature plots. Cluster 13: Osteoblasts. Cluster 9: Osteoblast precursors. Cluster 0, 2, 4, 5: Cxcl12+ reticular cells. Blue: high expression. (d): Split-dot-based visualization of representative gene expression. Circle size: percentage of cells expressing a given gene in a given cluster (0 – 100%), Color density: expression level of a given gene. (e) Flow cytometry-based analysis of Cxcl12-GFPhightdTomato+ cells (Hes1-E12.5 or Fgfr3- E12.5). Percentage of CD90+, CD105+ and CD200+ cells per total Cxcl12-GFPhightdTomato+ cells. n=4 per group. Two-tailed, Mann-Whitney’s U-test. Data are presented as mean ± s.d.

18 bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. bioRxiv preprint doi: https://doi.org/10.1101/2020.03.13.990853; this version posted March 14, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

Supplementary Figure 11. Comparative RNA-seq analysis of bone marrow stromal cells derived from fetal perichondrial cells and chondrocytes.

(a-c) Comparative bulk RNA-seq analysis of Cxcl12-GFPhigh stromal cells derived from fetal Hes1+ perichondrial cells and Fgfr3 + chondrocytes. (a): FACS-sorting of Cxcl12- GFPhightdTomato+ (Hes1-E12.5 and Fgfr3-E12.5) cells. Gating strategy for bone marrow cells isolated from Cxcl12GFP/+; Hes1-creER; R26RtdTomato and Cxcl12GFP/+; Fgfr3-creER; R26RtdTomato mice, pulsed at E12.5. (b): Principal component analysis (PCA) plot of four biological samples. Triangles: Cxcl12-GFPhighHes1-E12.5 cells, circles: Cxcl12-GFPhighFgfr3-E12.5 cells. x-axis: PC1, 79% variance, y-axis: PC2, 15% variance. n=2 biological replicates (Hes1-E12.5: pooled

from n=4 mice, Fgfr3-E12.5: pooled from n=5 mice). (c): MA plot, x-axis: log2FPKM, y-axis: high log2fold-change. y > 0.58 represents genes enriched in Cxcl12-GFP Hes1-E12.5 (red), y < - 0.58 represents genes enriched in Cxcl12-GFPhighFgfr3-E12.5 (blue). (d): Volcano plot, x-axis:

log2fold-change, y-axis: -log10adjusted p-value. x > 0.58 represents genes enriched in Cxcl12- GFPhighHes1-E12.5 (red), x < - 0.58 represents genes enriched in Cxcl12-GFPhighFgfr3-E12.5 (blue). Each dot represents a gene.