RSPO3 is important for trabecular bone and fracture risk in mice and humans

With increasing age of the population, countries across the globe are facing a substantial increase in osteoporotic fractures. Genetic association signals for fractures have been reported at the RSPO3 locus, but the causal gene and the underlying mechanism are unknown. Here we show that the fracture reducing allele at the RSPO3 locus associate with increased RSPO3 expression both at the mRNA and protein levels, increased trabecular bone mineral density and reduced risk mainly of distal forearm fractures in humans. We also demonstrate that RSPO3 is expressed in osteoprogenitor cells and osteoblasts and that osteoblast-derived RSPO3 is the principal source of RSPO3 in bone and an important regulator of vertebral trabecular bone mass and bone strength in adult mice. Mechanistic studies revealed that RSPO3 in a cell-autonomous manner increases osteoblast proliferation and differentiation. In conclusion, RSPO3 regulates vertebral trabecular bone mass and bone strength in mice and fracture risk in humans.

O steoporosis is a common skeletal disease, leading to a reduction in bone density and quality, and an increased fracture risk. One in two elderly women and one in four elderly men will at some point suffer from an osteoporotic fracture 1,2 . It is predicted that the number of people aged 65 years and older will triple globally by year 2100 3,4 as compared to those aged 12-64 years, and the incidence of fractures increases exponentially with age 5 . Therefore, the identification and treatment of patients at high fracture risk is an important public health goal.
Fracture risk is a moderately heritable trait 6,7 , and two recent large-scale genome-wide association studies (GWAS) have identified 15 loci associated with fractures at any bone site 3,8 . The identified fracture loci were all also associated with BMD, supporting the notion that BMD is a major regulator of fracture risk 3,8 . These previous large-scale human genetic studies reported that the strongest genetic determinant for fractures is located at the WNT16 locus and detailed translational studies revealed that osteoblast-derived WNT16 increases cortical bone mass [9][10][11] . The second strongest genetic determinant for fractures at any bone site (P = 8.9 × 10 −65 ) 8 is located at the RSPO3 locus. RSPO3 is a known WNT-signaling modulator 12,13 , and WNT-signaling is a major determinant of bone density and strength. We therefore hypothesized in this study that RSPO3 is the causal gene for the prominent fracture signal at the RSPO3 locus.
Previous human genetic studies have reported that the fracture reducing C-allele of rs7741021, the strongest genetic determinant for fractures at the RSPO3 locus, also is robustly associated with increased BMD (Table 1a) 8 . However, the causal gene(s) and the underlying mechanism are unknown. In addition, it is unknown if its associations with fractures at different bone sites differ and if the association with BMD is mediated via trabecular and/or cortical bone parameters. We, herein, used a combination of human genetic studies and extensive functional mechanistic studies to identify the causal gene at the RSPO3 locus and its mechanism to explain its effect upon fracture.

Results
Human genetic studies indicate that increased RSPO3 expression is associated with increased trabecular bone mineral density and reduced risk of distal forearm fractures. Previous large-scale human genetic studies have reported that the strongest genetic determinant at the RSPO3 locus for fractures at any bone site is located at the intronic SNP rs7741021 (Table 1a; P = 8.9 × 10 −65 ) 8 . A substantial part of the SNPs associated with evaluated phenotypes in GWAS is located in intronic regions and may directly or via highly linked SNPs affect regulatory regions and thereby gene transcription 32 . We, therefore, evaluated the association between rs7741021 and mRNA levels of RSPO3 using the GTEx resource. We observed that the C-allele of rs7741021, previously reported to associate with reduced fracture risk at any bone site and increased BMD (estimated BMD as determinated by ultrasound in the heel; Table 1a), was associated with increased RSPO3 mRNA levels in both subcutaneous adipose tissue (P = 2.4 × 10 −4 ; n = 581) and cultured fibroblasts (P = 1.8 × 10 −4 ; n = 483; Table 1a). Unfortunately, no bone-related tissue was available in the GTEx resource. From human genetic association studies, it is not possible to determine if a factor such as RSPO3 exerts local effects within the bone tissue. However, it is possible to determine if the genetic signals that associate with bone health parameters also associate with circulating levels of RSPO3. It should be emphasized that such association does not show that the effect on bone is mediated via circulating levels of RSPO3 as the circulating levels may only reflect the local levels, without being of biological importance as a systemic regulator of local bone metabolism.
We observed that SNP rs3734626, which is highly linked with SNP rs7741021 (r 2 = 0.84; coefficient of linkage disequilibrium measure, D′ = 0.99) at the RSPO3 locus, was strongly associated with circulating protein levels of RSPO3 (P = 1.8 × 10 −45 ; Table 1b) 33 . The association for SNP rs3734626 with the circulating protein levels of RSPO3 was derived from a previous report, demonstrating that this SNP was the most significant cisprotein quantitative trait loci (cis-pQTLs) for RSPO3 33 . The T-allele of rs3734626, associated with reduced fracture risk at any bone site and increased BMD, was associated with increased circulating levels of RSPO3 (Table 1b). Collectively, these human genetic studies indicate that the main fracture signal at the RSPO3 locus regulates RSPO3 expression and that the fracture-reducing allele associates with both increased RSPO3 expression and BMD.
We next evaluated the impact of rs7741021 on trabecular and cortical bone parameters analyzed separately by pQCT in the tibia in humans 9,34 . The fracture reducing C-allele of rs7741021 was significantly associated with increased trabecular volumetric BMD, while no significant association was observed with cortical volumetric BMD or cortical bone thickness (Table 1a).
As the distal radius is a bone compartment with relatively high proportion of trabecular bone, we next extended the fracture association studies by evaluating the associations for rs7741021 specifically with distal forearm fractures, defined by ICD codes, in the large UK Biobank cohort. The C-allele of rs7741021 was robustly associated with reduced risk of forearm fractures (OR 0.88, 95% confidence intervals (CI) 0.86-0.91 per C-allele; P = 9.6 × 10 −14 ; Table 1a) in age, age 2 , sex, weight and height adjusted analyses. We also evaluated the association between rs7741021 and hip fractures, defined by ICD codes. The C-allele of rs7741021 was associated with reduced risk of hip fractures (OR 0.94, 95% CI 0.90-0.99 per C-allele; P = 4.4 × 10 −3 ; Table 1a). The effect size for the association of rs7741021 with distal forearm fractures was significantly larger than the effect sizes for the association of rs7741021 with fractures at any bone site (Table 1a, OR 0.95, 95% CI 0.94-0.96 per C-allele; P = 1.6 × 10 −5 ; Z-test) and hip fractures (P = 3.5 × 10 −2 ; Z-test).
To estimate the impact of RSPO3 on bone parameters and fracture risk, we performed two-sample inverse varianceweighted (IVW) Mendelian randomization (MR) studies. We used either the above described strong cis-pQTL rs3734626 for circulating RSPO3 identified by Emilsson et al. 33 (Table 1b) or the linked SNP rs2489623 (r 2 = 0.74, D′ = 0.88; Table 1c), a cis-pQTL for circulating RSPO3 (P = 3.6 × 10 −28 ) identified by Sun et al. in an independent dataset 35 , as genetic instruments for circulating protein levels of RSPO3. These MR analyses indicated that increased circulating RSPO3 was strongly associated with increased trabecular vBMD, and reduced risk of distal forearm fractures, and as expected, with very similar effect sizes for the two genetic instruments (Table 2). However, it should be emphasized that this association might be completely driven by circulating RSPO3 being associated with local RSPO3 levels, and from these human associations studies it is, therefore, not possible to determine possible local effects of factors acting on bone remodeling. Thus, to determine the relative role of circulating RSPO3 and local RSPO3, functional studies in mice with cellspecific inactivation of RSPO3 is required.
To exclude reverse causality, wherein BMD influences levels of RSPO3, we performed a bidirectional MR using BMD as the exposure and circulating RSPO3 as the outcome. These studies revealed no evidence of reverse causality (P = 0.80).
Colocalization analyses may be used to determine if there is a shared genetic signal between two traits at a specific gene locus. Traditional colocalization analyses require that only one independent association signal exists at a specific locus. Because colocalization inference are very sensitive to small linkage disequilibrium (LD) changes in the tested populations, multiple causal signals presented in the same region may introduce  (Table S4) 35 (a) Betas are given in SD (or ORs) per effect allele (=C-allele). *The effect sizes for the eQTLs are given as normalized effect size according to the GTEx portal (https://gtexportal.org/home). eBMD = estimated bone mineral density measured by ultrasound in the heel. Trabecular vBMD = Trabecular volumetric BMD as measured by pQCT in the distal tibia metaphysis. Cortical vBMD = Cortical volumetric BMD as measured by pQCT in the tibia diaphysis. Cortical thickness = Cortical thickness measured by pQCT in the diaphysis of tibia. eQTL = expression quantitative trait locus. The effect size for the association of rs7741021 with forearm fractures was significantly higher than the effect size for the association of rs7741021 with fractures at any bone site (P = 1.6 × 10 −5 ; Z-test) and hip fractures (P = 3.5 × 10 −2 ; Z-test). The statistical tests were two-sided.
(b) Betas are given in SD (or ORs) per effect allele (=T-allele). eBMD = estimated bone mineral density measured by ultrasound in the heel. Trabecular vBMD = Trabecular volumetric BMD as measured by pQCT in the distal tibia metaphysis. Cortical vBMD = Cortical volumetric BMD as measured by pQCT in the tibia diaphysis. Cortical thickness = Cortical thickness measured by pQCT in the diaphysis of tibia. pQTL = protein quantitative trait locus. The effect size for the association of rs3734626 with forearm fractures was significantly higher than the effect size for the association of rs3734626 with fractures at any bone site (P = 2.1 × 10 −3 ; Z-test) and hip fractures (P = 3.0 × 10 −2 ; Z-test). The statistical tests were two-sided.
(c) Betas are given in SD (or ORs) per effect allele (=C-allele). eBMD = estimated bone mineral density measured by ultrasound in the heel. Trabecular vBMD = Trabecular volumetric BMD as measured by pQCT in the distal tibia metaphysis. Cortical vBMD = Cortical volumetric BMD as measured by pQCT in the tibia diaphysis. Cortical thickness = Cortical thickness measured by pQCT in the diaphysis of tibia. pQTL = protein quantitative trait locus. The effect size for the association of rs2489623 with forearm fractures was significantly higher than the effect size for the association of rs2489623 with fractures at any bone site (P = 2.1 × 10 −3 ; Z-test) and hip fractures (P = 3.0 × 10 −2 ; Z-test). The statistical tests were two-sided.
substantial bias 36 . As only one independent association signal was observed for trabecular volumetric BMD at the RSPO3 locus ( Supplementary Fig. 1), we used colocalization analyses to determine if there is a shared genetic signal for circulating RSPO3 and trabecular volumetric BMD at this locus. Bayesian colocalization analyses using COLOC 37 found that the plasma RSPO3 signal identified by Sun et al. was reasonably well colocalized with the single signal observed for trabecular volumetric BMD at the RSPO3 locus with a posterior probability of a single shared signal of 72% (posterior probability of the hypothesis 4 that the association with trait 1 and trait 2 has one shared SNP 37 , PP.H4 = 0.72), indicating that there is a single shared genetic signal in the 1MB locus of the RSPO3 cis-pQTL SNP (rs2489623), affecting both circulating RSPO3 and trabecular volumetric BMD ( Supplementary Fig. 1). Since there are multiple independent signals for both estimated BMD (eBMD) and forearm fractures at the RSPO3 locus, and since we do not have access to any individual level data to assume accurate LD, we did not undertake colocalization analyses for these parameters. These human genetic studies demonstrate that the strong fracture signal at the RSPO3 locus has the capacity to regulate RSPO3 expression, and that increased circulating RSPO3 levels associate with increased trabecular BMD, substantially reduced risk of distal forearm fractures and moderately reduced risk of hip fractures. Based on these human genetic data, we hypothesized that RSPO3 increases trabecular bone mass and thereby reduces fracture risk. To test this hypothesis, we performed extensive mechanistic studies using several conditional Rspo3-inactivated mouse models and cultured bone-derived cells.
RSPO3 is expressed in osteoprogenitor cells and osteoblasts but not in osteocytes or osteoclasts. We observed Rspo3 mRNA expression in several mouse tissues, revealing the highest Rspo3 expression in brain cortex and diaphyseal cortical bone with intermediate expression in the vertebral body and modest expression in flushed bone marrow (Fig. 1a). Rspo3 mRNA levels were high in primary calvarial osteoblasts after prolonged culture in osteogenic media ( Fig. 1b; Supplementary Fig. 2a). In contrast, no Rspo3 expression was observed in bone marrow-derived macrophages cultured with M-CSF or osteoclasts differentiated from bone marrow-derived macrophages stimulated with RANKL ( Fig. 1b; Supplementary Fig. 2b).
Next, the RSPO3 expression in mouse and human bone sections was evaluated using chromogenic in situ hybridization ( Fig. 1c-i). Assessment of Rspo3 expression in mouse vertebral bone revealed that Rspo3 mRNA is expressed in Runx2-expressing (Fig. 1c), Col1a1-expressing (Fig. 1d) and Dmp1expressing ( Fig. 1e) osteoblast-lineage cells on bone surface. In contrast, Rspo3 expression was not detectable in osteocytes ( Fig. 1e) or osteoclasts (Fig. 1f). Rspo3 expression was also observed in a subset of cells close to blood vessels and adipocytes in the bone marrow ( Supplementary Fig. 2c,d). Analyses of human bone sections ( Fig. 1g-i) confirmed moderate RSPO3 mRNA expression in immature osteoblast lineage cells, on eroded lining surfaces (reversal cells) and quiescent bone surfaces (bone lining cells) (Fig. 1g), and in a subset of cells close to blood vessels and adipocytes (Fig. 1h). No staining for RSPO3 mRNA expression was observed in osteocytes or osteoclasts in human sections (Fig. 1i).
Thus, in addition to expression in osteoblast-lineage cells on bone surfaces, RSPO3 expression was found in a subset of cells close to blood vessels, where mesenchymal skeletal stem cells are located within the bone marrow 38 . To determine Rspo3 expression specifically in mesenchymal stem cells in the bone marrow, we analyzed two independent single cell RNA-sequencing (scRNA seq) data sets of bone marrow cells. First, scRNA seq analysis of Cxcl12 expressing bone marrow cells showed that Rspo3 was expressed in stromal cells expressing both adipogenic and osteoblastic markers ( Fig. 2a-c) 39 . These cell clusters also expressed Lepr ( Supplementary Fig. 3), often used in combination with Cxcl12 as a marker for mesenchymal stem cells, supporting the notion that Rspo3 is expressed also in mesenchymal stem cells. No Rspo3 expression was found in Cd45-expressing hematopoietic cell clusters (Fig. 2a-c).
The expression of Rspo3 in Cxcl12 + Lepr + mesenchymal stem cells was further confirmed in a second recently published scRNA seq dataset ( Supplementary Fig. 4) 40 . This dataset includes scRNA seq analysis of cells expressing Cdh5 (found in vasculature), Lepr (perivascular stromal stem cells) and Col1a1 (osteoblastic cells). Using this dataset, we identified Rspo3 expression in stromal cells expressing both adipogenic and osteogenic markers ( Supplementary Fig. 4). Rspo3 mRNA was expressed also in pro-osteogenic Lepr + cells expressing Alpl and in the more mature osteoblast lineage Col1a1-expressing cells. No Rspo3 expression was found in the vascular cell clusters. Collectively, these two scRNA seq data sets of bone marrow cells reveal robust Rspo3 expression in mesenchymal stem cells expressing both osteogenic and adipogenic markers.
Osteoblast-derived RSPO3 is the principal source of RSPO3 in bone and is a major regulator of trabecular bone mass. As global Rspo3 −/− mice display early embryonic lethality 29 , and since we observed robust Rspo3 expression in osteoblast-lineage cells, we first evaluated the impact of osteoblast-derived RSPO3 for trabecular and cortical bone parameters. To this end, we generated a conditional Rspo3-inactivated mouse model targeting osteoblasts and osteocytes. A mouse model with exons 2-4 of Rspo3 flanked by LoxP sites (Rspo3 flox/flox , Fig. 3a) was used 41 . To achieve inactivation of Rspo3 early in the osteoblast lineage, we generated Runx2-creRspo3 flox/flox mice, which express Cre recombinase driven by the Runx2 promoter, a promoter expressed in early osteoblast lineage cells but not in osteoclasts ( Fig. 3a) 42 . Runx2-creRspo3 flox/flox mice had more than 90% reduced Rspo3 mRNA levels in both vertebral trabecular bone and tibial cortical diaphyseal bone with no reduction in brain cortex or liver compared to Rspo3 flox/flox mice (Fig. 3b). Thus, osteoblast-derived RSPO3 is the principal source of RSPO3 in both trabecular and cortical bones. Runx2-creRspo3 flox/flox mice were born apparently healthy and had a normal body weight growth at termination at 12 weeks of age and displayed normal weights of several visceral organs (Fig. 3c, d).  Fig. 5a-d). Furthermore, the trabecular BV/TV was positively correlated with Rspo3 mRNA levels in the vertebral body (r 2 = 0.76, P < 0.001; Fig. 3j), supporting the notion that local RSPO3 expression regulates trabecular bone mass. To evaluate whether the decreased trabecular bone mass in mice with Rspo3 inactivation in osteoblast-lineage cells resulted in decreased bone strength, lumbar vertebra L5 was evaluated using a compression test, demonstrating that the maximal load at failure was significantly decreased in Runx2-creRspo3 flox/flox mice compared with Rspo3 flox/flox mice (Fig. 3k). No significant effect of osteoblast-lineage specific Rspo3 inactivation was Fig. 1 RSPO3 is expressed in osteoblasts but not in osteocytes or osteoclasts. a Rspo3 mRNA expression pattern in various tissues of male, wildtype mice (n = 6). Mean of individual values are presented as horizontal lines and ±SEM as vertical lines. Data are presented as % of expression in cortical diaphyseal bone. b Rspo3 is expressed in primary calvarial osteoblasts (OBL), but not in bone marrow-derived macrophages (BMM) or RANKL-differentiated osteoclasts (OCL). Rspo3 gene expression increase from day 2 (d2) to day 7 (d7) in primary calvarial osteoblasts cultured in osteogenic media. Presented as % of expression in osteoblasts day 2. ND=not detectable. Mean of individual values are presented as horizontal lines and ±SEM as vertical lines. Difference in expression between day 2 and 7 was analyzed by two-sided Student's t test, n = 4 wells per group. c-f Representative in situ hybridization images in mice. Transverse sections across lumbar vertebra 5 in mouse showing the mRNA expression pattern (red) of Rspo3 (c-f, upper) and Runx2 (c, lower), Col1a1 (d, lower), Dmp1 (e, lower), and Acp5/Trap (f, lower). The lower images are consecutive to the upper images. Rspo3 mRNA could be observed in Runx2-expressing (c), Col1a1-expressing (d), and Dmp1-expressing (e) osteoblast lineage cells (OBL) on the bone surface. In contrast, Rspo3 was not detectable in osteocytes (Ocy; e), or osteoclasts (OCL; f). Scale bar 50 µm. g-i Representative in situ hybridization images in humans. Sections from human proximal femur (g) and iliac crest (h, i) confirmed moderate RSPO3 mRNA expression. Staining for RSPO3 mRNA (red) were observed in immature osteoblast lineage cells (OBL) on eroded lining surfaces (reversal cells), and on quiescent bone surfaces (bone lining cells), as well as in a subset of cells close to blood vessels (bv; g, h). No RSPO3 mRNA expression was observed in osteocytes (Ocy) or TRAP-immunostained osteoclasts (OCL, black; i). Scale bar 20 µm. Experiments were repeated one (a), two (g-i) or at least three (b-f) times. Source data are provided as a Source Data file. observed on cortical volumetric BMD or cortical thickness in the femur diaphysis (Fig. 3l, m).
Expression analyses in the vertebral body revealed reduced mRNA levels of the two WNT signaling inhibitors Sost and Dkk1 (Fig. 3n), suggesting a compensatory mechanism to counteract additional trabecular bone loss in the adult Runx2-creRspo3 flox/flox mice with life-long inactivation of Rspo3 in osteoblast-lineage cells. No compensatory regulation of Rspo1, -2 or -4 was observed (Fig. 3n).
Circulating levels of the bone formation marker procollagen type I N-terminal propeptide (P1NP; Fig. 3o, left) were reduced while the levels of the bone resorption marker C-terminal type I collagen fragments (CTX; Fig. 3n, right) were unchanged in We next determined if RSPO3 with an impact on trabecular bone is derived from osteoblast precursors/early osteoblasts or from osteocytes/late osteoblasts. To this end, we compared the phenotype of Runx2-creRspo3 flox/flox mice, having inactivation of Rspo3 in all osteoblast-lineage cells, with the phenotype of a mouse model with Rspo3 inactivated specifically in Dmp1expressing osteocytes and late osteoblasts. Rspo3 flox/flox mice were mated with Dmp1-cre mice expressing Cre recombinase specifically in late osteoblasts and osteocytes 43 , hereafter called Dmp1cre-Rspo3 flox/flox mice (Fig. 4a). Dmp1cre-Rspo3 flox/flox mice displayed a moderate reduction of Rspo3 mRNA levels in the vertebral body compared with Rspo3 flox/flox mice (−32%; P < 0.001; Fig. 4b). The substantially less pronounced reduction of Rspo3 mRNA levels in the vertebral body of Dmp1cre-Rspo3 flox/ flox mice compared with the substantial reduction (>90%, P < 0.001) in Runx2-creRspo3 flox/flox mice demonstrates that osteoblasts, and not osteocytes, are the major source of RSPO3 in vertebral trabecular bone (Figs. 3b and 4b).
Dmp1cre-Rspo3 flox/flox mice were born apparently healthy and had a normal body weight growth at termination at 15 weeks of age and displayed normal weights of several visceral organs (Fig. 4c, d). CT analyses revealed modestly reduced trabecular bone volume fraction (−9.3%; BV/TV, Fig. 4e-i) in Dmp1cre-Rspo3 flox/flox mice compared with Rspo3 flox/flox mice, reaching statistical significance in a two-way ANOVA considering genotype and sex (P = 0.027 for genotype). The reduction of trabecular bone volume fraction, however, was substantially more pronounced in the Runx2-creRspo3 flox/flox mice compared with the reduction observed in Dmp1cre-Rspo3 flox/flox mice (Figs. 3f, 4e), demonstrating that RSPO3 with an impact on trabecular bone is mainly osteoblast-and not osteocyte-derived. As expected, no effect on cortical bone density or cortical bone thickness was observed in the Dmp1cre-Rspo3 flox/flox mice (Fig. 4j,  k).
Inducible inactivation of Rspo3 reduces trabecular bone volume fraction. Although very informative, the studies using life-long osteoblast-lineage specific Rspo3 inactivation cannot separate between early developmental effects of RSPO3 and its effects on adult bone metabolism. To evaluate the effect of RSPO3 specifically on adult bone homeostasis and the underlying mechanism, before a compensatory new steady state of bone remodeling has been reached, we developed a mouse model with tamoxifen-inducible Rspo3 inactivation in adult mice. To this end, we bred Rspo3 flox/flox mice with CAGGCre-ER transgenic mice expressing a tamoxifen-inducible Cre-mediated recombination system (Fig. 4l) 44 . We have previously shown that CAGGCre-ER mice do not have a skeletal phenotype 45 . All mice evaluated were treated with the same tamoxifen regime and the phenotype of CAGGCre-ER-Rspo3 flox/flox and Rspo3 flox/flox mice were compared (Fig. 4l). The Rspo3 mRNA levels were lower (−38%, P = 0.05) in the vertebral body of tamoxifen-treated CAGGCre-ER-Rspo3 flox/flox mice compared with tamoxifentreated Rspo3 flox/flox mice (Fig. 4m). To determine the skeletal phenotype in the inducible Rspo3 knockout mice, 8-week-old Rspo3 flox/flox and CAGGCre-ER-Rspo3 flox/flox mice were treated with tamoxifen and the phenotype was evaluated 38 days after the first tamoxifen dose. The inducible inactivation of Rspo3 did not affect body weight or weights of several visceral organs (Fig. 4n,  o). Analyses of the vertebral trabecular bone volume fraction revealed a significant reduction both when evaluated using CT (BV/TV; −12.3%, P = 0.009, Fig. 4p-s) and histomorphometry (−18.0%, P = 0.038; Supplementary Fig. 6a-d) in tamoxifentreated CAGGCre-ER-Rspo3 flox/flox mice compared with tamoxifen-treated Rspo3 flox/flox mice. Inducible Rspo3 inactivation did not affect cortical bone density or cortical bone thickness (Fig. 4t, u). Expression analyses in the vertebral body revealed that inducible Rspo3 inactivation reduced Dkk1 mRNA levels (−32%, P = 0.049; Fig. 4v), suggesting a compensatory mechanism to counteract additional trabecular bone loss. Dynamic histomorphometry of vertebral trabecular bone demonstrated a 31% reduction of trabecular bone formation rate per tissue volume (P = 0.047, Supplementary Fig. 6e) and similar effect sizes were observed for number of osteoblasts per trabecular bone perimeter (−32%, P = 0.088; Supplementary Fig. 6f) and for levels of the circulating bone formation marker P1NP (−20%, P = 0.067; Supplementary Fig. 6g) in tamoxifen-treated CAGGCre-ER-Rspo3 flox/flox mice compared with tamoxifen-treated Rspo3 flox/ flox mice. In contrast, there was no indication of bone resorption being affected in mice with inducible Rspo3 inactivation (Supplementary Fig. 6f, g). To further explore the role of osteoblastderived RSPO3 on bone formation, we performed extensive studies using different osteoblast cell cultures.
To achieve a more efficient Rspo3 inactivation in calvarial osteoblasts, we used cells isolated from CAGGCre-ER-Rspo3 flox/flox mice treated with tamoxifen in vitro, resulting in an almost allele complete inactivation of Rspo3 mRNA expression (Fig. 5a), whereas tamoxifen-treated cells from Rspo3 flox/flox mice had unaltered Rspo3 mRNA levels ( Supplementary Fig. 7b). When these calvarial osteoblasts were cultured in osteogenic medium it was observed that inactivation of Rspo3 resulted in decreased osteoblast differentiation as assessed by substantially decreased ALP staining in culture using cells from tamoxifen-treated CAGGCre-ER-Rspo3 flox/flox mice as compared to genotypically identical cells not treated with tamoxifen, or with cells from Rspo3 flox/flox mice incubated in the absence or presence of tamoxifen (Fig. 5b). Importantly, osteoblasts from tamoxifen-treated CAGGCre-ER-Rspo3 flox/flox mice had almost completely lost the capability to form mineralized noduli, in contrast to cells from the same genotype not treated with tamoxifen (Fig. 5c, d), or cells from Rspo3 flox/flox mice with or without tamoxifen (Fig. 5c, Supplementary Fig. 7c). The effects on ALP activity and bone noduli formation were associated with lower expression of Alpl and Col1a1 mRNA in tamoxifen-treated CAGGCre-ER-Rspo3 flox/flox cells (Fig. 5e,f) as compared to the absence of effect by tamoxifen in cells from Rspo3 flox/flox mice ( Supplementary Fig. 7d, e). Deletion of Rspo3 in tamoxifen-treated cells from CAGGCre-ER-Rspo3 flox/flox mice resulted in less dense cell cultures due to decreased cell numbers as demonstrated by decreased amounts of DNA, an effect absent in tamoxifen-treated cells from Rspo3 flox/ flox mice or cells from Rspo3 flox/flox mice with or without tamoxifen (Fig. 5g, h, Supplementary Fig. 7f). Reduced cell numbers were due to an effect on cell proliferation as demonstrated by lower mRNA expression of Mki67, a wellestablished marker for cell proliferation (Fig. 5i, Supplementary  Fig. 7g). Collectively, these studies demonstrate that RSPO3 enhances osteoblast proliferation and differentiation in a cellautonomous manner.
In agreement with the in vivo observations, RSPO3 did not affect osteoclast formation as demonstrated by the observations that the osteoclastogenic response to PTH in the calvarial osteoblasts (increased Tnfsf11 and decreased Tnfrsf11b mRNA) was not affected by deletion of Rspo3 (Fig. 5j, k, Supplementary  Fig. 7h, i), nor did addition of recombinant RSPO3 protein affect osteoclast formation induced by RANKL in bone marrow macrophage cultures (Fig. 5l-n).
RSPO3 increases osteoblast proliferation and differentiation through WNT-β-catenin signaling. Through their Furin-like domain 1, RSPOs bind to extracellular domains of the E3 ligases ZNRF3/RNF43, which are important for ubiquitination and proteasomal degradation of LRP6 and Frizzleds 46 . When the ternary complex of RSPOs, LGRs and ZNRF3/RNF43 is formed, ZNRF3/RNF43 is endocytosed and degraded, resulting in decreased breakdown of Frizzleds and LRP6, and consequently to increased sensitivity to WNT canonical signaling 47,48 . We found that deletion of Rspo3 in tamoxifen-treated osteoblasts from CAGGCre-ER-Rspo3 flox/flox mice resulted in decreased amounts of total LRP6 as well as pLRP6 (Fig. 6a-d). In contrast, tamoxifen treatment did not affect total amounts of LRP6 in osteoblasts from Rspo3 flox/flox mice ( Supplementary Fig. 8a, b). Increased WNT canonical signaling is associated with enhanced mRNA expression of negative regulators such as the ligases Znrf3 and Rnf43 47, 48 . We noticed that the mRNA expression of Znrf3 and Rnf43 was decreased in tamoxifen-treated osteoblasts from CAGGCre-ER-Rspo3 flox/flox mice (Fig. 6e, f), whereas tamoxifen did not affect their expression in osteoblasts from Rspo3 flox/flox mice ( Supplementary Fig. 8c, d). These observations suggest that deletion of Rspo3 in the osteoblasts results in decreased WNT canonical signaling. To directly assess if downregulation of LRP6 and decreased phosphorylation of LRP6 reduced WNT canonical intracellular signaling, we analyzed the mRNA expression of Tcf7 and Lef1, two well-known WNT canonical target genes 49 . It was found that the mRNA expressions of both Tcf7 and Lef1 were significantly decreased in osteoblasts from CAGGCre-ER-Rspo3 flox/flox mice treated with tamoxifen (Fig. 6g, h). In contrast, tamoxifen treatment of osteoblasts from Rspo3 flox/flox mice did not affect Tcf7 or Lef1 mRNA expression (Supplementary Fig. 8e, f).

Discussion
Bone fractures are a major public health concern. Recent largescale human genetic studies have identified 15 loci associated with fractures at any bone site and the understanding of the causal genes and underlying mechanisms for these fracture signals may result in improved prevention and/or treatment of patients with high fracture risk 3,8 . A major genetic determinant for fractures at any bone site is located at the RSPO3 locus. We, herein, used a combination of human genetic studies and extensive functional mechanistic studies to identify the causal gene and its mechanism for the fracture signal at the RSPO3 locus. Human genetic studies revealed that the strong fracture association signal at the RSPO3 locus has the capacity to regulate RSPO3 expression, and that increased circulating RSPO3 levels associate with increased trabecular BMD and substantially reduced risk of distal forearm fractures. Studies using several different conditional Rspo3-inactivated mouse models revealed that osteoblast-derived RSPO3 is the principal source of RSPO3 in bone and a major regulator of trabecular bone volume fraction in mice. Further functional studies revealed that osteoblast-derived RSPO3 improves bone strength in vertebrae of mice. Mechanistic studies showed that osteoblast-derived RSPO3 increases osteoblast proliferation and differentiation via cell-autonomous effects, involving WNT canonical signaling. Given that the RSPO3 locus confers amongst the largest effects on fracture risk in the human genome, these findings point to a protein, whose effects may be a major contributor to fracture risk in humans.
Several previous studies have reported that genetic determinants at the RSPO3 locus are robustly associated with fractures at any bone site, BMD as measured by DXA at the hip and/or spine, and estimated BMD as determined by ultrasound in the  9,34 . An effect mediated via trabecular bone mass was supported by our findings from larg -scale bone site-specific fracture association analyses, revealing a pronounced association for the genetic signal at the RSPO3 locus with fractures at the distal forearm, a bone site dependent on trabecular bone. These human genetic studies, however, did not determine if RSPO3 is the causal gene. Our first evidence of RSPO3 being the causal gene was the observations that the prominent fracture association signal at the RSPO3 locus was associated with RSPO3 mRNA levels in both subcutaneous adipose tissue and cultured fibroblasts and circulating RSPO3 protein levels in two independent cohorts 33,35 . Importantly, the fracture reducing allele at the RSPO3 locus was associated with increased RSPO3 expression both at the mRNA and protein levels and increased trabecular volumetric BMD in the distal tibia. Based on these human genetic data, we hypothesized that RSPO3 increases trabecular bone mass and thereby reduces fracture risk. To test this hypothesis, we performed extensive mechanistic studies using several conditional Rspo3-inactivated mouse models and cultured bone-derived cells.
In descriptive studies, we first demonstrated that substantial RSPO3 expression was observed in both cortical and trabecular bone. Detailed cell-specific expression analyses using chromogenic in situ hybridization, single cell RNA sequencing of bone marrow cells and cultured bone-derived cells, revealed RSPO3 expression in osteoprogenitor cells and osteoblasts but not in osteocytes or osteoclasts in both humans and mice. The single cell RNA sequencing data showed RSPO3 expression in mesenchymal stem cells in the bone marrow, supporting previous studies describing RSPO3 expression in different types of stem cells 56 . These descriptive findings were verified by functional studies using mouse models with inactivation of Rspo3 in osteoblastlineage cells. To determine the source of RSPO3 in bone, we developed and compared conditional Rspo3-inactivated mouse models with inactivation of Rspo3 early in both osteoblasts and osteocytes (Runx2-cre) or in late osteoblasts and osteocytes (Dmp1-cre). Rspo3 mRNA levels in bone were almost completely abolished in the mice with inactivation of Rspo3 early in both osteoblasts and osteocytes while only marginal reduction was observed in mice with inactivation only in late osteoblasts and osteocytes. These findings demonstrate that osteoblasts and not osteocytes or osteoclasts are the major source of RSPO3 in both trabecular and cortical bone.
Importantly, inactivation of Rspo3 in both early osteoblasts and osteocytes resulted in a marked reduction of the trabecular bone volume fraction in the vertebrae, whereas inactivation of Rspo3 in late osteoblasts and osteocytes only marginally reduced the trabecular bone volume fraction, demonstrating that osteoblastderived, but not osteocyte-derived RSPO3, is a major regulator of trabecular bone mass. Furthermore, the trabecular bone volume fraction was strongly directly correlated with Rspo3 mRNA levels in the vertebral body, supporting the notion that local Rspo3 expression regulates trabecular bone mass in mice.
Although informative, the studies using life-long osteoblastspecific RSPO3 inactivation could not exclude early developmental effects, confounding the adult phenotype. Therefore, to evaluate the effect of RSPO3 on adult trabecular bone homeostasis, we developed a mouse model with tamoxifen-inducible Rspo3 inactivation in adult mice. A reduction of trabecular bone volume fraction was observed 5.5 weeks after the inducible Rspo3 inactivation in adult mice, excluding confounding developmental effects and demonstrating that endogenous RSPO3 exerts crucial effects on trabecular bone homeostasis in adult mice. In contrast, no effect on cortical bone thickness or cortical volumetric BMD was observed in mice with life-long osteoblastlineage specific Rspo3 inactivation or global adult inducible inactivation of Rspo3. These findings are in accordance with the human genetic finding in the present study, demonstrating that the most prominent fracture signal at the RSPO3 locus was associated with trabecular volumetric BMD, but not with cortical bone thickness or cortical volumetric BMD. Collectively, these findings indicate that RSPO3 mainly affects trabecular bone mass in both humans and mice.
It is well established that trabecular and cortical bone may respond differently to treatments 1 , and that WNT signaling has the capacity to regulate both the trabecular and the cortical bone compartments 10,[17][18][19]57,58 . Previous studies have demonstrated that osteoblast-derived WNT16 is the major known cortical bonespecific WNT and we recently reported that the secreted WNT lipase NOTUM is a major regulator specifically of cortical bone mass with no effect on trabecular bone 10,20,45,59 . In contrast, we, herein, report that the extracellular WNT-signaling modulator RSPO3 mainly regulates trabecular bone mass. Similarly, WNT10b protects against age-dependent trabecular bone loss 19 . Thus, the data presented in the present study corroborate the Fig. 5 Osteoblast-derived RSPO3 increases osteoblast proliferation and differentiation. a Rspo3 mRNA expression in primary calvarial osteoblasts (cOBL) isolated from CAGGCre-ER-Rspo3 flox/flox mice cultured for 2 and 6 days (d) with or without prior Rspo3 inactivation using tamoxifen (TAM). Veh=vehicle. b Representative photos of alkaline phosphatase staining of CAGGCre-ER-Rspo3 flox/flox and Rspo3 flox/flox cOBL cultured for 6 days after veh treatment or Rspo3 inactivation. c, d Representative photos (c) and quantification (d) of mineralized noduli in cultures of CAGGCre-ER-Rspo3 flox/flox and Rspo3 flox/flox cOBL cultured for 14 days after veh treatment or Rspo3 inactivation. e, f mRNA expression analysis of Alpl (e) and Col1a1 (f) in CAGGCre-ER-Rspo3 flox/flox cOBL cultured for 2 and 6 days with or without prior Rspo3 inactivation. g Haematoxylin (Htx) and Dapi staining of CAGGCre-ER-Rspo3 flox/flox cOBL cultured for 3 days with or without prior Rspo3 inactivation. Scale bar 100 μm. h Amount of DNA per well in CAGGCre-ER-Rspo3 flox/flox cOBL at the time of TAM addition (day −1), directly after removal of TAM (day 0) and after culture for 1, 3, 6, and 10 days. i Mki67 mRNA expression in CAGGCre-ER-Rspo3 flox/flox cOBL cultured for 2 days after Rspo3 inactivation. j, k mRNA expression of Tnfsf11 (j) and Tnfrsf11b (k) in CAGGCre-ER-Rspo3 flox/flox cOBL cultured with or without PTH and PGE2 for 5 days with or without prior Rspo3 inactivation. n = 3-4 wells. l-n TRAP staining (l), counting (m), and Acp5 mRNA expression (n) of bone marrow macrophages cultured in M-CSF (M) or in M-CSF and RANKL (RL) to induce osteoclastogenesis in the absence or presence of recombinant RSPO3 (rRSPO3). Scale bar 200 μm. Individual values are presented in all graphs with the mean presented as horizontal lines and ±SEM as vertical lines. If not otherwise stated, n = 4 wells per group. Experiments were repeated two (g, j-n) or at least three (a-f, h, i) times. For a, e, f, h, two-way ANOVA was used to determine the overall effect of Rspo3 inactivation by treatment of TAM or veh, time, as well as their interaction. For j, k, twoway ANOVA was used to determine the overall effect of PTH/PGE2 treatment, Rspo3 inactivation by treatment of TAM or veh, as well as their interaction. When only the effect of Rspo3 inactivation was evaluated, two-sided Student's t was used. Source data are provided as a Source Data file. clinical findings and suggest that the cortical and trabecular compartments are indeed biologically different, their homeostasis being independently regulated by different signaling molecules, all of which are involved in the regulation of bone mass 10 .
Mechanistic studies using a mouse model with inducible inactivation of Rspo3 in adult mice, and inducible Rspo3 inactivation in cultured osteoblasts, collectively revealed that osteoblast-derived RSPO3 increases osteoblast proliferation and differentiation in a cell-autonomous manner. A stimulatory effect in vivo of RSPO3 on bone formation is supported by a recent phase 1 clinical trial, evaluating the safety of different doses of an anti-RSPO3 antibody (OMP-131R10/rosmantuzumab) in patients with advanced solid tumors. High doses of the anti-RSPO3 antibody reduced serum levels of the bone formation marker P1NP in a majority of the evaluated subjects, indicating that RSPO3 enhances bone formation also in humans (https:// publications.oncomed.com/Ph-1ab-OMP-131R10-anti-RSPO3adv-solid-tumor-prev-treat-met-CRC-2016-EORTC-NCI-AACR. pdf Downloaded 2020.11.24).
RSPOs are potent enhancers of WNT signaling by making up a ternary complex with LGR receptors 4-6 and the E3 ubiquitin ligases ZNRF3/RNF43, through their furin-like domains 2 and 1, respectively 30,31,46 . A role of LGR4 for bone mass is suggested by the observations that a rare nonsense mutation within the LGR4 gene was reported to be strongly associated with low BMD and with osteoporotic fractures in Icelandic individuals 60 , and that global deletion of Lgr4 in mice results in reduced vertebral trabecular bone volume due to reduced bone formation 61 , suggesting that RSPO3 might influence bone formation in trabecular bone via activation of LGR4. Formation of the ternary complex results in membrane clearance of ZNRF3/RNF43 and thereby decreased proteasomal degradation of the WNT receptors Frizzleds and the Frizzled co-receptor LRP6 47,48 , known to be important in WNT canonical signaling 49 . In a feedback loop, the mRNA expressions of Znrf3 and Rnf43 are increased 47,48 . We found that LRP6 protein was substantially decreased, and that the mRNA expressions of Znrf3 and Rnf43 were downregulated in osteoblasts in which Rspo3 had been deleted, suggesting that WNT canonical signaling was decreased as a consequence of Rspo3 deletion. Interestingly, LRP6 has previously been reported to be important for trabecular bone and osteoblast differentiation 62 . The fact that the mRNA expressions of Tcf7 and Lef1, two classical signature genes induced by WNT canonical signaling 49 , were decreased in osteoblasts with deletion of Rspo3, provides direct evidence that WNT canonical signaling is enhanced by RSPO3. These findings indicate that RSPO3 stimulates osteoblast proliferation and differentiation through enhanced WNT canonical signaling. Similar to our present observations for RSPO3, it has been reported that also RSPO1 and RSPO2 can stimulate osteoblast differentiation in vitro and that deletion of Rspo2 in osteocalcinexpressing osteoblasts reduces trabecular bone mass due to decreased bone formation 22,63 . However, human genetic studies have not demonstrated any strong association between genetic variants at the RSPO1 or RSPO2 loci and bone mass or fracture susceptibility, suggesting that RSPO3 is the major RSPOregulating bone mass and fracture susceptibility in humans. A previous report claimed that RSPO3 regulates the differentiation of human adipose-derived stem cells via an activation of LGR4 64 . However, one effect of RSPO3-LGR4 activation in adiposederived stem cells in vitro in that study was to inhibit their osteogenic potential while a robust stimulatory effect of RSPO3 on osteoblast differentiation and proliferation of osteoprogenitors/osteoblasts both in in vivo and in vitro was observed in the present study. Differential effects of RSPO3 with capacity to both increase and reduce adipogenesis, depending on the origin of adipose progenitors, have recently been described for abdominalderived and gluteal-derived adipose progenitors 65 . Thus, RSPO3 may either enhance or reduce proliferation and/or differentiation in a cell-type specific and differentiation stage-dependent manner. In the present study, we observed RSPO3 expression in adipocytes within the bone marrow, in both mouse and human, as well as in scRNA seq of bone marrow adipogenic stromal cells. However, the functional role of RSPO3 in bone marrow adipocytes needs to be evaluated in future studies.
In conclusion, using a combination of human genetic studies and extensive functional mechanistic studies, we identified the causal gene and its mechanism for the prominent fracture association signal at the RSPO3 locus. We provide human genetic evidence that the fracture reducing allele at that locus increases RSPO3 expression, circulating RSPO3 level and trabecular bone mass and that it mainly reduces the risk of distal forearm fractures. The underlying mechanism includes a cell-autonomous effect of RSPO3 on osteoblast proliferation and differentiation.

Methods
Human genetic association studies Look ups of fracture signals at the RSPO3 locus in different available data sets. Fractures at any bone site: Association with fractures at any bone site for the most significant SNP at the RSPO3 locus (rs7741021) is available in Table S4 from Morris et al. 8 while the associations for the other evaluated SNPs at this locus (rs3734626 and rs2489623) are taken from the GWAS summary statistics for fractures at any bone site, available at the GEFOS website http://www.gefos.org.
Estimated BMD (eBMD) at the heel analyzed by ultrasound: Associations with eBMD for the evaluated SNPs at the RSPO3 locus are taken from the GWAS summary statistics from Morris et al. 8 , available at the GEFOS website http://www. gefos.org.
Trabecular vBMD, cortical vBMD and cortical thickness as analyzed by peripheral quantitative computed tomography (pQCT): Associations for the evaluated SNPs at the RSPO3 locus with trabecular vBMD in the distal metaphyseal region of tibia and cortical vBMD in the tibia diaphyseal region are taken from the GWAS meta-analyses summary statistics from Paternoster et al. 34 , while the association with cortical bone thickness in the tibia diaphyseal region are taken from Zheng et al. 9 .
eQTLs: Associations with eQTLs for the three SNPs at the RSPO3 locus were evaluated using the GTEx Portal https://www.gtexportal.org/home/ cis-pQTL: The associations for the most significant cis-protein QTLs (cis-pQTLs) for RSPO3 were taken from two independent data sets, including rs3734626 from Emilsson et al. (Supplemental Table 13) 33 and rs2489623 from Sun et al. (Supplementary Table 4) 35 .
New associations analyses of candidate SNPs at the RSPO3 locus with distal forearm fractures and hip fractures. Our new analyses of associations between candidate SNPs at the RSPO3 locus with distal forearm fractures (defined by ICD10 codes S52.5 and S52.6; N = 7324 distal forearm fracture cases) and hip fractures (defined by ICD10 codes S72.0, S72.1 and S72.2; N = 4035 hip fracture cases) were performed using data from the UK Biobank study (http://www.ukbiobank.ac.uk). Briefly, the UK Biobank is a large prospective cohort study of approximately a halfmillion adult (ages 40-69 years) participants living in the United Kingdom, recruited from 22 centers across the United Kingdom in 2006-2010 66 . We included 438,756 participants of white European descent with valid data on the candidate SNPs at the RSPO3 locus, the two fracture outcomes (distal forearm fractures and hip fractures) and relevant covariates (age, sex, height, weight). The UK Biobank has ethical approval from the North West Multi-Centre Research Ethics Committee (application no. 16/NW/0274), and informed consent was obtained from all participants. The present research was approved by the UK Biobank Research and Access Committee (application no. 51784).
Associations between the three candidate SNPs at the RSPO3 locus and the two fracture outcomes (distal forearm fractures and hip fractures) were performed using BOLT-LMM 67 . Briefly, this method uses a linear mixed model to account for relatedness and population structure. Fracture risk was corrected for age, age 2 , height, weight, sex, genotyping array and PC1 + … + PC20 in logistic regression models. Individuals were excluded based on unusually high heterozygosity or >5% missing genotype rate, a mismatch between self-reported and genetically inferred sex.
Mendelian randomization (MR). We performed MR analyses to determine the causal associations of RSPO3 on fracture risk and bone parameters using inverse variance-weighted (IVW) MR. The most significant cis-pQTL, either rs3734626 from Emilsson et al. 33 , or rs2489623 from the independent data set of Sun et al. 35 , was used as genetic instrument for circulating RSPO3. Genetic associations with the exposure (RSPO3) and outcomes (fractures and bone parameter) used in the Mendelian randomization are presented in Table 1b (rs3734626) and Table 1c (rs2489623). As genetic variants are randomly distributed at birth, they are unaffected by confounders. We then regressed the association of the cis-pQTL SNP on the outcome measure, weighing its effect by the magnitude of its effect upon the corresponding exposure.
In order to reduce potential horizontal pleiotropy, we used only cis-SNPs for serum RSPO3 levels. These cis-SNPs for RSPO3 were defined in the published GWAS as the lead SNPs within 300 kB, in the data set of Emilsson et al. 33 , or 1 MB, in the independent data set of Sun et al. 35 , of the RSPO3 gene. Neither of the cis-SNPs at the RSPO3 locus from the two different sources were associated with any other circulating protein levels included in the extensive circulating proteomic analyses, reducing the risk of potential horizontal pleiotropy. To assess the possibility of reverse causality that bone mass regulates circulating RSPO3, a bidirectional MR using eBMD as the exposure and circulating RSPO3 from Sun et al. 35 as the outcome was performed, using 1107 LD-independent SNPs from Morris et al. 8 as instrumental variables for eBMD, followed by IVW to metaanalyze their combined effect on the RSPO3 levels.
Colocalization. We tested for colocalization of the genetic signal for circulating RSPO3 and trabecular volumetric eBMD using colocalization analyses, which assesses potential confounding by linkage disequilibrium. A stringent Bayesian analysis implemented in COLOC R package was performed to estimate the posterior probability (PP) that the same causal signal in the 1 MB genomic locus centered on the cis-SNP affects both circulating RSPO3 and trabecular volumetric BMD 37 . Summary statistics for the RSPO3 locus were available for the cis-pQTL identified in the study by Sun et al. but not for the cis-pQTL identified in the study by Emilsson et al. 33 and consequently the analysis was done only for the cis-pQTL identified by Sun et al. 35 .
Animal experiments. Cell-specific and inducible Rspo3-inactivated mouse models were generated by breeding Rspo3 flox/flox mice (Rspo3 tm1.1Jcob /J, JAX stock #027313, Jackson Laboratories) 41 with mice expressing cre recombinase under the control of different promoters. In the Rspo3 flox mice, LoxP sites are introduced upstream from Rspo3 exon 2 and downstream from Rspo3 exon 4 (Fig. 3a) and in the presence of an active cre recombinase, the DNA fragment including exons 2-4 of the Rspo3 allele is excised.
All procedures involving animals were approved by the Ethics Committee in Gothenburg, Västra Götaland, and the care of the animals was in compliance with all relevant ethical regulations for animal testing and research. The mice were housed in a standard animal facility with a 12 h dark/light period, at standard temperature and humidity. Food and water were available ad libitum. At termination, the animals were anesthetized with Ketador/Dexdomitor (Richter Pharma/Orion Pharma), bled, and euthanized by cervical dislocation. Long bones and vertebrae were dissected and stored for future analyses, soft tissues were collected, weighed, and snap-frozen in liquid nitrogen.
Generation of Runx2-creRspo3 flox/flox mice. To generate specific inactivation of Rspo3 in the osteoblast-lineage, Rspo3 flox/flox mice were bred with Runx2-cre mice 10,42 . The Runx2-creRspo3 flox/flox offspring display early osteoblast-specific cre expression and have the capacity to recombine LoxP-flanked DNA sequences in an early osteoblast-specific manner and recombination occurs at all sites of endochondral and intramembranous bone formation, particularly in periosteal cells, osteoblasts and osteocytes but not osteoclasts, when Runx2-cre mice were crossed to a Rosa26 reporter strain 42 . The littermate control mice were Rspo3 flox/flox . Runx2-cre mice have an unchanged skeletal phenotype compared to WT mice 10 .
Generation of Dmp1cre-Rspo3 flox/flox mice. To generate mice depleted of Rspo3 in late osteoblasts and osteocytes, Rspo3 flox/flox female mice were crossed with heterozygous Rspo3 +/flox male mice expressing cre recombinase driven by the 10-kb Dmp1 promoter specifically expressed in late osteoblasts and osteocytes (Tg(Dmp1cre)1Jqfe mice) (Dmp1-cre mice) 43 . The littermate control mice were Rspo3 flox/flox . We have previously demonstrated that efficient recombination occurs in osteocytes but not osteoblasts or osteoclasts when Dmp1-cre mice were crossed to a Rosa26 reporter strain 68 .
Generation of mice with inducible Rspo3 inactivation. In order to study the acute adult effects of Rspo3, avoiding confounding developmental effects, inducible Rspo3 knockout mice were created by breeding the Rspo3 flox/flox mice with the CAGGCre-ER (B6.Cg-Tg(CAG-cre/Esr1*)5Amc/J, JAX stock #004682, Jackson Laboratories) transgenic mice 44,69 . The CAGGCre-ER-Rspo3 flox/flox offspring express a tamoxifen-inducible cre-mediated recombinase. To study the effects of Rspo3 on the adult bone tissue, a long-term study using adult CAGGCre-ER-Rspo3 flox/flox mice were performed. Male 8-week-old CAGGCre-ER-Rspo3 flox/flox and Rspo3 flox/ flox littermate control mice were injected with tamoxifen (50 mg kg −1 ; Sigma-Aldrich) for three consecutive days to activate the cre construct. The mice were terminated 38 days after the first tamoxifen injection at the age of 13.5 weeks.
The different mouse models were genotyped using the primers listed in Supplementary Table 1.
In situ hybridization. The chromogenic in situ hybridization was performed on vertebra L5 from 5-month-old mice, on 8 G needle biopsies from the posterior iliac crest of human healthy volunteers, as well as proximal femur specimens obtained from coxa valga patients undergoing corrective surgery. The bone specimens were fixated in 4% formaldehyde in phosphate buffered saline, followed by demineralization in 15% EDTA with 0.4% PFA for 3 weeks, and paraffin embedded. All human subjects have given their written informed consent, and the study was conducted according to the World Medical Association Declaration of Helsinki and approved by the Danish National Committee on Biomedical Research Ethics, journal no. S-20110112 and S-20120193. 3.5-µm-thick (human) and 6-µm-thick (mouse) consecutive dewaxed sections were in situ hybridized using a modified protocol of the RNAscope 2.5 HD manual protocol (322300, Advanced Cell Diagnostics, Bio-Techne Ltd., Abingdon, UK) 70 . The following probes were used: mouse R-spondin 3 (Rspo3, 402011, target sequence 731-2164), human R-spondin 3 (RSPO3, 490581, target sequence 38-1546), mouse Runt-related transcription factor 2 (Runx2, 414021, target sequence 3838-4821), mouse collagen 1a1 (Col1a1, 319371, target sequence 1686-4669), mouse dentin matrix protein 1 (Dmp1, 441171, target sequence 689-1543), and mouse tartrate resistant acid phosphatase analyzed in the nicheExplorer made available by Iannis Aifantis lab (http:// aifantislab.com/niche) in an interactive data browser as a supplement to their published scRNAseq data of mouse bone marrow vascular, perivascular, and osteoblast cell populations 40 .

Assessment of bone parameters
Bone histomorphometry. For dynamic histomorphometry, the mice were intraperitoneally injected with the fluorochromes calcein and alizarin (Merck GmbH, Darmstadt, Germany) 9 and 2 days prior to termination, respectively 10 . Upon dissection, vertebra L5 were fixated in 4% formaldehyde, dehydrated in EtOH, and imbedded in methyl methacrylate. Static trabecular bone parameters were determined in a 4-μm-thick plastic sections stained in Masson-Goldner Trichrome and dynamic trabecular bone parameters were determined in unstained 8-μm-thick sections. All parameters were analyzed using the OsteoMeasure7 histomorphometry system (OsteoMetrics, Atlanta, GA, USA), following the guidelines of the American Society for Bone and Mineral Research 71 .
High resolution micro-computed tomography. High-resolution micro-computed tomography (µCT) were used to analyze lumbar vertebra 5 (Skyscan, 1172; Bruker MicroCT, Aartselaar, Belgium) 10 . The lumbar vertebra were imaged with an X-ray tube voltage of 50 kV and a current of 201 µA, with a 0.5 mm aluminum filter, and the scanning angular rotation was 180°, and the angular increment was 0.70°. NRecon (version 1.6.9.8, Bruker MicroCT) was used to perform reconstruction after scans. The trabecular bone was analyzed 235 µm from the lower end of the pedicles and continued for approximately 229 µm. The data was analyzed using the CTAn software (version 1.13.2.1, Bruker MicroCT).
Peripheral quantitative computed tomography. Computed tomography (pQCT) scans were performed using the peripheral quantitative computed tomography XCT Research M (v.4,5B; Norland Stratec, Pforzheim, Germany), at a resolution of 70 µm 72 . The cortical bone parameters were analyzed in the mid-diaphyseal region of the left femur to determine cortical volumetric bone mineral density (BMD) and the cortical thickness 73 .
Mechanical strength. Intact lumbar vertebra 5 was axially loaded with a press head of 2 mm in diameter, with a 1 mm-thick holder through the vertebral foramen, at a speed of 0.155 mm s −1 using a mechanical testing machine (Instron 3366, Instron) 74 . The results were calculated by a custom Excel macro using the computer recorded load deformation raw data curves produced by Bluehill 2 software v2.6 (Instron).
Murine primary calvarial osteoblast culture. Murine primary calvarial osteoblasts were isolated from 4 to 6 days old mice by sequential collagenase treatments 75 . Cells in collagenase fractions 6-10 were cultured in complete α-MEM in T75 flasks for 4 days prior to re-seeding the cells at 20,000 cells cm −2 in 48-well plates. Osteoblasts were cultured in complete osteogenic α-MEM for indicated times with change of half of the culture media volume every 2-3 days. For gene expression analyses, cells were lysed in RNeasy lysis (RLT) buffer with βmercaptoethanol (Qiagen), followed by RNeasy micro RNA purification (Qiagen), cDNA synthesis and real-time PCR as described above. DNA was isolated using the DNeasy Blood and Tissue kit (Qiagen) following the manufacturer's instruction for animal cells. For ALP staining, cells were fixed in citrate buffered acetone and stained according to the commercial kit protocol (Sigma, 85L2-1KT). Mineralized nodules were visualized by fixation in 2.5% glutaraldehyde in 70% ethanol for 5 min, followed by three washes in 70% ethanol. For quantification of mineralization, images were taken using a Jenoptik Gryphax camera connected to a Nikon Eclipse 80i microscope and mineralized surface per well was quantified using the Bioquant OSTEO software (version 20.8.6 volume 1993).
In vitro ablation of Rspo3 in cultured calvarial osteoblasts. Primary calvarial osteoblasts from CAGGCre-ER-Rspo3 flox/flox and Rspo3 flox/flox control mice were isolated as above and cultured in complete α-MEM for 2-3 days prior to re-seeding in multiwell plates and culture in complete osteogenic media. Cells seeded in 48-well plates were cultured in osteogenic media for 24 h, followed by addition of tamoxifen (Sigma, ReadyMade solution, SML1666) for additional 24 h, before complete media change and continued culture in complete osteogenic media without tamoxifen. 50 or 100 nM tamoxifen was used and resulted in similar results. Osteoblastic differentiation was followed and analyzed as described above with the time of tamoxifen removal designated day 0. PTH (10 nM, Bachem, H-1660) and PGE2 (1 µM, R&D Systems, 2296) was added the day after tamoxifen removal (day 1) and continued until day 6, when cells were harvested for gene expression analysis as described above.
Murine bone marrow macrophages and osteoclasts cultures. Mouse bone marrow derived macrophages (BMM) were obtained from 10 to 12 weeks old male wild type C57Bl/6N mice by culture of bone marrow cells in suspension culture dishes in complete α-MEM supplemented with 30 ng ml −1 M-CSF (R&D Systems, 416-;L-050) 76,77 . BMMs were spot seeded in 48-well plates (20,000 cells in 20 μl complete α-MEM) and allowed to attach for 5 min before addition of complete α-MEM supplemented with 30 ng ml −1 M-CSF, with or without 4 ng ml −1 RANKL (R&D Systems, 462-TEC-010) and 100 ng ml −1 recombinant RSPO3 (R&D Systems, 4120-RS-025) for 3 days to induce osteoclast differentiation. Cells were stained for tartrate resistant acid phosphatase (TRAP) after 3 days using a commercial kit (Sigma, 387A-KT1). TRAP-positive cells with at least three nuclei were defined as osteoclasts and counted using the Bioquant OSTEO software. For gene expression analyses, cells were lysed in RLT buffer with β-mercaptoethanol (Qiagen), followed by RNeasy micro RNA purification (Qiagen), cDNA synthesis and real-time PCR as described above.
Protein preparation and analyses. Primary calvarial osteoblasts from CAGGCre-ER-Rspo3 flox/flox and Rspo3 flox/flox control mice were isolated as above and cultured in complete α-MEM for 2-3 days prior to re-seeding in 12-well plates at 20,000 cells cm −2 . Cells were then cultured in osteogenic media for 24 h, after which 100 nM tamoxifen was added for 24 h, followed by complete media change to osteogenic media without tamoxifen. Protein lysates were prepared 24 h after removal of tamoxifen by washing the cells once with cold PBS followed by addition and lysis in RIPA buffer (Sigma, R0278) with protease inhibitors (complete Mini EDTA-free; RocheDiagnostics, 5892970001) and phosphatase inhibitors (PhosSTOP; Roche Diagnostics, 04906845001). Lysates were transferred to tubes, centrifuged for 10 min at 11,000×g, +4°C, and the supernatant was used for protein analysis. Protein concentration was quantified using the DC Protein Assay Kit (Bio-Rad, 500-0112). Levels of LRP6 and pLRP6 were analyzed by capillary-based electrophoresis and immunodetection using the JESS ProteinSimple system and the Compass for SW Software (version 5.0.0, Protein Simple) as described by the manufacturer. Protein lysates were analyzed with anti-LRP6 (clone C47E12, Cell Signaling Technology, 3395, dilution 1:250) or anti-pLRP6 (Ser1490, Cell Signaling Technology, 2568, dilution 1:10) together with the Protein Normalization Assay Module (Pro-teinSimple, AM-PN01) and the Anti-rabbit Detection Module (ProteinSimple, DM-001) on 12-230 kDa capillary separation cassettes (ProteinSimple, SM-W004). LRP6 and pLRP6 are visually presented in the lane view of the Compass Software and quantified by normalization to total protein levels according to the Simple-Western protocol and following the guidelines for presentation and quantification of protein data by Journal of Biological Chemistry 78 .