Bone marrow adipogenic lineage precursors are the major regulator of bone resorption in adult mice

Bone resorption by osteoclasts is a critical step in bone remodeling, a process important for maintaining bone homeostasis and repairing injured bone. We previously identified a bone marrow mesenchymal subpopulation, marrow adipogenic lineage precursors (MALPs), and showed that its production of RANKL stimulates bone resorption in young mice using Adipoq-Cre. To exclude developmental defects and to investigate the role of MALPs-derived RANKL in adult bone, we generated inducible reporter mice (Adipoq-CreER Tomato) and RANKL deficient mice (Adipoq-CreER RANKLflox/flox, iCKO). Single cell-RNA sequencing data analysis, lineage tracing, and in situ hybridization revealed that Adipoq+ cells contain not only MALPs but also late mesenchymal progenitors capable of osteogenic differentiation. However, RANKLmRNA was only detected in MALPs, but not in osteogenic cells. RANKL deficiency in MALPs induced at 3 months of age rapidly increased trabecular bone mass in long bones as well as vertebrae within 1 month due to diminished bone resorption but had no effect on the cortical bone. Ovariectomy (OVX) induced trabecular bone loss at both sites. RANKL depletion either before OVX or at 6 weeks post OVX protected and restored trabecular bone mass. Furthermore, bone healing after drill-hole injury was delayed in iCKO mice. Together, our findings demonstrate that MALPs play a dominant role in controlling trabecular bone resorption and that RANKL from MALPs is essential for trabecular bone turnover in adult bone homeostasis, postmenopausal bone loss, and injury repair.


Introduction
Bone is critical for protecting internal organs, supporting the body, allowing movement, as well as hosting hematopoiesis.To maintain its essential structure and functions, bone undergoes continuous remodeling-a cyclic process involving osteoclastic bone resorption and osteoblastic/osteocytic bone formation 1 .In healthy adults, this remodeling process is preciselly balanced to preserve normal bone mass.In aged and diseased populations, this balance is shifted towards resorption rather than formation, leading to osteoporosis characterized by low bone mass, deteriorated bone structure, and high risk of fracture 1 .Following injuries such as fractures, bone remodeling is a crucial process that facilitates the bridging of the fracture gap 2 .
One important aspect of bone remodeling is to initiate osteoclast formation at the trabecular or cortical bone surface for bone resorption.Descended from myeloid progenitors of the hematopoietic lineage, osteoclasts are highly speci c, large multinucleated, and phagocytic cells secreting acid and catalytic enzymes to demineralize and degrade collagenrich bone extracellular matrix (ECM) 3 .For a long time, they were considered to be short-lived cells undergoing apoptosis quickly after fusion from mononuclear progenitors 4 .Recent in vivo research using advanced intravital imaging techniques overturned this dogma and discovered that they are actually long-lived cells constantly undergoing recycling through ssion and fusion mechanism [5][6][7] .Instead of apoptosis, the ssion products, osteomorphs, can refuse among each other or with existing osteoclasts to extend the longevity of osteoclasts 6 .
Past research has pointed out two cytokines as the most important regulatory factors for osteoclast formation and function: colony-stimulating factor (Csf1) and receptor activator of nuclear factor kappa Β ligand (RANKL).The former one promotes the proliferation of osteoclast precursors and their expression of receptor activator of nuclear factor kappa Β (RANK), a RANKL receptor 8 .The latter one is the predominant factor that drives the differentiation of osteoclast precursors into mature osteoclasts 9 .In addition, it stimulates osteoclasts to migrate and undergo cycles of fusion and ssion in vivo 6 .Encoded by Tnfsf11, RANKL belongs to tumor necrosis factor (TNF) superfamily and exists in two forms: membrane-bound and soluble 10 .Upon binding to RANK, it initiates the transcription of a cascade of osteoclast speci c genes via up-regulating the expression of a master transcription factor nuclear factor of activated T-cells, cytoplasmic 1 (NFATc1) 11 .Early research identi ed osteogenic cells, particularly bone matrix-embedded osteocytes, are the major source of RANKL that regulates osteoclast formation [12][13][14] .This nding ts well with the concept of bone remodeling as it emphasizes the crosstalk between bone forming and resorbing cells.
In the past several years, the application of advanced single cell transcriptomics approaches to bone research greatly expanded our knowledge about cellular components of bone tissue and their transcriptome pro les.In particular, single cell RNA-sequencing (scRNA-seq) of mesenchymal lineage cells revealed a new mesenchymal subpopulation that highly expresses adipogenic markers, including Pparg, Cebpa, Lpl, Adipq, etc., but does not contain lipid droplets 15,16 .Since they are precursors for mature adipocytes, we termed them marrow adipogenic lineage precursors (MALP) 15,17 .Other groups also identi ed similar cell types and named them Adipo-Cxcl12-abundant-reticular (Adipo-CAR) cells or marrow Adipoq + cells (MACs) 16,18 .Interestingly, scRNA-seq suggested that MALPs, but not osteoblasts nor osteocytes, are the major source of RANKL and Csf1 in bone 19,20 .Subsequent studies from our group and other groups con rmed that speci c deletion of either one of the factors in adipogenic lineage cells results in a drastically elevated trabecular bone mass due to diminished osteoclast number [19][20][21][22] .
However, those studies have limitations because they used a constitutive Cre, Adipoq-Cre, to examine the action of MALPs-derived factors.MALPs emerge in mouse bone right after birth 23 .While those studies analyzed mice up to 6 months of age, it is possible that changes in adult bone is due to developmental defect.In addition, since Adipoq-Cre also labels bone forming cells in adult mice, Adipoq + cells are considered to be bipotent bone marrow skeletal stem/progenitor cells 24 .Hence, we cannot exclude the possibility that osteoclast regulatory factors are also depleted in osteoblasts and osteocytes in those studies.To circumvent these limitations, in this study, we rst performed a lineage tracing experiment using inducible Adipoq-CreER Tomato (AdipoqER/Td) mice to delineate the relationship between MALPs and Adipoq + cells.Next, we adopted RNA uorescence in situ hybridization (FISH) to identify RANKL-expressing cells in vivo.Finally, we constructed inducible RANKL de cient mice using Adipoq-CreER and examined their adult bones under normal, estrogen depletion, and injury repair conditions.Our data revealed MALPs as the main source of RANKL in adult mice and demonstrated its essential role in controlling adult bone homeostasis, disorder, and healing.

Results
Adipoq + cells contain not only MALPs but also late, bipotent mesenchymal progenitors.
Since Adipoq is a marker for MALPs, we previously used Adipoq-Cre to study MALPs in vivo.To examine whether Adipoq + cells contain other progenitors, we integrated scRNA-seq datasets of bone marrow mesenchymal cells from 1-and 16-month-old mice we reported before (Fig. 1A) 15 .Pseudotime trajectory analysis revealed that early mesenchymal progenitors (EMPs) give rise to late mesenchymal progenitors (LMPs) and lineage committed progenitors (LCPs), which are then differentiated into either adipogenic lineage (MALPs) or osteogenic lineage (osteoblasts and osteocytes) (Fig. 1B).Violin plots clearly showed that while Adipoq is highly expressed in MALPs, it is also expressed in LCPs followed by osteoblasts and osteocytes in 1 month dataset at very low levels (Fig. 1C).Interestingly, EMPs started to express Adipoq at 16 months of age, albeit the level was low.
To analyze Adipoq + cells in adult mice, we generated inducible Td reporter mice driven by Adipoq-CreER.These mice, AdipoqER/Td, at 3 months of age received daily Tamoxifen (Tam) injections from day 1 to 3. At day 7, many Td + cells were observed inside the long bone (Fig. 1Da).Within the metaphysis region, Td + cells were made of 73.5 ± 1.0% Cd45-stromal cells, 2.7 ± 0.2% Perilipin + adipocytes, 9.0 ± 0.4% pericytes, and 14.8 ± 1.2% bone lining cells (Fig. 1Db-e, n = 5 mice).Although some bone surface lining cells were also Td+, they did not express Osterix, an osteogenic cell marker (Fig. 1De, f).Particularly at the endocortical bone surface, we observed a lot of Td + cells in the close proximity to Osterix + osteoblasts.Td did not label chondrocytes, osteocytes or periosteal cells (Fig. 1Df, g).In addition, in situ staining of Pparg, the master transcriptional factor for adipogenic differentiation 25 and another marker for MALPs 15 , showed that it is only expressed in Td + cells (Fig. 1E).These data indicate that Adipoq-CreER targets MALPs, but not bone forming cells (osteoblasts and osteocytes).Furthermore, CFU-F assay showed that almost all CFU-F colonies are Td-(Fig.1F, G), suggesting that Adipoq + cells lack the proliferation ability required by early progenitors.
To determine the fate of Adipoq + cells, we harvested long bones of AdipoqER/Td mice at 1, 4, 8 and 12 weeks post the rst Tam injection for lineage tracing experiment (Fig. 2A, B).Perilipin staining revealed that Td labels nearly all mature adipocytes throughout the tracing period.On the contrary, Td gradually labeled osteoblasts and osteocytes over time.While no Td + osteoblasts and osteocytes were detected at 1 week in both trabecular and cortical bone, the percentages of Td + osteoblasts increased to 46.1%, 77.1%, and 92.1% and the percentages of Td + osteocytes increased to 9.9%, 12.4%, and 27.6% in the trabecular bone at 4, 8 and 12 weeks, respectively.In the cortical bone, almost all endosteal osteoblasts became Td + after 4 weeks but few osteocytes (3.4%) became Td + even after 12 weeks of tracing.These data suggest that Adipoq + cells contain not only committed adipo-lineage cells but also uncommitted mesenchymal progenitors capable of osteogenic differentiation.
We noticed that Td + cells are not evenly distributed through the bone marrow.Thus, we counted them at four anatomic sites: subchondral bone, top metaphysis (region close to the growth plate), bottom metaphysis (region distal to the growth plate), and diaphysis (Fig. 2C).Interestingly, we found that the density of bone marrow Td + cells (excluding bone surface and embedded cells) is drastically reduced in the midshaft region compared to the trabecular bone region.During the 3-month tracing period, Td + cells in the area with high trabecular bone volume (subchondral bone and top metaphysis) remained unaltered, but Td + cells in the area with low trabecular bone volume (bottom metaphysis and diaphysis) decreased signi cantly (Fig. 2C, D).These data indicate that mesenchymal progenitors labeled by Adipoq-CreER are not early-stage progenitors with self-renewal ability.
MALPs are the major producer of osteoclast regulator factors in adult bone.
Our previous scRNA-seq of mouse bone marrow predicted that MALPs are the major producers of osteoclast regulatory factors, including RANKL and Csf1 15 .We recently pro led bone marrow from human femoral heads.Cell clustering revealed 6 mesenchymal cell clusters: Fibro-MSC (mesenchymal stromal cell), APOD+-MSC, Adipo-MSC, THY1+-MSC, Osteo-MSC, and Osteoblast (Fig. 3A).Among these clusters, Adipo-MSC and THY1+-MSC highly expressed adipogenic genes, and a major difference between them was THY1 expression level.Thus, we consider them both as human counterpart of MALPs (Fig. 3B).In line with mouse data, RANKL (TNFSF11) was mainly expressed in THY1-MSCs, albeit the level was low compared to mice.CSF1 was mainly expressed in Adipo-and THY1-MSCs followed by Fibro-MSCs.Their expression in osteolineage cells was much lower than in MALPs.We also examined their expression in other bone marrow cells (Fig. S1).While RANKL expression was restricted in mesenchymal lineage cells, CSF1 expression was broader, which also includes megakaryocyte-erythroid progenitor (MEP), erythroblasts, basophil/eosinophil/mast Cell (Ba/Eo/Ma), Vessel cells etc.However, the highest expression was still detected in Adipo-MSCs.
To con rm this nding, we stained RANKL in situ on 3-month-old AdipoqER/Td mouse femurs harvested at day 7 after the rst Tam injection.Interestingly, almost all RANKL-expressing cells were Td+ (Fig. 3C, n = 5 mice).Most of them resided in the metaphyseal and diaphyseal bone marrow and some were on the trabecular and cortical bone surface.Moreover, co-staining showed that RANKL + cells were also Pparg + cells (Fig. 3D).On the contrary, only 60 ± 0.8% of Csf1-expressing cells were Td+ (Fig. 3E, n = 5 mice).Importantly, we did not observe any Rankl and Csf1 mRNA expression in osteocytes in either trabecular bone or cortical bone, demonstrating that MALPs, but not osteogenic cells, are the major cell source of osteoclast regulatory factors.

MALP-derived RANKL supports bone resorption in adult mice.
To investigate the role of MALP-derived RANKL in adult bone remodeling, we constructed Adipoq-CreER RANKL ox/ ox (RANKL iCKO) mice.At 3 months of age, these mice displayed similar trabecular and cortical bone structures in femurs and vertebrae as WT siblings (Fig. S2).Next, we subjected both WT and iCKO mice to Tam injections for 3 days.Four weeks later, Rankl mRNA was reduced by 70.0% in bone marrow from iCKO mice but not in the cortical bone (Fig. 4A).This change did not alter their body weight (Fig. S3A) and longitudinal bone growth, as indicated by growth plate thickness and femoral bone length (Fig. S3B-D).Strikingly, compared to WT mice, iCKO mice exhibited a 3.3-fold increase in trabecular bone volume fraction (BV/TV), a 1.8-fold increase in trabecular number (Tb.N), a 2.0-fold increase in trabecular thickness, and a 69.7% decrease in trabecular separation (Tb.Sp) (Fig. 4B-D).However, their cortical bone structure was not altered (Fig. S4).Similar massive bone gain phenotype was also observed in vertebrae (Fig. S5).
We next performed bone histomorphometry to uncover the cellular changes.TRAP staining revealed that osteoclasts are greatly reduced by 63.0% at the trabecular bone surface, but not changed at the chondroosseous junction (COJ) and endosteal bone surface (Fig. 4E-F).Meanwhile, osteoblasts (Osterix + bone surface cells) was decreased by 17.4% (Fig. 4G, H) and their activity was also signi cantly reduced (Fig. 4I, J).Serum chemistry con rmed those changes, showing a 34.7% reduction in bone resorption marker CTX-1 and a 14.2% reduction in bone formation marker P1NP (Fig. 4K).Overall, these data show that MALP-derived RANKL is important for maintaining bone resorption in adult mice.
RANKL not only regulates bone metabolism but also immune system 9 .Since RANKL is expressed in MALPs that are distributed throughout the bone marrow, we examined hematopoietic cells in iCKO mice.However, ow analysis did not detect any changes in hematopoietic components in the bone marrow or peripheral blood (Fig. S6A, B).Their spleen weight was not altered either (Fig. S6C), suggesting that hematopoiesis is normal in iCKO mice.

RANKL depletion in MALPs attenuates ovariectomy (OVX)-induced bone loss.
OVX surgery in mice mimics human postmenopausal osteoporosis.To understand the functional role of MALP-derived RANKL in pathological bone loss, we injected Tam into 3-month-old female WT and iCKO mice for 3 days and subjected them to sham or OVX surgery the day after the last injection.Mice were euthanized 6 weeks later.Estrogen de ciency was con rmed by an 86.7% decrease in uterine weight and a 21.8% increase in body weight of WT (Fig. S7A, B).Similar changes were also observed in iCKO mice.In sham groups, iCKO mice displayed a drastically increase in femoral and vertebral trabecular bone mass (2.9-fold and 1.5-fold, respectively) compared to WT mice (Fig. 5A, B, S8).OVX reduced femoral trabecular BV/TV by 57.8% in WT mice and 36.9% in iCKO mice.Compared to WT mice, iCKO mice exhibited 4.5-, 1.9-, and 1.8-fold increases in BV/TV, Tb.N, and Tb.Th, respectively, and a 70.3% decrease in Tb.Sp at 6 weeks post OVX.This preservation of trabecular bone post OVX was more prominent in vertebrae, with 53.4% and 25.8% decreases in BV/TV in WT and iCKO mice (Fig. S8), respectively.OVX did not affect femoral cortical bones in both WT and iCKO mice (Fig. S9).
Bone histomorphometry revealed that OVX increased osteoclast surface in both WT and iCKO mice but iCKO mice with OVX have 49.8% less osteoclast surface compared to WT mice with OVX (Fig. 5C, D).OVX also increased osteoblast surface and osteoblast activity in WT and iCKO mice (Fig. 5E-G).Serum chemistry further con rmed that bone turnover is increased in both genotypes but bone resorption, marked by CTX-1, is 37% less in iCKO with OVX compared to WT with OVX (Fig. 5H).Taken together, the above data demonstrate that RANKL from MALPs contributes to the enhanced bone resorption in the OVX model.
OVX also induces bone marrow adiposity (Fig. 5I, J).Interestingly, while MALPs are precursors for marrow adipocytes, their number did not change after OVX (Fig. S10).Compared to WT, we did not observe any change in marrow adipocytes in iCKO mice after sham surgery.After OVX, adipocyte area and size in iCKO mice were increased similarly as WT mice (Fig. 5I, J).These data suggest that RANKL from MALPs does not participate in OVX-induced marrow adiposity.
RANKL depletion in osteoporotic bone restores bone mass.
Next, we investigated whether MALPs-derived RANKL can be targeted for osteoporosis treatment.To do so, we subjected 3-month-old mice to OVX.Six weeks later when trabecular bone mass is signi cantly reduced, iCKO mice received vehicle or Tam injections for 3 days to deplete RANKL expression in MALPs.As a control, WT mice received OVX surgery and similar injections.To our surprise, even 3 times of Tam injections signi cantly increased femoral and vertebral trabecular bone mass in WT mice by 1.6fold and 1.5-fold, respectively, at 4 weeks later (Fig. 6A, B, S11), suggesting that Tam alone has bene cial effects on bone.In comparison, Tam administration increased femoral trabecular bone mass in iCKO mice at a much higher level (3.3-fold), accompanied by a 1.5-fold increase in Tb.N, a 2.0-fold increase in Tb.Th.and a 49.0% decrease in Tb.Sp (Fig. 6A, B).Similar effects were also observed in vertebral trabecular bone (Fig. S11).
Subsequent bone histomorphometry revealed that Tam injections in WT mice decreased osteoclast surface by 18.7% (Fig. 6C, D) and increases osteoblast surface by 1.1-fold as well as osteoblast activity in WT mice (Fig. 6E-G).Strikingly, Tam injections in iCKO mice greatly reduced osteoclast surfaces by 53.1% (Fig. 6C, D).Osteoblast surface was reduced by 9.6% (Fig. 6E, F) and osteoblast activity was also reduced (Fig. 6G).Serum chemistry con rmed that iCKO mice have a greater reduction of bone resorption than WT mice after Tam treatment.Taken together, our data suggest that after OVX-induced osteoporosis is established, depletion of RANKL in MALPs is still effective in restoring trabecular bone within a short period of time.
MALP-derived RANKL contributes to bone healing after injury.
Osteoclasts play important roles in the cartilage and bone remodeling stages of fracture healing 2 .However, whether they are also required for healing after bone defect injury is not well studied.Since Adipoq + cells are located inside the bone, not at the periosteal bone surface, we next drilled non-critical size holes in the femoral cortex of iCKO and WT mice.In this injury model, trabecular bone appears rst in the bone marrow close to the cortical defect region and then is resolved after healing, indicating a bone remodeling process.Meanwhile, the defect area is lled with new bone via intramembranous ossi cation.We carried out the drill-hole injury on mice 4 days after daily Tam injections at day 1-3.
MicroCT analysis showed that the hole in WT mice is healed nicely at 4 weeks post injury with almost no intramedular trabecular bone left.However, iCKO mice still had a signi cant amount of trabecular bone remaining.Compared to WT mice, iCKO mice showed decreased BV/TV in the cortical bone area (25.7%) and increased BV/TV in the intramedullary area (2.0-fold) (Fig. 7A, B), indicating a delayed healing.Histomorphometry analysis showed that osteoclasts in iCKO mice are drastically reduced by 62.7% in the defect cortical bone area and 78.0% in the intramedular trabecular bone area (Fig. 7C, D), while osteoblasts are not affected (Fig. 7E, F).Our data indicate that MALP-derived RANKL drives osteoclastogenesis and bone remodeling in this type of bone repair.

Discussion
Using RNA FISH and an inducible conditional knockout model, the present study investigated the role of adipogenic precursors in regulating trabecular bone turnover in adult bone homeostasis, postmenopausal bone loss, and injury repair.Our prior studies, as well as others, revealed that MALPsderived osteoclast regulatory cytokines, RANKL and Csf1, are important for trabecular bone remodeling in young mice [19][20][21][22] .However, those studies used constitutive Adipoq-Cre and thus did not address their actions in adult bone tissue.In this report, we rst analyzed mouse and human scRNA-seq datasets and utilized in situ experiments to discover that RANKL and Csf1 are mainly expressed in MALPs but not in osteoblasts and osteocytes in adult animals.We then studied adult bone phenotypes of RANKL de cient mice at two anatomic sites (long bone and vertebra) using inducible Adipoq-CreER under normal and pathological conditions.Collectively, our data demonstrate that RANKL derived from MALPs plays a dominant role in stimulating osteoclast formation and promoting trabecular bone resorption under normal and pathological conditions.
Prior studies using osteocyte-speci c Cres, such as Dmp1-Cre and Sost-Cre, to ablate RANKL proposed that bone embedding osteocytes are crucial for trabecular bone remodeling [12][13][14] .Our research challenges this conventional view.First, scRNA-seq of bone marrow mesenchymal lineage cells in mouse and human samples revealed a speci c expression of RANKL in Adipoq + MALPs.Second, in situ staining of Rankl clearly showed that Rankl is mainly expressed in cells expressing Adipoq and Pparg, two markers for MALPs.To our surprise, we did not detect Rankl mRNA in osteocytes, which might re ect a relatively low sensitivity of in situ approach.A prior report detected RANKL expression in one third of osteocytes using RANKL antibody 26 .However, their immunohistological images showed many more RANKL + cells in the bone marrow.Third, RANKL iCKO mice exhibited a striking 3.3-fold increase of femoral trabecular bone mass within one month of RANKL depletion.This fold change is much higher than 1.6-, 1.7-, and 2.3-fold we previously detected in RANKL CKO mice using Adipoq-Cre at 1, 3, and 5 months of age 20 , and also higher than ~ 2.5-fold increase in 6-month-old Dmp1-Cre RANKL CKO mice 13 .Our lineage tracing showed that only 9.9% of osteocytes and 46.1% of osteoblasts in the trabecular bone are labeled by Td in AdipoqER/Td mice at this time point.Fourth, Dmp1-Cre is not speci c for osteocytes.Lineage tracing revealed that it also labels all osteoblasts and ~ 30% CAR cells 27 , a mesenchymal subpopulation highly overlapped with MALPs 17 .These data are in line with the low Dmp1 expression in LCP and osteoblast clusters in our scRNA-seq 17 .Thus, it is likely that Dmp1-Cre driven RANKL knockout depletes RANKL in MALPs as well.Sost-Cre is more speci c for osteocytes.However, it also labels many hematopoietic cells 12 .Some of them, such as B and T lymphocytes, express RANKL too [28][29][30] .Lastly, our proposal that MALPs are a predominant source of RANKL in the trabecular bone also ts well with the emerging view that osteoclasts are long-lived cells constantly undergoing recycling 3 .Using intravital microscope, McDonald et al. discovered that RANKL rapidly stimulates the recycle of osteoclasts through ssion and fusion via osteomorphs, small daughter cells of osteoclasts in the bone marrow 6 .Because of their location, osteocytes are unlikely to participate into this dynamic osteoclast turnover.Due to their abundance in the cortical bone, osteocytes are likely to be the major regulator of cortical bone turnover, as Xiong et al. showed that mice with Dmp1-Cre driven RANKL knockout are resistant to tail suspension-induced cortical bone loss.
In addition to osteoporosis, osteoclasts are also important for bone healing.Past research in this eld focused on fracture, which is mostly repaired via an endochondral ossi cation mechanism.During this process, a cartilaginous soft callus is rst formed and then replaced by a bony hard callus, which is eventually remodeled into new cortical bone 31 .Osteoclasts are responsible for resorption of soft callus and remodeling of hard callus.Previous studies showed that suppressed bone resorption, either by RANK depletion 32 or by pharmacological inhibition of RANKL 33 , delays cartilage dissolution and callus remodeling and thus reduces bony unions.On the contrary, increased RANKL activity by depleting OPG, the decoy receptor for RANK, stimulates osteoclastogenesis and accelerates bone fracture healing 34 .To our knowledge, this study is the rst investigation of osteoclasts in bone healing after drill hole injury, which is repaired via an intramembranous ossi cation mechanism.It is interesting to note that RANKL depletion in MALPs does not affect cortical bone during bone maintenance and after estrogen de ciency but delays cortical bone healing after drill hole injury.In RANKL iCKO mice, reduced osteoclastogenesis at the injury site causes persistent remaining of bony callus, leading to delayed healing at the injury site.
Bone surface is covered by osteoblasts, osteoclasts, and bone lining cells.Compared to osteoblasts with a large, cuboidal shape, bone lining cells are morphologically de ned as attened cells covering quiescent bone surface not undergoing bone remodeling.In the conventional view, they are descendants of osteoblasts and able to be quickly re-activated into osteoblasts upon stimulations, such as PTH, mechanical loading, and radiation [35][36][37] .A prior study also found that they can be a major source of osteoblasts during adulthood 38 .However, it is puzzling that scRNA-seq analyses performed so far have not identi ed a subpopulation matching the above characteristics of bone lining cells.To our surprise, we found many Td + cells on the trabecular and endocortical bone surface in adult AdipoqER/Td mice.Since we used uorescent imaging, we were unable to observe cell shape.Thus, we used Osterix staining to label osteoblasts.Those Adipoq + bone surface cells are Osterix-cells at the beginning of pulse chase, and hence are not osteoblasts.Some of them express Pparg, Rankl, or Csf1 mRNAs, which are highly speci c for MALPs based on scRNA-seq.These data clearly suggest that bone lining cells contain not only osteogenic cells but also adipogenic cells.Future research using spatial omics techniques will help us further de ne the composition of bone lining cells and provide new insights into bone remodeling.
One limitation of our study is that Adipoq-CreER does not solely label MALPs.In the past several years, comprehensive and unbiased scRNA-seq analyses from multiple groups have all identi ed a major mesenchymal subpopulation in mouse and human bone marrow that highly and speci cally expresses adipogenic markers 17,39 .This subpopulation was subsequently named as MALPs by our group 15 and adipo-CAR by another group 16 .While we have been utilizing Adipoq-Cre or CreER to label this cell population, the present data that this CreER instantly marks adipocytes and Pparg-expressing cells and gradually mark osteoblasts and osteocytes over time indicate that in addition to MALPs, it also labels mesenchymal progenitors capable of bilineage differentiation in Adipoq + cells.The same labeling pattern on adipocytes, osteoblasts and osteocytes was also reported by other researchers 23,40 .Those additional progenitors are likely to be LCPs identi ed in our scRNA-seq due to a lack of CFU-F forming ability of Adipoq + cells and the close proximity of those cells to the bone surface.This leads to a possible depletion of RANKL in osteogenic cells in our mouse model.Nevertheless, since our studies focus on 4-6 weeks after Tam injection, a time point when the majority of osteocytes are not labeled by Td, we believe our conclusion that MALP-derived RANKL plays a dominant role is still valid.
In conclusion, we have demonstrated that bone marrow adipoprogenitors control bone resorption at the trabecular bone region in adult mice during homeostasis and pathological conditions.Prior research from our group and others have shown that MALPs are a master regulator of bone marrow microenvironment 17 .In addition to bone resorption, they also regulate bone formation, angiogenesis, blood cell production etc.Our most recent study found that MALPs expand in leukemia patients, suggesting its potential contribution to blood disorders 39 .With the advance in drug design and delivery, it is imperative to develop novel approaches targeting this cell population for osteoporosis treatment and bone repair with minimum side effects.

Analysis of scRNA-seq datasets
Pre-aligned scRNA-seq matrix les were acquired from GEO GSE145477 and GSE176171 (mouse) and GSE253355 (human).Standard Seurat pipeline 41 was used for ltering, normalization, variable gene selection, dimensionality reduction analysis and clustering.For the integrated dataset, batch integration was performed using Harmony (version 1.0) 42 .Cell type was annotated according to the metadata from published datasets 15,39 .

Animals study design
All animal work performed in this report was approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Pennsylvania.Adipoq-CreER Rosa-tdTomato (AdipoqER/Td) mice were generated by breeding Rosa-tdTomato 43 mice with Adipoq-CreER mice 44 .To generate RANKL iCKO mice, we rst bred Adipoq-CreER with RANKL ox/ ox mice 20 to obtain Adipoq-CreER RANKL ox/+ , which were then crossed with RANKL ox/ ox to generate RANKL iCKO mice.Male RANKL iCKO mice was further crossed with female RANKL ox/ ox mice to generate RANKL iCKO mice and WT (RANKL ox/ ox ) siblings.
All mouse lines, except RANKL ox/ ox , were obtained from Jackson Laboratory (Bar Harbor, ME, USA).To induce Td expression and RANKL depletion, mice at 3 months of age received daily intraperitoneal injections of Tam (75 mg/kg) for 3 days.For OVX surgery, 3-month-old female mice received either OVX or sham operation and their femurs, tibiae, and vertebrae were collected 6 or 10 weeks later for analyses.For drill hole injury, 3-month-old female mice received a 0.8-mm diameter unicortical drill hole defect via a 21G needle at the diaphysis part of right femurs and their injured femurs were collected 4 weeks later for analyses.

Micro-computed tomography (microCT) analysis
MicroCT analysis (microCT 45, Scanco Medical AG, Brüttisellen, Switzerland) was performed at 7.4 µm isotropic voxel size as described previously 45 .Brie y, the distal end of femur corresponding to a region at 0 to 3.4 mm below the growth plate was scanned.The images of the secondary spongiosa regions (0.6 to 2.1 mm below the lowest point of the growth plate, ~ 200 slices) were contoured for trabecular bone analysis.At the femur midshaft, 100 slices located at 4.7-5.5 mm away from the distal growth plate were acquired for cortical bone analyses.In vertebrae, the region 50 slices away from the top and bottom end plates (~ 300 slices) was acquired for trabecular bone analysis.To analyze bone healing after drilling a hole, the contouring of defect area or intramedullary area were manually de ned.A total of 150 slices were used for trabecular bone analysis.Trabecular and cortical bones were segmented from soft tissue using a threshold of 487.0 mgHA/cm 3 and 661.6 mgHA/cm 3 , respectively, with a Gaussian noise lter (sigma = 1.2, support = 2.0).For trabecular bone analysis, trabecular bone volume fraction (BV/TV), trabecular thickness (Tb.Th), trabecular separation (Tb.Sp), and trabecular number (Tb.N) were recorded.For cortical bone analysis, periosteal perimeter (Ps.Pm), endosteal perimeter (Ec.Pm), cortical bone area (Ct.Ar), cortical thickness (Ct.Th), and tissue mineral density (TMD) were recorded.All calculations were performed based on 3D standard microstructural analysis 46 .
To obtain para n sections, femurs were xed in 4% PFA for 24 hr and decalci ed in a 10% EDTA for 4 weeks at 4°C.Samples were then embedded in para n, sectioned at 6 µm in thickness, and processed for H&E staining and Safranin O/fast green staining.

ELISA assays
Sera were collected during mouse euthanization for measuring bone turnover markers, collagen type I Ctelopeptide degradation products (mouse CTX-I ELISA Kit, MyBioSource) and N-terminal propeptide of type I procollagen (Immunotag™ Mouse PINP ELISA Kit, G-Bioscience) according to the manufacturer's instructions.
qRT-PCR analysis Bone marrow was centrifuged from long bones and mixed with Tri Reagent (Sigma Aldrich) for RNA puri cation.Cortical bone was dissected from the remaining marrow-free bones, crushed in liquid nitrogen, mixed and homogenized with Tri Reagent on ice (Sigma Aldrich) for RNA puri cation.A Taqman Reverse Transcription Kit (Applied BioSystems, Inc., Foster City, CA, USA) was used to reverse transcribe mRNA into cDNA.The power SYBR Green PCR Master Mix Kit (Applied BioSystems, Inc) was used for quantitative real-time PCR (qRT-PCR).Primers for Tnfsf11 gene are 5'-GGAAGCGTACCTACAGACTA-3' (forward) and 5'-TGCTCCCTCCTTTCATCA-3' (reverse), and primers for β-actin gene are 5'-TCCTCCTGAGCGCAAGTACTCT-3'(forward) and 5'-CGGACTCATCGTACTCCTGCTT-3' (reverse).

Statistical analyses
Data are expressed as means ± standard deviation (SD).For comparisons between two groups, unpaired two-sample student's t-test was applied.For comparisons amongst multiple groups across two xed effect factors (e.g., genotype and surgery), two-way ANOVA was applied, followed by Tukey-Kramer multiple comparison test to account for family-wise type I error using Prism 8 software (GraphPad Software).In all tests, the signi cance level was set at α = 0.05.For assays using primary cells, experiments were repeated independently at least three times and representative data were shown here.
Values of p < 0.05 were considered statistically signi cant.

Declarations
Figures