Hox genes are crucial regulators of periosteal stem cell identity Free

Bone homeostasis and repair are mediated by skeletal stem cells (SSCs) that self-renew and differentiate into the major skeletal/mesenchymal lineages; stromal, osteo, chondro and adipo (Ambrosi et al., 2019; Yue et al., 2016; Zhou et al., 2014). During aging, SSCs are depleted, resulting in weaker bones that are more likely to fracture and that repair less efficiently. Thus, SSCs harbor massive therapeutic potential as a source of differentiated cells to replace those lost and damaged in injury and disease, and are a promising cellular target for rejuvenating the aged skeleton. Nevertheless, while the signals that drive SSC differentiation have been well studied, we know relatively little about the molecular and genetic mechanisms that maintain the stem cell pool. Here, we uncover a previously unreported role for homeobox (Hox) genes in SSC maintenance.

Hox genes are evolutionarily conserved transcription factors that regulate positional identity and cell fate specification during embryonic development (Deschamps and van Nes, 2005). In mouse and human, there are 39 Hox genes grouped into four clusters, HoxA, HoxB, HoxC and HoxD, in 13 paralogs (Izpisúa-Belmonte et al., 1991; Krumlauf, 1994) that are expressed in a regionally restricted manner. During development, their precise expression pattern establishes the body plan of an organism. At post-natal stages, Hox genes are highly expressed in the stem cells of different species, tissues and organs (Issa et al., 2019; Svingen and Tonissen, 2003; Yoshioka et al., 2021), where they control self-renewal (Magnusson et al., 2007; Yoshioka et al., 2021) and differentiation (Bulajic et al., 2020; Song et al., 2020; Winnik et al., 2009). Therefore, in the adult, the Hox code may also impart positional information that is crucial for tissue homeostasis and regeneration (Pineault et al., 2019; Rux et al., 2016; Yoshioka et al., 2021). In the skeleton, Hox genes are expressed by stem and progenitor cells, and excluded from lineage-restricted cell types (Bradaschia-Correa et al., 2019; Gerber et al., 2018; Leucht et al., 2008; Lin et al., 2021; Pineault et al., 2002; Rux et al., 2016), suggesting that they play a stem cell-specific role. To date, studies have demonstrated that Hoxa11 is necessary for osteoblast maturation and successful fracture healing of the ulna (Rux et al., 2016; Song et al., 2020), but it is still unknown whether it plays a functional role in the SSC.

Here, we pinpoint Hox genes as being crucial for SSC maintenance. We show that Hox deficiency leads to a decrease in stem cell number and proliferation, and an increase in the propensity of stem cells to undergo spontaneous differentiation toward osteogenic and adipogenic fates. Conversely, an increase in Hoxa10 expression in skeletal stem and progenitor cultures reduces the differentiation potential of cells, while increasing their self-renewal capacity. During differentiation, SSCs give rise to many fate-restricted progenitors that have limited lineage potential and lifespan (Chan et al., 2015; Debnath et al., 2018). Here, we reveal that Hox overexpression can also drive reprogramming of skeletal progenitors to a more uncommitted stem cell-like state and establish that the reprogramming capacity of individual Hox genes is restricted to cells that are isolated from the specific anatomical region that they are endogenously expressed in. We have previously demonstrated that there is a loss of functional SSCs with age that underpins a decline in the regenerative capacity of the skeleton (Josephson et al., 2019). Here, we show that, during aging, Hox expression is reduced, coincident with the decrease in SSC number, and that Hoxa10 overexpression partially restores bone regeneration in aged mice. Therefore, reprogramming more prevalent progenitor populations could be an innovative method for rescuing SSC number and a promising strategy for combating bone disorders, such as age-associated bone loss.

In various tissues, including the placenta, bone marrow and skeleton, Hox genes are expressed by the stromal cells that mediate tissue repair (Bradaschia-Correa et al., 2019; Hwang et al., 2009; Liedtke et al., 2010; Rux et al., 2016), where they regulate morphogenesis and regeneration (Eckardt et al., 2003; Nogi and Watanabe, 2001; Orii et al., 1999; Thummel et al., 2007). Our RNA-sequencing dataset of whole-bone, hindlimb skeletal elements (Bradaschia-Correa et al., 2019) confirmed that Hox family members are transcriptionally enriched within CD45^–Ter119^–TIE2^–LEPR⁺ skeletal stem and progenitor cells (SSPCs) compared with cells of the bone microenvironment that comprise hematopoietic, endothelial, erythroid (CD45⁺Ter119⁺TIE2⁺LEPR^variable) and terminally differentiated skeletal/stromal lineage cells (CD45^–Ter119^–TIE2^–LEPR^–), with Hoxa10 being the most highly expressed (Fig. 1A, Fig. S1).

Skeletal stem and progenitor cells that reside within the periosteum, a thin membrane on the outer surface of the bone, are referred to as periosteal stem and progenitor cells (PSPCs), and are the main contributors to bone repair after injury (Colnot et al., 2012; Duchamp de Lageneste et al., 2018; Ferretti and Mattioli-Belmonte, 2014; Roberts et al., 2015). As such, they represent an ideal cellular therapeutic target for improving fracture healing. Still, the transcriptional signature of PSPCs is distinct from SSPCs found in other regions of the bone (Duchamp de Lageneste et al., 2018) and the role of Hox genes within this population is currently unknown. To examine this, we isolated PSPCs from adult tibiae – the most commonly fractured long bone – via serial collagenase digestions, and performed transcriptional analysis by RNA-sequencing. These data indicated that Hoxa10 is the most highly expressed family member in the adult tibial periosteum (Fig. 1B, Fig. S2A), which we validated by Nanostring nCounter gene expression analysis (Fig. S2B). PSPCs are a heterogeneous population encompassing naïve periosteal stem cells (PSCs: 6C3^–CD90^–CD49f^lowCD51^low CD200⁺CD105^–), and more committed periosteal progenitor 1 (PP1: 6C3^–CD90^–CD49f^lowCD51^lowCD200^–CD105^–) and periosteal progenitor 2 (PP2: 6C3^–CD90^–CD49f^lowCD51^lowCD200^variableCD105⁺) cells (Debnath et al., 2018) (Fig. 1C, Fig. S2C). We found that Hoxa10 expression was most abundant in PSCs and was significantly reduced as cells progressed along the lineage hierarchy to PP1 and PP2 (Fig. 1D). Correspondingly, in other anatomical locations within the bone, such as the bone marrow, Hox gene expression is specific to progenitor cells and absent in more differentiated lineages (Magnusson et al., 2007; Yoshioka et al., 2021). Here, we have investigated whether this was also the case in the periosteum by studying the early dynamics (the first two hours) of Hox expression during osteogenic and adipogenic differentiation of PSPCs. We observed that Hoxa10 was rapidly downregulated within 30 min, ahead of other canonical stem cell markers, including Pdgfra, Twist1 and cyclin D1 (Ccnd1) (Fig. 1E,F) (Chan et al., 2015; Duchamp de Lageneste et al., 2018; Morikawa et al., 2009; Pinho et al., 2013). Hoxa11 and Hoxc10, the family members with the second and third highest expression in the tibial periosteum (Fig. S2A,B), were also downregulated at the onset of differentiation (Fig. S2D). Thus, Hox transcription is strongly associated with the most primitive periosteal stem cell state.

Hox family members regulate stem cell function in multiple tissues and organisms (Bradaschia-Correa et al., 2019; Hwang et al., 2009; Liedtke et al., 2010; Rux et al., 2016). Hox genes are enriched in PSPCs that are responsible for injury repair (Fig. 1). In addition, we observed that Hox genes are downregulated in SSPC/PSPCs during aging (Fig. 2A,B, Fig. S3A), coincident with the loss of SSPC/PSPC number and function, and a decline in the regenerative capacity of the skeleton (Fig. 2C) (Josephson et al., 2019). Thus, we hypothesize that Hox genes play a role in regeneration of the skeleton after injury. To investigate this, we assessed the effect of loss of Hox expression on PSPC number, self-renewal and regenerative potential. In knockdown/knockout experiments, the role of individual Hox genes is frequently masked by the functional redundancy of other family members (Carpenter et al., 1993; Gavalas et al., 1998; Mark et al., 1993; McNulty et al., 2005; Rossel and Capecchi, 1999; Rux et al., 2016; Song et al., 2020; Studer et al., 1996). To circumvent these issues, we employed a mix of siRNAs that simultaneously targeted the most highly expressed Hox genes in the tibia (Hoxa10, Hoxa11, Hoxd10, Hoxd11 and Hoxc10), referred to as siHoxMix, which resulted in a significant decrease in their expression (Hoxa10, 85.3%; Hoxa11, 77.6%; Hoxd10, 85.4%; Hoxd11, 53.4%; Hoxc10, 82.8% expression reduction; Fig. S3B).

Self-renewal is a defining property of stem cells and is necessary to maintain the stem cell pool. We observed that treating isolated periosteal cells with siHoxMix caused a decrease in proliferation (Fig. 2D), cell number (Fig. 2E) and the proportion of cells expressing the PSC markers SCA-1 and PDGFRα (Houlihan et al., 2012; Rux et al., 2016) (Fig. 2F). We then quantified self-renewal at a single-cell resolution, using CellTrace, a fluorescent dye that is diluted with each cell division. The amount of dye in each cell after 6 days was measured by flow cytometry and indicates the generation number and cycling rate. Under control conditions, the majority of PSPCs underwent six or more cell divisions and are therefore considered high cycling, a characteristic associated with stemness (Fig. 2G, Fig. S3C,D). Upon Hox knockdown, there was a significant increase in the proportion of cells undergoing five cell divisions or fewer, and, most evidently, in the number of cells that divided only once (Fig. 2G, Fig. S3C). Moreover, sustained knockdown of multiple Hox genes (Hoxa11, Hoxc10 and Hoxd10) for 14 days (Fig. S3E) – or the earlier repression of all five Hox genes being studied – led to an upregulation of osteogenic and adipogenic genes at day 14 despite the absence of overt differentiation cues (Fig. S3F), suggesting a progressive exit of cells from the PSC state.

Next, we asked whether this was also the case in vivo, employing a Pdgfra^CreERT (Wosczyna et al., 2019) mouse line to conditionally delete Hox genes in skeletal stem cells, followed by tibial injury (Lee et al., 2021; Leucht et al., 2007) to stimulate PSPC/SSPC self-renewal (Fig. 2H). To circumvent potential Hox redundancy, we used an allele with loxP sites flanking the entire ∼100 kb span of the HoxA cluster (Kmita et al., 2005). Deletion of the HoxA cluster in Pdgfra⁺ PSPCs/SSPCs (Ambrosi et al., 2019; Morikawa et al., 2009; Pinho et al., 2013) (Fig. S3G) led to a reduction of PSCs/SSCs, a concomitant increase in PP1/pre-BCSP (pre-bone, chondro, stromal progenitors) and a significant loss of proliferative capacity compared with HoxA^flox/flox control animals (Fig. 2I). Thus, both in vitro and in vivo, loss of HoxA genes results in a decrease in PSC/SSC self-renewal and/or a loss of stemness, and consequential proportional accumulation of more-committed cell types.

To determine whether a reduction in Hox expression also impaired PSC regenerative function in vivo, we fractured the tibiae of young 3-month-old mice and locally delivered siHoxMix on a collagen sponge at the time of injury (we have verified that this delivery method results in widespread transfection; Fig. S3H). We used Col1a1^creERT;Rosa^Tomato mice to label newly formed osteolineage cells and administered tamoxifen at the onset of the differentiation phase of fracture healing (Lee et al., 2021) (5-6 days post-injury, Fig. 2J). At 7 days after injury, fracture sites transduced with siHoxMix displayed a 50% reduction in the proportion of Tomato⁺ cells compared with those transduced with a non-targeting control siRNA (Fig. 2K), demonstrating that the early depletion of Hox genes at the time of PSC self-renewal in response to injury leads to decreased osteolineage output. Together these data demonstrate that Hox genes are necessary to maintain PSPCs in an undifferentiated, self-renewing state, and that decreased Hox expression, as occurs in aging, is sufficient to impede bone regeneration.

As Hox knockdown resulted in loss of skeletal stem cell self-renewal, we asked whether, conversely, Hox overexpression promoted stemness and, thus, whether it might represent a strategy for rescuing skeletal stem cell number and function during aging. To this end, we constructed a lentiviral vector harboring the protein-coding sequence of Hoxa10, which is the Hox gene most highly expressed in the tibia, along with that of GFP, to fluorescently label infected cells (LV-Hoxa10/GFP), and a control vector containing only the GFP-coding sequence (LV-GFP, Fig. S4A). We isolated periosteal cells, infected them with LV-Hoxa10/GFP or LV-GFP and confirmed stable Hoxa10 overexpression (∼25-fold overexpression, Fig. S4B). Hoxa10 overexpression in periosteal cells increased the proportion of PSPCs (Fig. 3A). Moreover, when exposed to osteo-induction medium for 14 days (Fig. S4C), LV-Hoxa10/GFP-transfected cells maintained a significantly greater proportion of PSPCs (Fig. S4D) and expressed significantly lower levels of osteogenic genes [osterix (Sp7), osteocalcin (Bgalp) and Runx2] compared with cells infected with the control virus (Fig. S4E). Therefore, forced expression of Hoxa10 supports stem cell maintenance.

Our observation that Hoxa10 overexpression increases the number of stem cells in PSPC cultures (Fig. 3A) could be attributed to two mechanisms: an expansion of the stem cell pool through increased self-renewal/proliferation or reprogramming of more committed cells to a stem cell state. First, we examined the possibility that Hoxa10 overexpression enhances the proliferation of PSPCs, using CellTrace to determine the cycling rate of each stem and progenitor compartment infected with LV-Hoxa10/GFP or LV-GFP. Hoxa10 overexpression did not alter the proliferative rate of any PSPC population (Fig. 3B, Fig. S4F). Therefore, we subsequently investigated the possibility that Hoxa10 reprograms cells to a more-primitive state by isolating each periosteal stem and progenitor population (PSC, PP1 and PP2) separately, transducing them with LV-Hoxa10/GFP or LV-GFP and reassessing the lineage hierarchy after 7 days (Fig. 3C). Control and Hoxa10-overexpressing PSCs reconstituted a culture comprising predominantly PP1 cells (Fig. 3D), indicating that PSCs rapidly differentiate into more committed progenitors. This may also reflect the proliferative advantage of PP1 cells (Fig. 3B). However, there was no difference in the distribution of cell populations between LV-Hoxa10/GFP and LV-GFP transduced cells (Fig. 3D). When PP1 cells were transduced with LV-Hoxa10/GFP or LV-GFP, most cells remained in the PP1 cell state. A small fraction (5%) of LV-GFP-infected PP1 cells gave rise to PSCs. This limited amount of stem cells in the control-infected PP1 cells may reflect a basal level of stochastic sampling of neighboring cellular states, as has been described in many other purified cell populations (Gupta et al., 2011; Wang et al., 2014). Notably, the proportion of PSCs was significantly larger (16%) in LV-Hoxa10/GFP-infected PP1 cells at the expense of the PP1 population (Fig. 3E, Fig. S4G). A small fraction of PP2 cells also gave rise to PP1 cells but PP2 cells showed no difference in the relative abundance of the different cell populations when Hoxa10 was overexpressed (Fig. 3F). Four independent iterations of this experiment confirmed these results, in which Hoxa10 overexpression in PP1 cells led to a greater than threefold increase in the proportion of PSCs in the progenitor pool (Fig. 3G, Table S1) and an approximately twofold increase in the frequency of PSCs among total Hoxa10-overexpressing cells (Fig. 3H); the potential for contaminating PSCs during PP1 isolation was disproven by assessing the post-sorting frequency of PSCs (0%) in the PP1 samples (Fig. S4H). As the Hox and GFP molecules are produced in equimolar amounts from the bicistronic vectors used in these experiments, we used the GFP intensity during flow cytometry analysis to evaluate whether the phenotypic effects of Hox overexpression correlate with overexpression levels. We found that Hoxa10/GFP-expressing converted PSCs had significantly lower GFP expression than the transduced, non-reprogrammed PP1 cells, suggesting that PP1 to PSC conversion preferentially relies upon low overexpression of Hoxa10. PSCs and PP1 cells transduced with the GFP control, on the other hand, displayed no preference for high or low GFP expression (Fig. S4I). Although we showed that Hoxa10 overexpression did not alter PSPC proliferation (Fig. 3B), to unequivocally eliminate the possibility that this result reflected a PSC contamination of the PP1 fraction during cell sorting, and subsequent differential expansion of LV-Hoxa10/GFP and LV-GFP transduced cells, we repeated this experiment in the presence of mitomycin C, a potent inhibitor of proliferation. We confirmed an effective inhibition of proliferation in mitomycin C-treated PP1 cells transfected with LV-GFP or LV-Hoxa10/GFP (Fig. S5A) and, in the absence of cell division, observed an even greater increase in the proportion of PP1-derived PSCs in Hoxa10-overexpressing cells (Fig. 3I). Additionally, neither the lentiviral selection (LV-GFP or LV-Hoxa10/GFP) nor the presence of mitomycin C produced a difference in cell death (propidium iodide⁺) or apoptosis (annexin V⁺) when total cells, infected cells or individual PSPC compartments were assessed by flow cytometry (Fig. S5B-D). These data demonstrate that the emergence of PSCs from Hoxa10/GFP-infected PP1 cells is not due to an expansion of contaminating PSCs or differential cell survival and suggests that progenitors are instead reprogrammed to a less committed state.

To further interrogate the identity of PP1-derived PSC-like cells, we leveraged published single-cell gene expression data (Debnath et al., 2018) to identify novel transcriptional markers of PSCs. We sorted cells in silico using defined PSPC cell-surface markers [6C3 (Enpep), CD90 (Thy1), CD51 (Itgav), CD105 (Eng) and CD200 (Cd200)] and uncovered several factors that were enriched in periosteal stem cells versus PP1 and PP2 cells (Omd, Ucma and Frzb; Fig. S4J). All of these genes were significantly upregulated in PP1 cells transduced with LV-Hoxa10/GFP relative to those infected with LV-GFP (Fig. S4K), indicating a shift towards a PSC transcriptome.

Finally, we asked whether PP1-derived PSC-like cells also assume functional properties of PSCs. PSCs are at the top of the PSPC lineage hierarchy and thus possess a greater ability to self-renew than downstream progenitors (Debnath et al., 2018). Accordingly, we asked whether PP1-derived PSC-like cells have a greater self-renewal capacity than PP1 cells. We infected PP1 with LV-Hoxa10/GFP or LV-GFP, isolated 300-750 GFP⁺ cells and transplanted them (along with 100,000 bone marrow support cells) for 2 weeks underneath the renal capsule – an environment that is permissive for skeletal stem cell differentiation (Ambrosi et al., 2021; Debnath et al., 2018). GFP⁺ cells were then re-sorted and transplanted for another 2 weeks and the persistence of GFP⁺ PSCs (reprogrammed from PP1 cells) was determined after each transplantation (Fig. 3J). The majority of GFP⁺ PSCs derived from control LV-GFP-infected PP1 cells were lost after only a single round of transplantation, whereas PSCs derived from LV-Hoxa10/GFP-infected PP1 cells were maintained after multiple rounds of transplantation, indicating a greater self-renewal capacity (Fig. 3K). Therefore, Hoxa10 overexpression is sufficient to transcriptionally and functionally reprogram a fraction of PP1 progenitors to a more-primitive PSC state.

During development, individual Hox genes are expressed in different regions of the body. This spatial ‘Hox code’ is crucial to establish the body plan of the organism (Darbellay et al., 2019; Izpisúa-Belmonte et al., 1991; Papageorgiou, 2012). Hox genes are also expressed in adult skeletal elements and their expression pattern roughly mirrors that during embryogenesis (Fig. 4A) (Ackema and Charité, 2008; Bradaschia-Correa et al., 2019; Leucht et al., 2008; Rux et al., 2016; Rux and Wellik, 2017; Song et al., 2020). Thus, as in development, Hox genes may endow stem cells from different anatomical locations with the specific properties needed to regenerate the tissue in which they reside. Although we previously showed that periosteal cells from skeletal regions with or without Hox expression exhibit functional differences (Bradaschia-Correa et al., 2019; Leucht et al., 2008), the regional specificity of Hox function in the adult has yet to be established. To investigate this, we first precisely defined the adult Hox code, analyzing the expression profile of all 39 Hox genes in periosteal cells from various murine skeletal elements, including the pelvis, thoracic vertebrae 5-8 (spine^T5-8), radius/ulna and anterior ribs 1-4. Hoxa10 was the most highly expressed Hox gene in periosteal cells of the pelvis, radius/ulna and tibia (Figs 1B and 4B,C, Figs S2A,B and S6A), Hoxb8 was the most highly expressed in the adult spine^T5-8 (Fig. 4D, Fig. S6A) and Hoxa5 was the most highly expressed in the anterior ribs^1-4 (Fig. 4E, Fig. S6A).

We then asked whether the stemness-promoting activity of Hoxa10 is universal or limited to regions where it is most highly expressed (tibia, pelvis and radius/ulna). When periosteal cells from the pelvis, which express high levels of Hoxa10, were transduced with LV-Hoxa10/GFP, we observed a significant increase in the proportion of PP1-derived PSC-like cells (Fig. 4F, Table S2), as in the tibia (Fig. 3E). In contrast, there was no effect of LV-Hoxa10/GFP transduction on periosteal cells from the spine^T5-8 or anterior ribs^1-4, which express negligible Hoxa10 (Fig. 4H,I, Table S2), indicating that Hoxa10 has region-specific function in the adult skeleton. Although we also observed no effect of LV-Hoxa10/GFP transduction on periosteal cells from the radius/ulna that highly express both Hoxa11 and Hoxa10, only minor skeletal defects are observed in the radius/ulna of Hoxa10^−/− mice, whereas Hoxa11^−/− show severe perturbations (Boulet and Capecchi, 2002; Davis et al., 1995; Favier et al., 1996; Small and Potter, 1993; Wellik and Capecchi, 2003), suggesting that Hoxa11 plays a key role in that region, despite not being the most highly expressed Hox gene.

Next, we asked whether other Hox family members display the same reprogramming activity as Hoxa10 in periosteal cells from the skeletal element where they are most abundantly expressed. To this end, we overexpressed Hoxa11 in radius/ulna, Hoxb8 in spine^T5-8 and Hoxa5 in anterior rib^1-4 periosteal cells. In all cases, we observed a significant increase in the proportion PP1-derived PSC-like cells (Fig. 4J-L, Table S2). As embryonic development progresses anteriorly to posteriorly (and proximal-to-distal in the limb), each Hox gene cluster successively becomes more accessible and expressed, generating a nested pattern where more posterior tissues express a wider set of Hox genes than anterior ones (Tarchini and Duboule, 2006; Papageorgiou, 2012). Thus, posterior tissues may respond to anterior Hox gene expression. To test this, we infected tibial PP1 cells with Hoxa5 (expressed in the anterior rib^1-4) or control lentivirus (Fig. S6B). Hoxa5 overexpression in posterior progenitor cells did not induce reprogramming (Fig. S6C). Along with Hoxa10, Hoxa11 is also highly expressed in tibia periosteal cells and, when overexpressed, triggered tibia periosteal progenitor cells to revert to a more primitive state (Fig. S6C). Together, these data show that Hox genes promote periosteal stemness in a regionally restricted manner.

So far, our data demonstrate that Hox genes are both necessary to maintain a PSC identity and sufficient to induce PSC properties in more-committed progenitors. During aging, Hox genes are downregulated (Fig. 2A,B, Fig. S3A), coincident with a decrease in the number and function of SSPCs/PSPCs, and a consequent decline in the regenerative capacity of the skeleton (Ambrosi et al., 2021; Josephson et al., 2019). Thus, we hypothesized that modulating Hox expression would enhance fracture healing in elderly individuals. To address this hypothesis, we fractured tibiae of young (3-month-old) and middle-aged (13-month-old) wild-type mice, and administered either LV-GFP or LV-Hoxa10/GFP, adsorbed into a collagen sponge, to the fracture site at the time of injury (Fig. 5A). Bones were analyzed by μCT at 14 days post-injury, a key stage in the healing process when the injury site progresses to mineralized callus. Bone volume within the callus (BV/TV) is a standardized read-out of bone regeneration. As expected, middle-aged control LV-GFP-transduced tibiae showed a characteristic reduction in callus BV/TV compared with that of young mice (Fig. 5B,C). When middle-aged fracture sites were instead transduced with LV-Hoxa10/GFP, we observed a significant increase in BV/TV relative to the middle-aged controls, indicating a partial rescue of fracture healing (Fig. 5B,C). In summary, our data pinpoint Hox genes as essential location-specific regulators of PSC identity, and show that a short-term local increase in their expression is sufficient to augment bone regeneration, highlighting them as promising therapeutic targets for improving the healing of the skeleton in elderly and repair-compromised individuals.

Skeletal homeostasis and regeneration rely heavily on stem cells to replenish the tissue lost due to injury or wear and tear. To preserve this capacity, stem cell number has to be constantly maintained by cell division, one of the hallmarks of stemness. Currently, there is a relatively poor understanding of the molecular mechanisms that govern skeletal stem and progenitor cell maintenance over time and lineage progression during bone healing, and this presents one of the major hurdles to advancing cell-based therapies for treatment of bone fractures. The data generated in this study add to the growing appreciation that Hox genes have important functions in the adult skeleton and support a model in which high Hox expression in the uncommitted PSC compartment confers greater self-renewal capacity and moderates lineage progression towards more lineage-restricted cell fates. We also demonstrated that this role of Hox can be exploited to shift periosteal progenitors with limited self-renewal capacity (Debnath et al., 2018) to a more primitive state, thus increasing their functional potential.

Previous reports identified Hoxa11 as the primary Hox gene that patterns the embryonic tibia and a marker of the adult tibia (Rux et al., 2016; Wellik and Capecchi, 2003). Still, the tibia and fibula are only mildly affected in Hoxa11/Hoxd11 double mutants (Davis et al., 1995; Wellik and Capecchi, 2003), suggesting that other Hox family members are involved in hindlimb development. Both Hoxa10^–/– and Hoxd10^–/– mice display defects in the developing hindlimbs, with changes in these structures appearing with greater penetrance in Hoxa10^–/– mice (Wahba et al., 2001). As we also identified Hoxa10 as the most highly expressed Hox gene in the adult tibial periosteum (Fig. S2), we decided to investigate its specific role in this skeletal element. Although recent studies demonstrate that Hoxa11 marks a primitive mesenchymal stem cell (MSC) in the bone marrow and periosteum, and that Hoxa11 lineage-marked cells are long-term contributors to MSCs throughout life (Pineault et al., 2019; Rux et al., 2016; Song et al., 2020), we show for the first time that Hox genes play a functional role in skeletal stem cell maintenance.

Although loss-of-function studies in other tissues have demonstrated a stem cell maintenance role for Hox genes, phenotypes have usually been mild (Björnsson et al., 2003; Brun et al., 2004; Iyyanar and Nazarali, 2017; Magli et al., 1997; Owens and Hawley, 2002). Previous studies have also suggested that Hox genes do not play a role in the maintenance of skeletal stem cells. For example, Rux et al. showed that skeletal stem cells harvested from mice in which both alleles of Hoxd11 and one allele of Hoxa11 are knocked out display defects in tri-lineage differentiation but do not lose their stemness marker profile or self-renewal capacity (Rux et al., 2016). In addition, after the conditional deletion of Hoxa11 and Hoxd11 in 8-week-old forelimbs, skeletal stem cells lacking these genes were still maintained after 10 months (Song et al., 2020). Therefore, individual Hox genes may have distinct functions in the skeletal system. In support of this, previous studies have shown that adjacent Hox genes or Hox genes in the same paralogous group can have opposing functions. For example, in muscle satellite cells, Hoxa10 promotes proliferation whereas Hoxa9 and Hoxc10 disrupt mitosis (Schwörer et al., 2016; Yoshioka et al., 2021). Likewise, Hoxa5 inhibits endothelial cell differentiation and blocks vascular growth (Rhoads et al., 2005), whereas Hoxb5 promotes angiogenesis (Winnik et al., 2009). Furthermore, studies focusing on Hox in skeletal tissues typically rely on reducing the expression of one Hox paralogous group (or use single Hox knockout mouse models), and disregard the potential complementation by other highly expressed Hox genes in the cell compartments under study – as observed in the tibial periosteum (Fig. S2A,B). The examination of compound mutants or deficiencies within a Hox cluster or paralogous group may therefore yield more severe phenotypes that can elucidate the role of Hox in adult skeletal cells. In support of this, Hox gain-of-function studies, which can reveal activities masked by functional redundancy, show that Hox genes block differentiation in numerous tissues (Crooks et al., 1999; Fuller et al., 1999; Owens and Hawley, 2002; Schiedlmeier et al., 2003), including Hoxa10 in hematopoietic and cardiac progenitors (Behrens et al., 2013; Buske et al., 2001; Magnusson et al., 2007; Taghon et al., 2002; Thorsteinsdottir et al., 1997) and Hoxa2 in murine bone and cartilage development (Creuzet et al., 2002; Kanzler et al., 1998). Here, we have found that the expression of multiple posterior Hox paralogous groups must be decreased to detect a defect in number and function of tibial PSPCs. This functional redundancy may represent a protective evolutionary mechanism to maintain stem cells.

Hox genes have also been shown to regulate osteo-lineage specification. During differentiation of osteoblastic cell lines, Hoxa10 drives the early expression of osteogenic genes through chromatin remodeling, and the in vivo conditional deletion of Hoxa11 and Hoxd11 in the Hoxa11 domain leads to osteogenic differentiation defects (Hassan et al., 2007; Song et al., 2020). In keeping with this, we observed that Hoxa10 expression decreased within the first few hours of osteogenic and adipogenic differentiation, in line with a stem cell maintenance role, but by around 4 h post-induction, expression returned to the levels seen in non-induced PSPCs (Fig. 1E,F). Thus, Hox genes may play multiple dynamic roles in stem cell maintenance and fate decisions. We demonstrated that Hoxa10 overexpression leads to a partial restoration of in vivo bone regeneration in aged mice (Fig. 5), which could be driven both by an increase in stem cell self-renewal and, at later stages, osteo-lineage commitment. Future studies will precisely dissect the dynamic roles of Hox genes in this process.

Cellular reprogramming is associated with an opening of chromatin (Gaspar-Maia et al., 2011; Ugarte et al., 2015). Several Hox family members (including their Drosophila orthologues) function as pioneer factors, demonstrating a preference to bind inaccessible chromatin and, by doing so, increase accessibility (Beh et al., 2016; Bulajic et al., 2020). We also previously showed that Hox-positive periosteal cells have more open chromatin and are more stem-like than periosteal cells derived from Hox-negative tissue (Bradaschia-Correa et al., 2019). This, along with the increase in stem cell frequency upon Hox overexpression, led us to postulate that Hox expression drives skeletal progenitors to a more primitive state. We found that overexpression of Hoxa10 reprogrammed PP1 skeletal progenitors to a stem cell state. Intriguingly, blocking cell division resulted in a massive increase in the efficiency of this process (Fig. 3I). Mitomycin C arrests cells in the G1 phase of the cell cycle (Kang et al., 2001) and multiple studies have connected cell fate decisions to a prolonged G1 (Sela et al., 2012; Tapias et al., 2014). Thus, temporary cell cycle arrest may prolong the permissive window for cell reprogramming to occur. The ability of Hoxa10 overexpression to reprogram PP1 but not more committed PP2 progenitors is also noteworthy and could represent changes in the epigenetic landscape or a loss of key co-factors at this transition point. Detailed comparisons of the transcriptome and epigenome during progression along this lineage hierarchy will be necessary to gain further insights into the molecular mechanisms of reprogramming.

The continued regional specification of Hox gene expression in adult tissues has been demonstrated by several independent studies, largely by the characterization of cells in culture (Bradaschia-Correa et al., 2019; Chang et al., 2002; Leucht et al., 2008). Here, we corroborate this finding in the periosteal cell compartments of various anatomical regions and, importantly, find that this adult Hox code is functionally relevant to skeletal stem cell regulation. Similarly, in muscle satellite cells derived from the tibialis anterior, ectopic expression of Hoxa10 was able to rescue proliferation deficiency triggered by the conditional deletion of Hoxa10 in these cells, but the ectopic expression of Hoxa5, Hoxa9 or Hoxc10 had no such effect (Yoshioka et al., 2021). Importantly, region-specific Hox function in reprogramming and stem cell maintenance has implications for devising and predicting the engraftment potential of stem cell therapies that target specific segments of the skeleton, or potentially other tissues whose function is controlled by Hox expression profiles.

In contrast to the bone marrow, the periosteum is a relatively understudied tissue. Periosteal progenitor cells play a central role in bone repair and, as such, represent a promising source of cells for tissue engineering approaches. PSPCs also exhibit several characteristics that are advantageous for such strategies, including their high proliferative rate, which is necessary for efficient in vitro expansion (Sakaguchi et al., 2005; van Gastel et al., 2012; Yoshimura et al., 2007), and a greater osteogenic capacity than many other mesenchymal stem cell populations both in vitro and when transplanted in vivo (Roberts et al., 2011, 2015). Looking forward, advances in lineage reprogramming in many tissues have revealed a remarkable flexibility in cell identity (Morris, 2016), and unraveling the mechanisms of reprogramming in skeletal tissues can facilitate the development of cell fate engineering strategies. Further research elucidating the mechanisms by which Hox overexpression increases stem cell potency and how this affects lineage potential can help achieve this therapeutic goal.

C57BL/6 mice (8-16 weeks old), were purchased from the Jackson Laboratory and bred in the barrier facility at the New York University School of Medicine. Pdgfrα^CreERT/+ knock-in mice were obtained from by the Michael Wosczyna Laboratory at NYU Langone (New York, USA) (Wosczyna et al., 2019). Col1a1^CreERT/+ mice (B6.Cg-Tg(Col1a1-cre/ERT2)1Crm/J) were obtained from the Jackson Laboratory (016241). All mice were bred in the barrier facility at the New York University School of Medicine. To induce recombination in transgenic cre-ERT2 mice, tamoxifen (Sigma-Aldrich) was administered intraperitoneally at 2 mg/day according to the dosing protocol in either Fig. 2H or J. Mice were maintained on a 12-h light/dark cycle with food and water provided ad libitum.

All experiments involving human subjects were approved by the NYU Robert I. Grossman School of Medicine Institutional Review Board. After informed consent was obtained, bone marrow specimens were obtained during surgery. One cubic centimeter of bone marrow was immediately transferred into a microcentrifuge tube and placed on ice.

FPKM values for each Hox gene were derived from tibial periosteal RNA sequencing data previously published by our group (Bradaschia-Correa et al., 2019; deposited in GEO under accession number GSE173371). Nanostring read counts were determined using the nCounter platform and by generating a custom panel of target-specific oligonucleotide probes (CodeSet) of the 39 murine Hox genes (Table S3). Of the total 78 Hox isoforms produced by the four Hox clusters, only one isoform of Hoxc4 was not detectable by the custom CodeSet. Five housekeeping genes (Actb, Gusb, Pgk1, Tbp and Tubb) were used to normalize the read counts.

Primary periosteal stem and progenitor cells were obtained from the tibia, pelvis, anterior ribs (1-4), thoracic vertebrae (5-8) of the spine, radius/ulna or parietal/frontal calvaria. After careful dissection from 8- to 16-week-old wild-type (C57BL/6) mice, bones with intact periosteum were submitted to four serial collagenase digestions in 0.2% collagenase type 2 (ThermoFisher Scientific, 17101015) in DMEM (Life Technologies, 11885092) at 37°C for 20 min with gentle rocking. After each of the first three digestions, bones were subjected to light centrifugation (230 g) for 5 min and then transferred to a fresh tube of collagenase. After the last digestion, the bones and cell suspension from the last digestion were centrifuged at 450 g for 5 min and the pelleted cells were resuspended in growth media (GM): low glucose DMEM (Life Technologies, 11885092), 10% fetal bovine serum (Life Technologies, 10437-028) and 1% penicillin/streptomycin (Life Technologies, 15140122). Selective enrichment of periosteal stem/progenitor cells was confirmed using FACS analysis (Fig. S2C).

Cells were trypsinized, resuspended in HBSS (Life Technologies: 14170161), supplemented with 2% fetal bovine serum (Life Technologies, 10437-028), 1% penicillin/streptomycin (Life Technologies: 15140122) and 1% HEPES (Life Technologies, 15630080) (complete HBSS), and stained with 1:300 diluted CD45-PE (Miltenyi Biotec, 130-117-498), TER119-PE (Miltenyi Biotec, 130-117-512, TIE2-PE (ThermoFisher Scientific, 12-5987-82), 6C3-PE (ThermoFisher Scientific, 12-5891-82) and CD90-PE (Invitrogen, MA5-17749), and 1:200 diluted CD51-BV421 (BD Biosciences: 740062), CD105-PE-Cy7 (ThermoFisher Scientific: 25-1051-82) and CD200-BV711 (BD Biosciences: 745548) for 30 min on ice in the dark. Cells were then washed with 1 ml of the complete HBSS solution, centrifuged at 500 g for 5 min and finally resuspended with complete HBSS for flow cytometry. Cells were sorted on a Sony Biotechnology SY3200 cell sorter into a 50%/50% solution of complete HBSS and fetal bovine serum or analyzed on a Bio-Rad ZE5 Analyzer. Sorting was validated to result in >95% purity of the intended population in post-sort fractions. Beads (eBioscience 01-1111-41) were used to set initial compensation. Fluorescence minus one (FMO) controls were used for additional compensation and to assess background levels for each stain. We excluded doublets and gates were drawn as determined by internal FMO controls to separate positive and negative populations for each cell-surface marker. Mesenchymal cell populations negative for CD45, CD31 and TER119 cell-surface markers were analyzed according to the approach described in Fig. S4G,H.

All procedures followed protocols approved by the NYU Robert I. Grossman School of Medicine Committee on Animal Research. Mice were anesthetized with 1-4% isoflurane inhalation. A 4 mm incision was made over the proximal anteromedial tibia, and the tibial surface was exposed while carefully preserving the periosteum. A 1.0 mm hole was drilled through the anterior cortex with a high-speed dental drill (10,000 rpm). Incisions were closed with 5-0 Vicryl sutures. Before and after surgery, mice received subcutaneous injections of buprenorphine for analgesia and were allowed to ambulate freely. Twenty-four hours before euthanasia mice received 200 μg of EdU to label proliferating cells. Mice were euthanized at indicated days after surgery.

Mice were anesthetized using isoflurane inhalation. A 4 mm incision was made over the anteromedial tibial region. The medial and proximal tibia was exposed. Next, a 27G×1-1/4 needle was inserted through the medullary cavity of the tibia to stabilize the ensuing fracture and the tibia was then transected in the mid-diaphysis with small surgical scissors. Hox inhibition was achieved locally by adsorbing 50 nM HoxMix (10 nM each relevant Hox gene) or 50 nM NT Control siRNA into a 2×2×2 mm collagen sponge (DSI Dental Solutions) along with HiPerfect Transfection Reagent (Qiagen) according to the manufacturer's instructions and inserting the sponge under a muscle flap over the tibial fracture site. Transfection efficiency of the collagen sponge delivery method was similarly ascertained with a GFP-tagged siRNA (MISSION siRNA Fluorescent Universal Negative Control; Millipore Sigma, SIC003).

Conversely, to overexpress Hoxa10, either LV-GFP or LV-Hoxa10/GFP was combined with pLenti-rtTA (all lentiviral constructs at a M.O.I. of 150) and 2.5 μg/ml polybrene. The mixture was then adsorbed into a 2×2×2 mm collagen sponge (DSI Dental Solutions) that was then inserted under a muscle flap over the tibial fracture site.

For all experiments, incisions were closed with 5-0 Vicryl sutures. Before and after surgery, mice received subcutaneous injections of buprenorphine for analgesia and were allowed to ambulate freely.

Tibias were scanned using a high-resolution SkyScan μCT system (SkyScan 1172, Bruker). Images were acquired at 9 μm isotropic resolution using a 10 MP digital detector, 10 W energy (100 kV and 100A) and a 0.5 mm aluminum filter with a 9.7 μm image voxel size. A fixed global threshold method was used based on the manufacturer's recommendations and preliminary studies showed that mineral variation between groups was not high enough to warrant adaptive thresholding. The samples were oriented and the volume of interest (VOI) defined with the CTAn software (Bruker). The VOI was contoured manually to capture the entire callus region. The parameters selected to show variations between groups were total bone volume (BV), total tissue volume (TV), relative mineralized volume fraction (BV/TV), trabecular number (Tb.N), trabecular thickness (Tb.Th) and trabecular spacing (Tb.Sp) following the guidelines described by Bouxsein et al. (2010).

5×10⁴ PSPCs or sorted PP1 cells were seeded onto individual wells of 24-well plates [wells were first coated with a 1:100 dilution of fibronectin (Sigma: F0895) in PBS for 60 min] in GM and allowed to attach overnight. The next day, cells were stimulated with osteogenic media (OM) [DMEM, 10% FBS, 100 μg/ml ascorbic acid, 10 mM β-glycerophosphate and 1% penicillin/streptomycin]. Media were replenished every 2-3 days. For adipogenic differentiation, the cells were induced the next day using the MSC Adipogenic BulletKit (Lonza) induction media.

Primary PSPCs were transfected with a commercially available GeneSolution siRNAs targeting Hoxa10 (CACCACAATTCTCCCTATTTA; Qiagen, SI01068753), Hoxd10 (CCGAACAGATCTTGTCGAATA; Qiagen, SI00206542), Hoxc10 (TTCGAGGATGCTTCCACCTAA; Qiagen, SI01069306), Hoxa11 (CACCACTGATCTGCACCCAAA; Qiagen, SI01068788) and Hoxd11 (CCCGTCGGACTTCGCCAGCAA; Qiagen, SI01069558). AllStars Negative Control siRNA (Qiagen, 1027281) was used as a non-targeting control. Each component siRNA of HoxMix was delivered at 5 nM, yielding a total HoxMix concentration of 25 nM; non-targeting control siRNA was delivered at 25 nM. HiPerfect Transfection Reagent (Qiagen) was used as a transfection reagent according to the manufacturer's instructions. Transfection was carried out at the moment of seeding onto multiwell plates before the cells fully attached to the plates (fast-forward transfection). The seeded cells were treated with siRNAs every 3 to 4 days, and samples were assayed by qRT-PCR or flow cytometry after 2-14 days of knockdown.

5×10⁴ PSPCs were seeded onto wells of 24-well plates in GM and simultaneously administered either HoxMix or nontargeting siRNAs. After 24 h, the cells were incubated with 10 μM EdU at 37°C for 15 h, washed with PBS and then trypsinized. The Click-iT Plus EdU Alexa Fluor 488 Flow Cytometry Kit (ThermoFisher Scientific, C10632) was used to fix, permeabilize and label EdU-incorporated cells with a Click-iT reaction according to the manufacturer's instructions before subjecting the cells to flow cytometry analysis thereafter.

CellTrace Far Red Cell Proliferation Kit (ThermoFisher Scientific, C34564) was used according to the manufacturer's instructions. In the RNA interference experiments, PSPCs isolated from C57BL/6 wild-type mice were first expanded in vitro, and 5×10⁴ cells were then seeded onto individual wells of a 24-well plate with HoxMix or nontargeting siRNA. 24 h later, the cells from each well were trypsinized, incubated with CellTrace for 1 h at 37°C on day 0, then replated and cultured for 6 days. On day 6, cells were trypsinized and stained for PDGFRα (Invitrogen, 25-1401-82). A separate batch of cells was also trypsinized and incubated with CellTrace for 1 h at 37°C on day 6 to serve as a positive control.

For the overexpression experiments, PSCs, PP1 and PP2 cells were sorted from in vitro-expanded PSPCs. 3×10⁴ cells were incubated with either LV-GFP or LV-Hoxa10/GFP in individual wells of a 24-well plate. After 24 h, the procedure was then continued as described above. Cells in this case were stained with the lineage cell surface markers previously described (Debnath et al., 2018). Cells were then analyzed on a BD Biosciences LSRII UV cell analyzer for dye dilution and surface marker profile.

RNA was either isolated from cells immediately after periosteal isolation and FACS to observe in vivo gene expression or from cells expanded in vitro. The RNeasy Plus Kit (Qiagen, 74134) was used to isolate RNA and remove genomic DNA. RNA was then reverse-transcribed with the iScript cDNA Synthesis Kit (Bio-Rad, 170-8891). Quantitative real-time PCR was carried out using the Applied Biosystems QuantStudio3 system and RT² SYBR Green ROX PCR Master Mix (Qiagen, 330523). Specific primers were designed using Harvard PrimerBank (http://pga.mgh.harvard.edu/primerbank/) (Table S4). Results are presented as 2^–ΔΔCt values normalized to the expression of 18s. Mean and s.e.m. were calculated in GraphPad Prism 7 software.

Lentiviral DNA containing either a Hoxa10 CDS expression construct or a control construct lacking Hoxa10 CDS was generated at Genewiz using the Tet-ON system. In addition to the Hoxa10 sequence, the lentiviral vector used (ptetO) also include EGFP, luciferase and puromycin cloned downstream of the active CMV promoter. 2A peptide sequences are also included between each element (ptetO-Hoxa10-T2A-EGFP-P2A-Luciferase-T2A-Puromycin; LV-Hoxa10/GFP and ptetO-EGFP-P2A-Luciferase-T2A-Puromycin; LV-GFP) in order to produce multi-cistronic, equimolar expression of all four genes. EGFP was used to track the cells that have been infected in culture. Identical methods were used to generate lentiviral sequences containing Hoxa11, Hoxb8 and Hoxa5. pLenti-rtTA3 (Addgene #26429), pRSV-Rev (Addgene #12259), pMD2.G (Addgene #12259), and pMDLg/pRRE (Addgene #12251) plasmids were purchased and lentivirus was generated in the Lenti-X 293T Cell Line (TakaraBio, 632180), purified with a Lenti-X Maxi Purification Kit (TakaraBio, 631234) and titered with a Lenti-X qRT-PCR Titration Kit (TakaraBio, 631235).

3×10⁴ sorted PSCs, PP1, PP2 or in vitro-expanded PSPCs were seeded onto individual wells of 24-well plates [wells were first coated with a 1:100 dilution of fibronectin (Sigma, F0895) in PBS for 60 min] using GM made with 10% heat-inactivated fetal bovine serum and 1% penicillin/streptomycin. The cells were immediately transduced with pLenti-rtTA and LV-GFP, LV-Hoxa10/GFP, LV-Hoxa11/GFP, LV-Hoxb8/GFP or LV-Hoxa5/GFP at a M.O.I. of 75. The transduction efficiency was aided by the addition of 2.5 μg/ml polybrene (Sigma) and the cells were treated with 10 μg/ml doxycycline to activate expression downstream of the tetO promoter sequences. GM with 10 μg/ml doxycycline was used to replace the media every 2-3 days.

The periosteal scRNA-seq dataset is from a publicly available adult mouse femoral periosteum study (Debnath et al., 2018). We obtained the raw count matrix from the GEO deposit GSE106236 and annotated cells based on high expression levels of the genes associated with the cell-surface markers used for flow cytometry [6C3 (Enpep), CD90 (Thy1), CD51 (Itgav), CD105 (Eng) and CD200 (Cd200)]. The PSC, PP1 and PP2 cells in the count matrix were then sorted in silico according to a previously described gating strategy presented in the results (Debnath et al., 2018); genes with a high fold change between PSCs and PP1/PP2 cells were used to identify potential PSC markers.

A model of mesenchymal stem cell differentiation was used to compare the regenerative potential of PSPCs. 12- to 15-week-old syngeneic C57BL/6 mice were used as hosts for the renal capsule transplantation assay. An incision was made on the dorsal skin surface, followed by an incision through the peritoneum, and the kidneys were identified and then exteriorized. An incision in the renal capsule was made using a 27-gauge needle. Tibial bone marrow (2 μl, containing ∼100,000 cells) from 12-week-old C57BL/6 mice was used to resuspend 750 transduced (GFP⁺) PP1 cells. The mixture was then left exposed to open air for 1-2 min to allow a limited amount of coagulation to take place and subsequently grafted beneath the capsule. The kidney was placed back into its anatomical location and the peritoneum was closed using a Vicryl suture, followed by skin closure with a 6-0 Vicryl suture. Mice had ad libitum access to food and water (with dissolved 0.4 mg/ml doxycycline and 5% sucrose) and received subcutaneous buprenorphine for analgesia. Mice were euthanized 14 days post-surgery, and the renal grafts were harvested, digested in 0.2% collagenase type 2 (ThermoFisher Scientific) in DMEM at 37°C for 1 h, stained with antibodies and subjected to FACS. The GFP⁺ cells were collected and reused for the subsequent renal grafts.

We thank Gozde (Gina) Yildirim, MS (NYU College of Dentistry) for assistance with the μCT imaging [funded by the National Institutes of Health (S10OD010751)] and the Preclinical Imaging Laboratory [a shared resource partially supported by a Laura and Isaac Perlmutter Cancer Center support grant (NIH/NCI 5P30CA016087) and by a NIBIB Biomedical Technology Resource Center Grant (NIH P41 EB017183)]. Cell sorting/flow cytometry technologies were provided by NYU Langone's Cytometry and Cell Sorting Laboratory, which is supported in part by grant P30CA016087 from the National Institutes of Health/National Cancer Institute. RNAseq/ATACseq was performed in the NYU Langone's Genome Technology Center, partially supported by the Cancer Center Support Grant P30CA016087 at the Laura and Isaac Perlmutter Cancer Center. We thank the Genotyping Core Laboratory at NYU Langone for their assistance.