Targeted clearance of p21‐ but not p16‐positive senescent cells prevents radiation‐induced osteoporosis and increased marrow adiposity

Abstract Cellular senescence, which is a major cause of tissue dysfunction with aging and multiple other conditions, is known to be triggered by p16Ink4a or p21Cip1, but the relative contributions of each pathway toward inducing senescence are unclear. Here, we directly addressed this issue by first developing and validating a p21‐ATTAC mouse with the p21Cip1 promoter driving a “suicide” transgene encoding an inducible caspase‐8 which, upon induction, selectively kills p21Cip1‐expressing senescent cells. Next, we used the p21‐ATTAC mouse and the established p16‐INK‐ATTAC mouse to directly compare the contributions of p21Cip1 versus p16Ink4a in driving cellular senescence in a condition where a tissue phenotype (bone loss and increased marrow adiposity) is clearly driven by cellular senescence—specifically, radiation‐induced osteoporosis. Using RNA in situ hybridization, we confirmed the reduction in radiation‐induced p21Cip1 ‐ or p16Ink4a ‐driven transcripts following senescent cell clearance in both models. However, only clearance of p21Cip1+, but not p16Ink4a+, senescent cells prevented both radiation‐induced osteoporosis and increased marrow adiposity. Reduction in senescent cells with dysfunctional telomeres following clearance of p21Cip1+, but not p16Ink4a+, senescent cells also reduced several of the radiation‐induced pro‐inflammatory senescence‐associated secretory phenotype factors. Thus, by directly comparing senescent cell clearance using two parallel genetic models, we demonstrate that radiation‐induced osteoporosis is driven predominantly by p21Cip1‐ rather than p16Ink4a‐mediated cellular senescence. Further, this approach can be used to dissect the contributions of these pathways in other senescence‐associated conditions, including aging across tissues.


| INTRODUC TI ON
Cellular senescence is a non-proliferative and apoptosis-resistant state, which can be induced by different stressors including oxidative DNA damage, mitochondrial stress, proteostasis, and other stimuli and is characterized by the production of pro-inflammatory and matrix-degrading proteins that are components of the senescenceassociated secretory phenotype (SASP) (Tchkonia et al., 2013).
Senescent cells play important roles in tumor suppression, development, and tissue regeneration (Tchkonia et al., 2013). During normal physiological conditions, particularly in youth, senescent cells are cleared by the immune system (Prata et al., 2018). However, with a waning immune system and excessive accumulation of senescent cells with increasing age, disease, and post-therapy (e.g., radiation or chemotherapy), senescent cells are deleterious and associated with tissue dysfunction and overall morbidity .

Targeted clearance of p16 Ink4 -expressing cells in p16-INK-ATTAC
(p16 Ink4a apoptosis through targeted activation of caspase) mice in the presence of AP20187 (a synthetic drug that dimerizes a membranebound myristoylated FK506-binding protein-caspase-8  fusion protein) has been successfully used as a strategy to understand the effect of clearing p16 Ink4a -expressing senescent cells on multiple age-related conditions, including osteoporosis, frailty, cardiovascular disease, and metabolic dysfunction (for reviews, see (Tchkonia et al., 2013) and ). However, although the roles of p21 Cip1 or p16 Ink4a as markers of senescence have been well described (Baker et al., 2011(Baker et al., , 2013, their individual contributions in driving senescence more generally are largely unknown. In the present study, we directly address this issue by first constructing and validating a new mouse model, p21-ATTAC. Then, in parallel study designs using the p21-ATTAC and the analogous p16-INK-ATTAC mouse (Baker et al., 2011), we compare the effects of clearing either p21 Cip1 -or p16 Ink4a -expressing senescent cells in a condition where a tissue phenotype (bone loss and increased marrow adiposity) is clearly driven by cellular senescence-specifically, radiation-induced osteoporosis (Chandra et al., 2020). These studies both dissect the relative roles of the p21 Cip1 and p16 Ink4a pathways in causing radiation-induced osteoporosis and also establish a system where these mice can be used to evaluate the relative contributions of these pathways in causing senescence in other conditions, including age-related disorders.

| Generation and validation of the p21-ATTAC model
Details regarding the generation of the p21-ATTAC mice are provided in the Methods. Briefly, Figure S1a provides a schematic of the model, where the p16 Ink4a promoter is replaced by a 3.2-kb genomic fragment containing promoter sequences directly upstream of the mouse p21 Cip1 transcriptional start site (López-Domínguez et al., 2021). This 3.2-kb fragment contains the appropriate sequences necessary for induction of p21 Cip1 , including three p53-binding sites, which has been validated both in vitro (el-Deiry et al., 1995) and in vivo (Wang et al., 2021); further validation of this promoter is also provided below. As in the p16-INK-ATTAC model, in the p21-ATTAC mice the p21 Cip1 promoter drives the FKBP-Caspase-8 fusion protein, ATTAC, allowing for selective elimination of senescent cells expressing high levels of p21 Cip1 following treatment with AP20187, a synthetic drug with no known off-target effects. Mice expressing this transgene were then generated using site-specific integration into the Rosa26 locus (Figure S1b), as described in the Methods.
Prior to further characterization, we evaluated whether treating young (3-6-month old) p21-ATTAC mice with a low senescent cell burden but basal expression of p21 Cip1 across tissues with AP20187 senescent cells with dysfunctional telomeres following clearance of p21 Cip1 +, but not p16 Ink4a +, senescent cells also reduced several of the radiation-induced proinflammatory senescence-associated secretory phenotype factors. Thus, by directly comparing senescent cell clearance using two parallel genetic models, we demonstrate that radiation-induced osteoporosis is driven predominantly by p21 Cip1 -rather than p16 Ink4a -mediated cellular senescence. Further, this approach can be used to dissect the contributions of these pathways in other senescence-associated conditions, including aging across tissues.

K E Y W O R D S
bone, radiation, senescence would have any adverse effects. For this, we performed a doseescalation study (Figure S1c), where the final dose was the 10 mg/ kg of AP20187 used in the subsequent experiments. All mice tolerated this dose-response study with no abnormal physical signs and normal weight curves ( Figure S1d). In addition, at sacrifice, detailed pathological examination found no abnormal pathologies across tissues in the AP20187-versus the vehicle-treated mice.
We next performed both an in vivo validation of the promoter used in the p21-ATTAC construct and also a side-by-side comparison of the induction of the p21 Cip1 versus the p16 Ink4a promoters in a model known to induce both genes, radiation-induced osteoporosis (Chandra et al., 2020). Because the ATTAC construct contains an Egfp reporter ( Figure S1a), the expression of Egfp is specific for activation of the transgene in each model. For these and the subsequent studies, 24 Gy focal radiation treatment (FRT) was delivered to a 5 mm area of the right femoral metaphysis and the p21-ATTAC FRT has no bone-damaging effects on the contralateral femurs (Chandra et al., 2014(Chandra et al., , 2017(Chandra et al., , 2019. As shown in Figures 1 and S2, radiation treatment resulted in equivalent induction of both the p21 Cip1

| Targeted removal of p21 Cip1 -, but not p16 Ink4aexpressing cells reduces senescent cell burden and SASP following radiation
To further validate the p21-ATTAC mice, we used RNA-ISH, which demonstrated the presence of three populations of senescent cells (p21 Cip1 -expressing, p16 Ink4a -expressing, and p21 Cip1 /p16 Ink4a dualexpressing) ( Figure S3a). AP20187 treatment of the p21-ATTAC mice When exposing cells to DNA damaging agents, such as radiation, damage at telomeres is less efficiently repaired compared with the rest of the genome (Fumagalli et al., 2012;Hewitt et al., 2012). Therefore, telomeric lesions are extremely long-lived and in fact have been shown to persist for months to years both in vitro and in vivo (Anderson et al., 2019). Telomere-associated foci (TAF; also known as telomere dysfunction-induced foci [TIF]), which assess telomeric DNA damage, are increasingly considered the most definitive marker of senescence across tissues, including in bone cells (Chandra et al., 2020;Farr et al., 2016;Wang et al., 2012). We evaluated TAF+ osteocytes as they represent the largest fraction of bone cells and have been implicated as a major senescent cell component in age-related bone loss (Farr et al., 2017). As expected, the percentage of TAF+ osteocytes increased markedly in the radi- These data thus indicate that, in radiation-induced osteoporosis, clearance of p21 Cip1 -, but not p16 Ink4a -expressing cells results in a reduction in senescent osteocytes in bone.
Senescent cells are associated with the expression of proinflammatory SASP markers (Coppe et al., 2010). A reduction in SASP factors was one of the key findings associated with the pharmacological clearance of senescent cells by the senolytic cocktail of Dasatinib and Quercetin (D+Q) in age- (Farr et al., 2017) and radiation-related bone loss (Chandra et al., 2020), as well as during genetic clearance of p16 Ink4a -expressing senescent cells in p16-INK-ATTAC aged mice (Farr et al., 2017). Thus, we evaluated a panel of SASP factors that we have previously found to be upregulated following radiation in bone (Table S1) (Chandra et al., 2020) and Figure 3 shows the SASP factors that were downregulated following AP20187 treatment in either the p21-ATTAC or the p16-INK-ATTAC mice. Thus, Il6, Mmp12, Ccl2, and Ccl7 were significantly reduced upon AP20187 treatment in radiated bones of p21-ATTAC mice (Figure 3a), but not in p16-INK-ATTAC mice (Figure 3b). By contrast, Ccl4 was reduced in p16-INK-ATTAC mice following AP20187 treatment in the radiated bones, but not in p21-ATTAC mice.

| Targeted removal of p21-, but not p16-expressing senescent cells prevents radiation-induced bone loss
We next evaluated the skeletal consequences of clearance of either

| Targeted removal of p21-, but not p16 Ink4aexpressing senescent cells maintains BMSC cell fate and bone formation
Radiation-induced DNA damage induces cellular apoptosis (Chandra et al., 2014) and senescence (Chandra et al., 2020) of osteoblasts F I G U R E 1 Validation of the p21-ATTAC mouse model using radiation as an inducer of senescence. (a) Schematic showing the experimental design for the p21-ATTAC mice. The right legs of the mice were radiated (24 Gy) near the femoral metaphysis (5 mm above the growth plate), while the left leg served as control. Starting from day 1 post radiation, the animals received vehicle or AP20187 for 2 times per week for 6 weeks. Radiated (R) and nonradiated (NR) femurs at 42 days post-radiation were collected for qRT-PCR. (b) The Egfp transgene was activated in R bones, and the expression levels were significantly reduced in the AP20187-treated animals. Statistical comparisons between the groups (left panel in b) was done by an ordinary two-way ANOVA, with a Tukey's post-hoc analysis. (Right panel in b) The Egfp expression was normalized to the NR control leg for each animal. Statistical comparison was done using a two-tailed unpaired t test between the Veh-R and AP20187-R bones. (c) RNA-in situ hybridization (RNA-ISH) was performed using probes against p21 Cip1 (shown in red) and Egfp (shown in green) transcripts. (d) Four populations of bone marrow cells, one with DAPI alone (a′), second expressing p21 Cip1 (b′), a third expressing Egfp (c′), and a fourth expressing both p21 Cip1 and Egfp (d′) were used to generate two kinds of data, one quantifying Egfp and p21 Cip1 foci per cell as shown in (e), and second quantifying percentage of bone marrow cells that expressed p21 Cip1 and Egfp, Egfp alone and p21 Cip1 alone (f). Statistical comparisons were calculated using an ordinary two-way ANOVA, with a Tukey's post-hoc analysis p21/ Egfp

Egfp
Relative mRNA levels

Egfp+p21+ cells (%)
and osteocytes. These changes in osteoblast viability or function result in reduced mineral apposition rate (MAR) (Chandra et al., 2014(Chandra et al., , 2018Oest et al., 2015). We observed a decline in total osteoblasts in radiated bones of both vehicle-and AP-treated p21-ATTAC and

| Further characterization of p21 Cip1 + cells following radiation
Given the importance of p21 Cip1 + senescent cells in radiation-induced bone loss, we did additional studies to further characterize these F I G U R E 2 Radiation-induced senescent osteocytes with dysfunctional telomeres following clearance of p21 Cip1 -and p16 Ink4a -expressing cells. The right legs of p21-ATTAC mice were radiated (24 Gy, R) near the femoral metaphysis (5 mm above the growth plate), while the left leg (NR) served as control. The animals were assigned to vehicle (n = 4) or AP20187 (n = 5) treated groups. For the p16-INK-ATTAC mice, animals received the radiation dose identical to the p21-ATTAC mice and were assigned to vehicle (n = 3) or AP20187 (n = 4) treated groups. R-and NR-femurs at 42 days post-radiation were collected and processed for MMA embedding. (a,c) 5 μm deplasticized sections were processed for TAF staining and TAFs were detected. Telomeres (red) and γH2AX (green) are shown in osteocytes (OCY) of non-decalcified R femurs 42 days post-radiation. The co-localization is indicated by yellow TAF foci. Representative images of TAF+ (NR and R) osteocytic nuclei are shown and TAF+ osteocytes were quantified in NR and R femurs of vehicle-and AP20187-treated animals (b,d). Statistical comparisons were calculated using an ordinary two-way ANOVA, with a Tukey's post-hoc analysis

| DISCUSS ION
In the present study, we generated and validated a new mouse model, p21-ATTAC, and also used an established mouse model, (Baker et al., 2011), to directly compare the contributions of p16 Ink4a -versus p21 Cip1 -driven cellular senescence in a condition where the phenotype is known to be driven, at least in part, by cellular senescence-specifically, radiation-induced osteoporosis (Chandra et al., 2020). Despite the similar induction of These findings thus demonstrate that radiation-induced bone loss is predominantly, if not exclusively, a p21 Cip1 -driven process. To our knowledge, ours is the first study to use genetic models to dissect p16 Ink4a -versus p21 Cip1 -driven cellular senescence and provide an approach to further characterize these pathways in other conditions, including the myriad age-related disorders now associated with cellular senescence .
From a translational perspective, these studies may also guide the development of novel senolytic drugs that specifically target either p21 Cip1 -or p16 Ink4a -driven cellular senescence, perhaps allowing for more tailored approaches to treat various senescenceassociated conditions. We acknowledge, however, that our validation of the p21-ATTAC model was done in a specific condition, radiation-induced osteoporosis, and additional studies evaluating this model in other tissues and conditions associated with senescence need to be done.
Our findings are consistent with a previous report that found that in the absence of p21 Cip1 , mouse embryos were impaired in their ability to undergo senescence following radiation (Brugarolas et al., 1995). In addition, the time course of induction of p21 Cip1 or p16 Ink4a does differ following radiation both in vitro (López-Domínguez et al., (Chandra et al., 2020). Thus, in cultured mouse der- We also assessed whether the selective clearance of p21 Cip1 -or p16 Ink4a -expressing cells affected SASP expression. Interestingly, while radiation-induced upregulation of Il6, Mmp12, Ccl2, and Ccl7

2021) and in vivo
was prevented by clearance of p21 Cip1 -expressing cells, the clearance of p16 Ink4a -expressing cells did not affect these SASP markers.
Conversely, the radiation-induced upregulation of Ccl4 (macrophage inflammatory protein-1β (MIP-1β)) was attenuated following clearance of p16 Ink4a -, but not p21 Cip1 -expressing cells. These data further support that p21 Cip1 and p16 Ink4a are non-redundant pathways in our radiation-induced osteoporosis model.  bones and, based on the data noted above, the time course of increases in p21 Cip1 versus p16 Ink4a may also be of crucial importance.
The in vivo genetic clearance of p16 Ink4a -expressing senescent cells in INK-ATTAC mice has been used as an important tool to study primary mechanism(s) associated with several age-related conditions, including osteoarthritis and osteoporosis (Baker et al., 2016(Baker et al., , 2011Farr et al., 2017;Jeon et al., 2017;Palmer et al., 2019;Xu et al., 2015Xu et al., , 2018. Abrogation of CDKi's does come with the concern over lowering the threshold for neoplastic transformation; however, senescent cell deletion of p21 Cip1 in various mouse models of premature aging did not increase tumor incidence (Benson et al., 2009). Moreover, in the case of radiation-induced osteoporosis, approaches to pharmacologically clear senescent cells and perhaps more specifically, p21 Cip1 -expressing senescent cells, are likely to involve short-term treatment, minimizing concerns regarding tumorigenesis.
Bone marrow adiposity is negatively associated with bone mass across species (Devlin & Rosen, 2015). Using lineage tracing of mesenchymal cells, we previously reported that the decline in BMSCs following radiation was inversely correlated with increases in bone marrow adipocytes (Chandra et al., 2017). In a recent study, we demonstrated Conversely, and consistent with the skeletal changes in the two models, clearance of p16 Ink4a -expressing senescent cells, which has been shown to negatively regulate marrow adiposity in an aging (chronic senescence) p16-INK-ATTAC model (Farr et al., 2017), did not have any effect on radiation-induced marrow adiposity.
Our group has previously demonstrated that clearance of p16 Ink4a -expressing senescent cells using the p16-INK-ATTAC model prevents age-related bone loss in mice (Farr et al., 2017). Whether clearance of p21 Cip1 -expressing senescent cells using the p21-ATTAC mice will also prevent age-related bone loss remains to be experimentally defined, but our data would indicate that, in contrast to aging, clearance of p16 Ink4a -expressing senescent cells does not prevent bone loss following radiation, which appears to be driven principally by p21 Cip1 -expressing cells. Clearly, further studies using both the p16-INK-ATTAC and p21-ATTAC mice with aging and other conditions are needed to address the specific roles of p16 Ink4a -vesrus p21 Cip1 -expressing cells in mediating senescence in these conditions.
In summary, our studies dissect p21 Cip1 -versus p16 Ink4amediated senescence in radiation-induced osteoporosis, an established senescence-associated condition (Chandra et al., 2020). Using novel genetic models, our studies provide the first unequivocal demonstration of a dissociation in vivo between these pathways in driving cellular senescence. These studies also advance our understanding of the underlying mechanism(s) of radiation-induced bone damage, demonstrating that p21 Cip1 -expressing senescent cells and F I G U R E 6 Assessment of marrow adiposity in radiated bones following clearance of p21 Cip1 -and p16 Ink4a -expressing cells. The right legs of p21-ATTAC and p16-INKATTAC mice were radiated (24 Gy) near the femoral metaphysis (5 mm above the growth plate), while the left legs served as control. The animals received either vehicle or AP20187 as described above. (a,c) 5 μm Goldner's trichrome stained MMA-embedded sections from animals that received either vehicle (n = 9) or AP20187 (n = 10) in p21-ATTAC mice, and vehicle (n = 9) or AP20187 (n = 9) in p16-INK-ATTAC mice, were used to quantify adipocyte numbers (Ad.N) and adipocyte volume (Ad.V) and normalized against bone marrow area (BMA), where BMA is defined as total area minus bone area. Ad.N/BMA and Ad.V/BMA are calculated in vehicleand AP20187-treated NR-and R-bones. Data from R-bones are normalized against the control NR-bones from each animal (b,d). Statistical comparisons were done using a two-tailed unpaired t test between the Veh-R and AP20187-R bones Baker, and J.L. Kirkland) have been described previously (Baker et al., 2011). The p21-ATTAC-attB transgenic plasmid was constructed as follows ( Figure S1b). The pBT378 vector (provided by Tasic et al., 2011) was digested with PmeI and SwaI, to remove all DNA sequences between the two attB recombination sites, and a Geneblock (Integrated DNA Technologies, Coralville, IA) containing a single MluI site was inserted using the Gibson Assembly Master Mix (New England Biolabs, Ipswitch, MA). Next, a 3.8-kb MluI/BssHII fragment, containing FKBP-Casp8 and IRES-EGFP sequences from the p16-INK-ATTAC transgenic construct (devised by P.
Scherer) (Pajvani et al., 2005), was cloned into the MluI site. The mice received focal radiation as described previously (Chandra et al., 2020). Briefly, mice received a dose of 24 Gy (6.6 Gy/min) on day 0, delivered focally to 5mm of the femoral metaphyseal region using X-Rad-SmART (Precision X-Ray Inc.

| Methyl methacrylate tissue embedding and histology
Radiated and non-radiated femurs were processed as described previously (Chandra et al., 2020). Briefly, at day 42 post-radiation the bones were processed for routine methyl methacrylate (MMA) embedding. Non-decalcified femurs were sectioned into 5 μm sections that were used for static histomorphometry and TIF assay, while 8 μm sections were used for dynamic histomorphometry.

| Micro-computed tomography (μCT) analysis
Bones were harvested 42-day post-focal radiation and scanned using μCT (vivaCT 40, Scanco Medical AG, Brüttisellen, Switzerland). The distal femur was scanned corresponding to a 1-5 mm area above the growth plate. All images were first smoothed by a Gaussian filter

| Quantitative RT-PCR
Bones were collected for mRNA isolation and qRT-PCR as described previously (Chandra et al., 2020). Briefly, a 5-mm region below the growth plate of the distal metaphyseal femur was cut out from the R and NR legs. After removal of muscle tissue, the bone samples were homogenized and total RNA was isolated using RNeasy Mini Columns (QIAGEN, Valencia, CA). cDNA was generated from mRNA using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems by Life Technologies, Foster City, CA) according to the manufacturer's instructions, and RT-qPCR was performed as described in our previous studies (Chandra et al., 2020). All primer sequences have been authenticated in previous studies (Chandra et al., 2020). Primers were designed so that they overlapped two exons. A detailed list of primer sequences is provided in Table S2.

| Static and dynamic histomorphometry
Static histomorphometry was performed using 5 μm Goldner's trichrome-stained sections, counting osteoblast numbers (N.Ob) and osteoclast numbers (Oc.N)/mm of bone surface. Cuboidal shaped cells located on bone surfaces were characterized as osteoblasts, while multinucleated cells on the bone surface with a well-defined pit were characterized as osteoclasts. The region of interest for all histomorphometric assessments was within the 5mm region of radiation and a corresponding 5mm region in the contralateral leg, below the growth plate. An area of 1 mm 2 0.6 mm below the growth plate was used for static and dynamic histomorphometric analysis. Adipocyte numbers (Ad.N) and volume (Ad.V) were quantified in an area of 1.72 mm 2 , 0.6 mm below the growth plate, normalized against bone marrow area (total tissue area minus bone area). Dynamic histomorphometry was performed on unstained deplasticized sections, and mineral apposition rate (MAR) was calculated as described previously (Chandra et al., 2020).
Briefly, mineralized bone was identified by alizarin (red/orange) and calcein (green) fluorescent dyes, with mineralization being a marker of functional osteoblasts, and the distance between the two dye fronts indicative of new bone formed in 7 days, calculated by MAR.
Both static and dynamic histomorphometry was performed using Cy-3-labeled telomere-specific (CCCTAA) peptide nucleic acid probe (Panagene), followed by hybridization for 2 h at RT (minimum) in a humidified chamber in the dark. Next, slides were washed once with 70% formamide in 2 × SSC for 10 min, followed by 1 wash in SSC buffer and PBS for 10 min. Sections were mounted with DAPI (ProLong™ Diamond Antifade Mountant, Invitrogen, P36962). In-depth Z-stacking was used (a minimum of 135 optical slices using a 63× objective).
Number of TAFs/cell was assessed by manual quantification of partially or fully overlapping (in the same optical slice, Z) signals from the telomere probe and γH2A.X in z-by-z analysis. Images were deconvolved with blind deconvolution in AutoQuant X3 (Media Cybernetics).

| RNA in situ hybridization (RNA-ISH)
RNA-ISH was performed per the RNAScope protocol from Advanced Cell Diagnostics Inc. (ACD): RNAScope Multiplex Fluorescent Assay v2. Briefly, the assay allows simultaneous visualization of up to 4 RNA targets, with each probe assigned to a different channel (C1, C2, or C3 or C4). p21 (CDKN1A) signal amplified using the HRP-C1 linked with a secondary fluorophore, Opal 570 (detected in the Cy3 range), followed by a subsequent detection of p16/p19 (CDKN2A) -c3 or EGFP-c2 probe, the signal is amplified by the HRP-C3 or HRP-C2 (respectively) linked with a secondary fluorophore, Opal 650 (de- Following three PBS washes for 5 min, sections were then incubated with a rat fluorescent secondary antibody for 1 h (goat anti-rat 647 Alexa Fluor A-21247, followed by PBS washes and mounted using ProLong Gold Antifade Mountant with DAPI (Invitrogen). Sections were imaged using a Leica microscope (dmi8, ×40 dry objective).

| Statistical analysis
Sample sizes were based on previously published data (Chandra et al., 2020;Farr et al., 2017), in which statistically significant differences were observed on bone with various interventions in our laboratory. For all experiments, data are expressed as median with interquartile range. Statistical comparisons were done using twotailed unpaired t tests. For multiple comparisons in Figures 2c,d, 3b, 4b,d, and 3a,b, the data were analyzed using a two-way ANOVA with a Tukey post hoc analysis. All statistical analyses were performed by GraphPad Prism.

ACK N OWLED G M ENTS
This work was made possible by the Eagle's Cancer Research