Oncogene-induced senescence is a potent barrier to tumorigenesis that limits cellular expansion following certain oncogenic events. Senescent cells display a repressive chromatin configuration thought to stably silence proliferation-promoting genes while simultaneously activating an unusual form of immune surveillance involving a secretory program referred to as the senescence-associated secretory phenotype (SASP). Here, we demonstrate that senescence also involves a global remodeling of the enhancer landscape with recruitment of the chromatin reader BRD4 to newly activated super-enhancers adjacent to key SASP genes. Transcriptional profiling and functional studies indicate that BRD4 is required for the SASP and downstream paracrine signaling. Consequently, BRD4 inhibition disrupts immune cell–mediated targeting and elimination of premalignant senescent cells in vitro and in vivo. Our results identify a critical role for BRD4-bound super-enhancers in senescence immune surveillance and in the proper execution of a tumor-suppressive program.
Significance: This study reveals how cells undergoing oncogene-induced senescence acquire a distinctive enhancer landscape that includes formation of super-enhancers adjacent to immune-modulatory genes required for paracrine immune activation. This process links BRD4 and super-enhancers to a tumor-suppressive immune surveillance program that can be disrupted by small molecule inhibitors of the bromo and extra terminal domain family of proteins. Cancer Discov; 6(6); 612–29. ©2016 AACR.
See related commentary by Vizioli and Adams, p. 576.
This article is highlighted in the In This Issue feature, p. 561
Cellular senescence is a cell-cycle arrest program that limits proliferation of damaged cells and can be triggered in response to diverse forms of cellular stress (1, 2). Although initially defined as a phenotype of cultured fibroblasts (3), cellular senescence appears to have a broad impact on human health and disease. For example, cellular senescence can restrict tumorigenesis (4, 5) and modulate chemotherapy responses (6–8), and may exert a primordial role in wound healing, tissue remodeling, and repair (9–13). Moreover, senescent cells have been observed in aged or damaged tissues (14) and can contribute to the decline of tissue regeneration capacity with age (15, 16). Accordingly, disruption of senescence regulators can promote cancer or impair wound healing, and the elimination of senescent cells from tissues can enhance lifespan in mice (17, 18).
Beyond the cell-cycle arrest phenotype, senescent cells display marked changes in cell morphology, show increased senescence-associated β-galactosidase (SA-β-gal) activity, and undergo dramatic changes in higher-order chromatin organization that are associated with a distinct transcriptional profile (2). Mechanistically, the best-understood feature of cellular senescence is its stable cell-cycle arrest program, which renders senescent cells insensitive to mitogenic stimuli and undoubtedly contributes to key senescence phenotypes (2, 19). This arrest program involves a complex interplay between the RB and p53 tumor suppressor pathways (19), and is accompanied by marked changes in the chromatin landscape involving the redistribution of cellular heterochromatin and leading to stable repression of cell-cycle–regulatory genes (20, 21). Such a potently repressive chromatin environment presumably buffers against the activation of proliferation genes in response to mitogenic cues.
Conversely, senescent cells activate the expression of genes known to influence the tissue microenvironment, including matrix remodeling enzymes, growth factors, and immune-modulatory cytokines (22, 23). As a consequence, senescent cells display qualitative and quantitative changes in their secretome, a feature collectively referred to as the senescence-associated secretory phenotype (SASP; refs. 24, 25). This gene activation program is controlled in part by factors such as NFκB (26), C/EBPβ (27), mTOR (28, 29), MacroH2A.1 (30), and GATA4 (31). Among their many functions, SASP proteins mobilize immune cells and modify their activities, for example, by recruiting natural killer (NK) cells or altering macrophage polarization to kill and/or engulf senescent cells (9, 32–34). These immune-modulatory activities may be important for the resolution of certain wound-healing responses (9, 10, 13) and underlie a unique form of immune surveillance that can eliminate premalignant or malignant cells undergoing senescence in vivo (33, 35, 36).
As is clear from the above, the biological roles of senescence rely on gene repression and activation programs that substantially alter cell function and fate. In certain developmental and pathologic contexts, such dramatic gene expression changes rely on dynamic remodeling of enhancer landscapes, regulatory regions that when activated induce gene expression and maintain downstream biological circuits. For example, in embryonic stem cells (ESC), exceptionally large stretches of H3K27Ac-marked regulatory elements, termed super-enhancers (SE), are associated with genes that play prominent roles in ESC self-renewal (37, 38). In other cell types, SEs can contribute to lineage- and cellular state–specific transcription, for example, by sustaining the expression of the oncogene MYC in leukemia and some solid tumors (37, 39) or driving proinflammatory responses in endothelial cells (40). Importantly, transcription of SE-associated genes relies on dense accumulation of cofactors such as the bromo and extra terminal domain (BET) proteins, and is highly sensitive to perturbations in cofactor binding (39, 40). BET proteins such as BRD4 bind to chromatin in a manner that can be disrupted with small-molecule drugs, and, as such, BET inhibitors are currently in development to suppress certain hyperinflammatory diseases and cancer (41–43).
Studies to date have explored enhancer dynamics in ESC self-renewal, cell lineage specification, inflammation, and pro-oncogenic contexts. Here, we examine whether enhancer dynamics contribute to oncogene-induced senescence (OIS), a potently tumor-suppressive program that limits the expansion of premalignant cells (5, 19). Through genome-wide chromatin profiling, we find that H3K27Ac-enriched enhancers undergo global remodeling in senescence, with the appearance of new SEs adjacent to key SASP genes. We further show that genetic or pharmacologic suppression of BRD4 collapses SASP gene expression and thereby prevents the proper immune targeting of senescent cells in vitro and in vivo. Our results establish that enhancer remodeling and BRD4 are required for senescence-associated immune surveillance and, consequently, for the proper execution of the tumor-suppressive senescence program.
Enhancers Undergo Global Remodeling during Oncogene-Induced Senescence
The mechanisms underlying the execution and maintenance of OIS have been best studied in human diploid fibroblasts (2, 19, 20). In this model, senescence is triggered by the introduction of oncogenic HRASV12, which provokes a proliferative burst followed by induction of p53, p16Ink4a, cell-cycle exit, and the production of SA-β-gal and SASP (19, 44). Previous work using this model established the importance of heterochromatin formation and transcriptional repression in the senescence-associated cell-cycle arrest program (20, 21) and contributed to the characterization of the SASP (25). To study enhancer biology during senescence, we transduced IMR90 normal human diploid lung fibroblasts with HRASV12 and allowed the cells to senescence. For comparison, we also generated parallel populations of proliferating IMR90 cells, and cells made quiescent by serum withdrawal. Importantly, although such quiescent cells can be triggered to re-enter the cell cycle upon serum addition, senescent cells cannot (2).
Following lysis of cross-linked proliferating, senescent, and quiescent IMR90 cells, the enhancer landscape was visualized using chromatin immunoprecipitation coupled with high-throughput sequencing (ChIP-Seq) using the histone modification H3K27Ac as a marker of active enhancers (45). This analysis identified a total of approximately 33,000 H3K27Ac-detectable enhancers in proliferating, quiescent, and senescent IMR90 cells (union enhancers; Fig. 1A). Previous studies have documented distinct classes of enhancers termed SEs, which accumulate disproportional levels of specific chromatin modifications and are associated with key cell identity genes (38, 39). By implementing the previous definition of SEs based on H3K27Ac ChIP-Seq profiling (39), we identified approximately 1,250 SEs present across the three cell conditions and classified the approximately 32,000 remaining as typical enhancers (TE; Fig. 1B and C).
To reveal enhancer dynamics during OIS, we next compared the H3K27Ac levels at enhancers in senescent cells to those in proliferating cells. Based on a cut-off of 2-fold H3K27Ac enrichment over the proliferating condition, our analysis called approximately 6,500 senescence-activated TEs and approximately 7,090 senescence-inactivated TEs (Fig. 1B; Supplementary Table S1A and S1B), as well as 198 senescence-activated SEs and 191 senescence-inactivated SEs (Fig. 1C; Supplementary Table S1C and S1D), suggesting that over 40% of all enhancers in IMR90 cells are remodeled upon senescence. Interestingly, although the majority of the senescence-activated TEs and SEs displayed the H3K4Me1 mark in proliferating IMR90 cells (Supplementary Fig. S1A and S1B), senescence was found to selectively remodel the active mark H3K27Ac in a manner that distinguished this condition from both proliferating and quiescent cells. In particular, H3K27Ac signals gained and lost in senescent cells were not similarly altered during quiescence (Fig. 1D and E), indicating that such global enhancer remodeling was specific to the senescence condition and not an indirect consequence of cell-cycle arrest per se. H3K27Ac ChIP-Seq profiles of exemplary TEs and SEs specifically activated or inactivated during OIS are depicted in Fig. 1F–I. Importantly, these results were reproducible across independent biological replicates (Supplementary Fig. S2A and S2B; top plots). Altogether, these analyses reveal a global remodeling of active enhancers upon induction of OIS that occurs in a manner that distinguishes senescent cells from proliferating and quiescent counterparts.
Senescence-Activated SEs Correlate with the SASP Transcriptional Profile
To address the impact of enhancer remodeling observed during OIS on global gene expression changes, we next performed genome-wide transcriptome profiling in proliferating, quiescent, and senescent IMR90 cells by RNA sequencing (RNA-Seq). Unsupervised hierarchical clustering analysis grouped proliferating and quiescent samples closely together and clearly segregated them from the senescent samples (Fig. 2A). Importantly, gene set enrichment analysis (GSEA) showed a highly significant correlation between enhancer activity and transcriptional levels of the associated genes for both senescence-activated and senescence-inactivated TEs and SEs (Fig. 2B–E; FDR q values = 0.00). Consistent with the specificity of the global enhancer remodeling during senescence (see Fig. 1), GSEA failed to identify similar correlations between the senescence-associated enhancer changes and the transcriptional output of quiescent versus proliferating cells (Supplementary Fig. S3A–S3D).
We next used gene ontology (GO) analysis to gain insight into the biological processes controlled by both categories of enhancers remodeled during OIS. GO analysis of the genes associated with senescence-activated TEs indicated an enrichment of negative regulators of the MAPK pathway and the Toll-like receptor (TLR) signaling pathway (Fig. 2F and Supplementary Fig. S3E, top plot), which have been previously implicated in senescence (46, 47). On the other hand, TEs inactivated in senescent cells were linked to genes involved in telomere maintenance, chromosome localization, and DNA strand elongation/replication categories (Fig. 2G and Supplementary Fig. S3E, bottom plot), consistent with the cell-cycle arrest that accompanies senescence (19). These genes included many E2F targets (Supplementary Fig. S3E, bottom plot; Supplementary Table S1B), which are silenced during OIS through repressive histone marks (20, 21). Interestingly, despite the fact that E2F target genes are commonly downregulated in both quiescence and senescence (Supplementary Fig. S3F and S3G), their associated TEs were only found substantially remodeled in senescent cells (Supplementary Fig. S3H and S3I; and exemplified by the enhancer of the E2F target gene MCM6 shown in Fig. 1G). This result is consistent with the known resistance of senescent cells (compared with quiescent cells) to induce growth-promoting genes following a mitogenic stimulus (2) and suggests that senescence-specific enhancer inactivation contributes to, or is a prerequisite for, the stable repression of cell-cycle genes.
Dramatic changes were also observed in the SE landscape of senescent cells, with a robust activation of SEs in senescent compared with quiescent cells (Supplementary Fig. S3J). GO analysis on genes associated with such senescence-activated SEs indicated an enrichment of factors having cytokine and growth factor activity, including the well-described SASP factors IL1A, IL1B, IL8, INHBA, and BMP2 (Fig. 2H and Supplementary Fig. S3K; refs. 24, 26). In contrast, SEs inactivated in senescent cells were associated with genes related to DNA binding transcription factors (TF) and fibronectin binding categories (Fig. 2I and Supplementary Fig. S3K), the latter process likely reflecting the well-established reduction in extracellular matrix production that accompanies senescence (48). Examples of enhancer-regulated genes selectively activated or inactivated in senescent cells are depicted as heat maps in Fig. 2J–M. Together, these results connect enhancer remodeling to discrete senescence-associated transcriptional profiles and suggest that TE inactivation likely contributes to cell-cycle arrest, whereas SE activation might underlie the robust expression of SASP genes in senescent cells.
BRD4 Is Recruited to Senescence-Activated SEs Associated with Key SASP Genes
One of the proteins that bind acetylated histones at enhancers and are enriched in SEs is the transcriptional coactivator BRD4, which couples enhancer remodeling to RNA Pol II elongation (39). To address changes in BRD4 occupancy during OIS, we next performed ChIP-Seq profiling for BRD4 in proliferating, quiescent, and senescent IMR90 cells. Despite similar numbers of overall raw and mapped reads across all three conditions (see Supplementary Fig. S2C), BRD4 binding at H3K27Ac enhancer elements showed a progressive decline in signal strength from proliferating to quiescent to senescent cells (Fig. 3A). Although the reduced signal correlated with a decrease in total BRD4 protein in nonproliferating versus proliferating conditions (Fig. 3B), this may also reflect nuances of the ChIP-Seq assay in senescent cells, which is less sensitive owing to the global compaction of chromatin (49). Nevertheless, comparison of BRD4 occupancy at senescence-activated H3K27Ac enhancer elements in senescent versus proliferating cells (log2 fold change > 0.5) uncovered a robust and reproducible local enrichment of BRD4 at specific enhancer loci (see Supplementary Fig. S2A and S2B, bottom plots). In particular, 1,307 TEs and 37 SEs significantly gained BRD4 binding during the establishment of senescence (Fig. 3C and D; Supplementary Table S2A and S2B), which will be referred to as BRD4-gained TEs and BRD4-gained SEs, respectively.
Notably, although the BRD4 gain at senescence-activated TEs occurred significantly in both quiescent and senescent cells, albeit with different magnitudes (Fig. 3E), the gain of BRD4 at SEs was highly specific to the senescent cell state (Fig. 3F). Interestingly, BRD4-gained SEs were adjacent to key SASP genes with described regulatory and/or effector functions, including the cytokines IL1A, IL1B, IL8, and the matrix metalloproteases MMP1 and MMP10 (Fig. 3D and G; Supplementary Table S2B; refs. 24, 28). Consistently, such SEs were associated with genes related to the inflammatory response and immune system phenotype GO categories (Fig. 3H) and were selectively induced in senescent cells when compared with both proliferating and quiescent cells (Fig. 3I and J; FDR = 0.00). Given that BRD4 also cooperates with sequence-specific TFs for regulation of target genes, we performed de novo motif analysis on the BRD4-gained SE regions and identified several TFs such as NFκB, a bona fide SASP regulator (26), as well as FOS and ELK1 (Fig. 3K), known to be activated by RAS/MAPK signaling (50). Interestingly, motif analysis of all senescence-activated SEs also predicted sites for SMAD3, C/EBPβ, and GATA family of transcription factors (data not shown), previously linked to senescence (31, 51, 52). Altogether, these results demonstrate that OIS is accompanied by recruitment of BRD4 to a subset of senescence-activated SEs flanking key SASP genes, and suggest that the SASP program may be particularly sensitive to SE disruption by chromatin-targeting drugs.
The Bromodomain Protein BRD4 Is Critical for SASP Gene Expression
Having identified BRD4 enrichment at SE elements during OIS, we next sought to determine the impact of this factor on the transcriptional output of senescent cells. To this end, we inhibited BRD4 using a validated miR30-based shRNA (53) or the small molecule BET inhibitor JQ1 (42) and performed RNA-Seq in proliferating, quiescent, and senescent IMR90 cells. For comparison, we also examined transcriptional profiles of cells expressing an shRNA targeting the RELA/p65 subunit of NFκB, a well-characterized SASP regulator (26). As additional controls, we inactivated classic tumor suppressor genes, such as p53 and RB, known to direct the gene repression program leading to a permanent cell-cycle arrest (20), as well as the cell-cycle regulators p21 and p16 (CDKN1A and CDKN2A), which when inhibited in combination can bypass senescence in IMR90 cells (49). Collectively, we envisioned this panel of cell states would allow us to pinpoint direct and indirect regulators underlying distinct modules of the senescence transcriptional program.
Following RNA-Seq profiling, we first defined a senescence signature by identifying the top 100 induced and repressed genes in senescent cells compared with proliferating cells but not in quiescent cells (Supplementary Table S3A–S3C). This signature is similar to that produced by our laboratory using microarray technology (26, 49) and to those in previously published reports (54, 55). Hierarchical clustering analysis using this senescence signature revealed that BRD4 knockdown cells cluster away from the tumor suppressor gene knockdowns and closely together with p65 knockdown samples (Fig. 4A). This indicates that BRD4 controls a transcriptional module similar to that controlled by NFκB but distinct from the cell-cycle arrest program controlled by p53, RB, p16, and p21. Importantly, JQ1-treated senescent samples also clustered closely with BRD4 knockdown (Fig. 4A), supporting a bromodomain-dependent function of BRD4 in the regulation of gene expression programs during senescence, as previously described in other contexts (39, 40).
We next defined a BRD4 senescence signature by identifying senescence-induced genes that were significantly downregulated (2-fold cut-off; P < 0.005) upon BRD4 knockdown in senescent cells (Supplementary Table S3D). Principal component analysis (PCA) of this gene signature across the entire set of samples revealed a close clustering of BRD4- and p65-inhibited senescent samples with proliferating cells and away from the remaining senescence conditions, further supporting the distinct transcriptional output of BRD4 during OIS (Fig. 4B). GO analysis on the BRD4 senescence signature indicated a significant enrichment for terms related to immune system/inflammation, NFκB, and RAS (Fig. 4C), which included key SASP factors such as IL1A, IL1B, and IL8 (Supplementary Fig. S4A and S4B) and other genes herein found to gain distal SE activity and BRD4 binding during senescence (see Fig. 3I). In agreement, GSEA showed a significant enrichment of the BRD4 signature in genes associated with the BRD4-gained SEs (Fig. 4D; FDR = 0.00), supporting a mechanism of SE-mediated regulation of BRD4 targets during OIS. Despite sharing an overlapping transcriptional output (Fig. 4A and B), RNA-Seq profiling indicated that BRD4 regulates a larger set of genes than p65 during senescence (300 vs. 162; Fig. 4E; Supplementary Table S3D–S3G). Subsequent qRT-PCR validation using two independent BRD4 shRNAs and JQ1 similarly showed that not all BRD4-dependent SASP genes are NFκB targets, which include, among others, INHBA, BMP2, VEGFA, and VEGFC (Fig. 4F and G; Supplementary Fig. S4C and S4D). These data indicate that BRD4 is major SASP regulator during OIS.
To determine whether these phenomena were restricted to OIS in IMR90 cells, we assessed whether SASP expression also depends on BRD4 in different cell lines and/or by different senescence triggers. BJ human foreskin fibroblasts and WI-38 human diploid lung fibroblasts triggered to senescence by oncogenic HRASV12 also showed that the classic SASP factors IL1A, IL1B, IL6, IL8, CXCL1, and MMP10 were induced in both cell types, and that this induction was blunted in cells harboring BRD4 shRNAs or treated with JQ1 (Supplementary Fig. S4E and S4F). Furthermore, IMR90 cells triggered to senescence by treatment with the topoisomerase inhibitor etoposide (6, 7) displayed similar changes in the enhancer landscape as occurred during OIS (Fig. 4H; Supplementary Fig. S5A and S5B). However, the gain in H3K27Ac signal intensity adjacent to SASP genes was less than observed during OIS (Fig. 4H and Supplementary Fig. S5B), consistent with the lesser induction of SASP genes in this context (compare Fig. 4F and I, respectively). Still, genetic inactivation of BRD4 also blunted SASP gene expression in etoposide-treated cells (Fig. 4I and Supplementary Fig. S5C). Of note, in both etoposide-induced senescence and OIS, JQ1 treatment blunted SASP even after cells had established senescence (Fig. 4F, G, and I; Supplementary Figs. S4B–S4F and S5C), suggesting a key role for BRD4 in the maintenance of this senescence program. Altogether, these results suggest that senescence-induced enhancer remodeling contributes broadly to senescence biology and establish BRD4-bound SEs as a critical regulator of the SASP that can be disrupted pharmacologically.
BRD4 Is Needed for Acquisition of the Senescence-Associated Secretome
Given this unexpected association between BRD4 and senescence, we explored the impact of BRD4 inhibition on a variety of senescent phenotypes. BRD4 suppression in the IMR90 RAS model did not result in significant changes in tumor suppressor gene activation (Fig. 5A), or in other senescence markers, such as SA-β-gal activity (Fig. 5B), BrdUrd incorporation (Fig. 5C), or senescence-associated heterochromatin formation (SAHF; Fig. 5D). Colony formation at low cell density over 2 weeks was also not affected by BRD4 inhibition (Fig. 5E), further supporting that BRD4 inactivation does not bypass the senescence-associated cell-cycle arrest. These findings contrast with the phenotypes observed following suppression of RB and p53 pathway components on cellular senescence and are consistent with the distinct impact of BRD4 knockdown on the transcriptional profiles of senescent cells (see Fig. 4A). To examine the impact of BRD4 suppression on the ultimate readout of SASP, we next examined a subset of the senescent cell secretome using commercially available cytokine arrays. Using this approach, we detected a clear accumulation of secreted G-CSF, GM-CSF, CXCL1, IL6, IL8, and MIF in the conditioned media (CM) from senescent cells when compared with proliferating cells. CM from senescent cells upon BRD4 knockdown, however, showed a clear reduction in the levels of these cytokines, demonstrating that BRD4 inactivation affects the secretome of senescent cells (Fig. 5F and G). Collectively, these data indicate that BRD4 suppression selectively inhibits SASP but not other senescent phenotypes.
BRD4 Is Required for SASP-Mediated Paracrine Signaling Functions In Vitro
The SASP transcriptional module produces marked changes in the secretome, which sends signals that influence the fate and function of neighboring cells within the tissue microenvironment (2, 14). Having identified BRD4 as a factor linking remodeling of the enhancer landscape and transcription program leading to SASP, we next examined its ability to influence the behavior of other cell types in a series of in vitro assays. First, we examined the ability of BRD4 to affect paracrine senescence, a phenomenon whereby CM from senescent cells can transmit senescence signals to naïve cells (56, 57). Treatment of naïve proliferating IMR90 cells with CM from senescent IMR90 cells inhibited proliferation (Fig. 6A) and induced SA-β-gal activity (Fig. 6B and C), effects that were impaired when naïve cells were treated with CM from BRD4-inhibited senescent cells (Fig. 6A–C). Furthermore, the ability of CM from senescent cells to induce SASP factors such as IL1A, IL1B, IL6, and IL8 in proliferating cultures was also attenuated when using CM from senescent cells lacking BRD4 (Fig. 6D). Therefore, BRD4 is required for paracrine senescence, an effect that might amplify the impact of senescent cells in tissues.
We also examined the impact of BRD4 on the ability of senescent cells to modulate immune cell function and facilitate senescent cell targeting (26, 33). We previously showed that murine hepatic stellate cells undergoing senescence produce factors that could promote an antitumor M1 polarization of macrophages and recruit NK cells, leading to senescent cell elimination (9, 33). To study the impact of fibroblasts undergoing OIS on macrophage polarization, we generated human monocyte–derived macrophages and treated them with CM from IMR90 cells that were proliferating or triggered to senesce by oncogenic RAS. Strikingly, CM from senescent cells produced a substantial increase in the M1 polarization markers IL1A, IL1B, IL6, and CCR7 in human macrophages, an effect that was substantially reduced when CM was prepared from senescent cells expressing BRD4 shRNAs (Fig. 6E, left plots). Markers associated with M2 polarization were unaffected under all conditions (Fig. 6E, right plots). To study NK-cell–mediated cytotoxicity triggered by SASP, we cocultured the YT human NK cell line with GFP-labeled proliferating or senescent IMR90 cells and observed a significant reduction of viable senescent cells when compared with proliferating counterparts (9, 26). Importantly, this NK-mediated cytotoxicity was significantly reduced when NK cells were cocultured with BRD4-suppressed senescent cells (Fig. 6F–H; Supplementary Movies S1–S3). Therefore, BRD4 is required for key components of the immune surveillance mechanisms that trigger the clearance of senescent cells.
BRD4 Inhibition Impairs Immune Surveillance of Oncogenic NRAS-Expressing Senescent Cells In Vivo
The data presented above link SE remodeling and the emergence of SASP during senescence via BRD4, and imply that this factor ultimately contributes to the immune-modulatory and immune surveillance functions of the senescence program. To examine this in a physiologically relevant context, we took advantage of a powerful mouse model of OIS in vivo, where the induction of senescence can be acutely triggered and the fate of senescent cells can be monitored over time (36). In this model, sleeping beauty (SB) transposase and an SB transposon vector expressing oncogenic Nras are codelivered into the tail vein of mice via hydrodynamic injection, a procedure that results in stable integration of the transposon selectively into hepatocytes. Oncogenic NRAS rapidly triggers senescence and SASP, which in turn activate mechanisms of immune-mediated surveillance and clearance (36). We further adapted this model to coexpress an NrasG12D-IRES-GFP cassette and an shRNA targeting any gene in the same cells by codelivering two functionally linked transposon vectors: one in which NrasG12D is regulated from a tetracycline responsive promoter and the other constitutively expressing the reverse tet-transactivator protein (rtTA) together with a miR-30–based shRNA (Fig. 7A). We reasoned this system would enable tracing oncogene-expressing cells using the linked GFP reporter and assessment of both pharmacologic (systemic) and genetic (hepatocyte-specific) inhibition of BRD4 on senescent cell targeting and clearance in vivo.
We first assessed the impact of systemic BRD4 inhibition using the BET inhibitor I-BET762 (iBET), a compound analogous to JQ1 but with improved pharmacologic properties in vivo (41). Cohorts of mice that had been transduced with NrasG12D-IRES-GFP and a neutral shRNA (shRen) were treated daily with iBET or vehicle (Fig. 7B), and livers analyzed for GFP-expressing senescent cells and immune cell infiltration at 6 and 12 days after transduction. As shown in Fig. 7C (left plots) and D, both iBET- and vehicle-treated mice showed a similarly large fraction of GFP-positive cells at 6 days, indicating effective delivery of the transposon vectors. By day 12, livers isolated from vehicle-treated mice showed a dramatic reduction in GFP-positive hepatocytes (Fig. 7C, top middle plot, and D), consistent with their immune-mediated clearance (36), and clusters of immune cells in close proximity to the remaining GFP-positive foci (Fig. 7C, top right plot, and E, top plot). In stark contrast, livers from iBET-treated mice retained significantly more GFP-positive foci (Fig. 7C, bottom middle plot, and D). Importantly, although these foci were still SA-β-gal–positive (Supplementary Fig. S6A), they lacked the prominent immune cell infiltration observed in the livers of vehicle-treated animals (Fig. 7C, right plots, and E, and Supplementary Fig. S6B). Thus, pharmacologic inhibition of the BET protein family clearly impairs senescence immune surveillance in this model.
Besides their novel role in senescence surveillance as described above, BET proteins are crucial for the inflammatory response of various immune cells (41, 58). Because systemic iBET administration inhibits BET protein function in all cell types, it remains possible that the impaired senescence surveillance noted in iBET-treated animals could arise through compromised immune cell function. To address this issue, we took advantage of our genetic model to selectively suppress BRD4 in oncogene-expressing hepatocytes (see Fig. 7A). Specifically, we injected mice with shRNAs targeting a neutral gene (shRen) or Brd4 (shBrd4-1; shBrd4-2), treated the mice with doxycycline, and monitored GFP-positive foci and immune cell infiltration as above.
The results from these experiments decisively showed that genetic inhibition of BRD4 in oncogene-expressing hepatocytes is sufficient to impair their proper immune surveillance. Hence, livers from mice transduced with a control shRNA showed abundant clearance of GFP-positive cells between day 6 and day 12 after transfection, which was significantly impaired in livers from mice transduced with two independent Brd4 shRNAs (Fig. 7F and G). As was noted following pharmacologic BET inhibition, this impaired clearance was not due to a failure to activate the senescence program (Supplementary Fig. S6C) but instead was associated with reduced immune cell infiltration (Fig. 7H–J) and a blunted SASP response, as illustrated by the decreased levels of the chemokine MCP1 (CCL2; Fig. 7K), previously implicated in this model (36). All together, these data indicate that BRD4 plays a key role in the immune-mediated clearance of oncogene-expressing senescent cells in vivo, demonstrating its requirement for the full execution of the SASP-mediated tumor-suppressive program.
Cellular senescence occurs through a two-component process that leads to a stable cell-cycle arrest and the activation of an immune-modulatory program that collectively limits cell expansion. The senescence-induced cell-cycle program involves vast changes in the heterochromatin organization leading to transcriptional repression (20, 21), whereas the gene activation program can involve recruitment of key transcriptional activators (31, 40, 59). Here, we demonstrate that each component of the senescence program is dictated by distinct changes in the enhancer landscape'on the one hand, loss of TEs adjacent to the promoters of proliferation genes and, on the other, the acquisition of SEs flanking SASP genes, leading to a BRD4-controlled gene activation program. As a consequence of the latter program, BRD4 inhibition in premalignant senescent cells leads to a diminished SASP response and a compromised immune surveillance both in vitro and in vivo. These studies extend the role of BRD4-bound SEs in cancer beyond the establishment of tumor-promoting transcriptional circuits to a tumor-suppressive program that stimulates the clearance of premalignant senescent cells.
The Senescence-Associated Transcriptional Profile Is Shaped by Enhancer Remodeling
Cellular identity and plasticity rely on dynamic remodeling of enhancer landscapes (60). By comparing the active enhancer landscape during OIS to proliferative and quiescent conditions, we uncovered a global enhancer remodeling process that drives the gene expression program linked to senescence. Hence, the acquisition and loss of enhancer activity at particular loci correlated well with gene expression patterns previously implicated in senescence (46–48). Of note, although repression of E2F target genes was similarly observed in both quiescent and senescent cells, the TEs associated with these loci were exclusively inactivated during senescence. This result is particularly intriguing as it likely underlies the epigenetic mechanism of the stable cell-cycle arrest exhibited by senescent cells.
Most interestingly, SE activation resulted in induction of genes related to cytokine and growth factor activity, which include essential components of the SASP (24, 26) and also categorically distinguished senescent cells from growth-arrested quiescent counterparts. Interestingly, the SASP-associated SE loci that became activated during senescence displayed abundant H3K4Me1 marks in proliferating IMR90 cells, suggesting that they already preexist in inactive configurations (“primed”) in basal conditions and gain the activating H3K27Ac mark only upon RAS stimulation. This situation contrasts with that of macrophage activation in which a subset of enhancers are newly configured (“latent”) following a proinflammatory stimulus (61). Nevertheless, given the prominence of SEs in controlling cell identity and fate (37, 38), these observations underscore the importance of SASP as a key output of the senescence program.
SE-Associated BRD4 Is Required for the SASP
SEs frequently associate with transcriptional coregulators such as BET proteins to regulate gene expression (37, 38), and as such, SE-regulated genes are often highly sensitive to perturbation using BET inhibitors (39, 40). Indeed, using genome-wide chromatin profiling, we observed that BRD4 is recruited to a subset of SEs associated with SASP genes during senescence, and that genetic or pharmacologic inhibition of BRD4 collapses the senescence-induced SASP transcriptional program. Interestingly, although only a subset of SASP-associated enhancer loci gained substantial H3K27Ac mark and BRD4 binding during OIS, our RNA-Seq analysis indicated a high number of SASP genes that were repressed by BRD4 inactivation. In principle, this observation may reflect the reduced sensitivity of the ChIP-Seq assay in senescent cells. Alternatively, senescence-activated SEs may selectively trigger the expression of a specific class of SASP genes with key regulatory functions. For example, IL1A and IL1B, which are induced more than 1,000-fold through BRD4-gained SEs in senescence, are known SASP “amplifiers” (27, 28, 52, 57, 62) and, as such, could induce additional SASP factors through positive feedback loops without the need for direct BRD4 regulation. Consistent with this view, control of such amplifier genes may underlie the action of GATA4 on SASP during replicative senescence (31). Regardless, our findings establish BRD4-gained SEs as key players in the regulation of cellular senescence, unraveling novel regulatory nodes in senescent cells that can be targeted pharmacologically.
Besides binding acetylated histones that display an increase at senescence-activated SEs, the mechanism whereby BRD4 leads to transcriptional activation of SASP genes likely involves recruitment of specific TFs as described in other contexts (63–65). One of the TFs with consensus binding sites that were enriched in regions of senescence-activated SEs is NFκB, which is reminiscent of observations during the inflammatory response of macrophages and endothelial cells (40, 66). Importantly, however, our results clearly show that BRD4 can control non-NFκB targets involved in SASP, suggesting it has nonredundant functions with NFκB during senescence. It is likely that multiple TFs orchestrate the impact of BRD4-gained SEs on SASP gene transcription including FOS, ELK1, PRDM19, and others besides NFκB as indicated by our motif analysis. Moreover, analyses of all H3K27Ac-marked SEs identified additional TFs such as SMAD3, C/EBPβ, or the GATA family, previously linked to senescence (31, 51, 52) as well as to liver fibrosis (67), pointing to both specific and common regulatory mechanisms driving enhancer-mediated secretory pathway activation in different cellular contexts.
Our results also provide new insights into the diametrically opposed epigenetic mechanisms coordinating different components of the senescence program. Previous studies documented the importance of heterochromatin reorganization together with the recruitment of histone-modifying enzymes for the production of a repressive chromatin environment on the promoters of cell-cycle genes (20, 21). As shown here, these genes are often governed by TEs in proliferating cells that are inactivated during OIS and are not dependent on BRD4. By stark contrast, the gene activation program associated with SASP is ultimately directed by a redistribution of BRD4 within senescence-activated SEs associated with key SASP regulators. Importantly, as a consequence of the exquisite sensitivity of SE-regulated transcription to BET inhibition, genetic and pharmacologic inhibition of BRD4 collapses the SASP signature and diminishes SASP-mediated functions in vitro and in vivo while having no impact on the senescence-associated cell-cycle arrest. Thus, these results indicate that BRD4 inactivation uncouples the SASP from the cell-cycle arrest during senescence, and imply that BET inhibitors may be valuable at modulating SASP activity in vivo, such as has been recently reported for the mTOR inhibitor rapamycin (28, 29).
BRD4 Is Required for Paracrine Senescence and Immune Surveillance
BET protein–associated SEs are important in establishing biological circuits in cancer and other processes, such as ESC self-renewal, cell lineage specification, and inflammation (37–40). In contrast, our study links SE remodeling to senescence immune surveillance and, in doing so, reveals an unexpected role for BRD4 in the tumor-suppressive senescence program. Specifically, we show that BRD4, although dispensable for the senescence-associated cell-cycle arrest, is critical for paracrine senescence transmission, macrophage M1 polarization, and senescent-cell targeting by NK cells. More importantly, using a mouse model of senescence immune surveillance in the liver, we demonstrate that BRD4 inhibition in vivo disrupts SASP gene expression and the efficient clearance of premalignant senescent hepatocytes. Collectively, these results suggest that BRD4 is a master regulator that controls senescent cell communication with the microenvironment.
Our data linking BRD4 to a tumor-suppressive program stands in apparent contrast to the established role of BRD4 in tumors as a cancer maintenance gene (39, 53, 68). Still, consistent with a potential role for BRD4 in other tumor-suppressive programs, BRD4 inhibition can enhance oncogenic dedifferentiation in human breast cancer cells and cells from patients with premature aging syndrome (69, 70), and promotes hyperproliferation in the murine epidermis (71). Moreover, although BRD4 had no marked impact on p53 transcriptional profiles in our system, phosphorylated BRD4 can act as a transcriptional coactivator for p53 in certain cell types (72). In principle, the impaired SASP response produced by sustained BRD4 inhibition might enable rare oncogene-expressing cells to eventually evade senescence and progress to cancer. Consistent with this possibility, we observed an aggressive hepatocellular carcinoma in one of ten mice cotransduced with NrasG12D and Brd4 shRNAs 1 year following liver transduction, whereas no tumors were observed in a similar number of mice transduced with NrasG12D and a control shRNA (data not shown). In any event, our data support the view that BRD4 function is context-specific, and, as such, the impact of BET inhibition may have markedly different consequences depending on cell lineage or genotype. Thus, although BET inhibitors may have anticancer effects owing to their ability to block aberrant cell renewal, they may simultaneously disrupt immune cell function or immune surveillance, and, as such, their clinical impact will be determined by the balance of these effects.
Our results establish a key mechanism underlying the regulation of SASP gene expression that is responsible for the communication between cells undergoing OIS and the tissue microenvironment. Moreover, they begin to unravel the molecular framework of an immune surveillance network that, though understudied, might eventually be harnessed to engage the innate immune system for cancer therapy in a manner that complements current efforts to reawaken the adaptive immune system in cancer patients. Besides its role in cancer, the SASP response also contributes to tissue repair (10, 13), fibrosis (9, 66), and aging (73), and, in principle, BET inhibitors could provide strategies to modulate the SASP response and these processes at the organismal level.
Cell Culture and Retroviral Infection
Human diploid IMR90, WI38 lung fibroblasts, and human BJ foreskin fibroblasts were obtained from the American Type Culture Collection (ATCC), where they had been authenticated by short tandem repeat profiling, and were cultured in DMEM supplemented with 10% FBS and penicillin and streptomycin (GIBCO). Frozen stocks generated within 2 to 3 passages from date of purchase were used throughout the study. Human macrophages were generated from monocytes isolated from buffy coats from healthy human blood donors (New York Blood Center) and differentiated in Teflon-coated cell culture bags in the presence of human CSF1 (R&D Systems), as previously described (74). Resulting human monocyte–derived macrophages were cultured in RPMI-1640 with 2% human serum (Sigma) and penicillin and streptomycin (GIBCO). NK cells (from DSMZ; ref. 26) were cultured in RPMI-1640 with 10% FBS. For retroviral-mediated gene transduction, retroviruses were packed using Phoenix cells (G. Nolan, Stanford University, CA), and infections were performed as described elsewhere (20). The infected population was selected using 2 μg/mL puromycin (Sigma) and 75 μg/mL hygromycin B (Roche). For JQ1 treatments, cells were treated with 100 nmol/L of JQ1 or DMSO as vehicle for 48 hours prior to harvest. A working solution of JQ1 at 1 mmol/L was prepared to use a final concentration of vehicle (DMSO) no higher than 1/10,000.
Induction of Senescence and Quiescence
Senescence was induced in human cells by treatment with etoposide (100 μmol/L; Sigma) for 2 days, or by retroviral-mediated expression of HRASV12, as previously described (20). Vehicle-treated cells or cells transduced with empty vector served as proliferating counterpart controls, respectively. When indicated, cells were cotransduced with vectors encoding validated GFP-linked shRNAs targeting BRD4 (53), p65 (26), p53 (33), p53/RB (49), or p21/p16 (49). Infected cells were harvested 12 days after infection. Quiescence was induced by culturing IMR90 cells in 0.1% FBS for 4 days.
Senescence In Vitro Assays
BrdUrd incorporation, colony formation, and SA-β-gal assays in cells were performed as previously described (20, 26). Briefly, senescent IMR90 cells (postselection day 8) were plated on coverslips coated with 0.1% gelatin and fixed in 4% formaldehyde. For cell-cycle arrest analyses, cells were labeled with BrdUrd (100 μg/mL; Sigma) for 6 hours, and nuclei incorporating BrdUrd were visualized by immunolabeling with anti-BrdUrd antibodies (1:400; BD Pharmingen) as described previously (20). For colony formation assays, 1,250, 2,500, or 5,000 cells were plated in each well of a 6-well plate. Cells were cultured for 14 days, then fixed with 4% formaldehyde, and stained with crystal violet. Detection of SA-β-gal activity in mouse tissues was performed as described (26). For macrophage polarization experiments, 48-hour CM (containing FBS) were collected from proliferating (IMR90-vector) or senescent (IMR90-HRASV12) cells expressing the indicated shRNAs, and seeded at day 6 after pWZL-Hygro/shRNA transduction to have equal final cell numbers. CM were clarified by centrifugation, filtered through a 0.45 μm filter, diluted 1:3 with human monocyte–derived macrophage media, and added to 500,000 human monocyte–derived macrophages seeded in 6-well plates 24 hours earlier. Human macrophages were harvested following an 18-hour incubation with the indicated CM and processed for qRT-PCR analysis of previously described M1 or M2 polarization markers (75, 76). NK-cell cytotoxicity experiments were performed as described elsewhere (26, 33) with the following variations: NK cells were first stained with the CellTracker Red CMTPX Dye (Thermo Fisher Scientific) according to the manufacturer's instructions; briefly, cell staining was performed in RPMI supplemented with 10% FBS for 30 minutes using a final concentration of 1 μmol/L of CellTracker Red CMTPX Dye, and then washed twice with PBS. NK cells were then seeded into 96-well plates containing proliferating or senescent IMR90 cells expressing the indicated GFP-linked shRNAs. Cocultures were imaged over time using an INCell 6000 high-content imager (GE Healthcare Life Sciences), with the 488 nm and 561 nm laser excitation, using a 10× objective. The correct focal plane was maintained by the use of the INCell laser–based focus system at each time point, which maintains a focal plane relative to the bottom of the well. Images were captured every 30 minutes in a temperature-controlled incubation chamber, starting 30 minutes after NK-cell seeding onto IMR90 cultures. Cells were maintained in CO2-independent media (Life Technologies). Images for each channel were saved during the experiment and subsequently analyzed using GE Developer image analysis software (GE Healthcare Life Sciences). GFP-labeled IMR90 cells were identified and segmented from background using an intensity-based threshold method. Touching cells and clumps of cells were segmented using watershed segmentation. NK cells labeled with the CellTracker Red CMTPX Dye were identified using the same threshold and segmentation method as the IMR90 cells. The average area of the GFP-labeled IMR90 cells was used as a measure of cell health and viability. The association of NK cells with IMR90 cells was quantified by measuring the fraction of the IMR90 cell area per well, overlapped with the total NK-cell area per well. This value was normalized to the total area of GFP-positive IMR90 cells per well. Paracrine senescence assays were performed as previously described (57). In brief, CM were harvested from 2 × 106 (in a 10-cm dish) IMR90-vector proliferating cells or IMR90-HRASV12 senescent cells plated at day 10 after infection and incubated in DMEM 0.5% FBS for 72 hours. CM were clarified by centrifugation, filtered through a 0.45-μm filter, and mixed with DMEM 40% FBS in a proportion of 3 to 1 to generate CM containing 10% FBS. Cells were fixed for SA-β-gal staining after 7 days and collected for RNA extraction 5 days after incubation; fresh media were added on top on day 3. All experiments were performed with biological duplicates and were repeated at least twice.
ChIP-Seq and Bioinformatics Analysis
ChIP-Seq was performed as previously described (65). H3K27Ac and H3K4Me1 antibodies were purchased from Abcam (ab4729 and ab8895, respectively). The BRD4 antibody used for ChIP-Seq was purchased from Bethyl (A301-985A100). A total of 0.5 to 1.0 μg of each antibody per 0.5–1.0 million cells was used. ChIP-Seq libraries were constructed using the TruSeq ChIP Sample Prep Kit (Illumina) and sequenced using an Illumina HiSeq 2000 platform. Raw reads were mapped to the reference human genome assembly GRCh37 (hg19) using Bowtie and SAMtools. The Genomic Regions Enrichment of Annotations Tool (GREAT) was used to predict GO terms of ChIP-Seq–enriched regions (77). GSEA was performed according to the instructions. Motif enrichment was identified with a 1,000-bp window centered on 37 BRD4-associated enhancers using HOMER suite. Further details can be found in the Supplementary Methods. The GEO accession number for the raw and processed ChIP-Seq data reported in this article is GSE74238.
Identification of Senescence-Activated H3K27Ac-Enriched Enhancers
Peaks from all conditioned H3K27Ac ChIP-Seq samples were combined first. After stitching regions with a maximum distance of 12.5 Kb, 31,731 H3K27Ac TEs and 1,255 H3K27Ac SEs were identified (total union enhancers are 32,986). If the log2 fold changes of senescent versus proliferating were greater or less than two, they were assigned as senescence-activated or senescence-inactivated enhancers, respectively. Using this approach, 6,555 senescence-activated and 7,095 senescence-inactivated TEs were defined from 31,731 TEs. Similarly, 198 senescence-activated and 191 senescence-inactivated SEs were defined from 1,255 SEs.
Identification of BRD4-Gained Senescence-Activated SEs
BRD4 tag counts were calculated from 4,440 and 198 senescence-activated TEs or SEs, respectively. BRD4-gained enhancers were defined if log2 fold changes of BRD4 signals of senescence condition versus proliferating condition were greater than 0.5. With this criterion, 1,307 and 37 BRD4-gained TEs and SEs were identified, respectively.
Total RNA was isolated using the RNeasy Mini Kit (Qiagen), and cDNA was obtained using the TaqMan reverse transcription reagents (Applied Biosystems). qRT-PCR data analyses were performed as previously described (26). Details of qRT-PCR, RNA-Seq, cytokine arrays, protein immunoblotting in cultured cells, and histologic analyses in tissue sections are also provided in Supplementary Methods. The GEO accession number for the raw and processed RNA-Seq data reported in this article is GSE74324.
In Vivo Senescence Surveillance Assays upon Systemic or Hepatocyte-Specific BRD4 Inhibition
For induction of OIS in vivo, SB transposase vector and two SB transposon vectors expressing NrasG12D-IRES-GFP and a miR-30 shRNA (against Renilla or BRD4; see Supplementary Methods for additional details) were codelivered into C57BL/6J mice (purchased from Harlan-Teklad) via hydrodynamic injection. Mice were injected with a sterile 0.9% NaCl solution of 12.5 μg DNA of each transposon vector together with CMV-SB13 Transposase (1:5 ratio) via the lateral tail vein. Doxycycline was administered to mice via 625 mg/kg doxycycline-containing food pellets (Harlan Teklad) starting 24 hours before injections. For pharmacologic inhibition of BRD4, I-BET-762 (Selleck Chemicals) was reconstituted in DMSO and further diluted in PBS. Mice were treated with I-BET-762 or vehicle orally every day (15 mg/kg). Mice were sacrificed at the indicated time points and chunks of liver tissue either fixed in 10% neutral buffered formalin or fresh-frozen in optimal cutting temperature compound for histologic analyses performed as detailed in Supplementary Methods. All experiments with mice were performed in accordance with protocols approved by the Memorial Sloan Kettering Cancer Center's Internal Animal Care and Use Committee and Federal Office of Laboratory Animal Welfare guidelines.
Statistical tests performed to assess the significance of the Chip-Seq, RNA-Seq, qRT-PCR, histologic, and functional data indicated above are described in the Supplementary Methods.
Disclosure of Potential Conflicts of Interest
C.R. Vakoc has ownership interest in a patent related to BET inhibitor use in leukemia and is a consultant/advisory board member for Syros. No potential conflicts of interest were disclosed by the other authors.
Conception and design: N. Tasdemir, A. Banito, D. Alonso-Curbelo, J.E. Bolden, A. Chicas, S.W. Lowe
Development of methodology: N. Tasdemir, D.F. Tschaharganeh, M. Fennell, A. Chicas
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Tasdemir, A. Banito, J.-S. Roe, D. Alonso-Curbelo, M. Camiolo, D.F. Tschaharganeh, O. Aksoy, J.E. Bolden, M. Fennell
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Tasdemir, A. Banito, J.-S. Roe, D. Alonso-Curbelo, M. Camiolo, J.E. Bolden, C.-C. Chen, M. Fennell, V. Thapar
Writing, review, and/or revision of the manuscript: N. Tasdemir, A. Banito, J.-S. Roe, D. Alonso-Curbelo, S.W. Lowe
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N. Tasdemir, A. Banito, J.-S. Roe, D. Alonso-Curbelo, C.-H. Huang, O. Aksoy
Study supervision: C.R. Vakoc, S.W. Lowe
Other (helped with some of the experiments): O. Aksoy
This study was supported by funds from an R01 grant (AG16379) from the NIH (to S.W. Lowe), CA013106 (NIH/NCI), and NIH/NCI Cancer Center Support Grant P30 CA008748. N. Tasdemir was supported by the Lindsay and Goldberg Fellowship from the Watson School of Biological Sciences. A. Banito was supported by an EMBO long-term fellowship. D. Alonso-Curbelo is recipient of a postdoctoral fellowship from the Spanish Fundación Ramón Areces. J.-S. Roe is supported by the Martin Sass Foundation and the Lauri Strauss Leukemia Foundation. S.W. Lowe is an investigator at the Howard Hughes Medical Institute and is the Geoffrey Beene Chair of Cancer Biology.
The authors thank Sha Tian, John P. Morris IV, Leila Akkari, Florian Klemm, Johanna Joyce, Valery Krizhanovsky, Lars Zender, and Amaia Lujambio for expert technical assistance and advice; Jonathan Brown, Charles Lin, and Jay Bradner for key insights and technical advice; and members of the Lowe laboratory and Charles J. Sherr for helpful comments.
Note: Supplementary data for this article are available at Cancer Discovery Online (http://cancerdiscovery.aacrjournals.org/).
- Received February 19, 2016.
- Revision received April 14, 2016.
- Accepted April 14, 2016.
- ©2016 American Association for Cancer Research.