PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Cancer Cell. Author manuscript; available in PMC 2010 April 6.
Published in final edited form as:
PMCID: PMC2829755
NIHMSID: NIHMS152196

Functional Identification of Tumor Suppressor Genes Through an in vivo RNA Interference Screen in a Mouse Lymphoma Model

SUMMARY

Short hairpin RNAs (shRNAs) capable of stably suppressing gene function by RNA interference (RNAi) can mimic tumor suppressor gene loss in mice. By selecting for shRNAs capable of accelerating lymphomagenesis in a well-characterized mouse lymphoma model, we identified over ten candidate tumor suppressors, including Sfrp1, Numb, Mek1, and Angiopoietin 2. Several components of the DNA damage response machinery were also identified, including Rad17, which acts as a haploinsufficient tumor suppressor that responds to oncogenic stress and whose loss is associated with poor prognosis in human patients. Our results emphasize the utility of in vivo RNAi screens, identify and validate a diverse set of tumor suppressors, and have therapeutic implications.

INTRODUCTION

Tumor suppressors act in signaling networks that restrict cellular proliferation and present barriers to malignant transformation. For example, the p53 tumor suppressor gene encodes a transcription factor that can limit proliferation by promoting cell cycle arrest, senescence, or apoptosis (Riley et al., 2008). p53 is activated to inhibit proliferation in response to stress, including replicative stress produced by mitogenic oncogenes, and thus acts as part of a failsafe mechanism that halts the expansion of aberrantly proliferating cells. Many other tumor suppressors have been identified, each pointing towards programs or pathways that naturally limit tumor growth. Although generally not considered direct drug targets, their loss-of-function can create cellular dependencies that can be exploited therapeutically (Vassilev et al., 2004).

RNA interference facilitates loss-of-function genetics in mammalian cells and has been used to explore various aspects of cancer biology, including the function of tumor suppressor genes. Moreover, the availability of genome-wide libraries of shRNAs capable of stably repressing gene expression has enabled genetic screens for determinants of oncogenic transformation as well as potential therapeutic targets (Berns et al., 2007; Westbrook et al., 2005). To study cancer phenotypes not readily modeled in vitro, we have adapted RNAi technology to suppress tumor suppressor gene function in mice and have used this technology to study aspects of tumorigenesis, tumor maintenance, and treatment response (Hemann et al., 2003).

The Eμ-Myc lymphoma model expresses the c-myc oncogene in B cells (Adams et al., 1985) and has been used extensively for identifying lesions that promote tumorigenesis, either through retroviral-based insertional mutagenesis, by intercrossing with various transgenic or knockout mice or, more rapidly, by engrafting Eμ-Myc derived hematopoietic stem and progenitor cells (HSPCs) transduced with a gene or shRNA into syngeneic recipient mice (Schmitt and Lowe, 2002). Using the latter approach, we have shown that shRNAs targeting p53 or certain pro-apoptotic genes can mimic the corresponding gene deletion by promoting tumorigenesis (Hemann et al., 2003; Hemann et al., 2004). We therefore reasoned that it should be possible to introduce complex pools of shRNAs into Eμ-Myc progenitors, allowing for the selection of those capable of promoting tumorigenesis in transplanted recipients.

RESULTS

A p53 shRNA can be recovered from low complexity pools at high efficiency

To identify appropriate conditions for an in vivo RNAi screen using the Eμ-Myc model, we initially determined the complexity of shRNAs that could be effectively screened in our HSPC transduction/reconstitution assay using a shRNA targeting p53 as a positive control. All shRNAs were based on the miR30 design, where sequences homologous to the targeted gene are inserted into a natural microRNA structure and thus are efficiently incorporated into the RNAi pathway and capable of potent knockdown when integrated at single copy in the genome (Dickins et al., 2005; Silva et al., 2005). The p53 shRNA (p53.1224) was cloned into a retroviral vector that co-expressed green fluorescent protein (GFP), thereby enabling cells expressing shRNAs to be tracked by flow cytometry or whole body fluorescence imaging (Figure 1A).

Figure 1
Strategy for an efficient in vivo RNAi screen in the Eμ-Myc lymphoma model

As expected, Eμ-Myc HSPCs transduced with undiluted p53.1224 typically produced tumors in recipient mice by 10 weeks, albeit with incomplete penetrance (Figure 1B). Similarly, Eμ-Myc HSPCs transduced with 1:10 – 1:100 dilutions of p53.1224 in empty vector or in a vector containing a control shRNA produced tumors with a similar penetrance and slightly longer latency. Tumors promoted by pools containing p53.1224 dilutions were invariably GFP positive (although only a subset of the transplanted cells were infected) and sequence analysis indicated that there was a strong selection for the p53 shRNA relative to control vector. The tumors also displayed the B220+, IgM immunophenotype and histopathology reminiscent of lymphomas arising in Eμ-Myc transgenic animals (Figure 1C, D; data not shown), implying Myc was required for tumorigenesis. Eμ-Myc HSPCs transduced with the empty retroviral vector or shRNAs targeting the human CCND1 or CDK5 genes did not produce tumors, and the few tumors that eventually arose from negative controls were GFP negative (data not shown). Therefore, tumor acceleration did not result from insertional mutagenesis or a general perturbation of the RNAi machinery but required a biologically active shRNA that could be enriched from diluted pools.

The background frequency of Eμ-Myc lymphomas was less than is observed for germline Eμ-Myc transgenic mice, which typically develop lymphomas at complete penetrance between 15 and 30 weeks (Adams et al., 1985). This reduced penetrance likely reflects the inability of Eμ-Myc HSPCs to provide long-term reconstitution of the hematopoietic system of recipient mice under the sublethal irradiation conditions used in our transplantation experiments. Indeed, we see that virally-transduced cells eventually were depleted from the peripheral blood of recipient mice within about 20 weeks (data not shown). We reasoned that the low background of tumors arising in negative controls would facilitate the identification of tumor-accelerating shRNAs by creating a defined window in which a particular shRNA could trigger malignant transformation. Based on these pilot experiments we designated 20 weeks post-transplantation as the end point for our screen.

In vivo screening identifies candidate shRNAs capable of promoting lymphomagenesis

Although the progenitor cell transplantation procedure described above is scalable, it is not amenable to genome-wide shRNA screening, which prompted us to seek out strategies to filter the larger shRNA library into categories enriched for cancer relevant genes. To start, we decided to survey shRNAs targeting the “cancer 1000” set of genes containing putative cancer-related genes compiled from microarray expression data and literature mining (Witt et al., 2006). The list contained potential tumor suppressors, as well as oncogenes that were not predicted to have an impact in our model and, in principle, would serve as negative controls.

Approximately 2300 shRNAs targeting the mouse orthologs of the cancer 1000 list were obtained from the CODEX RNAi library (http://katahdin.cshl.edu:9331/RNAi_web/scripts/main2.pl) and transferred into the LMS vector in pools of 48. DNA sequencing of a subset of pools confirmed that an appropriate representation of shRNAs was retained. While p53 shRNAs promoted tumorigenesis at a 1:100 dilution (Figure 1B), we reasoned that a large pool size in the range of 100 or greater would increase the probability that high potency shRNAs would outcompete weaker shRNAs, and therefore chose a smaller pool size of 48. Each pool was introduced into at least three independent progenitor populations and transplanted into irradiated recipient mice, which were subsequently monitored for lymphoma formation by lymph node palpation and fluorescence imaging. Positive (p53.1224) and negative controls (either empty vector or a control shRNA) were included to rule out variations between HSPC populations and behaved as expected from our pilot studies (Supplementary Figure 1A).

Of the 48 pools tested, 27 produced GFP-positive lymphomas in one, two or three recipient mice (Figure 2A, B), suggesting they contained one or more tumor-promoting shRNAs. To identify these shRNAs, we isolated genomic DNA from lymphomas, amplified the integrated shRNAs using PCR, and sequenced the amplified products. Most tumors showed enrichment of a subset of the transduced shRNAs, typically containing 1–3 shRNAs per tumor. As expected, the positive control hairpin, p53.1224 in pool 9AD, was among the shRNAs that were enriched (Supplementary Table 1). An example of a scoring shRNA, is one that targets Angiopoietin2 (Ang2.2112), which was identified in tumors obtained from three mice, and comprised more than 80% of sequence reads in one lymphoma (Figure 2C).

Figure 2
shRNAs cooperate with Myc during tumorigenesis

In vivo validation of candidate tumor suppressors using multiple shRNAs

From the analysis of the entire library, we identified >80 different shRNAs that were present in tumors (Supplementary Table 1). Since this exceeded the number of genes we could feasibly validate in vivo, we reasoned that the shRNAs that featured the highest enrichment in tumors from the primary screen would be most likely to validate when tested as individual shRNAs. We decided to retest the first 15 of those identified that were highly enriched (>80% of sequence reads from a single tumor), as well as shRNAs that were identified as enriched in more than one independent tumor. These shRNAs were reintroduced as individual clones into Eμ-Myc progenitor cells and assessed for their ability to promote tumorigenesis in transplanted recipients. Ten out of fifteen shRNAs analyzed showed accelerated tumor onset in at least a subset of animals (Supplementary Table 2).

We also selected five shRNAs that were present in tumors but were not highly enriched by the criteria described above. Of these, only two promoted tumorigenesis when introduced as individual shRNAs. One targeted the ataxia telangiectasia mutated (ATM) gene, whose loss is known to accelerate tumorigenesis in Eμ-Myc transgenic mice (Pusapati et al., 2006). The other targeted Rad51C, which has been implicated in homologous recombination and the DNA damage response (Sharan and Kuznetsov, 2007). By contrast, shRNAs targeting Max, Edg5 and Fgf20 did not validate upon retesting, suggesting that either not all shRNAs in the lymphomas were tumor promoting, or that a cooperating shRNA from the pool was required.

We also retested six shRNAs from the library that were not observed in tumors or fell far below our cutoff for follow up validation (Supplementary Table 2). As expected, none of these shRNAs validated in vivo. Therefore, while our screen probably did not uncover all tumor-promoting shRNAs contained in our library, many of the enriched shRNAs target genes that have properties of tumor suppressors. Still, that each shRNA (including our positive control) produced tumors with incomplete penetrance suggested that the combination of Myc and each shRNA was not sufficient for lymphomagenesis, but that additional lesions were required. While insertional mutagenesis might supply some cooperative events, it is probably not sufficient since: (i) our negative controls never showed tumor acceleration within the given time period (Figure 2A, see also Figure 3A) and (ii) many pools and shRNAs tested in an identical manner were not tumor promoting.

Figure 3
Validation of tumor suppressor gene activity in vivo

Genes silenced by five of the validated shRNAs were chosen for further investigation because they targeted potentially important pathways in cancer development. Specifically, Mitogen Activated Protein Kinase Kinase 1 (MEK1) is a component of the MAPK pathway; Angiopoietin2 (Ang2) is a regulator of angiogenesis; Secreted Frizzled-Related Protein 1 (Sfrp1) and Numb are negative regulators of the Wnt and Notch pathways, respectively; and Rad17 is involved in the DNA damage response. Tumors that arose from shRNAs against these genes were confirmed by RT-QPCR to have reduced expression of the targeted gene (Supplementary Figure 1B).

Several independent shRNAs targeting each of these genes were generated and introduced into our HSPC transplantation assay (Figure 3A). In all cases examined, multiple shRNAs targeting each gene accelerated tumorigenesis in at least a subset of recipient mice, thereby ruling out off-target effects of individual shRNAs. Immunophenotyping revealed that all lymphomas were of pre-B cell origin (B220+, IgM), suggesting that most tumor-promoting shRNA acted directly on tumorigenesis rather than modulating the cell of origin of the disease (Supplementary Figure 3). Interestingly, when introduced into Myc-overexpressing ARF null lymphomas in vitro, shRNAs targeting p53, Rad17, and to a lesser extent Mek1 enhanced proliferation in a competition assay (Figure 3B). In contrast, those targeting Ang2, Sfrp1, and Numb did not confer any competitive advantage, implying that the tumor-promoting effect of repressing these genes depends on the in vivo microenvironment or pathways spontaneously altered during lymphomagenesis, underscoring the value of an in vivo screen.

Mek1 can have tumor suppressive properties

In validating our findings, we were surprised that some of the genes we identified as tumor suppressors have pro-oncogenic properties in other contexts. Thus, while angiopoietin 2 was identified as an anti-angiogenic protein (Maisonpierre et al., 1997), it can also be pro-angiogenic in vivo (Lobov et al., 2002). Likewise, Mek1 can transmit oncogenic signals downstream of ras (de Vries-Smits et al., 1992), but is also required for the transmission of checkpoint signals in response to both oncogenic and genotoxic stress (Lin et al., 1998; Zhu et al., 1998; Yan et al., 2007). As expected, tumors triggered by Mek1 shRNAs displayed reduced Mek1 expression, corresponding to a lower level of phospho-ERK1/2, two downstream targets (Figure 4A). Interestingly, acute activation of Myc triggered the phosphorylation of the Mek targets Erk1/2 in a Mek-dependent manner (Figure 4B), and treatment of cells with a Mek1 inhibitor attenuated Myc-induced cleavage of the apoptosis effector PARP, as well as activation of p53 and the DNA damage response proteins RAD17 and γH2AX (see below) (Figure 4C,D). Furthermore, primary B-cells co-expressing Myc and a Mek1 shRNA were selected for in an in vitro competition assay, whereas cells expressing the Mek1 shRNA alone depleted over time (Figure 4E-F). Together, these data imply that Mek1 is a context dependent tumor suppressor whose anti-proliferative action is revealed in Myc-expressing cells.

Figure 4
Mek1 can have tumor suppressive properties

RAD17 is activated by Myc

The DNA damage response (DDR) promotes checkpoint activation following DNA damage, including that produced by exogenous DNA damaging agents or following replication stress (Halazonetis et al., 2008). Consistent with a crucial role of these checkpoints in limiting malignant transformation, a substantial number of our validated shRNAs target DDR genes. Among these were Rad17, ATM, Rad51C, and Prkdc (DNA PK catalytic subunit) – all of which were validated as tumor suppressors in vivo (Figure 3A). Previous studies suggest that Atm and Prkdc can have tumor suppressive effects in the hematopoietic system, and loss of rad51C can promote tumorigenesis in mice (Xu et al., 1996; Jhappan et al., 1997; Kuznetsov et al., 2009). However, despite its central role in the DNA damage response, the contribution of Rad17 to cancer development has not been examined.

Rad17 acts as part of a complex that assembles the DNA-damage repair sliding clamp onto DNA at sites of damage and, in fission yeast, is required for both the DNA damage and the DNA replication cell cycle checkpoints (Parker et al., 1998). Rad17 activity is positively regulated by ATR through phosphorylation at serines 635/645 and facilitates phosphorylation of Chk1 by ATR in response to replication stress and DNA damage to maintain genomic stability (Wang et al., 2006; Wang et al. 2003; Bao et al., 2001). Four different shRNAs targeting Rad17 consistently promoted lymphomagenesis, albeit with different latencies and variable penetrance (Figure 3A). As expected, the resulting tumors showed knockdown of Rad17 (Figure 5A).

Figure 5
The DNA damage and replication checkpoint protein Rad17 is phosphorylated after Myc induction, and shRNA-mediated knockdown of Rad17 attenuates effects of Myc-induced stress responses

While DDR pathways may indirectly limit tumorigenesis by maintaining genomic stability, they may act more directly by mediating anti-proliferative responses to cellular stress. Indeed, ectopic activation of the myc oncogene, which serves as the primary genetic lesion in our screen, triggers a replication stress-induced DNA damage response leading to delayed S phase progression and/or apoptosis (Karlsson et al., 2003; Dominguez-Sola et al., 2007). Accordingly, we found that Eμ-Myc lymphomas expressed much more phosphorylated, hence activated, Rad17 than control lymphocytes in vivo (Figure 5B). Furthermore, enforced Myc expression rapidly triggered Rad17 and Chk1 phosphorylation in a manner that paralleled its ability to induce ARF and p53 (Figure 5C top, D and E; Supplementary Figure 4). Of note, oncogenic Ras expression also triggered RAD17 phosphorylation in normal diploid fibroblasts (Figure 5C, bottom), indicating that other mitogenic oncogenes can activate RAD17.

Interestingly, the Rad17 shRNAs that were most effective at promoting lymphomagenesis (Rad17.1169, Rad17.232) also most prominently attenuated anti-proliferative responses to Myc. Hence, cells co-expressing Myc and Rad17.1169 displayed reduced phospho-p53 and p19ARF levels (Figure 5E), suggesting attenuation of these oncogene-induced failsafe programs. More importantly, these Rad17 shRNAs enhanced colony formation in Myc-expressing MEFs, which also displayed reduced apoptosis and enhanced proliferation (Figures 5F-H). Consistent with the possibility that Rad17 is required for the response to oncogene-induced replicative stress, the selective advantage produced by Rad17 suppression was more pronounced in primary B-cells expressing Myc compared to normal controls (Supplementary Figure 2A). Intriguingly, the attenuation of this replication stress checkpoint also appears to be an indirect outcome of Mek1 inhibition (see Figure 4C, D), suggesting its central role in tumor suppression.

Rad17 is a haploinsufficient tumor suppressor

While confirming the above results using multiple Rad17 shRNAs, we noted dramatic shRNA-dependent differences in colony formation assays. For example, some Rad17 shRNAs (Rad17.1169 and Rad17.232) enhanced colony formation, while others (Rad17.2567 and Rad17.2159) had the opposite effect (Supplementary Figure 2B). Accordingly, in a competition assay using Eμ-Myc-ARF−/− lymphoma cells, those shRNAs that decreased proliferation were selected against whereas, as shown above, those that enhanced proliferation were enriched (Figure 6A). Although, in principle, these differences might reflect off target effects of RNAi, at least two independent shRNAs targeting Rad17 were able to confer a proliferative advantage or disadvantage, respectively, and all were tumor-promoting to at least some degree (see Figure 3A).

Figure 6
Rad17 acts as a haploinsufficient tumor suppressor

Immunoblotting of cell lysates in this setting of acute Rad17 knockdown revealed an inverse correlation between Rad17 protein levels and proliferative advantage, with shRNAs that intrinsically produced the most potent knockdown being more strongly selected against (Figure 6A, B). Furthermore, those Rad17 shRNAs that conferred a proliferative advantage initially showed only modest Rad17 suppression, although Rad17 levels decreased upon further propagation (data not shown). Presumably, these polyclonal populations contained cells harboring a discreet range of Rad17 levels such that those with optimal knockdown eventually dominated the population. Consistent with these in vitro results, analysis of peripheral blood from mice reconstituted with Eμ-Myc HSPCs showed that B cells harboring weak RAD17 shRNAs were more rapidly enriched in vivo than those harboring potent Rad17 shRNAs (Figure 6C, D). These observations apparently explain the inverse correlation between the ability of Rad17 shRNAs to suppress Rad17 expression and promote tumorigenesis, with the most potent shRNAs being the least oncogenic (Figure 3A). By contrast, the potency of each Rad17 shRNA was directly proportional to their ability to attenuate Chk1 phosphorylation in response to exogenous DNA damaging agents or activation of Myc (Supplementary Figure 4), implicating Chk1 as one downstream effector for the observed effects. Presumably suppression of Rad17 beyond a crucial threshold is deleterious to proliferation.

The above results are consistent with observations that Rad17 null and Chk1 knockout mice die during embryogenesis (Budzowska et al., 2004; Liu et al., 2000) and that deletion of Rad17 in certain tumor lines leads to endoreduplication, chromosomal aberrations, and apoptosis (Wang et al., 2003). Accordingly, the most potent Rad17 shRNAs (Rad17 shRNAs 2567 and 2159) triggered a DNA damage response as assessed by their ability to promote more phospho-H2AX (γH2AX) foci in MEFs relative to controls and the less potent Rad17 shRNAs (1169 and 232) (Figure 6E-F). Moreover, as described for Rad17-deficient cells, nocodazole treatment of cells expressing the most potent Rad17 shRNAs (but not those with weaker suppressive activity) showed less G2/M accumulation and increased polyploidy, compared with controls (Figure 6G). Together, these results suggest that partial suppression of Rad17 confers a proliferative advantage by enabling cells to evade an oncogene-induced replicative stress response, whereas further suppression of Rad17 is deleterious to proliferation owing to catastrophic failure in DNA repair and excessive genomic instability. As such, Rad17 has properties of a haploinsufficient tumor suppressor.

RAD17 and human cancer

The identification of shRNAs targeting p53 and ATM - genes affected by loss-of-function mutations in human tumors - highlights the potential of this in vivo RNAi screen to identify clinically-relevant tumor suppressor genes. Available literature on hits from our screen further supports this notion: in human tumors, the promoter of SFRP1 is found methylated and NUMB has been attributed tumor suppressor functions and it is found underexpressed in breast cancer (Pece et al., 2004; Stylianou et al., 2006). By surveying public gene expression databases (http://www.oncomine.org), we found that RAD17 was significantly underexpressed in a substantial fraction of human diffuse large B cell lymphomas (Figure 7A), which correlated with poor prognosis (Figure 7B). Consistent with the data from primary human lymphomas, several human lymphoma lines displayed varying degrees of RAD17 expression (Figure 7C).

Figure 7
Rad17 is underexpressed in human B-cell lymphoma and its status impacts the survival of lymphoma patients

In order to further validate the function of RAD17 as a tumor suppressor, we re-expressed its cDNA coupled to a GFP-reporter in Bjab human Burkitt lymphoma cells (featuring the lowest levels of RAD17) and performed cell competition assays. As expected, cells expressing exogenous RAD17 were outcompeted compared to controls (Figure 7D). Similarly, forced RAD17 expression has been shown to slow tumor growth in nude mice (Beretta et al., 2008). Of note, a non-phosphorylatable (S635A/S645A) mutant Rad17, was still selected against albeit to a lesser degree that wild-type Rad17, suggesting that its tumor suppressive properties are partially independent of phosphorylation at these residues (data not shown). Together, these data support a role of Rad17 as a tumor suppressor in both murine and human cells.

By surveying a database of copy number alterations at Cold Spring Harbor Laboratory, we noted that RAD17 is frequently deleted in breast and colon cancer (Figure 7E), where its underexpression is common and, in breast cancer, correlates with poor prognosis (http://www.oncomine.org). Similarly, previous studies have suggested that RAD17 can be deleted in head and neck cancer (Zhao et al., 2008). Although deletions encompassing RAD17 are often large and may include other tumor suppressor genes, their frequency in breast and colon cancer approaches that seen for established tumor suppressors such as PTEN (Figure 7E). Together with our functional studies, which demonstrate that RAD17 acts is a haploinsufficient tumor suppressor in mice, these observations suggest that RAD17 has tumor suppressor activity in humans and heterozygous loss may promote tumorigenesis.

DISCUSSION

Here we identified tumor suppressor genes targeting an array of biological processes by conducting a forward genetic screen using a biologically relevant endpoint – tumorigenesis. Although further studies will explore how each gene acts to suppress tumorigenesis, several have biological activities not readily assayed in vitro. Notably, our screen was not exhaustive: improvements in shRNA knockdown efficiency, a broader screen, a larger cohort of animals, and/or expansion to other tumor models will undoubtedly yield additional relevant genes. Hence, this study, when placed in the context of other studies to functionally identify cancer genes, implies that there are a surprisingly large number of genes that, when deregulated in an appropriate genetic background, can contribute to malignancy.

Our approach conceptually parallels the replication competent retrovirus-based insertional mutagenesis screens that have identified candidate oncogenes in the Eμ-Myc model and other systems (Uren et al., 2005). However, none of our top 15 candidate tumor suppressors were identified as sites of common insertions in Eμ-Myc or other lymphoid-based insertional mutagenesis screens (Akagi et al., 2004), suggesting that shRNA screening interrogates a distinct set of genes. Our shRNA-based approach allows a defined selection of genes to be screened and, owing to the trans-acting effects of RNAi, one integration is in principle sufficient to inactivate gene expression from two alleles. Thus, our approach complements insertional mutagenesis screens and identifies yet additional uses for the well-characterized Eμ-Myc mouse model.

In parallel to the current study, we also conducted an RNAi screen using a mouse model of hepatocellular carcinoma (Zender et al.; Cell 2008). In this setting we chose shRNAs targeting the mouse orthologs of genes deleted in human HCC as a guide to enrich the RNAi library for tumor suppressor genes. By expanding to a different in vivo model in this study and employing a more broadly defined set of shRNAs, we discovered tumor suppressor genes that would not have been identified based on genomics data alone. Interestingly, preliminary screening of the Cancer1000 library in the HCC model uncovered candidate tumor suppressors not identified in the lymphoma screen, while several hits from the lymphoma screen did not accelerate liver tumorigenesis (L.Z., W. Xue, S.W.L., unpublished observations). These observations indicate that many tumor suppressor genes function in a context dependent manner and highlight the value of conducting shRNA-based screens in multiple tumor models.

We chose to investigate in detail one of our candidate tumor suppressor genes, rad17. Together, our data support a model where Rad17 acts as a haploinsufficient tumor suppressor by mediating replication stress from oncogenes to both p53-dependent and independent anti-proliferative responses. Alleviation of this effect allows proliferation to continue inappropriately. Of note, lymphomas triggered by the most oncogenic Rad17.1169 shRNA retained a wild type p53 gene and intact p53 response, suggesting that p53-loss is not required for Rad17 suppression to promote tumorigenesis (data not shown). Therefore, while attenuation of Rad17 activity may eventually lead to genomic instability and contribute indirectly to tumorigenesis, we believe the more direct effect on the cell cycle described here is likely to explain its action as a tumor suppressor in our system. In line with previous studies, we found Chk1 activation to be Rad17 dependent. Interestingly, Chk1 also displays phenotypes consistent with a haploinsufficient tumor suppressor, namely deregulated cell cycle entry, accelerated tumor development and, if homozygously deleted, embryonic lethality due to excessive DNA damage (Liu et al., 2000; Lam et al., 2004)

The identification of Rad17 as a tumor suppressor demonstrates the potential of shRNA-based screens to discover and validate haploinsufficient tumor suppressors whose partial loss of expression is pro-oncogenic, but whose complete loss of function is deleterious for pre-neoplastic cells (Payne and Kemp, 2005). Based on the variable potencies of different shRNAs that target the same gene, in vivo RNAi screens are able to survey a broad dynamic range of target gene expression for which those cells with optimal knockdown will be selected for during tumorigenesis. In support of this concept, Rad17 shRNAs that induce distinct levels of knockdown following acute introduction into cell populations feature a more homogeneous suppression of Rad17 in the outgrown tumors (compare Figures 5A and and6B6B).

Importantly, genomic deletions found in human tumor samples are often hemizygous, and it is often assumed that relevant tumor suppressors must display concomitant loss or suppression of the remaining wild type allele. Indeed, reduced expression of Rad17 is observed in human diffuse large B-cell lymphoma (DLBCL) and this correlates with poor prognosis. While it remains to be determined whether hemizygous deletions involving Rad17 occur in DLBCL, they occur and are common in human colon and breast cancer. As such hemizygous deletions in cancer cells can be quite large, there may be many other genes that can contribute to cancer when reduced to a single copy. Since, in these instances there is no clear mutation in a second allele, it is difficult to determine their relevance though genomic approaches alone.

We were surprised that some genes with putative oncogenic properties were identified in our screen, implying that many genes can act both pro- or anti-oncogenic depending on genetic or cellular context. As an example, Mek1, a critical effector in the MAPK pathway, scored in all our assays. While seemingly paradoxical, these studies are consistent with previous work showing that Mek is required for DNA checkpoint activation in response to genotoxic stress (Yan et al., 2007). Antiproliferative functions of Mek have furthermore been corroborated by studies demonstrating that high dose MAPK signaling can produce antiproliferative responses (Olson et al., 1998), and studies suggesting that, in premalignant cells, Mek is required for Ras-induced senescence – a tumor suppressive program (Lin et al., 1998; Zhu et al., 1998). In addition, Mek1 inhibition may destabilize Myc (Sears et al., 2000), enabling proliferation without apoptosis (Murphy et al., 2008), or interfere with a feedback mechanism that would otherwise dampen proliferation (Pratilas et al., 2009). Whatever the precise mechanism whereby Mek1 suppression accelerates tumorigenesis, our data, together with published reports, emphasize that the physiological response to Mek1 inhibition is highly context dependent and strongly influenced by the genetic background in which it occurs.

Although our goal was to identify genes that limit tumorigenesis, our results have therapeutic implications. First, since many chemotherapeutic agents trigger a DNA damage response whose integrity can influence treatment outcome, knowledge of RAD17 status in tumors may help guide the use of chemotherapy in patients. Second, owing to their pro-oncogenic activities in certain settings, some of the tumor suppressors we identified (e.g. MEK1 and ANG2) are targets of inhibitors in clinical trials (Rinehart et al., 2004) – our observations hint that contextual information may be required for the effective use of these inhibitors in the clinic. Finally, our screen identified several tumor suppressor genes that encode secreted proteins, including Sfrp1, Ang2, Fgf15, Wnt1, Shbg, and Bmp3 (Supplementary Table 2, see also Zender et al., 2008). As shRNAs targeting these genes were isolated from pools of cells in which only a portion contain a particular shRNA, it is likely that these factors operate either in an autocrine manner, or as short range paracrine signals that alter the microenvironment in ways that stimulate tumorigenesis. Still, if loss of these proteins is required to sustain tumor progression, systemic delivery of recombinant proteins or peptides may have therapeutic utility [(see, for example, (Wajapeyee et al., 2008)]. It seems likely that these and other high-throughput methods to functionally identify cancer genes will produce further insights into the complexities of cancer development and point towards new therapeutic targets.

EXPERIMENTAL PROCEDURES

Short hairpin RNA vectors

A miR-30-based shRNA library targeting the cancer 1000 gene set (~2300 shRNAs) was subcloned into LMP and LMS (MSCV-based vectors) (Dickins et al., 2005) in pools of 96 or 48 shRNAs, respectively. Individual shRNAs for validation were synthesized as 97bp oligos (Sigma Genosys), PCR-amplified, cloned into LMS and LMP and verified by sequencing. Targeting sequences were selected based on RNAi Codex (Silva et al., 2005) or BIOPREDsi algorithms (Huesken et al., 2005) and are available upon request.

shRNA Recovery, Identification and Determination of Representation

Genomic DNA was isolated from tumor tissues (Puregene, Gentra Systems) and the integrated proviral sequences were amplified with primers flanking the miR30 cassette. The PCR product was digested with EcoRI/XhoI and directionally cloned into LMS. 30–100 bacterial colonies were sequenced by standard capillary sequencing for each tumor. To identify the shRNAs and determine their distribution, the sequence reads were aligned to a list of all shRNAs used in the screen using the blat algorithm (Kent, 2002).

Stem Cell Isolation and Adoptive Transfer

All mouse experiments were performed in accordance with institutional and national guidelines and regulations and approved by the Institution Animal Care and Use Committee (IACUC#06-02-97-17). Pregnant Eμ-Myc (C57BL/6) mice were sacrificed to obtain embryonic 12.5–13.5 (E12.5-E13.5) fetal livers. For hematopoietic reconstitution experiments, 6 to 8 week old C57BL/6J recipient mice received a single 7 Gy dose of total body γ-irradiation (137Cesium source), and were reconstituted 24 hr later with approximately 3 × 106 viable fetal liver cells by tail vein injection. Flow cytometry analysis was performed on a Becton Dickinson LSRII cell analyser with FACSVantage DiVa software and the Guava EasyCyte System with CytoSoft software.

Lymphoma Monitoring and Analysis

Reconstituted animals were monitored for illness by lymph node palpation, overall morbidity, and, in some cases, whole-body fluorescence imaging (Schmitt and Lowe, 2002). Overall survival was defined as the time from stem cell reconstitution until the animal reached morbidity and was sacrificed. Statistical analysis was performed using a one-way ANOVA (analysis of variance) test using Graph Pad Prism version 3.0 (Graph Pad Software). Immunohistochemistry was performed using anti-caspase 3 and anti-PCNA antibodies. Tumor cell DNA content was determined by flow cytometry with propidium iodide staining of ethanol-fixed cells.

- For additional experimental procedures refer to supplemental material.

Supplementary Material

01

Acknowledgments

We thank B. Ma, K. Diggins-Lehet, J. Simon, M. Yang, E. Earl and L. Bianco for excellent technical assistance, and E. Hernando and L. Chiriboga for immunohistochemistry. We also thank R. Dickins and M. Spector for critical reading of the manuscript, and members of the Lowe laboratory for advice and helpful discussions. This work was supported by postdoctoral fellowships from the NCI (A.B.), the Terri Brodeur Breast Cancer Foundation (C.U.B.), the Deutsche Forschungsgemeinschaft (DFG Mi 1210/1-1, C.C.M.), and a clinical fellowship from Alan and Edith Seligson (L.Z., J.Z.), as well as research support from the Don Monti Memorial Research Foundation and grants from the National Cancer Institute (CA105388, CA13106, S.W.L., G.H.). G.H. and S.W.L. are Howard Hughes Medical Institute investigators.

Footnotes

SIGNIFICANCE

Tumor suppressor genes act in a variety of cellular processes to restrict oncogenic transformation and their mode of action can be context dependent. During tumorigenesis, genetic changes that circumvent these failsafe mechanisms are selected for. By employing libraries of shRNAs in an in vivo RNAi screen in the Eμ-Myc model of B cell lymphoma, we speed up this selective process and functionally identify more than ten potential tumor suppressor genes whose suppression accelerates lymphomagenesis. Our results illustrate how functional genetic approaches in mice complement genomic studies for identifying tumor suppressor genes, and point towards new genes and processes that influence cancer development. In addition, they highlight the complexity of cancer gene action and suggest avenues for therapeutic intervention.

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

References

  • Adams JM, Harris AW, Pinkert CA, Corcoran LM, Alexander WS, Cory S, Palmiter RD, Brinster RL. The c-myc oncogene driven by immunoglobulin enhancers induces lymphoid malignancy in transgenic mice. Nature. 1985;318:533–538. [PubMed]
  • Akagi K, Suzuki T, Stephens RM, Jenkins NA, Copeland NG. RTCGD: retroviral tagged cancer gene database. Nucleic Acids Res. 2004;32:D523–527. [PMC free article] [PubMed]
  • Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Rosenwald A, Boldrick JC, Sabet H, Tran T, Yu X, et al. Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature. 2000;403:503–511. [PubMed]
  • Bao S, Tibbetts RS, Brumbaugh KM, Fang Y, Richardson DA, Ali A, Chen SM, Abraham RT, Wang XF. ATR/ATM-mediated phosphorylation of human Rad17 is required for genotoxic stress responses. Nature. 2001;411:969–974. [PubMed]
  • Beretta GL, Gatti L, Cesare MD, Corna E, Tinelli S, Carenini N, Zunino F, Perego P. The human homolog of fission yeast Rad17 is implicated in tumor growth. Cancer Lett 2008 [PubMed]
  • Berns K, Horlings HM, Hennessy BT, Madiredjo M, Hijmans EM, Beelen K, Linn SC, Gonzalez-Angulo AM, Stemke-Hale K, Hauptmann M, et al. A functional genetic approach identifies the PI3K pathway as a major determinant of trastuzumab resistance in breast cancer. Cancer Cell. 2007;12:395–402. [PubMed]
  • Budzowska M, Jaspers I, Essers J, de Waard H, van Drunen E, Hanada K, Beverloo B, Hendriks RW, de Klein A, Kanaar R, et al. Mutation of the mouse Rad17 gene leads to embryonic lethality and reveals a role in DNA damage-dependent recombination. Embo J. 2004;23:3548–3558. [PubMed]
  • de Vries-Smits AM, Burgering BM, Leevers SJ, Marshall CJ, Bos JL. Involvement of p21ras in activation of extracellular signal-regulated kinase 2. Nature. 1992;357:602–604. [PubMed]
  • Desmedt C, Piette F, Loi S, Wang Y, Lallemand F, Haibe-Kains B, Viale G, Delorenzi M, Zhang Y, d’Assignies MS, et al. Strong time dependence of the 76-gene prognostic signature for node-negative breast cancer patients in the TRANSBIG multicenter independent validation series. Clin Cancer Res. 2007;13:3207–3214. [PubMed]
  • Dickins RA, Hemann MT, Zilfou JT, Simpson DR, Ibarra I, Hannon GJ, Lowe SW. Probing tumor phenotypes using stable and regulated synthetic microRNA precursors. Nat Genet. 2005;37:1289–1295. [PubMed]
  • Dominguez-Sola D, Ying CY, Grandori C, Ruggiero L, Chen B, Li M, Galloway DA, Gu W, Gautier J, Dalla-Favera R. Non-transcriptional control of DNA replication by c-Myc. Nature. 2007;448:445–451. [PubMed]
  • Halazonetis TD, Gorgoulis VG, Bartek J. An oncogene-induced DNA damage model for cancer development. Science. 2008;319:1352–1355. [PubMed]
  • Hemann MT, Fridman JS, Zilfou JT, Hernando E, Paddison PJ, Cordon-Cardo C, Hannon GJ, Lowe SW. An epi-allelic series of p53 hypomorphs created by stable RNAi produces distinct tumor phenotypes in vivo. Nat Genet. 2003;33:396–400. [PubMed]
  • Hemann MT, Zilfou JT, Zhao Z, Burgess DJ, Hannon GJ, Lowe SW. Suppression of tumorigenesis by the p53 target PUMA. Proc Natl Acad Sci U S A. 2004;101:9333–9338. [PubMed]
  • Hicks J, Krasnitz A, Lakshmi B, Navin NE, Riggs M, Leibu E, Esposito D, Alexander J, Troge J, Grubor V, et al. Novel patterns of genome rearrangement and their association with survival in breast cancer. Genome Res. 2006;16:1465–1479. [PubMed]
  • Huesken D, Lange J, Mickanin C, Weiler J, Asselbergs F, Warner J, Meloon B, Engel S, Rosenberg A, Cohen D, et al. Design of a genome-wide siRNA library using an artificial neural network. Nat Biotechnol. 2005;23:995–1001. [PubMed]
  • Hummel M, Bentink S, Berger H, Klapper W, Wessendorf S, Barth TF, Bernd HW, Cogliatti SB, Dierlamm J, Feller AC, et al. A biologic definition of Burkitt’s lymphoma from transcriptional and genomic profiling. N Engl J Med. 2006;354:2419–2430. [PubMed]
  • Jhappan C, Morse HC, 3rd, Fleischmann RD, Gottesman MM, Merlino G. DNA-PKcs: a T-cell tumour suppressor encoded at the mouse scid locus. Nat Genet. 1997;17:483–486. [PubMed]
  • Karlsson A, Deb-Basu D, Cherry A, Turner S, Ford J, Felsher DW. Defective double-strand DNA break repair and chromosomal translocations by MYC overexpression. Proc Natl Acad Sci U S A. 2003;100:9974–9979. [PubMed]
  • Kent WJ. BLAT--the BLAST-like alignment tool. Genome Res. 2002;12:656–664. [PubMed]
  • Kuznetsov SG, Haines DC, Martin BK, Sharan SK. Loss of Rad51c leads to embryonic lethality and modulation of Trp53-dependent tumorigenesis in mice. Cancer Res. 2009;69:863–872. [PMC free article] [PubMed]
  • Lam MH, Liu Q, Elledge SJ, Rosen JM. Chk1 is haploinsufficient for multiple functions critical to tumor suppression. Cancer Cell. 2004;6:45–59. [PubMed]
  • Lin AW, Barradas M, Stone JC, van Aelst L, Serrano M, Lowe SW. Premature senescence involving p53 and p16 is activated in response to constitutive MEK/MAPK mitogenic signaling. Genes Dev. 1998;12:3008–3019. [PubMed]
  • Liu Q, Guntuku S, Cui XS, Matsuoka S, Cortez D, Tamai K, Luo G, Carattini-Rivera S, DeMayo F, Bradley A, Donehower LA, Elledge SJ. Chk1 is an essential kinase that is regulated by Atr and required for the G(2)/M DNA damage checkpoint. Genes Dev. 2000;14:1448–1459. [PubMed]
  • Lobov IB, Brooks PC, Lang RA. Angiopoietin-2 displays VEGF-dependent modulation of capillary structure and endothelial cell survival in vivo. Proc Natl Acad Sci U S A. 2002;99:11205–11210. [PubMed]
  • Ma XJ, Wang Z, Ryan PD, Isakoff SJ, Barmettler A, Fuller A, Muir B, Mohapatra G, Salunga R, Tuggle JT, et al. A two-gene expression ratio predicts clinical outcome in breast cancer patients treated with tamoxifen. Cancer Cell. 2004;5:607–616. [PubMed]
  • Maisonpierre PC, Suri C, Jones PF, Bartunkova S, Wiegand SJ, Radziejewski C, Compton D, McClain J, Aldrich TH, Papadopoulos N, et al. Angiopoietin-2, a natural antagonist for Tie2 that disrupts in vivo angiogenesis. Science. 1997;277:55–60. [PubMed]
  • Minn AJ, Gupta GP, Siegel PM, Bos PD, Shu W, Giri DD, Viale A, Olshen AB, Gerald WL, Massague J. Genes that mediate breast cancer metastasis to lung. Nature. 2005;436:518–524. [PMC free article] [PubMed]
  • Murphy DJ, Junttila MR, Pouyet L, Karnezis A, Shchors K, Bui DA, Brown-Swigart L, Johnson L, Evan GI. Distinct thresholds govern Myc’s biological output in vivo. Cancer Cell. 2008;14:447–457. [PMC free article] [PubMed]
  • Olson MF, Paterson HF, Marshall CJ. Signals from Ras and Rho GTPases interact to regulate expression of p21Waf1/Cip1. Nature. 1998;394:295–299. [PubMed]
  • Parker AE, Van de Weyer I, Laus MC, Verhasselt P, Luyten WH. Identification of a human homologue of the Schizosaccharomyces pombe rad17+ checkpoint gene. J Biol Chem. 1998;273:18340–18346. [PubMed]
  • Payne SR, Kemp CJ. Tumor suppressor genetics. Carcinogenesis. 2005;26:2031–2045. [PubMed]
  • Pece S, Serresi M, Santolini E, Capra M, Hulleman E, Galimberti V, Zurrida S, Maisonneuve P, Viale G, Di Fiore PP. Loss of negative regulation by Numb over Notch is relevant to human breast carcinogenesis. J Cell Biol. 2004;167:215–221. [PMC free article] [PubMed]
  • Pratilas CA, Taylor BS, Ye Q, Viale A, Sander C, Solit DB, Rosen N. (V600E)BRAF is associated with disabled feedback inhibition of RAF-MEK signaling and elevated transcriptional output of the pathway. Proc Natl Acad Sci U S A. 2009;106:4519–4524. [PubMed]
  • Pusapati RV, Rounbehler RJ, Hong S, Powers JT, Yan M, Kiguchi K, McArthur MJ, Wong PK, Johnson DG. ATM promotes apoptosis and suppresses tumorigenesis in response to Myc. Proc Natl Acad Sci U S A. 2006;103:1446–1451. [PubMed]
  • Riley T, Sontag E, Chen P, Levine A. Transcriptional control of human p53-regulated genes. Nat Rev Mol Cell Biol. 2008;9:402–412. [PubMed]
  • Rinehart J, Adjei AA, Lorusso PM, Waterhouse D, Hecht JR, Natale RB, Hamid O, Varterasian M, Asbury P, Kaldjian EP, et al. Multicenter phase II study of the oral MEK inhibitor, CI-1040, in patients with advanced non-small-cell lung, breast, colon, and pancreatic cancer. J Clin Oncol. 2004;22:4456–4462. [PubMed]
  • Schmitt CA, Lowe SW. Apoptosis and chemoresistance in transgenic cancer models. J Mol Med. 2002;80:137–146. [PubMed]
  • Schmitt CA, McCurrach ME, de Stanchina E, Wallace-Brodeur RR, Lowe SW. INK4a/ARF mutations accelerate lymphomagenesis and promote chemoresistance by disabling p53. Genes Dev. 1999;13:2670–2677. [PubMed]
  • Sears R, Nuckolls F, Haura E, Taya Y, Tamai K, Nevins JR. Multiple Ras-dependent phosphorylation pathways regulate Myc protein stability. Genes Dev. 2000;14:2501–2514. [PubMed]
  • Serrano M, Lin AW, McCurrach ME, Beach D, Lowe SW. Oncogenic ras provokes premature cell senescence associated with accumulation of p53 and p16INK4a. Cell. 1997;88:593–602. [PubMed]
  • Sharan SK, Kuznetsov SG. Resolving RAD51C function in late stages of homologous recombination. Cell Div. 2007;2:15. [PMC free article] [PubMed]
  • Silva JM, Li MZ, Chang K, Ge W, Golding MC, Rickles RJ, Siolas D, Hu G, Paddison PJ, Schlabach MR, et al. Second-generation shRNA libraries covering the mouse and human genomes. Nat Genet. 2005;37:1281–1288. [PubMed]
  • Stylianou S, Clarke RB, Brennan K. Aberrant activation of notch signaling in human breast cancer. Cancer Res. 2006;66:1517–1525. [PubMed]
  • Uren AG, Kool J, Berns A, van Lohuizen M. Retroviral insertional mutagenesis: past, present and future. Oncogene. 2005;24:7656–7672. [PubMed]
  • van de Vijver MJ, He YD, van’t Veer LJ, Dai H, Hart AA, Voskuil DW, Schreiber GJ, Peterse JL, Roberts C, Marton MJ, et al. A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med. 2002;347:1999–2009. [PubMed]
  • Vassilev LT, Vu BT, Graves B, Carvajal D, Podlaski F, Filipovic Z, Kong N, Kammlott U, Lukacs C, Klein C, et al. In vivo activation of the p53 pathway by small-molecule antagonists of MDM2. Science. 2004;303:844–848. [PubMed]
  • Wajapeyee N, Serra RW, Zhu X, Mahalingam M, Green MR. Oncogenic BRAF induces senescence and apoptosis through pathways mediated by the secreted protein IGFBP7. Cell. 2008;132:363–374. [PMC free article] [PubMed]
  • Wang X, Zou L, Zheng H, Wei Q, Elledge SJ, Li L. Genomic instability and endoreduplication triggered by RAD17 deletion. Genes Dev. 2003;17:965–970. [PubMed]
  • Wang Y, Klijn JG, Zhang Y, Sieuwerts AM, Look MP, Yang F, Talantov D, Timmermans M, Meijer-van Gelder ME, Yu J, et al. Gene-expression profiles to predict distant metastasis of lymph-node-negative primary breast cancer. Lancet. 2005;365:671–679. [PubMed]
  • Wang X, Zou L, Lu T, Bao S, Hurov KE, Hittelman WN, Elledge SJ, Li L. Rad17 phosphorylation is required for claspin recruitment and Chk1 activation in response to replication stress. Mol Cell. 2006;23:331–341. [PubMed]
  • Westbrook TF, Martin ES, Schlabach MR, Leng Y, Liang AC, Feng B, Zhao JJ, Roberts TM, Mandel G, Hannon GJ, et al. A genetic screen for candidate tumor suppressors identifies REST. Cell. 2005;121:837–848. [PubMed]
  • Witt AE, Hines LM, Collins NL, Hu Y, Gunawardane RN, Moreira D, Raphael J, Jepson D, Koundinya M, Rolfs A, et al. Functional proteomics approach to investigate the biological activities of cDNAs implicated in breast cancer. J Proteome Res. 2006;5:599–610. [PMC free article] [PubMed]
  • Xu Y, Ashley T, Brainerd EE, Bronson RT, Meyn MS, Baltimore D. Targeted disruption of ATM leads to growth retardation, chromosomal fragmentation during meiosis, immune defects, and thymic lymphoma. Genes Dev. 1996;10:2411–2422. [PubMed]
  • Xue W, Krasnitz A, Lucito R, Sordella R, Vanaelst L, Cordon-Cardo C, Singer S, Kuehnel F, Wigler M, Powers S, et al. DLC1 is a chromosome 8p tumor suppressor whose loss promotes hepatocellular carcinoma. Genes Dev. 2008;22:1439–1444. [PubMed]
  • Yan Y, Black CP, Cowan KH. Irradiation-induced G2/M checkpoint response requires ERK1/2 activation. Oncogene. 2007;26:4689–4698. [PubMed]
  • Zender L, Xue W, Zuber J, Semighini CP, Krasnitz A, Ma B, Zender P, Kubicka S, Luk JM, Schirmacher P, et al. An oncogenomics-based in vivo RNAi screen identifies tumor suppressors in liver cancer. Cell. 2008;135:852–864. [PMC free article] [PubMed]
  • Zhao M, Begum S, Ha PK, Westra W, Califano J. Downregulation of RAD17 in head and neck cancer. Head Neck. 2008;30:35–42. [PubMed]
  • Zhu J, Woods D, McMahon M, Bishop JM. Senescence of human fibroblasts induced by oncogenic Raf. Genes Dev. 1998;12:2997–3007. [PubMed]