We have recently adapted miRNA-based shRNA gene silencing to in vivo applications in the Eμ-myc lymphoma mouse, a well-established model of B cell lymphoma^7-9. This approach has subsequently been expanded to examine small sets of shRNAs in a cohort of mice¹⁰. However, despite the disseminated and effective use of cell-based RNAi screens^11-17, adaptation of these approaches to mouse models has been limited. Here, we use RNAi to interrogate loss of function phenotypes for large gene sets in the context of individual mice. Results from this study identify a set of genes involved in cytoskeletal organization and cell migration that are important for lymphoma progression in vivo. Importantly, this study demarcates critical determinants of tumor behavior in vivo, information that could not be obtained using conventional cell-based screening approaches.

Previous studies in the Eμ-myc system have suggested that Eμ-myc lymphomas are largely composed of cells with tumor initiating potential. Specifically, tumor cell dilution experiments have shown that as few as 10 tumor cells can produce tumors following tail vein injection into syngeneic recipient mice ¹⁸. Thus, we reasoned that if nearly all tumor cells have the capacity to contribute to tumor formation following transplantation, then this system might accommodate the introduction of a diverse set of shRNA-infected cells into a given tumor. To test this, we performed a dilution experiment using lymphoma cell cultures that were partially transduced with a retroviral vector co-expressing an shRNA targeting Topoisomerase 2α (Top2A), an essential mediator of the cytotoxic effects of the chemotherapeutic doxorubicin ¹⁹, and the gene encoding Green Fluorescent Protein (GFP) to mark infected cells. Viral multiplicity of infection (MOI) was titered, such that either 2.0% or 0.2% of lymphoma cells were infected. These partially transduced lymphoma cell populations were injected into syngeneic recipient mice, and mice were treated with doxorubicin at the time of lymphoma presentation. At both infection efficiencies, we saw a significant increase in the percentage of GFP-positive cells following treatment (Fig. 1a). These results showed that cells representing as little as 1/500^th of the injected lymphoma cell population were retained at the time of doxorubicin treatment. These data are consistent with large numbers of lymphoma cells (at least 500) contributing to an individual tumor following transplantation in vivo, and suggest the Eμ-myc lymphoma model may be appropriate for in vivo, pool-based RNAi screens with complex shRNA libraries.

Based on these preliminary results, we performed an shRNA screen with a pool of approximately 2250 hairpins targeting 1000 genes with known or putative roles in cancer¹⁹. Retroviral plasmids expressing these shRNAs, as well as GFP, were pooled and packaged as a mixture of retroviruses. This retroviral pool was then used to infect lymphoma cells, and the resulting transduced cells were injected into recipient mice, maintained in culture, or collected immediately to serve as a reference sample (Fig. 1b). At the time of lymphoma presentation, tumors were harvested and genomic DNA was extracted from primary tumors and cultured cells. Following PCR amplification of shRNAs from genomic DNA using common primers (Fig. 1b), hairpin representation was analyzed by high throughput sequencing.

Approximately 1600 of the original 2250 unique hairpins could be identified from each in vitro cultured sample, and, surprisingly, between 600 and 900 unique hairpins could be identified in lymphomas from individual mice (Fig. 1c). Thus, a large, diverse shRNA set can be introduced in vivo and a significant proportion of the initial library complexity can be maintained in this setting. Sequencing of shRNAs from genomic DNA derived from outgrown single cell clones showed that approximately 90% of cells were infected with only a single shRNA (Supplementary Fig. 1), suggesting that, at a minimum, 900 cells can contribute to the lymphoma burden in an individual mouse. Thus, not only can a large percentage of lymphoma cells give rise to a tumor following transplantation, but, in fact, many of these cells contribute to the growth of the resulting tumor.

Unsupervised hierarchical clustering showed that the three in vivo samples were more similar to one another than to any of the cultured in vitro samples, based on the hairpins that enriched or depleted in each setting (Fig. 1d). Thus, the set of genes that impacts cancer cell homeostasis in vitro is distinct from those that are central to tumor growth in vivo. Importantly, the number of sequencing reads obtained was sufficient to see both enrichment and depletion of shRNAs from the initial injected population. While many shRNAs displayed similar changes in representation in clustered samples, significant variation was also present in samples of the same type. This is likely due to the stochastic gain or loss of shRNAs following introduction into mice or serial replating in culture.

Hairpins whose representation decreased at least 10-fold in all three replicates or enriched at least 5-fold in two out of three replicates, relative to their representation in the control cell population collected shortly after retroviral transduction, scored as candidate hits. The set of scoring shRNAs in vitro by this criteria was largely non-overlapping with the set of shRNAs that scored in vivo (Fig. 1e, Supplementary Table 1), indicating that by performing shRNA screens in the context of a normal tumor microenviroment, we were able to identify a set of genes that exclusively impacts growth in a physiologically relevant setting.

We focused our follow-up studies on shRNA sets that specifically affected tumor growth in vivo. As a more stringent criterion, we selected genes for which 2 or more cognate shRNAs depleted on average at least 10-fold in mice (Table 1). Based on gene ontology (GO) classifications and manual curation, shRNAs targeting genes involved in cell motility, including dynamic actin reorganization and cell adhesion, were highly represented in the set of hairpins that specifically depleted in vivo (8 out of 11). Several of these genes were chosen for validation (Supplementary Fig. 2a). These included genes encoding Rac2, a hematopoetic-specific Rho GTPase important for the formation of lamellipodia during cell migration^20,21, CrkL, an adaptor protein reported to be involved in the activation of Rac²², and Twinfilin (Twf1), an actin monomer-binding and actin filament capping protein²³. For these genes, multiple independent shRNAs (Fig. 2a) recapitulated the initial screening phenotype. Specifically, while cells grown in the presence of these shRNAs grew robustly in culture, they all showed a selective depletion in lymph nodes following tail vein injection (Fig. 2b). To further characterize the efficiency of our screen, we examined a non-motility gene targeted by two depleted shRNAs and two genes targeted by a single scoring shRNA. shRNAs targeting the genes encoding IL-6 (two depleted shRNAs in our screen) and Lyn kinase (one depleted shRNA in our screen) depleted following introduction in mice, while shRNAs targeting the gene encoding AIF-1 (one depleted shRNA in our screen) failed to deplete (Supplementary Fig. 2b and data not shown). Thus, shRNAs targeting four out of four genes from the gene set identified using our most stringent criteria validated as single constructs, while the validation rate for single scoring shRNAs was lower.

To further characterize the role of genes identified in our screen in lymphoma homeostasis, we subjected shRNA-infected cells to secondary functional assays. Lymphoma cells suppressing Rac2, CrkL, or Twf1 showed motility defects in transwell migration assays, consistent with a role for these proteins in lymphoma cell migration (Fig. 2c and d and Supplementary Fig. 3a and b). Additionally, Rac2-deficient lymphoma cells showed defects in SDF-1α induced migration on fibronectin (Supplementary Movies 1 and 2). Suppression of Rac2 also resulted in impaired lymphoma cell migration in short term in vivo engraftment assays. Specifically, lymphoma cells suppressing Rac2 were depleted in the spleen and bone marrow two and twenty-four hours after tail vein injection (Fig. 3a). Similar to the effect Rac2 knockdown, lymphoma cells suppressing Wave2, an important mediator of cell migration known to function downstream of other Rac proteins²⁴, showed chemotaxis defects in vitro and were selectively depleted in the lymph nodes at the time of disease presentation (Fig. 3b and Supplementary Fig. 3c and d).

Suppression of Rac2 was also selected against in common sites of lymphoid metastasis, such as the liver, as seen by histology and by GFP enrichment analysis (Figs. 3c and d). suggesting that Rac2 might represent a meaningful lymphoma drug target. In fact, suppression of Rac2 in lymphoma cells extended both tumor free and overall survival following tail vein injection (Fig. 3e and f). Similarly, lymphoma-bearing mice treated with NSC23766²⁵, an inhibitor of Rac1 and Rac2 (Supplementary Fig. 4a)²⁶, survived significantly longer than untreated controls (Fig. 4a). Notably, suppression of Twf1 also delayed tumor progression following lymphoma tail vein injection (Supplementary Fig. 4b). Thus, multiple proteins involved in actin reorganization and cell motility represent potential drug targets in B cell malignancies. Additionally, combinations of shRNAs produced synergistic effects on lymphoma growth (Supplementary Fig. 4c).

Interestingly, when lymphomas were treated with the chemotherapeutic vincristine, knockdown of Rac2 or Twf1 extended animal survival following drug treatment (Fig. 4b and Supplementary Fig. 4d). Importantly, this effect was specific for therapeutic response in vivo, as suppression of Rac2 did not sensitize lymphoma cells to vincristine treatment in culture (data not shown). These results suggest that there is a requirement for tumor cell mobilization in lymphoma relapse and that Rac2 and Twf1 activity is important in this mobilization (Fig. 4c). It further suggests that Rac2 or Twf1 inhibition, or inhibition of lymphoma cell motility by another mechanism, may synergize with conventional chemotherapeutics in the treatment of lymphoma.

Using a well-established mouse model of B cell lymphoma, we have adapted RNAi screens to in vivo applications. With this system, we were able to screen over 900 unique hairpins in individual mice. While improvements in RNAi technology will be required to perform saturating loss of function studies, this work demonstrates the feasibility of using large, unbiased shRNA sets for in vivo screens. Importantly, unlike conventional microarray studies, this approach investigates gene function, rather than gene expression – implicating “scoring” genes as directly relevant to interrogated phenotypes. We can envision using a similar approach to examine modulators of therapeutic response or tissue-specific tumor dissemination. Importantly, identification of similarly tractable genetic systems may permit the adaptation of this screening methodology to study tissue development or the biology of solid tumors.

Our screen identified modulators of lymphoma cell motility and chemotaxis as key determinants of tumor homeostasis. This group included known regulators of actin-based cell motility, as well as proteins, like Twf1, with known roles in actin dynamics but no established role in mammalian cell migration. Suppression of Rac2 and Twf1 in lymphoma cells impaired lymphoma cell migration to the lymph nodes and other organs that represent common sites of lymphoid metastasis. Additionally, pharmacological inhibition of Rac2 synergized with a conventional chemotherapeutic to extend the lifespan of tumor-bearing mice. These results highlight a potential therapeutic strategy in hematopoietic cancers, suggesting that, in instances where there is minimal metastasis at the time of initial treatment with traditional chemotherapeutics, suppression of cell migration may improve therapeutic outcome.

1