Cancer can be effectively treated using targeted therapy, as exemplified by Imatinib 1 or Sorafenib 2. There are increasing efforts to fulfill the promise of targeted therapy, using antibodies, peptides and small molecules that selectively affect cancer cells. In each case, the key is to identify target molecules that play a unique role in tumor cells.

Genes encoding such target molecules can be discovered by either comparative or functional genomic approaches. Comparative approaches analyze cytogenetic data, genomic sequences, mRNA expression profiles or proteomic profiles, and select target genes or proteins based on differential expression or mutation status. For example, high-throughput sequencing of cancer cell genomes identified BRAF 3 and PIK3CA 4 as frequently mutated genes in multiple human tumors. On the other hand, functional approaches involve perturbing cells with agents, such as cDNAs, small RNAs, or small molecules, and searching for those that induce specific phenotype changes. Subsequent target identification may lead to the discovery of cancer therapeutic targets. Indeed, the RAS oncogenes were identified using an expression cloning strategy that searched for human genes that transform the mouse fibroblast cell line NIH3T3 5.

Among the agents used for functional genomic approaches, small RNAs are increasingly appealing, because RNA-interference (RNAi) mediated by small RNAs enables gene silencing in mammalian cells. RNAi is a naturally occurring phenomenon involved in the silencing of genes, which results in regulation of gene expression or activation of an antiviral defense system 6. The RNAi pathway involves DICER, which processes double-stranded RNAs into small RNA duplexes (approximately 22 nucleotides). One strand of the small RNA duplex is incorporated into an effector complex known as the RNA-induced silencing complex (RISC) and acts as a guide molecule in translational repression or mRNA cleavage, depending on the degree of base-pair match with the target mRNA 7.

The conserved RNAi pathway is also activated by experimentally designed double-stranded RNAs or short hairpin RNAs (shRNAs), which make it possible to knock down genes of interest in mammalian cells. Consequently, RNAi libraries targeting large numbers of mRNAs have been generated and used for conducting high-throughput, loss-of-function screens in tissue culture systems. For example, RNAi libraries were used to identify novel tumor suppressors 89, regulators of cell death and survival 10, and novel components of p53 signaling 11. Moreover, RNAi libraries were used for understanding the mechanisms of action of novel compounds 12, for characterizing determinants of sensitivity to clinically used drugs 13, and for identifying novel targets for anti-cancer therapy, using a pair of isogenic cell lines 14.

Isogenic cell lines are useful for discovering therapeutic agents and probing the biology of transformation. They may consist of cancer cells at different stages of malignancy, or a specific cancer gene can be deleted to create an isogenic cell line counterpart. Another approach is to isolate primary cells and induce transformation by sequential addition of oncogenic elements. This system provides a series of genetically defined cell lines, and thereby allows for identification of tumor-cell-selective, or even genotype-selective, lethal agents. The successful use of such a system has been described for identification of small molecules with potentially high therapeutic indices 15.

Here we utilized an RNAi library consisting of shRNAs targeting human kinases to find kinases whose inactivation induces tumor-cell-selective lethality or growth arrest. The initial screening was conducted in two sarcoma cell lines; then, a series of isogenic cell lines derived from primary fibroblasts were used for selecting tumor-cell-specific cytotoxic shRNAs. We report that knocking down CSNK1E, a clock gene encoding casein kinase 1-epsilon (CK1ε), induces tumor-cell-selective cytotoxicity. Subsequent validation experiments showed that tumor cells depend more on the kinase activity of CK1ε than normal cells do. The use of a kinase inhibitor specific to CK1ε revealed that another clock protein, PERIOD2, is a key substrate of CK1ε and modulates tumor cell growth.

Our RNAi library was made of lentivirus solutions in 384-well plates. Each well contains lentiviruses harboring expression plasmids encoding a single shRNA that is designed to target a single mRNA. The arrayed library targets 1,006 human genes, including 571 kinases; most of them are protein kinases, while other kinases acting on nucleic acids, lipids, and carbohydrates are included (Figure 1a).

We began the screen with two different sarcoma cell lines, with the goal of pre-selecting shRNAs that are lethal to these tumor-derived cell lines. We infected U-2-OS, osteosarcoma-derived cells, and HT1080, fibrosarcoma-derived cells, in triplicate and incubated them for three days. This allows time for the shRNAs to be expressed, to bind to their target mRNAs, and cause a reduction in expression of the encoded protein, as the protein is turned over. Percent growth inhibition was determined by adding alamar blue to the culture, and by measuring fluorescence.

A number of shRNA clones displayed growth inhibitory effects (Figure 1b,c and Additional data file 1). Statistical analysis of primary screening data revealed 195 genes whose knockdown inhibited growth of either cell lines more than 25% (P < 0.01; Figure 1b and Additional data file 1). Fifty-two genes affected cell growth in both cell lines, while other genes had specific effects on each cell line, which may reflect the different tissue origin of these two sarcomas. Some of the hit genes in common between the two cell lines are well-known regulators of the cell cycle or cell survival, but there were nine genes whose functions have not been described (Additional data file 1). We were most interested in genes whose functions are most critical to the survival of these two cancer cell lines. Accordingly, we calculated the sum of the percent growth inhibition in each cell line and selected seven shRNAs whose summed values were >90% (Figure 1c). Six out of these seven genes were among the 52 common hits in Figure 1b. One gene, PPM1E, was not statistically significant and, as expected, shRNAs targeting PPM1E were not active in our follow-up analysis (data not shown).

Reducing expression of these six target genes causes growth arrest or cell death in two different sarcoma-derived cancer cell lines; however, we were concerned that knockdown of these six genes may affect normal cells to the same degree. To identify shRNA clones that have cancer cell selectivity, we created fresh batches of lentivirus harboring the six shRNA clones and retested them in a pair of nearly isogenic cell lines, BJ-TERT and BJ-TERT/LT/ST/RAS^V12. Both cell lines were derived from primary human BJ foreskin fibroblasts 16. These BJ primary cells were engineered successively to express the catalytic subunit of human telomerase (hTERT), the SV40 large T and small T oncoproteins (LT and ST), and an oncogenic allele of HRAS (HRAS^G12V). We refer to these cells lines as BJ-TERT, BJ-TERT/LT/ST, and BJ-TERT/LT/ST/RAS^V12. Only the BJ-TERT/LT/ST/RAS^V12 cells form tumors in nude mice. Therefore, testing of shRNAs in BJ-TERT and BJ-TERT/LT/ST/RAS^V12 should enable one to identify genes with a function that is essential in tumor cells, but not normal cells. We measured trypan blue exclusion to evaluate the cytotoxic or growth inhibitory effect of these shRNA clones on normal cells and their isogenic engineered tumor cell counterparts.

Out of these six shRNA clones, five did not show differential activity in the two cell lines; however, the shRNA targeting CSNK1E (hereafter, shCSNK1E) had a tumorigenic-cell-line-specific activity (Figure 2a). The activity of shCSNK1E was further tested in four different BJ-derived cell lines, namely, BJ-TERT, BJ-TERT/LT/ST, BJ-TERT/LT/ST/RAS^V12, and DRD cells. DRD cells were engineered to express hTERT, ST, HRAS^G12V, dominant negative p53, and constitutively active cyclin-dependent kinase (CDK)4/cyclin D, which inactivates the RB protein 17. The p53DD/CDK4/cyclin D1 combinations substitute for LT. DRD cells are tumorigenic in nude mice, which is expected from the fact that they are also derived from BJ primary cells and the effects of mutations in both cell lines should be similar. The growth inhibitory potential of shCSNK1E increased as the cell doubling time decreased, suggesting that the activity is proliferation-rate dependent rather than genotype dependent (Figure 2b).

Theoretically, the length of shRNA involved in base paring with the target mRNA is long enough to ensure specificity of the shRNA clone. However, mismatches between a shRNA and target mRNAs are tolerable; RISC is able to suppress expression of off-target mRNAs whose sequences do not perfectly complement the guide strand of the shRNA 7. In order to confirm our hypothesis that knocking down expression of CSNK1E is responsible for the observed growth inhibition, we tested multiple shRNA clones targeting the CSNK1E gene; each shRNA clone binds to different regions of the CSNK1E mRNA. If more than a single shRNA clone induces growth inhibition, CSNK1E is likely to be the relevant target, because the probability of a common off-target effect of multiple shRNA clones with unrelated sequences is low. We found that four shRNAs targeting CSNK1E induced strong growth inhibition in HT1080 cells (Figure 2c). The level of CSNK1E mRNA decreased upon expression of these shRNAs, as assessed by real-time quantitative PCR analysis (Figure 2d). Note that one of these shRNAs, clone 1838, did not display stronger growth inhibition effects even though the mRNA level was decreased significantly. This is likely to reflect an off-target effect of this particular shRNA.

The CSNK1E gene encodes the CK1ε protein, whose main function is to regulate the circadian rhythm by phosphorylating other clock gene products 18. The role of CK1ε in cancer has been speculated upon, because CK1ε was shown to phosphorylate key proteins in cancer signaling pathways, such as p53 19 and β-catenin 20. However, the significance of these phosphorylation events in carcinogenesis is not known, and the possibility of using CK1ε as a pharmacological target for cancer treatment has not been considered. Therefore, we analyzed the expression level of CSNK1E in human tumor samples to obtain support for its involvement in human cancer. Some genes that are specifically required for tumor maintenance are overexpressed in cancer cells over normal cells. We analyzed the gene-expression database Oncomine for differential expression patterns in normal versus tumor in different tissue types 21. The Oncomine database contained microarray expression data for CSNK1E from ten different tissues, including brain, head and neck, renal, bladder, leukemia, lung, melanoma, prostate, salivary gland, and seminoma. Interestingly, all tumor tissues in the database showed upregulated CSNK1E expression compared to normal tissues, suggesting a positive role of CK1ε in cancer maintenance or neogenesis (Figure 3).

The proliferation-rate-dependent action of shCSNK1E (Figure 2b) raises the possibility that shRNA treatment induces cell cycle arrest; thus, fast growing cells have greater growth inhibition. To test this hypothesis, we stained the DNA of shRNA-treated cells with propidium-iodide and analyzed the cell cycle distribution by flow cytometry. The cell cycle distribution profile indicates that, after expression of shCSNK1E, HT1080 cells were arrested in the second gap (G2) phase of the cell cycle, with a concomitant increase in the population of cells harboring less than the normal diploid DNA content (that is, sub-G1), implying apoptosis had occurred (Figure 4a).

The cell cycle is primarily regulated by the activity of cyclins and CDKs. Among CDK/cyclin complexes, CDK1-cyclin A promotes the transition from G2 to mitosis (M), while CDK1-cyclin B governs maturation of M phase 22. We examined whether shCSNK1E treatment affected expression of cyclin A2 and cyclin B1 in HT1080 cells. Real-time PCR analysis revealed that shCSNK1E decreased mRNA levels of cyclin B1 and cyclin A2 (Figure 4b). In contrast, mRNA levels of cyclin D1, whose function is important for the G1 to S transition, were slightly increased (Figure 4b). These data are consistent with the cell cycle distribution pattern after shCSNK1E treatment observed by flow cytometry; knocking down CSNK1E expression caused down-regulation of cyclin B1 and cyclin A2, which results in cell cycle arrest at the G2/M phase.

In addition to G2/M phase cell cycle arrest, shCSNK1E treatment induced apoptotic cell death, as evidenced by the appearance of small, fragmented cells and a sub-G1 population (Figure 4a). To confirm the apoptotic phenotype of shCSNK1E-treated cells, we examined cleavage of poly(ADP-ribose)polymerase-1 (PARP1), which is cleaved by caspases during apoptosis. Western blot analysis with antibodies specific to PARP1 showed that cells treated with a non-targeting shRNA contained only full length PARP1, whereas those treated with shCSNK1E or staurosporine, a known inducer of caspase-dependent apoptosis, contained a diagnostic PARP1 fragment, indicating that apoptotic caspases were activated by these treatments (Figure 4c). Activation of apoptotic caspases was further confirmed by western blot, which detected the active form of caspase-3 only in shCSNK1E or staurosporine-treated samples (Figure 4c). Thus, shCSNK1E induces caspase-mediated apoptosis in sensitive cancer cells.

These results suggest that chemotherapeutic reagents targeting CK1ε may induce growth arrest and apoptosis with some degree of cancer cell selectivity. To test this hypothesis, we examined the effect of IC261, a kinase inhibitor of CK1ε, in cell culture. IC261 was reported to selectively inhibit casein kinase 1 compared to other protein kinases, by an ATP-competitive mechanism. Moreover, it showed an order of magnitude greater selectivity for CK1δ and CK1ε over other casein kinase 1 isoforms 23. Treatment with IC261 started to inhibit the growth of HT1080 cells at submicromolar concentrations (Figure 5a). When we tested IC261 in BJ-TERT and BJ-TERT/LT/ST/RAS^V12 cells, the sensitivity of BJ-TERT/LT/ST/RAS^V12 cells was greater than that of BJ-TERT cells, which was consistent with the results obtained with shCSNK1E (Figures 5b and 2a). These data suggest that inhibition of the kinase activity of CK1ε is crucial for the observed growth arrest and apoptosis, as opposed to other functions of this protein, such as those mediated by protein-protein interactions. Moreover, as shRNAs targeting CK1δ were not effective in suppressing cell growth during primary screening, the cancer-cell-selective activity of IC261 can likely be attributed to its inhibition of CK1ε (Additional data file 1).

CK1ε is known to control the circadian rhythm by phosphorylating clock proteins, such as PERIOD and CRYPTOCHROME 24. These clock proteins are also reported to regulate the cell cycle, suggesting they have a role in linking the circadian system and the cell cycle machinery 2526. Mammalian cells have three isoforms of PERIOD proteins and two isoforms of CRYPTOCHROME proteins, which are encoded by PER1, PER2, PER3, CRY1 and CRY2 genes, respectively. In order to define the role of each isoform in CK1ε-mediated growth regulation, we conducted counter-screening with shRNAs targeting these genes to identify suppressors of IC261-induced growth inhibition in HT1080 cells. Knocking down expression of PER1, PER3, CRY1, or CRY2 did not affect growth inhibition by IC261 (Additional data file 2). However, four different shRNA clones targeting PER2 suppressed IC261-induced growth inhibition, implying that PERIOD2 is the most crucial substrate of CK1ε in controlling cell proliferation (Figure 5c,d). Note that the maximum growth inhibition by IC261 in Figure 5c is smaller than that in Figure 5a, though they have similar EC50 values of 0.1 μg/ml. This is because cells have been growing for three days before being treated with IC261 in order to express shRNAs targeting PER2, whereas in Figure 5a, IC261 was added to culture at the time of cell seeding.

As we showed that the proliferation rate of target cells is an important determinant of growth inhibition by CSNK1E knockdown (Figure 2b), we measured the proliferation rate of HT1080 cells upon PER2 knockdown using the alamar blue assay (Figure 5e). None of the shRNA clones targeting PER2 changed the proliferation rate of HT1080 cells, indicating that the protective effect of PER2 knockdown on IC261-induced growth inhibition is not caused by slowing cell growth. It has been shown that a major function of CK1ε in the circadian rhythm is to phosphorylate PERIOD2, which drives proteosome-mediated degradation of PERIOD2 27. In several independent reports, overexpression of PERIOD2 has been shown to exert anti-tumor effects in both cell culture and mouse models 262829. Therefore, treatment with IC261 is likely to stabilize PERIOD2, which activates the PERIOD2-mediated tumor suppression pathway.

Here, we report the identification of CK1ε as a potential target for developing anticancer reagents. The mammalian CK1 family consists of at least seven isoforms (α, β, γ1, γ2, γ3, δ and ε), as well as additional splice variants 18. They share highly conserved kinase domains, but differ significantly in the length and primary structure of their amino- and carboxy-terminal non-catalytic domains, implying that each isoform may play a specific role in regulating biological processes 18. Defining isoform-specific functions will aid us in developing agents with enhanced specificity and reduced off-target effects. As the specificity of RNAi agents is potentially high, it allows us to differentiate among these isoforms, which is challenging for some chemical inhibitors.

In our screening, knocking down other isoforms of CK1 was not effective at inducing growth arrest, implying that CK1ε has a unique function in promoting the integrity and proliferation of tumor cells. The nature of the signaling pathway that CK1ε uses to control cell growth remains elusive, but several lines of evidence support a positive role of this kinase in oncogenesis. First, in our gene expression analysis, cancer cells have a high level of CK1ε compared to normal cells, regardless of the tissue origin, implying that a high level of CK1ε causes a growth or survival advantage during tumorigenesis (Figure 3). Second, in a recent report, enforced expression of myristoylated-CK1ε, but not other isoforms, induced colony formation in soft-agar-growing engineered human epithelial cells 30. Third, deletion of PERIOD2 in mice caused increased tumor development upon gamma-radiation, suggesting a tumor suppressive role of PERIOD2 26. CK1ε is a major kinase that phosphorylates and degrades the PERIOD2 protein through the proteasome 31; therefore, it is likely that CK1ε exerts its oncogenic effect by inhibiting the tumor suppressive function of PERIOD2. In accordance with this model, we showed that knocking down PER2 abrogated the growth inhibitory effect of IC261, a kinase inhibitor of CK1ε (Figure 5c).

RNAi libraries and isogenic cell lines make it possible to identify target genes and proteins for cancer therapeutic development. We found that CSNK1E is one such target gene upon which cancer cells depend more than normal cells. As kinase inhibitors of CK1ε displayed the same phenotype as shRNA treatments, efforts to develop kinase inhibitors of CK1ε with enhanced potency and selectivity would be valuable. Future work involving the screening of larger shRNA libraries might reveal additional potential drug targets.

CDK, cyclin-dependent kinase; CK1ε, casein kinase 1-epsilon; DMEM, Dulbecco's modified Eagle's medium; GI, growth inhibition; hTERT, catalytic subunit of human telomerase; LT, SV40 large T oncoprotein; PARP1, poly(ADP-ribose)polymerase-1; PBS, phosphate-buffered saline; RISC, RNA-induced silencing complex; RNAi, RNA-interference; shCSNK1E, shRNA targeting CSNK1E; shRNA, short hairpin RNA; ST, SV40 small T oncoprotein; VCM, viral collection media.

WSY and BRS conceived the study, designed the experiments, analyzed the data, and wrote the manuscript. WSY collected the data. BRS supervised the research.

The following additional data are available. Additional data file 1 displays the growth inhibitory activity of all shRNAs used in this study, statistical analysis, and list of hits. Additional data file 2 is a figure showing the results of testing shRNAs targeting PER1, PER3, CRY1, and CRY2 for suppressing IC261-induced growth arrest.