METHODS FOR IDENTIFYING PROTEINS ESSENTIAL FOR HUMAN CELL PROLIFERATION AND THERAPEUTIC AGENTS

THAT TARGET THEM

BACKGROUND OF THE INVENTION

This invention relates to the fields of kinases, cancer, and pharmaceutical compounds.

Mammalian cells in culture have provided a powerful and immensely useful system to study many aspects of eukaryotic cell physiology. Tissue culture cells are superb systems for many biochemical and cell biological studies, such as the study of signal transduction pathways or the analysis of protein translocation. They have also been excellent sources from which purified systems for the in vitro study of many cell processes have been developed. However, mammalian tissue culture systems have not generally been useful for complex genetic studies.

Genetically tractable organisms have provided powerful models for the study of many aspects of biology in unbiased manners (see for example Hartwell et al., 1970; Hartwell et al, 1974; Hafen et al, 1987; Simon et al., 1991; Nurse et al., 1998; Moon et al., 2004). When genetic screens are done to saturation in these systems, they allow comprehensive identification of the genetic components which play a role in the biological event under study. Mammalian tissue culture cells have hitherto not been tremendously useful for such genetic screening studies for several reasons. It is difficult to construct and test homozygous null genetic backgrounds in mammalian cells because they are diploid and there is no effective system for mating or exchange of genetic information between cells. Many of the cells that grow well in culture are genetically unstable, making any long-term selection or assessment more difficult.

Recent advances in methods for manipulation of cDNA clones and the development of RNA interference (RNAi) made cells in culture more amenable to genetic manipulation (Wu et al., 2002; Carpenter and Sabatini, 2004). Using RNAi it is possible to alter the levels of a given protein and measure the resulting phenotype. Protein levels can be raised by synthesis from cDNA expression vectors or reduced by introducing an inhibitory RNA. These types of changes can be done in high throughput to screen large numbers of proteins, with the ultimate goal of screening the entire coding potential of the genome, termed the proteome (Ziauddin and Sabitini, 2001; Chanda et al. 2003; Matsuda et al., 2003; Berns et al., 2004; Kittler et al. 2004; Paddison et al., 2004; Zheng et al., 2004; MacKeigan et al., 2005; Pelkmans et al., 2005).

Three siRNA-based screens have looked at cell survival (Kittler et al., 2004), induction of apoptosis (MacKeigan et al, 2005), or changes in endocytosis (Pelkmans et al., 2005). All of these siRNA screens have been done using parallel transfection of oligonucleotide siRNAs into appropriate cells. In the study from Kittler et al. (2004), the siRNAs were generated by RNAseIII cleavage of double strand RNA precursors made by transcription of cDNAs for the target genes. In the other cases the siRNAs were prepared synthetically and purchased from a commercial supplier. Three large-scale shRNA screens have been reported studying the NFkappaB signal transduction pathway (Zheng et al., 2004), proteasome function (Paddison et al., 2004), and p53 signaling (Berns et al., 2004). Of the reported siRNA and shRNA screens, some relied on pooling strategies and the others on parallel screens.

Comparisons of the various RNAi screens point to several interesting observations. The most striking observation is that in assays that should be largely overlapping, for example counting changes in cell number versus changes of cell viability or hits that induce apoptosis as a subset of hits that inhibit proliferation, there is only an overlap of 10% to 25% of the proteins scored as positives. In each case the follow up work suggests that a good proportion of the identified hits are real; however, this is less overlap than might have been expected. There are a number of reasons that may cause these

differences. The first is an informatics issue. Most of the studies have invoked in-house naming conventions for gene identification, and it is often difficult to compare results from one lab to another. The second reason may be due to the statistical methods used to determine what is considered a "positive". All of the reports have used replicate assays of the same DNA or virus preparation to establish average values and standard deviations. Third, inherent differences may exist between experimental approaches. Even simple changes such as how long an assay is extended before a final reading is taken can greatly influence the number and quality "positives." Finally, the shRNA or siRNA molecules that are currently available target different regions of a given mRNA, this may dramatically affect the level of knockdown and could therefore affect scoring. In sum, current screens may not be performed to saturation. It would be extremely valuable to conclusively determine the critical targets for a given class of proteins, for examples, kinases. Identification of such class of proteins would allow those in the field of drug discovery to engage in systematic identification of therapeutics in a manner which is previously unmatched in terms of both breadth and depth.

SUMMARY OF THE INVENTION

The invention features methods for identifying and utilizing kinases, and other proteins, whose down-regulation induces a desirable outcome. For example, the methods provided may be employed to identify kinases that are essential to proliferation in cancer cells. The methods for identifying kinases employ contacting cells with one or more shRNAs that target a particular protein and determining the effect of down-regulation.

The invention also features a method of identifying a therapeutic compound for treating cancer by assaying a candidate therapeutic compound for inhibition of the activity of a kinase selected from the group consisting of ANPB, RSK2, PAK6, NLK, CONT PACGFP, ULK4, DDR2, PDGFRB, TRRAP, EPHBl, TSSK2, CAMLCK, MST2, CONT TIE, CK2A2, TAKl, CONT KIT, JNK3-1937, CDKI l, ALK, NEK7, STK33, FYN, PAK3, PAK3,

CONT ERBB2, HER3, PKD2, CDKlO ₅ MAP3K8, PKNl, MYO3B, EPHB4, BRD2, CDK3, JNK2, HER3, PKD3, PITSLRE, PEK, CDK7, JNK3-, SLK, SLK, CONT BCR, PLKl, MELK, PCTAIRE, CKlE, CONT MTOR ₅ CDK9, PITSLRE, CLIKlL, JNK3, SURTK106, PKGl, ADCK4, PBK, TYRO3, LATS2, SGK2, ALK4, CAMK4, CKlE ₅ AMPKAl, CDK4, HIPK2, CLK3, FGFR3, BTK, LKBl, CDKlO, PDHK2, AAKl, CAMKlG, SGK495, MYTl, SRPK2, TNKl, HUNK ₅ MISR2, CKlE, HER4, PDGFRA, VRK3, EEF2K, CAMK2A, ROS, MAP2K1, RSK3, KHSl, ZC1/HGK, MET, TLKl, MAPAPK3, FER, BUBRl, PKD2, CAMKKl ₅ CDK9, CONT SCRAM, TRAD, RNASEL, IRR, CDK6, JNK3, VACAMKL, CONT 447, and CONT GFP, wherein the level of inhibition is indicative of the therapeutic efficacy of the candidate compound. The invention also features methods of drug screening wherein at least 5, 10, 15, 20, 30, 40, 50, 60, 70, 80, or 90 of the above kinases are used in parallel to assay for anti-cancer therapeutics.

The assaying, for example can include contacting the candidate therapeutic compound with the kinase protein in vitro or in vivo and measuring the activity of the kinase. Alternatively, the assaying includes contacting the therapeutic compound with a cell capable of transcribing the mRNA encoding the kinase and measuring the level of expression of the mRNA or other characteristics associated with a change in the level of biological activity of the encoded kinase or the level of the kinase so encoded. Assays may also be performed in animal models. Assays known in the art for monitoring kinase activity, transcription, translation, protein stability, and availability may all be used in the methods of the invention.

The invention further features a method of generating a database using data obtained from methods that include contacting cells expressing a class of mRNA, each member having known sequence, with one or more shRNAs specific to each sequence of the class; measuring the ability to proliferate of the cells, wherein a reduction in the ability identifies an mRNA as essential to proliferation; and creating a record in the database, wherein the record includes the identity of an mRNA of the class that is essential. A record may also

include other information on the mRNA (and the protein it encodes), such as the phenotypic effect of reducing the activity of the protein. For this method, by a class of mRNA is meant a class so grouped by the user, typically based on the function of the encoded proteins. Exemplary classes include kinases, receptors, phosphatases, and transcription factors.

The invention also features databases containing information on proteins, such as kinases, obtained as described herein.

By the genes "ANPB, RSK2, PAK6, NLK, CONT PACGFP, ULK4, DDR2, PDGFRB, TRRAP, EPHBl, TSSK2, CAMLCK, MST2, CONT TIE, CK2A2, TAKl, CONT KIT, JNK3-1937, CDKl 1, ALK ₅ NEK7, STK33, FYN, PAK3, PAK3, CONT ERBB2, HER3, PKD2, CDKlO, MAP3K8, PKNl, MYO3B, EPHB4, BRD2, CDK3, JNK2, HER3, PKD3, PITSLRE, PEK, CDK7, JNK3-, SLK, SLK, CONT BCR, PLKl, MELK, PCTAIRE, CKlE, CONT MTOR, CDK9, PITSLRE, CLIKlL, JNK3, SURTK106, PKGl, ADCK4, PBK, TYRO3, LATS2, SGK2, ALK4, CAMK4, CKlE, AMPKAl, CDK4, HIPK2, CLK3, FGFR3, BTK, LKBl, CDKlO, PDHK2, AAKl, CAMKlG, SGK495, MYTl, SRPK2, TNKl, HUNK, MISR2, CKlE, HER4, PDGFRA, VRK3, EEF2K, CAMK2A, ROS, MAP2K1, RSK3, KHSl, ZC1/HGK, MET, TLKl, MAPAPK3, FER, BUBRl, PKD2, CAMKKl, CDK9, CONT SCRAM, TRAD, RNASEL, IRR, CDK6, JNK3, VACAMKL, CONT 447, and CONT GFP" is meant the genes with the corresponding Genebank ID numbers as set forth in Table 2. Exemplary Genebank accession numbers for the human form of these genes are also set forth in Table 2. By the genes listed above is also meant a nucleic acid with at least 30%, 40%, 50%, 60%, 70%, 80%, 90%, 95%, or 99% percent sequence identity to any of the genes listed above as determined by the NCBI BLAST program. Additionally and alternatively, a gene listed above is defined as a nucleic acid that hybridizes under high stringency conditions to a nucleic acid of a gene listed above.

The "percent sequence identity" of two nucleic acid or polypeptide sequences can be readily calculated by known methods, including but not

limited to those described in Computational Molecular Biology, Lesk, A. M., ed., Oxford University Press, New York, 1988; Biocomputing: Informatics and Genome Projects, Smith, D. W., ed., Academic Press, New York, 1993; Computer Analysis of Sequence Data, Part I, Griffin, A. M., and Griffin, H. G., eds., Humana Press, New Jersey, 1994; Sequence Analysis in Molecular Biology, von Heinje, Academic Press, 1987', and Sequence Analysis Primer, Gribskov, and Devereux, eds., M. Stockton Press, New York, 1991; and Carillo and Lipman, SIAM J. Applied Math. 48:1073, 1988.

Methods to determine identity are available in publicly available computer programs. Computer program methods to determine identity between two sequences include, but are not limited to, the GCG program package (Devereux et al., Nucleic Acids Research 12:387, 1984), BLASTP, BLASTN, and FASTA (Altschul et al., J. MoI. Biol. 215:403, 1990). The well known Smith Waterman algorithm may also be used to determine identity. The BLAST program is publicly available from NCBI and other sources (BLAST Manual, Altschul, et al., NCBI NLM NIH Bethesda, Md. 20894). Searches can be performed in URLs such as the following: http ://www.ncbi.nlm.nih.gov/BLAST/unfmishedgenome.html; or http^/www.tigr.org/cgi-bin/BlastSearch/blast.cgi. These software programs match similar sequences by assigning degrees of homology to various substitutions, deletions, and other modifications. Conservative substitutions typically include substitutions within the following groups: glycine, alanine; valine, isoleucine, leucine; aspartic acid, glutamic acid, asparagine, glutamine; serine, threonine; lysine, arginine; and phenylalanine, tyrosine.

By "hybridize" is meant pair to form a double-stranded complex containing complementary paired nucleobase sequences, or portions thereof, under various conditions of stringency. (See, e.g., Wahl. and Berger, Methods Enzymol 152:399 (1987); Kimmel, Methods Enzymol 152:507 (1987))

By "hybridizes under high stringency conditions" is meant under conditions of stringent salt concentration, stringent temperature, or in the presence of formamide. For example, stringent salt concentration will

ordinarily be less than about 750 mM NaCl and 75 mM trisodium citrate, preferably less than about 500 mM NaCl and 50 mM trisodium citrate, and most preferably less than about 250 mM NaCl and 25 mM trisodium citrate. Low stringency hybridization can be obtained in the absence of organic solvent, e.g., formamide, while high stringency hybridization can be obtained in the presence of at least about 35% formamide, and most preferably at least about 50% formamide. Stringent temperature conditions will ordinarily include temperatures of at least about 30° C, more preferably of at least about 37° C, and most preferably of at least about 42° C. Varying additional parameters, such as hybridization time, the concentration of detergent, e.g., sodium dodecyl sulfate (SDS), and the inclusion or exclusion of carrier DNA, are well known to those skilled in the art. Various levels of stringency are accomplished by combining these various conditions as needed. In a preferred embodiment, hybridization will occur at 30° C in 750 mM NaCl, 75 mM trisodium citrate, and 1% SDS. In a more preferred embodiment, hybridization will occur at 37° C in 500 mM NaCl, 50 mM trisodium citrate, 1% SDS, 35% formamide, and 100 μg/ml denatured salmon sperm DNA (ssDNA). In a most preferred embodiment, hybridization will occur at 42° C in 250 mM NaCl, 25 mM trisodium citrate, 1% SDS, 50% formamide, and 200 μg/ml ssDNA. Useful variations on these conditions will be readily apparent to those skilled in the art.

For most applications, washing steps that follow hybridization will also vary in stringency. Wash stringency conditions can be defined by salt concentration and by temperature. As above, wash stringency can be increased by decreasing salt concentration or by increasing temperature. For example, stringent salt concentration for the wash steps will preferably be less than about 30 mM NaCl and 3 mM trisodium citrate, and most preferably less than about 15 mM NaCl and 1.5 mM trisodium citrate. Stringent temperature conditions for the wash steps will ordinarily include a temperature of at least about 25° C, more preferably of at least about 42° C, and most preferably of at least about 68° C. In a preferred embodiment, wash steps will occur at 25° C. in 30 mM

NaCl, 3 ITiM trisodium citrate, and 0.1% SDS. In a more preferred embodiment, wash steps will occur at 42° C. in 15 mM NaCl, 1.5 mM trisodium citrate, and 0.1% SDS. In a most preferred embodiment, wash steps will occur at 68° C. in 15 mM NaCl, 1.5 mM trisodium citrate, and 0.1% SDS. Additional variations on these conditions will be readily apparent to those skilled in the art. Hybridization techniques are well known to those skilled in the art and are described, for example, in Benton and Davis {Science 196:180 (1977)); Grunstein and Hogness {Proc Natl Acad Sci USA 72:3961 (1975)); Ausubel et al. {Current Protocols in Molecular Biology, Wiley Interscience, New York (2001)); Berger and Kimmel {Guide to Molecular Cloning Techniques, Academic Press, New York, (1987)); and Sambrook et al. {Molecular Cloning: A Laboratory Manual, Cold Spring Harbor Laboratory Press, New York). Preferably, hybridization occurs under physiological conditions. Typically, complementary nucleobases hybridize via hydrogen bonding, which may be Watson-Crick, Hoogsteen or reversed Hoogsteen hydrogen bonding, between complementary nucleobases. For example, adenine arid thymine are complementary nucleobases that pair through the formation of hydrogen bonds.

By "inhibition of kinase activity" is meant any reduction of the biological activity of the kinase attributable to any mechanism, including reduction in transcription, translation, or stability of the mRNA encoding the kinase or by binding, degrading, or otherwise inhibiting the enzymatic activity of the kinase or its substrate.

By "kinase activity" is meant the activity whereby an enzyme phosphorylates an amino-acid residue on a substrate (e.g., a protein, lipid, or carbohydrate substrate).

By a "compound," "candidate compound," or "factor" is meant a chemical, be it naturally occurring or artificially derived. Compounds may include, for example, peptides, polypeptides, synthetic organic molecules, naturally occurring organic molecules, nucleic acid molecules, and components or combinations thereof.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 is a graph showing the correlation between the relative growth of cells in two screening assays.

Figure 2 is a graph showing the correlation between the relative growth of cells in two screening assays.

Figure 3 is a diagram showing the overlap of shRNA hits in two different cell lines.

Figure 4 is a graph showing growth of cells transfected with virus containing shRNA constructs in the presence and absence of puromycin.

Figure 5 is a histogram showing the titer of viruses produced in uninfected and infected cells.

Figure 6 is a set of two histograms showing the number of cells containing the indicated relative DNA content. The left panel shows the DNA content of cells transfected with a luciferase construct and the right panel shows the DNA content of cells transfected with a shRNA construct.

Figure 7 (top panels) is a set of two histograms showing the growth rate of cells expressing both an shRNA construct against ErbB3 and a cDNA.

Figure 7 (bottom panels) is a set of photomicrographs of cells expressing both shRNA against ErbB3 and a cDNA encoding MAP3K14 (bottom left and bottom center panels), or control cells (bottom right panel).

Figure 8 is a diagram showing an experimental plan for identifying and characterizing essential kinases.

Figure 9 is a diagram showing follow up experiments for proteins identified in the screens of the invention.

Figure 10 is a photomicrograph showing MAP LC3 staining of FIELA cells.

Figure 11 is a photomicrograph showing MAP LC3 staining of HDELA cells.

Figure 12 is a photomicrograph showing SA-beta Gal staining of HELA cells.

Figure 13 (left panel) is a photomicrograph MAP LC3 staining of HELA cells; Figure 13 (right panel) is a photomicrograph of Activated Caspase-3 staining of HELA cells.

Figure 14 is a set of four histograms showing the number of cells containing the indicated relative DNA content in cells transfected with the indicated shRNA.

DETAILED DESCRIPTION OF THE INVENTION

In the post-genomic era, researchers can use the tools of molecular biology to identify specific proteins with roles in cancer cell growth and/or survival and then develop chemical compounds (drugs) that disrupt function of these proteins, with the idea that these drugs may be used as cancer therapies. One of the challenges of the approach is to first identify appropriate protein targets for specific cancer types. An ideal target protein might be one that is essential for viability in a cancer cell, but not essential for viability in healthy, non-cancerous cells in the same tissue or organ. In general, methods of the invention are employed:

1. To identify genes that encode proteins essential for viability and/or proliferation in specific cell types using an RNA interference (RNAi)-based approach, in which small hairpin RNAs (shRNAs) directed against specific gene targets and introduced into cells, e.g., via a lentiviral vector, cause gene- specific knock-down of mRNA and protein levels.

2. To parse the list of genes identified in (1) further by grouping the genes into broad categories that describe how disruption of the genes affects viability and/or proliferation. These modes of action include induction of cell death (apoptosis), induction of autophagy, and/or induction of cellular senescence in the presence of a shRNA that disrupts the gene target.

3. Identifying and developing drugs that disrupt the function of those target proteins identified in the screen as good candidates, e.g., based on the fact that they are essential in cancerous cells but not in corresponding normal cells.

4. To use the results of multiple iterations of the above steps to develop new rules for choosing a 'good' therapeutic target.

Although the following discussion focuses on kinases, any class of genes, e.g., those encoding proteins from given receptor class, phosphatases, and transcription factor families, could be tested and grouped in a similar manner using these methods

The "100 Hits" Essential Kinase Project

We have developed the following methods to identify drug targets. First, a set of genes essential for viability and/or proliferation of two different cell lines (HeLa and 293 T cells) were identified using a collection of more than 2,000 shRNAs directed against a set of 416 kinase-encoding genes (the shRNAs were developed by The RNAi Consortium). Next, the set of genes identified as essential in either cell line (the "100 hits") were used to screen normal (i.e., non-transformed) and cancerous (i.e., transformed) cell lines believed to be relevant to specific cancer types (for example, normal MCFlOA and tumorogenic MCF 7 cells originating from breast tissue). The screen may employ transfection of the cell lines and an alamar blue readout of proliferation/viability. Alternatively, screens may employ lentiviral transduction and alamar blue readout, followed by secondary assays, e.g., for apoptosis, autophagy, and senescence (anti-activated-Caspase-3 immunofluorescent detection, anti-LC3 immunofluorescent detection, and SA- beta-Gal assays, respectively). It will be appreciated that the readout may be generated using any number of systems known to those skilled in the art.

As mentioned, kinases shown to be essential in cancerous but not noncancerous cells from the same tissue type are considered better targets than kinases essential for viability in both cell types since the ideal therapeutic approach is to kill cancerous cells without affecting normal cells. In addition, there are at least two other ways in which the data from the "100 hits" screens could influence target selection. First, the accumulation of data on many different cell types should lead to identification of a sub-set of kinases essential in most, or many, cell types and, conversely, a sub-set of kinases that are essential only in a small sub-set of cells. Second, the accumulation of data on many different cell types allows one to group the action of kinases as generally acting through a particular mode or modes of action, or as acting in different modes in different cell types. Moreover, after several rounds of the approach (screen, categorize, identify and develop therapeutics), one may also learn that genes that act via one mode or modes have better potential for cancer therapies than genes that act in a different mode or modes (for example, a reasonable a priori guess is that proteins whose knock-down via RNAi leads to apoptosis may be better candidates for the development of targeted cancer therapies than those that work via senescence).

To study kinases on a proteomic scale, large repositories of shRNAs and cDNAs for the majority of the human kinases have been prepared. Using these tools, levels of individual kinases in cells can be raised by synthesis of proteins from appropriately designed cDNA expression vectors or reduced by expression of shRNAs specific for the targeted kinase. When these approaches are used in high-throughput parallel screens where protein levels are altered one gene at a time, an unbiased list of candidate genes that contribute to a cell's phenotype can be scored. With appropriate secondary assays and counter- screens, the direct involvement of these hits with the process under study can be confirmed. This approach is analogous to many types of gain-of-function or loss-of-function screens used in powerful genetic model systems and promises to expand greatly our knowledge of the cellular physiology of human cells in culture.

shRNA repositories and expression strategies.

One of the most exciting discoveries of recent years is RNA interference or RNAi (Fire et al., 1998; Bernstein et al, 2001; Elbashir et al., 2001; Hammond et al., 2001, Ketting et al, 2001). Many eukaryotic cells have enzymatic machinery that recognizes double strand RNA sequences, processes them to active short duplex sequences, and then uses them to modify gene expression. RNAi has been reviewed extensively (for example see Huppi et al., 2005, Tomari and Zamore, 2005). siRNA are short RNA duplexes typically between 19 and 27 nucleotides in length that are recognized by a multimeric protein complex known as the RNA-induced silencing complex (RISC). The RISC complex unwinds the duplex, uses one strand of the duplex for recognition of homologous sequences in mRNA, and then cleaves the niRNA. The outcome is the lowering of specific mRNA levels and the subsequent reduction of specific protein levels. Two methods are available to generate siRNAs. The more straightforward but expensive approach is to synthesize both strands of the siRNA, hybridize the two strands to form a duplex, and transfect the oligonucleotides into cells. Oligonucleotides transfect more readily than longer double strand sequences making this an efficient process in many cells. The other popular method is introducing an expression construct into cells that transcribes a short hairpin RNA (shRNA) where the duplexed sense and antisense RNA sequences are connected by a small loop. The shRNA is recognized by a cellular enzyme called Dicer that removes the hairpin and releases the siRNA, which in turn is processed by RISC.

RNAi is amenable to high throughput approaches and can be used to test large collections of siRNAs or shRNAs for cellular changes (Berns et al., 2004; Paddison et al., 2004; Zheng et al., 2004; Kittler et al., 2005; Mackeigan et al., 2005; Pelkmans et al., 2005). In the present invention, siRNAs are transfected into cells or shRNA-expressing vectors are transfected or transduced into cells, and phenotypic changes are measured. Because it may not be possible to predict which siRNA/shRNA sequences will most efficiently lead to mRNA

degradation, multiple siRNA/shRNA for each gene are often used in these screens. The higher number of constructs has led to the adopting of two general strategies. The RNAi' s can be added in a pool, and the RNA that is driving the change identified in subsequent deconvolution experiments, or each RNAi can be introduced separately in parallel, and tested individually. The types of cell- based assays that can be utilized are considerably more informative when using parallel strategies. Pooling selects for stronger phenotypes given that background RNAi' s can dilute out weaker phenotypes. With pooling, the assay is frequently limited to positive selections, and therefore deleterious outcomes cannot easily be scored. With parallel screening, both strong and weak phenotypes can be scored. Finally, when using parallel screening, no deconvolution steps are required, and the results of each RNAi can be recorded for comparison.

There is also a significant strategic decision about whether to use siRNA, shRNA, or both for high throughput screens. Our approach has been to develop and use shRNA strategies for screening. The reasons for this decision are straightforward and based on four main points. First, shRNA-expression vectors are prepared from a plasmid resource and therefore are in an inexhaustible supply. Second, the plasmids are prepared using standard lab methods, making them less expensive to use than commercial siRNAs. For high throughput analysis, this is an important parameter when considering the supply costs encountered with a range of screens and the needed repetitions to establish statistical significance. Third, the shRNAs that may be used in these studies include those in a lentivirus backbone which can thus be used to transduce almost all cell types, whether growing or arrested. This eliminates the concerns which accompany working with cells or under conditions where transfection of oligonucleotides may fail. Fourth, stable viral integration of the shRNA enables protein down-regulation over the life span of the cell. This is especially useful when protein turnover is slow.

There are three large academic repositories of shRNA, one prepared by the Bernards lab (Berns et al., 2004), one by the Harmon and Elledge labs

(Paddison et al., 2004), and one by the RNAi Consortium (TRC). The TRC shRNA vectors are based on a lentiviral backbone, allowing infection of essentially all human cells. Although this vector is easily used as a lentivirus, it also works quite well as a plasmid when transfected into cells, albeit with the limitations of transfection.

It should also be noted that RNAi has technical limitations of its own. It is now well documented that siRNAs and shRNAs often induce off-target effects (Jackson et al., 2003; Jackson and Linsley, 2004). These off-target effects are thought to come about by the unexpected recognition of mRNAs by RISC loaded with the siRNA. No clear rules exist on how to predict these off- target effects, and this emphasizes the value of validation and follow-up experimentation in assigning a functional role to a protein based on a siRNA/shRNA hit.

The kinoine

Kinases are important for reasons including: (1) their established roles in many important cell pathways, (2) the extensive knowledge of their structure and function, and (3) the ability to find and design small molecular inhibitors that specifically block kinase activity. In addition, although there has been extensive work in the study of kinases, there remains a great deal which is unknown or poorly understood. One hundred one kinases in our current target list have never been mentioned in a refereed publication, and 260 are discussed in 5 or fewer publications.

Although most is known about the protein kinases, the methods of the invention are applicable to all kinases, including the lipid, nucleotide, and carbohydrate kinases. Non-protein kinases, like the protein kinases, also help control cell metabolism. Any splice isoform of a kinase may be assayed.

Transfections with shRNAs can be used to identify essential kinases. shRNAs to 434 kinases (on average 4.5 shRNAs per kinase) were individually transfected into both 293T and HeLa cells. The DNA was

introduced using a standard lipid carrier (Fugene), and the cells transferred to drug selection at 24 hrs. At 4 or 5 days, the number of viable cells was measured by Alamar blue assay. The Alamar blue method was chosen after comparison with tetrazolium salt assays, such as MTT or MTS, or other more specific tests for individual phenotypes, e.g. apoptosis measured by DNA fragmentation, because of its robust nature, sensitivity, lack of toxicity, ease of use, low cost, and ability to identify a number of interesting classes of response.

The HeLa and 203T lines were chosen for two reasons. First, they are among the most prevalent cell lines used for mammalian cell experimentation. Second, both of these cell lines are known to be easily transfectable. In the development of these assays HeLa transfections have now been done three times and the 293 T transfections have been done four times to establish methods for analysis and to test for reproducibility. Within one experiment, each transfection is done in triplicate, and the results are averaged. Figure 1 compares the values recorded for the second and third trials in HeLa cells. The reproducibility of this assay is high, as demonstrated by a correlation coefficient of 0.897, between these two data sets.

The shRNAs that produced greater than a 50% reduction in cell growth over the 5 -day assay period in two or more independent screens were classified as potential growth inhibitors; a representative data set blown up from Figure 1 is shown in Figure 2. The vast majority of these kinases has not been studied in detail.

Similar sets of hits have been prepared from the comparisons of independent screens done in 293T cells. The HeLa and 293T hits generated from shRNAs to this set of kinases have been used for the subsequent validation and characterization experiments as described below. Comparisons of the hits in 293.T and HeLa cells allow us to draw several additional conclusions. Before beginning these experiments it was unclear how the collection of essential kinases would compare from one cell line to another. The comparison between hits in HeLa and 293 T show remarkable differences

(Figure 3). Notably, only 20 shRNA hits were found to be common to the two lines. These results indicate that some cells in culture will show clear differences in metabolic events that are most sensitive to shRNA-mediated blocks.

Validation of the shRNA hits.

To show that individual shRNAs could be validated as the physiological cause of the loss of cell proliferation or survival, several experiments were performed. First, we reasoned that scoring multiple shRNAs for the same kinase was a strong indication that the intended target mRNA was knocked down. In some initial screens, multiple hairpins for the same kinase were scored and in other initial screens one hairpin was scored, requiring the synthesis of new shRNAs. Second, by utilizing shRNAs that target untranslated regions, a rescue experiment could be set up by expressing the ORF from the corresponding cDNA vector in our kinase collection. This was possible because all of the ORFs within this repository contain the coding region from start to stop and lack the native untranslated regions. Third, changes in the mRNA levels were measured by standard real-time PCR methods. Our work in this area has shown that all three methods are valuable, and the order in which they are presented directly corresponds with the ease of performing the experiments in high throughput. Success in the first two has strongly predicted that mRNA levels are down-regulated.

Infection.

The shRNA vector backbone was designed by the TRC to allow lentivirus production. This is a useful viral vector system that allows infection of almost every cell line, whether proliferating or post-mitotic. One of the underlying reasons we began screening essential kinases by transfection was the concern that growth-inhibiting hairpins, targeting essential kinases, might kill the 293 T viral packaging cells before adequate virus was harvested. This concern proved valid. Some of the strongest hits from the 293T set were used to test lentiviral production in 293 T cells, and in a fraction of the cases the

resulting supernatants did not contain infectious virus particles. The "no viral" phenotype was determined by the cells ability to grow in the absence but not the presence of puromycin (Figure 4). Approximately 10% of the essential kinase hits identified by transfection were able to block lentivirus production. There are three possible explanations for the "no virus" phenotype, including (1) shRNA expression may kill or block cells at a stage in which lentiviral production cannot proceed; (2) the targeted kinase may be important for a key element of the lentivirus life cycle; and (3) the quality of the DNA template used to initiate viral production was poor. The third possibility has been ruled out for two reasons. First, multiple carefully quantified DNA preparations have yielded the same result, and second, the same DNA preparations work to inhibit cell proliferation when introduced by transfection in the presence of puromycin. We have now identified several possible methods to allow production of the missing lentiviruses from these sets, including making the missing lentivirus stocks in mouse cells where the shRNAs do not recognize the targeted mRNAs, producing virus in cells in which the RNAi machinery is crippled, or making virus in cells in which the target mRNAs are greatly increased by expression from a cDNA vector from our collection.

The same experiments described above provide another measure of reproducibility by allowing comparison of the results from transfection, (where the hits were originally identified), and the results from transduction. Most of the hits identified by transfection gave identical results by transduction. By viral infection, cell growth was diminished in the absence and presence of puromycin, suggesting viral production was very efficient and arguing that the slower growth was due to the shRNA affecting its target. When the hairpins are ranked by their effectiveness for killing, most show either strong or weak growth inhibition. Five of the tested hairpins did not appear to inhibit growth. We attribute this lack of effect to a false positive score in the initial transfection experiments.

Determination of viral titer

For experiments involving delivery of shRNAs by transduction, knowledge of viral titers is helpful for two main reasons: low viral titers can lead to false negative results, and unusually high titers can produce false positives. For example, excessive numbers of integrated proviruses resulting from a high multiplicity of infection (MOI) can slow the growth of cells non- specifically.

Standard methodologies for titering virus are laborious and slow. The number of infectious particles per ml can be determined by performing serial dilutions of the viral stock followed by infection of NIH3T3 cells. Twenty-four hours after infection, cells are incubated with puromycin and over time singly dispersed cells grow into resistant colonies that can be visualized with crystal violet staining and counted. Alternatively, GFP-expressing viruses can be used, and the ratio of GFP-positive and GFP-negative infected cells as determined by flow cytometry used to calculate titer.

We have adopted a HT method for assessing relative viral titers, either directly from virus-containing producer cell supernatants or in infected cells to 96-well format. This method detects the signal from DNA oligonucleotides hybridized to a target RNA rather than the target RNA itself. We chose as our target RNA the puromycin acetyl-transferase gene that serves as a selectable marker on all of our lentiviras constructs and one set of our Moloney cDNA expression constructs. Figure 5 demonstrates the sensitivity and reproducibility of assessing viral titers in lentivirus-infected NIH3T3 cells by this method; error bars show the average of triplicate values. We have obtained similar results when determining the relative numbers of both Moloney and lentivirus particles directly, using only 0.1-10 microliters of producer cell supernatant.

Arrested cells can be characterized for the type of block

The next step for analyzing the shRNA hits is to determine the aspect of cellular physiology that is affected. Our strategy is to use flow cytometry analysis to first identify which shRNAs can cause cell cycle blocks. Figure 6

shows the results of one round of flow cytometry analysis with a test set of 30 shRNAs identified in the primary screens (Figure 6). We find that greater than 25% of the kinase shRNA hits when transfected into cells give a specific cell cycle block, and about 10% lead to increased apoptosis. The G2 cell cycle block shown in Figure 5 is caused by an shRNA to PDGFRb, a kinase not previously known to act in G2.

cDNA/shRNA screens can be used to find downstream events that rescue blocks

The methods of the invention may be used to identify the cellular context in which newly identified kinases function by establishing and ordering them in signal transduction pathways. One method to determine these types of relationships is to look for genetic interactions. This may be done by employing the equivalent of genetic modifier screens using the shRNA-induced phenotype as an initial alteration. A cDNA expression construct or a second shRNA is then introduced to permit a search for genes that suppress or enhance the initial shRNA-dependent block.

In one experiment, an shRNA targeting ErbB3, which was identified in HeLa cells as an essential kinase, was packaged as a lentivirus and used to infect HeLa cells where it induced a cell proliferation defect. This phenotype was modified by infection of cells with individual Moloney virus vectors expressing kinases. In initial experiments it was determined that the Moloney virus infections were best done before addition of the ErbB3 shRNA lentivirus. The Moloney virus carries the blasticidin resistance gene and 24 hrs after infection, blasticidin was added. Forty-eight hours post Moloney virus infection, cells were infected with the lentivirus carrying an ErbB3 shRNA and containing a different selectable marker, a puromycin resistance gene. Puromycin selection started 24 hrs after infection (now 72 hrs after the initial Moloney virus infection), and Alamar blue assays were performed after 3 days (144 hrs after the initial Moloney virus infection). In addition, cell images were collected by phase contrast microscopy. Cell proliferation was effectively

blocked in all controls. The strongest rescue of the cell block imposed by either of two different shRNAs for ErbB3 was seen by the vector that expressed MAP3K14 (Figure 7). This kinase is known to act downstream of the EGF family of receptor tyrosine kinases (Habib et al., 2001).

A second type of genetic interaction screen was performed to examine whether similar epistasis studies can be done by transfection. The cell proliferation block imposed by transfection of CDKlO kinase was examined. An shRNA for CDKlO that gives a strong block was co-transfected individually with kinase cDNA expression vectors. The same backbone vector was used for the Moloney virus production discussed in the paragraph above, but used here as simple transfection vectors. Twenty- four hours post- transfection, cells were selected with puromycin to ensure that the shRNA causing growth inhibition was present, and cell number and viability was scored 4 days later. Rescue of the proliferation block imposed by either of two independent shRNAs for CDKlO was seen by the vectors that expressed TOPK, MAP2K2, PFKM, PTK2B, PGK2, FN3K, CHLK, PRPS2, DGKA, and SGKL in each of three independent experiments.

Experimental Design

Completion of sequencing the human genome has resulted in the generation of a growing list of genes that encompass the protein coding potential of the human genome. Definition of the protein coding regions has in turn allowed the development of reagents to study the human proteome in its entirety. As described in the Background section above, advances in high- throughput cloning of open reading frames and shRNA vectors allows the construction of useful repositories that can be used to alter the levels of each protein in a directed manner. With these resources, mammalian cells in culture can be treated essentially as genetically tractable systems. At full capacity we seek to be able to raise or lower the levels of each human protein in individual microplate wells using cell-based assays in order to examine most of cell metabolism in a highly parallel fashion.

Figures 8 and 9 outline the process for identifying kinase targets. First, shRNA expression constructs are used to determine which kinases are essential for survival and proliferation in a small set of commonly used tissue culture cell lines representing both highly transformed and more normal cell lines. Second, the mechanism of action of each of these essential kinases is classified, and those kinases showing a stage-specific cell cycle block when down- regulated are selected for more study. The proteins that have roles in cell death, in regulating the rate of cell proliferation, or in processes that lead to arrests, irrespective of cell cycle position, are also be identified. Third, the context of action of the cell cycle-specific kinases is examined by determining the genetic interactions of the selected kinase with other kinases and by biochemical studies.

High throughput shRNA screens to identify essential kinases Genes can be divided into two functional classes — those genes that are essential for proliferation and survival and those genes that are not. Genes can be assigned to either group following assays for cell viability after shRNA expression (Figure 9). Genes whose functions are essential for proliferation or survival can be tentatively identified when shRNA expression leads to inhibition of cell proliferation or loss of viable cells. Cell viability and number can be measured quantitatively in high throughput with agents such as Alamar Blue or tetrazolium salts, both of which measure cellular oxidative metabolism. As with all screens of this type, classification is only tentative and will need to be confirmed by validation studies described below. At this stage a protein is tentatively classified as essential if any one of the five shRNAs for each protein leads to decreased proliferation or survival. False positives will be removed at later stages.

Complete transfection of shRNA for all kinases in HeLa and 293T.

To establish an initial data set of essential kinases, HeLa and 293 T cells were transfected with shRNAs targeting 434 human protein kinases. There were approximately 4.5 shRNA constructs per kinase, and each hairpin was

assayed independently. Transfected cells were selected with puromycin, allowed to grow for several days, and cultures scored for viability by an Alamar blue assay. Varying the experimental parameters showed that assays carried out 5 days after transfection gave the best balance between maximizing the effects of introduced shRNAs and sensitivity of the Alamar Blue assay. For each experiment, the transfections were done in triplicate, and the entire experiment repeated either three times for HeLa or four times for 293 T. These comparisons have identified 60 kinases (67 shRNAs) that are essential for growth in HeLa cells and 48 kinases (53 shRNAs) for 293T. Twenty of these kinases were essential in both HeLa and 293T. Loss of each of these kinases leads to robust and reproducible blocks to proliferation or survival. Kinases are scored as hits only when they inhibit growth 50% or more relative to all other shRNAs in multiple independent experiments. For 10 kinases, two independent shRNAs that target different regions of the mRNA scored as positive, and for one kinase three independent shRNAs scored to inhibit cell growth. We consider two or more independent hairpin hits to be strong candidates for essential kinases.

Infection.

In addition to transfection, shRNAs may be introduced through infection, e.g., via lentivrus. The group of kinases that are needed for lentivirus production might be of interest in their own right. Some may generate a block that is important for cell survival, but others might be specific for steps in the virus life cycle.

Screen HeLa, 293T, WI38, MCFlOA, A549, and MCF7 cells by transduction.

Once the choice of production methods for the entire lentivirus collection is made, a small set of reference cell lines is used to generate a list of essential kinases. In addition to HeLa and 293 T cells, this set may include a pair of commonly used cells, e.g., WB 8 and MCFlOA, representing a nonpathological phenotype. WB 8 cells are normal human diploid lung

fibroblast cells that undergo senescence through passage in culture. MCFlOA cells are a breast epithelial line commonly used for many comparative experiments with more highly transformed breast cancer cell lines. In addition, a pair of cell lines, A549 and MCF7, are used for screening to represent more typical tumor lines. A549 and MCF7 cells were chosen because they are good representative carcinoma and adenocarcinoma cell lines, respectively, that have been analyzed for mRNA profiling and sequencing of protein tyrosine kinase mutations. A549 are from a lung tumor and MCF7 from a breast tumor. HeLa and 293T cells are included to compare the findings from transfections and infections, and they will provide a direct comparison with the shRNA lentivirus results in the other four lines. It will be appreciated that additional cell lines may be employed.

The search for essential kinases may also be done lines in a fashion analogous to transfection, except that shRNAs will be introduced by viral transduction. The six cell lines described, or others, may be utilized. Cells are infected independently with each shRNA-expressing lentivirus, and then cultured for 5 days when differences in cell viability and cell number are determined by Alamar blue assays. Each infection will be done in triplicate, and mean values and standard deviations will be used to record the response to each hairpin.

Determination of the type of block. shRNAs that lead to decreases in cell viability or proliferation are classified further. There are at least three categories under which these hits are classified— arrest in a specific stage of the cell cycle, increases in cell death, or a general arrest irrespective of cell cycle stage. These three categories can be initially distinguished by, for example, flow cytometry, as described below. Other classes and phenotypes may appear and are readily used in the classified schemes

Some of the shRNA-induced cell proliferation and viability blocks will result in cells arrested in specific stages of the cell cycle. These correspond to

many cell cycle regulating proteins whose role is required once and only once during the cell cycle. Such experiments have helped establish the technical approaches required here. Treated cells are typically assayed for changes in their cell cycle profiles by flow cytometry after fixation and propidium iodide staining. Requirement for progression through a particular stage is detected by an increase in percentage of cells in one stage compared to controls. Although flow cytometric analysis is not generally considered a high throughput method, it can be done with reasonable efficiency and is considered a gold standard of cell cycle analysis. This approach is preferred, as the assay is robust and the results are readily characterized.

The second class of essential kinases includes those that stop or greatly slow cell division, but do so without specificity for cell cycle position. These include kinases involved in energy regulation, general transcription, or other events that are important throughout a division cycle. Such kinases are identified by virture of flow cytometry profiles that resemble the profiles untreated cell cultures.

The final class of essential kinases identified include those that induce cell death. Most types of cell death including all types of apoptosis can be initially identified by flow cytometry profiles as cell bodies with chromosome numbers below the Gl population of the untreated cultures. All of the hits from the essential kinase screens are tested for cell death by histone H3 staining for chromosome condensation that accompanies cell death. Further apoptotic assays which may be used are known in the art and include assays for caspase activation and DNA breaks.

Validate shRNA knockdowns that give stage-specific cell cycle block. shRNA hits may also be validated to determine whether the observed phenotype is the result of the kinase's action on its intended gene target or an off-target effect. A traditional approach for validation is the measurement of changes in mRNA and protein levels using quantitative RT-PCR (for example,

TaqMan) and western blotting assays, respectively. Alternatively, validation of these hits may be done by establishing similar phenotypes with a second independent shRNA (or siRNA). Having two distinct shRNAs that target the same transcript generating the same phenotype is a clear indicator that the loss the targeted mRNA is responsible for the phenotype. Whenever possible a second shRNA is targeted to the non-coding regions of the mRNA to enable rescue experiments described below.

If identical phenotypes are found with the second shRNA, then validation can be completed by several methods. When an RNAi to the non- coding regions of the protein is available, a preferred method is to rescue the phenotype by expressing the protein from an available cDNA expression libraries. This can be done on a large number of shRNA hits in a high throughput manner. A second manner of rescue is to make a mutant kinase cDNA where the mutation destroys or reduces the recognition by the shRNA. Either approach is sufficient to establish validation — two shRNAs to different regions of the mRNA both independently inhibit growth, and this proliferation defect can be overcome by over-expression of the targeted protein. If the second shRNA is not in the untranslated region or is not readily available, direct measurements in mRNA levels will be performed on the single hairpin hit using TaqMan analysis. Where available, antibodies to the targeted protein are also used to determine changes in protein levels by western blotting. The combination of these approaches provides multiple levels of validation without undue efforts.

Establish biochemical characterization of stage-specific arrests.

Stage-specific blocks are characterized in more detail, in particular to determine as accurately as possible the precise point at which the block has occurred and find any unusual features of the block. This is done both biochemically and immunochemically by examining other cell cycle events associated with the arrest phenotype.

The state of chromosomal DNA in the blocked cells is further characterized. For further characterization of putative Gl or S phase blocks, cells are, for example, labeled with 5-bromo-2'-deoxyuridine (BrdU), which is incorporated into DNA and identifies cells replicating their DNA in S phase. Here we analyze BrdU and PI staining by flow cytometry and determine whether the block is Gl, early S, or late S phase. As for G2/M, cell cycle analysis using flow cytometry and PI staining clusters G2 blocks with mitotic blocks. To distinguish these different blocks and fully classify the mitotic arrest into prophase, metaphase, anaphase, or telophase, fluorescence microscopy are performed using the nuclear stain Hoechst in combination with a phosphorylation-specific antibody against histone H3 conjugated to fluorescein. Phosphorylation-specific antibodies to histone H3 detect cells in late G2 (speckled pattern) and mitosis, and is a useful marker for mitosis and chromosome condensation. For example, G2 and mitotic arrests can be characterized by an overall increase in mitotic index (% cells in mitosis) coincident with a homogenous nuclear morphology (all speckled G2, all in prophase, all in metaphase, etc.) showing mitosis stage specific blocks. Blocks in telophase and cytokinesis defects are monitored by including standard counterstains for microtubules or actin.

The activation state of cell cycle-dependent enzymes such as the cyclin- dependent kinases is examined in cells blocked with each validated shRNA. Activation of cdk2 is a late Gl event, activation of cdk4 (if present) indicates a mid-to-late Gl phase has been passed, and cdc2 activation is a late G2 to M event. The activity of other kinases such as the PLK and aurora kinases can also help characterize G2 or M phase blocks. These kinase activity assays are conveniently done by immunoprecipitation with specific antibodies from shRNA treated cells followed by an in vitro kinase assay with the addition of appropriate kinase substrate. Alternatively, the kinase activity can be monitored by microscopy and immunofluorescence with a phosphospecific antibody that is conjugated to a fluorochrome. Moreover, in G2 there are important phosphatases whose activity is carefully controlled by cell cycle

position. The activities of G2-specific phosphatases can be monitored in extracts from shRNA-treated cells with antibodies specific to the appropriate phosphorylated substrates.

Many stage-specific transcription events can help characterize the specific cell cycle blocks. The expression of several cyclins has been well characterized and can provide useful cell cycle landmarks. In addition, other cell cycle specific transcripts have been found in recent years by the use of expression profiling; the exact candidates that are employed will depend upon the stages of cell cycle blocks that are obtained. This can be measured by synchronizing cells and testing the timing of gene activation by standard quantitative measure of RNA levels. Synchronizing cells can be done by a number of methods depending on cell type.

Signaling pathways of essential kinases necessary for cell cycle progression

The experiments described above result in the generation of a list of essential kinases whose depletion produces a stage-specific cell cycle block. Using approaches based in principle on genetic modifier screens, the roles of these kinases is compared with each other to establish groups of kinases that may function together in a particular process and which, in some cases, may form some portion of a signaling pathway.

These results demonstrate that the equivalent to classic genetic modifier screens can be done in mammalian cells, and proteins that act downstream of the point of inhibition can be identified.

Taken together these screens provide a simpler and faster use of classic full genetic systems. The efficiency is due, in part, to the fact that the rescue is imparted by known cDNAs or shRNAs, and, therefore, no mapping, cloning, or sequencing is required.

It would be appreciated that other specific protocols to characterize the essential kinases are known in the art and may be employed in the scheme for

characterization we provide. For example, biochemical studies of the essential kinases can be used.

Overcome the cell cycle block imposed by the down-regulation of an essential kinase.

The blocks imposed by these shRNAs are challenged by expression of a protein kinase from the cDNA expression vectors described above. The experimental protocol is to infect cells with the shRNA-expressing lentivirus (marked with the puromycin resistance gene) and a cDNA-expressing Moloney virus (marked with the blasticidin resistance gene) and select with both puromycin and blasticidin. Our results indicate that the best rescues occur when the cDNA-expression is established prior to the infection with the shRNA-expression lentivirus. Cell proliferation is measured by Alamar blue assays. While subsets of the kinase cDNA expression libraries are used (for example, only those that correspond to the shRNA hits), the entire kinase cDNA repository only numbers about 500. Therefore, assaying with all of the cDNA expression vectors is not difficult. However, one may predict that not every shRNA block will be rescued by cDNA expression. Some of the kinases targeted by the shRNA will act at points where no downstream kinases are used or the point of action may be at a required branch point where multiple required downstream effectors are found and a single kinase may not be sufficient. In general, we anticipate that these screens will identify downstream positively acting cell cycle regulators or perhaps in rare cases identify an independent pathway that can compensate for the loss of the shRNA-targeted pathway. Ordering the action of these and other regulators will be possible, and the relevant experiments are described below.

Rescue by shRNA expression is also possible and will identify downstream negative regulators. All of the shRNA constructs have been screened for their ability to block cell cycle but not for their ability to overcome blocks by other shRNAs.

Epistasis and biochemical analysis.

After defining a group of kinases that rescue the block imposed by treating cells with shRNA, determination of the order of action for the proteins in the pathway is made and confirmed with biochemical studies.

From the work above we have chosen two essential kinases and two sets of cDNAs and shRNAs that rescue the loss of the essential kinase ("rescue kinases") for further analysis as archetypal. For our lead targets, we have established a matching set of cDNAs and shRNAs. Another version of the protein kinases that is helpful in these experiments is a kinase-dead version of the kinase. The available mutants change the conserved lysine in the ATP- binding loop to methionine. The other mutant that is planned is changing the conserved aspartic acid, phenylalanine, glycine sequence (DFG) in the active site to asparagine, phenylalanine, glycine (RFG). These mutants may be used to study the action of the chosen pathways.

This analysis is divided into two phases. One line of experimentation utilizes the information generated in Figure 9, where the initial shRNA- dependent, stage-specific cell cycle block was characterized in biochemical and immunocytochemical detail. Studies that have been shown to provide informative markers, whether monitoring stage-specific transcription or phosphorylation events, or aspects of chromosome morphology or cytoskeletal architecture, are now repeated in cells treated with shRNA alone, shRNA plus rescuing construct, and rescuing construct alone. By these combinations of analyses, we establish an order of events at the level of cellular physiology that, in turn, enables us to order the sequence of actions of the interacting proteins. For example, an shRNA may cause growth inhibition with the loss of phosphorylation of a well-known substrate P of a certain cdk. We may find that one rescuing kinase X restores the activity of this cdk, thereby rescuing the cell cycle block. Alternatively, we may find a second rescuing kinase, kinase Y, rescues the block without the activation of the cdk nor phosphorylation of P. This suggests that Y acts downstream of X. A second rescuing kinase Z may rescue with phosphorylation of protein P but without cdk activity, suggesting

that Z acts downstream of X, and further that the well-known substrate of this cdk may also be the substrate of a second kinase, possibly even kinase Z itself. Any feature of the arrested cell identified in Figure 9 above can be studied in this way.

One of the biochemical markers that is the most helpful in ordering these new pathways is the phosphorylation of the kinases themselves. It is common that kinases are regulated by phosphorylation events, and since this is their enzymatic role, the patterns of phosphorylation promise to be a powerful method to order the activity of these proteins. These experiments are based on the same logic as above but here the readout will phosphorylation of the kinase itself.

The basal level of phosphorylation of a rescuing kinases is established first. Cells are infected with the kinase as a fusion protein (both amino- and carboxy-terminal fusion proteins with a number of tags are available for optimization) of the kinase dead version of the kinase, and then the tagged protein will be precipitated from lysates of the cells and a phosphopeptide tryptic map of the protein established by mass spectroscopy. Changes in the pattern of phosphorylation are then examined by co-infecting with the cDNAs or shRNAs for the rescue kinases in the set. Since the rescuing kinases were selected by their ability to overcome the initial cell cycle block, the action of these kinases as studied is robust. Rescue kinases, either over-expressed by cDNAs or down-regulated by shRNAs, are tentatively classified as working upstream of given kinase if they induce changes in the phosphorylation pattern. Any relationships detected in these experiments are considered putative ordering of the pathway. These types of kinase and downstream substrate tests are then carried out either in the presence or absence of the original blocking shRNA or in stable or transient assays.

Examine cellular metabolism by large-scale shRNA screens.

The analysis of kinase function using shRNAs provides a starting point to examine their roles in cellular events such as proliferation and survival.

However, the same technical approaches developed here for 650 kinases can be expanded to studies of other protein families and eventually to large proteome- scale studies. While shKNA-induced loss of a protein from cells is a dramatic change, these types of larger-scale screens tentatively classify protein function and serve as useful starting points for more detailed biochemical studies.

Database Development

The methods described herein may also be employed to generate a database containing the identify of proteins, whose down-regulation produces a particular result, either desirable or undesirable.

Drug Screening

AU studies that identify proteins that are essential for cell proliferation or survival have the potential to generate putative targets for drug discovery. While shRNA assays are a surrogate for drug action, the lists of kinases that are essential for tumor cell growth may be examined for possible new drug development targets. Examples of desirable characteristics would be kinases for use as a development target are, for example, that the kinase is required in tumor cells but not in non-pathological cells and that the kinase has a role in cellular proliferation which was not previously appreciated.

The activities of target genes identified by assaying with shRNA is exploited to identify candidate therapeutic compounds for cancer or other hyperproliferative disorders, e.g., to inhibit tumor growth, to inhibit angiogenesis, to decrease inflammation associated with a lymphoproliferative disorder, to inhibit graft rejection, or neurological damage due to tissue repair. Other hyperproliferative disorders include atherosclerosis, graft coronary vascular disease after transplantation, vein graft stenosis, peri-anastomatic prosthetic graft stenosis, restenosis after angioplasty or stent placement, psoriasis, and endometriosis. Tumors of interest include carcinomas, e.g. colon, duodenal, prostate, breast, ovarian, melanoma, ductal, hepatic, pancreatic, renal, endometrial, stomach, dysplastic oral mucosa, polyposis, invasive oral cancer, non-small cell lung carcinoma, transitional and squamous

cell urinary carcinoma etc.; neurological malignancies, e.g. neuroblastoma, gliomas, etc.; hematological malignancies, e.g. childhood acute leukemia, non- Hodgkin's lymphomas, chronic lymphocytic leukemia, malignant cutaneous T- cells, mycosis fungoides, non-MF cutaneous T-cell lymphoma, lymphomatoid papulosis, T-cell rich cutaneous lymphoid hyperplasia, bullous pemphigoid, discoid lupus erythematosus, and lichen planus.

Cancers of interest include breast cancers, (e.g., ductal carcinoma in situ, infiltrating (or invasive) ductal carcinoma (IDC), infiltrating (or invasive) lobular carcinoma (ILC)), non-small cell lung carcinoma including epidermoid carcinoma (also called squamous cell carcinoma), adenocarcinoma , large cell carcinoma, carcinoid, cylindroma, mucoepidermoid, malignant mesothelioma, and skin cancer (e.g., melanoma).

Methods for identifying therapeutic compounds that inhibit the activity of a kinase typically detect the level of expression of the kinase or its activity. Essentially, any compound that reduces the activity of a kinase that is essential to proliferation in hyperproliferative cells is of interest. Mechanisms for reducing activity include reducing transcription, translation, or post- transcriptional alteration of the kinase, and inhibiting the binding of the kinase substrate, e.g., by competitive inhibition or covalent modification of the kinase. For screening purposes, the chemical identity of the candidate therapeutic compounds is not critical. Compounds may be randomly screened for activity, or rational drug design may be employed based on the nucleic acid sequence encoding the kinase or the structure of the protein.

Appropriate assays for kinase activities are known in the art, e.g., Davies et al. Biochem. J. 2000, 351 :95, Bain et al. Biochem. J. 2003, 371 :199, and Copeland Anal. Biochem. 2003, 320:1. For example, therapeutic compounds and kinases may be labeled with FRET moieties to indicate binding, or the rate or amount of phosphorylation of a substrate for the kinase may be monitored. Substrates for a kinase, if unknown, may be identified by methods known in the art, e.g., highthroughput screening of random peptide sequences.

Example 1.

The methods described herein have been employed to identify kinases that are essential for proliferation in a variety of cells. The results of these experiments are tabulated in Table 1.

The "Gene" column lists the gene name. In parentheses is the number of "mode of action" assays in which inhibition of the expression of the indicated gene scored positive. "Mode of action" assays include, for example, autophagy assays (e.g., Figures 10 and 11) SA-beta Gal assays (e.g., Figure 12) and apoptotic assays (e.g., Figures 13 and 14).

In the columns labeled 786, HeLa, MCF7, 293T, and MCFlOA, the number in these columns show relative survival (from the alamar blue colorometric assay readout). A higher number means better survival, and conversely, a low score means that the effect of the shRNA was to reduce viability and/or proliferation. The latter identify putative essential kinases.

Shading within a "PUBS" column indicates the number of peer-reviewed journals associated with the indicated gene (i.e., greater than 300, fewer than 10, or fewer than 100).

The "normal/tumor" column indicates the effect of inhibition of expression of the indicated gene on the survival of cancer cells relative to the survival of normal cells under the same conditions (i.e., the ratio of survival of the MCFlOA (normal) and MCF7 (cancerous) cell lines.) A value of greater than one indicates that inhibition of the expression of the indicated gene selectively decreases the survival of cancer cells. Shaded entries indicate the greatest selective effect.

The genes in Table 1 can be divided into three groups, indicated in Table 2.

TABLE 2

91807 MLCK CAMLCK NM 003016

140469 MYO3B MYO3B NM 012433

140609 NEK7 NEK7 NM 003016

Category 2

PKN 1 PKN 1 NM 002741

149420 PDIK1 L CLIK1 L NM 001358

6046 BRD2 BRD2 NM 006802

7301 TYRO3 TYRO3 NM 004660

2050 EPHB4 EPHB4 NM 012426

5592 PRKG1 PKG 1 NM 003016

1022 CDK7 CDK7 NM 012311

1025 CDK9 CDK9 NM 005850

6041 RNASEL RNASEL NM 006711

701 BUB1 B BUBR1 NM 012433

984 CDC2L1 PITSLRE NM 003016

91 ACVR1 B ALK4 NM 003016

8295 TRRAP TRRAP NM 001356

8558 CDK10 CDK10 NM 007165

1018 CDK3 CDK3 NM 003675

3645 INSRR IRR NM 012426

2047 EPHB1 EPHB1 NM 004941

6195 RPS6KA1 RSK3 NM 001358

269 AMHR2 MISR2 NM 005243

5164 PDK2 PDHK2 NM 004660

6788 STK3 MST2 NM 012426

9448 MAP4K4 ZC1/HGK NM 012311

25865 PRKD2 PKD2 NM 003016

26524 LATS2 LATS2 NM 003089

28996 HIPK2 HIPK2 NM 012311

Category 3

1019 CDK4 CDK4 NM 003016

5602 MAPK10 JNK3 NM 003016

2065 ERBB3 HER3 NM 003016

6098 ROS1 ROS NM 007165

1021 CDK6 CDK6 NM 003016

5604 MAP2K1 MAP2K1 NM 003675

4233 MET MET NM 005131

238 ALK ALK NM 007165

2066 ERBB4 HER4 NM 001358

2534 FYN FYN NM 004941

6794 STK11 LKB1 NM 003016

5601 MAPK9 JNK2 NM 006559

2261 FGFR3 FGFR3 NM 006802

695 BTK BTK NM 007165

5347 PLK1 PLK1 NM 001358

5159 PDGFRB PDGRFB NM 006711

5156 PDGFRA PDGFRA NM 003016

6197 RPS6KA3 RSK2 NM 001356

Literature Cited

Bender, A., and Pringle, J. R. (1989). Multicopy suppression of the cdc24 budding defect in yeast by CDC42 and three newly identified genes including the ras-related gene RSRl. Proc Natl Acad Sci U S A 86, 9976-9980.

Berns, K., Hijmans, E. M., Mullenders, J., Brummelkamp, T. R., Velds, A., Heimerikx, M., Kerkhoven, R. M., Madiredjo, M., Nijkamp, W., Weigelt, B., et al. (2004). A large-scale RNAi screen in human cells identifies new components of the p53 pathway. Nature 428, 431-437.

Bernstein, E., Caudy, A. A., Hammond, S. M., and Hannon, G. J. (2001). Role for a bidentate ribonuclease in the initiation step of RNA interference. Nature 409, 363-366.

Brizuela, L., Braun, P., and LaBaer, J. (2001). FLEXGene repository: from sequenced genomes to gene repositories for high-throughput functional biology and proteomics. MoI Biochem Parasitol 118, 155-165.

Carpenter, A. E., and Sabatini, D. M. (2004). Systematic genome-wide screens of gene function. Nat Rev Genet 5, 11-22.

Chanda, S. K., White, S., Orth, A. P., Reisdorph, R., Miraglia, L., Thomas, R. S., DeJesus, P., Mason, D. E., Huang, Q., Vega, R., et al. (2003). Genome-scale functional profiling of the mammalian AP-I signaling pathway. Proc Natl Acad Sci U S A 100, 12153-12158.

Elbashir, S. M., Harborth, J., Lendeckel, W., Yalcin, A., Weber, K., and Tuschl, T. (2001). Duplexes of 21 -nucleotide RNAs mediate RNA interference in cultured mammalian cells. Nature 411, 494-498.

Fire, A., Xu, S., Montgomery, M. K., Kostas, S. A., Driver, S. E., and Mello, C. C. (1998). Potent and specific genetic interference by double- stranded RNA in Caenorhabditis elegans. Nature 391, 806-811.

Habib, A. A., Chatterjee, S., Park, S. K., Ratan, R. R., Lefebvre, S., and Vartanian, T. (2001). The epidermal growth factor receptor engages receptor interacting protein and nuclear factor-kappa B (NF-kappa B)-inducing kinase to activate NF-kappa B. Identification of a novel receptor-tyrosine kinase signalosome. J Biol Chem 276, 8865-8874.

Hafen, E., Basler, K., Edstroem, J. E., and Rubin, G. M. (1987). Sevenless, a cell-specific homeotic gene of Drosophila, encodes a putative transmembrane receptor with a tyrosine kinase domain. Science 236, 55-63.

Hammond, S. M., Boettcher, S., Caudy, A. A., Kobayashi, R., and Hannon, G. J. (2001). Argonaute2, a link between genetic and biochemical analyses of RNAi. Science 293, 1146-1150.

Hartley, J. L., Temple, G. F., and Brasch, M. A. (2000). DNA cloning using in vitro site-specific recombination. Genome Res 10, 1788-1795.

Hartwell, L. H., Culotti, J., Pringle, J. R., and Reid, B. J. (1974). Genetic control of the cell division cycle in yeast. Science 183, 46-51.

Hartwell, L. H., Culotti, J., and Reid, B. (1970). Genetic control of the cell-division cycle in yeast. I. Detection of mutants. Proc Natl Acad Sci U S A 66, 352-359.

Huppi, K., Martin, S. E., and Caplen, N. J. (2005). Defining and assaying RNAi in mammalian cells. MoI Cell 17, 1-10.

Jackson, A. L., Bartz, S. R., Schelter, J., Kobayashi, S. V., Burchard, J., Mao, M., Li, B., Cavet, G., and Linsley, P. S. (2003). Expression profiling reveals off-target gene regulation by RNAi. Nat Biotechnol 21, 635-637.

Jackson, A. L., and Linsley, P. S. (2004). Noise amidst the silence: off- target effects of siRNAs? Trends Genet 20, 521-524.

Ketting, R. F., Fischer, S. E., Bernstein, E., Sijen, T., Hannon, G. J., and Plasterk, R. H. (2001). Dicer functions in RNA interference and in synthesis of small RNA involved in developmental timing in C. elegans. Genes Dev 15, 2654-2659.

Kittler, R., Putz, G., Pelletier, L., Poser, L, Heninger, A. K., Drechsel, D., Fischer, S., Konstantinova, L, Habermann, B., Grabner, H., et al. (2004). An endoribonuclease-prepared siRNA screen in human cells identifies genes essential for cell division. Nature 432, 1036-1040.

Mackeigan, J. P., Murphy, L. O., and Blenis, J. (2005). Sensitized RNAi screen of human kinases and phosphatases identifies new regulators of apoptosis and chemoresistance. Nat Cell Biol.

Marsischky, G., and LaBaer, J. (2004). Many paths to many clones: a comparative look at high-throughput cloning methods. Genome Res 14, 2020- 2028.

Matsuda, A., Suzuki, Y., Honda, G., Muramatsu, S., Matsuzaki, O., Nagano, Y., Doi, T., Shimotohno, K., Harada, T., Nishida, E., et al. (2003). Large-scale identification and characterization of human genes that activate NF-kaρρaB and MAPK signaling pathways. Oncogene 22, 3307-3318.

Moon, R. T., Kohn, A. D., De Ferrari, G. V., and Kaykas, A. (2004). WNT and beta-catenin signalling: diseases and therapies. Nat Rev Genet 5, 691-701.

Nurse, P., Masui, Y., and Hartwell, L. (1998). Understanding the cell cycle. Nat Med 4, 1103-1106.

Paddison, P. J., Silva, J. M., Conklin, D. S., Schlabach, M., Li, M., Aruleba, S., Balija, V., O'Shaughnessy, A., Gnoj, L., Scobie, K., et al. (2004). A resource for large-scale RNA-interference-based screens in mammals. Nature 428, 427-431.

Parada, L. F., Tabin, C. J., Shih, C, and Weinberg, R. A. (1982). Human EJ bladder carcinoma oncogene is homologue of Harvey sarcoma virus ras gene. Nature 297, 474-478.

Park, J., Hu, Y., Murthy, T. V. S., Vannberg, F., Shen, B., Rolfs, A., Hutti, J. E., Cantley, C. C, LaBaer, J., Harlow, E., and Brizuela, L. (In press). Building a human kinase gene repository: Bioinformatics, molecular cloning, and functional validation. Proc Natl Acad Sci U S A.

Pelkmans, L., Fava, E., Grabner, H., Hannus, M., Habermann, B., Krausz, E., and Zerial, M. (2005). Genome- wide analysis of human kinases in clathrin- and caveolae/raft-mediated endocytosis. Nature.

Shih, C, and Weinberg, R. A. (1982). Isolation of a transforming sequence from a human bladder carcinoma cell line. Cell 29, 161-169.

Simon, M. A., Bowtell, D. D., Dodson, G. S., Laverty, T. R., and Rubin, G. M. (1991). Rasl and a putative guanine nucleotide exchange factor perform

crucial steps in signaling by the sevenless protein tyrosine kinase. Cell 67, 701- 716.

Sternberg, N., Hamilton, D., and Hoess, R. (1981). Bacteriophage Pl site-specific recombination. II. Recombination between loxP and the bacterial chromosome. J MoI Biol 150, 487-507.

Taparowsky E, Suard Y, Fasano O, Shimizu K, Goldfarb M, Wigler M. (1982) Activation of the T24 bladder carcinoma transforming gene is linked to a single amino acid change. Nature 300, 762-765.

Tomari, Y., and Zamore, P. D. (2005). Perspective: machines for RNAi. Genes Dev 19, 517-529.

Verkade, H. M., Bugg, S. J., Lindsay, H. D., Carr, A. M., and O'Connell, M. J. (1999). Radlδ is required for DNA repair and checkpoint responses in fission yeast. MoI Biol Cell 10, 2905-2918.

Walhout, A. J., Temple, G. F., Brasch, M. A., Hartley, J. L., Lorson, M. A., van den Heuvel, S., and Vidal, M. (2000). GATEWAY recombinational cloning: application to the cloning of large numbers of open reading frames or ORFeomes. Methods Enzymol 328, 575-592.

Wu, R. Z., Bailey, S. N., and Sabatini, D. M. (2002). Cell-biological applications of transfected-cell microarrays. Trends Cell Biol 12, 485-488.

Zheng, L., Liu, J., Batalov, S., Zhou, D., Orth, A., Ding, S., and Schultz, P. G. (2004). An approach to genomewide screens of expressed small interfering RNAs in mammalian cells. Proc Natl Acad Sci U S A 101, 135-140.

Ziauddin, J., and Sabatini, D. M. (2001). Microarrays of cells expressing defined cDNAs. Nature 411, 107-110.

OTHER EMBODIMENTS

The description of the specific embodiments of the invention is presented for the purposes of illustration. It is not intended to be exhaustive or to limit the scope of the invention to the specific forms described herein. Although the invention has been described with reference to several embodiments, it will be

understood by one of ordinary skill in the art that various modifications can be made without departing from the spirit and the scope of the invention, as set forth in the claims. All patents, patent applications, and publications referenced herein are hereby incorporated by reference.

Other embodiments are in the claims.

What is claimed is: