Kinase requirements in human cells: I. Comparing kinase requirements across various cell types

Abstract

shRNA loss-of-function screens were used to identify kinases that were rate-limiting for promoting cell proliferation and survival. Here, we study the differences in kinase requirements among various human cells, including freshly prepared primary cells, isogenic cells, immortalized cells, and cancer cell lines. Closely related patterns of kinase requirements among the various cell types were observed in three cases: (i) in repeat experiments using the same cells, (ii) with multiple populations of freshly prepared primary epithelial cells isolated from the same tissue source, and (iii) between nearly isogenic cells that differ from each other by the expression of a single gene. Other commonly used cancer cell lines were distinct from one another, even when they were isolated from similar tumor types. Even primary cells of different lineages isolated from the same tissue source showed many differences. The differences in kinase requirements among cell lines observed in this study suggest that the control of proliferation and survival may be significantly different between cell lines and that simple comparisons from any one cell to another may be misleading. Although the regulation of cell proliferation and survival are heavily studied areas, we did not see a bias in these screens toward the identification of previously known and well studied kinases, suggesting that our knowledge of molecular events in these areas is still meager.

The introduction of siRNAs into cells either by transfection of siRNAs themselves or processing shRNAs into their active siRNA form within a cell has been used by many investigators to reduce mRNA levels for a particular protein and then to study the effects (for general reviews see refs. 1 and 2). When RNAi-mediated changes are done on scale, the assays resemble genetic screens and allow many aspects of mammalian cell biology to be studied that were inaccessible previously.

siRNA-based screens have been used to study many cell biological processes, including cell survival, induction of apoptosis, changes in endocytosis, and chromosome integrity (3–6). shRNA-based screens have examined NF-κB signal transduction, proteasome function, p53 signal transduction, modulation of RAS oncogene activity, modulation of mitosis, and mammary cell transformation (7–12). Our current studies have focused on the differences in kinase requirements in various human cells. The kinase family was chosen for several reasons. Kinases are known to play key roles in many cell decision-making processes (13, 14). Their biochemistry is well known, and although many kinases have been studied in detail, there are numerous family members that are essentially unstudied to date. Finally, many chemical screens and subsequent medicinal chemistry have shown that effective antikinase, small-molecule inhibitors can be found that have cell-specific effects. Some of these small-molecule inhibitors have been advanced as clinical candidates for drug development, and the first antikinase drugs are now used effectively in the clinic (15, 16).

We have used high-throughput shRNA screens to identify functional differences in kinase requirements among human cells, identifying kinases whose roles in proliferation and survival differ in various cells. Twenty-one different cell populations from various sources were examined in different settings to make these comparisons. The tested cells included multiple collections of two types of primary cells from the same tissue but from different individuals, essentially isogenic cells that vary only by the introduction and expression of a single gene, immortalized but not transformed cells, and numerous tumor cell lines derived from various tissues. Interestingly, many of the identified kinases in these screens were poorly characterized, and their roles in cells have not been predicted previously. Overall, the major finding from these studies is how frequently cells in culture have developed different sensitivities to kinase loss. The range of differences raises the question of when it is appropriate to compare biochemical data from one cell line to another.

Results

Identification of Essential Kinases in HeLa and 293T Cell Lines.

While screening the commonly used human cell lines HeLa and 293T for kinases that were essential for proliferation and survival, we noted that the lists of required kinases for these two cell lines had only a small overlap [Table 1 and supporting information (SI) Table S1]. Potential kinase requirements were identified by transfection of individual shRNA expression constructs in parallel assay formats with each well receiving a different shRNA construct. Approximately five shRNAs for each human kinase were tested in our screening set, which represented ≈85% of the full kinome. The shRNA expression cassettes were in a lentivirus backbone, and they were prepared and characterized previously by The RNAi Consortium (TRC; Broad Institute, Cambridge, MA; www.broad.mit.edu/genome_bio/trc) (12). Cell number and viability were assayed 5 days after transfection by reduction of alamarBlue, a measure of mitochondrial fitness (17, 18). AlamarBlue is regarded as quantitative method for measuring cell proliferation and survival, and in our hands we were able to generate a linear standard curve over a broad range of cell concentrations where absorbance was directly proportional to cell number (Fig. S1). Upon visual inspection, alamarBlue colorimetric readout correlated with the number of crystal violet-stained cells (Fig. S2). Data points were determined by the average of triplicate points within each experiment, and each cell line was tested three times. For these experiments essential kinases were scored when alamarBlue readings were 50% below the average effect of all shRNAs tested and observed in replicate experiments. The vast majority of shRNAs have less than a 2-fold effect and follow a normal Gaussian distribution where the density of data points fall in the normal limits of a bell-shaped curve. Our 50% cutoff for choosing a “hit” was based on data points falling two or more standard deviations from the median of all values. We have also tested why cells give a low reading as measured by alamarBlue and find that all of the reductions can be accounted for by cell apoptosis, autophagy, senescence, or cell cycle block (Fig. S3).

Table 1.

Combined list of essential kinase shRNA hits

shRNA	HeLa	293T	shRNA	HeLa	293T	shRNA	HeLa	293T	shRNA	HeLa	293T
AAK1 1945	x	x	CK2a2 611	x		MAPAPK3 6155	x		PLK1 6247		x
ADCK4 7332	x		CLIK1L 2296	x		MELK 1644		x	RNASEL 924	x
ALK4 1810	x	x	CLK3 745		x	MET 396		x	ROS 956	x
AMPKa1 859		x	DDR2 1421	x		MISR2 1957	x	x	RSK2 1394	x	x
ANPb 430	x	x	EEF2K 6222	x		MST2 2176	x		RSK3 1384	x
BRD2 6312	x		EPHB1 821		x	MYO3B 2405	x	x	SGK2 2111	x
BTK 360		x	EPHB4 1775	x		NEK7 1966		x	SgK495 1817	x	x
CAMK2A 10286	x		ERBB3 619	x		NLK 2068		x	SRPK2 6274	x
CamK4 580	x	x	ERBB4 1410	x	x	PAK3 3245	x		STK33 2077		x
CAMKK1 1982		x	FER 2347	x		PAK6 1748	x	x	SURTK106 1742	x
caMLCK 1846		x	FGFR3 373		x	PBK 1809		x	TAK1 1557	x
CDK3 482	x	x	FYN 3098	x		PDGFRa 1422	x	x	TLK1 7056	x
CDK4 365		x	HIPK2 3201	x		PDGFRb 1997		x	TNK1 743	x	x
CDK6 487	x		HUNK 2271	x		PDHK2 2316	x		TRAD 1428		x
CDK7 593	x		IRR 3189	x		PEK 1402	x	x	TRRAP 5365		x
CDK9 495	x		JNK2 1012	x		PITSLRE 6207	x		TSSK2 3219	x
CDK9 498	x		JNK3 1018	x	x	PITSLRE 6210	x		TYRO3 2181	x
CDK10 1821		x	JNK3 1021	x		PKD2 1948		x	ULK4 2202		x
CDK10 1822		x	JNK3 1937		x	PKD2 1949	x	x	VACAMKL 10253	x
CDK11 3141	x		JNK3 1940	x	x	PKD3 1412	x	x	VRK3 912	x	x
CK1e 1834	x	x	KHS1 2187	x		PKG1 997	x		YSK4 2393	x
CK1e 1835		x	MAP2K1 2328	x		PKN1 1485		x	ZC1/HGK 1830		x
CK1e 603	x

HeLa-63 shRNA hits for 58 kinases. 293T-46 shRNAs hits for 41 kinases. Twenty shRNAs hits (19 kinases) in common. x indicates hits.

Of 89 shRNAs that blocked cell proliferation and survival by 50% or more in either line, only 20 shRNAs inhibited both HeLa and 293T cells. Sixty-three shRNAs inhibited HeLa, and 46 inhibited 293T. Even though there were considerable differences between the two cell lines, the assays themselves were robust, showing a coefficient of correlation of 0.85 or higher among repeats in the same cell line (Fig. 1 and Table S2).

Fig. 1.

Proliferation and survival screen. Relative cell proliferation and survival was measured by alamarBlue assay after transfection of ≈2,000 shRNAs targeting 430 kinases into HeLa cells. Fluorescence readings from two independent HeLa experiments are compared on the x- and y axis with a correlation coefficient of 0.89 demonstrating reproducibility between replicate screens. See Table S1 for a full list of kinases assayed in the primary screen.

Although we expected to find a significant number of differences between these two cell lines, we were surprised to find such a small number of commonly required kinases. The different sensitivities in these two cell lines raised several interesting questions. The first question considered here is whether cells grown in culture routinely show significant differences in kinase requirements. In accompanying papers, we study specific settings where cell-to-cell differences are small (19, 20) or large (21).

Kinase Expression Levels and Cell-to-Cell Differences.

We have identified kinases that are essential for the survival of one cell but not the other. It is important to know whether the kinases are expressed, and if so, how the relative RNA levels vary between the cell lines being compared. Here, we have included microarray-based RNA expression data for 31 of 80 essential kinases in both HeLa and 293T cell lines (Fig. S4). These data show that it is not possible to predict kinase requirements based on RNA expression data, and therefore this information does not aid us in identifying a cell's sensitive point. The different kinases are expressed at different levels in both cell lines, and for many, the RNA expression levels do not significantly vary between cell lines, although one cell line dies and the other lives after kinase loss.

Viral Transduction.

For all of the remaining experiments reported here, we chose viral transduction rather than transfection for delivery into cells. This approach was used to minimize variations of shRNA delivery and expression across various cell lines, especially for those cells that are difficult to transfect, and to allow cellular responses to be assessed over extended time periods. Successful high-throughput production of high-titer lentiviral stocks was measured by three methods. First, for each batch of lentivirus, formal titers were determined for a set of lentiviral standards. Titers were determined by counting colonies after transduction of mouse NIH 3T3 cells at limiting dilution and under puromycin selection. The viral titer was sufficiently high for successful transductions and found to be ≈2 × 10⁶ cfu/ml (data not shown). Second, we measured the levels of genomic RNA sequences released from virions in the supernatants prepared during a production run. Comparisons of the levels of RNA sequences for the puromycin acetyl transferase gene found in a selection of ≈180 lentivirus vectors for 36 kinases showed no more that a 2-fold difference from the mean values, demonstrating that the viral production of the vast majority of shRNA expression vectors was similar (Fig. S5). Third, the generation of infectious particles was confirmed by comparing inhibition of cell proliferation and survival in the presence or absence of puromycin (examples of 10 plots are shown in Fig. S6). When the comparisons of relative alamarBlue values derived from plus/minus puromycin selection showed a positive linear corrleation with a slope of ≈1, then viral transductions were approaching 100%. Taken together, these results confirm that the various cell lines can be reproducibly transduced and that the viral titer is in an appropriate range for these tests.

Other Cell Lines Also Show Significant Differences in Kinase Requirements.

To determine whether HeLa and 293T cells were a serendipitous choice of cell lines for comparison in our initial experiment, we examined kinase requirements for four related tumor cell lines all isolated from nonsmall cell lung carcinomas (NSCLC). Fig. 2 shows that even when cell lines are derived from similar tissue sources they still can display significant differences in kinase requirements. In pairwise comparisons, A549, H23, H1229, and H358 lung carcinoma cell lines show similar ranges of overlapping kinase requirements to those seen when comparing HeLa and 293T cells (Fig. 2 A vs. B). Similarly, if we scored for the frequency at which the same kinases were required in any two cell lines, in any three cell lines, in all four cell lines, or in only one cell line, the differences among the cell lines were quite apparent, with only 5% of kinases required in all four cell lines and 40% unique to a particular cell line (Fig. 2C). These experiments suggest that individual cell lines in culture have different points of maximum sensitivity for kinase loss and therefore may rely on different molecular pathways or have different rate-limiting steps to manage proliferation and survival.

Fig. 2.

Differential kinase requirements comparing cell lines derived from different tissue origins and the same tissue origin. (A and B) Relative cell proliferation and survival was measured by alamarBlue assay 5–6 days after transfection of HeLa (cervical) and 293T (renal) cell lines, and after transduction of A549 (lung), H358 (lung), H1975 (lung), and H23 (lung) cell lines with ≈2,000 shRNAs targeting 430 kinases. Kinase requirements were defined as a 50% or more growth inhibition relative to the mean value of examined shRNAs. Total number of essential kinases that scored per cell line is shown in parentheses. Hatch bars show similarities in kinase requirements between (A) two cell lines for HeLa and 293T (B) and four pairs of lung cell lines. (C) Of 278 essential kinases that scored among four lung lines, 109 kinases scored in any two cell lines, 44 kinases scored in any three cell lines, and 14 in all four cell lines.

Cell-to-Cell Differences in a Panel of Representative Human Cells.

To examine whether different kinase requirements were found among other human cells, we used the 89 shRNAs identified in the HeLa and 293T cell comparison as probes to measure cell-to-cell differences in a broad panel of cell lines including normal cells isolated from a number of different tissues sources. Also incorporated into this set were 11 shRNAs that scored in a pilot screen of HeLa and 293T cells, which brought the total of tested shRNAs to 100. The cells that were chosen for this comparison included a selection of human tumor and immortalized lines commonly used in the laboratory, with an oversampling of kidney and cervical carcinoma cells representing the origin of the 293T and HeLa cells. We also included two separate populations of primary human foreskin keratinocytes (HFKs) and human foreskin fibroblasts (HFFs) and two well studied and essentially isogenic colon carcinoma cell lines that differ only by the expression of the human papillomavirus (HPV) E7 oncoprotein.

Comparisons of cell response are presented in Fig. 3 as heat maps using unsupervised hierarchical clustering with Euclidean distance to link shRNAs and cell lines with related growth inhibition patterns. Transduction of each of the 21 cell lines or primary cell populations with the 100 shRNAs gave similar values whether puromycin was included or not, showing that viral titers were sufficient to infect all of the cultured cells and that cells were amenable to lentivirus transduction and viral gene expression (see Fig. S6 and data not shown). For the data in Fig. 3, the extent of inhibition of proliferation and survival was displayed as rank order using percent arrest values. Percent arrest was calculated for each individual shRNA within a cell line by normalizing to the values of proliferation of a GFP-expressing lentiviral vector (LKO-GFP). For data presented in the heat map (Fig. 3), no puromycin was added, but analogous results were seen in the presence of puromycin. The most potent inhibitors of cell growth are shown in red in Fig. 3, and the least effective inhibitor is in green.

Fig. 3.

Rank order of HeLa-293T hits across a large panel of cell lines. Relative cell proliferation and survival was measured by alamarBlue assay 6 days after transduction of lentiviruses expressing 89 shRNAs targeting 80 kinases and controls into 19 different cell lines. The RKO-pc and RKO-E7 cells were run in replicate, and separate experiments were listed as RKO-pc-1 and RKO-pc-2 and RKO-E7–1 and RKO-E7–2, respectively. Reduction in viability induced by each shRNA was calculated relative to a lentiviral vector expressing GFP and assembled by rank order. Color scales represent the greatest decrease in viability by an shRNA (red) to the least (green). Columns display 21 different screens (19 cell lines tested + two duplicates). Rows display shRNAs. Data were analyzed by hierarchical clustering with Euclidean distance to link shRNAs and cell lines with related growth inhibition properties.

For accessing reproducibility between transduction experiments, comparative experiments were performed in quadruplicate within each experiment, and each cell line was tested two to three times. For example, the heat map presented in Fig. 3 includes two experiments of human colon carcinoma cell line (RKO-pc) engineered to express HPV type-16 (HPV16) E7 oncoprotein (RKO-E7). The patterns of kinase requirements from replicate experiments were very reproducible with a resulting correlation coefficient of 0.95 between the same cell line in two independent screens.

We used both Euclidean distances (Table S3) and correlation coefficients (R; data not shown) to determine statistical relationships among experimental pairs of cell lines. Both approaches led to identical conclusions. The range of Euclidean distances between any 2 of the 21 cell lines was from 95 to 400. As expected, the experimental repeats with the same cell lines showed the lowest Euclidean distance values, indicating that they were most similar to one another. The repeats of kinase requirements of RKO-pc cells showed a Euclidean distance of 127, whereas the repeats in RKO-E7 recorded 95.

The results between pairs of cell lines or cell populations closely mirrored the general conclusions from the HeLa-to-293T cell comparison with several notable exceptions. In most cases the patterns of kinase requirements showed considerable differences between cell lines. Reassuringly, the most closely related cell comparisons were between sets of primary cells prepared from the same tissues but isolated from different populations of individuals on separate dates; the two populations of primary HFKs (HFK-A and HFK-B) had a relative distance of 146, and the two populations of primary HFFs (HFF-1 and HFF-2) had a distance of 123. The comparisons between the RKO cells with and without E7 showed the next best correlation with relative distances of 180, 181, 193, and 219 between each of the four pairs of potential comparisons of these two cell lines, each repeated once. The comparisons for the RKO colon carcinoma cell lines with or without E7 identified 19 kinases with changes in percent arrest of 30 or more. These differences are further investigated in the accompanying paper (19).

All other pairs were more distant, some considerably so. Three classes of comparisons are worth special note. First, when the HeLa-to-293T comparison is used as a standard to measure the relationship of other cell pairs, the vast majority of the pairs are equal to or more distant. A total of 193 cell pairs scored equal to or more distant than HeLa-to-293T, whereas only 18 pairs were more closely related. Second, even though two primary cell populations of keratinocytes and fibroblasts were isolated from the same pools of foreskin tissue they exhibited different patterns of kinase requirements. Third, regarding commonly used tumor cells there were isolated cases that showed close comparisons between tumor cell pairs, and that was when the cells came from related tissue origins. For example, the cervical carcinoma lines HeLa and CaSki, both infected originally by HPV, were among the most closely related of the tumor cell lines, as were the two lung carcinoma cell lines A549 and CALU-1. Interestingly, the other HPV-associated cervical carcinoma cell line, SiHa, was no more closely related to HeLa and CaSki than the lung, renal, or breast tumor cell lines in our test set. This result is surprising because HeLa is derived from an HPV18-positive adenocarcinoma, whereas CaSki and SiHa each are derived from HPV16-positive squamous cell carcinomas. Other pairs that have similar kinase requirements include 293T and BJ Tert cells (a kidney epithelial cell immortalized by adenovirus and a foreskin fibroblast immortalized by telomerase, respectively) and A549 and ACHN (from a NSCLC and a renal carcinoma, respectively).

Validation of shRNA Effects in a Panel of Closely Related Human Tumor Cell Lines.

Because we chose only the most potent hits in the initial screens, many of the kinases were identified only by a single shRNA. This process left open the possibility that some unknown number of the responses seen were caused by off-target effects of the shRNAs. Rather than performing the expensive analysis of RNA levels for each of the hits in a large number of cells to validate each shRNA result, we used a simpler test for on-target effects. We expanded the collection of shRNAs used in Fig. 3 to identify two hairpins that target different regions of the mRNA but that generated similar responses in a panel of test cells. The identification of two shRNAs that are specific for different regions of the mRNA and exhibit similar experimental outcomes is now widely accepted as a good initial test for on-target effects, thus indicating that the shRNAs have specifically down-regulated expression of their intended target. Although this test is not a conclusive measurement of mRNA knockdown for each kinase in all of the tested cell lines, it does allow us to interpret the overall essential kinase patterns determined with confidence. The cells that were chosen for this validation step were all from same tissue source to avoid possible complications that might arise by comparing very different cell types.

We collected 444 shRNAs representing ≈5 shRNAs for 80 of the kinases identified in Table 1. These shRNAs were tested for growth inhibitory effects in five NSCLC cell lines (H23, H358, H647, H1229, and H1975). We identified two shRNAs for each kinase that behaved most similarly across the five NSCLC lines, where at least one of the two shRNAs exhibited inhibitory effects (>50% inhibition in at least one of the five cell lines). This process yielded 75 pairs of shRNAs for 75 of the 80 kinases. The remaining five kinases were not considered further because none of their shRNAs exhibited >50% inhibitory effect in any of the five cell lines. We then plotted how these five cell lines respond to these two shRNAs as percent arrest, which was calculated for each individual shRNA within a cell line by normalizing to the proliferation values of a control lentiviral vector expressing a scrambled shRNA. Fig. S7 shows the growth inhibitory patterns for the two shRNAs that show the most similar response in five NSCLC cell lines for all of the 75 kinases, and Fig. 4 shows the responses of the five kinases whose pairs of shRNA showed the best similarities across the five cell lines. As shown in Fig. 4 and Fig. S7, for all but a few of these kinases, we are able to find two shRNAs that behave similarly, suggesting that in the vast majority of cases the target of the shRNA is the intended kinase. Importantly, the five cell lines frequently respond differently to both of these two representative shRNAs. These data are entirely consistent with earlier experiments and suggest that even potentially closely related cells often show remarkably different profiles of kinase requirements.

Fig. 4.

Viral transduction of five NSCLC cell lines using additional shRNAs targeting the same kinase. Presented here are five kinases whose two shRNAs show the most similar patterns across the five NSCLC lines. The plots for all 75 pairs of shRNAs are in Fig. S7.

Discussion

We have used a series of RNAi screens to identify a collection of kinases whose loss is damaging to a cell's proliferation and survival. For each tested cell, the kinases identified from these screens mark components of cell metabolism that are rate-limiting for survival and proliferation. Two features of these kinase screens have drawn our attention. The first is the lack of bias toward kinases that have been identified in previous work. Fig. S8 shows the number of published papers reported in PubMed for each human kinase and compares it to the number of papers published for the kinase hits studied in Table 1 and Fig. 3. The shape of the graft of papers published for all kinases is remarkably similar to that generated for the kinases identified in our studies. The only small difference is the skew from the handful of kinases studied in enormous depth and not found in our studies. Many of the kinases identified in our studies have only a few publications or still are not reported in the literature. We interpret this to mean that methods and approaches used to detect and study kinases to date have not exhaustively identified all kinases with key roles in cell proliferation and survival.

The second feature that we have noted is how frequently cells show different patterns of kinase requirements. We have been surprised to detect strong differences in kinase requirements at least across many of the cells examined in these studies. The initial observation that we found surprising was that among the 89 shRNAs required in HeLa or 293T cells only 20 were common to both cell lines. This difference in kinase requirements is reinforced by three general observations: (i) if the level of differences found while comparing HeLa and 293T cells is used as a standard to judge other cell pairs, >90% of the other comparisons we studied were more distant, (ii) using the shRNA screens described here, we found only 14 kinases in lung that were required in all of the cells we tested, and (iii) although using the 50% inhibition cutoff described here identified on average ≈50–100 kinases as required for proliferation and survival of any one cell line, over one-half of all kinases scored as being essential in at least one cell. Although judging the expected level of similarity between cells in culture is a highly subjective argument, a widely held view of cell metabolism is that there are many key features held in common across cells, which is likely to be true for many activities of cell biology. However, the differences reported here bring into question how broadly this view should be applied, particularly in regard to the roles that kinases might play in regulating cell proliferation and survival decisions.

The concept that many processes and regulatory events are similar across multiple cell lines is reinforced by several practices commonly used in the signal transduction field. For example, it is common to apply findings in one cell line to other cells or even to other species and to define an archetypical view of a mammalian cell. This observation extends to the assembly of results into composite cell signaling charts, many of which now grace freezer doors across molecular biology laboratories. Similarly, it has become common practice to use results from any transfectable cell to consider how an exogenously expressed protein may influence another, sometimes via cotransfection. Given the demonstration of how different individual cell lines can be in culture, these and other related practices may be misleading. This would also bring into consideration whether some of the widely held views on cell network theories or complicated signaling pathways may be built at least in part on inappropriate simplifications generated by assembling summary views from different cell systems that should not be compared so directly. Resolving these issues is clearly a large and important task.

Why Do Cells Respond Differently to Kinase Down-Regulation?

Our data demonstrate that the requirement for essential kinases varies among many commonly used cell lines. At present, it is impossible to know fully how these differential requirements develop, but a few first principles are clear. One source of variation is the original cell type. Similarly prepared and cultured primary cell types, for example, within populations of keratinocytes isolated from different individuals on different occasions, are among the most closely related cell comparisons tested here. However, we observed large differences between types of primary cells, for example, between keratinocytes and fibroblasts, even though both cell types were isolated from the same tissue source at the exact same time. These cells showed little relationship to one another and scored among the most distant of the cell pairs tested. Therefore, cell source will clearly be a major cause of cell-to-cell differences. However, we also found little evidence for extensive similarities among tumor cells that arose from similar cell origins. Therefore, at some stage, either in tumor development or adaptation to culture, major changes in cell regulation must have occurred.

Given the similarity of primary cells from the same tissue and same cell type, other sources for differential responses must come from mutations or epigenetic changes that have occurred in tumor development, in selection for proliferation in tissue culture, or through random stochastic changes that might occur between any two cell lineages. We anticipate that the overall differences in patterns of kinase requirements will be caused by a compilation of multiple changes, because two cell lines that differ only by the expression of a single gene yield patterns that are quite similar.

Finally, it is worth noting that the differences identified here make the identification of commonly used targets for tumor chemotherapy more challenging than ever. There are many previous studies that point out that tumor cells from the same tissue with the same histology, and sometimes even with some similar mutations, behave differently in response to drug treatment. Our findings suggest that these differences may be even more common than originally suspected. Nonetheless, the use of RNAi screens provides a rapid and powerful method to identify sensitive rate-limiting steps that are essential for cell survival and proliferation. Among these potential targets will be many new candidates for drug discovery programs.

Materials and Methods

Tissue Culture.

Normal HFKs and HFFs were obtained from neonatal foreskins. HFKs were maintained in keratinocyte-SFM supplemented with 1% penicillin-streptomycin, 0.1% gentamicin, and 0.2% amphotericin B (Fungizone), and HFFs were maintained in DMEM supplemented with 1% penicillin-streptomycin and 10% calf serum.

NCI-H1299, NCI-H358, NCI-H647, NCI-H1975, and NCI-H23 were maintained in RPMI medium 1640 supplemented with 1% penicillin-streptomycin and 10% FBS. HeLa, CaSki, and SiHa were maintained in DMEM supplemented with 1% penicillin-streptomycin and 10% FCS. 786-O, A498, ACHN, Calu-1, A549, MCF7, 293T, WI38, and BJ-TERT in DMEM were supplemented with 1% penicillin-streptomycin, 4 mM l-glutamine, and 10% FBS. RKO cells stably expressing the pcDNA3 control vector (RKO-pc) or HPV16 E7 (RKO-E7) were maintained in McCoy's medium supplemented with 1% penicillin-streptomycin, 0.5 mg/ml G418, and 10% calf serum (22). MCF10A cells were cultured in DMEM/F12 supplemented with 1% penicillin-streptomycin, 20 ng/ml EGF, 0.5 μg/ml hydrocortisone, 10 ng/ml cholera toxin, 10 μg/ml insulin, and 5% horse serum as described (23). All cell lines were grown at 37°C with humidified atmosphere and 5% CO₂.

AlamarBlue Assay.

Five or six days after infection, media were removed from 96- or 384-well plates, and alamarBlue reagent (Biosource/Invitrogen) diluted into IX supplemented DMEM and added to each well. Plates were then incubated for 2–4 h at 37°C before reading on a microtiter well plate reader Spectrafluor Plus (Tecan) at 595 nm.

Library Manipulation.

Library glycerol stocks were handled as recommended by TRC. High-throughput, high-quality DNA preparation has been described, and DNA concentration was determined by using Hoescht dye (24).

100 Hits DNA and Virus Production.

High-quality DNA preparations for 100 hits were obtained with a large-scale plasmid purification kit (Qiagen). For high-throughput lentiviral production in a 96-well format, 293T packaging cells were cotransfected with shRNA-encoding replication-deficient viral vectors and the necessary helper plasmids for virus production. The virus was pseudotyped with the envelope glycoprotein from vesicular stomatitis virus (VSV-g) as described (12, 25, 26).

Transductions in 96- and 384-Well Formats.

For 96-well experiments, all cell lines were seeded between 2,000 and 5,000 cells per well by using a 96-well format in a final volume of 100 μl per well. Twenty-four hours later, 50 μl of media was removed and different amounts of viral supernatant were added depending on the experiment (ranging from 1.25 to 20 μl), all in the presence of 8 μg/ml polybrene final concentration. Each viral supernatant was added in quadruplicate. Plates were spun at 1,175 × g for 30 min. Transduced cells were washed between 12 and 16 h after transduction, and 24 h later 0.5–2 μg/ml puromycin was added to select wells, with the exception of MCF7 cells where puromycin was added 48 h after transduction. Cells were harvested 5 or 6 days after transduction for alamarBlue measurements. For 384-well experiments, cell lines were seeded between 500 and 800 cells per well in a final volume of 40 μl per well. Twenty-four hours later, different amounts of viral supernatants were added depending on the cell lines (ranging from 2.5 to 5 μl), all in the presence of 8 μg/ml polybrene final concentration and processed as above.

mRNA and Genomic RNA Analysis.

For mRNA quantitation of transduced HeLa and 786-O cells, the transduction protocol was performed in 12-well plates with 120 μl of viral supernatant. After transduction, plates were spun at 1,175 × g for 30 min. Transduced cells were washed 12 h after transduction and harvested at different time points by lysing cells in buffer supplied with Panomics QuantiGene assay according to the manufacturer's recommended procedure. Target-specific probes were incubated with 80 μl of cell lysate. GAPDH control-specific probes were incubated with 5 μl of cell lysate that was diluted 1:10. To calculate the percent of knockdown, the target values were normalized to GAPDH values and then normalized to scrambled shRNA control. To assess the relative titers of viruses produced in a 96-well format, the puromycin-N-acetyl transferase genomic RNA levels from viral supernatants were measured according to the manufacturer's protocol.