The Concept of Synthetic Lethality in the Context of Anticancer Therapy

 

 

 

 

Review

Nature Reviews Cancer 5, 689-698 (September 2005) |doi:10.1038/nrc1691

The Concept of Synthetic Lethality in the Context of Anticancer Therapy

William G. Kaelin, Jr1  About the author

top

Two genes are synthetic lethal if mutation of either alone is compatible with viability but mutation of both leads to death. So, targeting a gene that is synthetic lethal to a cancer-relevant mutation should kill only cancer cells and spare normal cells. Synthetic lethality therefore provides a conceptual framework for the development of cancer-specific cytotoxic agents. This paradigm has not been exploited in the past because there were no robust methods for systematically identifying synthetic lethal genes. This is changing as a result of the increased availability of chemical and genetic tools for perturbing gene function in somatic cells.

The bottleneck to the development of safe and effective anticancer drugs does not lie in an inability to identify chemicals that will kill cancer cells. In fact, thousands of compounds have been identified over the past 50 years that will accomplish this feat. Instead, the bottle-neck lies in our inability to identify chemicals that will kill cancer cells at concentrations that do not harm patients. Most of the chemotherapeutic agents used today have remarkably low THERAPEUTIC INDICES and narrowTHERAPEUTIC WINDOWS. The therapeutic window is influenced by a number of factors, including the shape of the curve that relates the intended biological effect of the drug to changes in the activity of its intended target (‘on-target’), and the propensity of the drug to affect unintended targets (‘off-targets’) at higher doses. Off-target effects can cause toxicity and, in some cases, antagonize on-target biological effects. Most anticancer drugs in use today were discovered based on their ability to kill rapidly dividing cancer cells in vitro. Predictably, when administered to patients, many of these drugs also injure rapidly dividing normal cells, such as bone-marrow haematopoietic precursors and gastrointestinal mucosal epithelial cells. In addition, many of these drugs are toxic to normal cells that are not rapidly dividing. Examples include doxorubicin (toxic to the heart), bleomycin (toxic to the lung) and cytarabine (toxic to the cerebellum). These other forms of organ damage become particularly important (dose-limiting) in settings in which toxicity to rapidly dividing cells can be partially ameliorated through supportive-care measures (such as bone-marrow transplantation). For these reasons, it is imperative that anticancer drugs be developed that can kill cancer cells at clinically achievable concentrations, with therapeutic indices that are higher than those of classic cytotoxic agents.

Therapeutic index

Many factors influence the therapeutic index of a drug. Some relate to the quality of the drug itself — for example, its ability to distinguish between intended and unintended targets. Others relate to the nature of its target — for example, its distribution, its normal function(s), and the degree to which those functions must be altered to achieve the desired effect. Most antibacterial agents are remarkably safe because their targets are present in the organisms they are designed to kill but not in normal host cells. However, many other relatively ‘safe’ drugs — such as anti-hypertensives, anti-anxiety drugs and cholesterol-lowering agents — inhibit normal cellular proteins. These drugs are clinically useful because their effects are titratable (through changes in dose and schedule), and quantitative changes in the activities of their targets lead to the desired changes in host physiology.

Two paths can be envisioned to arrive at an anticancer drug that would selectively kill cancer cells. The first, which is modelled on the development of anti-infectious agents, would be to identify drug targets that are essential for the viability of cancer cells but are not present in normal cells (the so-called ‘target-driven therapeutic index’)1, 2 (Fig. 1). The fusion proteins generated by cancer-associated chromosomal translocations might, at first glance, seem to be ideal in this regard. However, this presumes that drugs can be developed that will discriminate between a particular protein (or functional subdomain) in its normal context and in its pathogenic, fused state. This might be difficult. For example, it is fallacious to argue that the efficacy and safety of imatinib mesylate (Glivec) for the treatment of chronic myelogenous leukaemia (CML) stems from the fact that its target, breakpoint cluster region (BCR)–Abelson murine leukaemia viral oncogene homologue (ABL), is unique to CML cells because imatinib mesylate inhibits the kinase activities of both BCR–ABL and ABL (in addition to several other cellular kinases)3. So, the relatively high therapeutic index of imatinib mesylate cannot be explained by the restriction of its target(s) to CML cells (see below for potential alternative explanations). Similarly, it might be difficult to develop drugs that directly inhibit oncoproteins that result from point mutations without affecting their normal counterparts.

Figure 1 | Framework for developing anticancer drugs with a high therapeutic index.

Figure 1 : Framework for developing anticancer drugs with a high therapeutic index. Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.comAn anticancer drug might have a high therapeutic index because its target is uniquely present in cancer cells (a), or because the requirement for its target is quantitatively or qualitatively different in cancer cells than in normal cells (b and c). This differential requirement might be because of intrinsic differences in the cells (b), such as genetic (red) and epigenetic (blue) differences, or extrinsic differences in the cells (c), such as loss of survival signals provided by normal cell–cell and cell–matrix interactions. Modified with permission from Ref. 2 © (2002) Elsevier Science.


A second way to achieve enhanced cancer-cell selectivity, however, would be to identify situations where the requirement for a particular target was enhanced in the context of a cancer cell compared with normal cells (the so-called ‘context-driven therapeutic index’)1, 2 (Fig. 1). The requirement for a particular target might be increased because of changes that are intrinsic to the cancer cell (for example, through epigenetic or genetic changes), extrinsic to the cancer cell (for example, as a result of microenvironmental changes leading to altered cell–matrix and cell–cell interactions), or both.

All of the anticancer drugs in use today affect targets that are shared between normal cells and cancer cells, including enzymes involved in fundamental processes such as DNA replication. The fact that their therapeutic indices, however small, exceed unity, coupled with the observation that they can, in certain settings, induce striking remissions and occasionally cures (for example, cisplatinum-based regimens for testicular cancer), indicates that contextual differences between normal cells and cancer cells are therapeutically exploitable. So, can our growing knowledge of cancer genetics, coupled with a more sophisticated understanding of gene–gene interactions, be used to identify drug targets that have enhanced therapeutic indices by virtue of such contextual differences? Studies of gene–gene interactions in model organisms have provided a conceptual framework for this task.

Synthetic lethality

Two genes (‘A‘ and ‘B‘) are said to be ‘synthetic lethal’ if mutation of either gene alone is compatible with viability but simultaneous mutation of both genes causes death4, 5, 6, 7, 8, 9 (Fig. 2). This concept can be extended to situations in which simultaneous mutation of two genes impairs cellular fitness more than mutation of either gene alone (‘synthetic sick’). In either of these two situations, A buffers the effect of changes in B and vice-versa, but this buffering is lost when both A and B are mutated at the same time4, 6, 10. Synthetic lethal interactions have most commonly been described for loss-of-function alleles, but can also involve gain-of-function alleles. For example, gene B might become essential for survival when a particular gene A is overexpressed (known as synthetic dosage lethality)11, 12, 13. Approximately 20% of genes in the budding yeastSaccharomyces cerevisiae are individually essential, but genetic screens in this organism suggest that synthetic lethal interactions are common among the remaining 80% (perhaps on the order of 10 interactions per gene)10, 14, 15.

Figure 2 | Gene–gene interactions: synthetic lethal and suppressive interactions for two genes.

Figure 2 : Gene|[ndash]|gene interactions: synthetic lethal and suppressive interactions for two genes. Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.comTwo genes (‘A‘ and ‘B‘) are said to be ‘synthetic lethal’ if mutation of either gene alone is compatible with viability but simultaneous mutation of both genes causes death. B is an extragenic suppressor of A if mutation of Bsuppresses the phenotype observed whenA is mutated. A lowercase letter denotes a mutant.


Loss-of-function alleles having a synthetic lethal (or synthetic sick) relationship can often, but not always, be easily rationalized based on the functions of their protein products. They might, for example, be uniquely redundant with respect to an essential function (as occurs in some PARALOGUES), be two subunits of an essential multiprotein complex, be two interconnected components in an essential linear pathway (with each mutation decreasing the flux through the pathway), or participate in parallel pathways that are together essential for survival (for example, a crucial metabolic pathway and an alternative or salvage pathway). The concept of synthetic lethality can be further extended to embrace the situation where mutation of A is lethal only in combination with mutations that affect several non-essential genes B, C, D and so on2, 6.

It has been suggested that the concept of synthetic lethality could be used to choose anticancer drug targets1, 7, 16. In particular, the protein products of genes that are synthetic lethal to known cancer-causing mutations, if amenable to pharmacological attack (for example, if they encode an enzyme), should theoretically represent excellent targets for anticancer therapy. This approach simultaneously tackles two vexing problems in cancer pharmacology. The first relates to the fact that many cancer-associated mutations, like most drugs, induce a loss of function1, 2. Therefore, it is not immediately obvious how to pharmacologically approach cancer cells in which, for example, a particular tumour-suppressor protein is crippled (or worse yet, absent). Targeting a protein that is synthetic lethal to such a lost or crippled protein provides an elegant solution to this problem. The second problem relates to whether it is possible to achieve selectivity by inhibiting proteins that are also important for cellular homestasis. If A and B are synthetic lethal (or synthetic sick), then inhibitors of B should selectively kill (or inhibit) cancer cells with mutant A. In the ideal situation, complete neutralization of B, genetically or pharmacologically, would have no effect on normal cells, and even partial inhibition of B in cancer cells would cause death (because of mutant A; Fig. 3, left panel). However, Binhibitors might display a significant therapeutic index even when these ideal conditions are not met. This would require that the Amutation shifts or alters the fitness dose–response curve of the Binhibitor such that keeping B activity below a certain threshold selectively impairs cells with mutant A (Fig. 3, middle and right panels).

Figure 3 | Theoretical fitness curves for wild-type andA-/- cells in response to a drug that inhibits the B gene product.

Figure 3 : Theoretical fitness curves for wild-type and A|[minus]|/|[minus]| cells in response to a drug that inhibits the B gene product. Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.comA reading of 0% fitness denotes death, whereas 100% fitness denotes the wild-type state (for simplicity, fitness >100% is not considered in these examples). In the middle panel, a therapeutic window is created by a shift in the fitness curve when gene A is absent. In the left and right panels the therapeutic window is created by changes in the shapes of the fitness curves when gene A is absent.


It could be argued that some (and perhaps most) anticancer drugs in use today are, at least in hindsight, exploiting synthetic lethal, or synthetic sick, interactions. For example, synthetic lethal relationships between DNA-replication genes (such as certain DNA polymerases) and DNA-repair genes (such as mismatch-repair genes) are well documented in model organisms7, 16. It seems likely that the efficacy of the many anticancer drugs that interfere with DNA synthesis is due, at least in some cases, to the presence of tumour-associated mutations that affect DNA repair or the response to DNA damage. Another example of synthetic interactions is provided by certain chemotherapeutic agents and mutations that directly or indirectly compromise the function of the retinoblastoma protein (pRB, encoded by the RB1 gene) tumour suppressor. Inactivation of pRB has been documented in many cancers and leads to an increase in E2F activity, which, in turn, activates various genes involved in S-phase entry17. One of these, topoisomerase II, causes DNA strand breaks and apoptosis when bound to topoisomerase inhibitors such as etoposide. As would be predicted, pRB-pathway mutations sensitize cells to drugs that inhibit topoisomerase II (Refs 18–21). In addition, E2F1, like the oncoprotein MYC, increases the expression of many pro-apoptotic genes, including the p53 paralogue p73, which might sensitize pRB-defective cells to drugs that elicit additional apoptotic signals (such as DNA-damaging agents)22, 23, 24, 25.

Two newer anticancer agents also exploit contextual differences between cancer cells and normal cells. Studies in model organisms suggest that mutations affecting chaperones that are involved in protein folding can unmask the deleterious consequences of various mutations26. Preclinical data indicate that HSP90 (heat-shock protein of 90kDa) inhibitors have anticancer activity, and that certain mutant oncoproteins, such as mutant BRAF and mutant EGFR (epidermal growth factor receptor), have an increased requirement for HSP90 function27,28, 29. One HSP90 inhibitor, 17AAG, has completed phase I testing and is entering phase II studies. The accumulation of mutated and/or misfolded proteins might also alter the requirement of a cell for proteasomal function30. The proteasomal inhibitor bortezomib is well tolerated in humans and was recently approved for the treatment of multiple myeloma31.

Discovery of human synthetic lethal interactions

Our knowledge of the molecular networks that are established in normal cells and cancer cells is too rudimentary to allow reliable predictions of the genes that will be synthetic lethal to a given cancer gene. Nonetheless, a few ideas have been put forward for how synthetic lethal combinations might be achieved, based on first principles. Many oncoproteins, including E2F1 and MYC, represent a double-edged sword for cancer cells because they deliver both pro-mitogenic and pro-apoptotic signals. A counterintuitive approach to treating cancer cells that have hyperactive oncoproteins such as these would be treating them with drugs that enhance their action further, in the hope of crossing an apoptotic threshold. For example, E2F1 is negatively regulated by both pRB and cyclin A32, 33, 34, 35. Loss of the pRB pathway establishes a positive-feedback loop in which E2F1 activates its own promoter36, and blocking the remaining interaction of cyclin A with E2F1 kills transformed cells but not their normal counterparts37, 38, 39. Unfortunately, inhibiting the activity of the cyclin-A partner CDK2 (cyclin-dependent kinase 2) does not have the same effect40, possibly because another catalytic partner can substitute for CDK2 in its absence41, 42. Synthetic lethal interactions might also be predicted based on the loss of particular cell-cycle checkpoints16. For example, S-phase cells, in contrast to G1 cells, can be induced to undergo premature chromosomal condensation under certain conditions, such as treatment with caffeine at doses that inhibit ATR (ataxia telangiectasia and RAD3-related protein)43, 44, 45, 46. Cells that lack p53, which has a role in G1 control, are more susceptible to caffeine than their wild-type counterparts47.

There are now multiple examples of cancers that seem to be dependent on or ‘addicted’ to certain activated oncogenes (gene-replacement experiments suggest that tumour cells can also become addicted to the inactivation of tumour-suppressor genes). Oncogene addiction might underlie the success of the kinase inhibitor imatinib mesylate for CML (in which the oncogene is BCR–ABL) and gastrointestinal stromal tumours (in which the oncogene is KIT)3 and of the EGFR inhibitor gefitinib for EGFR-mutated non-smallcell lung cancer48, 49, 50, 51. Bernard Weinstein, who coined the term ‘oncogene addiction’, initially envisioned that this phenomena was related to the ability of such oncogenes, which can be viewed as nodes in complex molecular networks, to simultaneously deliver proliferative and antiproliferative signals52 (Fig. 4A). As long as the oncogene signal is sustained, the proliferative signal — which might promote mitogenesis, survival, or both — would dominate. However, if the oncogene is acutely silenced, the antiproliferative signal dominates, leading to cessation of growth or cell death (in this scenario it must be invoked that the antiproliferative signal ‘decays’ more slowly than the proliferative signal when the oncogene is inhibited)2.

Figure 4 | Models of oncogene addiction.

Figure 4 : Models of oncogene addiction. Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.coma | Many oncogenes paradoxically induce pro-mitogenic signals as well as anti-mitogenic (or pro-apoptotic) signals. Growth stimulation results from oncogene activation presumably because the former is dominant to the latter. However, acute inactivation of the oncogene might cause growth cessation or death if the anti-mitogenic/pro-apoptotic signals decay more slowly than the mitogenic signals (for example, because of differences in mRNA and protein half-life). Adapted from Ref. 53. b | Oncogene dependency due to gene–gene interactions. Cancer cells accumulate mutations (arrows) over time that cumulatively lead to a transformed phenotype. Selection favours acquisition of mutations that are neutral or beneficial (adaptive) in the context of the mutations that preceded them. However, some of these changes might be deleterious (red arrow) were it not for the changes that preceded them. If true, correcting early genetic changes (yellow arrow) will unmask these deleterious effects. In this model, cancer cells behave like a molecular ‘house of cards’. c | Activation (indicated by bold arrow) of an oncogenic pathway diminishes selection pressure to maintain collateral signalling pathways. Silencing of these collateral pathways over time, because of genetic or epigenetic changes, leads to oncogene dependency. Adapted from Ref. 57.


Superimposed on the network abnormalities that are induced by activated oncogenes are network abnormalities that are induced by mutations at other loci. The resulting abnormalities in molecular circuitry create additional opportunities for oncogene addiction1, 2, 53, 54, including those that arise as a result of gene–gene interactions, such as synthetic lethality and extragenic suppression. Cancers arise through sequential genetic changes that ultimately convert a normal cell to a fully transformed one. These mutations are under selective pressure to be adaptive or neutral, from the point of view of the cancer, in the context of the mutations that preceded them (Fig. 4B). It seems likely, a priori, that some of the mutations that occur late in the evolution of a cancer cell might only be advantageous, or indeed even tolerated, because of the mutations that preceded them (or put another way, these mutations would be deleterious if not for the mutations that had preceded them). In the extreme case, an early A mutation might be an extragenic suppressor of the lethality that would otherwise be caused by a late B mutation (Fig. 2, right panel). If this is true, correcting the A mutation should cause death because of the acquisition of the B mutation. For example, RB1 inactivation, as described above, leads to increased E2F activity, which can stimulate S-phase entry but can also promote p53-dependent apoptosis55, 56. So, a tumour in which TP53 was already mutated might derive an additional benefit from mutating RB1 but at the price of becoming addicted to p53 loss (in the sense that restoring p53 function would lead to apoptosis).

Similarly, Mills and colleagues have suggested that oncogene addiction might arise because of the loss of collateral signalling pathways. This is due to genomic instability coupled with the loss of selection pressure to maintain the collateral signalling pathways57, a process referred to as ‘genetic streamlining’58(Fig. 4C). Collectively, these ideas suggest that the pathways that are activated early in the course of tumour progression (owing to oncogene activation or tumour-suppressor-gene inactivation) are likely to be excellent therapeutic targets because of synthetic interactions with the mutational changes that followed them. Silencing these pathways should reveal the deleterious consequences of these subsequent changes, whether these changes did or did not contribute to tumour progression. The potential interrelationship between oncogene addiction and synthetic lethality is illustrated by the phosphatase and tensin homologue (PTEN) tumour-suppressor protein, which negatively regulates the phosphatidylinositol 3-kinase (PI3K) pathway, and mTOR (mammalian target of rapamycin). PTEN-/- cells are reported be more sensitive to the antiproliferative effects of mTOR inhibitors than their wild-type counterparts59. This observation indicates that PTEN-/- cells are ‘addicted’ to PI3K–mTOR signalling, and that PTEN and mTOR have a synthetic sick relationship.

Chromosomal deletions in cancer cells lead to the loss of one or both copies of many genes. Frei suggested that cancer-cell vulnerabilities to pharmacological attack might also be gleaned by examining the functions of contiguous genes that are homozygously deleted along with tumour-suppressor genes60. For example, the gene encoding methylthioadenosine phosphorylase (MTAP) — which has a role in a salvage pathway for adenosine biosynthesis — is often co-deleted with the adjacent CDKN2A locus, which encodes the tumour-suppressor proteins INK4A and ARF on 9p21 (Ref. 61). As would be predicted, cells that lack MTAP have increased sensitivity to L-alanosine — a potent inhibitor of de novo AMP synthesis — and to an inhibitor of de novo purine-nucleotide synthesis, 6-methylmercaptopurine riboside (MMPR)62.

Kamb suggested that expression databases be mined for paralogous genes in which one or more members were underexpressed in cancer cells relative to normal cells (for example, as a result of haploinsufficiency or homozygous deletion)58. A drug that inhibited the remaining paralogue(s), but not the differentially expressed paralogue, would, theoretically, be cancer-cell selective. This approach, however, presumes that it is possible to develop drugs that can discriminate between paralogous proteins. Moreover, synthetic lethal screens in yeast indicate that paralogous pairs represent a minority of the potential synthetic lethal combinations in a cell10, 15, 63. Therefore, unbiased chemical and genetic screens are likely to be the most fruitful methods for identifying novel synthetic lethal relationships on which to base new cancer treatments.

Screens for synthetic lethal interactors

The example of topisomerase II inhibitors, as cited above, demonstrates that proteins bound to drugs might have effects that are very different from those predicted by true null mutations, or by techniques such as RNA interference (RNAi) that cause quantitative reductions in protein abundance. For example, a drug might interfere with one function of a multifunctional protein, or cause a protein to act in a dominant-negative or dominant-positive manner. For this reason, screens for synthetic lethality that are carried out using libraries of chemical compounds are likely to be complementary to screens that are carried out using genetic tools (such as RNAi or short interfering RNA; siRNA).

Chemical screens. Hartwell and Friend pioneered the idea of screening for drug-like chemicals that specifically kill yeast deletion mutants with defects in cell-cycle checkpoints or DNA repair16, 64. This paradigm can be extended to human cells. A number of groups have identified chemicals from collections of pure compounds, or that are present in complex mixtures (for example, extracts or broths), that selectively inhibit cells with cancer-relevant genetic alterations using isogenic human cell-line pairs grown in multiwell plates (Fig. 5). Schreiber and co-workers identified marine sponge extracts that preferentially inhibited the proliferation of Trp53-/- mouse embryonic fibroblasts, as determined by BROMODEOXYURIDINE (BRDU) INCORPORATION, relative to wild-type mouse embryonic fibroblasts65. However, the chemical entities responsible for these effects were not identified. Kinzler and co-workers co-cultured KRAS-mutated colon cancer cells (engineered to produce blue fluorescent protein) with a subclone in which the mutant KRAS allele was eliminated by homologous recombination (and engineered to produce yellow fluorescent protein), and monitored differential killing using the ratio of blue/yellow fluorescence66 (Fig. 6A). Several chemical entities, including a novel cytidine nucleoside, were found that selectively killed cells containing mutant KRAS. A fluorescence-based mammalian synthetic lethal assay, which was modelled after earlier yeast assays67, was also developed by Canaani and colleagues68, 69 (Fig. 6B). Leder and co-workers discovered a small molecule called F16, which selectively killsERBB2 (also known as HER2/NEU)-overexpressing mammary epithelial cells, compared with their normal counterparts70, 71. The toxicity of F16 correlates with its selective uptake in, and disruption of, mitochondria of cells that are transformed with ERBB2. Stockwell and co-workers identified a number of compounds that preferentially killed primary human cells that were transformed in vitro with human telomerase reverse transcriptase (TERT), RAS, and oncoproteins that affect pRB, p53 and/or protein phosphatase 2A (PP2A)21. Included among these were clinically useful inhibitors of topoisomerase I and II. In a focused screen of pro-apoptotic agents Quon and colleagues discovered that human cells overexpressing MYC displayed increased sensitivity to the death receptor DR5 agonist tumour-necrosis-factor-related apoptosis-inducing ligand (TRAIL) in vitroand in vivo, and linked this to p53-independent induction of DR5 by MYC72. Recent studies suggest that it is possible to screen pairwise combinations of drugs against ISOGENIC cell lines to uncover novel drug–gene and drug–drug interactions73, 74.

Figure 5 | Synthetic lethal screening with chemical or interfering RNA libraries.

Figure 5 : Synthetic lethal screening with chemical or interfering RNA libraries. Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.comIsogenic cell-line pairs that do or do not harbour a cancer-relevant mutation (in the case illustrated, the cell-line pair differs only with respect to a particular tumour-suppressor gene (TSG)) are grown in multiwell plates to which different chemical or genetic (short interfering RNAs, short hairpin RNAs or other interfering RNAs) perturbants are added. In time, such assays might be carried out using microarrays spotted with chemicals or siRNA species104, 105. A ‘hit’ is a perturbant that is cytostatic or cytotoxic to the cell with the cancer-relevant mutation (arrow). It should be noted that the interpretation of such assays needs to consider potentially confounding effects, such as differences in proliferation rate and cell-cycle distribution.


Figure 6 | Fluorescence-based mammalian synthetic lethal assay.

Figure 6 : Fluorescence-based mammalian synthetic lethal assay. Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.coma | The Kinzler method66. Isogenic cell-line pairs that do/do not harbour a cancer-relevant mutation are engineered to produce blue fluorescent protein (BFP) and yellow fluorescent protein (YFP), respectively, and are co-cultured in multiwell plates to which different chemicals are added. Selective killing of blue cells is indicative of a synthetic lethal interaction (yellow well). b | The Canaani method68, 69. Cells lacking a tumour-suppressor gene (TSG) are engineered to stably produce a green fluorescent protein (GFP) with an emission wavelength of ‘1’. These cells are transfected with an unstable episomal plasmid encoding theTSG along with a GFP that has a different emission wavelength (‘2′). Retention of the episomal plasmid after exposure to chemical or genetic perturbants is indicative of a synthetic lethal relationship. WT, wild type.


The use of isogenic cell-line pairs to identify compounds that selectively kill cancer cells as a result of synthetic interactions is a powerful approach for the following reason. It is not uncommon for approx1% of the compounds in a chemical library to inhibit the growth of human cancer cells at the concentrations used in typical high-throughput screens. This translates into thousands of potential anticancer drugs from a screen conducted with 105 to 106 compounds (such as might be found at a large pharmaceutical company or public consortium). Without the use of a filter, such as differential killing in a genotype-specific manner, there are too many ‘hits’ to pursue. In the past, this has led to ‘hits’ being prioritized on the basis of factors such as ease of synthesis, potency, intellectual-property issues and the likelihood of having desirable absorption, distribution, metabolism and excretion (ADME) properties based on accepted criteria such as ‘LIPINSKI’S RULES75, 76. Although they are important, none of these latter considerations address selectivity. Furthermore, these factors can sometimes be addressed by modifying the chemical structure of the initial compound (medicinal chemistry). It would be ironic if chemicals that can selectively kill cancer cells through synthetic lethal interactions were present but missed for this reason during the countless cytotoxic screens that have been conducted since the mid-twentieth century.

A generic problem for cell-based screening of libraries of chemical compounds relates to successful target identification. In some cases, it is possible to use a chemical entity identified in such a screen to capture its protein target by affinity chromatography77,78. For chemicals that induce a phenotype in yeast, mutants that display increased or decreased resistance (fitness) can be sought79, 80. Such mutants often provide clues as to the pathways that are affected by a compound, and therefore its potential target (or targets). A conceptually attractive approach to target identification would be to generate compendia of molecular signatures (for example, gene-expression profiles) for various loss-of-function mutations in a suitable host (for example, yeast or human cells)81. The signature generated by the compound of interest could then be compared in silico to the compendium, with the rationale that the compound signature and target-disruption signature should be near(est) neighbours in an ideal situation. The search for targets of chemicals identified in cell-based synthetic lethal screens should also be expedited by a knowledge of the genes that score as synthetic lethal in genetic screens carried out in model organisms and human cells, as described below.

Genetic screens. In the past, genetic screens for synthetic lethal interactors have been largely relegated to model organisms such as yeast, the fruitfly Drosophila melanogaster and the wormCaenorhabditis elegans that are amenable to forward-genetic approaches. Typically, these approaches have combined random mutagenesis with phenotypic screens, reflecting the retention of the query gene linked to a suitable reporter. Synthetic lethal screens in yeast have been invaluable for elucidating certain principles surrounding synthetic lethal interactions. Unfortunately, many tumour-suppressor genes and oncogenes do not have clear yeast orthologues. Although forward-genetic screens are more cumbersome in fruitflies and worms than in yeast, they offer the advantage that their genomes do contain orthologues of most human cancer genes. In worms the RB1 orthologue, lin-35, has been well studied in the context of vulvar development82. Fay and co-workers reported that a gene encoding a ubiquitin-conjugating enzyme related to human UBCH7 is synthetic lethal to lin-35 (Ref. 83), as is the worm homologue of CDH1 (Ref. 84). Using a fruitfly-based screen in which the fruitfly RB1-like geneRbf1 was conditionally inactivated in the eye, Belvin and co-workers discovered that RBF1 is synthetic lethal to a novel prolyl isomerase85. It is not yet known whether these synthetic lethal interactions will hold true in all cell types, nor whether they will hold true across species.

However, forward-genetic approaches such as these are now giving way to genome-wide reverse-genetic approaches. Successful studies have been carried out in yeast (Box 1) but, for the reasons cited above, metazoan models are usually more appropriate than yeast for synthetic lethal screens for human cancer genes.

RNAi is a powerful method for silencing genes in worms and fruitflies, and collections of interfering RNAs have been created to facilitate high-throughput genome-wide screens in these organisms86, 87, 88, 89 (for an excellent review, see Ref. 90). RNAi can be conveniently achieved in wild-type or mutant worms by growing them on lawns of Escherichia coli carrying a plasmid that produces the interfering RNAs of interest, which are then ingested. Alternatively, interfering RNAs can be delivered to worms by soaking them in a solution that contains the appropriate molecules. An interfering RNA that exacerbated the mutant phenotype without affecting wild-type animals would indicate a synthetic lethal, or synthetic sick, interaction. High-throughput screens have also been conducted to identify interfering RNAs that inhibit the proliferation of fruitfly cells grown in multiwell plates88. Such screens could easily be adapted to carry out synthetic lethal screens. In this scenario, the identification of interfering RNAs that do not affect wild-type fruitfly cells but kill fruitfly cells in which the gene of interest was mutated or silenced would be desired. If required, silencing could be accomplished by simultaneously administering two interfering RNAs (one corresponding to the query gene and one corresponding to the gene of interest).

Many cancer-relevant genes are linked to specific types of cancer despite being ubiquitously expressed and performing functions that are thought to be generic rather than tissue specific. In addition, there are now many examples where different phenotypes have been observed following heterozygous inactivation of a particular tumour-suppressor gene in both mice and humans. These observations indicate that context, with respect to cell-type and species, is important. As a corollary, they indicate that synthetic lethal relationships ultimately need to be discovered or validated in relevant human cells, and that caution needs to be exercised when extrapolating cell-culture results to intact organisms. In the past, the use of RNAi in mammalian cells was problematic because double-stranded RNA elicits an antiviral response on entry into mammalian cells. In 2001, however, Tuschl and co-workers showed that siRNAs can be used to silence genes in mammalian cells without triggering a nonspecific host response91. Soon thereafter several groups showed that the actions of siRNAs in cells can be mimicked with short hairpin RNAs (shRNAs) encoded by plasmid or viral vectors92, 93, 94,95, 96. siRNA libraries and shRNA vector libraries are being created, and proof-of-concept experiments indicate that these libraries can be used to carry out genome-wide phenotypic screens in mammalian cells (including human cells)97, 98, 99,100. In theory, these libraries could be used to carry out synthetic lethal screens using isogenic cell-line pairs, scoring for siRNA (or shRNA) species that specifically kill cells with a cancer-relevant mutation in a one well/one siRNA (or shRNA) species format (Fig. 5). Alternatively, several groups are incorporating DNA ‘bar codes’ (Box 1) into shRNA vectors, modelled after the use of DNA bar codes in yeast and E. coli (or have used the shRNA sequence itself as a bar code)97, 98. If successful, it should be possible to infect isogenic cell-line pairs with pools of vectors encoding different shRNAs, and then identify those shRNAs that cause a fitness defect specifically in those cells that harbour the cancer-relevant mutation under investigation.

Combination therapy

Random mutations that lead to gene inactivation should theoretically decrease the genetic buffering capacity of an individual cancer cell. As outlined above, therapies predicated on synthetic lethal relationships are one way to exploit this. At the same time, random mutations and genome plasticity, viewed at the level of a tumour, markedly increase the likelihood that rare therapy-resistant subclones will emerge. Decades of clinical experience, including recent examples of imatinib mesylate resistance101, 102, as well as tumour models incorporating the use of conditionally expressed oncogenes103, support this view. A 1-cm3 tumour already contains >109 cells. So, the likelihood of clinical success will increase with early diagnosis (to minimize the number of cells in the pool from which resistant cells might arise) and the use of effective drug combinations. The use of drug combinations to minimize chemotherapeutic resistance is a well-established pharmaceutical principle. It is based on the knowledge that the probability of a given cell being simultaneously resistant to a combination of non-cross-resistant drugs varies as the product of the probabilities of becoming resistant to each of the individual components. The choice of which drugs to combine might be based on a knowledge of cancer molecular biology (for example, by simultaneously targeting two or more cancer-relevant mutations), empirical testing (for example, by systematically testing combinations of active agents for additive or synergistic effects) or both.

Implications and future directions

Over the decades, the medical therapy of metastatic cancer has, with a few notable exceptions, been a frustrating and often futile exercise. This has contributed to the view that each mutation within a cancer cell is another plate of armour that serves as a barrier to successful therapy. However, our empirical knowledge of the susceptibility of cancer cells to drugs in humans stems from an armamentarium that was largely discovered and developed using the same paradigm. Moreover, there is every reason to believe that certain genetic changes within cancer cells will create liabilities under the appropriate conditions. There are now tools to systematically search for mutated oncogenes that encode molecules, such as kinases, that can be targeted by drugs, as well as the tools to reveal vulnerabilities created by synthetic lethal interactions. Understanding how the phenotypes created by cancer genotypes (for example, tumour type and resistance to therapy), as well as synthetic lethal relationships, are influenced by contextual differences (for example, cell type and species) remains a formidable task. Nonetheless, we are clearly poised to move away from empirically discovered cytotoxics and towards new agents that are based on a knowledge of cancer genetics and a more sophisticated view of gene–gene interactions.

top

Acknowledgements

I would like to thank S. Elledge, A. Reddy, M. Tyers, P. Silver and M. Vidal for their critical reading of this manuscript and/or helpful comments. I apologize to colleagues whose work was not cited due to space limitations or my ignorance. Dedicated to the memory of Nancy P. Kaelin.

Competing interests statement

The author declares no competing financial interests.

top

References

  1. Kaelin, W. G. Jr. Choosing anticancer drug targets in the postgenomic era. J. Clin. Invest. 104, 1503–1506 (1999).

  2. Reddy, A. & Kaelin, W. G. Jr. Using cancer genetics to guide the selection of anticancer drug targets. Curr. Opin. Pharmacol. 2, 366–373 (2002).

  3. Kaelin, W. G. Jr. Gleevec: prototype or outlier? Sci. STKE2004, PE12 (2004).
    References 1–3 provide counter-arguments to naysayers who suggest that genetically complex cancers will never be successfully treated with drugs.

  4. Hartman, J. T., Garvik, B. & Hartwell, L. Principles for the buffering of genetic variation. Science 291, 1001–1004 (2001).

  5. Guarente, L. Synthetic enhancement in gene interaction: a genetic tool come of age. Trends Genet. 9, 362–366 (1993).

  6. Kamb, A. Mutation load, functional overlap, and synthetic lethality in the evolution and treatment of cancer. J. Theor. Biol. 223, 205–213 (2003).
    This paper and reference 58 are thoughtful essays on maladaptive genetic changes in cancer cells that might render them vunerable to pharmacological attack.

  7. Friend, S. & Oliff, A. Emerging uses for genomic information in drug discovery. N. Engl. J. Med. 338, 125–126 (1998).

  8. Dobzhansky, T. Genetics of natural populations. XIII. Recombination and variability in populations of Drosophila pseudoobscura. Genetics 31, 269–290 (1946).

  9. Lucchesi, J. C. Synthetic lethality and semi-lethality among functionally related mutants of Drosophila melanogaster.Genetics 59, 37–44 (1968).

  10. Sharom, J. R., Bellows, D. S. & Tyers, M. From large networks to small molecules. Curr. Opin. Chem. Biol. 8, 81–90 (2004).
    Excellent introduction to systems biology as applied to cancer and cancer pharmacology.

  11. Kroll, E. S., Hyland, K. M., Hieter, P. & Li, J. J. Establishing genetic interactions by a synthetic dosage lethality phenotype. Genetics 143, 95–102 (1996).

  12. Measday, V. & Hieter, P. Synthetic dosage lethality. Methods Enzymol. 350, 316–326 (2002).

  13. Li, J. J. & Herskowitz, I. Isolation of ORC6, a component of the yeast origin recognition complex by a one-hybrid system.Science 262, 1870–1874 (1993).

  14. Tong, A. H. et al. Systematic genetic analysis with ordered arrays of yeast deletion mutants. Science 294, 2364–2368 (2001).

  15. Tong, A. H. et al. Global mapping of the yeast genetic interaction network. Science 303, 808–813 (2004).
    References 14 and 15 provide a glimpse into the complexity of synthetic lethal networks in yeast.

  16. Hartwell, L., Szankasi, P., Roberts, C., Murray, A. & Friend, S.Integrating genetic approaches into the discovery of anticancer drugs. Science 278, 1064–1068 (1997).
    This seminal paper argues that synthetic lethal interactions be exploited to arrive at safer, more efficacious cancer drugs.

  17. Sellers, W. R. & Kaelin, W. G. Jr. Role of the retinoblastoma protein in the pathogenesis of human cancer. J. Clin. Oncol.15, 3301–3312 (1997).

  18. Nip, J. et al. E2F-1 cooperates with topoisomerase II inhibition and DNA damage to selectively augment p53-independent apoptosis. Mol. Cell. Biol. 17, 1049–1056 (1997).

  19. Almasan, A. et al. Deficiency of retinoblastoma protein leads to inappropriate S-phase entry, activation of E2F-responsive genes, and apoptosis. Proc. Natl Acad. Sci. USA 92, 5436–5440 (1995).

  20. Banerjee, D. et al. Role of E2F-1 in chemosensitivity. Cancer Res. 58, 4292–4296 (1998).

  21. Dolma, S., Lessnick, S. L., Hahn, W. C. & Stockwell, B. R.Identification of genotype-selective antitumor agents using synthetic lethal chemical screening in engineered human tumor cells. Cancer Cell 3, 285–296 (2003).

  22. Evan, G. I. & Vousden, K. H. Proliferation, cell cycle and apoptosis in cancer. Nature 411, 342–348 (2001).

  23. Zaika, A., Irwin, M., Sansome, C. & Moll, U. M. Oncogenes induce and activate endogenous p73 protein. J. Biol. Chem.276, 11310–11316 (2001).

  24. Meng, R., Phillips, P. & El-Deiry, W. p53-independent increase in E2F-1 expression enhances the cytoxic effects of etoposide and of adriamycin. Intl J. Oncol. 14, 5–14 (1999).

  25. Irwin, M. S. et al. Chemosensitivity linked to p73 function.Cancer Cell 3, 403–410 (2003).

  26. Rutherford, S. L. & Lindquist, S. HSP90 as a capacitor for morphological evolution. Nature 396, 336–342 (1998).

  27. Isaacs, J. S., Xu, W. & Neckers, L. Heat shock protein 90 as a molecular target for cancer therapeutics. Cancer Cell 3, 213–217 (2003).

  28. Workman, P. Altered states: selectively drugging the HSP90 cancer chaperone. Trends Mol. Med. 10, 47–51 (2004).

  29. Neckers, L. & Neckers, K. Heat-shock protein 90 inhibitors as novel cancer chemotherapeutics – an update. Expert Opin. Emerg. Drugs 10, 137–149 (2005).

  30. Goldberg, A. L. Protein degradation and protection against misfolded or damaged proteins. Nature 426, 895–899 (2003).

  31. Rajkumar, S. V., Richardson, P. G., Hideshima, T. & Anderson, K. C. Proteasome inhibition as a novel therapeutic target in human cancer. J. Clin. Oncol. 23, 630–639 (2005).

  32. Krek, W., Xu, G., & Livingston, D. M. Cyclin A-kinase regulation of E2F1 DNA binding function underlies suppression of an S phase checkpoint. Cell 83, 1149–1158 (1995).

  33. Dynlacht, B. D., Flores, O., Lees, J. A. & Harlow, E.Differential regulation of E2F transactivation by cyclin/CDK complexes. Genes Dev. 8, 1772–1786 (1994).

  34. Krek, W. et al. Negative regulation of the growth-promoting transcription factor E2F-1 by a stably bound cyclin A-dependent protein kinase. Cell 78, 161–172 (1994).

  35. Xu, M., Sheppard, K. A., Peng, C-Y., Yee, A. S. & Piwnica-Worms, H. Cyclin A/CDK2 binds directly to E2F1 and inhibits the DNA-binding activity of E2F1/DP1 by phosphorylation.Mol. Cell. Biol. 14, 8420–8431 (1994).

  36. Parr, M. J. et al. Tumor-selective transgene expression in vivomediated by an E2F-responsive adenoviral vector. Nature Med. 3, 1145–1149 (1997).

  37. Chen, Y. et al. Selective killing of transformed cells by cyclin/cyclin-dependent kinase 2 antagonists. Proc. Natl Acad. Sci. USA 96, 4325–4329 (1999).

  38. Chen, W., Lee, J., Cho, S. Y. & Fine, H. A. Proteasome-mediated destruction of the cyclin A/cyclin-dependent kinase 2 complex suppresses tumor cell growth in vitro and in vivo.Cancer Res. 64, 3949–3957 (2004).

  39. Mendoza, N. et al. Selective cyclin-dependent kinase 2/cyclin A antagonists that differ from ATP site inhibitors block tumor growth. Cancer Res. 63, 1020–1024 (2003).

  40. Tetsu, O. & McCormick, F. Proliferation of cancer cells despite CDK2 inhibition. Cancer Cell 3, 233–245 (2003).

  41. Berthet, C., Aleem, E., Coppola, V., Tessarollo, L. & Kaldis, P.CDK2 knockout mice are viable. Curr. Biol. 13, 1775–1785 (2003).

  42. Ortega, S. et al. Cyclin-dependent kinase 2 is essential for meiosis but not for mitotic cell division in mice. Nature Genet.35, 25–31 (2003).

  43. Schlegel, R. & Pardee, A. B. Caffeine-induced uncoupling of mitosis from the completion of DNA replication in mammalian cells. Science 232, 1264–1266 (1986).

  44. Nishimoto, T., Ishida, R., Ajiro, K., Yamamoto, S. & Takahashi, T. The synthesis of protein(s) for chromosome condensation may be regulated by a post-transcriptional mechanism. J. Cell. Physiol. 109, 299–308 (1981).

  45. Hall-Jackson, C. A., Cross, D. A., Morrice, N. & Smythe, C.ATR is a caffeine-sensitive, DNA-activated protein kinase with a substrate specificity distinct from DNA-PK. Oncogene 18, 6707–6713 (1999).

  46. Sarkaria, J. N. et al. Inhibition of ATM and ATR kinase activities by the radiosensitizing agent, caffeine. Cancer Res.59, 4375–4382 (1999).

  47. Nghiem, P., Park, P., Kim, Y., Vaziri, C. & Schreiber, S. ATR inhibition selectively sensitizes G1 checkpoint-deficient cells to lethal premature chromatin condensation. Proc. Natl Acad. Sci. USA 98, 9092–9097 (2001).

  48. Sordella, R., Bell, D. W., Haber, D. A. & Settleman, J.Gefitinib-sensitizing EGFR mutations in lung cancer activate anti-apoptotic pathways. Science 305, 1163–1167 (2004).

  49. Lynch, T. J. et al. Activating mutations in the epidermal growth factor receptor underlying responsiveness of non-small-cell lung cancer to gefitinib. N. Engl. J. Med. 350, 2129–2139 (2004).

  50. Paez, J. G. et al. EGFR mutations in lung cancer: correlation with clinical response to gefitinib therapy. Science 304, 1497–1500 (2004).

  51. Pao, W. et al. EGF receptor gene mutations are common in lung cancers from ‘never smokers’ and are associated with sensitivity of tumors to gefitinib and erlotinib. Proc. Natl Acad. Sci. USA 101, 13306–13311 (2004).

  52. Weinstein, I. B. et al. Disorders in cell circuitry associated with multistage carcinogenesis: exploitable targets for cancer prevention and therapy. Clin. Cancer Res. 3, 2696–2702 (1997).

  53. Weinstein, I. B. Disorders in cell circuitry during multistage carcinogenesis: the role of homeostasis. Carcinogenesis 21, 857–864 (2000).

  54. Weinstein, I. B. Cancer. Addiction to oncogenes — the Achilles heal of cancer. Science 297, 63–64 (2002).
    References 52–54, introduced the term ‘oncogene addiction’.

  55. Adams, P. & Kaelin, W. J. Jr. The cellular effects of E2F overexpression. Curr. Top. Microbiol. Immunol. 208, 79–93 (1996).

  56. Sherr, C. The Pezcoller lecture: cancer cell cycles revisited.Cancer Res. 60, 3689–3695 (2000).

  57. Mills, G., Lu, Y. & Kohn, E. Linking molecular therapeutics to molecular diagnostics: inhibition of the FRAP/RAFT/TOR component of the PI3K pathway preferentially blocks PTEN mutant cells in vitro and in vivo. Proc. Natl Acad. Sci. USA98, 10031–10033 (2001).

  58. Kamb, A. Consequences of nonadaptive alterations in cancer.Mol. Biol. Cell 14, 2201–2205 (2003).

  59. Neshat, M. et al. Enhanced sensitivity of PTEN-deficient tumors to inhibition of FRAP/mTOR. Proc. Natl Acad. Sci. USA98, 10314–10319 (2001).

  60. Frei, E. D. Gene deletion: a new target for cancer chemotherapy. Lancet 342, 662–664 (1993).

  61. Cairns, P. et al. Frequency of homozygous deletion at p16/CDKN2 in primary human tumours. Nature Genet. 11, 210–212 (1995).

  62. Li, W. et al. Status of methylthioadenosine phosphorylase and its impact on cellular response to L-alanosine and methylmercaptopurine riboside in human soft tissue sarcoma cells. Oncol. Res. 14, 373–379 (2004).

  63. Wong, S. L. et al. Combining biological networks to predict genetic interactions. Proc. Natl Acad. Sci. USA 101, 15682–15687 (2004).

  64. Simon, J. A. et al. Differential toxicities of anticancer agents among DNA repair and checkpoint mutants of Saccharomyces cerevisiae. Cancer Res. 60, 328–333 (2000).

  65. Stockwell, B., Haggarty, S. & Schreiber, S. High-throughput screening of small molecules in miniaturized mammalian cell-based assays involving post-translational modifications.Chem. Biol. 6, 71–83 (1999).
    References 64 and 65 are two early examples of using isogenic cell lines to isolate compounds that kill cells in a genotype-specific manner.

  66. Torrance, C., Agrawal, V., Vogelstein, B. & Kinzler, K. Use of isogenic human cancer cells for high-throughput screening and drug discovery. Nature Biotechnol. 19, 940–945 (2001).

  67. Bender, A. & Pringle, J. R. Use of a screen for synthetic lethal and multicopy suppressee mutants to identify two new genes involved in morphogenesis in Saccharomyces cerevisiae. Mol. Cell. Biol. 11, 1295–1305 (1991).

  68. Simons, A., Dafni, N., Dotan, I., Oron, Y. & Canaani, D.Establishment of a chemical synthetic lethality screen in cultured human cells. Genome Res. 11, 266–273 (2001).

  69. Simons, A., Dafni, N., Dotan, I., Oron, Y. & Canaani, D.Genetic synthetic lethality screen at the single gene level in cultured human cells. Nucleic Acids Res. 29, E100 (2001).

  70. Fantin, V. R. & Leder, P. F16, a mitochondriotoxic compound, triggers apoptosis or necrosis depending on the genetic background of the target carcinoma cell. Cancer Res. 64, 329–336 (2004).

  71. Fantin, V. R., Berardi, M. J., Scorrano, L., Korsmeyer, S. J. & Leder, P. A novel mitochondriotoxic small molecule that selectively inhibits tumor cell growth. Cancer Cell 2, 29–42 (2002).

  72. Wang, Y. et al. Synthetic lethal targeting of MYC by activation of the DR5 death receptor pathway. Cancer Cell 5, 501–512 (2004).

  73. Haggarty, S. J., Clemons, P. A. & Schreiber, S. L. Chemical genomic profiling of biological networks using graph theory and combinations of small molecule perturbations. J. Am. Chem. Soc. 125, 10543–10545 (2003).

  74. Borisy, A. A. et al. Systematic discovery of multicomponent therapeutics. Proc. Natl Acad. Sci. USA 100, 7977–7982 (2003).

  75. Lipinski, C. A., Lombardo, F., Dominy, B. W. & Feeney, P. J.Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv. Drug Deliv. Rev. 46, 3–26 (2001).

  76. Lipinski, C. A. Drug-like properties and the causes of poor solubility and poor permeability. J. Pharmacol. Toxicol. Methods 44, 235–249 (2000).

  77. Knockaert, M. et al. Intracellular targets of cyclin-dependent kinase inhibitors: identification by affinity chromatography using immobilised inhibitors. Chem. Biol. 7, 411–422 (2000).

  78. Hultsch, T., Albers, M. W., Schreiber, S. L. & Hohman, R. J.Immunophilin ligands demonstrate common features of signal transduction leading to exocytosis or transcription. Proc. Natl Acad. Sci. USA 88, 6229–6233 (1991).

  79. Baetz, K. et al. Yeast genome-wide drug-induced haploinsufficiency screen to determine drug mode of action.Proc. Natl Acad. Sci. USA 101, 4525–4230 (2004).

  80. Giaever, G. et al. Genomic profiling of drug sensitivities via induced haploinsufficiency. Nature Genet. 21, 278–283 (1999).

  81. Marton, M. et al. Drug target validation and identification of secondary drug target effects using DNA microarrays. Nature Med. 4, 1293–1301 (1998).

  82. Lu, X. & Horvitz, H. R. lin-35 and lin-53, two genes that antagonize a C. elegans Ras pathway, encode proteins similar to RB and its binding protein RBAp48. Cell 95, 981–991 (1998).

  83. Fay, D. S., Large, E., Han, M. & Darland, M. lin-35/Rb andubc-18, an E2 ubiquitin-conjugating enzyme, function redundantly to control pharyngeal morphogenesis in C. elegans. Development 130, 3319–3330 (2003).

  84. Fay, D. S., Keenan, S. & Han, M. fzr-1 and lin-35/Rb function redundantly to control cell proliferation in C. elegans as revealed by a nonbiased synthetic screen. Genes Dev. 16, 503–517 (2002).

  85. Edgar, K. A. et al Synthetic lethality of retinoblastoma mutant cells in the Drosophila eye by mutation of a novel peptidyl prolyl isomerase gene. Genetics 170, 161–171 (2005).

  86. Kamath, R. S. et al. Systematic functional analysis of theCaenorhabditis elegans genome using RNAi. Nature 421, 231–237 (2003).

  87. Ashrafi, K. et al. Genome-wide RNAi analysis ofCaenorhabditis elegans fat regulatory genes. Nature 421, 268–272 (2003).

  88. Cherry, S. et al. Genome-wide RNAi screen reveals a specific sensitivity of IRES-containing RNA viruses to host translation inhibition. Genes Dev. 19, 445–452 (2005).

  89. Rual, J. F. et al. Toward improving Caenorhabditis elegansphenome mapping with an ORFeome-based RNAi library.Genome Res. 14, 2162–2168 (2004).

  90. Willingham, A. T., Deveraux, Q. L., Hampton, G. M. & Aza-Blanc, P. RNAi and HTS: exploring cancer by systematic loss-of-function. Oncogene 23, 8392–8400 (2004).

  91. Elbashir, S. et al. Duplexes of 21-nucleotide RNAs mediate RNA interference in cultured mammalian cells. Nature 411, 494–498 (2001).

  92. Brummelkamp, T. R., Bernards, R. & Agami, R. Stable suppression of tumorigenicity by virus-mediated RNA interference. Cancer Cell 2, 243–247 (2002).

  93. Brummelkamp, T. R., Bernards, R. & Agami, R. A system for stable expression of short interfering RNAs in mammalian cells. Science 296, 550–553 (2002).

  94. Lee, N. S. et al. Expression of small interfering RNAs targeted against HIV-1 rev transcripts in human cells. Nature Biotechnol. 20, 500–505 (2002).

  95. Paddison, P. J., Caudy, A. A., Bernstein, E., Hannon, G. J. & Conklin, D. S. Short hairpin RNAs (shRNAs) induce sequence-specific silencing in mammalian cells. Genes Dev. 16, 948–958 (2002).

  96. Sui, G. et al. A DNA vector-based RNAi technology to suppress gene expression in mammalian cells. Proc. Natl Acad. Sci. USA 99, 5515–5520 (2002).

  97. Berns, K. et al. A large-scale RNAi screen in human cells identifies new components of the p53 pathway. Nature 428, 431–437 (2004).

  98. Paddison, P. J. et al. A resource for large-scale RNA-interference-based screens in mammals. Nature 428, 427–431 (2004).
    References 97 and 98 suggest that it should eventually be possible to carry out synthetic lethal screens in isogenic human cell-line pairs using bar-coded shRNA libraries.

  99. Shirane, D. et al. Enzymatic production of RNAi libraries from cDNAs. Nature Genet. 36, 190–196 (2004).

  100. Aza-Blanc, P. et al. Identification of modulators of TRAIL-induced apoptosis via RNAi-based phenotypic screening. Mol. Cell 12, 627–637 (2003).

  101. Gorre, M. et al. Clinical resistance to STI-571 cancer therapy caused by BCRABL gene mutation or amplification. Science293, 876–880 (2001).

  102. Shah, N. P. et al. L. Multiple BCR–ABL kinase domain mutations confer polyclonal resistance to the tyrosine kinase inhibitor imatinib (STI571) in chronic phase and blast crisis chronic myeloid leukemia. Cancer Cell 2, 117–125 (2002).

  103. Jonkers, J. & Berns, A. Oncogene addiction: sometimes a temporary slavery. Cancer Cell 6, 535–538 (2004).

  104. Bailey, S. N., Sabatini, D. M. & Stockwell, B. R. Microarrays of small molecules embedded in biodegradable polymers for use in mammalian cell-based screens. Proc. Natl Acad. Sci. USA101, 16144–16149 (2004).

  105. Wheeler, D. B. et al. RNAi living-cell microarrays for loss-of-function screens in Drosophila melanogaster cells. Nature Methods 1, 127–132 (2004).

  106. Ooi, S. L., Shoemaker, D. D. & Boeke, J. D. DNA helicase gene interaction network defined using synthetic lethality analyzed by microarray. Nature Genet. 35, 277–286 (2003).
    Describes the use of DNA bar codes coupled with oligonucleotide microarrays to conduct synthetic lethal assays in yeast.

  107. Shoemaker, D. D., Lashkari, D. A., Morris, D., Mittmann, M. & Davis, R. W. Quantitative phenotypic analysis of yeast deletion mutants using a highly parallel molecular bar-coding strategy. Nature Genet. 14, 450–456 (1996).

  108. Winzeler, E. A. et al. Functional characterization of the S. cerevisiae genome by gene deletion and parallel analysis.Science 285, 901–906 (1999).

  109. Eason, R. G. et al. Characterization of synthetic DNA bar codes in Saccharomyces cerevisiae gene-deletion strains.Proc. Natl Acad. Sci. USA 101, 11046–11051 (2004).

  110. Hensel, M. et al. Simultaneous identification of bacterial virulence genes by negative selection. Science 269, 400–403 (1995).

Author affiliations

  1. Howard Hughes Medical Institute, 44 Binney Street, Mayer 457, Boston, Massachusetts 02115, USA.
    Email: william_kaelin@dfci.harvard.edu