| Human Molecular Genetics | Pages |
©1999 Oxford University Press |
Forward genetics in mammalian cells: functional approaches to gene discovery
Selection Of Mutant Cell Lines
Obtaining a starting cell line
Mutagenesis and selection
Dominance tests and complementation groups
Cloning by complementation
Uses of mutant cell lines
Examples Of Forward Genetics In Mutagenized Mammalian Cells
Interferon-stimulated pathways
Abnormal constitutive expression
Mutant cell lines unresponsive to IL-1 or TNF-[alpha]
Responses to TGF-[beta]
Mutations in the T cell receptor
Loss of HLA expression or antigen presentation
Genetic analysis of integrin function
Expression Cloning
Transfection of cellular DNA for functional gene discovery
Expression selection of retroviral cDNA libraries
Retroviral insertional mutagenesis plus activated proto-oncogene: an approach to identify new transforming genes
Functional Cloning Of Recessive Genes
Use of antisense expression libraries to clone pro-apoptotic genes
Random gene inactivation by antisense promoter insertions
Genetic suppressor elements
GSEs from individual genes: proof of the principle
Using the GSE approach to identify new drug sensitivity genes
GSEs and identification of novel candidate tumor suppressors
The SETGAP technique: isolation of growth suppressing GSEs
Perspectives
Note Added In Proof
Somatic cell mutants resistant to retrovirus replication
Acknowledgements
References
Forward genetics in mammalian cells: functional approaches to gene discovery
Received June 9, 1999; Accepted June 16, 1999
Definitive proof of function in biological systems requires genetic analysis. Only when the loss of a particular protein corresponds to the loss of a specific function can one be sure that the protein truly affects the function. Changing the pattern of gene expression through random mutagenesis or by introducing expression libraries, followed by selection of mutant or variant cells and identification of a missing or overexpressed protein, has the power to reveal or confirm the roles of specific components of signaling pathways and to provide mutant cell lines and cDNA reagents to be used in defining detailed mechanisms through structure-function analyses. These examples of forward genetics contrast with reverse genetic approaches, where the function of a known gene product is explored by knockout or replacement. Here we review a broad range of techniques that have been used to alter gene expression randomly in mammalian cells, with examples of specific discoveries that have resulted from these applications of forward genetics.
SELECTION OF MUTANT CELL LINES
Random alteration of gene expression in forward genetics must be coupled with a highly effective means of selecting or sorting variant cells. To obtain mutant mammalian cell lines, one must: (i) set up a selection or screen in which the expression of a marker gene is induced or in which constitutive expression is lost; (ii) mutagenize the cells; (iii) select mutant clones and characterize them for dominance and complementation of similarly derived but independent mutant clones; (iv) clone complementing DNAs from an expression library; (v) analyze expression of the mRNA and protein encoded by the complementing DNA in a set of similar but independent mutant clones; and (vi) select the best clone for a structure-function analysis of the role of the target protein in the pathway. The same basic cell line used for mutagenesis can also be used with cDNA libraries intended to induce or suppress function by expression of full-length proteins, protein fragments or antisense RNAs.
Obtaining a starting cell line
The great majority of mutants derived from cells treated with the frameshift mutagen ICR-191 are recessive, lacking the target proteins. From this result, it is obvious that ploidy is an important consideration in choosing cell lines for analysis, since the number of alleles that must be inactivated to cause complete loss of expression will surely affect the mutation frequency. We have used pseudodiploid cells only, since the mutation frequency is very likely to be lower in hyperploid cells. Chromosome loss can be stimulated by treating cells with cytochalasin or other drugs that disrupt microtubules and it may be possible to increase mutation frequencies by this stratagem. If expression of a cell surface protein is induced in response to ligand binding, the cells can be used directly, first to obtain mutants and then to complement them, with the aid of the fluorescence-activated cell sorter (FACS) or by complement-mediated lethal selection. This strategy has worked well to obtain mutant cells that fail to induce cell surface expression of endogenous class MHC proteins (see below). The appearance on the cell surface of a protein not normally found in the target cells, expressed from an inducible construct, can also be useful (1), but in our recent experience this approach has been inferior to lethal selection, because four to five rounds of FACS sorting are required, which is slow and relatively expensive, and because unstable mutants are sometimes obtained (see below). Lethal selection with complement might give better results.
Lethal selection has also been used in a system in which the expression of a selectable marker is regulated by the upstream element of a gene that responds to induction. For example, to obtain mutants unresponsive to IFN-[alpha],[beta], expression of the Escherichia coli guanine phosphoribosyltransferase (gpt) gene was driven by 1.8 kb of upstream sequence from the human 6-16 gene, which is tightly regulated by IFN-[alpha],[beta]. In a genetic experiment, one does not need to understand how the upstream element is regulated, only that it is sufficient to drive inducible expression. Natural promoters are complex and are often regulated by more than one pathway, a property that may be either useful or an unwanted complication. An alternative, more focused approach is to use a small, well-defined element (i.e. a [kappa]B site), usually as a multimer, to drive the expression of a selectable marker in response to the activation of a single transcription factor.
An interesting additional possibility, demonstrated by the work of Gogos et al. (2), employs the retroviral gene-trap vector pRSAHyTk, in which a promoterless hygromycin resistance-thymidine kinase (TK) fusion gene is placed in front of a splice acceptor sequence. When inserted into an intron, expression of the fusion gene comes under the control of the 5[prime] regulatory region of the targeted endogenous gene. Hygromycin or gancyclovir can then be used to select cell clones in which expression of the marker is off in the absence of inducer or on in its presence. This stratagem is efficient and has several additional important advantages: no prior knowledge or selection of the inducible gene is required and full-length endogenous promoters are used, ensuring that regulation of the marker gene will be physiological.
Since the selections are based on protein expression, the pathways targeted need not depend only on transcriptional regulation. As an example, one might be able to obtain mutations in pathways that regulate mRNA stability by using selectable markers, driven from constitutive promoters, encoded in mRNAs that include AU-rich destabilizing sequences in their 3[prime]-untranslated regions. Which markers to choose? Our experience with cell surface expression of cd2 or cd4 has ranged from success (1) to frustration. Using the IL-1-responsive E-selectin promoter to regulate expression of cd2, cd4 or both (separately transfected) in human 293 cells led to the isolation of unresponsive clones after four or five rounds of FACS sorting. However, these clones were invariably unstable, yielding spontaneous revertants at a frequency (~10-5) high enough to make complementation difficult (Xiaoxia Li and G.R. Stark, unpublished data). The same promoter driving TK and zeo, in the same cells, has worked well for lethal selection (3). We suspect that metastable DNA methylation may be responsible for the instability observed. The silencing of gene expression by methylation is common in tissue culture cells and has been an issue in previous studies of mutagenesis (4). A likely explanation is that one allele of the target gene is silenced by methylation and that the other has been inactivated by the mutagen. Spontaneous reactivation of the methylated allele then yields revertants. This explanation is supported by our observation that treatment with 5-aza-2[prime]-deoxycytidine, which inhibits DNA methylation, greatly increased the reversion frequency of a metastable mutant clone (M. Agarwal and G.R. Stark, unpublished data). It may be possible to obtain stable mutants by re-mutagenizing revertant clones, but an easier solution is to rely on lethal selection, which has never led to an unstable mutant in our hands.
The E.coli gpt gene, expressed under control of the human 6-16 promoter in HPRT- HT1080 cells, has worked very well for the selection of IFN-[alpha],[beta]-unresponsive mutants with 6-thioguanine and for the selection of complemented cells with hypoxanthine/aminopterin/thymidine (HAT) medium (5). However, this scheme depends on the unusual 6-16 promoter, which is tightly off in the absence of inducer. Even a low level of basal expression, normal for many promoters, precludes the use of 6-thioguanine in our experience. A more general strategy, which also eliminates the need for HPRT- cells, employs the Herpes simplex TK gene and gancyclovir. For reasons that we do not understand, one can readily find low concentrations of gancyclovir that allow cells with a low basal expression of TK to live, but that kill cells with induced expression of ~10-fold or more (3). A variety of markers for positive selection can be used in conjunction with TK, including genes encoding zeocin resistance (3) or hygromycin resistance (2). Regulated expression of a single hybrid protein, providing both TK and hyg activities (2), is an attractive alternative.
Signaling pathways have two normal states: off in the absence of inducer and on in its presence. Mutants can be selected that do not respond to inducer (i.e. with gancyclovir for failure to induce TK) or that do not keep constitutive gene expression off (i.e. with zeocin or hygromycin in the absence of inducer). Mutant cells that express markers constitutively, obtained in several instances (see below), are likely to lack the expression of negative regulators of signaling and should be very useful in identifying and characterizing these regulators. A variant approach that has been very productive in the Stark laboratory involves setting up a cell line in which the pathway to be studied is turned on constitutively, obviating the need for induction by external stimuli, then obtaining mutants in which constitutive expression is lost. Four different examples are currently being explored. (i) The IL-1 pathway can be driven by overexpressing the receptor-proximal proteins MyD88, IRAK or TRAF6 (6). Mutagenesis should lead to the selective inactivation only of components that function downstream of these proteins, narrowing the range of mutants to be explored. A similar strategem is being used to study TNF-dependent signaling. (ii) NF[kappa]B is released from I[kappa]B and activated by serine phosphorylation in response to ras-dependent signaling (7). In a cell line overexpressing the Val12 mutant of K-ras, NF[kappa]B is active and [kappa]B-dependent genes are expressed. Mutagenesis yields cells in which this constitutive pathway has been inactivated (N. Sizemore and G.R. Stark, unpublished data). (iii) Mutants in which NF[kappa]B is constitutively free of I[kappa]B and active have been obtained in three or more complementation groups. The mutations have probably caused the loss of negative regulatory proteins (S. Sathe, X. Li and G.R. Stark, unpublished data). A second round of mutagenesis, to inactivate constitutive expression, should inactivate positive effectors in these primary mutant cell lines. (iv) The tumor suppressor p53 is overexpressed when Ras-dependent signaling is hyperactive (8,9). Constitutive p53-dependent expression of thymidine kinase is lost in mutagenized HT1080-derived cells, which have a dominant N-ras mutation and wild-type p53. Several different complementation groups have been obtained (M. Agarwal, S. DePrimo and G.R. Stark, unpublished data).
Mutagenesis and selection
We have used ICR-191, an intercalating agent that causes frameshifts and deletions, exclusively, because we wanted to maximize the loss of gene expression, to obtain recessive mutants lacking a protein required for signaling as efficiently as possible (10). Indeed, our experience has been that the great majority of mutant clones obtained with ICR-191 lack both the protein and the mRNA. The loss of mRNA expression following frameshift mutations can be explained by a process called nonsense-mediated mRNA decay, which specifically targets mRNAs with premature stop codons for degradation. The steady-state levels of transcripts of genes containing frameshift mutations may be up to 20 times lower than the wild-type levels (11). Others have employed point mutagens (EMS or MNNG), which are more likely to cause missense or nonsense mutations, with good success in obtaining recessive mutants (12-14). If one wants to obtain dominant mutants in which an altered protein inhibits function, it would probably be better to use EMS or MNNG rather than ICR-191.
Mammalian cells have two copies of most genes and it is usually necessary to inactivate both to eliminate expression of the encoded protein. When cells are treated with ICR-191 to ~50% lethality, the frequency of inactivating the single copy X-linked HPRT gene is ~10-5 and the frequency of losing expression of any one of the ~10 genes required for IFN-[alpha],[beta] signaling is 10-8-10-9, in good agreement with expectations (15). Four or five rounds of mutagenesis increase the frequency of inactivating responses to IFN-[alpha],[beta], IFN-[gamma], TNF or IL-1 to ~10-6 (3,15), allowing one to isolate tens of independent mutant clones in a straightforward manner. Although each mutant clone undoubtedly has a very large number of mutations in other genes, each grows well (selected for growth during expansion of mutagenized pools of cells) and behaves as expected for a `clean' mutation of the pathway under study, i.e. a single protein is missing and when that protein is restored the pathway functions as well as in wild-type cells.
Complementation of mutants by using expression libraries in order to clone the missing gene has been successful, but is still arduous. We are exploring an alternative stratagem, involving retrovirus-mediated insertional mutagenesis, which would avoid the need for functional complementation. It is impractical to use insertional mutagenesis de novo to inactivate two alleles of a target gene, but it may be possible to use it in conjunction with chemical mutagenesis. One can obtain a population of heavily mutagenized cells that contain no mutations in the pathway by selection with, for example, zeocin in the presence of inducer, followed by an objective assay for lack of mutants by selection with gancyclovir in the presence of inducer. This selected population will contain many cells that express only one copy of some gene required for induction and retroviral insertion can then be used to inactivate and mark that gene. The expected frequency can be estimated roughly at 10-8-10-9 (probability of inactivating a single allele after five rounds of mutagenesis ~10-4, assuming 10 target genes per pathway, assuming a target size of 10-5-10-6 of the total genome), which is within the range of an experiment.
Dominance tests and complementation groups
Mutant clones that arise from different mutagenized pools are certain to be independent and mutant clones that arise from the same pool but have distinctly different phenotypes are likely to be so. The objective placement of mutants into complementation groups and analysis of dominance can be done in a straight- forward way by analyzing the properties of heterokaryons. Different constitutive drug resistance markers are put into each of the pair of cells to be tested, followed by polyethylene glycol-mediated fusion, selection of A-B hybrids with both drugs (simultaneously applied) and analysis of the phenotype 7-10 days after fusion (10). Analysis of heterokaryons after a short time, rather than of true hybrids after a longer time, is a precaution to minimize the unwanted consequences of chromosome loss. If the heterokaryons derived from a mutant clone and its wild-type parent have a wild-type phenotype, the mutation is recessive, which has been the case in virtually all the mutants analyzed so far following mutagenesis with ICR-191. Point mutagens (EMS, MNNG, etc.) might give a detectable frequency of dominant negative mutations. Fusion of two recessive mutant clones gives a wild-type phenotype if they are in different complementation groups. We designate mutants as in the following example: I1A is unresponsive to IL-1, in complementation group 1, independent isolate A. Usually recessive mutants in different complementation groups lack different proteins, but more complexity is certainly possible, in particular when oligomeric proteins are involved.
Cloning by complementation
If the mutant clone has been obtained by random chemical mutagenesis, the only choice for cloning an unknown gene that has been inactivated is to introduce wild-type coding sequences. This has proved to be tricky and difficult, although there have been important successes. The source of a wild-type coding sequence can be genomic DNA (16), genomic DNA libraries or cDNA expression libraries (see for example ref. 17). An important practical problem is how to regulate the level of expression in a cDNA library. Signaling proteins are usually present in low concentrations and, if overexpression is deleterious to the cell, one does not expect to see complementation with a cDNA library driven by a strong promoter. Retroviral cDNA libraries are very attractive for ease of use and efficient infection and the attenuated LTR used to drive expression in many cases gives moderate levels of expression (18). Expression from genomic libraries is likely to be regulated appropriately by endogenous promoters, but identification and recovery of the complementing gene can be difficult and the unknown size of a complementing gene makes library selection uncertain at best. Specific examples of functional complementation are given below.
Uses of mutant cell lines
The interferon-insensitive mutants U3A, U4A and [gamma]2A lack Stat1, Jak1 and Jak2, respectively. Since these three proteins are used widely for signaling, the three mutant cell lines have been especially useful in helping to define the functions of the missing components in many different signaling pathways (5,19). Furthermore, expression of variants of the three proteins in the null backgrounds of the mutant cell lines has helped to define the functional motifs of each protein and to work out important mechanistic details of the pathways (5,19). Two recent examples from our laboratory follow. (i) Stat1-null cells are deficient in the constitutive expression of many different genes, including those encoding caspases 1-3 (20) and LMP2 (21). This novel function of Stat1 does not require tyrosine phosphorylation, which is essential for all known signaling functions of this protein, and involves the binding of Stat1 to other transcription factors (M. Chatterjee-Kishore and G.R. Stark, unpublished data). (ii) IFN-[gamma]-activated Stat1 dimers negatively regulate c-myc and c-jun expression directly, through GAS elements, and both Tyr701 and Ser727 must be present and phosphorylated for negative regulation. Surprisingly, in the absence of Stat1, IFN-[gamma] stimulates c-myc and c-jun through a novel pathway (R. Chilakamarti, K.C. Guo, B.R.G. Williams and G.R. Stark, submitted for publication).
EXAMPLES OF FORWARD GENETICS IN MUTAGENIZED MAMMALIAN CELLS
The systems analyzed so far are summarized in Table 1 and the strategies employed in Table 2.
Table 1. Signaling defects in mutagenized mammalian cells (laboratories in which the work was done are indicated; see text for references)
| I. Loss of response to a specific external signal (identifies proteins required for response to a defined ligand) | ||
| IFN-[alpha] | (Stark-Kerr, Flavell) | |
| IFN-[gamma] | (Stark-Kerr) | |
| IL-1 or TNF-[alpha] | (Stark, Seed) | |
| TGF-[beta] | (Howe) | |
| Antigens (via the T cell receptor) | (Weiss, Crabtree, Abraham) | |
| II. Abnormal constitutive expression of normally inducible genes (identifies negative regulatory proteins) | ||
| IFN-regulated genes | (Stark-Kerr) | |
| NF[kappa]B-regulated genes | (Stark) | |
| III. Loss of constitutive expression (identifies proteins required for signaling in known or unknown pathways) | ||
| NF[kappa]B-regulated genes | (Stark; three different examples) | |
| p53-regulated genes | (Stark) | |
| HLA | (DeMars) | |
| Antigen presented by class II MHC | (DeMars) | |
| Integrin [alpha]IIb[beta]3 | (Ginsberg) | |
Table 2. Successful mutant strategies
| Selectable or cell surface marker | Positive selection | Negative selection | Examples of pathways investigated |
| Escherichia coli gpt in HPRT-null cells | HAT medium | 6-Thioguanine | IFN-[alpha], IFN-[gamma] |
| Thymidine kinase plus dominant marker (i.e. zeo, neo, hyg) or TK-hygro hybrid protein | Zeo, neo, hyg, etc. | Gancyclovir | TNF-[alpha], IL-1, p53 |
| Endogenous cell surface protein | Antibody + FACS | Antibody + FACS or + complement | Class II induction by IFN-[gamma]; antigen presentation |
| Exogenous cell surface protein (cd2, cd4) | Antibody + FACS | Antibody + FACS or + complement | IFN-[gamma] |
Interferon-stimulated pathways
Mutant cell lines unresponsive to IFN-[alpha] were selected using IFN-responsive promoters to drive expression of selectable markers in human HT1080 fibrosarcoma cells. In total, six clones unresponsive to IFN-[alpha] and two unresponsive to IFN-[gamma] were isolated. The mutated genes encode receptor subunits, members of the Jak family of kinases, members of the signal transducer and activator of transcription (Stat) family of transcription factors and p48, a member of the IRF family of transcription factors.
Analysis of the phenotypes of cells lacking individual components provided powerful tests for their roles in signaling.
(i) The IFN-[alpha],[beta] and IFN-[gamma] signaling pathways, although discrete, share two signaling components, Stat1 and Jak1, so that cells lacking either of these proteins, selected for failure to respond to IFN-[alpha], also do not respond to IFN-[gamma]. The Stat1-null mouse lacks all responses to the IFNs and is a valuable experimental tool (22,23).
(ii) Cells lacking tyk2 are completely unresponsive to IFN-[alpha] but retain a significant residual response to IFN-[beta] (10). Therefore, these two type I IFNs, which utilize the same receptor, signal somewhat differently. Another clear difference between IFN-[alpha] and IFN-[beta] emerges from the discovery of [beta]R1, a gene that responds only to IFN-[beta] by requiring a [beta]-specific signal in addition to the activation of ISGF3, which is generated by both subtypes (24).
(iii) The availability of mutants lacking Jak1, Jak2, tyk2, Stat1 and Stat2 has helped to clarify the roles of these components in many other pathways (25). Tyk2 and Stat2 appear to function only in IFN-mediated signaling. Jak1 and Jak2 are activated by many different interleukins and growth factors and the use of mutant cell lines has helped to define the primary importance of each kinase in signaling. For example, even though both Jak1 and Jak2 are activated by growth hormone or erythropoietin, only Jak2 is required for signaling. Even though the same two kinases are activated by EGF and PDGF, only Jak1 is required. Thus, the use of mutant cell lines lacking single specific components complements and extends biochemical measurements of the phosphorylation or activation by other means of specific components.
(iv) IFN-[gamma]-dependent expression of endogenous class II MHC proteins on the cell surface was used to isolate mutant cells in which this pathway was defective (26). Further study has revealed that the mutant cell line G3A is defective in IFN-[gamma]-induced expression of the class II transactivator protein CIITA (27) and that mutant G1B lacks the function of the protein RFX5, a component of a complex transcription factor for class II genes (27). Interestingly, study of the G3A cell line has also revealed that CIITA is essential for IFN-[gamma]-induced expression of both class I and class II MHC genes (28).
Abnormal constitutive expression
Selection with HAT of mutagenized cells in the absence of IFN led to the isolation of mutant cell lines that express IFN-regulated genes constitutively (29,30), allowing negative regulators of IFN signaling pathways to be identified. The first mutant cell lines isolated expressed IFN-[alpha] or IFN-[beta] constitutively, thus accounting indirectly for their constitutive expression of IFN-regulated genes (29). More recently, using cells in the U4 complementation group that lack Jak1 and thus cannot respond to any IFN, the mutant P2.1 was isolated, which expresses several different IFN-stimulated genes at ~5-10 times their normal basal levels (30). Surprisingly, in P2.1 cells activation of the IFN-[beta] gene is defective in response to the normal inducer, double-stranded RNA (dsRNA), as is the expression of other dsRNA-dependent genes. PKR, the kinase that responds to dsRNA, is present and functional in P2.1 cells. The defect is probably in an as yet unknown component of dsRNA-mediated signaling that lies downstream of PKR. How the repression of IFN-stimulated genes and the activation of IFN genes are connected, as revealed by the phenotype of mutant P2.1, is still a mystery.
Mutant cell lines unresponsive to IL-1 or TNF-[alpha]
These cytokines are of great interest because of their central role in inflammation. Much information concerning the signaling pathways has been gathered by using protein-protein associations to identify interacting components and by studying the phosphorylation of these components in response to ligand binding. Genetic information from knockout mice is now available concerning the roles of signaling components identified in this way (for a recent example see ref. 31). Ting et al. (13) reported the isolation of mutant cell lines unresponsive to TNF-[alpha], using a multimeric [kappa]B element to drive cell surface expression of cd14. One mutant cell line lacks the serine/threonine kinase RIP, which is thus implicated in TNF-[alpha]-mediated activation of NF-[kappa]B. We have used the E-selectin promoter to drive expression of TK or zeo. These two constructs were co-transfected into human 293 cells and a clone with the desired properties was mutagenized with ICR-191. Selections with IL-1 have led to the isolation of several independent mutant clones. All the mutations are recessive and inter-mutant fusions have led to the identification of at least three different complementation groups (3). Work is in progress to identify mutants lacking known components and to complement the remaining mutants with expression libraries. Selection of the same mutagenized pools with TNF-[alpha] has yielded additional clones.
Responses to TGF-[beta]
Hocevar and Howe (32) obtained recessive mutants in three different complementation groups by negative selection of ICR-191-mutagenized HT1080 cells carrying a TGF-[beta]-responsive promoter that regulates the expression of E.coli gpt. In all three, TGF-[beta] fails to stimulate a TGF-[beta]-responsive reporter luciferase construct and gives attenuated levels of TGF-[beta]-induced plasminogen activator inhibitor-1 gene expression.
Mutations in the T cell receptor
The leukemic T cell line Jurkat has provided a series of interesting and useful mutant clones. Goldsmith and Weiss (33) combined a selection for lectin-mediated receptor-dependent growth inhibition with FACS sorting, using the Ca2+ indicator indo-1, to isolate a mutant defective in signal transduction but retaining a normal T cell receptor. Serafini et al. (34) used a NF-AT-responsive promoter to drive expression of the diphtheria toxin A chain gene. The two independent mutant clones obtained could be complemented by constitutively active calcineurin but not by the Ca2+ regulatory protein CAML, localizing the defects between these two signaling components. Since the selection employed does not permit revertants to be isolated, functional complementation was not attempted. Williams et al. (35) obtained a mutant clone lacking the known signaling component ZAP-70 using a strategy similar to that employed by Goldsmith and Weiss (33). All of these mutants have provided new information about T cell receptor-mediated signaling, a discussion of which is beyond the scope of this review. It is interesting to note that each of the three laboratories used a single round of mutagenesis successfully, with a different mutagen in each case. Since the mutant clones are recessive, the ease of obtaining them implies that Jurkat cells are functionally haploid for several loci involved in receptor-mediated signaling.
Loss of HLA expression or antigen presentation
Kavathas et al. (36) selected human lymphoblastoid cell mutants with antisera and complement for loss of a single HLA haplotype after exposure to [gamma]-radiation. The frequency (4 × 10-5) was about as expected for loss of a single allele. Closely linked loci were also lost, suggesting substantial deletions, which were confirmed by karyotypic analysis in some of the mutant clones. Restoration of the HLA-DR[alpha] and [beta] genes to a mutant cell line from which all class II genes had been lost failed to restore an associated phenotype, the lack of class I MHC-dependent antigen presentation, even though expression of HLA on the cell surface was restored (37). The additional locus that had been deleted by mutagenesis included the TAP1 transporter gene, required for antigen presentation in the context of class I MHC.
Genetic analysis of integrin function
Chinese hamster ovary (CHO) cells transfected with plasmids encoding an activated form of the human integrin [alpha]IIb[beta]3 were mutagenized with EMS and clones that had lost cell surface integrin expression were isolated using a specific antibody and FACS (14,38). CHO cells were chosen because they are hypodiploid and random loss of chromosomes was expected to facilitate the isolation of recessive mutants. This stratagem was apparently successful, since only one round of mutagenesis was necessary. Mutations of the transfected integrin genes were obtained that disrupted either ligand binding or the capacity for activation, allowing a structure-function analysis of [alpha]IIb[beta]3 to be performed. In addition, interesting mutations were obtained in CHO cell functions required for integrin activation. Expression cloning (see below) has also been used to good advantage in this system. Hughes et al. (39) selected cDNA clones that suppressed the activation of an exogeneously expressed integrin in CHO cells, thus identifying H-RAS and RAF1 as negative regulators. Fenczik et al. (40) first suppressed integrin activation by overexpressing the cytoplasmic domain of integrin [beta]1, then complemented this defect with a cDNA expression library. CD98 was thus identified as a regulator of integrin-ligand interaction. Ramos et al. (41) identified PEA-15 as a protein that blocks RAS-mediated suppression of integrin activation, again by expression cloning. This series of studies illustrates well the potential power of combining mutagenesis and expression cloning in a variety of settings to explore a single pathway. See Note added in proof.
EXPRESSION CLONING
An overview is given in Figure 1. Cellular phenotypes can be divided into those caused by overexpression or dysfunction of genes (dominant mutations) and those that result from loss of gene expression (recessive mutations). Activation by mutation or gene amplification of proto-oncogenes or of genes whose products facilitate drug resistance provide examples of dominant and co-dominant phenotypes, whereas loss or inactivation of tumor suppressor genes or of genes whose products facilitate drug sensitivity, also leading to drug resistance or transformation, are examples of recessive genetic alterations. Often, knowledge of the genetic mechanism responsible for a particular gain of function or lesion has been obtained by using the techniques of classical genetics, involving either gene mapping by linkage analysis and subsequent positional cloning or the cloning and characterization of genomic regions involved in disease-related chromosomal rearrangements. Although these approaches have been very productive, years of work are often required to identify a single gene and, yet more difficult, to define its functional relationship to a particular phenotype.
Figure 1. Strategy for expression cloning. The first step involves determining the dominant or recessive nature of the phenotype of interest by characterizing cell-cell hybrids. Depending on the result, two differet types of expression libraries might be constructed: a library of full-length cDNA in the case of a dominant phenotype or a GSE or antisense library in the case of a recessive phenotype. The libraries are then delivered to the target (presumably wild-type) cells, which are then selected for the desired phenotypic alteration. Library-derived inserts are then rescued from the selected cells, recloned into an expression vector to generate a less complex library, enriched for biologically active clones. The sub-library is subjected to a second round of selection and the resulting clones are tested individually. GSEs with confirmed activity are used as probes to isolate the corresponding full-length cDNAs, which are also used for biological testing.
The genetic event underlying a particular phenotypic alteration can be identified relatively easily for dominant or co-dominant genes, which can be effectively identified and studied using many different techniques for transfer and isolation of amplified or overexpressed DNA (42-46). For example, expression selection was used to clone several cellular oncogenes (47). An extremely important contribution has been made as a result of studying retroviruses, which have served not only as a natural system for cDNA cloning and transfer, but have also provided a wide variety of mutated, truncated, overexpressed and hybrid genes for subsequent analysis by expression/selection (48,49). The dominant nature of oncogenes also facilitates studies of their functions, both in cell culture and in transgenic animals (50,51).
Transfection of cellular DNA for functional gene discovery
Dominant mutations that cause selectable phenotypes can, in principle, be identified by the direct transfer of genomic DNA from mutant to wild-type cells. This approach was applied successfully in the famous work of Michael Wigler's group, leading to the discovery of the activated Ha-ras oncogene in human bladder carcinoma T24 cells (52,53). DNA derived from this cell line induced the morphological transformation of mouse NIH 3T3 cells. The functional human sequences responsible were cloned from a genomic library, prepared from transformed mouse cells, using human Alu repeats as a probe. In later work, instead of total cellular cDNA, libraries including large pieces of genomic DNA were used for transfection, thus simplifying the identification of exogenous DNA in the cell, especially if the clones were derived from pre-mapped or pre-selected genomic regions (54,55). Probably the most impressive example of this kind is the recent cloning of a mouse clock gene that controls circadian rhythms, found by rescuing a mutation in the germline of genetically aperiodic mice using a series of contiguous genomic clones that span a previously mapped genomic region (56). Although this general approach is very straightforward, it is complicated by several technical problems and, for these reasons, has not been used broadly. (i) It depends on the availability of a screening system in which the desired phenotype can be caused by a single genetic event. However, for example, for many oncogenes there is no system as effective as NIH 3T3 cells are for ras. (ii) The length of many mammalian genes makes it almost impossible to use direct transfection of genomic DNA, due to fragmentation. (iii) The identification of exogenous sequences integrated in the genome of a target cell remains laborious.
Expression selection of retroviral cDNA libraries
The use of cDNA rather than genomic DNA for functional selection has obvious advantages. cDNAs are compact, surrounded by DNA sequences from other organisms and expressed from strong promoters, making identification of the introduced DNA relatively easy. cDNA cloning has been used successfully to identify genes whose expression is responsible for a variety of phenotypes (45,46,57). However, the transfection of plasmid-borne cDNA is neither efficient nor selective enough, since too many different clones are delivered simultaneously to each transfected cell. As the techniques have evolved further, there are now high expectations for retroviruses as a delivery system and, in fact, these viruses have already been used for the successful expression selection of several important genes (58-63). Oncogenes are acquired by viruses in a stochastic process in which rare sporadic recombinants are generated between viral and cellular RNAs, the great majority of which are never seen since they do not induce a selectable phenotype. Viruses that have incorporated transforming oncogenes do not survive well in natural populations because they are toxic to the host, but they can be isolated and maintained relatively easily under laboratory conditions. Thus, oncogenes were found by screening naturally arising retroviral cDNA libraries for transforming viruses (48,49).
The use of retroviral vectors has several major advantages over cDNA transfections (64). First of all, the stable transfer efficiencies are at least 100-fold higher, facilitating the transfer and functional screening of very large libraries. Retrovirus infection is not associated with any stress or cytotoxicity to the recipient cells. In virus-infected populations, the majority of cells contain one integrated provirus, facilitating further identification of the biologically active insert. Packaging systems have been developed that allow genetic material to be delivered to cells of many different types and species of origin (65-74). In one of the very first applications, retroviral cDNA libraries were screened for clones that induce the oncogenic transformation of NIH 3T3 fibroblasts, measured by the loss of contact inhibition (58). Among 19 different cDNA clones isolated from a total of 300 000, three were derived from the known oncogenes raf-1, lck and ect2, while nine others were derived from genes that had been cloned previously but were not known to transform fibroblasts ([beta]-catenin, thrombin receptor, phospholipase c-[gamma]2, etc.). Seven novel cDNAs with transforming activities, including those encoding three new members of the CDC24 family of guanine nucleotide exchange factors, were also cloned.
A very important additional advantage of using retroviral libraries is that the integrated viruses can be rescued for secondary screening, thus allowing one to do selections in the presence of a relatively high background. Recovery can be achieved by several means: (i) isolation of inserts by PCR, with subsequent cloning back into the same vector (75,76); (ii) PCR of full provirus, followed by transfection into packaging cells (77); (iii) polyethylene glycol-mediated fusion with packaging cells, leading to the release of infectious virus (78); and (iv) transfection of expression plasmids encoding the retroviral genes gag-pol and env (or G-protein of vesicular stomatitis virus) into the selected cells (79,80). In certain cases, the selection can be done directly with packaging cells infected with the cDNA library (81; see also below).
Recently, a novel technique of provirus rescue has been developed, based on a new type of retroviral vector containing lox sites in the LTRs and E.coli plasmid replicon sequences within the provirus (82). Integrated proviruses, excised from genomic DNA by the Cre recombinase, delivered by retrovirus infection, are used directly to transform competent bacterial cells. This system was first applied in a genetic screen performed to identify cDNAs that abrogated the sensitivity to TGF-[beta] of mink lung epithelial cells, leading to the important observation that ectopic expression of the murine double minute 2 (mdm2) gene rescued TGF-[beta]-induced growth arrest in a p53-independent manner by interfering with the function of the retinoblastoma susceptibility gene product (Rb). Thus, MDM2 confers TGF-[beta] resistance in a subset of tumors and may promote tumorigenesis by interfering with two different tumor suppressors, p53 and Rb (83).
Although not many reports describe functional screens of retroviral cDNA libraries, there is little doubt that this rapidly emerging field has strong potential. Recent advances in developing more sophisticated and effective vectors, generating different selection systems and construction of many retroviral libraries from different sources make it clear that many genes that participate in a variety of processes will be identified by this technique shortly. However, retroviruses have a serious limitation, which reduces their value as gene discovery tools somewhat. The length of viral RNA is limited by the capacity of the viral capsid, ~10 kb. Therefore, long cDNAs may be missing from retroviral libraries. This problem can be resolved in part by using vectors that contain only the essential viral sequences (no internal promoters and no selectable markers), leaving the rest of the space for the cDNA inserts (80).
Retroviral insertional mutagenesis plus activated proto-oncogene: an approach to identify new transforming genes
Retroviruses can induce tumors not only by transducing oncogenes. Replication-competent, slowly transforming retroviruses can also induce transformation by deregulating the expression of cellular oncogenes that happen to be near the site of integration. Thus, the provirus both activates and marks the proto-oncogene (84,85). One of the best studied examples is that of the mouse mammary tumor virus (MMTV), which promotes breast cancer by activating any one of several proto-oncogenes: int-1, int-2, etc. (86). Since the sites of virus integration are essentially random, the infected cell population includes many independent integrations that lead to modulation of expression of the nearby genomic regions. Natural selection picks from this population those cells that are capable of unconstrained growth. This process was used by Anton Berns and co-workers to create an elegant experimental approach for the systematic identification of novel oncogenes (87,88). To increase the efficiency of transformation by slowly transforming viruses, they used oncogene collaboration, introducing retroviruses into transgenic animals that already carried an activated oncogene. Infection of Eµ-myc transgenic mice (expressing the c-myc transgene under the control of the immunoglobulin heavy chain promoter) with Moloney murine leukemia virus (Mo-MLV) resulted in a dramatic acceleration of pre-B cell lymphomagenesis (89). Proviral tagging identified genes that cooperate with the c-myc transgene in transforming B cells. Four loci (pim-1, bmi-1, pal-1 and bla-1), occupied by proviruses in 35, 35, 28 and 14% of the tumors, respectively, were identified, each representing a putative proto-oncogene, some of which have been characterized in detail (87,88). The protocol was then modified to search for genes that collaborate effectively with a transgene in later stages of tumor development. Propagation of tumors induced by Mo-MLV in Eµ-Pim1 or H2-K-myc transgenic mice, after transplantation to syngeneic hosts, permitted proviral tagging of genes required for tumor progression. Molecular cloning of common proviral insertion sites that were detected preferentially in transplanted tumors led to the identification of the novel gene Frat1 (90). This work demonstrates the power of proviral tagging, which can be combined with germline gene disruption and overexpression in many mouse cancer models to investigate the biochemical pathways through which oncogenes act.
FUNCTIONAL CLONING OF RECESSIVE GENES
Expression selection approaches cannot be applied directly to clone genes which, when overexpressed, cause a phenotype that cannot be selected by means of drug resistance, transformation or expression of a surface marker. Thus, overexpression of tumor suppressor genes or of genes that confer drug sensitivity results in growth arrest, apoptosis or sensitization to drug treatment, making it very difficult to select them in this way. Fortunately, inactivation of such genes does result in a selectable phenotype, but the mutations are recessive and cannot be detected in the presence of the wild-type allele. This situation has led to an interesting sequence of gene discovery in which dominant genetic mechanisms were understood faster than recessive ones, resulting in a disproportionately large amount of information on oncogenes compared with tumor suppressors (91-93). More rapid progress in understanding recessive genetic mechanisms has been made possible by using general procedures to select recessive genes functionally, followed by proof of their roles through gene inactivation. Several such techniques have now been developed and used successfully.
Use of antisense expression libraries to clone pro-apoptotic genes
Antisense RNAs can effectively and specifically inhibit expression of the corresponding gene. Antisense cDNA expression libraries usually contain many clones capable of inactivating genes and, therefore, can be used to generate a selectable phenotype for gene discovery. This approach was first developed and applied to the isolation of intracellular death-promoting genes (94). The TKO method involves transfecting cells with an antisense cDNA library made in an episomal vector, followed by selection of those transfectants that survive exposure to a killing cytokine, IFN-[gamma]. This approach has yielded five novel genes involved in apoptosis (95-98). These death-associated proteins (DAP) represent a diverse spectrum of biochemical activities, including a novel calcium/calmodulin-regulated kinase with a death domain (DAP-kinase), a nucleotide-binding protein (DAP-3), a small proline-rich cytoplasmic protein (DAP-1) and a novel homolog of the eIF4G translation initiation factor (DAP-5). Extensive further work has proved that these genes are critical for mediating cell death following exposure to IFN-[gamma] and that they also function in other cases in which apoptosis is stimulated through death receptors. Moreover, DAP-kinase has strong tumor suppressive activity, coupling the control of apoptosis to metastasis (99). These experiments reveal the power of the TKO approach, which is likely to lead to the discovery of many new pro-apoptotic genes as it is applied to different systems.
Random gene inactivation by antisense promoter insertions
An interesting variant of the antisense RNA approach has been developed recently and applied successfully for the functional identification of a novel candidate tumor suppressor gene (100). The technique is based on a specially designed expression vector that can inactivate transcribed genes functionally. It contains a selectable marker (neo) without a promoter and an IPTG-inducible promoter in the antisense orientation, upstream of the selectable marker. This cassette of elements is flanked by lox sites for excision by the Cre recombinase. The selectable marker can be transcribed effectively only within a natural transcription unit, allowing one to select cells in which the insert has integrated into a transcribed gene. After G418 selection, IPTG is added to activate antisense transcription, generating an antisense RNA corresponding to the gene that has acquired the transfected insert. The antisense RNA may inhibit expression from the unaffected allele of the target gene, leading to a functional knockout. Introduction of Cre recombinase reverses the phenotype. Using mouse 3T3 fibroblasts, a transduced cell population was selected to isolate clones that grew in 0.5% agar and formed metastatic tumors in nude mice and the novel gene tsg101 was identified. Its homozygous disruption leads to cell transformation and removal of the transactivator restores normal growth. The protein encoded by tsg101 has a coiled-coil domain that interacts with stathmin, a cytosolic phosphoprotein previously implicated in tumorigenesis.
Genetic suppressor elements
Despite the relative simplicity of generating libraries encoding antisense RNAs, many antisense constructs fail to cause a significant biological effect (101,102). Some of the most efficient antisense RNAs correspond to only a portion of the target mRNA, often the 5[prime]-end (103,104). There is, however, no clear rule about how to predict the optimal region of the mRNA to target (104,105). Although antisense RNA libraries have been used for expression selection of recessive genes (see above), they work only when expression can be suppressed efficiently by antisense RNAs transcribed from the full-length or 3[prime]-terminal portions of cDNAs that are characteristic of such libraries (94). Another approach is to use dominant mutant proteins that interfere with the function, localization, stability or assembly of a wild-type protein. Dominant negative proteins frequently correspond to truncated forms of the normal protein that include specific domains involved in essential protein-protein interactions (106). Aside from the relatively rare cases of proteins whose domain structure is well understood, one is not likely to be able to predict which portion of a particular protein will act in a dominant-negative fashion. Thus, both antisense and dominant-negative strategies require the expression of specific portions of the target gene in either the sense or antisense orientation. The main difficulty is how to choose an appropriate sequence. This obstacle has been overcome by using a general method to clone genetic suppressor elements (GSEs) that may act either through antisense RNA or through dominant-negative truncated proteins. The strategy involves the use of an expression library containing randomly fragmented cDNAs corresponding to the genes targeted for suppression, followed by the isolation of biologically active GSEs by functional selection. The approach combines within a single expression library the advantages of retroviral gene transfer, antisense RNA-based and dominant negative protein-based gene suppression techniques (78).
GSEs from individual genes: proof of the principle
The random fragment strategy was originally tested by Holzmayer et al. (107) in a prokaryotic system and then extended to mammalian cells, where it was first used to derive a set of GSEs from a simple library, constructed from fragments of the cDNA for topoisomerase II[alpha], an essential nuclear enzyme that determines sensitivity to a variety of compounds (epipopdophyllotoxins, acridines and anthracyclines) used in chemotherapy (75). Topoisomerase II can thus be defined as a drug sensitivity gene (108) and GSEs interfering with its expression were expected to confer drug resistance. A library was generated as a mixture of recombinant retroviruses carrying random 200-500 bp fragments of the human topoisomerase II[alpha] cDNA. The fragments were cloned in both orientations and each insert was provided with an initiation codon at the 5[prime]-end and with stop codons in three reading frames at the 3[prime]-end. By selection in HeLa cells, a set of retroviruses that could confer resistance to etoposide were isolated that encode either antisense RNAs or short proteins (66-124 amino acid residues) representing different portions of topoisomerase II[alpha]. This work provided the first direct indication that suppression of this enzyme can promote drug resistance and allowed the general principles of the GSE methodology to be established.
Since tumor suppressors are involved in negative growth regulation (109), GSEs against them should behave as dominant oncogenes. The principles were developed using p53 as a model (76,110). A GSE library derived from rat p53 cDNA was delivered to mouse embryo fibroblasts (MEFs) and two different selections were applied to isolate GSEs: focus formation and resistance to DNA damage (76). Unexpectedly, the two selections yielded non-overlapping sets of sense-oriented GSEs that also differ in their ability to cooperate with an activated Ha-ras oncogene, to suppress apoptosis and to induce immortalization of rodent fibroblasts, suggesting that p53-mediated control of cellular senescence, response to DNA damage and regulation of growth involve different functional domains of the p53 protein. These experiments confirmed the efficiency of the GSE approach in mapping functional domains and also allowed selection protocols to isolate GSEs against tumor suppressor genes to be established.
Using the GSE approach to identify new drug sensitivity genes
The isolation of GSEs from complex cDNA populations extends the strategy to the cloning of new genes whose suppression is associated with a selectable phenotype. A large GSE library prepared from total cellular cDNA should, in principle, contain suppressors of most genes expressed in the cell. The GSE function is likely to be opposite to the function of the corresponding genes: thus, GSEs against drug sensitivity or tumor suppressor genes act as drug resistance genes or oncogenes, respectively. GSEs isolated from complex libraries can be used as probes to clone the full-length cDNAs. A large GSE library was constructed from fragmented, normalized total mouse cDNAs and screened to isolate elements conferring resistance to etoposide (81). The selection was carried out directly on packaging cells transduced with the library. Virus produced by cells surviving the first round of selection were reselected from a freshly infected cell population. After the second round, three GSEs were isolated that were able to confer resistance to etoposide. Two carried cDNA fragments corresponding to unknown genes and one represented an antisense fragment of the kinesin heavy chain (KHC), a protein that has never before been associated with drug resistance. The anti-KHC GSE induced an unusual phenotype characterized by resistance to DNA damage and by hypersensitivity to anti-microtubule drugs (111). Moreover, it also induced the immortalization of rodent fibroblasts, thus resembling some of the anti-p53 GSEs (81). Reduced expression of KHC was found in several etoposide-resistant cell lines, indicating that the anti-KHC GSE had uncovered a novel, naturally occurring mechanism of drug resistance. This finding initiated a research program that has established a novel mechanism of drug response involving kinesin-mediated secretion of growth inhibitory factors by cells subjected to a genotoxic stress (81,111; K.V. Gurova, S.A. Axenovich and A.V. Gudkov, in preparation).
GSEs and identification of novel candidate tumor suppressors
GSEs against tumor suppressor genes should act as dominant oncogenes in gene transfer assays, providing a strategy to identify new tumor suppressors, defined as `elements whose loss or inactivation induces one or another phenotype of neoplastic growth deregulation' (91). Although normalized libraries are expected to contain GSEs against most cellular genes, their high complexity may cause difficulties in screening. Therefore, a protocol has been established to construct less complex GSE libraries from cDNAs enriched for the desired sequences. Assuming that tumor suppressor genes may be lost or inactivated in breast tumors but not in normal breast cells, we prepared cDNAs specific for normal breast cells by using a subtraction procedure borrowed from a method of representational difference analysis (112). Sequences in breast tumor cell cDNAs were subtracted from a library of normal breast cDNAs, which was then used as a probe to screen a conventional cDNA library (113). cDNA inserts from ~300 phage clones so identified were combined and used to prepare a random fragment library. The library was delivered to mouse mammary cells, which were then selected for anchorage-independent growth. One of the biologically active GSEs was used to isolate a homologous full-length cDNA, representing a new gene encoding a new nuclear protein, p33ING1. The overexpression of ING1 leads to strong growth inhibition of the two cell lines tested, causing G1 arrest. Further characterization of the biological properties of p33ING1 and the corresponding GSE demonstrated that this protein cooperates with p53 in effecting control of cell growth (114). Further analysis of the structure and expression of ING1 showed that this gene encodes, in addition to p33ING1, a longer protein, p37ING1, translated from an alternative transcript. Interestingly, p37ING1 acts as a suppressor of p53, indicating that ING1 represents an unusual case in which a single gene encodes both a candidate tumor suppressor and a candidate oncogene, expressed from alternative transcripts (M. Zeremski, J.E. Hill, L. Diatchenko, I.V. Garkavtsev, E.V. Koonin and A.V. Gudkov, submitted for publication; I.A. Grigorian, K.V. Gurova, M. Zeremski and A.V. Gudkov, in preparation).
In summary, both of the GSEs chosen for detailed analysis (the anti-KHC GSE and the GSE against ING1) have led to new insights into mechanisms of drug response and transformation, demonstrating the power of the approach. Many other GSEs with similar properties have been isolated, each of which may lead to equally interesting and important genes (115). Moreover, new versions of the GSE methodology are being developed, broadening the potential applications of this technique. For example, a large GSE library, prepared from normalized fragmented chicken cDNA in a replication-competent avian retroviral vector, has been screened successfully for GSEs that inhibit replicative senescence and prolong the lifespan of chicken fibroblasts in culture (E. Feinstein, A.P. Komarov and A.V. Gudkov, in preparation). The use of replication-competent vectors opens attractive opportunities for isolating GSEs in vivo, using selection protocols similar to those applied by Anton Berns and co-workers for the in vivo isolation of new oncogenes by retroviral insertional mutagenesis (see above).
The SETGAP technique: isolation of growth suppressing GSEs
Pestov and Lau have developed another functional approach to identify genetic elements with growth suppressive activity (116,117). The basic idea is that the temporary expression of growth-arresting elements can protect cells against selective killing of proliferating cells. The essential components of the technology, named SETGAP (selectable expression of transient growth arrest phenotype), include: (i) construction of expression libraries in an inducible vector (IPTG-inducible vectors were used); (ii) delivery of the library to target cells that support inducible expression (mouse Balb 3T3 cells expressing a mutated lac repressor allowing transcription in the presence of IPTG); (iii) induction of expression, followed by selective killing of proliferating cells (by labeling the DNA of replicating cells with bromodeoxyuridine, followed by staining with Hoechst dye and killing by illumination with visible light); (iv) reversal of induction and expansion of surviving cells, to be used as a source of the responsible genetic elements. This method was first used to isolate growth-suppressing GSEs from a pool of random cDNA fragments of 19 growth-related genes associated with the G0/G1 transition, based on the premise that blocking the function of an essential gene should lead to growth inhibition. As a result, GSEs encoding potential dominant negative variants of c-fos, junB and p44 MAPK were isolated that interfere with the growth-related functions of these genes (116). Later, SETGAP was extended to isolate growth-inhibitory clones from a mouse cDNA expression library, constructed in a similar vector. Two such cDNA clones encoded the ubiquitin-conjugating enzyme UbcM2 and a truncated form of a novel WD40 repeat protein, Bopl, which is conserved from yeast to humans (117). This work indicates that SETGAP is a useful technology to identify growth-suppressing clones, although its applications are limited to those genetic elements that can cause reversible growth arrest of the target cell without causing cell death.
PERSPECTIVES
Recent technological advances in molecular genetics have already resulted in the accumulation of an enormous amount of information on new genes, generated within a variety of genomic and cDNA sequencing projects and by the application of powerful new techniques that allow us to analyze differential gene expression in large sets of known and unknown sequences. These techniques include different versions of the array hybridization technology (118) and cDNA subtraction methods, such as RDA (112), SSH (119) or differential display (120). Another source of information on new genes is the two-hybrid system (121), which efficiently generates lists of suspects. However, none of these powerful techniques provides proof of gene function. The efficiency of the functional gene cloning approaches described in this review would be increased substantially if enriched pools of sequences were used. Thus, screening full-length cDNA and GSE libraries would be much easier if the libraries were prepared from pre-selected, differentially expressed sequences determined by array hybridization, two-hybrid screening or subtraction. The GSE work that resulted in the identification of the ING1 gene is an example of a study in which a cDNA subtraction method (RDA) was successfully combined with functional GSE selection. At present, there is a wide gap between the power of enriching techniques and functional analysis of new cDNA sequences, making the final step, from the enriched material to a defined functional gene, difficult. Successful resolution of this problem depends on how quickly sequence enrichment can be combined with the techniques described in this review, or with newer approaches, to achieve more efficient gene discovery by functional selection.
The mammalian cell mutants obtained so far have been in pathways that are not essential for cell growth or viability. Extension of the methodology to obtaining conditional mutations in essential genes will necessitate the use of replica plating techniques specifically adapted for mammalian cells (see for example ref. 122). Although several different types of mutant have been obtained (Table 1), it should still be possible to extend the range of pathways that can be investigated by random mutagenesis. New horizons might include projects to investigate regulation of the processing, transport or stability of mRNAs or the translocation, modification or stability of proteins. Forward genetics in mammalian cells is now at the end of the beginning.
NOTE ADDED IN PROOF
Somatic cell mutants resistant to retrovirus replication
To identify cellular functions involved in the early phase of the retroviral life cycle, rat2 fibroblasts were treated with chemical mutagens and individual virus-resistant clones were recovered after selection for resistance to infection. Two clones characterized in detail were resistant to infection by both ecotropic and amphotropic murine viruses, as well as by human immunodeficiency virus type 1 pseudotypes. One clone showed a strong block to reverse transcription of the retroviral RNA, and the second showed normal levels of viral DNA synthesis but did not allow formation of the circular DNAs normally found in the nucleus. The properties of these two mutant lines suggest that host gene products play important roles both before and after reverse transcription. This interesting approach (123) may be applicable to other viruses as well.
ACKNOWLEDGEMENTS
This work was supported by grants from the National Institutes of Health, CA60730 and CA75179 to A.V.G. and CA62220 to G.R.S.
REFERENCES