Expression genetics in cancer: Shifting the focus from DNA to RNA

Abstract

Expression genetics is a conceptually different approach to the identification of cancer-related genes than the search for mutations at the genome level. While mutations lie at the heart of cancer, at least in its early stages, what is recognized here are phenotypic changes usually many steps removed from the initiating mutation. Classically cancer geneticists have concentrated on genomic changes and have ignored the productive potential of examining downstream events based on screening for differential gene expression between tumor cells and well matched normal counterparts. Genes involved in cancer affect the normal functions of many cellular processes: not only proliferation but cell–cell and cell–matrix interactions, DNA repair, invasion and motility, angiogenesis, senescence, apoptosis, and others. Yet very few cancer-related genes affecting these processes have been identified in human cancers by classical methods to find mutated genes despite enormous efforts. I report here our success in readily isolating more than 100 candidate tumor suppressor genes from human tissue, estimated to represent roughly 20% of the total genes recoverable by this approach. Half of the genes are unknown and the other half include representatives of most known cancer processes. Because their expression is lost during cancer progression, they may be useful tumor markers for diagnosis and prognosis. Because these genes are not mutated, they provide opportunities for pharmacological intervention by inducing their reexpression.

Cancer genetics is an enormous puzzle. Cancer genes are involved in the dys-regulation of a broad array of normal cellular functions including cell proliferation (1), DNA repair (2), chromosome stability (3), cell–cell communication (4, 5), cell–matrix interactions (6–8), angiogenesis (9), tumor invasion, motility, and metastasis (10, 11), senescence (12, 13), and apoptosis (14). But how fragmentary is the list of mutated genes involved in human cancers (15) when compared with the multitude of expression changes underlying the cancer phenotypes!

More than 100 oncogenes have been identified in animal systems, but only a small subset have been found consistently as mutated genes in human cancer (16–18). Those involved in multiple tumor types include those encoding the growth factor receptors erbB and erbB2, ras, myc, and Bcl-1. On the tumor suppressor side, ≈12 genes are currently accepted as bona fide examples of genes whose mutation or deletion is associated with tumorigenesis in one or more types of human cancer (15–17). They include Rb, p53, WT1, BRCA1, BRCA2, VHL, APC, NF-1, NF-2, and MTS-1 (encodes p16, a cell cycle inhibitor). The known factors affect principally cell cycle regulation. None are known to affect invasion or metastasis, to name but a few of the cellular and systemic processes involved in cancer.

Thus the genes that have been identified do not begin to account for the diversity of cancer phenotypes. Clearly, vastly improved methods are needed to recognize and recover human cancer-related genes. And, more profoundly, we need to broaden our criteria for the identification of these genes.

Cancer genes are defined operationally by their altered expression, leading to an abnormal phenotype in a significant subset of cancers. The altered expression may facilitate initiation or progression of a neoplasm, as oncogenes do, or may inhibit it, as do tumor suppressor genes. Conventionally, only mutated genes have been considered as candidate cancer genes, but cancer phenotypes result from altered gene expression, and there is no simple 1:1 relationship between mutated genes and cancer phenotypes. For example, cancer genes such as myc, p53, and WT-1 encode transcription factors (15, 16), which in turn regulate the expression of multiple downstream genes. Other recognized cancer genes such as erbB2 encode transmembrane receptors that regulate downstream genes through signal transduction pathways.

Paradoxically, some genes are mutated in cancer whereas others with similar functions are not. Examples include Rb, which is often mutated in particular forms of cancer, and p107, which is not, and the INK4 gene p16, which is mutated, and p15, which is not (19, 20). Indeed, underexpression of p107 (and p300) may be oncogenic. Similarly, overexpression of the wild-type epidermal growth factor receptor and estrogen receptor may be oncogenic by virtue of altering expression of signal transduction pathways and downstream genes that they modulate. Amplification leading to overexpression is another mode of dys-regulation stabilized at the genomic level, as for example cyclin D1 (20). But overexpression of D1 may also occur without amplification, suggesting that cells carrying this amplification are selected in response to a preexisting local cancer-promoting environment.

Should only the mutated genes be designated as cancer genes? Clearly, many more genes are altered in expression in cancer cells than are mutated. Nonmutated genes with stably altered expression patterns in significant subsets of cancers are a key component of the cancer genetics puzzle, both for their contribution to understanding the molecular bases of cancer and for their potential role in the design of chemotherapeutic agents.

Individual genes may be altered in expression either by mutation or by changes in their regulation. This dichotomy, first recognized in classical studies of bacterial genetics (i.e., β-galactosidase mutation and adaptation), provides the basis for grouping cancer genes into two classes: class I genes are mutated or deleted, whereas class II genes are not altered at the DNA level. Rather they affect the phenotype by expression changes (21).

Thus, retroviral oncogenes exemplify class I mutated oncogenes, and epidermal growth factor receptor is an example of a class II oncogene. Rb is a class I tumor suppressor gene, and maspin (discussed below) is a class II tumor suppressor gene.

Some genes may be class I in certain situations and class II in others. An example is MTS-1 (p16), a gene involved in cell cycle inhibition, in which DNA methylation is one mode of gene silencing without mutation (22). Another is BRCAl, seen as a mutated gene in familial breast cancer (23, 24). As yet, however, few if any sporadic mutations have been found in somatic tissues of patients without a family history of breast cancer. Rather, reduced amounts of BRCAl mRNA, representing down-regulation of the wild-type gene, were found in primary tumors of the nonfamilial disease (25). Thus while BRCA1 is class I in familial cancers, it may function as a class II tumor suppressor gene in sporadic tumors.

Identifying Expression Changes in Class II Cancer Genes

In principle, genetic changes can be identified by screening at various levels: cellular phenotypes, DNA, RNA, protein. In practice, human cell genetics is unwieldy; and two-dimensional protein gels display only a fraction of cellular proteins, which are difficult to recover and utilize. Most screening has focused on changes at the DNA level by two major methods. Pedigree analysis provides linkage data for localizing genes involved in inherited defects. Chromosome analysis, either by loss of heterozygosity (26) or by comparative genomic hybridization (27), can localize aberrant chromosomal events. Both approaches detect only a small subset of cancer genes, and provide at best localization to the megabase level. Positional cloning, a complex, slow, laborious, and very expensive process, is then required to recover the actual gene. Representational difference analysis is a new method in which subtraction of fragmented genomic DNA is coupled to PCR to select for alterations in DNA (28).

The RNA alternative to screening for mutational differences between normal and tumor cells is to screen for differences in gene expression. Differential gene cloning (29) and subtractive hybridization (30), both successful methods to screen for differentially expressed genes at the RNA level, are now being superseded by differential display (DD), a powerful and novel procedure (31–33). In DD screening, partial cDNAs, produced by reverse transcription-PCR of randomly primed RNAs, identify the majority of expressed mRNAs; and they can be directly recovered from sequencing gels. Nanogram amounts of total RNA suffice, and the time from RNA extraction of cells to gene identification can be as short as 1–2 weeks. Several different cell populations can be compared in adjacent lanes of the same gel. Both overexpressed and underexpressed genes can be identified in the same experiment, making it possible to recognize and recover novel genes whose expression changes during development, carcinogenesis, or any other process under investigation.

DD is the current method of choice in hundreds of laboratories, as evidenced by the September 1996 Cold Spring Harbor Symposium that attracted more than 400 participants worldwide. DD has sparked more elaborate methods such as the RNA modification of representational difference analysis (34) and serial analysis of gene expression (SAGE) (35), as well as modifications and improvements from many laboratories including our own (36) that retain the simplicity of the original procedure.

The widespread and expanding application of differential gene expression analysis to a broad range of questions in fundamental biology bodes well for its contributions to cancer research. A technical obstacle has been the acquisition of RNA from tumor cells in heterogeneous specimens containing many cell types. Current methods to solve this problem include dissection of fixed or frozen tissues (37) and cell sorting (38). Currently in development are the use of fine needle aspirates containing more than 90% tumor cells, and preparation of cDNA libraries from small numbers of cells.

Recovery of genes as cDNAs is another advantage of expression cloning, because cDNAs can be used as probes in Northern blot analyses and in isolating genomic sequences, in vectors for recombinant protein production, for sequence information in designing peptides for antibody production, and for antisense constructs. Thus, recovering the gene of interest as a cDNA rather than as a genomic clone, greatly simplifies the subsequent research.

The thesis of this essay is that cancer genes of class II have been neglected by cancer geneticists. Until now the focus has been on class I genes and the screening methods have been largely limited to searches for alterations in DNA. Yet, with the application of expression cloning methods, such as subtractive hybridization and DD, a rich yield of candidate tumor suppressor genes can be recovered.

Class II genes offer novel therapeutic possibilities because they are present as unmutated wild-type alleles in cancer cells. Class II genes could provide molecular assay systems with which to screen candidate therapeutic agents for reversal of their altered regulation. Tumor suppressor genes in particular are important targets in which reversing the block to gene expression could lead to stable tumor suppression. Ideally, the pharmacological agent of choice would re-induce expression of a coordinate set of down-regulated tumor suppressor genes, and in so doing, would inhibit proliferation and switch the cells into a previously interrupted differentiation program. Terminal differentiation would then remove them from the cancer pool.

Application of DD in Breast Cancer

Some years ago, my coworkers and I began to apply expression genetics to identify tumor suppressor genes in an effort to unlock the stores of these genes that were predicted but undetected (39). To date we have identified more than 100 genes whose expression is abolished or strongly down-regulated in breast carcinoma cell lines (36, 39). More than half of those identified by DD are novel, unknown genes. Several of the known genes have been examined for expression in normal and tumor tissue by immunostaining and shown to exhibit patterns of protein expression consistent with our cell culture data. Protein expression levels may diverge from mRNA levels by translational and posttranslational modifications, but screening at the transcription level remains by far the most powerful method yet available to acquire a large subset of differentially expressed genes.

Hundreds of genes that play regulatory or inhibitory roles in particular cellular processes may qualify as tumor suppressors. This estimate is supported by evidence from DD gels comparing normal and tumor-derived cDNAs from human mammary epithelial cells and carcinomas, in which about 1% of the displayed cDNAs are overexpressed in the normal cells (31–33, 36). Because cells of a given tissue contain 10,000–15,000 different mRNAs, the method detects roughly 150 candidate tumor suppressor genes in a given cell type. This value is probably an underestimate, but important, because it predicts numbers in the hundreds, not thousands. A similar conclusion was reached by Bauer et al. (40). On this basis, there is a reasonable chance to recover most of the tumor suppressor genes that are transcriptionally regulated in breast and other cancers.

As with all screening methods, the choice of which genes to investigate requires further knowledge or additional assays. Clearly, some of these genes will have stronger tumor suppressor activity than others, depending in part on redundancy and regulatory interactions, as well as function. Some of the genes we have identified are either known or related to known genes, providing clues to function. Examples are genes encoding proteases and protease inhibitors, cell surface and extracellular matrix proteins, keratins, regulatory proteins, and cytokines.

For instance, maspin, a serine protease inhibitor or serpin, is a class II gene discovered by its differential expression: on in normal mammary epithelial cells and off in metastatic mammary carcinomas (41). The recombinant protein is inhibitory in the invasion assay, and the inhibition is reversed by an antimaspin antibody preparation that recognizes the reactive center of the protein (42). As shown by time-lapse video microscopy and Boyden chamber studies (43), maspin has strong antimotility activity that may provide the mechanism for its ability to inhibit invasion. Direct evidence of its tumor suppressor activity comes from nude mouse studies in which transfected tumor cells expressing maspin were inhibited in growth and metastasis (41). Therapeutic applications of this class II tumor suppressor gene are under investigation.

We have recovered largely, perhaps entirely, class II genes by expression cloning. The evidence is first their consistent wild-type restriction fragment patterns in Southern blot analysis. Thus, no evidence of gross rearrangement was found. Second, we have induced reexpression of mRNA in 7 of the 20 genes so far tested by treatment of tumor cells with azadeoxycytidine (unpublished data). This result suggests that the genes were down-regulated in expression in the tumor cells by DNA methylation, a process that blocks transcription. Reexpression of other down-regulated genes has been induced by treatment with phorbol ester (39). These findings suggest that far more genes are down-regulated at the transcription level in cancer cells than are mutated. This result is not surprising, in the light of rapidly growing evidence of the complexity of transcriptional mechanisms of regulation.

Tjian and Maniatis (44) have emphasized the fact that relatively few transcription factors control the expression of all the 10,000–15,000 genes expressed in a given eukaryotic cell. Although detailed mechanisms to account for this 100-fold or more amplification remain largely unexplored, two general considerations apply: (i) multiple transcription factors act together as complexes regulating individual promoters, and (ii) the same factors play regulatory roles in the expression of many different genes. One transcription factor may contribute to modulated expression of several downstream genes. Recall also that there are cascades of regulation, and autoregulatory feedback pathways, which amplify the opportunities for network stabilization. Thus, the clonally stable changes in mRNA expression seen in cancer may ultimately be traced back to mutations in upstream genes, such as those encoding transcription factors. Another possibility, not mutually exclusive, is that some aspects of carcinogenesis may be akin to development, in which no mutations are involved in the entire program of organismic differentiation.

Differentiation Programs as Modulators of Cancer

Differentiation programs are highly relevant to cancer therapy. More than 30 years ago, Barry Pierce (45) proposed the concept of differentiation therapy: the rerouting of malignant cells back onto a normal differentiation pathway. He developed this concept over the subsequent years with elegant experiments using embryonal carcinoma cells. Leo Sachs, in his extensive investigations of normal and leukemic blood cells, reached similar conclusions: namely, that “genetic abnormalities that give rise to malignancy can be by-passed and their effects nullified by inducing differentiation” (46).

An important example of induced differentiation of tumor cells is the HL-60 leukemic cell line that can be induced to differentiate either as granulocytes by, for example, retinoic acid, or as monocytes/macrophages by phorbol 12-myristate 13-acetate and other agents (47). Recently, all-trans retinoic acid has been highly successful in the clinical treatment of acute pro-myelocytic leukemia by inducing the cancer cells to differentiate (48). More generally, retinoic acid and less toxic derivatives are being developed for the treatment of many kinds of cancer by induced differentiation (49).

An important natural mechanism of gene silencing without mutation is DNA methylation (50). Although methylation occurs at the DNA level, its effect is upon transcription, and the process is readily reversible. Genes can be demethylated in culture or in vivo by treatment with deoxyazacytidine. A new sequencing method for identifying methylcytosine is available (51). The role of DNA methylation as a suppressor of gene expression in cancer is becoming widely recognized (52–54).

A new insight into the mechanism of differentiation comes from the long studied murine erythroleukemia (MEL) cells treated with hexamethylenebisacetamide (HMBA), an inducer of differentiation (55). When MEL cells are treated with HMBA under specified conditions, cell division is blocked and the cells differentiate. The cell cycle block has now been shown to result from the rapid degradation of cdk4, a kinase that complexes with cyclin D3 (56). The complex phosphorylates Rb to promote the normal cell cycle, but when Rb is not appropriately phosphorylated, the cycle is blocked. What remains unknown is the link connecting the inhibition of cell cycle progression with the induction of differentiation.

These new findings call to mind a similar gap in linking cell cycle inhibition through the p16 (INK-4a)-D1-Rb pathway with senescence (20). Both senescence and terminal differentiation are irreversible processes, driven by molecular equivalents of clock-like programs. Cancer is not irreversible, as shown by the many examples of suppression induced by gene transfer or differentiation induced by drugs. Thus filling in the molecular gap between cell cycle blockage and the irreversible processes of differentiation and senescence may offer novel opportunities for therapy. The crucial molecules may not be mutated.

In summary, expression genetics offers a powerful research approach. Genes whose expression is altered at the mRNA level in cancer cells include previously known genes as well as the important class of unknown genes. Characterization of these genes and their functions provides a penetrating insight into the regulatory interactions that have been upset in cancer. Most important, dys-regulated genes are at least as important potential targets for therapy as mutated genes, and there are many more of them. Inducing reexpression of these genes by pharmacological intervention offers a promising approach to normalization or differentiation of tumor cells and thereby to cancer therapy.