18.3.1 Animal tumor viruses provided the first evidence of oncogenes

For many years it has been known that some animal leukemias, lymphomas and cancers are caused by viruses. A few human examples are also known (Table 18.1). Tumor viruses fall into three broad classes:

Table 18.1

Human and animal tumor viruses.

• DNA viruses normally infect cells lytically. They cause tumors by rare anomalous integrations into the DNA of non-permissive host cells (cells that do not support lytic infection). One way or another, integration of the viral genome implants the transcriptional activation or replication signals of the virus into the host genome and triggers cell proliferation. Some of the viral genes involved have been identified, such as those for the T-antigen of SV40 or E1A and E1B of adenoviruses. Unlike the classic retroviral oncogenes (see below) these genes are virus-specific and do not have exact cellular counterparts.

• Retroviruses have a genome of RNA. They replicate via a DNA intermediate, which is made using a viral reverse transcriptase (Figure 18.2). These viruses do not normally kill the host cell (HIV is an exception), and only rarely transform it. The genome of a typical retrovirus consists of three genes, gag, pol and env (Figure 18.3A).

• Acute transforming retroviruses are retrovirus particles which, unlike normal retroviruses, transform the host cell rapidly and with high efficiency. Their genomes include an additional gene, the oncogene (Figure 18.3B). Usually the oncogene replaces one or more essential viral genes, so that these viruses are replication-defective. To propagate them, they are grown in cells which are simultaneously infected with a replication-competent helper virus that supplies the missing functions. Studies of acute transforming retroviruses have revealed more than 50 different oncogenes.

Figure 18.2

Retroviral life cycle. The virus particle (top) contains the RNA genome and viral reverse transcriptase within an outer lipoprotein envelope and inner protein capsid. A double-stranded DNA copy of the viral genome integrates into the host DNA. Here it (more...)

Figure 18.3

A normal and an acute transforming retrovirus. The RNA genome has terminal repeats (R), subterminal unique sequences (U5, U3) and three genes, gag, pol and env. A complicated scheme of splicing and post-translational processing results in a variety of (more...)

18.3.2 An in vitro transfection assay confirmed that cancer cells contain activated oncogenes

An entirely independent way of discovering oncogenes came from a cell transformation assay. The NIH-3T3 mouse cell line readily undergoes transformation in vitro - probably it has already acquired several of the successive genetic changes on the pathway to cancer, and one further change suffices to transform it. In the 3T3 test, the cells are transfected with random DNA fragments from human cancer cells. Potential oncogenes can then be identified by selecting transformants and recovering the human DNA present in them (Figure 18.4). Transformants are obtained when DNA from tumor cells is used, but not with DNA from nontumor cells. Thus tumor cells, even from nonviral tumors, contain activated oncogenes. This route led to the identification of essentially the same set of oncogenes as were found in acute transforming retroviruses.

Figure 18.4

The NIH-3T3 assay. Mouse 3T3 cells are transfected with random fragments of DNA from a human tumor. Any transformed cells (identified by their altered growth) are isolated, and a phage library constructed from their DNA. Phage are then screened for the (more...)

18.3.3 Oncogenes are mutated versions of genes involved in a variety of normal cellular functions

It quickly became apparent that normal cells had counterparts of all the retroviral oncogenes (Table 18.2), and in fact that v-onc genes were transduced cellular genes. With a few exceptions, the v-onc gene products differ from their c-onc (proto-oncogene) counterparts by amino acid substitutions or truncations, which serve to activate the proto-oncogene.

Table 18.2

Viral and cellular oncogenes.

Functional understanding of oncogenes began with the discovery in 1983 that the viral oncogene v-sis was derived from the normal cellular platelet-derived growth factor B (PDGFB) gene. Uncontrolled over-expression of a growth factor would be an obvious cause of cellular hyperproliferation. The roles of many cellular oncogenes (strictly speaking, proto-oncogenes) have now been elucidated (Table 18.2). Gratifyingly, they turn out to control exactly the sort of cellular functions that would be predicted to be disturbed in cancer. Five broad classes can be distinguished:

• secreted growth factors (e.g. SIS);

• cell surface receptors (e.g. ERBB, FMS);

• components of intracellular signal transduction systems (e.g. the RAS family, ABL);

• DNA-binding nuclear proteins, including transcription factors (e.g. MYC, JUN);

• components of the network of cyclins, cyclin-dependent kinases and kinase inhibitors that govern progress through the cell cycle (e.g. MDM2).

18.4.1 Activation of some oncogenes can occur by amplification

Many cancer cells contain multiple copies of structurally normal oncogenes. Breast cancers often amplify ERBB2 and sometimes MYC; a related gene NMYC is usually amplified in late-stage neuroblastomas. Hundreds of extra copies may be present. They can exist as small separate chromosomes (double minutes) or as insertions within the normal chromosomes (homogeneously staining regions, HSRs). The genetic events producing HSRs may be quite complex because they usually contain sequences derived from several different chromosomes (reviewed by Pinkel, 1994). Similar gene amplifications are often seen in noncancer cells exposed to strong selective regimes - for example amplified dihydrofolate reductase genes in cells selected for resistance to methotrexate. In all cases the result is greatly to increase the level of gene expression.

Comparative genome hybridization (CGH) (Forozan et al., 1997) can in principle reveal all regions of amplification in a single experiment, together with any regions of allele loss or aneuploidy, which may point to tumor suppressor genes (see below). The CGH test (Figure 18.5) uses a mixture of DNA from matched normal and tumor cells in competitive fluorescence in situ hybridization. With the aid of image-processing software, chromosomal regions can be picked out where the ratio of FISH signal from normal and tumor DNA deviates from expectation. Depending on the direction of deviation, these mark regions of amplification or of allele loss in the tumor. The smallest alteration visible on standard CGH analysis is 5–10 Mb. By using high-density oligonucleotide arrays instead of metaphase chromosomes, CGH analysis can be carried to much higher resolution.

Figure 18.5

Comparative genome hybridization. Tumor DNA and normal control DNA were labeled with green and red fluorescent labels respectively, then hybridized in situ together in equal quantities to chromosomes of a normal cell. The curves show computer-generated (more...)

18.4.2 Some oncogenes are activated by point mutations

The H-RAS1 gene (Table 18.2) is one of a family of ras genes that encode proteins involved in signal transduction from G-protein-coupled receptors. A signal from the receptor triggers binding of GTP to the RAS protein, and GTP-RAS transmits the signal onwards in the cell. RAS proteins have GTPase activity, and GTP-RAS is rapidly converted to the inactive GDP-RAS. Specific point mutations in RAS genes are frequently found in cells from a variety of tumors including colon, lung, breast and bladder cancers. These lead to amino acid substitutions that decrease the GTPase activity of the RAS protein. As a result, the GTP-RAS signal is inactivated more slowly, leading to excessive cellular response to the signal from the receptor.

18.4.3 Chromosomal translocations can create novel chimeric genes

Tumor cells typically have grossly abnormal karyotypes (Figure 18.6), with multiple extra and missing chromosomes, many translocations and so on. Most of these changes are random, and reflect a general genomic instability which is a normal part of carcinogenesis (see below). A huge research effort has been devoted to picking out tumor-specific changes superimposed on the background of random changes. Over 150 different tumor-specific breakpoints have now been recognized (Mitelman et al., 1997), and they reveal an important common mechanism in tumorigenesis.

Figure 18.6

Multicolor FISH (M-FISH) karyotype of a human myeloid leukemia-derived cell line. Note the numerous numerical and structural (arrowheads) abnormalities revealed by 24-color whole chromosome painting. These include t(5;15), der(7)t(7;15), der(8)ins(8;11), (more...)

The best-known tumor-specific rearrangement produces the Philadelphia (Ph¹) chromosome, a very small acrocentric chromosome seen in 90% of patients with chronic myeloid leukemia. This chromosome turns out to be produced by a balanced reciprocal 9;22 translocation. The breakpoint on chromosome 9 is within an intron of the ABL oncogene. The translocation joins most of the ABL genomic sequence onto a gene called BCR (breakpoint cluster region) on chromosome 22, creating a novel fusion gene (Chissoe et al., 1995). This chimeric gene is expressed to produce a tyrosine kinase related to the ABL product but with abnormal transforming properties (Figure 18.7A).

Figure 18.7

Chromosomal translocations which activate oncogenes. (A) activation by qualitative change in the t(9;22) in chronic myeloid leukemia. The chimeric BCR-ABL fusion gene on the Philadelphia chromosome encodes a tyrosine kinase which does not respond to normal (more...)

Many other rearrangements are known that produce chimeric genes (Table 18.3). The products are normally transcription factors (or sometimes tyrosine kinases) which take their target specificity from one component gene, but couple it to an activation or ligand-binding domain from the other. This has been one of the most satisfying stories to emerge from cancer research, with several examples of clinical phenotypes being elegantly explained by a combination of cytogenetic and molecular genetic findings. The whole topic of chromosomal translocations in cancer and the underlying genetic events has been reviewed by Rabbitts (1994) and Sanchez-García (1997).

Table 18.3

Chimeric genes produced by cancer-specific chromosomal rearrangements.

18.4.4 Oncogenes can be activated by transposition to an active chromatin domain

Burkitt's lymphoma is a childhood tumor common in malarial regions of Central Africa and Papua New Guinea. Mosquitoes and Epstein-Barr virus are believed to play some part in the etiology, but activation of the MYC oncogene is a central event. A characteristic chromosomal translocation, t(8;14)(q24;q32) is seen in 75–85% of patients (Figure 18.7B). The remainder have t(2;8)-(p12;q24) or t(8;22)(q24;q11). Each of these translocations puts the MYC oncogene close to an immunoglobulin locus, IGH at 14q32, IGK at 2p12 or IGL at 22q11. Unlike the tumor-specific translocations shown in Table 18.3, the Burkitt's lymphoma translocations do not create novel chimeric genes. Instead, they put the oncogene in an environment of chromatin that is actively transcribed in antibody-producing B-cells. Usually exon 1 (which is noncoding) of the MYC gene is not included in the translocated material. Deprived of its normal upstream controls, and placed in an active chromatin domain, MYC is expressed at an inappropriately high level.

Many other chromosomal rearrangements put one or another oncogene into the neighborhood of either an immunoglobulin (IGG) or a T-cell receptor (TCR) gene (Rabbitts, 1994; Sanchez-García, 1997). Presumably the rearrangements arise by random malfunctioning of the recombinases that rearrange IGG or TCR genes during maturation of B and T cells (Section 8.6), and are then selected for their growth advantage. Predictably, these rearrangements are characteristic of leukemias and lymphomas, but not solid tumors.

18.5.1 Retinoblastoma exemplifies Knudson's two-hit hypothesis

Retinoblastoma (MIM 180200) is a rare, aggressive childhood tumor of the retina. 60% of cases are sporadic and unilateral; the other 40% are inherited as an imperfectly penetrant autosomal dominant trait, which was mapped to 13q14. In familial retinoblastoma bilateral tumors are common. In 1971 AG Knudson proposed that two successive mutations (‘hits’) were required to turn a normal cell into a tumor cell (Figure 18.8), and that in familial forms one of the hits was inherited. A seminal study by Cavenee et al. (1983) both proved Knudson's hypothesis, and established the paradigm for laboratory investigations of TS genes.

Figure 18.8

Knudson's two-hit hypothesis. Suppose there are 1 million target cells and the probability of mutation is 10^-5 per cell. Sporadic retinoblastoma requires two hits and will affect 1 person in 10 000 (10⁶ × 10^-5 × 10^-5 = 10^-4), while the (more...)

Cavenee and colleagues sought evidence of somatic mutations at the RB1 locus in sporadic retinoblastoma by typing surgically removed tumor material with a series of markers from chromosome 13. When they compared the results on blood and tumor samples from the same patients, they noted several cases where the constitutional (blood) DNA was heterozygous for one or more chromosome 13 markers, but the tumor cells were apparently homozygous. They reasoned that what they were seeing was one of Knudson's ‘hits’: loss of one functional copy of a tumor suppressor gene. Combining cytogenetic analysis with studies of markers from different regions of 13q, Cavenee et al. were able to suggest a number of mechanisms for the loss (Figure 18.9). Later studies confirmed this interpretation by showing that in inherited cases, it was always the wild-type allele that was lost in this way.

Figure 18.9

Mechanisms of loss of wild-type allele in retinoblastoma. (A) Loss of a whole chromosome by mitotic non-disjunction. (B) Loss followed by reduplication to give (in the case studied by Cavenee et al.) three copies of the Rb chromosome. (C) Mitotic recombination (more...)

Loss of one marker allele but retention of the other is often seen because in both sporadic and inherited disease one hit is usually a point mutation while the other often involves loss of all or part of a chromosome. Inherited TS mutations are usually small-scale mutations - large chromosomal deletions would probably be lethal if carried in every cell of the body. Individual tumor cells may well be viable with a large deletion in heterozygous form, but large homozygous deletions are likely to be lethal even at the cell level. Thus both familial and sporadic tumors tend to retain markers surrounding a TS gene on one chromosomal homolog, but lose them from the other. However, if the wild-type allele is silenced by methylation rather than by deletion (Section 18.5.4), no loss of heterozygosity will be seen.

18.5.2 Rare familial cancers identify many TS genes

Following the example of retinoblastoma, many rare mendelian cancers are believed to involve TS genes via a two-hit mechanism (Table 18.4). As with retinoblastoma, mapping the genes in these rare families opens the way to positional cloning of TS genes. In many cases (APC, NF2, PTC for example) the TS gene identified in this way turns out to be important in the corresponding sporadic cancer, though this is not always the case - BRCA1 mutations are not found in sporadic breast tumors.

Table 18.4

Rare familial cancers caused by TS gene mutations.

18.5.3 Loss of heterozygosity (LoH) screening identifies locations of TS genes

By screening paired blood and tumor samples with markers spaced across the genome, we can discover candidate locations for TS genes (Figure 18.10). For meaningful results, a large panel of tumors must be screened with closely spaced markers. Highly polymorphic microsatellite markers are used in order to minimize the number of uninformative cases where the constitutional DNA is homozygous for the marker. Not all tumors will show the pattern of one small and one large abnormality that is needed to produce visible LoH. Advanced cancer cells often show LoH at as many as one quarter of all loci, so large samples are needed to tease out the specific changes from the general background chromosomal instability. Finally, most pathological tumor samples contain a mixture of intergrowing tumor and non-tumor (stromal) tissue, so that LoH shows as a decreased relative intensity (allelic imbalance) rather than total loss of the band from one allele (Figure 18.11A). Thus screening for LoH is quite laborious. Comparative genome hybridization (Figure 18.5) allows much easier screening for large deletions, but lacks the resolution required to detect small deletions. Using the CGH principle of competitive hybridization, but to DNA microarrays rather than chromosome spreads, may solve this problem (Pinkel et al., 1998).

Figure 18.10

Possible tumor suppressor genes on chromosome 3p. On the right are the results of typing constitutional and tumor DNA from a series of patients with oral tumors, using various markers (D3S1038, etc) from 3p (Wu et al., 1994). Color signifies loss of heterozygosity (more...)

Figure 18.11

Genetic changes in tumors. (A) Loss of heterozygosity. The normal tissue sample (N) is heterozygous for the marker D8S522 (arrows), while the tumor sample (T) has lost the upper allele. The bands higher up the gel are ‘conformation bands’, (more...)

Small homozygous deletions also occur, and are common at certain loci, such as CDKN2A (Section 18.6.3). Being small, homozygous deletions give an excellent pointer to the location of the TS gene. However, they also set a trap for the investigator. Markers that are homozygously deleted in the tumor tend to amplify from the contaminating stromal material, which is chromosomally normal. When there are overlapping deletions on the two homologs (Figure 18.12), the LoH results can appear to show evidence for two non-existent tumor suppressor genes, located either side of the one true gene. FISH is a good technique for confirming homozygous deletions. Alternatively, cell lines can be used where there is no problem of stromal contamination.

Figure 18.12

A pitfall in interpreting LoH data. The true event in this tumor is homozygous deletion of TSG. Because of amplification of contaminating stromal material (pale lines), LoH data show retention of heterozygosity at the TSG locus, with LoH at two flanking (more...)

18.5.4 Tumor suppressor genes are often silenced epigenetically by methylation

Tumor supressor genes may be silenced by deletion (reflected in loss of heterozygosity) or by point mutations, but there is increasing evidence for a third mechanism - DNA methylation (Versteeg, 1997; Jones and Laird, 1999). As explained in Section 8.4, cytosines in CpG dinucleotides are liable to be methylated, and when the cytosine lies in a CpG island within the promoter region of a gene, methylation is often associated with lack of expression of the gene. Many tumors show genome-wide disturbances of the normal methylation pattern. More specifically, CpG island methylation has been demonstrated for several tumor suppressor genes in a variety of cancers (Jones and Laird, 1999). At least in the cases of the CDKN2A, VHL, RB1 and MLH1 genes, there is evidence that promoter methylation causes silencing of the gene. In one study, 84% of colorectal tumors with microsatallite instability showed methylation of the MLH1 promoter (Herman et al., 1998), and in cell lines from such tumors MLH1 expression can be restored by treatment with the demethylating agent 5-aza-2′-deoxycytidine. Gene silencing by methylation is an example of an epigenetic mechanism (Section 16.4.2), a heritable change in gene expression that does not depend on a DNA sequence change. Standard techniques for mutation screening overlook changes in methylation, so their importance has probably been underestimated.

Box 18.2

Two-hit mechanisms may explain patchy mendelian phenotypes. It is possible that a two-hit mechanism similar to the Knudson model (Figure 18.9) may explain some non-cancer phenotypes that are mendelian in families but patchy in individuals. Why for example (more...)

18.6.1 Function of pRb, the RB1 gene product

The RB1 gene is widely expressed, encoding a 110-kd nuclear protein, pRb. In normal cells pRb is inactivated by phosphorylation and activated by dephosphorylation. Active (dephosphorylated) pRb binds and inactivates the cellular transcription factor E2F1, function of which is required for cell cycle progression (Figure 18.14). For reviews of pRb and E2F1, see Weinberg (1995, 1996). The G₁-S checkpoint seems to be the most crucial in the cell cycle; 2–4 hours before a cell enters S-phase, pRb is phosphorylated. This releases the inhibition of E2F1 and allows the cell to proceed to S phase. Phosphorylation is governed by a cascade of cyclins, cyclin-dependent kinases and cyclin kinase inhibitors.

RB1 gene mutations produce sporadic or inherited retinoblastoma, this being the classic example of Knudson's two-hit hypothesis (Figure 18.8). It is not clear why constitutional mutation of a gene so fundamental to cell cycle control should result specifically in retinoblastoma and a small number of other tumors, principally osteosarcomas. However, this is a common theme in molecular pathology: mutation of a gene produces a phenotypic effect in only a subset of the cells or tissues in which the gene is expressed and appears to have a function (Section 16.7.1). The product of the MDM2 oncogene (which is amplified in many sarcomas) binds and inhibits pRb, thus favoring cell cycle progression. Several viral oncoproteins (adenovirus E1A, SV40-T antigen, human papillomavirus E7 protein) also bind and sequester or degrade pRb.

18.6.2 Function of p53, the TP53 gene product

p53 was first described in 1979 as a protein found in SV40-transformed cells, where it associated with the T-antigen. Later, the TP53 gene which encodes p53 appeared as a dominant transforming gene in the 3T3 assay (Figure 18.4), and so was classed as an oncogene. Subsequently it transpired that while p53 from some tumor cells was oncogenic, p53 from normal cells positively suppressed tumorigenesis.

Loss or mutation of TP53 is probably the commonest single genetic change in cancer. This reflects the central importance of p53, which has several functions in the cell. One is as a transcription factor. Tetramers of p53 bind DNA and can activate transcription of reporter genes placed downstream of a p53 binding site. However, p53 is believed to have a much broader role in the cell, which has been summarized as ‘the guardian of the genome’. One of its guardian functions is to stop cells replicating damaged DNA (Figure 18.14). Normal cells with damaged DNA arrest at the G₁-S cell cycle checkpoint until the damage is repaired, but cells that lack p53 or contain a mutant form do not arrest at G₁. Replication of damaged DNA presumably leads to random genetic changes, some of which are oncogenic, similar to cells with a defective mismatch repair system (see below).

Probably related to this, p53 has a crucial role in cell death. In response to oncogenic stimuli, cells undergo apoptosis (programmed cell death). Apoptosis has come to occupy a central place in our understanding of the cancer process (reviewed by Fisher, 1994). It is one of the main higher-level controls that protect the organism against the consequences of the natural selection among its constituent cells described at the start of this chapter. Tumor cells lacking p53 do not undergo apoptosis, and so escape the control. p53 may be knocked out by deletion, by mutation or by the action of an inhibitor such as the MDM2 gene product (which binds p53 and targets it for degradation; MDM2 also binds pRb, see above) or the E6 protein of papillomavirus.

Loss of heterozygosity assays confirmed the status of TP53 as a tumor suppressor gene. TP53 maps to 17p12, and this is one of the commonest regions of loss of heterozygosity in a wide range of tumors. Tumors that have not lost TP53 very often have mutated versions of it. To complete the picture of TP53 as a TS gene, constitutional mutations in TP53 are found in families with the dominantly inherited Li-Fraumeni syndrome (MIM 151623). Affected family members suffer multiple primary tumors, typically including soft tissue sarcomas, osteosarcomas, tumors of the breast, brain and adrenal cortex, and leukemia (Figure 18.15).

Figure 18.15

A typical pedigree of Li-Fraumeni syndrome. Malignancies typical of Li-Fraumeni syndrome include bilateral breast cancer diagnosed at age 40 (I₂); a brain tumor at age 35 (II₁); soft tissue sarcoma at age 19 and breast cancer at age 33 (II₃); breast cancer (more...)

18.6.3 Function of CDKN2A and ARF, the CDKN2A gene products

The remarkable gene variously called MTS1, INK4A and CDKN2A at 9p13 encodes two structurally unrelated proteins (Figure 18.16). Exons 1α, 2 and 3 encode the CDKN2A (p16^INK4A) protein. A second promoter starts transcription further upstream at exon 1β. Exon 1β is spliced on to exons 2 and 3, but the reading frame is shifted, so that an entirely unrelated protein ARF (p19^ARF) (Alternative Reading Frame) is encoded.

Figure 18.16

The two products of the CDKN2A gene. This gene (also known as MTS and INK4A) encodes two completely unrelated proteins. CDKN2A, or p16^INK4A, is transcribed from exons 1α, 2 and 3, and ARF, or p19^ARF, from exons 1β, 2 and 3 - but with a (more...)

Both gene products function in cell cycle control (Figure 18.14). CDKN2A functions upstream of the RB1 protein in control of the G₁-S cell cycle checkpoint. Cyclin-dependent kinases inactivate pRb by phosphorylation, but CDKN2A inhibits the kinases (Weinberg, 1995). Thus loss of CDKN2A function leads to loss of RB1 function and inappropriate cell cycling. The other product of the CDKN2A gene, ARF, mediates G₁ arrest by destabilizing MDM2 (Pomeranz et al., 1998). MDM2 binds to p53 and induces its degradation. Thus ARF acts to maintain the level of p53. Loss of ARF function leads to excessive levels of MDM2, excessive destruction of p53, and hence loss of cell cycle control.

Inherited mutations, usually involving just the CDKN2A product, are seen in some families with multiple melanoma, but somatic mutations are very much more frequent. Homozygous deletion of the CDKN2A gene inactivates both the RB1 and the p53 arms of cell cycle control, and is a very common event in the development of many tumors. Other tumors have mutations that affect the CDKN2A but not the ARF product (e.g. inactivation of the 1α promoter by methylation). These tumors tend also to have p53 mutations, showing the importance of inactivating both arms of the control system shown in Figure 18.14.

18.7.1 Nucleotide instability is manifest as defects in DNA replication or repair

Replication error repair defects

Loss of heterozygosity studies on colon cancers (Section 18.5.3) produced novel and unexpected results in some patients. Rather than lacking alleles present in the constitutional DNA, some tumor specimens appeared to contain extra, novel, alleles of the microsatellite markers used. LoH is a property of particular chromosomal regions, but the microsatellite instability (MSI or MIN) seemed to be general. Tumors could be classified into MSI⁺ or MSI^-. MSI⁺ tumors gained alleles for a good proportion of the markers tested, regardless of their chromosomal location. Figure 18.11B shows an example.

Microsatellite instability is a characteristic feature of the autosomal dominant hereditary non-polyposis colon cancer. HNPCC genes were mapped to two locations, 2p15-p22 and 3p21.3. In a wonderful example of lateral thinking, Fishel et al. (1993) related the MSI⁺ phenomenon to so-called mutator genes in E. coli and yeast. These genes encode an error-correction system that checks the DNA for mismatched base pairs (Figure 18.17). Because the E. coli Dam system methylates adenine in GATC sequences, but not until some time after DNA replication, the system can recognize the newly-synthesized strand if it has not yet been methylated, and it can cut out and resynthesize the DNA surrounding a mismatch on this strand. Mutations in the genes that encode the MutHLS error-correction system lead to a 100 to 1000-fold general increase in mutation rates. Fishel and colleagues cloned the human homolog of one of these genes, MutS, and showed that it mapped to the location on 2p of one of the HNPCC genes and was constitutionally mutated in some HNPCC families. In all, five human homologs of the E. coli MutS or MutL genes have been implicated in cancer (Eshleman and Markowitz, 1996; Peltomaki and de la Chapelle, 1997; see Table 18.5.

Figure 18.17

The MutHLS error correction system in E. coli. A replication error introduces a mismatch (A). The MutS protein binds to mismatched base pairs (B). In an ATP-dependent reaction, a MutS-MutL-MutH complex is formed which probably brings any GATC sequence (more...)

Table 18.5

Genes involved in DNA replication error repair.

Patients with HNPCC are constitutionally heterozygous for a loss-of-function mutation. Their normal cells still have a functioning mismatch repair system and do not show the MSI⁺ phenotype. In a tumor, the second copy is lost by one of the mechanisms shown in Figure 18.9. Interestingly, these tumors are unusual in having relatively normal karyotypes - showing that while genetic instability is important in cancer, it may be achieved either through instability at the nucleotide level or by chromosomal instability.

Microsatellite instability is seen in about 13% of colorectal, gastric and endometrial carcinomas, but only occasionally in other tumors. Mismatch repair defects should affect all cells - why are they a feature of only a fairly restricted set of tumors? One explanation may be that in MSI⁺ cells mutations in longer homopolymeric runs occur 1000 times as often as mutations of microsatellites. The TGFβ receptor II gene, which has an A₁₀ run, has frameshifting mutations in 90% of MSI⁺ cancers. Loss of TGFβ signalling seems to be an important step in development of colorectal cancer. In MSI^- colorectal cancer, this system is often inactivated by loss of a downstream effector, SMAD4 on chromosome 18 (White, 1998).

18.7.2 Chromosomal instability is a very common feature of cancer cells

Malignant tumor cells that do not show microsatellite instability usually have bizarrely abnormal karyotypes, with many losses, gains and rearrangements of chromosomes (Figure 18.6). DNA studies reinforce this picture of chromosomal instability: cells from a typical advanced colon, breast or prostate cancer show loss of heterozygosity at around one quarter of all loci. Only a few of the changes seem to be causally connected with the cancer; mostly they are a reflection of a general chromosomal instability.

Cell fusion experiments (Lengauer et al., 1998) suggest that chromosomal instability is a dominant phenotype, and so presumably the result of a single mutant allele at the relevant locus. In yeast, mutations in many different genes can lead to chromosomal instability. The genes concerned are involved in chromosome condensation, functioning of the centromere and kinetochore, or checkpoints such as a spindle checkpoint that prevents chromatids separating until the chromosome is correctly aligned on the spindle. Identification of the corresponding human genes and investigation of their role in cancer has only just begun.

18.7.3 A DNA damage checkpoint that prevents cells containing damaged DNA from entering mitosis is often inactivated in cancer cells

Normal cells with unrepaired DNA damage do not enter mitosis. Details of this checkpoint mechanism are coming to light, mainly from experiments in yeast (Nurse, 1997; Weinert, 1998). In the presence of DNA damage the CDC2 cyclin-dependent kinase, which is the immediate controller of entry into mitosis, is inactive, and levels of p53 protein are raised. Among the mammalian genes involved in the damage checkpoint are ATM and maybe ATR, BRCA1 and BRCA2.

ATM is the gene responsible for ataxia telangiectasia (AT, MIM 208900), a rare recessive combination of cerebellar ataxia, telangiectasia (dilation of blood vessels in the conjunctiva and eyeballs), immunodeficiency, growth retardation and sexual immaturity. AT patients have a strong predisposition to cancer. Homozygotes usually die of malignant disease before age 25, and there have been suggestions that heterozygotes have a raised risk of cancer - for example a 3.9-fold increased risk of breast cancer among women (Easton, 1994). AT affects about one person in 100 000 in the UK and USA, so the Hardy-Weinberg distribution (Section 3.3.1) suggests that one person in 158 of the population is heterozygous. If their raised risk of cancer is confirmed, this would represent a significant cancer risk at the population level.

In vitro, cells of AT patients show chromosomal instability and hypersensitivity to ionizing radiation or radiomimetic chemicals, even though DNA repair appears to be normal. Unlike normal cells, AT cells fail to accumulate p53 after irradiation. The ATM gene product has protein kinase activity, and activates p53 by phosphorylation of serine-15 (reviewed by Nakamura, 1998). Thus part of the DNA-damage checkpoint involves the ATM protein reacting to damage by activating p53. A related protein, ATR, appears to have an opposite effect: overexpression (rather than loss of function) of ATR inhibits the p53 response to DNA damage (Friend and Tapscott, 1998). Possibly the breast cancer genes, BRCA1 and BRCA2, may also form part of the checkpoint mechanism. In all cases, if cells replicate damaged DNA, not only may mutations be propagated, but also DNA breaks or crosslinks may predispose to the chromosome deletions, translocations and inversions that are such a common feature of tumor cells.

18.7.4 Telomerase, the stability of chromosomal ends and the immortality of cancer cells

As we saw in Section 2.3.5, the ends of human chromosomes are protected by a repeat sequence (TTAGGG)_n, that is maintained by a special RNA-containing enzyme system, telomerase. Telomerase is present in the human germ line but is absent in most somatic tissues, and telomere length declines with time in normal somatic cells, a phenomenon which may contribute to the ‘mitotic clock’ that limits the number of divisions a cell can go through. There has been much excitement over the discovery that 90% of human primary tumors possess telomerase activity. Maybe this is the key to their immortality, and maybe an anti-telomerase agent would limit their mitotic potential. Indeed, ectopic expression of hTERT, the catalytic subunit of human telomerase, allows cells that were destined to senesce to multiply indefinitely - but in many cell types this also requires inactivation of the G₁-S controls shown in Figure 18.14 (Weinberg, 1998).

18.8.1 Vogelstein's concept of ‘gatekeeper’ genes describes, even if it does not explain, the tissue specificity of many genetic cancers

The first mutation in the multistep evolution of a tumor is critical because it should confer some growth advantage on an otherwise normal cell. Sometimes the reason why a genetic change causes a particular type of cancer and not another is understandable - translocating the MYC oncogene into an immunoglobulin locus, for example, is likely to cause trouble for cells that express immunoglobulins at high levels. More often, however, no such connection is apparent. Why should mutations in the RB1, TP53 and MSH2 genes cause retinoblastoma, breast cancer and HNPCC, respectively, when the proteins encoded by these genes play important roles in virtually all cells? According to the gatekeeper hypothesis (Kinzler and Vogelstein, 1996), in a given renewing cell population one particular gene is responsible for maintaining a constant cell number. A mutation of a gatekeeper leads to a permanent imbalance of cell division over cell death, whereas mutations of other genes have no long-term effect if the gatekeeper is functioning correctly.

The tumor suppressor genes identified by studies of mendelian cancers are the gatekeepers for the tissue involved - NF2 for Schwann cells, VHL for kidney cells, and so on. The reason why one particular widely expressed gene should be the gatekeeper for just one cell type must be buried in the complex networks of interactions that connect input to output in cell behavior (Figure 18.13C). An accumulation of minor differences leaves a certain gene with a predominant effect in one cell type but not another. For those cancers that do not have mendelian versions, maybe there is not a single gatekeeper.

18.8.2 The colorectal cancer model

Most colon cancer is sporadic. Familial cases fall into two categories:

• Familial adenomatous polyposis (FAP or APC: MIM 175100) is an autosomal dominant condition in which the colon is carpeted with hundreds or thousands of polyps. The polyps (adenomas) are not malignant, but if left in place, one or more of them is virtually certain to evolve into invasive carcinoma. The condition has been mapped to 5q21 and the gene responsible, APC, identified.

• Hereditary non-polyposis colon cancer (HNPCC; MIM 120435, 120436) is also autosomal dominant and highly penetrant, but unlike FAP there is no preceding phase of polyposis.

Malignant colorectal carcinomas develop from normal epithelium through microscopic dysplastic aberrant crypt foci to benign epithelial growths called adenomas. Adenomas can be classified into early (less than 1 cm in size), intermediate (more than 1 cm but without foci of carcinoma) or late (more than 1 cm and with foci of carcinoma). Adenomas develop into carcinomas, which eventually metastasize. Although there is not one invariant sequence of mutations in the development of every colorectal carcinoma, the most likely sequence is one where each successive step confers a growth advantage on the cell. Pointers to the commonest sequence include the following observations:

• In FAP, constitutional loss of one copy of the APC gene on 5q21 is sufficient to carpet the colon with adenomatous polyps. The very earliest detectable lesions in either sporadic colorectal cancer or FAP, dysplastic aberrant crypt foci, lack all APC expression. In FAP, about 1 epithelial cell in 10⁶ develops into a polyp, a rate consistent with loss of the second APC allele being the determining event. The APC product binds β-catenin, and thus represses transcription by the TCF7L2 transcription factor. Sporadic colorectal cancers sometimes have an intact APC gene but achieve the same effect by activating mutations of β-catenin, which therefore acts as an oncogene in this context.

• About 50% of intermediate and late adenomas, but only about 10% of early adenomas, have mutations in the KRAS oncogene (a relative of HRAS, Table 18.2). Thus KRAS mutations may often be involved in the progression from early to intermediate adenomas.

• About 50% of late adenomas and carcinomas show loss of heterozygosity on 18q. This is relatively uncommon in early and intermediate adenomas. It seems likely that the relevant gene is SMAD4 rather than the initial candidate, DCC (see White, 1998).

• Colorectal cancers, but not adenomas, have a very high frequency of mutations in the TP53 gene.

These are not the only changes seen in colorectal carcinomas, but they are the ones which can be most readily associated with specific stages, and they lead to the model shown in Figure 18.18. Analogous schemes could no doubt be constructed for other cancers if we had better knowledge of the early stages. The mutator genes MSH2, MSH1, etc., that are mutated in HNPCC are not directly involved in this pathway. Several genes have been identified that are frequently mutated in cancers with microsatellite instability (TGFBR2, RAS, E2F-4, BAX, etc.), and it seems likely that the mutator phenotype can either accelerate progress along the pathway shown in Figure 18.18, or facilitate alternative routes to malignant transformation.

Figure 18.18

Fearon and Vogelstein's model for the development of colorectal cancer. This is primarily a tool for thinking about how tumors develop, rather than a firm description. Every colorectal carcinoma is likely to have developed through the same progression (more...)