Introduction

Top Abstract Introduction Materials and Methods Results and Discussion References

Persistent smoking dramatically increases the risk of death from lung cancer (1). Of all of the carcinogenic effects of tobacco smoke, its mutagenic action is believed today to be the major cause of human lung malignancy (2, 3). The main data in support of that viewpoint came from the p53 tumor suppressor gene mutations. p53 is the "emergency brake" gene that expresses its tumor-preventing apoptotic and cell cycle checkpoint functions in physiologically stressful situations (4, 5). Mutational alteration of these functions often gives p53 mutant cells a proliferative advantage over wild-type p53⁺ cells. Of the 3,800 potential mutagens contained in tobacco smoke (6), the best candidate for the role of a causative mediator between smoking, p53 mutations, and lung cancer is benzo[a]pyrene diol epoxide (BPDE), a metabolite of benzo[a]pyrene, one of the polycyclic aromatic hydrocarbons found in cigarette smoke. BPDE binds to DNA and forms bulky adducts at the N2 position of guanines (7). If the adducts remain unrepaired, their misreading by DNA polymerase during replication results predominantly (70%) in GT transversions (8, 9); this prevalence is, in fact, the mutational signature of BPDE. GT transversions are frequent (35%) in human lung cancer, but they are uncommon (<10%) in most other cancers, thus indicating a possible link between BPDE and lung cancer (2, 9). This hypothesis gained support (10) when the frequency distributions of in vitro BPDE adduct formation and in vivo p53 mutations of lung cancers along the human p53 gene were compared. Strong BPDE adduct signals were detected at the guanines of codons 157, 248, and 273, the mutational hot spots in human lung cancer. This coincidence led the authors (10) to conclude that "targeted adduct formation rather than phenotypic selection appears to shape the p53 mutational spectrum in lung cancer." Whereas codons 248 and 273 are among major mutational hot spots in virtually all cancers, codon 157 was claimed to be a hot spot unique to lung cancer (10).

In mutationally characterizing the p53 gene, one way is to plot the frequency of each of 12 different types of base substitutions in the gene as a whole (Fig. 1A), which we will call hereon a p53 mutational pattern. Another way (Fig. 2) is to plot a frequency of the particular 1 of these 12 types as a function of a position in the p53 gene, which we refer to as a p53 mutational spectrum.

View larger version (26K): [in this window] [in a new window]   Fig. 1.   p53 mutational patterns of lung adenocarcinomas from smokers (99 mutations) and nonsmokers (50 mutations). Because the strand with original lesions is identifiable only for (GT, CA), (GC, CG), and (CT, GA) pairs of complementary substitutions, we used only these three pairs to test the "mutagenesis" vs. "strand-asymmetric repair" alternative (see text). (A) In a standard (strand-nonspecific) representation, the patterns of smokers and nonsmokers show a difference, although statistically insignificant (P value 0.151). (B) The same patterns as in A, but organized as pairs of complementary transitions and transversions; the upper part may represent a nontranscribed strand, and the lower part, a transcribed strand.

View larger version (20K): [in this window] [in a new window]   Fig. 2.   Comparison of the p53 spectra of GT transversions from lung cancer (radon-, asbestos-, and mustard gas-associated cases excluded) of ever smokers and cancers in non-lung tissues least accessible to smoke (see Table 1). Hot- and "warm"-spot codons are indicated. Codon 248, one of the strongest BPDE targets, appears not to be a GT transversion hot spot in either lung cancer or non-lung cancer spectra.

Thus, the previously proposed causative chain was as follows: (i) the major risk factor for lung cancer is smoking, (ii) smoke contains benzo[a]pyrene, (iii) BPDE forms bulky adducts at G bases of DNA, (iv) the adducts cause GT transversions, (v) these transversions are a hallmark of mutant p53 genes from lung cancers, and (vi) hot spots of GT transversions coincide with preferential sites of p53 DNA-BPDE adducts. However, as noted in ref. 11, some pivotal epidemiological evidence required was missing. If BPDE is a major initiating mutagen that shapes the lung cancer p53 mutational pattern and spectrum of smokers, then one would expect both the pattern and the spectrum to be essentially different in the lung tumors of confirmed nonsmokers. Until recently, published p53 mutational data in nonsmokers were sparse; consequently, reported differences (3, 12, 13) should be viewed skeptically. Furthermore, tumorigenesis is a multistep process, and the interaction of BPDE with p53 DNA may represent only one step. Therefore, even if the p53 pattern and spectral differences between smokers and nonsmokers were reliably demonstrated, other contributory causes, e.g., repair and selection, must be considered in addition to possible primary, BPDE-like adduct formation. With this in mind, we compared p53 mutational patterns and spectra in contrasting epidemiological groups by using the information obtained from the p53 mutation catalog (14) and updated by the recently available data.

First, we found a significant difference between smokers and nonsmokers in the frequency of mutations presumably originating in the transcribed and nontranscribed strands. However, within-strand p53 mutational patterns of the two epidemiological groups are, in fact, identical. Second, the distribution of GT transversions along the p53 gene is also the same in lung cancers of smokers and in cancers of other, smoke-inaccessible tissues. Third, smokers do not differ from nonsmokers in the frequency of tumors with silent p53 mutations. Additionally, silent G:CT:A transversions in lung cancer do not correlate with the "silent spectrum" of BPDE adducts.

These results suggest the need to revise the tobacco smoke-associated etiology of lung cancers with p53 mutations. Although not entirely denying the implication of BPDE-like carcinogens as specific exogenous mutagens, the main purpose of this revision is to emphasize endogenous sources of mutations such as oxidative DNA damage (15, 16) and nongenotoxic stresses (4, 17). We propose that smoking aggravates these stresses and, thus, intensifies the processes of strand-asymmetric repair and selection of tumorigenic stem cells carrying mutations in stress-responsive genes such as p53. Although lung cancers with p53 mutations are more frequent in smokers than in nonsmokers, the increased selection rather than mutation rate may account for this difference in tissues insulted by smoking.

	Materials and Methods

Top Abstract Introduction Materials and Methods Results and Discussion References

Table 1 introduces the main groups of p53 mutations analyzed. The mutations were retrieved from the International Agency for Research on Cancer p53 database (April 1999 update; 10,396 entries) and augmented by later data. The associated habits of smoking, snuffing, and chewing tobacco and drinking alcohol and hot beverages, etc., were confirmed from original papers according to the references in ref. 14. The relatively infrequent dubious cases such as "passive smokers" and "low drinkers" were not considered. The resulting lung cancer p53 data set was partitioned into three risk groups: smokers, nonsmokers, and patients with an unknown smoking history. To achieve a sufficient sample size, we pooled GT transversions from tumors of several, supposedly smoke-inaccessible tissues (Table 1). The resulting sample of GT transversions was large enough to be reliably compared with that of lung cancer.

View this table: [in this window] [in a new window]   Table 1.   Tumor tissues classified in regard to the frequency of G:CT:A transversions from the coding part of the p53 gene as reported in the International Agency for Research on Cancer p53 database (April 1999 update)*

For statistical analysis of p53 mutational patterns and spectra, we used a number of tests. The popular Pearson ² test, although nonparametric, is based on the expectation that, within any category, mutation frequencies are distributed normally. Therefore, if the observed mutation frequencies for some categories are very low, the results of the ² test become invalid. This is the case with the p53 mutational spectra; it covers more than 100 sites but, except for the hot spots, many of these sites contain too few mutations for ² analysis. The Adams and Skopek test (20) uses a Monte-Carlo simulation to estimate a P value of the hypergeometric bivariate (or multivariate) tabular analysis test and is recognized as superior to the ² test by virtue of being able to cope with infrequent mutations. Lower P values indicate greater dissimilarity of the two spectra in question. For example, a P value of less than 0.05 indicates a statistically significant (at 5% level) difference between spectra. Similarly, high P values (e.g., 0.9) mean that both samples generated essentially the same spectrum.

We used HG-PUBL software (http://metalab.unc.edu/dnam/di-hypg.htm) with number of Monte-Carlo iterations set at 10,000 to make sure the sufficient test power was achieved (at least 1,700 iterations are recommended in ref. 21).

	Results and Discussion

Top Abstract Introduction Materials and Methods Results and Discussion References

Complementary Base Substitutions in p53 from Lung Cancer Distinguish Smokers from Nonsmokers. In Fig. 1, p53 mutational pattern is plotted in two ways. A standard way (Fig. 1A) reveals a difference between smokers and nonsmokers that could be attributable to a direct action of BPDE-like mutagens on p53 DNA. However, the second method (Fig. 1B) makes the same patterns point to a different cause. All 12 types of base substitutions are grouped into 6 pairs, each consisting of two complementary substitutions: (CT, GA) (GT, CA), etc. The overall p53 pattern (Fig. 1A) then is shown as two complementary subpatterns in the following fashion. In each pair, we select the substitution that is more frequent than the other in lung cancers of smokers. Six such substitutions plotted in decreasing order of magnitude form the first subpattern (above the line in Fig. 1B). Their complementary substitutions form the second subpattern (under the line in Fig. 1B). This new representation of the p53 mutational pattern shows that the excesses of GT over CA, GC over CG, AG over TC, etc., observed in smokers turned out to be either less pronounced in nonsmokers (in the case of GT over CA) or even reversed (excesses of CG over GC and GA over CT). At the same time, the differences in Fig. 1A disappear in Fig. 1B. In fact, the two parallel subpatterns (upper as well as lower in Fig. 1B) are almost perfectly correlated. The following six tests develop this surprising observation.

Testing the Strand-Specific p53 Patterns of Smokers and Nonsmokers. The majority of p53 base substitutions in lung cancer involve G:C pairs, both in smokers (82.5%; 250 of 303) and in nonsmokers (82.9%: 58 of 70). Ninety-eight of 250 (39.2%) of these substitutions in smokers and 23 of 58 (39.6%) in nonsmokers occur at CpG sites. Two transitions in CpGs, CT and GA, are believed to originate from one event, namely, deamination of 5-mC, but in different strands (22, 23). If deamination occurs in the nontranscribed (coding) strand (NTS), the result is a CpGTpG transition; if the mirror 5-mC is deaminated in the transcribed (noncoding) strand (TS), replication converts it to a complementary CpGCpA transition in the coding strand. As to the complementary transversions (GT, CA and GC, CG), both presumed exogenous and endogenous sources, e.g., BPDE and oxidative DNA damage, respectively, point to modified Gs as primary lesions (reviewed in ref. 16). When such lesions occur in the NTS, they produce GT and GC transversions. Accordingly, the complementary CA and CG transversions most probably are derived from the similar lesions originating in the transcribed strand (24).

Testing p53 Spectra of GT Transversions in Lung Cancer of Smokers, Nonsmokers, and Non-Lung Cancers. This surprising lack of within-strand differences prompted us to look closer at how the lung cancer p53 mutational spectrum of smokers was compared (and contrasted) so far to (i) the lung cancer p53 spectrum of nonsmokers and (ii) the p53 spectrum of cancers in smoke-inaccessible tissues. The BPDE hypothesis predicts that these two control groups will show different p53 spectra for GT transversions whereas the alternative selection hypothesis implies no such difference. Therefore, to find out whether origin or selection of mutations is the major shaper of the lung-specific p53 spectrum, one should focus on GT transversions for both smoker vs. nonsmoker and lung vs. non-lung comparisons because it is precisely GT transversions that BPDE predominantly generates. Instead, all comparisons published so far (3, 12, 13) have been lumped together, thus making the spectrum-shaping roles of BPDE and selection indiscernible.

The Mutational Pattern at Codon 157 Is Not Unique to Lung Cancer. In the p53 spectra limited to GTs, codon 157 is equally frequent in lung and non-lung cancers (Fig. 2). However, codon 157 appears to be a hot spot unique to lung cancer in previous reports (3, 12, 13) in which mutations of different types were pooled together along p53. Fig. 3 explains this "paradox." Two neighboring codons, 156 (CGC) and 157 (GTC), form a singular CpG between them (Fig. 3A). The guanine of this CpG is one of the most preferred BPDE-binding targets (10) and also is a GT transversion hot spot in lung cancer. However, it was overlooked that in this particular codon, GT prevails over any other type of mutations virtually in all tumor tissues (Fig. 3B). Colon tissue may serve as the most instructive example because this tissue rarely contacts benzo[a]pyrene and the overall p53 pattern of colon cancer is notable for having more than 50% of endogenous CT/GA transitions at CpG sites. However, in codon 157, both of these cancers show virtually the same excess of GTs: 54.5% in lung vs. 50% in colon (Fig. 3B). In fact, this excess is not lung-specific at all (Fig. 3B). The explanation suggests itself: compared with GTs, all other substitutions in codon 157 are much less tumorigenic.

View larger version (29K): [in this window] [in a new window]   Fig. 3.   (A) The scheme explaining high selectivity of GT compared with all other changes at the CpG composed by codons 156 and 157. Underlined are guanines that show strong BPDE binding (10). (B) p53 mutational pattern of codon 157 showing cancer-common excess of GT transversions; radon-associated cases are excluded.

Silent p53 Mutations in Tumors Do Not Correlate with BPDE Adducts. Most silent p53 mutations in human tumors seem to be selectively neutral somatic changes (27). Therefore, they turn up in tumors only as hitchhikers of nonsilent, tumor-driving mutations (25, 26). The hypothesis of BPDE-induced p53 mutations predicts that lung cancers should contain many more silent GT transversions in smokers than in nonsmokers and more often in codons 248, 267, and 282 (all of CGG structure) because their second guanines show rather strong BPDE adduct signals (10). Free of interference by selection, silent GTs in the p53 gene as a whole (and in these three codons in particular) are more suitable than nonsilent GTs for testing BPDE-induced mutagenesis in tumors. In reality, of 23 somatic silent p53 mutations in smokers (pooled for lung, oral cavity, and esophagus), there is only 1 GT (in esophagus), and it is located in codon 173 (GTG), which is not a BPDE target (10).

The Hypothesis of Smoke-Induced Selection of p53 Mutations. Regarding the mutagenesis vs. selection alternative, it is important to distinguish the following two comparisons of p53 mutations in smokers and nonsmokers. The first comparison quantitatively sets off smokers against nonsmokers within the group of lung cancers carrying p53 mutations. The overall ratio is 344:75 = 4.6:1, tumors with multiple p53 mutations being particularly contrasting (Table 2). It may seem quite natural to relate this difference solely to direct mutagenic effects of smoking. However, this might be a simple effect of the difference in the number of cell divisions because smoking is known to increase the rate of cell proliferation in the bronchial epithelium. Moreover, the role of selection cannot be excluded, even in the case of p53 multiplets. The point is that cells use the p53 gene nonconstitutively, only under stressful conditions (4, 5), in response to either DNA-damaging insults like ionizing radiation and drugs, or, more commonly, "nongenotoxic" hypoxia (4, 17). Wild-type p53 protein prevents tumor development either by committing stressed cells to apoptosis or by arresting them mainly at the G₁ stage of the cell cycle, before DNA replication (29, 30). Cells with mutant p53 often escape apoptosis and G₁ arrest and enter the S phase followed by cell division and proliferation. Our previous retrospective analysis of tumors with multiple p53 mutations (26) showed that the majority of their primary, mutation-prone changes also might originate in nondividing stem cells, during p53-regulated G₁ arrest, i.e., precisely when a potential carcinogenic advantage of mutant p53 protein over a normal one may be brought into effect by selection. If so, the higher frequency of p53 mutations in smokers (Table 1) may reflect not only the smoke-induced mutagenesis but also the increased selection pressure on the p53 gene in tissues insulted by smoking.

Silent p53 Singlets in Tumors as Markers of Mutability. Finally, we address the key question of whether tobacco smoking raises a mutation rate in p53 during lung tumorigenesis. All of the above comparisons were inconclusive in this regard because of direct or indirect effects of selection. The only class of p53 mutations that can be used for this purpose are single, silent base substitutions (Table 2). Because they are not accompanied by nonsilent p53 mutations, changes in other genes most likely "picked them up" during tumorigenesis. This means that silent p53 singlets represent tumors driven by p53-independent selection. Then, the frequency of tumors with silent p53 singlets reflects only the mutation rate in p53. Remarkably, smokers and nonsmokers turned out to be indistinguishable in this frequency (Table 2).

Concluding Remarks. There is a tendency of current cancer research to downplay rather evident physiological insults caused by exogenous carcinogens in favor of their cancer-prone mutagenic action. Primarily nongenotoxic, these insults nevertheless change conditions for genetic processes such as strand- and sequence-specific repair and selection of tumorigenic mutations in genes such as p53. What if physiologically induced repair and selection may affect not only the base-to-base spectrum of p53 mutations, but even their overall pattern? Comparison of the standard (Fig. 1A) and our new graphic representation of p53 mutational patterns (Fig. 1B) gives a positive answer. Parallel patterns in Fig. 1B may reflect repair-associated differences between the two complementary DNA strands of p53 that acquire mutations and how smoking, as many other stressors of cells, exacerbates this difference.