- Email: [email protected]
Role in tumorigenesis of silent mutations in the TP53 gene
Mutation Research 457 (2000) 93–104
Role in tumorigenesis of silent mutations in the TP53 gene Bernard S. Strauss∗ Department of Molecular Genetics and Cell Biology, The University of Chicago, 920 East 58th Street, Chicago, IL 60637, USA Received 15 July 2000; received in revised form 8 September 2000; accepted 18 September 2000
Abstract Over 10,000 mutations in the TP53 suppressor gene have been recorded in the International Agency for Research on Cancer (IARC) tumor data base. About 4% of these mutations are silent. It is a question whether these mutations play a role in tumor development. In order to approach this question, we asked whether the reported silent mutations are randomly distributed throughout the TP53 gene. The p53 data base was searched exon by exon. From the frequency of codons with no silent mutations, the average number of silent mutations per codon for each exon was calculated using the Poisson distribution. The results indicate the distribution to be non-random. About one-third of all silent mutations occur in “hot-spots” and after subtraction of these hot-spots, the remaining silent mutations are randomly distributed. In addition, the percentage of silent mutations among the total in the silent mutation hot-spots is close to that expected for random mutation. We conclude that most of the silent mutations recorded in tumors play no role in tumor development and that the percentage of silent mutation is an indication of the amount of random mutation during tumorigenesis. Silent mutations occur to a significantly different extent in different tumor types. Tumors of the esophagus and colon have a low frequency of silent mutations, tumors of the prostate have a high frequency. © 2000 Elsevier Science B.V. All rights reserved. Keywords: TP53; Silent mutation; Hyper-mutability; Selection; Genetic instability
1. Introduction Tumors often include new mutations not present in the surrounding tissue [1]. Two explanations have been proposed to account for this finding. The first supposes that tumors are genetically unstable and produce orders of magnitude more point mutations than normal tissue [1,2]. The finding that organisms (including humans) with deficiencies in DNA repair pathways are both hyper-mutable and tumor-prone provides strong support for this view [3,4]. A second explanation supposes that normal mutability accounts for the production of mutations which accumulate in the tumor tissue because they are selected in successive waves ∗ Tel.: +1-773-702-1628; fax: +1-773-702-3172. E-mail address: [email protected] (B.S. Strauss).
of clonal development [5,6]. These hypotheses are not mutually exclusive. Tumors might be hyper-mutable at some stage but selection might still determine which mutations are found. Since the primary observation is the finding that tumors include many new mutations, the demonstration of hyper-mutability requires a marker which does not provide a selective advantage. TP53 silent mutations, in which a nucleotide has been altered to produce a synonymous codon without amino acid change [7], may provide such a marker [8–12]. However, it must first be demonstrated that such mutations are truly non-selective since there is evidence that not all synonymous codons are utilized to an equal extent. Li [7] gives as an example the finding that for a particular E. coli protein, 21 of 23 leucine residues are encoded by CUG although there are five other codons available. Synonymous codons are used at very differ-
0027-5107/00/$ – see front matter © 2000 Elsevier Science B.V. All rights reserved. PII: S 0 0 2 7 - 5 1 0 7 ( 0 0 ) 0 0 1 3 5 - 4
94
B.S. Strauss / Mutation Research 457 (2000) 93–104
ent frequencies and the different p53 codons are not used equally [13]. One explanation is that the abundance of the different tRNA species differs and that the codon corresponding to the most abundant tRNA might have an advantage over one specifying a rare species [7,14]. A mutation from a codon using a rare tRNA to one using a more abundant species would have a selective advantage although silent. Alternatively, a silent mutation at an exon/intron border which changes a splice site would be expected to result in a major change in the protein gene product [15]. Such a change has been observed in a mutation leading to human muscular dystrophy [16] and results in a changed pattern of transcription. Therefore, before silent mutations in TP53 can be used as an indicator of mutability it is necessary to show that they are really neutral and do not confer some selective advantage to the tumor. Mutation is assumed to be a random process in a hypothetical DNA of uniform structure. This hypothesis supposes that each base in a sequence be equally susceptible to change and mutates to each of the other three with equal frequency. In practice, this ideal situation is rarely, if ever, observed. Particular bases are more susceptible than others, transitions are unexpectedly more frequent than transversions [17,18] and mutation can be greatly affected by factors such as sequence, particularly by the occurrence of CpG islands [19]. Transitions (C → T) are particularly likely at CpG sites because of the frequent methylation of the C residues and the subsequent deamination of 5-methylcytosine to give a T [20]. If mutations occur randomly, the frequency with which they occur should be described by the Poisson distribution. Several factors might affect the distribution. If mutations tend to occur at particular positions their distribution would not be random. For example, a natural polymorphism might result in what appears to be a concentration of mutational change at a particular position [21]. In addition, mutagenesis studies often disclose sections of the DNA sequence that are prone to mutation (e.g. [22]). The reason for such “hot-spots” is often not clear but does relate to the structure of the DNA. The existence of hot-spots might be expected to confound an analysis for random mutation. In addition, a mutation conferring a selective advantage would be enriched in the population. I assume that selection occurs at the level of the protein product and so the analysis
is carried out at the codon rather than the nucleotide level. In this paper, the distribution of silent mutations in the TP53 gene is analyzed to answer the question: are the silent mutations found in tumors non-selective? It is assumed that if the mutations can be shown to occur randomly this amounts to a demonstration that they are not selected, since selection of particular codons would skew the distribution and make it non-random. We conclude that the majority of silent mutations are non-selected but the analysis is made complicated by the existence of hot-spots for both total and for silent mutation. Silent mutations not found in hot-spots are distributed randomly. Analysis of the silent mutations in hot-spots shows that most of these include the theoretical percentage to be expected if the mutations were not selected. The majority of silent mutations in TP53 are therefore neutral and not selected and their frequency is an indication of the amount of mutation that occurred during the development of a particular tumor. The finding that different tumor types have characteristic frequencies of silent mutation indicates their different developmental history.
2. Materials and methods Sequence data on the p53 protein were obtained using Genbank (locus DNHU53) and were manipulated using the MacVector 6.5.3 program (Oxford Molecular Ltd, Madison, WI). The data base of TP53 mutations compiled by the International Agency for Research on Cancer (IARC) provided the material for this study. The data presented are based on the April 1999 release. This data base tabulates the mutations in the TP53 tumor suppressor gene recorded in peer-reviewed journals [23]. The information has already been much used, e.g. to analyze the possible role of mutagenic agents in different tumors and for studies of the positions at which frameshifts are likely to occur [24–26]. Some 10,396 mutations (10,036 in codons) including 426 silent mutations (4.1%) are recorded in the 1999 data base. An update of this data base (July 2000) has just been made available but is not used in this analysis. The new data base includes 689 silent mutations out of a total of 13,952 (4.9% silent mutations) [27] continuing the upward trend in reported silent mutations previously commented upon [10]. The
B.S. Strauss / Mutation Research 457 (2000) 93–104
reason for this upward trend is not clear. An additional problem with the database is that in only a minority of the investigations has the whole gene been sequenced. A survey of the literature in the 1997 version of the data base indicated that only 18% of 550 publications reported sequencing the entire gene [11]. Most of the papers report only on exons five through eight. The data were therefore analyzed exon by exon. The data base was sorted into exons using the Microsoft “Excel” program and the total number of mutations and the number of silent mutations in each codon was tabulated. “Hot-spots” for silent mutation were identified by determining codons for which six or more silent mutations had been recorded. Since there are 393 codons in the p53 protein, the average number of silent mutations per codon is 426/393 = 1.08. The probability of finding any number of events is given by the Poisson distribution as: P (i) = mi e−m /i!, where m is the mean number of events and i the number of occurrences. The probability of finding six events (i) with an average (m) of 1.08 events is 0.00075 and codons with this number of events or more were arbitrarily classified as hot-spots for silent mutation.
3. Results and discussion Given the codon usage in p53, random mutation of single bases should result in 23.5% of mutations being silent (830 possible silent out of 3537 possible mutations, see below). Overall there are 4.1% silent mutations recorded. This indicates that the overall distribution of mutations is not random, a result to be expected if most of the mutations (silent plus non-silent) recorded in the TP53 gene play a selective role. The p53 data base was examined codon by codon for total and silent mutations (Fig. 1). The average number of mutations per codon is 26.5 and the average number of silent mutations per codon is 1.1. Some codons appear to be “hot-spots” for total mutation with over 700 occurrences (Table 1). Analysis of the silent mutations in the April 1999 version of the data base also discloses codons in which a high frequency of silent mutations has been recorded. The number of mutations and the number of silent mutations for those codons with over 100 total mutations or with six or more silent mutations is listed in Table 1.
95
4. Distribution of silent mutations If silent mutations are non-selective and random, the occurrence of such mutations should be described by the Poisson distribution. The average number of silent mutations per codon should be predictable from the frequency of codons with no mutations recorded since f (0) = e−m . In general, investigators sequence only codons five through eight but when other exons are sequenced, the whole exon is sequenced. The calculations were therefore done exon by exon. The data base was scanned and the frequency within each exon of codons with no silent mutations was recorded (Table 2). This value was then used to calculate an average number of silent mutations per codon and the calculated number was compared with that observed. The deviations between calculated and observed in this comparison are greater than would be expected by chance. A Chi-square of 88.4 is obtained from the data. Since there are eight classes (exons) with silent mutations (Table 2), there are 7 degrees of freedom. The P value is therefore 0.001 indicating that the silent mutations are not randomly distributed. Particular regions of a gene are more likely to harbor mutations and such regions are considered “hot-spots” [28]. The reasons for the occurrence of such sites are diverse and range from sequence effects on the mutation process itself to the selective effects of a particular change [29–31]. CpG sites are particularly mutation prone and transitions at such sites account for 35% of all p53 mutations (Holmquist, personal communication and [8]). The silent hot-spot codons are particularly rich in C + G (Table 3) with a G + C content of 82% compared to a G + C content of 57% for p53 cDNA. In contrast, the distribution of bases in the codons in which any silent mutations occur (minus the hot-spots) is similar to the p53 cDNA base composition (Table 3). Since “hot-spots” are by definition regions with a greater than expected number of mutations they are not randomly distributed. For example, a rare polymorphism might be observed as a hot-spot in a survey of a random population. A corrected average number of mutations was therefore calculated by subtracting the silent mutations in “hot-spots” from the total (Table 2). Neither tryptophan (W) or methionine (M) have synonymous codons which can yield silent changes so these were removed from the codon count
96
B.S. Strauss / Mutation Research 457 (2000) 93–104
Fig. 1. (A) Total number of TP53 mutations at different codons recorded in the IARC data base. (B) Total number of silent mutations at different codons recorded in the IARC data base.
B.S. Strauss / Mutation Research 457 (2000) 93–104
97
Table 1 Hot-spots in TP53 for total and silent mutationsa Hot-spots for total mutation Codon
AA
Exon Silent
Total
157 158 163 173 175 176 179
Val Arg Tyr Val Arg Cys His
GTC CGC TAC GTG CGC TGC CAT
3 2 2 12 3 0 4
132 141 101 114 467 194 164
196
Arg
CGA
4
111
213∗
Arg
CGA
12
165
237 241
Met Ser
ATG TCC
0 2
103 103
245
Gly
GGC
11
344
248 249
Arg Arg
CGG AGG
6 4
707 316
273 278 280 282
Arg Pro Arg Arg
CGT CCT AGA CGG
2 0 2 12
708 127 126 284
5 5 5 5 5 5 5 5 5 5 5 5 6 6 6 7 7 7 7 7 7 7 7 7 8 8 8 8 8 8 8
Hot-spots for silent mutation Codon
AA
Silent
Total
137 140 152 153 154
Leu Thr Pro Pro Gly
CTG ACC CCG CCC GGC
6 6 7 7 9
19 20 72 22 52
173
Val
GTG
12
114
211 213∗
Thr Arg
ACT CGA
8 12
26 165
226
Gly
GGC
9
11
244 245 247 248
Gly Gly Asn Arg
GGC GGC AAC CGG
7 11 7 6
95 344 33 707
250
Pro
CCC
6
64
266
Gly
GGA
6
94
282 293 299
Arg Gly Leu
CGG GGG CTG
12 7 7
284 29 12
a The IARC p53 data base was sorted by codon and then by the initial (before mutation) and final (after mutation) amino acid. The table lists codons with 100 or more total mutations (left) or with six or more silent mutations (right). The asterik marks the site of a known polymorphism in the TP53 gene.
(Table 2 column (c) and (g)). Chi-square values for the distribution were calculated from the observed numbers of silent mutations in each exon and the expected numbers as calculated from the P(0) term of the Poisson including hot-spots. A Chi-square value of 9.83 (7 d.f., P ∼ 0.2) is obtained after subtraction of hot-spots. This latter value fits the hypothesis that the 281 silent mutations not in hot-spots out of the total of 426 (66%), appear randomly in the data base. We suppose that mutations that occur randomly are not selected and therefore are neutral.
What of the 426 − 281 = 145 silent mutations located in the hot-spots for silent mutation? Of these, 53 are located in codons which are hot-spots for total mutation (over 100 mutations per codon) for a total of 1614 mutations or 3.28% silent mutations. The remaining 92 silent mutations were in silent-only hot-spot codons with a reported total of 549 mutations of all types for a silent mutation frequency of 16.8%. If mutations were random, given the amino acid distribution in the p53 protein, 23.5% of all mutations (129 of the 549) would be silent. Among the
98
B.S. Strauss / Mutation Research 457 (2000) 93–104
B.S. Strauss / Mutation Research 457 (2000) 93–104 Table 3 Base composition of hot-spots for silent mutationa Percent
p53 cDNA Codons with any silent mutation Silent hot-spot codons Codons with any silent mutation minus silent hot-spots
A +T
C +G
43.1 36.9 18.5 39.2
56.9 63.1 81.5 60.8
The A + T and G + C composition of the p53 cDNA, of the codons with any silent mutation, and of the hot-spots was determined by entering these as a sequence into a MacVector file and requesting the program to calculate the mononucleotide frequency. a
silent hot-spots not also hot-spots for total mutation there were eight codons: numbers 137, 140, 153, 211, 226, 247, 293 and 299 in which the percentage of silent mutations was individually greater or equal to the 23.5% expected if mutation had been random (Table 1). There were 57 silent mutations in these codons and because of their occurrence in high frequency we suppose them to be in codons that were not selected. Therefore, in addition to the 281 silent mutations not in hot-spots that the Poisson analysis indicates to be non-selected, there are an additional 57 in hot-spots that are also not selected. This means that at least 281 + 57 = 338 silent mutations out of a total of 426 in the data base, or 338/426 × 100 = 79.3 percent are not selected. The existence of hot-spots for missense mutation and the very different percentages of silent mutations in the hot-spots for total mutation (3.3%) compared to the hot-spots for silent mutation only (16.8% silent mutation in these codons) is compatible with the general assumption that certain amino acid positions in the p53 protein are more critical for tumorigenesis than others [30]. Just as the existence of hot-spots for missense mutation with a low frequency of silent mutations indicates codons which are selected and presumably important for carcinogenesis, the hot-spots for silent mutations which include the theoretical percentage of silent mutations expected if mutation is random, indicate codons whose mutation is not critical for carcinogenesis. If this is correct, then not only the silent mutations but all the mutations in these codons are not selected. Some amino acid substitutions reported in p53 must therefore have no physiological role in the
99
development of the tumor. On the other hand, germinal p53 mutations, recognized because they are found in members of cancer prone families, are presumed to play a selective role in the development of tumors. As might be expected, none of the 122 germinal p53 mutation entries in the IARC data base (updated to 3/98) are silent. 5. Analysis of codons with no silent mutations What should be the composition of codons with no silent mutations if silent mutation is random? Given the amino acid composition of the protein, what should be the distribution of these codons, i.e. how many leucines, how many methionines, etc.? The question is similar to one considered by students of molecular evolution [7,33] but the following approximation is based on the actual usage of codons in the p53 cDNA. The different amino acids differ with respect to the number of silent mutations possible. There are nine possible single base substitutions in any codon. For two codons (methionine and tryptophan) there are no synonyms. Such codons are non-degenerate [32]. Nine codons, have two synonyms, for one codon there are three synonyms, for five codons there are four synonyms and for three codons there are six synonyms. The number of silent mutations expected for amino acids with six synonyms depends on the codon. For example, four out of nine leucine CTA mutations will be silent but only two TTG single base mutations are silent. The calculations in Table 4 are based on the actual codon usage in the interval studied, exons five through nine. The calculation of the expected percentage of silent mutations in p53 is based on the actual codon usage in the complete cDNA. If there were equal numbers of amino acids each with an equal number of mutations, then we would expect more tryptophan and methionine codons in the non-silent category because these have no synonyms. The same reasoning suggests that there would be fewer leucine and serine codons (six synonyms) without silent mutations as compared to cysteine and aspartic acid (two synonyms) because the leucine and serine codons have a greater chance for silent mutation than the cysteine and aspartic acid. Since every codon has nine different single base substitutions, the relative probability with which a particular codon will not have a silent mutation is proportional to the number
100
B.S. Strauss / Mutation Research 457 (2000) 93–104
B.S. Strauss / Mutation Research 457 (2000) 93–104 Table 5 Comparison of the found and expected distribution of codons with no silent mutation in exons five through ninea Codons per amino acid
Number of codons found with no silent mutation
Number expected = 59∗ (i)
(O−E)2 /E
2 3 4 6 Chi-square = 4.429
35 1 13 10 P = 0.2–0.3
27.02 1.829 16.933 13.216
2.356 0.376 0.914 0.783
a The expected number of codons of each type with no silent mutations was calculated by multiplying the fraction of non-silent mutations (column i, Table 4) by the observed total number of codons with no silent mutations. This number was compared with the number found and a Chi-square value was calculated.
of non-silent mutations possible, i.e. nine out of nine tryptophan mutations are non-silent, so the probability of a tryptophan mutation being non-silent is 1. For leucine (CTA), three of nine mutations are silent and therefore the probability of a non-silent mutation is 6/9. If the amino acids were present in equal amounts, the relative probability of a mutation in a tryptophan codon being non-silent as compared to a (CTA) leucine would be (9/9)/(6/9) or 1.5 times higher. But the amino acids are not present in equal amounts. Even though no tryptophan mutations can be silent, if there were one tryptophan and 100 leucines and few mutations, the chances are that any particular codon with no silent mutation recorded would be a leucine. In calculating the relative chance of finding a particular codon not having a silent mutation, we need to consider both the probability of not producing a silent mutation in any codon and the relative number of codons. For example, the expected number of two synonym (two-fold degenerate) codons without a silent mutation is the probability of that particular type of codon not giving a silent mutation multiplied by the number of such codons in the sample. For this analysis, all codons with no silent, one silent, two silent, three silent and five silent mutations were pooled. Exons five to nine were scanned to determine the codons in which silent mutations were not observed (Tables 4 and 5). The remaining exons have so few silent mutations recorded that an analysis seemed unlikely to be significant. There are 206 codons with 9,653 recorded mutations in p53 exons five through nine, and 59 of these codons excluding tryptophan
101
and methionine do not have silent mutations. As in the analysis above (Table 2), the seven tryptophan and methionine codons were excluded from the analysis since there is no chance for them to give a silent mutation. These null sites were classified as to type of codon (2, 3, 4 or 6 synonyms) and the expected fraction of non-silent mutations in each type was calculated (Table 4). This fraction was used to calculate the number of codons of each type expected to have no silent mutations among the 59 codons with no silent mutation found. This expected number was then compared to that found and a Chi-square value was calculated. The value (Chi-square = 4.43 with 3 degrees of freedom given the four types of codon) has a probability of 0.2–0.3 and is therefore not significantly different from what might be expected by chance, as expected if the silent mutations are distributed randomly.
6. Silent mutations in different tumor types This analysis indicates that at least 80% of all silent mutations are randomly distributed and hence unlikely to have been subject to selection. The frequency of silent mutations can therefore be used as a selection-free indication of the mutational processes that occur in TP53 during the genesis of tumors. Ideally, we would like to know the frequency of TP53 silent mutations in an unselected sample of tumors, some of which will have no mutations in p53. Unfortunately, the data base is so constructed that only tumors with at least one mutation in p53 are entered. Since the data on the total number of tumors analyzed is not available, the percentage of silent mutations among all mutations has been used based on the following argument. A “driver” mutation is required to produce a tumor. This mutation may be in p53 or in some other gene which is part of the p53 cascade. The silent mutations, as shown above are for the most part hitch-hikers and are not selected. Most (∼67%) of the silent mutations reported are the only mutations found in the p53 gene in the tumor analyzed [10] and it is assumed must be accompanied by some driver mutation elsewhere. Insofar as the “hitchhiker” mutation does not occur in the same event as the “driver”, the frequency of silent mutations in the data base is an estimate of (or at least proportional to) the actual frequency in all tumors.
102
B.S. Strauss / Mutation Research 457 (2000) 93–104
Table 6 TP53 silent mutation in tumors of various typesa Total mutations (b)
Bladder Bone & connective Brain Breast Bronchus & lung Colon Esophagus Female generative Head & neck Hematopoietic Liver & kidney Prostate & male
498 352 591 1020 1190 1128 624 792 772 803 583 234
30 26 18 47 63 17 11 19 31 24 13 31
6 7.4 3 4.6 5.3 1.5 1.8 2.4 4 3 2.2 13.2
20.4 14.4 24.2 41.7 48.7 46.1 25.5 32.4 31.6 32.8 23.8 9.6
4.6 9.4 1.6 0.7 4.2 18.4 8.3 5.5 0.0 2.4 4.9 48.0
231 458 481 76
8 29 28 7
3.5 6.3 5.8 9
9.4 18.7 19.7 3.1
0.2 5.6 3.5 4.9
9833
402
4.08
Generative Pancreas Skin Stomach Thyroid Total in tumors listed
Total silent (c)
% Silent (d)
Expected silent (e)
(O−E)2 /E (f)
Tumor source (a)
402
P (g) 0.05 0.01
0.05 0.001 0.01 0.05
0.05 0.001
0.05 0.05
122.2
a
The IARC database was scanned by morphological site (a) and total mutations (b) and total silent mutations (c) were tabulated. The percent silent mutations were calculated both for each site and overall (d). From the overall percentage of silent mutations and the total mutations recorded, the expected silent mutations were calculated (e) = (b) × 0.0408. Chi-square for the overall distribution was calculated by summation of the individual values (f ) = [(c) − (e)]2 /(e). The individual values of (f) are equivalent to the calculation of Chi-square for deviation from the overall average. The values in column (g) indicate probabilities of less than the value indicated.
Frequency is the resultant of many factors and although a difference in frequency may reflect difference in mutation rate it may also reflect difference in the number of cell generations in which the mutation could occur. An observation that tumor types differ in the frequency with which mutations are found therefore tells little more than that the biology of the tumors is different. The frequency of silent mutations does appear to differ in tumors of different types (Table 6). Data from tumors from 16 sites total 9833 mutations including 402 silent mutations for an average of 4.08% silent mutations. The tumors can be classified into groups based on comparison with the average. Tumors of the colon, esophagus, female generative tissue and liver and kidney have percentages of silent mutations statistically lower than the overall average. Tumors of the prostate and testis and less certainly of bone and connective tissue have very high percentages of silent mutation. The special nature of mutations in the prostate has been pointed out earlier [33]. There is additional evidence indicating the unique nature of the development of this tumor. Mul-
tiple mutations are reported in many tumors and 14 out of 24 of the silent mutations reported in prostate tumors occur in tumors with multiple mutations [23]. There are 66 total mutations in the 29 prostate tumors with multiple mutations giving an uncorrected 21.2% of silent mutations. Adjusting for ascertainment by supposing that a first mutation is selected to produce the tumor and the remainder are superfluous gives a corrected 38% silent mutations in the TP53 gene in prostate tumors. The analysis of the p53 data base presented in this paper indicates that the majority of silent mutations in tumors are selectively neutral and may be used to measure mutability. Such mutations observed in tumor populations do not play an etiologic role but may be epi-phenomena of the process of carcinogenesis. It is a reasonable speculation that other non-silent, but conservative, mutations may also be neutral and not play a role in the development or progression of tumors. Finally, the difference in the proportion of silent mutations seen in different tumors may serve as a sign of their differing evolutionary pathways.
B.S. Strauss / Mutation Research 457 (2000) 93–104
Acknowledgements This work was supported in part by a grant from the National Cancer Institute, NIH (CA32436). I wish to acknowledge the advice of the University of Chicago Cancer Center Biostatistics Core Support Grant. I would also like to thank Dr. Gerald Holmquist, Dr. Larry Loeb, and Dr. Sergei Rodin for their critical comments on an earlier version of the mansucript. Correspondence about this paper may be addressed to the author at the email address given in the beginning.
References [1] L.A. Loeb, Cancer cells exhibit a mutator phenotype, Adv. Cancer Res. 72 (1998) 25–56. [2] L. Loeb, C. Springgate, N. Battula, Errors in DNA replication as a basis of malignant change, Cancer Res. 34 (1974) 2311– 2321. [3] A.B. Buermeyer, S.M. Deschenes, S.M. Baker, R.M. Liskay, Mammalian DNA mismatch repair, Annu. Rev. Genet. 33 (1999) 533–564. [4] E.C. Friedberg, J.P. Bond, D.K. Burns, D.L. Cheo, M.S. Greenblatt, L.B. Meira, D. Nahari, A.M. Reis, Defective nucleotide excision repair in xpc mutant mice and its association with cancer predisposition, Mutat. Res. 459 (2000) 99–108. [5] I.P. Tomlinson, M.R. Novelli, W.F. Bodmer, The mutation rate and cancer, Proc. Natl. Acad. Sci. U.S.A. 93 (1996) 14800–14803. [6] I. Tomlinson, W. Bodmer, Selection, the mutation rate and cancer: ensuring that the tail does not wag the dog, Nature Med. 5 (1999) 11–12. [7] W.-H. Li, Molecular Evolution, Sinauer Assiciates, Sunderland, MA, 1997, p. xv, 487. [8] S.N. Rodin, G.P. Holmquist, A.S. Rodin, CpG transition strand asymmetry and hitch-hiking mutations as measures of tumorigenic selection in shaping the p53 mutation spectrum, Int. J. Mol. Med. 1 (1998) 191–199. [9] S.N. Rodin, A.S. Rodin, Strand asymmetry of CpG transitions as indicator of G1 phase-dependent origin of multiple tumorigenic p53 mutations in stem cells, Proc. Natl. Acad. Sci. U.S.A. 95 (1998) 11927–11932. [10] B.S. Strauss, Hypermutability and silent mutations in human carcinogenesis, Semin Cancer Biol. 8 (1998) 431–438. [11] B.S. Strauss, Silent and multiple mutations in p53 and the question of the hypermutability of tumors, Carcinogenesis 18 (1997) 1445–1452. [12] B.S. Strauss, Hypermutability in carcinogenesis, Genetics 148 (1998) 1619–1626. [13] P.M. Sharp, M. Averof, A.T. Lloyd, G. Matassi, J.F. Peden, DNA sequence evolution: the sounds of silence, Phil. Trans. Roy. Soc. Lond. B. 349 (1995) 241–247.
103
[14] J.R. Powell, E.N. Moriyama, Evolution of codon usage bias in Drosophila, Proc. Natl. Acad. Sci. U.S.A. 94 (1997) 7784– 7790. [15] T. Hongyo, G.S. Buzard, D. Palli, C.M. Weghorst, A. Amorosi, M. Galli, N.E. Caporaso, J.F. Fraumeni Jr., J.M. Rice, Mutations of the K-ras and p53 genes in gastric adenocarcinomas from a high-incidence region around Florence, Italy, Cancer Res. 55 (1995) 2665–2672. [16] I. Richard, J.S. Beckmann, How neutral are synonymous codon mutations? Nature Genet. 10 (1995) 259. [17] D.W. Collins, T.H. Jukes, Rates of transition and transversion in coding sequences since the human-rodent divergence, Genomics 20 (1994) 386–396. [18] N.K. Sinha, M.D. Haimes, Molecular mechanisms of substitution mutagenesis: an experimental test of the Watson–Crick and Topal–Fresco models of base mispairings, J. Biol. Chem. 256 (1981) 10671–10683. [19] G.P. Pfeifer, M. Tang, M.F. Denissenko, Mutation hotspots and DNA methylation, Curr. Top. Microbiol. Immunol. 249 (2000) 1–19. [20] G.P. Pfeifer, p53 mutational spectra and the role of methylated CpG sequences, Mutat. Res. 450 (2000) 155–166. [21] M. Galliano, U. Kragh-Hansen, A.L. Tarnoky, J.C. Chapman, M. Campagnoli, L. Minchiotti, Genetic variants showing apparent hotspots in the human serum albumin gene, Clin. Chim. Acta 289 (1999) 45–55. [22] C. Milstein, M.S. Neuberger, R. Staden, Both DNA strands of antibody genes are hypermutation targets, Proc. Natl. Acad. Sci. U.S.A. 95 (1998) 8791–8794. [23] P. Hainaut, T. Hernandez, A. Robinson, P. Rodriguez-Tome, T. Flores, M. Hollstein, C.C. Harris, R. Montesano, IARC database of p53 gene mutations in human tumors and cell lines: updated compilation, revised formats and new visualisation tools, Nucleic Acids Res. 26 (1998) 205–213. [24] S.P. Hussain, C.C. Harris, p53 mutation spectrum and load: the generation of hypotheses linking the exposure of endogenous or exogenous carcinogens to human cancer, Mutat. Res. 428 (1999) 23–32. [25] M. Hollstein, D. Sidransky, B. Vogelstein, C.C. Harris, p53 mutations in human cancers, Science 253 (1991) 49– 53. [26] M.S. Greenblatt, A.P. Grollman, C.C. Harris, Deletions and insertions in the p53 tumor suppressor gene in human cancers: confirmation of the DNA polymerase slippage/misalignment model, Cancer Res. 56 (1996) 2130–2136. [27] T. Hernandez-Boussard, P. Rodriguez-Tome, R. Montesano, P. Hainaut, p53 mutation database: a relational database to compile and analyze p53 mutations in human tumors and cell lines, International Agency for Research on Cancer. Hum. Mutat. 14 (1999) 1–18. [28] S. Benzer, On the topography of the genetic fine structure, Proc. Natl. Acad. Sci. U.S.A. 87 (1961) 403–415. [29] Z. Hatahet, M. Zhou, L.J. Reha-Krantz, S.W. Morrical, S.S. Wallace, In search of a mutational hotspot, Proc. Natl. Acad. Sci. U.S.A. 95 (1998) 8556–8561. [30] D.R. Walker, J.P. Bond, R.E. Tarone, C.C. Harris, W. Makalowski, M.S. Boguski, M.S. Greenblatt, Evolutionary
104
B.S. Strauss / Mutation Research 457 (2000) 93–104
conservation and somatic mutation hot spot maps of p53: correlation with p53 protein structural and functional features, Oncogene 18 (1999) 211–218. [31] J.C. Delaney, J.M. Essigmann, Context-dependent mutagenesis by DNA lesions, Chem. Biol. 6 (1999) 743–753. [32] W.H. Li, C.I. Wu, C.C. Luo, A new method for estimating
synonymous and non-synonymous rates of nucleotide substitution considering the relative likelihood of nucleotide and codon changes, Mol. Biol. E 2 (1985) 150–174. [33] S.G. Chi, R.W. deVere White, F.J. Meyers, D.B. Siders, F. Lee, P.H. Gumerlock, p53 in prostate cancer: frequent expressed transition mutations, J. Natl. Cancer Inst. 86 (1994) 926–933.
Role in tumorigenesis of silent mutations in the TP53 gene Bernard S. Strauss∗ Department of Molecular Genetics and Cell Biology, The University of Chicago, 920 East 58th Street, Chicago, IL 60637, USA Received 15 July 2000; received in revised form 8 September 2000; accepted 18 September 2000
Abstract Over 10,000 mutations in the TP53 suppressor gene have been recorded in the International Agency for Research on Cancer (IARC) tumor data base. About 4% of these mutations are silent. It is a question whether these mutations play a role in tumor development. In order to approach this question, we asked whether the reported silent mutations are randomly distributed throughout the TP53 gene. The p53 data base was searched exon by exon. From the frequency of codons with no silent mutations, the average number of silent mutations per codon for each exon was calculated using the Poisson distribution. The results indicate the distribution to be non-random. About one-third of all silent mutations occur in “hot-spots” and after subtraction of these hot-spots, the remaining silent mutations are randomly distributed. In addition, the percentage of silent mutations among the total in the silent mutation hot-spots is close to that expected for random mutation. We conclude that most of the silent mutations recorded in tumors play no role in tumor development and that the percentage of silent mutation is an indication of the amount of random mutation during tumorigenesis. Silent mutations occur to a significantly different extent in different tumor types. Tumors of the esophagus and colon have a low frequency of silent mutations, tumors of the prostate have a high frequency. © 2000 Elsevier Science B.V. All rights reserved. Keywords: TP53; Silent mutation; Hyper-mutability; Selection; Genetic instability
1. Introduction Tumors often include new mutations not present in the surrounding tissue [1]. Two explanations have been proposed to account for this finding. The first supposes that tumors are genetically unstable and produce orders of magnitude more point mutations than normal tissue [1,2]. The finding that organisms (including humans) with deficiencies in DNA repair pathways are both hyper-mutable and tumor-prone provides strong support for this view [3,4]. A second explanation supposes that normal mutability accounts for the production of mutations which accumulate in the tumor tissue because they are selected in successive waves ∗ Tel.: +1-773-702-1628; fax: +1-773-702-3172. E-mail address: [email protected] (B.S. Strauss).
of clonal development [5,6]. These hypotheses are not mutually exclusive. Tumors might be hyper-mutable at some stage but selection might still determine which mutations are found. Since the primary observation is the finding that tumors include many new mutations, the demonstration of hyper-mutability requires a marker which does not provide a selective advantage. TP53 silent mutations, in which a nucleotide has been altered to produce a synonymous codon without amino acid change [7], may provide such a marker [8–12]. However, it must first be demonstrated that such mutations are truly non-selective since there is evidence that not all synonymous codons are utilized to an equal extent. Li [7] gives as an example the finding that for a particular E. coli protein, 21 of 23 leucine residues are encoded by CUG although there are five other codons available. Synonymous codons are used at very differ-
0027-5107/00/$ – see front matter © 2000 Elsevier Science B.V. All rights reserved. PII: S 0 0 2 7 - 5 1 0 7 ( 0 0 ) 0 0 1 3 5 - 4
94
B.S. Strauss / Mutation Research 457 (2000) 93–104
ent frequencies and the different p53 codons are not used equally [13]. One explanation is that the abundance of the different tRNA species differs and that the codon corresponding to the most abundant tRNA might have an advantage over one specifying a rare species [7,14]. A mutation from a codon using a rare tRNA to one using a more abundant species would have a selective advantage although silent. Alternatively, a silent mutation at an exon/intron border which changes a splice site would be expected to result in a major change in the protein gene product [15]. Such a change has been observed in a mutation leading to human muscular dystrophy [16] and results in a changed pattern of transcription. Therefore, before silent mutations in TP53 can be used as an indicator of mutability it is necessary to show that they are really neutral and do not confer some selective advantage to the tumor. Mutation is assumed to be a random process in a hypothetical DNA of uniform structure. This hypothesis supposes that each base in a sequence be equally susceptible to change and mutates to each of the other three with equal frequency. In practice, this ideal situation is rarely, if ever, observed. Particular bases are more susceptible than others, transitions are unexpectedly more frequent than transversions [17,18] and mutation can be greatly affected by factors such as sequence, particularly by the occurrence of CpG islands [19]. Transitions (C → T) are particularly likely at CpG sites because of the frequent methylation of the C residues and the subsequent deamination of 5-methylcytosine to give a T [20]. If mutations occur randomly, the frequency with which they occur should be described by the Poisson distribution. Several factors might affect the distribution. If mutations tend to occur at particular positions their distribution would not be random. For example, a natural polymorphism might result in what appears to be a concentration of mutational change at a particular position [21]. In addition, mutagenesis studies often disclose sections of the DNA sequence that are prone to mutation (e.g. [22]). The reason for such “hot-spots” is often not clear but does relate to the structure of the DNA. The existence of hot-spots might be expected to confound an analysis for random mutation. In addition, a mutation conferring a selective advantage would be enriched in the population. I assume that selection occurs at the level of the protein product and so the analysis
is carried out at the codon rather than the nucleotide level. In this paper, the distribution of silent mutations in the TP53 gene is analyzed to answer the question: are the silent mutations found in tumors non-selective? It is assumed that if the mutations can be shown to occur randomly this amounts to a demonstration that they are not selected, since selection of particular codons would skew the distribution and make it non-random. We conclude that the majority of silent mutations are non-selected but the analysis is made complicated by the existence of hot-spots for both total and for silent mutation. Silent mutations not found in hot-spots are distributed randomly. Analysis of the silent mutations in hot-spots shows that most of these include the theoretical percentage to be expected if the mutations were not selected. The majority of silent mutations in TP53 are therefore neutral and not selected and their frequency is an indication of the amount of mutation that occurred during the development of a particular tumor. The finding that different tumor types have characteristic frequencies of silent mutation indicates their different developmental history.
2. Materials and methods Sequence data on the p53 protein were obtained using Genbank (locus DNHU53) and were manipulated using the MacVector 6.5.3 program (Oxford Molecular Ltd, Madison, WI). The data base of TP53 mutations compiled by the International Agency for Research on Cancer (IARC) provided the material for this study. The data presented are based on the April 1999 release. This data base tabulates the mutations in the TP53 tumor suppressor gene recorded in peer-reviewed journals [23]. The information has already been much used, e.g. to analyze the possible role of mutagenic agents in different tumors and for studies of the positions at which frameshifts are likely to occur [24–26]. Some 10,396 mutations (10,036 in codons) including 426 silent mutations (4.1%) are recorded in the 1999 data base. An update of this data base (July 2000) has just been made available but is not used in this analysis. The new data base includes 689 silent mutations out of a total of 13,952 (4.9% silent mutations) [27] continuing the upward trend in reported silent mutations previously commented upon [10]. The
B.S. Strauss / Mutation Research 457 (2000) 93–104
reason for this upward trend is not clear. An additional problem with the database is that in only a minority of the investigations has the whole gene been sequenced. A survey of the literature in the 1997 version of the data base indicated that only 18% of 550 publications reported sequencing the entire gene [11]. Most of the papers report only on exons five through eight. The data were therefore analyzed exon by exon. The data base was sorted into exons using the Microsoft “Excel” program and the total number of mutations and the number of silent mutations in each codon was tabulated. “Hot-spots” for silent mutation were identified by determining codons for which six or more silent mutations had been recorded. Since there are 393 codons in the p53 protein, the average number of silent mutations per codon is 426/393 = 1.08. The probability of finding any number of events is given by the Poisson distribution as: P (i) = mi e−m /i!, where m is the mean number of events and i the number of occurrences. The probability of finding six events (i) with an average (m) of 1.08 events is 0.00075 and codons with this number of events or more were arbitrarily classified as hot-spots for silent mutation.
3. Results and discussion Given the codon usage in p53, random mutation of single bases should result in 23.5% of mutations being silent (830 possible silent out of 3537 possible mutations, see below). Overall there are 4.1% silent mutations recorded. This indicates that the overall distribution of mutations is not random, a result to be expected if most of the mutations (silent plus non-silent) recorded in the TP53 gene play a selective role. The p53 data base was examined codon by codon for total and silent mutations (Fig. 1). The average number of mutations per codon is 26.5 and the average number of silent mutations per codon is 1.1. Some codons appear to be “hot-spots” for total mutation with over 700 occurrences (Table 1). Analysis of the silent mutations in the April 1999 version of the data base also discloses codons in which a high frequency of silent mutations has been recorded. The number of mutations and the number of silent mutations for those codons with over 100 total mutations or with six or more silent mutations is listed in Table 1.
95
4. Distribution of silent mutations If silent mutations are non-selective and random, the occurrence of such mutations should be described by the Poisson distribution. The average number of silent mutations per codon should be predictable from the frequency of codons with no mutations recorded since f (0) = e−m . In general, investigators sequence only codons five through eight but when other exons are sequenced, the whole exon is sequenced. The calculations were therefore done exon by exon. The data base was scanned and the frequency within each exon of codons with no silent mutations was recorded (Table 2). This value was then used to calculate an average number of silent mutations per codon and the calculated number was compared with that observed. The deviations between calculated and observed in this comparison are greater than would be expected by chance. A Chi-square of 88.4 is obtained from the data. Since there are eight classes (exons) with silent mutations (Table 2), there are 7 degrees of freedom. The P value is therefore 0.001 indicating that the silent mutations are not randomly distributed. Particular regions of a gene are more likely to harbor mutations and such regions are considered “hot-spots” [28]. The reasons for the occurrence of such sites are diverse and range from sequence effects on the mutation process itself to the selective effects of a particular change [29–31]. CpG sites are particularly mutation prone and transitions at such sites account for 35% of all p53 mutations (Holmquist, personal communication and [8]). The silent hot-spot codons are particularly rich in C + G (Table 3) with a G + C content of 82% compared to a G + C content of 57% for p53 cDNA. In contrast, the distribution of bases in the codons in which any silent mutations occur (minus the hot-spots) is similar to the p53 cDNA base composition (Table 3). Since “hot-spots” are by definition regions with a greater than expected number of mutations they are not randomly distributed. For example, a rare polymorphism might be observed as a hot-spot in a survey of a random population. A corrected average number of mutations was therefore calculated by subtracting the silent mutations in “hot-spots” from the total (Table 2). Neither tryptophan (W) or methionine (M) have synonymous codons which can yield silent changes so these were removed from the codon count
96
B.S. Strauss / Mutation Research 457 (2000) 93–104
Fig. 1. (A) Total number of TP53 mutations at different codons recorded in the IARC data base. (B) Total number of silent mutations at different codons recorded in the IARC data base.
B.S. Strauss / Mutation Research 457 (2000) 93–104
97
Table 1 Hot-spots in TP53 for total and silent mutationsa Hot-spots for total mutation Codon
AA
Exon Silent
Total
157 158 163 173 175 176 179
Val Arg Tyr Val Arg Cys His
GTC CGC TAC GTG CGC TGC CAT
3 2 2 12 3 0 4
132 141 101 114 467 194 164
196
Arg
CGA
4
111
213∗
Arg
CGA
12
165
237 241
Met Ser
ATG TCC
0 2
103 103
245
Gly
GGC
11
344
248 249
Arg Arg
CGG AGG
6 4
707 316
273 278 280 282
Arg Pro Arg Arg
CGT CCT AGA CGG
2 0 2 12
708 127 126 284
5 5 5 5 5 5 5 5 5 5 5 5 6 6 6 7 7 7 7 7 7 7 7 7 8 8 8 8 8 8 8
Hot-spots for silent mutation Codon
AA
Silent
Total
137 140 152 153 154
Leu Thr Pro Pro Gly
CTG ACC CCG CCC GGC
6 6 7 7 9
19 20 72 22 52
173
Val
GTG
12
114
211 213∗
Thr Arg
ACT CGA
8 12
26 165
226
Gly
GGC
9
11
244 245 247 248
Gly Gly Asn Arg
GGC GGC AAC CGG
7 11 7 6
95 344 33 707
250
Pro
CCC
6
64
266
Gly
GGA
6
94
282 293 299
Arg Gly Leu
CGG GGG CTG
12 7 7
284 29 12
a The IARC p53 data base was sorted by codon and then by the initial (before mutation) and final (after mutation) amino acid. The table lists codons with 100 or more total mutations (left) or with six or more silent mutations (right). The asterik marks the site of a known polymorphism in the TP53 gene.
(Table 2 column (c) and (g)). Chi-square values for the distribution were calculated from the observed numbers of silent mutations in each exon and the expected numbers as calculated from the P(0) term of the Poisson including hot-spots. A Chi-square value of 9.83 (7 d.f., P ∼ 0.2) is obtained after subtraction of hot-spots. This latter value fits the hypothesis that the 281 silent mutations not in hot-spots out of the total of 426 (66%), appear randomly in the data base. We suppose that mutations that occur randomly are not selected and therefore are neutral.
What of the 426 − 281 = 145 silent mutations located in the hot-spots for silent mutation? Of these, 53 are located in codons which are hot-spots for total mutation (over 100 mutations per codon) for a total of 1614 mutations or 3.28% silent mutations. The remaining 92 silent mutations were in silent-only hot-spot codons with a reported total of 549 mutations of all types for a silent mutation frequency of 16.8%. If mutations were random, given the amino acid distribution in the p53 protein, 23.5% of all mutations (129 of the 549) would be silent. Among the
98
B.S. Strauss / Mutation Research 457 (2000) 93–104
B.S. Strauss / Mutation Research 457 (2000) 93–104 Table 3 Base composition of hot-spots for silent mutationa Percent
p53 cDNA Codons with any silent mutation Silent hot-spot codons Codons with any silent mutation minus silent hot-spots
A +T
C +G
43.1 36.9 18.5 39.2
56.9 63.1 81.5 60.8
The A + T and G + C composition of the p53 cDNA, of the codons with any silent mutation, and of the hot-spots was determined by entering these as a sequence into a MacVector file and requesting the program to calculate the mononucleotide frequency. a
silent hot-spots not also hot-spots for total mutation there were eight codons: numbers 137, 140, 153, 211, 226, 247, 293 and 299 in which the percentage of silent mutations was individually greater or equal to the 23.5% expected if mutation had been random (Table 1). There were 57 silent mutations in these codons and because of their occurrence in high frequency we suppose them to be in codons that were not selected. Therefore, in addition to the 281 silent mutations not in hot-spots that the Poisson analysis indicates to be non-selected, there are an additional 57 in hot-spots that are also not selected. This means that at least 281 + 57 = 338 silent mutations out of a total of 426 in the data base, or 338/426 × 100 = 79.3 percent are not selected. The existence of hot-spots for missense mutation and the very different percentages of silent mutations in the hot-spots for total mutation (3.3%) compared to the hot-spots for silent mutation only (16.8% silent mutation in these codons) is compatible with the general assumption that certain amino acid positions in the p53 protein are more critical for tumorigenesis than others [30]. Just as the existence of hot-spots for missense mutation with a low frequency of silent mutations indicates codons which are selected and presumably important for carcinogenesis, the hot-spots for silent mutations which include the theoretical percentage of silent mutations expected if mutation is random, indicate codons whose mutation is not critical for carcinogenesis. If this is correct, then not only the silent mutations but all the mutations in these codons are not selected. Some amino acid substitutions reported in p53 must therefore have no physiological role in the
99
development of the tumor. On the other hand, germinal p53 mutations, recognized because they are found in members of cancer prone families, are presumed to play a selective role in the development of tumors. As might be expected, none of the 122 germinal p53 mutation entries in the IARC data base (updated to 3/98) are silent. 5. Analysis of codons with no silent mutations What should be the composition of codons with no silent mutations if silent mutation is random? Given the amino acid composition of the protein, what should be the distribution of these codons, i.e. how many leucines, how many methionines, etc.? The question is similar to one considered by students of molecular evolution [7,33] but the following approximation is based on the actual usage of codons in the p53 cDNA. The different amino acids differ with respect to the number of silent mutations possible. There are nine possible single base substitutions in any codon. For two codons (methionine and tryptophan) there are no synonyms. Such codons are non-degenerate [32]. Nine codons, have two synonyms, for one codon there are three synonyms, for five codons there are four synonyms and for three codons there are six synonyms. The number of silent mutations expected for amino acids with six synonyms depends on the codon. For example, four out of nine leucine CTA mutations will be silent but only two TTG single base mutations are silent. The calculations in Table 4 are based on the actual codon usage in the interval studied, exons five through nine. The calculation of the expected percentage of silent mutations in p53 is based on the actual codon usage in the complete cDNA. If there were equal numbers of amino acids each with an equal number of mutations, then we would expect more tryptophan and methionine codons in the non-silent category because these have no synonyms. The same reasoning suggests that there would be fewer leucine and serine codons (six synonyms) without silent mutations as compared to cysteine and aspartic acid (two synonyms) because the leucine and serine codons have a greater chance for silent mutation than the cysteine and aspartic acid. Since every codon has nine different single base substitutions, the relative probability with which a particular codon will not have a silent mutation is proportional to the number
100
B.S. Strauss / Mutation Research 457 (2000) 93–104
B.S. Strauss / Mutation Research 457 (2000) 93–104 Table 5 Comparison of the found and expected distribution of codons with no silent mutation in exons five through ninea Codons per amino acid
Number of codons found with no silent mutation
Number expected = 59∗ (i)
(O−E)2 /E
2 3 4 6 Chi-square = 4.429
35 1 13 10 P = 0.2–0.3
27.02 1.829 16.933 13.216
2.356 0.376 0.914 0.783
a The expected number of codons of each type with no silent mutations was calculated by multiplying the fraction of non-silent mutations (column i, Table 4) by the observed total number of codons with no silent mutations. This number was compared with the number found and a Chi-square value was calculated.
of non-silent mutations possible, i.e. nine out of nine tryptophan mutations are non-silent, so the probability of a tryptophan mutation being non-silent is 1. For leucine (CTA), three of nine mutations are silent and therefore the probability of a non-silent mutation is 6/9. If the amino acids were present in equal amounts, the relative probability of a mutation in a tryptophan codon being non-silent as compared to a (CTA) leucine would be (9/9)/(6/9) or 1.5 times higher. But the amino acids are not present in equal amounts. Even though no tryptophan mutations can be silent, if there were one tryptophan and 100 leucines and few mutations, the chances are that any particular codon with no silent mutation recorded would be a leucine. In calculating the relative chance of finding a particular codon not having a silent mutation, we need to consider both the probability of not producing a silent mutation in any codon and the relative number of codons. For example, the expected number of two synonym (two-fold degenerate) codons without a silent mutation is the probability of that particular type of codon not giving a silent mutation multiplied by the number of such codons in the sample. For this analysis, all codons with no silent, one silent, two silent, three silent and five silent mutations were pooled. Exons five to nine were scanned to determine the codons in which silent mutations were not observed (Tables 4 and 5). The remaining exons have so few silent mutations recorded that an analysis seemed unlikely to be significant. There are 206 codons with 9,653 recorded mutations in p53 exons five through nine, and 59 of these codons excluding tryptophan
101
and methionine do not have silent mutations. As in the analysis above (Table 2), the seven tryptophan and methionine codons were excluded from the analysis since there is no chance for them to give a silent mutation. These null sites were classified as to type of codon (2, 3, 4 or 6 synonyms) and the expected fraction of non-silent mutations in each type was calculated (Table 4). This fraction was used to calculate the number of codons of each type expected to have no silent mutations among the 59 codons with no silent mutation found. This expected number was then compared to that found and a Chi-square value was calculated. The value (Chi-square = 4.43 with 3 degrees of freedom given the four types of codon) has a probability of 0.2–0.3 and is therefore not significantly different from what might be expected by chance, as expected if the silent mutations are distributed randomly.
6. Silent mutations in different tumor types This analysis indicates that at least 80% of all silent mutations are randomly distributed and hence unlikely to have been subject to selection. The frequency of silent mutations can therefore be used as a selection-free indication of the mutational processes that occur in TP53 during the genesis of tumors. Ideally, we would like to know the frequency of TP53 silent mutations in an unselected sample of tumors, some of which will have no mutations in p53. Unfortunately, the data base is so constructed that only tumors with at least one mutation in p53 are entered. Since the data on the total number of tumors analyzed is not available, the percentage of silent mutations among all mutations has been used based on the following argument. A “driver” mutation is required to produce a tumor. This mutation may be in p53 or in some other gene which is part of the p53 cascade. The silent mutations, as shown above are for the most part hitch-hikers and are not selected. Most (∼67%) of the silent mutations reported are the only mutations found in the p53 gene in the tumor analyzed [10] and it is assumed must be accompanied by some driver mutation elsewhere. Insofar as the “hitchhiker” mutation does not occur in the same event as the “driver”, the frequency of silent mutations in the data base is an estimate of (or at least proportional to) the actual frequency in all tumors.
102
B.S. Strauss / Mutation Research 457 (2000) 93–104
Table 6 TP53 silent mutation in tumors of various typesa Total mutations (b)
Bladder Bone & connective Brain Breast Bronchus & lung Colon Esophagus Female generative Head & neck Hematopoietic Liver & kidney Prostate & male
498 352 591 1020 1190 1128 624 792 772 803 583 234
30 26 18 47 63 17 11 19 31 24 13 31
6 7.4 3 4.6 5.3 1.5 1.8 2.4 4 3 2.2 13.2
20.4 14.4 24.2 41.7 48.7 46.1 25.5 32.4 31.6 32.8 23.8 9.6
4.6 9.4 1.6 0.7 4.2 18.4 8.3 5.5 0.0 2.4 4.9 48.0
231 458 481 76
8 29 28 7
3.5 6.3 5.8 9
9.4 18.7 19.7 3.1
0.2 5.6 3.5 4.9
9833
402
4.08
Generative Pancreas Skin Stomach Thyroid Total in tumors listed
Total silent (c)
% Silent (d)
Expected silent (e)
(O−E)2 /E (f)
Tumor source (a)
402
P (g) 0.05 0.01
0.05 0.001 0.01 0.05
0.05 0.001
0.05 0.05
122.2
a
The IARC database was scanned by morphological site (a) and total mutations (b) and total silent mutations (c) were tabulated. The percent silent mutations were calculated both for each site and overall (d). From the overall percentage of silent mutations and the total mutations recorded, the expected silent mutations were calculated (e) = (b) × 0.0408. Chi-square for the overall distribution was calculated by summation of the individual values (f ) = [(c) − (e)]2 /(e). The individual values of (f) are equivalent to the calculation of Chi-square for deviation from the overall average. The values in column (g) indicate probabilities of less than the value indicated.
Frequency is the resultant of many factors and although a difference in frequency may reflect difference in mutation rate it may also reflect difference in the number of cell generations in which the mutation could occur. An observation that tumor types differ in the frequency with which mutations are found therefore tells little more than that the biology of the tumors is different. The frequency of silent mutations does appear to differ in tumors of different types (Table 6). Data from tumors from 16 sites total 9833 mutations including 402 silent mutations for an average of 4.08% silent mutations. The tumors can be classified into groups based on comparison with the average. Tumors of the colon, esophagus, female generative tissue and liver and kidney have percentages of silent mutations statistically lower than the overall average. Tumors of the prostate and testis and less certainly of bone and connective tissue have very high percentages of silent mutation. The special nature of mutations in the prostate has been pointed out earlier [33]. There is additional evidence indicating the unique nature of the development of this tumor. Mul-
tiple mutations are reported in many tumors and 14 out of 24 of the silent mutations reported in prostate tumors occur in tumors with multiple mutations [23]. There are 66 total mutations in the 29 prostate tumors with multiple mutations giving an uncorrected 21.2% of silent mutations. Adjusting for ascertainment by supposing that a first mutation is selected to produce the tumor and the remainder are superfluous gives a corrected 38% silent mutations in the TP53 gene in prostate tumors. The analysis of the p53 data base presented in this paper indicates that the majority of silent mutations in tumors are selectively neutral and may be used to measure mutability. Such mutations observed in tumor populations do not play an etiologic role but may be epi-phenomena of the process of carcinogenesis. It is a reasonable speculation that other non-silent, but conservative, mutations may also be neutral and not play a role in the development or progression of tumors. Finally, the difference in the proportion of silent mutations seen in different tumors may serve as a sign of their differing evolutionary pathways.
B.S. Strauss / Mutation Research 457 (2000) 93–104
Acknowledgements This work was supported in part by a grant from the National Cancer Institute, NIH (CA32436). I wish to acknowledge the advice of the University of Chicago Cancer Center Biostatistics Core Support Grant. I would also like to thank Dr. Gerald Holmquist, Dr. Larry Loeb, and Dr. Sergei Rodin for their critical comments on an earlier version of the mansucript. Correspondence about this paper may be addressed to the author at the email address given in the beginning.
References [1] L.A. Loeb, Cancer cells exhibit a mutator phenotype, Adv. Cancer Res. 72 (1998) 25–56. [2] L. Loeb, C. Springgate, N. Battula, Errors in DNA replication as a basis of malignant change, Cancer Res. 34 (1974) 2311– 2321. [3] A.B. Buermeyer, S.M. Deschenes, S.M. Baker, R.M. Liskay, Mammalian DNA mismatch repair, Annu. Rev. Genet. 33 (1999) 533–564. [4] E.C. Friedberg, J.P. Bond, D.K. Burns, D.L. Cheo, M.S. Greenblatt, L.B. Meira, D. Nahari, A.M. Reis, Defective nucleotide excision repair in xpc mutant mice and its association with cancer predisposition, Mutat. Res. 459 (2000) 99–108. [5] I.P. Tomlinson, M.R. Novelli, W.F. Bodmer, The mutation rate and cancer, Proc. Natl. Acad. Sci. U.S.A. 93 (1996) 14800–14803. [6] I. Tomlinson, W. Bodmer, Selection, the mutation rate and cancer: ensuring that the tail does not wag the dog, Nature Med. 5 (1999) 11–12. [7] W.-H. Li, Molecular Evolution, Sinauer Assiciates, Sunderland, MA, 1997, p. xv, 487. [8] S.N. Rodin, G.P. Holmquist, A.S. Rodin, CpG transition strand asymmetry and hitch-hiking mutations as measures of tumorigenic selection in shaping the p53 mutation spectrum, Int. J. Mol. Med. 1 (1998) 191–199. [9] S.N. Rodin, A.S. Rodin, Strand asymmetry of CpG transitions as indicator of G1 phase-dependent origin of multiple tumorigenic p53 mutations in stem cells, Proc. Natl. Acad. Sci. U.S.A. 95 (1998) 11927–11932. [10] B.S. Strauss, Hypermutability and silent mutations in human carcinogenesis, Semin Cancer Biol. 8 (1998) 431–438. [11] B.S. Strauss, Silent and multiple mutations in p53 and the question of the hypermutability of tumors, Carcinogenesis 18 (1997) 1445–1452. [12] B.S. Strauss, Hypermutability in carcinogenesis, Genetics 148 (1998) 1619–1626. [13] P.M. Sharp, M. Averof, A.T. Lloyd, G. Matassi, J.F. Peden, DNA sequence evolution: the sounds of silence, Phil. Trans. Roy. Soc. Lond. B. 349 (1995) 241–247.
103
[14] J.R. Powell, E.N. Moriyama, Evolution of codon usage bias in Drosophila, Proc. Natl. Acad. Sci. U.S.A. 94 (1997) 7784– 7790. [15] T. Hongyo, G.S. Buzard, D. Palli, C.M. Weghorst, A. Amorosi, M. Galli, N.E. Caporaso, J.F. Fraumeni Jr., J.M. Rice, Mutations of the K-ras and p53 genes in gastric adenocarcinomas from a high-incidence region around Florence, Italy, Cancer Res. 55 (1995) 2665–2672. [16] I. Richard, J.S. Beckmann, How neutral are synonymous codon mutations? Nature Genet. 10 (1995) 259. [17] D.W. Collins, T.H. Jukes, Rates of transition and transversion in coding sequences since the human-rodent divergence, Genomics 20 (1994) 386–396. [18] N.K. Sinha, M.D. Haimes, Molecular mechanisms of substitution mutagenesis: an experimental test of the Watson–Crick and Topal–Fresco models of base mispairings, J. Biol. Chem. 256 (1981) 10671–10683. [19] G.P. Pfeifer, M. Tang, M.F. Denissenko, Mutation hotspots and DNA methylation, Curr. Top. Microbiol. Immunol. 249 (2000) 1–19. [20] G.P. Pfeifer, p53 mutational spectra and the role of methylated CpG sequences, Mutat. Res. 450 (2000) 155–166. [21] M. Galliano, U. Kragh-Hansen, A.L. Tarnoky, J.C. Chapman, M. Campagnoli, L. Minchiotti, Genetic variants showing apparent hotspots in the human serum albumin gene, Clin. Chim. Acta 289 (1999) 45–55. [22] C. Milstein, M.S. Neuberger, R. Staden, Both DNA strands of antibody genes are hypermutation targets, Proc. Natl. Acad. Sci. U.S.A. 95 (1998) 8791–8794. [23] P. Hainaut, T. Hernandez, A. Robinson, P. Rodriguez-Tome, T. Flores, M. Hollstein, C.C. Harris, R. Montesano, IARC database of p53 gene mutations in human tumors and cell lines: updated compilation, revised formats and new visualisation tools, Nucleic Acids Res. 26 (1998) 205–213. [24] S.P. Hussain, C.C. Harris, p53 mutation spectrum and load: the generation of hypotheses linking the exposure of endogenous or exogenous carcinogens to human cancer, Mutat. Res. 428 (1999) 23–32. [25] M. Hollstein, D. Sidransky, B. Vogelstein, C.C. Harris, p53 mutations in human cancers, Science 253 (1991) 49– 53. [26] M.S. Greenblatt, A.P. Grollman, C.C. Harris, Deletions and insertions in the p53 tumor suppressor gene in human cancers: confirmation of the DNA polymerase slippage/misalignment model, Cancer Res. 56 (1996) 2130–2136. [27] T. Hernandez-Boussard, P. Rodriguez-Tome, R. Montesano, P. Hainaut, p53 mutation database: a relational database to compile and analyze p53 mutations in human tumors and cell lines, International Agency for Research on Cancer. Hum. Mutat. 14 (1999) 1–18. [28] S. Benzer, On the topography of the genetic fine structure, Proc. Natl. Acad. Sci. U.S.A. 87 (1961) 403–415. [29] Z. Hatahet, M. Zhou, L.J. Reha-Krantz, S.W. Morrical, S.S. Wallace, In search of a mutational hotspot, Proc. Natl. Acad. Sci. U.S.A. 95 (1998) 8556–8561. [30] D.R. Walker, J.P. Bond, R.E. Tarone, C.C. Harris, W. Makalowski, M.S. Boguski, M.S. Greenblatt, Evolutionary
104
B.S. Strauss / Mutation Research 457 (2000) 93–104
conservation and somatic mutation hot spot maps of p53: correlation with p53 protein structural and functional features, Oncogene 18 (1999) 211–218. [31] J.C. Delaney, J.M. Essigmann, Context-dependent mutagenesis by DNA lesions, Chem. Biol. 6 (1999) 743–753. [32] W.H. Li, C.I. Wu, C.C. Luo, A new method for estimating
synonymous and non-synonymous rates of nucleotide substitution considering the relative likelihood of nucleotide and codon changes, Mol. Biol. E 2 (1985) 150–174. [33] S.G. Chi, R.W. deVere White, F.J. Meyers, D.B. Siders, F. Lee, P.H. Gumerlock, p53 in prostate cancer: frequent expressed transition mutations, J. Natl. Cancer Inst. 86 (1994) 926–933.