Epidemiology of Doublet/Multiplet Mutations in Lung Cancers: Evidence that a Subset Arises by Chronocoordinate Events

Epidemiology of doublet mutations in the EGFR and TP53 genes in lung cancer

Analysis by smoking status, gender, and ethnicity does not reveal a significant difference in doublet frequency in the EGFR gene (Table 1, Table S1). The average age of patients with singlets and doublets is similar (61.4±11.3 vs 65.8±10.3).

The results for the TP53 gene in lung cancer are similar with one exception. The frequency of TP53 doublets is significantly higher in Asians than in Caucasians (3.2% vs 1.8%, P = 0.03) (Table 1, Table S2), but this is not significant when corrected for multiple comparison (six Fisher's Exact Tests performed for the EGFR and TP53 genes). Additional data are necessary to evaluate this point.

Overall, 98/1627 (6.0%) of EGFR mutations and 54/2387 (2.3%) of TP53 mutations in human lung cancers are doublets, a much higher frequency than that observed in the lacI gene in Big Blue mice (8-fold and 3-fold, respectively) (Table 2).

Doublets tend to cluster in the EGFR and TP53 genes in lung cancer: the “half-life of mutation spacing” is 9 bp and 15 bp, respectively

Spontaneous lacI doublets (0.7% of total mutations) are clustered with a spacing fitting an exponential distribution and show a similar mutation pattern to singlets, indicating a chronocoordinate event [13], [14]. Based on the doublet spacing between two mutations, EGFR doublets can be divided into two groups: the proximal group (<100 bp, average spacing 22.8±24.4-bp) and the distal group (>809 bp, average spacing 12.3±5.1 kb) (Table S1). Analogous to lacI doublets, a subset [40% (39/98)] of EGFR doublets occurring in the same exon with proximal spacing shows an exponential (R² = 0.9979) rather than a quasi-uniform distribution (Figure S1). Half the doublets have mutations separated by 9 bp or less (the “half-life of mutation spacing”; see Terminology in Materials and Methods).

The spacing distribution for TP53 doublets in lung cancer occurring in the same exon (43%, 23/54) also shows proximal spacing (<86 bp, average 20.2±20.9 bp) and fits to an exponential distribution (R² = 0.979) (Table S2, Figure S1) with a half-life of mutation spacing of 15 bp. Doublets occurring in different exons have distal spacing (131–4504 bp, average 1317.9±1000 bp) separated by an intron. A similar spacing to that in lung cancer was found for TP53 doublets within the same exon in breast and colorectal cancers, with a half-life of mutation spacing of 24 and 16 bp, respectively (Tables S4 and S5).

Thus, about one-third of EGFR and TP53 doublets are highly clustered (≤ 30 bp) and all doublets occurring within a single exon have an inter-mutation spacing that is exponentially distributed rather than the expected quasi-uniform distribution, consistent with chronocoordinate events in lacI in normal mouse tissue.

The OMIDI pairs are strong evidence for clustered chronocoordinate mutations

The great majority of spontaneous human germlime or somatic mouse MIDIs are OMIDIs (FIX, Big Blue); indeed, about half are single base deletions [15]–[18]. In contrast, there is a very low fraction of OMIDIs (0.26%, 4/1526) and nonsense mutations (0.13%, 2/1526) in EGFR singlets, consistent with a strong selection for mutations that alter, rather than delete, function. Eight of 98 (6%) EGFR doublets contain a pair of frameshift mutations (OMIDI pairs; see abbreviations in Materials and Methods). All eight of the OMIDI pairs are closely spaced (Table 3, Figure S2), with the net result being an in-frame mutation. Given one OMIDI mutation, a second random mutation 3′ to the first (either IMIDI or OMIDI) will restore reading frame only 1/3 of the time, and a sub-fraction of that third is still protein truncating because a nonsense mutation occurs prior to the second mutation. In total 0.49% of reported EGFR mutations (8/1627) are clustered OMIDI pairs with no truncation mutation occurring before the second OMIDI. These eight OMIDI pairs are expected to reside on the same allele; otherwise, compound heterozygous OMIDI doublets will lead to two allelic truncations, eliminating all EGFR gene function. Meanwhile, both of the clustered OMIDIs should also occur in rapid succession (chronocoordinate mutations) within the same cell cycle because the frequency of secondary mutations randomly occurring in a proliferation-deficient cell resulting from a single OMIDI is rare. The remaining 14 deletions (in exons), 9 indels and 3 doublets are all in-frame (IMIDI) mutations (Table 2; Table S1).

The mutations in the single OMIDI pair reported in the TP53 gene are 2309 bp apart, in different exons (Table S2, ID # 14373), and, unlike the OMIDI pairs in the EGFR gene, do not together restore the reading frame.

EGFR multiplets are consistent with a singlet and a highly clustered chronocoordinate second mutation

Four triplets and one quadruplet in EGFR were found at a frequency of 0.3% (5/1627) of lung cancers with mutations prior to treatment with tyrosine kinase inhibitor (TKI) (Table 4). It is of interest that pairs of mutations in each of the triplets and three mutations within the quadruplet are tightly clustered with proximal spacing (3–60 bp), similar to doublets in lacI in normal mouse tissues. The third mutation is located in a different exon with distal spacing. The data are consistent with sequential selection of two mutations in which one of the mutations is a chronocoordinate doublet or triplet.

TP53 multiplets (6 triplets and one quadruplet) also account for 0.3% (7/2387) of total mutations (Table S3). Five multiplets (71%, 5/7) reside in the same exon or two exons, consistent with one mutation and one chronocoordinate doublet. Seven multiplets had mutations separated by 15 spacing regions. Of these, six were highly clustered (≤30 nt) and nine were within 100 nucleotides or less. Eighty percent (16/20) of multiple mutations in breast cancer and 76% (25/33) in colorectal cancer also reside in one or two exons (Tables S6, S7).

A comparison of OMIDI pairs and closely spaced doublet and multiplet mutations in the EGFR and TP53 genes is shown in Table 5. The frequency of OMIDI pairs in the EGFR gene is approximately 10-fold greater than in the TP53 gene (and the one OMIDI pair in the TP53 gene does not result in a net in-frame mutation), reflecting the more frequent selection of mutations that alter, rather than delete, EGFR protein function. The frequency of closely spaced mutations in doublets and multiplets in both genes is similar.

All characterized EGFR doublets are allelic: a significant subset of chronocoordinate mutations provides an alternative explanation

At face value, the allelism data suggests that the driver/driver mutations in EGFR need to be on the same molecule [10]. However, the existence of a substantial subset of chronocoordinate mutations provides an alternative explanation. Sixteen heterozygous EGFR doublets from three different studies have been analyzed by cloning or allele-specific amplification (Table S1) [10], [19], [20]. Subsequent sequencing of the products revealed that all these alleles were either doubly mutated or wild type, indicating that, in every case, both mutations were located on the same allele. If the doublet pairs result from independent events, half may be expected to be compound heterozygotes. The observation of 100% allelism (16/16) vs. zero compound heterozygotes (p = 0.002) may reflect that, functionally, the driver/driver mutations are required to be on the same molecule even though the EGFR protein forms dimers [10]. However, the data are also consistent with 50% chronocoordinate events, putatively during strand replication or patch repair, and 50% independent sequential mutations without a requirement that both mutations be on the same molecule (8 of 8 allelic mutations observed when 4 of 8 are expected at random; p≤0.08).

Unfortunately, to our knowledge, allelic analysis has not been performed on heterozygous TP53 doublets.

Data are consistent with chronocoordinate mutations (with caveats)

Several aspects of the data are consistent with a subset of EGFR and TP53 doublets arising from choronocoordinate events rather than from hypermutability (random distributed spacing) or sequential selection. About 32–35% of the doublets found in the TP53 (35%, 19/54) and EGFR (32%, 31/98) genes are either OMIDI pairs or highly clustered doublets separated by ≤30 nt. However, there are caveats for each inference that must be ruled out in future work. The main value of the present analysis is to generate data that constrain our hypotheses and, by Occam's razor, generate a simple hypothesis (the subset of chronocoordinate events) consistent with multiple data.

(1) The OMIDI pairs are the best evidence for chronocoordinate mutations. For most genes, the majority of spontaneous MIDIs are OMIDIs, but in the EGFR gene, OMIDIs are rare, consistent with the overall selection for altered function rather than lack of function. It is notable that the eight EGFR doublets containing OMIDI pairs all have closely spaced mutations that result in a net in-frame mutation, thus producing an EGFR protein with potentially altered, rather than absent, function. However, no experimental evidence of a protein containing one of these doublet mutations has been demonstrated. The most reasonable explanation for this occurrence is through chronocoordinate mutations that are selected together for altered protein function. An alternative explanation is that OMIDI pairs might somehow be sequential, although random mutations within a few nucleotides in the genome must be extremely unlikely.

(2) The relatively high frequency of clustered doublet mutations in the EGFR gene is consistent with chronocoordinate mutations. The exponential distribution of the spacing between the two mutations in EGFR and TP53 doublets in the same exons (R² = >0.98) is highly unlikely by chance.

A simulation was performed previously for each of the TP53 exons 5–9 to test the null hypothesis that the mutations in TP53 doublets are independent events and that the distribution of the spacing between the two mutations depends only on the spectrum of mutations in the TP53 gene and the size of the mutation target [13]. The first mutation of each simulated doublet was drawn randomly based on the observed distribution of singlet mutations in a given exon. Using the singlet distribution from the TP53 IARC database (release 8) accounts for bias that may result from closely spaced hot spots. The second mutation in the doublet was drawn randomly based on a uniform distribution over the TP53 coding sequence for that exon, since it is likely to be a “hitchhiker” (passenger) mutation, rather than a driver mutation for the tumor.

The mutation spacing in the simulated doublets was compared with 402 actual TP53 doublets in the IARC database from all types of cancer. The results demonstrated that two mutations within doublets are not independent events by statistical comparison of the observed and expected distributions, and indicated that more doublets occurred with a mutation spacing of less than 30 nucleotides than expected by chance (P = 0.03, 0.02, 0.0006, and 0.01 for exons 5–8). These results are consistent with the occurrence of chronocoordinate mutations in the human TP53 gene. An alternate explanation for the clustering of doublets is that they may represent some as yet unappreciated functional constraints.

(3) The allelic nature of doublets is consistent with chronocoordinate mutations. Among 98 doublet pairs reported in the literature, only a few have been analyzed to determine whether they are present on the same or different alleles. Among 16 doublet pairs tested by cloning or allele-specific amplification, all 16 pairs were found to be present on the same allele. However, for the remaining doublet pairs, in which cloning or allele-specific amplification was not performed, the possibility that the mutations were present on different alleles cannot be ruled out.

(4) The relatively high frequency of multiplets with two or more clustered mutations and an additional mutation at a distal location also argues for a chronocoordinate mutational event plus an independent single mutational event. Again, an alternative explanation could involve functional constraints on the clustered mutations.

In aggregate, these data are consistent with a contribution of clustered chronocoordinate mutations to human cancer. An excess of doublets or multiplets with a subset of clustered distribution occurs more frequently than predicted by chance in a wide range of organisms, including riboviruses, DNA viruses, prokaryotes, yeast, and eukaryotic cell lines and tissues [23], [24].

Data support driver plus passenger TP53 doublets in lung cancer

About 22% of the second mutations in doublets are expected to be silent if they are “passenger” mutations, rather than “driver” mutations [25]. In the TP53 doublets, there are five silent and two intronic nucleotide changes (13%, 7/54), not significantly different from the 23.5% silent nucleotide changes expected to occur with random passenger mutations (p = 0.46) [26], [27]. Silent nucleotide changes (18%, 4/22) within TP53 multiplets are also close to 23.5%. Together, these data imply that TP53 doublets and multiplets consist of a driver plus passenger mutation pattern, in contrast to the driver/driver pattern found in EGFR doublets [10].

Doublet mutations are associated with mutation showers in mouse [9]. The advent of massively parallel sequencing can facilitate the analysis of sufficient numbers of samples to define any doublets and then determine whether these doublets are associated with mutation showers in cancer. The non-random clustering mutations in mutation showers should provide more definitive data for the occurrence of chronocoordinate mutations in human cancers. The contribution of mutation showers to cancer remains to be determined.

Terminology and abbreviations

Samples and mutational analysis

In order to investigate the mechanism and characteristics of doublets and multiplets, EGFR mutations identified by sequencing were collected from 66 published papers and our updated EGFR mutation database [10], [31] (www.egfr.org). TP53 mutations were collected from IARC version 12 (www-p53.iarc.fr) and excluded mutations in pulmonary fibrosis and in lung cancer in patients exposed to smoky coal emissions, radon, mustard gas, asbestos, heavy metals, and atomic bomb radiation (γ-rays). We also excluded two papers in which multiple mutations comprised >50% of the total mutations, most likely due to PCR artifacts, since multiple mutations generally make up only ∼3% of total mutations. TP53 mutations in breast and colorectal cancers also were retrieved and the spacing distribution of doublets/multiplets was analyzed.

Statistical Analysis

Mutation patterns and other categorical count distributions were tested for significant differences by the Fisher's Exact Test or unordered R×C contingency tables using the “Fisher-Freeman-Halton” test implemented by the StatXact statistical analysis software package (CYTEL Software Corporation, Cambridge, MA).