Inverse PCR-Based RFLP Scanning Identifies Low-Level Mutation Signatures in Colon Cells and Tumors Free

Large numbers of mutations are postulated to occur as early events in carcinogenesis (1, 2) and, for certain types of human tumors, to be a driving force in generating clonogenic, causative genetic changes required for multistage carcinogenesis (2, 3, 4). Cancers of the mutator phenotype are estimated to contain genetic alterations on the order of thousands per cell (2, 5). Widespread mutations occurring almost randomly at lower levels [mutation frequency (MF) <10⁻²] throughout the genome are in fact estimated to have an even higher incidence (6) and possibly occur in a variety of human tumors (7). These putative low-level genomic events are assumed to take place very early in malignant transformation, and their presence is highly significant due to their potential use as molecular markers for early identification of genomic instability that can lead to cancer and their putative influence on the ability of tumors to resist drug treatment and/or metastasize (8, 9).

However, because such genetic events fall below the detection threshold of practically all mutation scanning methods, they have been very difficult to detect (7, 10). Phenotypic mutation detection methods applied to the identification of low-level nucleotide changes in the hypoxanthine phosphoribosyltransferase (HPRT) and lacI genes can, in fact, detect a very high number of low-level nucleotide changes in mismatch repair-deficient colon cancer human cell lines (11, 12) and in transgenic rat tumors (13). However, the inability of applying these mutation detection systems to nonclonogenic cells and to genes relevant to cancer has precluded direct examination of low-level mutations in surgical tumor samples that potentially contain repair defects similar to those observed in the cell lines. Due in part to this technical hurdle, the influence of low-level mutations in tumor onset, progression, and response to treatment has been difficult to assess (1, 14), and the origins of genomic instability in human cancer remain a subject of controversy (15).

We describe inverse PCR-based, amplified restriction fragment length polymorphism (iFLP), a new technique designed to allow high-throughput, genome-wide scanning for low-level mutations in cell lines and human tumors. DNA sequences are circularized after a self-ligation reaction and then exposed to a TaqI restriction enzyme digestion (Fig. 1,A). DNA circles that do not normally contain a natural restriction site are converted to double-stranded linear DNA fragments only if they have acquired nucleotide changes leading to formation of TaqI site(s). By ligating TaqI-specific adaptors, these DNA fragments are PCR amplified selectively over their wild-type counterparts. Digestion by TaqI anywhere along the sequence leads to formation of a single DNA size that allows highly accurate and sensitive detection by size-resolving methods. We developed a system that harnesses this simple principle on a genome-wide scale by using high-throughput inverse PCR in conjunction with automated denaturing high-precision liquid chromatography (dHPLC)-based quantitation (Fig. 1 B). Because the number of genes examined depends on the number of inverse PCR reactions performed and each DNA circle examined contains 80–100 positions that can generate TaqI sites, several thousand genomic positions can be examined per iFLP experiment at a selectivity of ∼1 mutant in 10⁵ wild-type sequences. Using this technology, widespread, low-level mutations associated with genomic instability at the nucleotide level [single-nucleotide instability (SNI)] were discovered in repair-deficient colon cancer cells and colon cancer surgical specimens, and their extent and diversity were assessed. To our knowledge, this is the first report describing low-level (MF <10⁻²) mutational signatures and their diversity in human solid tumor surgical specimens.

The colorectal carcinoma cell lines SW-480 (ATCC CCL-228), DLD-1 (ATCC CCL-221) and HCT-116 (ATCC CCL-247) were purchased from the American Type Culture Collection. DLD-1 and HCT-116 cells are deficient in mismatch repair activity and display microsatellite instability (MSI) and a high spontaneous mutation rate at coding sequences (11, 16, 17). SW-480 cells do not display mismatch repair deficiency (18) or MSI (16, 19, 20, 21), but they harbor a G→A mutation in codon 273 of p53 gene (Arg→His substitution; Ref. 22). We observed that an NlaIII restriction site is generated in the mutated p53 sequence, and this property was used as a positive control in certain experiments. The osteosarcoma (HOS) cell line (ATCC CRL-1543), which is negative for mutations in both p53 and HPRT genes, was also used. After initial treatments, cell lines were grown for 32 generations in DMEM medium supplemented with 10% fetal bovine serum (Life Technologies). To perform the HPRT assay, DLD-1 cells grown for 32 generations were treated with 6-thioguanine (6TG). 6TG, 6 μg/ml, was added along with DMEM and 10% fetal bovine serum after 2 h of cell settlement. Cells were allowed to grow for another ∼12 doublings in the presence of 6TG to allow for phenotypic expression of the HPRT mutations or lack thereof. After this, cells were washed with PBS (Life Technologies), trypsinized with 0.25% trypsin-EDTA solution (Sigma), and counted before genomic DNA extraction. Ten human colon cancer surgical biopsies and corresponding normal tissue from the same patients were anonymously obtained from the Massachusetts General Hospital Tumor Bank. These specimens were removed immediately after surgery and snap-frozen in liquid nitrogen to preserve DNA. Genomic DNA was extracted from frozen samples using a commercial kit (QIAamp DNA mini kit; Qiagen, Inc.).

A single-tube procedure was developed for genomic digestion via MseI, genome-wide circularization, exonuclease treatment, TaqI digestion, and TaqI-specific linker-ligation. Genomic DNA, 1–1000 ng depending on experiment, was suspended in 50 mm Tris-HCl, 10 mm MgCl₂, 10 mm DTT, 1 mm ATP, and 25 μg of BSA (pH 7.5) and digested with MseI (New England Biolabs, Beverly, MA) for 1.5 h to produce sticky ends for subsequent circularization. After heat inactivation of MseI (70°C for 30 min), DNA was circularized by addition of 1,200 units T4 DNA ligase (New England Biolabs) and incubation at 14°C overnight. After ligase inactivation (70°C for 30 min), circularized DNA was treated with 0.6 μl (12 units) of Escherichia coli exonuclease I (New England Biolabs) plus 2.4 μl (12 units) of λ exonuclease (New England Biolabs) at 37°C for 1.5 h to eliminate noncircularized DNA. The temperature was then raised to 80°C to inactivate exonucleases. Circularized DNA was then digested with TaqI (New England Biolabs) followed by heat inactivation of the enzyme at 80°C for 40 min. In certain experiments, TaqI was substituted by the enzymes NlaIII, or HpyCH4IV. Linkers were then added to digested DNA, annealed at 50°C, and then slowly cooled to 10°C. Linkers used were: 5′-AGG CAA CTG TGC TAT CCG AGG GAA-3′ (2 μl from 2 μg/μl) and 5′-CGT TCC CTC GGA-3′ (2 μl from 1 μg/μl) for TaqI and HpyCH4IV; and 5′-AGG CAA CTG TGC TAT CCG AGC ATG-3′ (2 μl from 2 μg/μl) for NlaIII. To this mixture was added 1 μl of T4 DNA ligase (2000 units/μl; New England Biolabs), followed by incubation at room temperature for 30 min. This single-tube procedure takes about 30 h to perform, including the overnight circularization.

In certain experiments, iFLP was performed on a genomic fraction rather than the full genome to enrich target DNA sequences and to allow higher selectivity in mutation detection. For this purpose, after the initial MseI digestion, DNA was separated on a 1% agarose gel and DNA in the molecular weight size region 400 to 700 bp was gel-purified (QIAquick gel extraction kit; Qiagen, Inc.). The iFLP procedure was subsequently performed with the purified enriched genomic DNA, which is estimated to contain about 5–10% of the original genome.

After ligation in the single-tube protocol, 4–8 μl of the ligated genome were PCR-amplified using linker-specific primers 5′-AGG CAA CTG TGC TAT CCG AGG GAA-3′ for TaqI or HpyCH4IV-digested DNA and 5′-AGG CAA CTG TGC TAT CCG AGC ATG-3′ for NlaIII-digested DNA. PCR was carried out in a Perkin-Elmer Gene-Amp PCR 9600 system (PE Biosystems) using a Titanium TaqPCR kit (Clontech) with thermocycling conditions: 72°C for 8 min followed by 22 cycles of 95°C for 30 s; 70°C for 1 min and 68°C for 5 min; and 4°C hold. A set of second, gene-specific quantitative PCR reactions (high-throughput inverse PCR) was then performed to detect presence of amplified inverted sequences, by using primers amplifying a short segment (∼50–200 bp) from the genomic fragment of interest. For example, gene-specific PCR using the inverse primers 5′-AATGCCGTTTTCTTCTTGACTG-3′ (forward) and 5′-TTCCGTCCCAGTAGATTACCAC-3′ (reverse) results in a ∼200-bp sequence. Presence of an amplicon following PCR indicates mutations leading to new restriction sites in a 611-bp circularized segment of p53, nucleotide 14361–nucleotide 14972, which was originally negative for restriction sites. To perform the high-throughput PCR reactions, a 0.5-μl sample from the first PCR product was used in each 10-μl reaction, using Titanium TaqDNA polymerase and the thermocycling conditions: 95°C for 1 min followed by 30 cycles of 95°C for 30 s; 68°C for 1 min and 68°C for 5 min; and 4°C hold.

Genomic regions selected for amplification via high-throughput PCR do not contain TaqI regions and therefore are not expected to yield a product unless a mutation modified the sequence to generate a TaqI digestion site. A 500-bp DNA fragment contains, on average, 80–100 mutable sites that can form a TaqI site following a single-nucleotide substitution, deletion, or insertion. Therefore each PCR reaction from the high-throughput PCR set evaluates 80–100 potential sequence changes simultaneously. In control experiments, use of an alternative enzyme whose recognition sequence is present in the wild-type sequences was used to serve as positive (+) control for iFLP, i.e., 100% mutant fragments. For example, a circularized DNA fragment with a HpyCH4IV restriction site in the wild-type sequence when digested with HpyCH4IV instead of TaqI yields a same-size PCR product with the one obtained following a TaqI-forming mutation. Conversely, some sequences that were examined were chosen to obtain no wild-type HpyCH41V while they also contained a wild-type TaqI site. These samples used the TaqI enzyme as a positive control and HpyCH41V as a mutation-scanning enzyme.

Inverse PCR products were analyzed either by ethidium bromide-stained 1% agarose gel electrophoresis or via HPLC chromatography on a WAVE dHPLC system (Transgenomics, Inc., Omaha, NE). The WAVE system is equipped with UV and fluorescent detectors and with two 96-well autosamplers that enable high-throughput analysis of PCR products. Eight μl of product from each PCR reaction were injected and analyzed at nondenaturing conditions, at 48°C. DNA elution profiles were recorded and displayed as chromatograms. To estimate the approximate MF for mutations discovered by iFLP, the positive iFLP controls were diluted into negative samples to produce samples with known ratios of mutant to wild-type alleles. High-throughput, inverse PCR was carried out using samples of known MF in parallel with unknown samples, and inverse PCR amplicons were quantitated via peak integration using the Navigator software of the WAVE system. The use of the high-sensitivity fluorescent detector allowed quantitation of PCR amplicons in the exponential phase of the PCR reaction, and therefore quantitation of MFs was not affected by PCR saturation effects. Experiments were repeated three independent times, and only PCR products demonstrating concordance among all three experiments were considered to be regions positive for TaqI-forming mutations.

Independent verification that iFLP-generated DNA amplicons correspond to genuine mutations at the corresponding circularized fragment was obtained by PCR amplifying directly from genomic DNA the indicated DNA region, followed by amplification via primer ligation at the mutation (23, 24). This approach uses the formation of a TaqI site at the mutation to ligate a linker and selectively PCR-amplify mutation-containing alleles. Amplification via primer ligation has been recently used to detect mutations with a selectivity of ∼10⁻³ mutant-to-wild type alleles (24). For verification of the iFLP-detected mutation at the OGG1 gene, a 652-bp genomic DNA fragment encompassing the segment in which the mutation was detected by iFLP was amplified directly from genomic DNA, using primers 5′-AGGGGCAATATAAGGCTTAGAG-3′ (forward) and 5′-AGGGCCACATAATGTTATCAGG-3′ (reverse). One μg of genomic DNA was used in each PCR reaction using Advantage high-fidelity PCR amplification and touch-down PCR (94°C for 30 s; 10 cycles of 94°C for 20 s, 65°C for 20 s, and 68°C for 1min, with the annealing temperature decreasing by 1°C/cycle; 20 cycles of 94°C for 20 s, 55°C for 20 s, and 68°C for 1min; 68°C for 6 min; 4°C hold). Gel-purified PCR products (50 μg) were digested with TaqI enzyme, purified by QIAquick PCR purification kit (Qiagen, Inc.), and ligated with linkers 5′-AACTGTGCTATCCGAGGGAATGGAACCTGAGTCCTCTCACCGCA-3′ and 5′-CGT GCG GTG AGA-3′, as described previously (24). Ligated products were PCR-amplified using a linker-specific primer (5′-AAC TGT GCT ATC CGA GGG AA-3′), corresponding to the first 20 bases of the 44-mer oligonucleotide that was ligated, and the reverse primer used in the original amplification from genomic DNA. The PCR products were gel-purified and sequenced at the Dana-Farber Cancer Institute Molecular Biology Core Facility to identify the position and type of mutation that formed the new TaqI site on OGG1.

Studies were also carried out using single-gene fragments obtained via amplification from genomic DNA and applying the iFLP protocol with small modifications. The high-throughput PCR step is not necessary in single-gene iFLP, because the amplicon from the first PCR already conveys the information on presence or absence of a mutation. A 611-bp region from the p53 gene (nt14361–nt14972) was PCR-amplified using a high-fidelity polymerase (Advantage HF2; Clontech). The primers used for PCR amplification were 5′-TTG GGA GTA GAT GGA GCC TGG T-3′ (forward) and 5′-TTT TTG AAA GCT GGT CTG GTC C-3′ (reverse). This DNA fragment contains MseI restriction sites (5′-TTAA-3′) close to the 5′ and 3′ ends, and this enables circularization after MseI digestion. Ten PCR reactions were carried out, each using 1 μg of genomic DNA, and the PCR products were mixed in a single tube to ensure that copies of low-level mutated alleles, even at a MF of ∼10⁻⁶, would be included in the amplicons. PCR thermocycling conditions were: 94°C for 30 s; 10 cycles of 94°C for 20 s, 65°C for 20 s, and 68°C for 1min, with the annealing temperature decreasing by 1°C/cycle, touchdown PCR; 20 cycles of 94°C for 20 s, 55°C for 20 s, and 68°C for 1 min; 68°C for 6 min; 4°C hold. After PCR amplification, the PCR product was gel-purified to remove traces of genomic DNA (QIAquick gel extraction kit; Qiagen, Inc.), and DNA was quantitated with picogreen (Molecular Probes). The iFLP single-tube protocol was then applied to the amplified DNA fragment. To verify that single-gene iFLP can detect the single NlaIII restriction site present in SW-480 cells, which is otherwise absent in the wild-type sequence, NlaIII digestion was used instead of TaqI.

Using single-gene iFLP, a 550-bp region from the HPRT gene was amplified and tested via iFLP for mutations forming NlaIII restriction sites. The primers used were 5′-TTT AAC GTC AGT CTT CTC T-3′ (forward) and 5′-CCA GTT CTA AGG ACG TCT GTA-3′(reverse), and the PCR thermocycling conditions were the same as for the p53 region described above. The amplified HPRT fragment does not contain NlaIII sites, however we observed that a low-level mutation generating an NlaIII site (C→T missense mutation, C39838T, in codon 508 of the HPRT gene exon 7, in GenBank sequence no. M26434) was previously reported in DLD-1 mismatch repair-deficient cells (17, 21). To evaluate iFLP detection of this mutation and to quantitate the detected MF, the amplified HPRT fragment was mutagenized using mutagenesis via overlap extension (25), and a mutation-positive DNA fragment was obtained. The mutation-positive sample was diluted into negative samples to produce mixed samples with decreasing ratio of mutant-to-wild type alleles to a MF of 1 × 10⁻⁶. Ethidium bromide-stained agarose gels of the PCR products were quantitated in a gel-digitization system (α-Innotech Corp.) equipped with a low-level light-detecting camera and software for quantitative determination of fluorescent gel bands. The reproducibility of single-gene iFLP and genome-wide iFLP was verified by performing at least three independent experiments.

Experiments for determination of MSI were performed using the dHPLC method (26, 27) for assessment of BAT25 and BAT26 microsatellite markers (28). The procedures described by Pan et al. (26) were followed with minor modifications to amplify PCR fragments containing the microsatellite markers BAT25 and BAT 26 and assess fragment length changes via the dHPLC. In brief, primers used for BAT26 were: 5′-GCAGTCAGAGCCCTTAAC-3′ (forward) and 5′-CCATTCAACATTTTTAACCC-3′ (reverse); and primers for BAT25 were: 5′-GGGAGTGATTCTCTAAAGAGT-3′ (forward) and 5′-CTAAAATGCTCTGTTCTC-3′ (reverse). A touchdown PCR reaction using titanium Taq polymerase (BD-Clontech) was used: 94°C for 30 s; 10 cycles of 94°C for 20 s, 65°C for 20 s, and 68°C for 1 min, with the annealing temperature decreasing by 1°C/cycle, touchdown PCR; 20 cycles of 94°C for 20 s, 55°C for 20 s, and 68°C for 1 min; 68°C for 6 min; 4°C hold. PCR products were injected in the dHPLC at 50°C using dHPLC gradient conditions described by Pan et al. (26), and presence of MSI for the specific marker was detected as a shift in the obtained dHPLC peak.

Generation of an NlaIII site by a single-point mutation anywhere along a circularized DNA sequence produces a PCR amplicon of a single size in the iFLP assay. This facilitates detection using size-separation visualization methods (gel electrophoresis, dHPLC) and distinguishes mutations from PCR mis-priming events. Single-gene iFLP was carried out in a p53 region that contains a G→A mutation generating a new NlaIII site in SW-480 cells, as well as in HOS cells that are wild type and contain no NlaIII site. A strong gel band appears at ∼600 bp for cells carrying the mutation, which is absent for wild-type cells (Fig. 2,A, Lanes 1 and 7, respectively). The amplified PCR product was excised from the gel and sequenced, and the expected mutation was confirmed. To examine the selectivity of mutation detection (i.e., the minimum mutant-to-wild type alleles detectable), gradual dilutions of the PCR product from mutation-positive genomic DNA into wild-type DNA were used (Fig. 2,A, Lanes 2–6). The mutation was detectable within a million-fold excess wild-type DNA. SW-480 cells were then examined for mutations generating NlaIII sites in a second DNA region, within the HPRT gene, which does not contain NlaIII site in the wild type. No agarose-resolved bands were evident (Fig. 2 B, Lane 6). In the presence of NlaIII site, the anticipated iFLP amplicon is 558 bp. An HPRT DNA fragment mutagenized to contain NlaIII-forming mutation (C→T missense mutation in codon 508) was then gradually diluted within the wild-type HPRT fragment and a ∼550-bp band with an intensity inversely proportional to the MF was observed, in the presence of up to a million-fold excess wild-type DNA. Given that there are 87 mutable positions that can generate NlaIII sites via a single-base substitution, insertion, or deletion on the examined HPRT fragment, these data indicate that single-gene iFLP scans the circularized HPRT fragment at 87 positions for specific base changes, at a selectivity of ∼10⁻⁶ at each position.

To examine whether single-gene iFLP can detect random, low-level mutations generated at the HPRT gene by genomic instability, the mismatch repair-deficient cell line DLD-1 was cultured for 32 generations so that development of previously reported base changes in HPRT can be anticipated, including an NlaIII-forming mutation at codon 508 (17, 21). Initially, a 6TG treatment was applied to select for HPRT mutants, and genomic DNA extracted from such mutants was examined via single-gene iFLP at HPRT. A strong 558-bp gel band was resolved when 6TG-treated DLD-1 mutants were screened (Fig. 2,C, Lane 2). Excision of this band from the gel followed by sequencing revealed the anticipated C→T transition mutation that forms a new NlaIII site. Next, DLD-1 and SW-480 cells were examined in the absence of 6TG treatment. The 558-bp band was absent for the repair-proficient SW-480 cells but present for the repair-deficient DLD-1 cells, albeit with a ∼5-fold lower intensity than when DLD-1 were 6TG-treated (Fig. 2,D, Lanes 1and 2, respectively), as determined by microdensitometry performed on the two gel pictures using a common standard. After excision of this gel band, the DNA was sequenced, and the anticipated mutation was confirmed. Semiquantitation of the MF revealed in DLD-1 cells was performed. By including dilutions of an artificially mutagenized HPRT fragment at codon 508 within the same experiment (Fig. 2 D), semiquantitation of gel band intensity via digital densitometry revealed a MF of approximately 10⁻⁴ for the NlaIII-forming HPRT mutation in DLD-1. Given that DLD-1 cells were allowed to grow for 32 cell doublings, a mutation rate of ∼0.31 × 10⁻⁵ mutations/cell/generation for the identified mutation is derived. From the data of Bhattacharyya et al. (17), an expected mutation rate of ∼0.27 × 10⁻⁵ mutations/cell/generation is inferred, because codon 508 mutation makes up about 16% of an overall reported HPRT mutation rate of 1.51–1.9 × 10⁻⁵ for DLD-1 (17, 18). Therefore the mutation rate determined via single-gene iFLP is in agreement with previously reported data.

To apply iFLP to high-throughput mutation screening, we tested whether the single-tube procedure can detect the NlaIII-forming G→A missense mutation in the p53 gene of SW-480 cells. Genomic DNA extracted from HOS (mutation-negative) cells and SW-480 cells were subjected to the single-tube iFLP protocol to examine presence of mutation-generated NlaIII sites on circularized segments from p53. After inverse PCR, strong bands appeared on ethidium gels for the mutant SW-480 (Fig. 3,A, Lanes 2, 4, 6, 8, and 10) but not for the wild-type HOS cells. Variation of the amount of starting genomic DNA (1–200 ng) demonstrated that iFLP applied on 1 ng of genomic DNA is sufficient to generate a strong signal (Fig. 3,A, Lane 10). A particularly strong gel band was obtained when 200 ng of genomic DNA enriched for the examined sequences (400–700-bp fraction of MseI-digested genome) was used as the starting material in the single-tube procedure (Fig. 3 A, Lane 4). Enrichment allows the highest selectivity in iFLP mutation detection, because it ensures the presence of several copies of mutated sequences in the sample even at low MFs, ∼10⁻⁵ mutant-to-wild type ratio.

Genome-wide iFLP was first applied to screening 30 DNA segments from a number of genes for widespread mutations in the mismatch repair-deficient cell lines DLD-1 and HCT-116 and the repair-proficient SW-480 and analyzed by gel electrophoresis. Inverse PCR products for DLD-1 and SW-480 cells along with mutation-positive controls (+) for each cell line are depicted in Fig. 3,B. Out of 30 regions screened, four regions belonging to diverse genes demonstrated formation of new restriction sites for DLD-1 cells (Fig. 3 B, arrows), whereas all 30 regions from SW-480 cells were negative.

Next, genome-wide iFLP was combined with automated dHPLC detection and applied to analyze a total of 48 gene regions using genomic DNA from the colon cancer cell lines DLD-1, HCT-116, and SW-480, as well as from 10 colon cancer and corresponding normal tissue surgical samples. The results confirmed the gel electrophoresis findings for the DLD-1 and SW-480 cell lines (Fig. 4). Thus, chromatographs from the OGG1, BRCA2, and MSH2 regions of interest from DLD-1 cells showed small peaks at the same retention time as the mutation-positive controls, whereas SW-480 cells showed no peaks. To estimate MFs for the identified mutations, dHPLC was used in the high-sensitivity fluorescence mode, and experiments using dilutions of the mutation-positive controls (obtained by digesting the circularized DNA with an enzyme containing a wild-type restriction site) into negative samples were performed together with the DLD-1 and SW-480 samples. Fig. 4,E depicts a typical dilution experiment carried out for the BRCA2 region for HCT-116 and SW-480 cells, from which a MF ∼2 × 10⁻⁴ was derived for HCT-116 cells. This approach also indicated that the detection limit of the method is approximately 1:10⁵ mutant:wild type ratio (Fig. 4 E). Therefore, for starting genomic DNA material of 500 ng, which contains about 10⁵ genomic copies, the lowest MF detectable is MF ∼10⁻⁴–10⁻⁵. Similarly conducted dilution experiments indicated that, for the OGG1 gene in DLD-1 cells, the MF observed was MF = 6 × 10⁻³ (average of three independent experiments). Confirmation and identification of mutations in the OGG1 gene of DLD-1 cells was obtained by performing an independent mutation assay that PCR-amplifies the mutated sequence via ligation of a primer at the mutation (24), as described in “Materials and Methods.” Sequencing of the mutated fragment identified a TaqI-forming G→A mutation in the coding region of OGG1 (codon 346; GenBank accession no. Y13277; missense Gly→Arg).

Similar to results obtained with repair-deficient colon cancer cell lines, dHPLC screening of colon tumor, and corresponding normal tissue surgical samples resulted to PCR products present in the tumor and absent in the corresponding normal tissues. In Fig. 5, four representative results from tumor/normal samples 1–4 are depicted, together with dilutions of the positive controls that indicate the expected position of TaqI-forming mutations. By using serial dilutions of the mutation-positive controls, MFs for the 48 gene regions were subsequently derived. In Fig. 6, typical results are plotted for the mismatch repair-deficient cells DLD-1 and HCT-116, as well as for four colon tumor samples. The overall iFLP-identified mutation rates for DLD-1 regions (MF∼2–9 × 10⁻³) are generally higher than for HCT-116 regions (MF∼1–8 × 10⁻⁴), and the two cell lines contain mutations at distinct sequences. In addition, the total number of DLD-1 mutations in the 48 regions examined (sum of each mutation-positive segment × its own MF) is about 10-fold higher than for HCT-116. Bhattacharyya et al. (17) observed a similar relation among MFs in DLD-1 (MF ∼4 × 10⁻³) and HCT-116 (MF ∼ 6 × 10⁻⁴) and found distinct mutation spectra among the two cell lines by using the HPRT phenotypic assay. Therefore the iFLP data indicate that the genomic instability previously observed in HPRT is indicative of “global instability” occurring throughout the DLD-1 and HCT-116 genomes, as hypothesized (11, 12, 17), and that widespread, low-level mutations underline the presence of a mutator phenotype in these cell lines. The difference in repair defects among the two cell lines may explain the differences in the mutation spectra. Mutation spectra depicted in Fig. 6 for the four colon tumor specimens contain regions that appear commonly mutated across most of the samples (e.g., MSH6, PTEN_C), indicating that these may be “hotspots” for the generation of mutations. In general however, the low-level mutational spectrum is unique to the particular tumor sample and cell line. Overall, by examination of 48 gene regions via dHPLC, an extensive pattern of TaqI-forming mutations was observed in seven out of 10 samples, whereas three samples contained no iFLP-detectable mutations (Table 1). These data indicate widespread genomic instability in sporadic colon tumors that manifests as numerous low-level mutations in gene-coding and noncoding regions. To estimate the overall number of diverse low-level mutations that would be discovered if iFLP was applied throughout the genome, the mean number of tumor-specific mutations per region examined (e.g., for tumor 1, 11 mutations in 48 regions, i.e., ∼0.23 mutations per ∼500 bp) is scaled by the size of the human genome (3 × 10⁹ bp), yielding ∼1.4 × 10⁶ distinct low-level mutations. The overall number of single-point mutations is probably larger, because only a fraction of all possible nucleotide changes generate TaqI sites and can be detected by iFLP.

The MSI status of the colon cancer cell lines and the 10 paired colon samples was examined by screening the mononucleotide markers BAT25 and BAT26. Changes in mononucleotide repeat size manifested as significant shifts in the obtained chromatograms (Fig. 7). Samples were classified as MSI-H if both BAT25 and BAT26 were shifted in the tumor relative to the normal colon samples, MSI-L if one of the two markers was shifted, and MSI-S if none of the two markers were shifted. Assessment of MSI status in this manner was recently shown (26) to be in excellent agreement to the results obtained by the Bethesda guidelines using multiple microsatellite markers. Repair-deficient cell lines DLD-1 and HCT116 demonstrated mononucleotide repeat changes for both BAT25 and BAT26 relative to the repair-proficient cell line SW480 (data not shown) and were thus classified as MSI-H. Two of 10 colon tumors, samples 1 and 5, demonstrated significant shifts in the dHPLC chromatograms compared with their normal tissue counterparts on both BAT26 and BAT25 and were classified as MSI-H. The remaining eight colon tumor samples were negative for both microsatellite markers and were classified as MSI-S. Table 1 summarizes the MSI results.

Carcinomas are generally heterogeneous and intrinsically redundant populations of cells generated by transformations including diverse mutations in individual genes and the loss or gain of entire chromosomes (aneuploidy; Ref. 8). To this end, diverse and widespread low-level mutations are expected to have profound implications in the ability of tumors to resist cytotoxic drug treatment (8). For example, specific mutations conferring resistance to treatment by the drug STI-571 in chronic myelogenous leukemia are known to preexist as low-level mutations in the untreated tumor (29, 30, 31, 32). After selection by the drug treatment, some of these low-level mutations become causative, prevalent genetic changes (29, 33, 34, 35). Accordingly, assessment of the extent and diversity of low-level mutations in tumors would be very useful for estimation of the likelihood of drug resistance and tumor progression (8). iFLP provides a methodology for achieving this purpose in surgical specimens, and demonstrates the existence of low-level mutations in sporadic colon cancers at MF ∼10⁻²–10⁻⁴. DNA fingerprinting methods, arbitrarily primed (AP)-PCR, and inter-simple-sequence (SS) PCR, have also been applied to the identification of mutations in colon cancer surgical specimens (5, 36, 37). However, because these methods do not detect mutations at levels below MF ∼10⁻¹–10⁻², low-level mutations are not detected, and the incidence of potential resistance-causing mutations is probably underestimated. This is illustrated in Fig. 8, in which the number of diverse mutations detected in Fig. 6 in the colon cancer samples is replotted versus their MF and indicates that mutations scored decrease rapidly as the selectivity of mutation detection decreases. For example, for tumor 1, a method with a selectivity of 10⁻³ mutant-to-wild type alleles would detect seven mutations, whereas another method with a selectivity of 10⁻² would only detect one mutation. Therefore, to effectively estimate the heterogeneity and number of mutations in tumors, one must detect the diverse low-level mutation load of a surgical specimen, which is directly dependent on the selectivity of the detection method.

Our results show similar numbers of low-level mutations in colorectal cell lines and a substantial fraction of colon tumors and indicate that millions of diverse mutations are present in the majority of sporadic colon carcinomas examined. Although most of these mutations are probably not functional (2), their presence is indicative of an ongoing genomic instability at the single-nucleotide level (SNI), which may induce mutations with the potential to affect tumor development and progression. A similar SNI has been observed in cell lines and transgenic animals using phenotypic mutation assays (38, 39). MSI (21, 40) and chromosomal instability (41) are two established forms of genomic instability in colorectal cancer and other tumors (9, 10). However, many tumors display neither MSI nor chromosomal instability but different forms of genomic instability (42, 43, 44). For example, as demonstrated in Table 1, SNI of the form detected by iFLP often exists in the absence of MSI (38, 39), indicating that different repair defects may be responsible for each type of instability. Unlike MSI that only affects those protein-coding regions containing repeat sequences amenable to instability, SNI may potentially impact every coding sequence, thereby actively influencing processes that lead to tumor formation. Accordingly, it is possible to use iFLP to clarify types and subtypes of surgical specimens with SNI, the coexistence or absence of MSI and chromosomal instability and how early in the tumor formation process, such low-level point mutations build-up. It is also evident from Fig. 6 that, when SNI is significant, a single iFLP experiment can efficiently identify numerous low-level mutations and their frequencies, allowing generation of a mutational signature for each sample. These signatures can potentially be used for matching metastatic lesions with primary tumor sites or for detecting tumor recurrence by tracing the tumor signature in surgical margins or bodily fluids (9).

The described iFLP fills a current need in mutation detection methodology, i.e., the identification of unknown, random point mutations at low MFs. As Fig. 6 indicates, distinct sequences become mutated in DNA depending on the specific repair deficiency leading to instability (45, 46), thus it is difficult to predict a priori where a mutation is likely to occur to apply techniques for known mutation detection. Thus, despite development of RFLP-based methods that detect mutations at specific sequence positions down to a MF of 10⁻⁷ (47, 48), random, low-level mutations generated by mismatch repair-induced genomic instability have proven hard to identify (49). Accordingly, mutation scanning rather than RFLP-based mutation detection at known sequence positions is required. On the other hand, PCR-based mutation scanning methods such as DGGE, dHPLC, and CDCE are ultimately limited by PCR errors (50, 51) and often have to be combined with phenotypic selection systems to acquire the necessary selectivity (52). iFLP is not affected by polymerase errors and allows mutation detection at thousands of sequence positions per experiment combined with the high selectivity of RFLP at each position. The number of sequences examined can be scaled at will by adjusting the number of inverse PCR reactions performed in the last step. By magnifying the effective “target size,” iFLP multiplies the capabilities of RFLP detection by the total number of mutable sites, and a limited mutation scanning occurs that increases the probability of detecting widespread, low-level mutations, i.e., a hallmark of genomic instability (2, 53). It is estimated that about 3–5% of all possible mutations in each sequence can generate the TaqI recognition sequence via a single-nucleotide change and therefore can be identified by iFLP. By replacing TaqI with different restriction enzymes, more sequence changes can be detected. Because the assay can be applied from nanogram amounts of genomic DNA (Fig. 3 A), it should be possible to perform high-throughput mutation scanning from minute biopsies obtained by needle aspiration or laser capture microdissection. By using methylation-sensitive enzymes, it should be possible to adapt iFLP to the discrimination of methylated from nonmethylated DNA sequences. Additional envisioned uses of iFLP include evaluating tissues for carcinogenic exposures that generate random, low-level mutations and identifying genetic variations in pooled DNA from animals or plants.

We acknowledge the assistance of Mohamet Miri and Dr. Frank Haluska in obtaining tissue specimens from the Massachusetts General Hospital Tumor Bank and of the Cooperative Human Tissue Network.