COLD-PCR enhanced melting curve analysis improves diagnostic accuracy for KRAS mutations in colorectal carcinoma

Background KRAS mutational analysis is the standard of care prior to initiation of treatments targeting the epidermal growth factor receptor (EGFR) in patients with metastatic colorectal cancer. Sensitive methods are required to reliably detect KRAS mutations in tumor samples due to admixture with non-mutated cells. Many laboratories have implemented sensitive tests for KRAS mutations, but the methods often require expensive instrumentation and reagents, parallel reactions, multiple steps, or opening PCR tubes. Methods We developed a highly sensitive, single-reaction, closed-tube strategy to detect all clinically significant mutations in KRAS codons 12 and 13 using the Roche LightCycler® instrument. The assay detects mutations via PCR-melting curve analysis with a Cy5.5-labeled sensor probe that straddles codons 12 and 13. Incorporating a fast COLD-PCR cycling program with a critical denaturation temperature (Tc) of 81°C increased the sensitivity of the assay >10-fold for the majority of KRAS mutations. Results We compared the COLD-PCR enhanced melting curve method to melting curve analysis without COLD-PCR and to traditional Sanger sequencing. In a cohort of 61 formalin-fixed paraffin-embedded colorectal cancer specimens, 29/61 were classified as mutant and 28/61 as wild type across all methods. Importantly, 4/61 (6%) were re-classified from wild type to mutant by the more sensitive COLD-PCR melting curve method. These 4 samples were confirmed to harbor clinically-significant KRAS mutations by COLD-PCR DNA sequencing. Five independent mixing studies using mutation-discordant pairs of cell lines and patient specimens demonstrated that the COLD-PCR enhanced melting curve assay could consistently detect down to 1% mutant DNA in a wild type background. Conclusions We have developed and validated an inexpensive, rapid, and highly sensitive clinical assay for KRAS mutations that is the first report of COLD-PCR combined with probe-based melting curve analysis. This assay significantly improved diagnostic accuracy compared to traditional PCR and direct sequencing.

Background KRAS (Kirsten rat sarcoma virus homolog 2) is a membrane-anchored G-protein that acts downstream of the epidermal growth factor receptor (EGFR) to activate pro-growth and anti-apoptotic pathways, including the MAP kinase and PI3 kinase pathways [1]. Mutations in codons 12 and 13 of the KRAS gene confer resistance to drugs targeted at EGFR by impairing GTPase activity, which results in constitutive EGFR-independent signaling. KRAS is one of the most frequently mutated oncogenes in human cancer, with significant mutation rates in common epithelial malignancies such as colon cancer (~40%), lung cancer (~20%), gastric cancer (~10%), and pancreatic cancer (~65%) (COSMIC database; http://www.sanger.ac.uk/genetics/CGP/cosmic/) [2,3].
The role of KRAS mutation status in clinical decision making is best defined for colon cancer. Several large randomized-controlled trials demonstrated no benefit from expensive anti-EGFR drugs such as cetuximab (Erbitux™) in patients with KRAS-mutant colon cancer [4,5]. The National Comprehensive Cancer Network (NCCN) now recommends KRAS testing prior to initiation of anti-EGFR therapy in colon cancer patients [6].
Accumulating evidence also suggests a role for KRAS testing to guide therapy in patients with non-small cell lung cancer [7,8].
CO-amplification at Lower Denaturation temperature (COLD)-PCR is a recently described method to selectively amplify mutant alleles in a wild type background that does not require any additional instrumentation or reagents to implement [9]. Two forms of COLD-PCR are described: fast COLD-PCR and full COLD-PCR (reviewed in [10]). Fast COLD-PCR enriches G:C to A:T mutations that slightly, but predictably lower the melting temperature (T m ) of the PCR amplicon, by using a critical denaturation temperature (T c ) that favors PCR amplification of the mutant allele. Full COLD-PCR theoretically enhances detection of any type of mutation via conditions which promote annealing of WT:mutant pairs and selective denaturation of these heteroduplexes at an empirically determined T c . As the name implies, fast COLD-PCR has the advantage of being more rapid than full COLD-PCR (1-2 hours of instrument time compared to 5-8 hours) and is also easier to troubleshoot and implement in our experience. Fast COLD-PCR is ideal for KRAS codon 12 and 13 mutational analysis because >90% of the mutations are G:C to A:T changes (G12 D, G12V, G12 S, G12C, G13D), and the remaining common mutations are T m -neutral G:C to G:C changes (G12R and G12A).
Here, we report a novel strategy for mutation detection that combines fast COLD-PCR with two-probe melting curve analysis using the LightCycler® instrument. By incorporating COLD-PCR we achieve >10-fold enhancement for the majority of KRAS mutations with a sensitivity of~1% mutant allele in a wild type background.

Patient tumor samples and DNA extraction
All clinical specimens consisted of DNA prepared from formalin-fixed paraffin-embedded (FFPE) tissue from patients with sporadic colorectal carcinoma. De-identified residual clinical specimens were obtained in accordance with the declaration of Helsinki and ethics guidelines of the local institutional review board. We obtained a total of 61 de-identified specimens: 10 DNA specimens from the University of Texas MD Anderson Cancer Center, 4 specimens of unstained FFPE tissue sections with corresponding hematoxylin-eosin (HE) stained slides from the University of Chicago, and 47 DNA specimens prepared from FFPE tissue from colorectal cancer specimens at the University of Washington. HE-stained slides were used as a guide to manually dissect areas of tumor tissue from unstained slide sections. Genomic DNA was prepared with the Gentra Puregene DNA Isolation Kit (Qiagen, Cat. No. 158489).

KRAS mutational detection by Regular-and COLD-PCR enhanced melting curve analysis
Our basic melting-curve strategy to identify KRAS mutations is a modification of a procedure described by Nikiforova and colleagues [11]. The assay is a PCR fluorescence resonance energy transfer (FRET) hybridization probe assay performed with a LightCycler® carousel instrument (Roche Molecular Biochemicals, Manheim Germany). This format requires a pair of primers and two fluorescently labeled oligonucleotide probes and relies on FRET between two probes that hybridize adjacent to one another ( Figure 1). For the detection of KRAS codon 12/13 mutations, the Cy5.5labeled probe spans this region, and the products are distinguished based on their distinct T m , which reflects the thermodynamic stability of the perfectly complementary and mismatched probe-target duplexes. The wild-type T m is 69.5 ± 0.5°C, and the mutant T m is 62.8 ± 1°C. The assay cannot reliably determine the exact base change based on the T m . The primers and probes are given in Table 1 909 339) and run on both LightCycler® 1.5 and LightCy-cler® 2.0 carousel instruments. For the regular PCR melting curve assay (not COLD-PCR enhanced) the cycling program steps were as follows: 1) UNG step to remove carry-over amplicon, 40°C for 10 min, 2) activation of antibody-inhibited Taq polymerase, 94°C for 10 min, 3) cycling 94°C 1 s, 52°C 20 s, 72°C 10 s for 45 total cycles, 4) melting curve with continuous temperature ramping at 0.15°C/s from 45°C to 80°C, and 5) Cooling to 40°C. The COLD-PCR enhanced cycling program is shown in Table 2. An initial 10 cycles of regular PCR are performed to build up product, followed by 40 cycles of COLD-PCR with a critical denaturation temperature (T c ) of 81°C which favors amplification of mutant alleles. We determined the T c empirically by progressively lowering the denaturation temperature from 84°C 82°C 81°C 80°C 79°C. Melting curves were produced in both mutant and wild type samples down to T c = 80°C. We chose the T c at 81°C because the melting curves were difficult to interpret at T c = 80°C due to reduced PCR efficiency and increased signal-to-noise ratio (Additional file 1 Figure S1).

Regular-and COLD-PCR enhanced Sanger DNA sequencing
We used KRAS-F and KRAS-R primers (Table 1) for both PCR amplification and DNA sequencing. For DNA sequencing, regular-or COLD-PCR reactions were performed as described above except that the anchor and sensor probes were not added to the reactions and thermocycling was performed on a Bio-Rad C1000 instrument. PCR products were purified using DNA Clean & Concentrator™-5 (Zymo Research) and eluted in a 30 μl volume of elution buffer (Cat No D4003, D4004, D4013 & D4014). Each 10 μl sequencing reaction included 2 μl Big Dye Terminator v3.1 (Applied Biosystems, ABI) 2 μl 5× sequencing buffer (ABI), 4 pmol of sequencing primer (either KRAS-F or KRAS-R), and 200-300 ng purified PCR product. Sequencing reactions were cleaned up using AutoSeq G-50 Sephadex columns (Amersham Table 1 Primer and Probe Sequences (adapted from [11])  We sequenced a total of 30 out of 61 samples including all 4 discrepant samples that were re-classified as KRAS-mutant by the COLD-PCR melting curve analysis (Additional file 2 Table S1). Two of the four discrepant samples had very low mutant peaks by COLD-PCR sequencing using the same thermocycling conditions (T c = 81°C, Table 2) that produced easily discernable mutant peaks in the COLD-PCR melting curve assay, presumably because the melting curve analysis is inherently more sensitive than sequencing. For this reason, we lowered the T c to 79.4°C to further enhance amplification of minority mutant alleles and confirm the KRAS mutations we detected in the 4 discrepant samples. We selected the T c of 79.4°C after we tried T c ranging from 79-81°C at 0.2°C increments and determined that the T c of 79.4°C was the lowest denaturation temperature in which PCR product from a wild type sample was still visible an agarose gel. We repeated COLD-PCR sequencing with T c = 79.4°C a total of 3 times for each of the 4 discrepant samples, obtaining concordant KRASmutant results and appropriately negative KRAS-wild type controls on each run.

Confirmation of Discrepant Specimens With An Alternative Assay
In addition to COLD-PCR sequencing, we confirmed the 4 discrepant specimens using the Lightmix KRAS mutation detection kit following the manufacturer's instructions (TIB Molbiol, Cat#04-0416-08). This assay uses a locked nucleic acid oligomer competitor and has a stated sensitivity of at least 1% for mutant allele in a wild type background (Additional file 1 Figure S2).

Mixing/Sensitivity Studies
We performed two types of mixing studies to estimate sensitivity: (1) KRAS-mutant colorectal cancer cell lines mixed with KRAS-WT peripheral blood leukocyte DNA, and (2) KRAS-mutant FFPE DNA mixed with KRAS-WT FFPE DNA. We used a total of three KRAS-mutant colorectal cancer cell lines, SW480 (homozygous G12V; GGT to GTT), HCT116 (heterozygous G13D; GGC to GAC) and LS174T (heterozygous G12D; GGT to GAT), and two KRAS-mutant FFPE colorectal cancer samples (one G12 D and one G13D). The undiluted G12 D and G13 D FFPE samples both had~25% KRAS mutant allele based on a visual estimate of tumor nuclear content and on the mutant:WT peak height ratios in regular PCR melting curve analysis (data not shown). To control for potential differences in ploidy between samples and variation in DNA preparations that may confound the interpretation of mixing studies, we first adjusted each DNA preparation so that the cycle thresholds (Ct) of all mutant and WT samples were within 0.5 of each other (PCR-efficiency matched). Mutant cell lines or FFPE DNA samples were mixed in the following ratios (mutant%/WT%): 50/50, 20/80, 10/90, 5/95, 2/98, 1/99. We performed mixing studies in duplicate on a total of 5 unique mutant:WT pairs.

Establishing and Validating COLD-PCR Conditions
To establish the critical denaturation temperature (T c ) for COLD-PCR melting curve analysis we selected a G12C KRAS-mutant FFPE colorectal cancer specimen with~10% mutant allele based on a visual estimate of tumor nuclei and the mutant/WT melting peak height ratio using regular PCR. The mutant/WT peak height ratio increased by an average of 6.3-fold in this "low-positive" patient sample as the T c was lowered from 84°C 82°C 81°C ( Figure 2). COLD-PCR DNA sequencing confirmed a proportional increase in the mutant T base at the first position of codon 12 (G12C = GGT TGT) as the T c was lowered (Figure 2 insets). We found that using a T c lower than 81°C further enhanced the relative proportion of mutant allele, but reduced PCR efficiency to a degree that made it more difficult to interpret melting peaks (Additional file 1 Figure S1).
We next validated the COLD-PCR melting curve assay using FFPE colorectal cancer specimens with distinct characterized KRAS mutations. COLD-PCR enhanced detection of G:C to A:T mutations (G12 D, G13 D, G12C, G12V) between 9.8 and 14.0 fold compared to regular PCR (Summarized in Table 3). Importantly, the COLD-PCR reaction conditions we established did not diminish detection of T m -neutral G:C to G:C mutations (G12R and G13R). Surprisingly, the fast COLD-PCR reaction conditions actually enhanced the detection of a G13R (GGC to CGC) mutation, which we did not expect to enhance based on a T m -neutral G > C change. We hypothesize that this enrichment is due to local sequence effects or to a small amount of WT:mutant heteroduplex formation during the primer annealing step, which favors amplification of mutant alleles by the same mechanism as full COLD-PCR (see Background for discussion of fast vs. full COLD-PCR) [10].

Sensitivity of the COLD-PCR Enhanced Melting Curve Assay
To determine the sensitivity of the COLD-PCR melting curve assay to detect low quantities of mutant allele in WT background we adopted two complementary strategies: (1) we mixed KRAS-mutant cell line DNA with KRAS-WT peripheral blood leukocyte DNA, and (2) KRAS-mutant FFPE tumor DNA with KRAS-WT FFPE tumor DNA at mutant:WT ratios of 50/50, 20/80, 10/90, 5/95, 2/98, and 1/99. Prior to mixing we ensured the PCR-efficiency of each sample in a pair was matched within 0.5 Ct because variations in DNA quality, ploidy, and sample preparation may favor the amplification of one sample over the other and confound the estimation of sensitivity [12]. In a total of 5 unique sample pairings which included G12 D, G13 D, and G12V-mutant DNA, the assay could readily detect KRAS mutations down to 1% mutant allele in each case with COLD-PCR enhancement, but only down to 10% mutant allele with regular PCR (Figure 3).
As a second measure of sensitivity, we ran commercial 20% (G13C) and 1% (G12C) KRAS-mutant control DNA (TIB Molbiol, Lot No. 4160902). The COLD-PCR enhanced melting curve assay correctly classified both controls as KRAS-mutant with a mutant:WT peak height ratio of 3.5 for the 20% G13C control, and 0.3 for the 1% G12C control (Additional file 1 Figure S3).
We next assessed if the COLD-PCR enhanced melting curve assay was sensitive enough to eliminate the need for tumor dissection in a KRAS-mutant sample with a low percentage (~3%) of tumor cells. We confirmed that both regular-PCR melting curve analysis and Sanger sequencing detected a KRAS mutation (G12D) when DNA was prepared with morphology-guided dissection to enrich for tumor cells in our standard fashion. We then tested a DNA sample prepared from the same tumor specimen without dissection (scraping the whole slide). The COLD-PCR enhanced melting curve assay readily detected the KRAS mutation when the whole slide was scraped, whereas the result with regular PCR melting curve analysis was equivocal (Figure 4).

COLD-PCR Enhanced Melting Curve Analysis Detects Mutations That Are Missed By Regular PCR and Direct Sanger Sequencing
We compared the COLD-PCR melting curve assay to regular PCR melting analysis in a cohort of 61 FFPE colorectal cancer specimens which included 14 blinded samples that had KRAS testing done previously at outside institutions. KRAS mutations were detected in   (Table 4; Additional file 2 Table S1). Regular Sanger sequencing did not detect mutations in the 4 samples that were reclassified from KRAS-WT to KRAS-mutant by COLD-PCR, presumably because the mutant allele % was below the limit of detection (sensitivity~20 percent for standard sequencing). However, COLD-PCR enhanced DNA sequencing confirmed that the 4 reclassified samples each harbored a clinically significant KRAS mutation (Additional file 3; Figure 5; see methods section for a detailed discussion of COLD-PCR sequencing). KRAS mutations in the 4 re-classified samples were also confirmed using the TIB Molbiol Lightmix assay, which has a stated sensitivity of at least 1% (Additional file 1 Figure S2). Sequencing of an additional 26 specimens that were concordant by regular-and COLD-PCR confirmed the KRAS melting curve analysis result in all cases (Additional file 2 Table S1). Among the 14 specimens from outside institutions, KRAS results were concordant using the less sensitive regular PCR melting curve analysis and Sanger sequencing (7/14 KRAS-mutant; 7/14 KRAS-WT). However, we re-classified 1/14 of these specimens as KRAS-mutant by both COLD-PCR melting curve analysis and COLD-PCR sequencing. The method used by the outside laboratory to identify KRAS mutations was regular sequencing, which explains why the KRAS mutation in that specimen was not originally detected.

Discussion
Cancer tissue presents challenges for KRAS mutation analysis because each specimen is a unique mixture of tumor and non-tumor cells. A typical formalin-fixed paraffin-embedded colorectal cancer specimen might have between 2-50% tumor cells, which corresponds to 1-25% mutant allele when the tumor is diploid and a heterozygous mutation is present. To enrich the proportion of tumor cells most laboratories (including ours) manually dissect areas of tumor tissue from unstained sections guided by an HE-stained slide. This approach improves mutation detection, but tumors with a diffuse growth pattern or dense inflammatory infiltrate may still have a significant proportion of "contaminating" nontumor DNA after tumor dissection. For this reason, methods such as COLD-PCR and peptide nucleic acid (PNA) clamping have been developed to increase the sensitivity of mutation analysis by selectively amplifying mutant alleles in a wild type background [9,10,13].
Here, we demonstrate that COLD-PCR combined with probe-based melting curve analysis can reproducibly detect down to 1% KRAS-mutant allele in a wild type background. This is comparable to the sensitivity The left panel shows the average and SD (error bars) of the %mutant peak height across the 5 mixing studies using regular-(gray triangles) or COLD-PCR (black squares) conditions. Mutant peak heights were consistently above the limit of detection down to 1% mutant allele using COLD-PCR compared to 10% mutant allele with regular PCR. The right panel shows an example of one of the mixing studies (SW480 cell line; homozygous G12V) used to generate the sensitivity graph in the left panel. Mutant melting peaks are not clearly discernible for 1%, 2%, and 5% KRAS-mutant samples using regular PCR (top right panel), but are readily detected using COLD-PCR (bottom right panel).
reported with PNA clamping methods applied to melting curve analysis [14,15]. We show that our COLD-PCR method can readily detect the 20% and 1% KRASmutant controls used in the nucleic acid clamping-based TIB Molbiol commercial kit for the LightCycler® instrument, with substantially less expensive reagents (~$5 compared to~$100 per reaction). Limitations of our study include the relatively small sample size (n = 61), and lack of validation of the full range of rare KRAS mutations. A larger scale study is required to determine the assay's performance on the full complement of rare KRAS mutations.
A recent editorial on COLD-PCR technology suggested that "The utility of COLD-PCR will be greatly enhanced if this approach can be demonstrated to completely eliminate microdissection steps or at least to substantially reduce the tumor percentage required.... [16]." We show that the COLD-PCR melting curve assay is sensitive enough to readily detect a KRAS mutation without tumor dissection in a sample with only~3% tumor cells. Despite this, we believe it is prudent to continue dissection for tumor enrichment because additional mutational testing may be required on the same  DNA preparation, and these add-on tests may have lower sensitivity. Among a cohort of 61 sporadic colorectal cancer specimens we found that 4/33 (12.1%) specimens harboring KRAS mutations were incorrectly classified as KRAS-WT using regular (conventional) PCR melting curve analysis and Sanger sequencing, despite dissection of samples for tumor enrichment. An additional 10/33 KRAS-mutant specimens were close to the limit of detection using standard melting curve analysis (Table 4 "Lo-Pos"). These findings are supported by recent reports of false negative or equivocal KRAS results when less sensitive methods are used [13,17,18]. One study identified 10/57 (17.5%) additional KRAS-mutant samples when COLD-PCR was incorporated into high resolution melting curve analysis [17]. Kobunai and colleagues detected KRAS mutations in an additional 25 out of 224 colorectal tumors when a highly sensitive PNA-clamping method was compared to direct sequencing [13]. A third group reported equivocal KRAS results in 10% (18/180) of FFPE tumor specimens using a melting curve approach that was similar to our assay without COLD-PCR enhancement [18].
Misclassification of KRAS mutation status has important clinical consequences because KRAS testing of tumor tissue is often used as the sole determinant for selection of anti-EGFR therapy [19,20]. Taken together, our data and the results of other recent studies suggest that more than 10% of patients with KRAS codon 12/13 mutations are incorrectly classified as wild type when less sensitive methods are used. This has implications for patient care and for the interpretation of results from clinical trials that assessed KRAS mutations by direct sequencing, such as the 2008 landmark New  Figure 5 Re-classified specimen. Shown are results from one of the 4 specimens that was re-classified from KRAS-WT to KRAS-mutant by COLD-PCR (sample #3 in Additional file 3). With regular PCR no definite mutant allele is detected by the melting curve assay, or by direct sequencing (left panel). Using COLD-PCR a clear mutant peak is visible in the melting curve assay and confirmed as a G12C mutation by sequencing (c.34 G > C). Arrows point to the position of the mutant melting peak and mutant base pair position. Note that a T c of 79.4°C was used for COLD-PCR sequencing compared to a T c of 81°C for the COLD-PCR melting curve assay (see Methods), which is the likely explanation for the higher proportion of mutant allele observed by COLD-PCR sequencing (bottom right).
England Journal of Medicine study by Karapetis and colleagues [4]. Previous studies have reported COLD-PCR applied to KRAS mutational analysis [9,17,21] and COLD-PCR combined with high-resolution melting curve analysis [17,22], but to our knowledge this is the first report of COLD-PCR combined with fluorescent probe-based melting analysis. Mancini and colleagues applied COLD-PCR to KRAS mutation detection using the Corbett Research RotorGene 6000 high resolution melting instrument [17]. The method described by  utilized a~7 hour full COLD-PCR protocol followed by a second reaction in a pyrosequencing instrument and achieved 1.1 to 4.7-fold enhancement KRAS-mutant signal compared to regular PCR [21]. Milbury and colleagues used single-step and nested PCR conditions followed by high resolution melting curve in a separate instrument [22]. There are important advantages to the novel strategy we describe here which make it appealing for the clinical setting compared to earlier reports. Most importantly, our method is performed as a single~1 hour reaction and is entirely closed-tube, which cuts down substantially on labor and instrument time, while minimizing the risk of sample handling errors and contamination. Additionally, compared to high resolution melting curve analysis, our probe-based method offers theoretical increased specificity afforded by 2 primers and 2 probes (compared to 2 primers only), and better suitability for laboratories that already own or have experience with the Roche LightCycler® instrument.

Conclusion
We report a rapid, single-reaction, closed-tube clinical assay for KRAS mutations that combines COLD-PCR with melting curve analysis using the LightCycler® instrument. The assay has a sensitivity of at least 1%, which is comparable to or better than many commercial kits and previous studies, although methods with superior sensitivity have been reported [22][23][24]. In a cohort of 61 colorectal cancer samples, the assay improved diagnostic accuracy by correctly identifying 4 KRAS-mutant samples that were not detected by regular melting curve analysis or direct sequencing. Incorporation of COLD-PCR with probe-based melting curve analysis was straight-forward to optimize and implement, and is therefore an appealing strategy for assays designed to detect low-level minority alleles.