Efficient Genotyping of KRAS Mutant Non-Small Cell Lung Cancer Using a Multiplexed Droplet Digital PCR Approach

Droplet digital PCR (ddPCR) can be used to detect low frequency mutations in oncogene-driven lung cancer. The range of KRAS point mutations observed in NSCLC necessitates a multiplex approach to efficient mutation detection in circulating DNA. Here we report the design and optimisation of three discriminatory ddPCR multiplex assays investigating nine different KRAS mutations using PrimePCR™ ddPCR™ Mutation Assays and the Bio-Rad QX100 system. Together these mutations account for 95% of the nucleotide changes found in KRAS in human cancer. Multiplex reactions were optimised on genomic DNA extracted from KRAS mutant cell lines and tested on DNA extracted from fixed tumour tissue from a cohort of lung cancer patients without prior knowledge of the specific KRAS genotype. The multiplex ddPCR assays had a limit of detection of better than 1 mutant KRAS molecule in 2,000 wild-type KRAS molecules, which compared favourably with a limit of detection of 1 in 50 for next generation sequencing and 1 in 10 for Sanger sequencing. Multiplex ddPCR assays thus provide a highly efficient methodology to identify KRAS mutations in lung adenocarcinoma.


Introduction
Lung cancer is the leading cause of cancer-related mortality worldwide [1] and over 20 000 cases of non-small cell lung cancer (NSCLC) were diagnosed in the UK in 2012 [2]. The most frequently mutated oncogenes in lung adenocarcinoma are the RAS family GTPases and EGFR (25% and 15% respectively [3]). Knowledge of the molecular profile of advanced lung adenocarcinoma is critical for therapeutic decision-making [4], particularly in the use of EGFR and ALK tyrosine kinase inhibitors [5,6]. The presence of a KRAS mutation may also be of therapeutic relevance if MEK inhibitor and taxane combinations prove efficacious in this patient cohort [7]. Obtaining adequate tumour tissue for conclusive genotyping in lung cancer can be problematic, however [8]. Droplet digital PCR (ddPCR) is a sensitive method of quantitative mutation detection [9,10] that has the potential to accurately genotype patient-derived material from a small amount of starting material.
Detection of KRAS hotspot mutations by ddPCR has been limited by the variety of potential alleles within adjacent loci, although some KRAS mutations occur more commonly in cancer than others (Tables 1 and 2). The four commonest mutations account for 80% of all KRAS nucleotide changes found in human cancers (85% of all changes in NSCLC), while the nine commonest mutations account for 95% of all changes and 97.5% of changes in NSCLC. Fluorophore -labelled digital PCR probes complement a particular mutant DNA sequence and so only detect one specific KRAS mutation. Using these assays in duplex with a probe to detect the mutant allele and a probe to detect the wild-type allele allow mutant allele fraction calculation for a given mutation, but this approach requires potentially multiple assays and the use of more material before correct identification of the genotype. Development of a multiplex assay combining several different mutant probes in the same reaction is therefore an attractive alternative. Multiplex KRAS digital PCR assays have been described using the RainDrop™ Digital PCR system (Rain-Dance Technologies, Billerica, Massachusetts, USA) in advanced colorectal cancer [11] but not as yet with the Bio-Rad QX100 system, an affordable digital PCR system with commerciallyavailable probes for several KRAS mutations, nor has a multiplex tool been used in lung cancer.
We set out to design multiplex digital PCR assays which would accurately identify nine different KRAS mutations and to demonstrate the application of these assays to patient-derived material using the Bio-Rad QX100 system.

Digital PCR analysis
To assess the mutant allele fraction, the concentration of mutant DNA (copies of mutant DNA per droplet) was estimated from the Poisson distribution. Number of mutant copies per droplet Mmu = -ln (1-(nmu/n)), where nmu = number of droplets positive for mutant FAM probe and n = total number of droplets. The DNA concentration in the reaction was estimated as follows: MDNAconc = -ln (1-(nDNAcon/n)), where nDNAconc = number of droplets positive for mutant FAM probe and/or wild-type HEX probe and n = total number of droplets. The mutant allele fraction = Mmu/ MDNAconc. The measured mutant allele frequency describes the mutant allele fraction expressed as a percentage.

Ion torrent proton sequencing
Sequencing libraries were prepared with the Ion AmpliSeq™ Cancer Hotspot Panel v2 (Life Technologies) using the Ion AmpliSeq Library Preparation protocol with 3-5ng of DNA, according to manufacturer's instructions. Following barcoding, libraries were quantified using qPCR and diluted to 100 pM. Libraries were templated with the Ion OneTouch2 system (Life Technologies) and sequenced on a PI chip using the Ion PI OT2 200 Kit (Life Technologies), 520 flows and an average amplicon length of 112 bases to a mean depth of x2721307. The sequencing resulted in 45272-11767856 reads per sample. Ion torrent Variant caller v4.0-r73742 with no Hotspot region and configuration "Germ Line Low Stringency" was used for calling variants. Read counts for all positions were computed using pileup (SAMtools v1.1 [12]) and this data was analysed for possible variants using custom Perl and R scripts. Variants at > 3% reported by both analysis methods and not reported in 1000 Genomes Project database (www.1000genomes.org) were identified as possible somatic mutations. The data was cross referenced against the Cosmic database v70 (cancer. sanger.ac.uk) to identify possible hotspot mutations.

Statistical analysis
Linear regression was performed using the formula r 2 = 1-(SS reg /SS tot ) where SS reg refers to the sum of the squares of distances to best-fit linear regression line and SS tot refers to the sum of the squares of vertical distances from the null hypothesis (y = mean of all y values) using GraphPad Prism version 6.0a (GraphPad Software, La Jolla, California, USA).
KRAS multiplex assay specificity 92 wells of KRAS wild-type gDNA were analysed with each multiplex. The limit of detection was set at 0.05% to reflect our measured results with spiked KRAS mutant gDNA species (S8 Fig). Individual droplets were clustered according to their HEX amplitude into two groups using the kMeans algorithm. A threshold for mutants was estimated based on 1.5 times the median of the FAM values of those drops that belong to the cluster with higher HEX categories. False positive 'mutant' droplets were classified as droplets that were in the lower HEX cluster, but whose FAM value exceeds the threshold to reflect the FAM and HEX amplitudes measured for any of the KRAS mutant species tested for in that multiplex assay. A false positive result was defined as three false positive 'mutant' droplets as per the Rare Mutation Detection Best Practices Guidelines [13]. To assess the specificity, we calculated the binomial probability of achieving fewer than three 'mutant' droplets per 6000 wild-type droplets for each multiplex assay.

KRAS mutation duplex ddPCR
We tested all KRAS ddPCR assays separately across an annealing temperature gradient to optimise thermocycling conditions. Each assay was tested with the appropriate gDNA or oligonucleotide alone and then in duplex with wild-type and KRAS mutant DNA and both relevant FAM and HEX probes present. Decreasing annealing temperature increased FAM amplitude of the mutant probe to a plateau at 54°C for KRAS G12V, D, A, S, R and C and G13C and 13D probes (Fig 1). The KRAS Q61H probe had a further minimal increase in FAM amplitude at 53.4°C as compared to 54°C. All probes tested showed good separation of the four different droplet groups at 54°C, allowing clear identification and quantification of different DNA populations.

KRAS multiplex ddPCR assay design
We designed a digital PCR-based multiplex tool to screen for the most common KRAS mutations in lung adenocarcinoma; G12C, G12D and G12V (http://www.sanger.co.uk/cosmic). Wild-type probe assays for each mutation were tested in combination with varying concentrations of mutant probe assays, giving rise to mutant droplet populations of varying FAM amplitude (Fig 2). The HEX amplitude of all wild-type probe assays was found to be very similar and so the WT for G12C, WT for G12V and WT for G12D assays were selected as references for all subsequent KRAS G12 and G13 mutant assays. Of the different combinations of mutant assays trialled, the multiplex assay which gave best separation of the droplet populations used 900 nM primers and 500 nM G12C probe, 562.5 nM primers and 312.5 nM G12D probe and 225 nM primers and 125 nM G12V probe (Fig 2, top left panel).
The ability to test for multiple KRAS mutations would help to detect sub-clonal populations by applying the multiplex approach to clinical samples on which starting material is scarce. To model this, we employed cell-line derived gDNA for the three mutations and tested it with each probe in duplex to ensure specificity (S1 Fig). In the presence of the G12D mutant probe and KRAS G12D, V and C mutant DNA, a second mutant population of droplets was identified at lower FAM amplitude to the G12D mutant DNA population (red hatched box, left uppermost panel). This population was not observed when each mutant DNA species was tested in duplex with the G12D probe (left second panel). A similar second mutant population was observed with the G12V mutant probe and all three mutant DNA species as compared with each mutant DNA in duplex (left lower panels). This second population, particularly as seen with the G12D mutant probe, may fall at the expected FAM amplitude of a different KRAS mutation in the multiplex assay and lead to false positive mutation detection. To further explore this, FFPE tissue DNA from an individual known to have a KRAS 12/13 mutation (F124), as verified by Cobas 1 testing (Cobas 1 KRAS Mutation Test, Roche Molecular Systems, Inc., Branchburg, NJ, USA) was analysed (S1 Fig, right panels). The multiplex assay identified a mutant droplet population with FAM amplitude that was interpreted as a G12D or G12V mutation. When the sample was tested with each duplex assay individually, no mutant population in either G12V or G12C was observed, whilst a population with a lower FAM amplitude than expected with the G12D mutant assay was observed. On subsequent sequencing and further development of the digital PCR analysis of the tissue DNA, this sample was later identified as KRAS G12A mutant (Table 3). Cross-reactivity of KRAS probe assays on combination Due to the close similarities between the DNA sequences of the various KRAS mutations, significant cross-reactivity was noted between probes designed for mutations within the same region. This was particularly evident with probes designed for substitutions at the same nucleotide i.e. KRAS G12V, D and A and KRAS G12S, R and C. The cross-reactivity of the probe assay with 'mismatched' DNA carrying a different KRAS mutation (S2 and S3 Figs) results in a droplet population of varying FAM and VIC amplitudes relatively close to the empty droplet population and another population close to the true wild-type droplet population. The position of these additional droplet populations dictates the design of multiplex assays to reduce potential misinterpretation of KRAS genotypes. Therefore, three multiplex assays were devised to each combine a probe for a mutation at nucleotide position 35, a probe for a mutation at position 34 and a probe for a mutation at either position 37, 38 or 183.

KRAS multiplex ddPCR assay optimisation
Multiplex A is a combination of FAM assays for KRAS G13C, G12C and G12V. Several different combinations of mutant probe concentrations were tested and the use of a wild-type probe for G12C and G13C at varying concentrations (S4 Fig). There was significant overlap between the HEX amplitude of the wild-type droplet populations and due to the proximity of the relevant nucleotide bases and the estimated length of the probe, it was felt that 450 nM primers and 250 nM G12C, V or D wild-type probe was adequate for quantification of both G12 and G13 wildtype populations. The optimal multiplex assay (S4 Fig, top left panel) comprised 900 nM primers and 500 nM G13C mutant probe, 450 nM primers and 250 nM G12C probe and 225 nM primers and 125 nM G12V probe. Multiplex B is a combination of FAM assays for KRAS G12S, G12D and G13D. Of all the concentrations tested, the optimal combination of mutant probe assays was 675 nM primers and 375 nM G12S probe, 450 nM primers and 250 nM G12D probe and 225 nM primers and 125 nM G13D probe (S5 Fig, top left panel). Multiplex C analyses KRAS mutations at G12R, G12A and Q61H. It requires a wild-type probe for both the G12/13 position and the Q61 position for accurate quantification of mutant allele frequency. The assay was optimised with 450 nM primers and 250 nM G12C probe and 900 nM primers and 500 nM Q61H probe as the wild-type assays. The best separation of the mutant droplet populations was achieved using 675 nM primers and 375 nM G12R probe, 450 nM primers and 250 nM G12A probe and 900 nM primers and 500 nM Q61H mutant probe (S6 Fig, top left panel). All optimised multiplex assays (Fig 3) were tested for cross-reactivity with various species of KRAS mutant DNA (S7 Fig). There is no overlap between the desired droplet populations and the populations for the other KRAS mutant DNA species in all three multiplexes. In addition, the position of droplet populations due to cross-reactivity is highly reproducible and can be used to start to identify which KRAS genotype is present in a multiplex assay containing other mutant probes.

KRAS multiplex ddPCR assay characterisation
Decreasing amounts of NCI-H358 (KRAS G12C) and A549 (KRAS G12S) gDNA were spiked into KRAS wild-type gDNA (NCI-H1975) and tested in the appropriate KRAS multiplex assay. The limit of detection for KRAS G12C mutant DNA in multiplex A was 0.03% and for KRAS G12S DNA in multiplex B was 0.045% (S8 Fig, top panels) The measured mutant allele frequency correlated well with decreasing amounts of spiked KRAS mutant gDNA in multiplexes A and B (r2 = 0.9992 and 0.0998 respectively), demonstrating the linearity of mutation detection in the multiplex assays. Mutant allele frequency showed little intra-well variability for either multiplex assay and high reproducibility between two different operators on three alternate days for a range of allele frequencies (S8 Fig, middle and bottom panels).
Similar KRAS allele frequency was observed using multiplex or duplex assays in both cell line DNA (single species KRAS mutant gDNA and combinations of KRAS mutant gDNA at varying allele frequencies) or oligonucleotides and FFPE tissue DNA (Fig 4; r2 = 0.9302 and 0.9542 respectively).
We analysed multiple wells of NCI-H1975 KRAS wild-type gDNA with each of the three KRAS multiplex assays and calculated the probability of false positive mutation detection, setting the limit of detection at 0.05% to reflect our measured results with spiked KRAS mutant gDNA species (S8 Fig). The specificity of each multiplex is 99.99995%, 99.99857% and 99.85578% for multiplexes A, B and C respectively (S1 Table).

Comparison of KRAS multiplex assay mutant detection with Sanger sequencing
A range of samples containing decreasing amounts of NCI-H358 (KRAS G12C) or A549 (KRAS G12S) gDNA spiked into KRAS wild-type DNA (NCI-H1975) were simultaneously analysed using digital PCR and Sanger sequencing. KRAS G12C mutant DNA was detectable using multiplex A down to a mutant allele frequency of 0.2%, whereas the mutant peak on the chromatogram is only visible at a mutant allele frequency of 31.5% and not below 10% (S9 Fig, top panels). KRAS G12S mutant DNA remains detectable by digital PCR using multiplex B at a mutant allele frequency of 0.1%, but is only visible using Sanger sequencing at 17%. (S9 Fig, lower panels).

Detection of KRAS mutations in FFPE tissue DNA
FFPE tissue DNA extracted from 11 cases of advanced KRAS mutant NSCLC (S2 Table) was analysed using each multiplex assay. The presence of a KRAS 12/13 mutation was known from prior Cobas 1 testing (Cobas 1 KRAS Mutation Test, Roche) but investigators were blinded to the specific KRAS genotype. At least one KRAS mutation was detected in each case and two cases had two detectable KRAS clones (Fig 5). Mutations were confirmed with the appropriate duplex assay. The low frequency G12D mutation in case S011 may represent FFPE artefact due to the frequent deamination of guanine nucleotides during the tissue preservation process [14]. The G12F mutation in S018 was identified by subsequent sequencing of the tissue sample after observation of the cross-reactivity droplet population in all multiplex assays. In addition, two species of KRAS mutant gDNA were spiked into a background of KRAS wild-type DNA (NCI-H1975) gDNA to create three samples containing a major and minor KRAS clone; C1 is  a combination of NCI-H358 (KRAS G12C) and A549 (KRAS G12S) gDNA, C2 is a combination of NCI-H358 (KRAS G12C) and A427 (KRAS G12D) gDNA and C3 is a combination of A427 (KRAS G12D) and A549 (KRAS G12S) gDNA. All KRAS species were detectable using the KRAS multiplex assays, including the minor clones down to a mutant allele frequency of 0.3% measured for the KRAS G12D mutant DNA in sample C3 (Fig 6).  Comparison of KRAS multiplex assay mutant detection with next generation sequencing FFPE tissue DNA from all 11 cases and gDNA from the 3 biclonal cell line DNA samples were sequenced using the Ion Ampliseq Cancer Hotspot Panel v2 on the Ion Proton TM platform (Life Technologies) to analyse the KRAS 12 and 13 loci. Next generation sequencing could not detect KRAS mutant allele frequencies below 4% in tissue DNA and none of the minor clones below 2% were detectable in the biclonal gDNA samples in either of the duplicate library preparations (Tables 3 and 4). The frequency of KRAS mutant alleles that were detectable on next generation sequencing correlate well with those measured in the KRAS multiplex digital PCR assays (Fig 7, r2 = 0.8973).

Discussion
We have used commercially available primer probe assays for nine common KRAS mutations to design and optimize a multiplex assay approach to genotype KRAS mutant cancers using limited material using digital droplet PCR. The Bio-Rad QX200 system has been used to analyse copy number variation in a multiplexed fashion with EvaGreen DNA binding dye [15] but this is the first report of KRAS mutation detection multiplex assays on the QX100 system that are based on varying concentrations of the primer probe assays.
These multiplex assays permit the identification of the specific KRAS genotype and we have observed that this can be deduced from the reproducible droplet populations caused by crossreactivity in any of the multiplex assays as well as the positive mutant droplet populations at the predicted FAM amplitude. These multiplexes can therefore be used to identify the presence of rarer KRAS mutations at the G12/13 and Q61 positions other than those already tested for in the multiplex assays, as demonstrated with case S018.
Sub-clonal mutations can also be detected using this ddPCR approach to KRAS mutation detection. In two cases, S010 and S011, two different KRAS mutations were detected both in multiplex and duplex. We have noted, however, that all the mutations tested in multiplex B, are G>A nucleotide changes. Low frequency single nucleotide changes as an artifact of the tissue preservation process have been observed [16]. The KRAS G12D mutation observed in S011 (mutant allele frequency 0.05%) may therefore be a false positive result. The low frequency KRAS G12C mutation in S010, however, is a G>T change and therefore much more likely to be a true result.
We have demonstrated that the multiplex assays are more sensitive than Sanger and next generation sequencing and highly specific. The assays have linearity down to a limit of Table 4. Detection of KRAS mutant clones in biclonal gDNA samples using next generation sequencing (NGS) and KRAS multiplex digital PCR assays. Two individual library preparations were analysed for each sample. detection of 0.03-0.045%. Due to the false positive signal seen from FFPE tissue on NGS below mutant allele frequencies of 1.5% [17], the limit of detection is at best 1 mutant KRAS molecule in 50 wild-type KRAS molecules. The KRAS G12S minor clone in the C1 gDNA sample that was measured at 1.9% on multiplex digital PCR, for example, was not detectable on NGS in either of two independent library preparations. Sanger sequencing, in comparison, has a limit of detection of only 1 in 10 and over a range of mutant allele frequencies, a sensitivity of 20%. Digital PCR maintains a very high specificity despite this increased sensitivity and is approaching the 100% specificity seen with NGS [17]. The accuracy of KRAS multiplex ddPCR is reflected by the correlation observed with the mutant allele frequency detected by Ion Torrent Proton TM sequencing, an established method of determining this variable. In addition, we have demonstrated the robustness of the multiplex assays across a range of mutant allele frequencies both within the same experiment and between operators on three non-consecutive days.

ID
We have described the combination and optimization of commercially available digital PCR assays to develop novel multiplex assays that allow the accurate detection and discrimination of KRAS genotypes in patient-derived material. Multiplexing of assays reduces the amount of clinical material required for testing, which will be particularly important in settings where the proportion of mutant relative to wild-type DNA is likely to be low, such as with circulating free tumour DNA from plasma samples. The three multiplexed assays described here between them will identify the nucleotide changes in 95% of the cancer cases where the KRAS gene is mutated.  compared with Sanger sequencing of gDNA samples. KRAS G12C mutant DNA samples at allele frequencies ranging from 0.2-31.5% analysed by digital PCR in KRAS multiplex assay A (upper panels) or Sanger sequencing (upper chromatograms). KRAS G12S mutant DNA samples at allele frequencies ranging from 0.1-17% analysed by digital PCR in KRAS multiplex assay B (lower panels) or Sanger sequencing (lower chromatograms). All panels in the upper row show a FAM amplitude up to 10000 and a HEX amplitude up to 6000. Panels in the lower row have a FAM amplitude up to 14000 and a HEX amplitude up to 11000. Key: black-empty droplets, blue-mutant DNA FAM positive droplets, green-wild-type DNA HEX positive droplets, brown-wild-type and mutant DNA double positive droplets; black arrow-mutant nucleotide not discernible from background signal on Sanger sequencing, red arrow-mutant thymine nucleotide detectable using Sanger sequencing, green arrow-mutant adenine nucleotide detectable using Sanger sequencing. (EPS) S1 Table. KRAS multiplex assay specificity. (DOC) S2 Table. Clinical characteristics of 11 patients with lung adenocarcinoma included in the FFPE tissue analysis. Key: F-former smoker, S-smoker, N-never smoker, Unk-unknown. (DOC)