METHODS FOR LUNG CANCER DETECTION

[0001] This application claims the benefit of U.S. Provisional Application Ser. No. 62/323,589, filed Apr. 15, 2016, which is hereby incorporated by reference in its entirety.

[0002] The disclosed inventions relate generally to methods for detecting nucleic acid mutations and fusions using amplification methods such as the polymerase chain reaction (PCR).

[0003] Detection of mutations associated with cancers whether prior to diagnosis, in making a diagnosis, for disease staging or to monitor treatment efficacy has traditionally relied or solid tumor biopsy samples. Such sampling is highly invasive and not without risk of potentially contributing to metastasis or surgical complications. Better and less invasive methods are needed for detecting mutations associated with cancer.

[0004] Provided herein in one embodiment, is a method for determining the single nucleotide variants present in a lung squamous cell carcinoma. The method in this embodiment, includes generating a set of amplicons by performing a multiplex amplification reaction on nucleic acids isolated from a sample of blood or a fraction thereof from an individual suspected of having a lung squamous cell carcinoma, wherein each amplicon of the set of amplicons spans at least one single nucleotide variant loci of a set of single nucleotide variant loci known to be associated with lung cancer; and

[0005] determining the sequence of at least a segment of each amplicon of the set of amplicons, wherein the segment comprises a single nucleotide variant loci, thereby determining the single nucleotide variants present in the squamous cell carcinoma.

[0006] In another embodiment, provided herein is a method for supporting a lung cancer diagnosis for an individual suspected of having lung cancer from a sample of blood or a fraction thereof from the individual. The method includes generating a set of amplicons by performing a multiplex amplification reaction on nucleic acids isolated from the sample, wherein each amplicon of the set of amplicons spans at least one single nucleotide variant loci of a set of single nucleotide variant loci known to be associated with lung cancer; and

[0007] determining the sequence of at least a segment of each amplicon of the set of amplicons, wherein the segment comprises a single nucleotide variant loci, thereby determining whether one or more single nucleotide variants are present in the plurality of single nucleotide variant loci. According to illustrative embodiments,

[0008] the absence of a single nucleotide variant supports a diagnosis of stage 1a, 2a, or 2b adenocarcinoma,

[0009] the presence of a single nucleotide variant supports a diagnosis of squamous cell carcinoma or a stage 2b or 3a adenocarcinoma, and/or

[0010] the presence of 5, 10, 15 or more single nucleotide variants supports a diagnosis of squamous cell carcinoma or a stage 2b or 3 adenocarcinoma.

[0011] In certain embodiments, the presence of 5, 10, or 15 or more single nucleotide variants supports a diagnosis of squamous cell carcinoma or a stage 3 adenocarcinoma

[0012] In illustrative examples of any of the method embodiments provided herein that include an amplification step, the amplification reaction is a PCR reaction, the annealing temperature is between 1 and 15° C. greater than the melting temperature of at least 50, 60, 70, 85, 80, 90, 95, or 100% of the primers of the set of primers, the length of the annealing step in the PCR reaction is between 15 and 60 minutes, the primer concentration in the amplification reaction is between 1 and 10 nM, and the primers in the set of primers, are designed to minimize primer dimer formation.

[0013] In any of the method embodiments of the invention, that include determining or detecting the presence of an SNV using an amplification method, an efficiency and an error rate per cycle can be determined for each amplification reaction of the multiplex amplification reaction of the set of single nucleotide variance loci, and the efficiency and the error rate can be used to determine whether a single nucleotide variant at the set of single variant loci is present in the sample. In some of these exemplary embodiments, a confidence is determined and a SNV call is made if a cutoff confidence value is exceeded, such as 90%, 95%, or 98% confidence.

[0014] In other embodiments, inventive compositions and solid supports are provided herein.

[0015] Other embodiments and features and advantages of the disclosed inventions will be apparent from the following detailed description and from the claims.

[0039] The above-identified figures are provided by way of representation and not limitation.

[0040] Methods and compositions provided herein improve the detection, diagnosis, staging, screening, treatment, and management of lung cancer. Methods provided herein, in illustrative embodiments analyze single nucleotide variant mutations (SNVs) in circulating fluids, especially circulating tumor DNA. The methods provide the advantage of identifying more of the mutations that are found in a tumor and clonal as well as subclonal mutations, in a single test, rather than multiple tests that would be required, if effective at all, that utilize tumor samples. The methods and compositions can be helpful on their own, or they can be helpful when used along with other methods for detection, diagnosis, staging, screening, treatment, and management of lung cancer, for example to help support the results of these other methods to provide more confidence and/or a definitive result.

[0041] Accordingly, provided herein in one embodiment, is a method for determining the single nucleotide variants present in a lung squamous cell carcinoma by determining the single nucleotide variants present in a ctDNA sample from an individual, such as an individual having or suspected of having, squamous cell carcinoma, using a ctDNA SNV amplification/sequencing workflow provided herein.

[0042] In another embodiment, provided herein is a method for detecting lung sqaumous cell carcinoma in a sample of blood or a fraction thereof from an individual, such as an individual suspected of having a cancer, that includes determining the single nucleotide variants present in a sample by determining the single nucleotide variants present in a ctDNA sample using a ctDNA SNV amplification/sequencing workflow provided herein. The presence of 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, or 15 SNVs on the low end of the range, and 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 30, 40, or 50 SNVs on the high end of the range, in the sample at the plurality of single nucleotide loci is indicative of the presence of squamous cell carcinoma.

[0043] In another embodiment, provided herein is a method for detecting a clonal single nucleotide variant in a lung tumor of an individual. The method includes performing a ctDNA SNV amplification/sequencing workflow as provided herein, and determining the variant allele frequency for each of the SNV loci based on the sequence of the plurality of copies of the series of amplicons. A higher relative allele frequency compared to the other single nucleotide variants of the plurality of single nucleotide variant loci is indicative of a clonal single nucleotide variant in the tumor. Variant allele frequencies are well known in the sequencing art. Support for this embodiment, is provided, for example in FIGS. 12-14.

[0044] In certain embodiments, the method further includes determining a treatment plan, therapy and/or administering a compound to the individual that targets the one or more clonal single nucleotide variants. In certain examples, subclonal and/or other clonal SNVs are not targeted by therapy. Specific therapies and associated mutations are provided in other sections of this specification and are known in the art. Accordingly, in certain examples, the method further includes administering a compound to the individual, where the compound is known to be specifically effective in treating lung squamous cell carcinoma having one or more of the determined single nucleotide variants.

[0045] In certain aspects of this embodiment, a variant allele frequency of greater than 0.25%, 0.5%, 0.75%, 1.0%, 5% or 10% is indicative a clonal single nucleotide variant. These cutoffs are supported by the data in tabular form FIG. 20.

[0046] In certain examples of this embodiment, the squamous cell carcinoma is a stage 1a, 1b, or 2a squamous cell carcinoma. In certain examples of this embodiment, the squamous cell carcinoma is a stage 1a or 1b squamous cell carcinoma. In certain examples of the embodiment, the individual is not subjected to surgery. In certain examples of the embodiment, the individual is not subjected to a biopsy.

[0047] In some examples of this embodiment, a clonal SNV is identified or further identified if other testing such as direct tumor testing suggest an on-test SNV is a clonal SNV, for any SNV on test that has a variable allele frequency greater than at least one quarter, one third, one half, or three quarters of the other single nucleotide variants that were determined.

[0048] In some embodiments, methods herein for detecting SNVs in ctDNA can be used instead of direct analysis of DNA from a tumor. Results provided herein demonstrate that SNVs that are much more likely to be clonal SNVs have higher VAFs (See e.g. FIGS. 12-14).

[0049] In certain examples of any of the method embodiments provided herein herein, before a targeted amplification is performed on ctDNA from an individual, data is provided on SNVs that are found in a tumor from the individual. Accordingly, in these embodiments, a SNV amplification/sequencing reaction is performed on one or more tumor samples from the individual. In this methods, the ctDNA SNV amplification/sequencing reaction provided herein is still advantageous because it provides a liquid biopsy of clonal and subclonal mutations. Furthermore, as provided herein, clonal mutations can be more unambiguously identified in an individual that has lung cancer, if a high VAF percentage, for example, more than 1, 2, 3, 4, 5, 6, 7, 8, 9, 10% VAF in a ctDNA sample from the individual is determined for an SNV.

[0050] In certain embodiment, method provided herein can be used to determine whether to isolate and analyze ctDNA from circulating free nucleic acids from an individual with lung cancer. First, it is determined whether the lung cancer is an adenocarcinoma or a squamous cell carcinoma. If the lung cancer is a squamous cell carcinoma circulating free nucleic acids are isolated from individual. The method in some examples, further includes determining the stage of the lung cancer, wherein if the lung cancer is squamous cell carcinoma or stage 3a adenocarcinoma, circulating free nucleic acids are isolated from the individual. Results provided in FIG. 15 and in tabular form in FIG. 20 demonstrate that SNVs are prevalent in squamous cell carcinoma or stage 3a adenocarcinoma, However, SNVs are much less prevalent in earlier stage ADCs. Accordingly, important health care savings can be realized by saving from testing for SNVs in stage 1a, 1b, and/or 2a ADC patients.

[0051] In examples, if the lung cancer is squamous cell carcinoma or stage 3a adenocarcinoma, circulating free nucleic acids are isolated from the individual. Furthermore, in some examples, if

[0052] the lung cancer is stage squamous cell carcinoma or stage 3a adenocarcinoma nucleic acids are not isolated from a lung tumor of the individual.

[0053] In some methods, provided herein are inventive compositions and/or solid supports. F1. A composition comprising circulating tumor nucleic acid fragments comprising a universal adapter, wherein the circulating tumor nucleic acids originated from a lung squamous cell carcinoma tumor.

[0054] In some embodiments, provided herein is an inventive composition that includes circulating tumor nucleic acid fragments comprising a universal adapter, wherein the circulating tumor nucleic acids originated from a sample of blood or a fraction thereof, of an individual with lung squamous cell carcinoma. Results presented in Example 1 demonstrate the surprising advantage of ctDNA SNV amplification/sequencing test methods. These methods typically include formation of ctDNA fragment that include a universal adapter. Furthermore, such methods typically include the formation of a solid support especially a solid support for high throughput sequencing, that includes a plurality of clonal populations of nucleic acids, wherein the clonal populations comprise amplicons generated from a sample of circulating free nucleic acids, wherein the ctDNA. In illustrative embodiments based on the surprising results provided herein, the ctDNA originated from a lung squamous cell carcinoma tumor.

[0055] Similarly, provided herein as an embodiment of the invention is a solid support comprising a plurality of clonal populations of nucleic acids, wherein the clonal populations comprise nucleic acid fragments generated from a sample of circulating free nucleic acids from a sample of blood or a fraction thereof, from an individual with lung squamous cell carcinoma.

[0056] In certain embodiments, the nucleic acid fragments in different clonal populations comprise the same universal adapter. Such a composition is typically formed during a high throughput sequencing reaction in methods of the present invention, as performed in Example 1.

[0057] The clonal populations of nucleic acids can be derived from nucleic acid fragments from a set of samples from two or more individuals. In these embodiments, the nucleic acid fragments comprise one of a series of molecular barcodes corresponding to a sample in the set of samples.

[0058] Detailed analytical methods are provided herein as SNV Methods 1 and SNV Method 2 in the analytical section herein. Any of the methods provided herein can further include analytical steps provided herein. Accordingly, in certain examples, the methods for determining whether a single nucleotide variant is present in the sample, includes identifying a confidence value for each allele determination at each of the set of single nucleotide variance loci, which can be based at least in part on a depth of read for the loci. The confidence limit can be set at least 75%, 80%, 85%, 90%, 95%, 96%, 96%, 98%, or 99%. The confidence limit can be set at different levels for different types of mutations.

[0059] The method can performed with a depth of read for the set of single nucleotide variance loci of at least 5, 10, 15, 20, 25, 50, 100, 150, 200, 250, 500, 1,000, 10,000, 25,000, 50,000, 100,000, 250,000, 500,000, or 1 million. FIG. 20 provides depth of read data for SNV loci successfully analyzed in Example 1.

[0060] In certain embodiments, a method of any of the embodiments herein includes determining an efficiency and/or an error rate per cycle are determined for each amplification reaction of the multiplex amplification reaction of the single nucleotide variance loci. The efficiency and the error rate can then be used to determine whether a single nucleotide variant at the set of single variant loci is present in the sample. More detailed analytical steps provided in SNV Method 2 provided in the analytical method can be included as well, in certain embodiments.

[0061] In illustrative embodiments, of any of the methods herein the set of single nucleotide variance loci includes all of the single nucleotide variance loci identified in the TCGA and COSMIC data sets for lung cancer, or for lung adenocarcinoma and/or especially lung squamous cell carcinoma.

[0062] In certain embodiments of any of the methods herein the set of single nucleotide variant loci include 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 40, 50, 75, 100, 250, 500, 1000, 2500, 5000, or 10,000 single nucleotide variance loci known to be associated with lung cancer, lung ADC, and/or especially lung SCC on the low end of the range, and, 5, 6, 7, 8, 9, 10, 15, 20, 25, 30, 40, 50, 75, 100, 250, 500, 1000, 2500, 5000, 10,000, 20,000 and 25,000 on the high end of the range.

[0063] In any of the methods for detecting SNVs herein that include a ctDNA SNV amplification/sequencing workflow, improved amplification parameters for multiplex PCR can be employed. For example, wherein the amplification reaction is a PCR reaction and the annealing temperature is between 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10° C. greater than the melting temperature on the low end of the range, and 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 or 15° on the high end the range for at least 10, 20, 25, 30, 40, 50, 06, 70, 75, 80, 90, 95 or 100% the primers of the set of primers.

[0064] In certain embodiments, wherein the amplification reaction is a PCR reaction the length of the annealing step in the PCR reaction is between 10, 15, 20, 30, 45, and 60 minutes on the low end of the range, and 15, 20, 30, 45, 60, 120, 180, or 240 minutes on the high end of the range. In certain embodiments, the primer concentration in the amplification, such as the PCR reaction is between 1 and 10 nM. Furthermore, in exemplary embodiments, the primers in the set of primers, are designed to minimize primer dimer formation.

[0065] Accordingly, in an example of any of the methods herein that include an amplification step, the amplification reaction is a PCR reaction, the annealing temperature is between 1 and 10° C. greater than the melting temperature of at least 90% of the primers of the set of primers, the length of the annealing step in the PCR reaction is between 15 and 60 minutes, the primer concentration in the amplification reaction is between 1 and 10 nM, and the primers in the set of primers, are designed to minimize primer dimer formation. In a further aspect of this example, the multiplex amplification reaction is performed under limiting primer conditions.

[0066] In another embodiment, provided herein is a method for supporting a lung cancer diagnosis for an individual, such as an individual suspected of having lung cancer, from a sample of blood or a fraction thereof from the individual, that includes performing a ctDNA SNV amplification/sequencing workflow as provided herein, to determine whether one or more single nucleotide variants are present in the plurality of single nucleotide variant loci. In this embodiment, the following elements, statements, guidelines or rules apply:

[0067] the absence of a single nucleotide variant supports a diagnosis of stage 1a, 1b, or 2a adenocarcinoma,

[0068] the presence of a single nucleotide variant supports a diagnosis of squamous cell carcinoma or a stage 2b or 3a adenocarcinoma, and/or

[0069] the presence of ten or more single nucleotide variants supports a diagnosis of squamous cell carcinoma or a stage 2b or 3 adenocarcinoma.

[0070] The above elements, statements, guidelines or rules are supported by the results of Example 1 (See e.g. the tabular data in FIG. 20). These results identify analysis using a ctDNA SNV amplification/sequencing workflow of lung ADC and SCC samples from an individual as a valuable method for identifying SNVs found in an ADC tumor, especially for stage 2b and 3a ADC tumors, and especially an SCC tumor at any stage (See e.g. FIG. 15 and FIG. 20).

[0071] In certain examples, this embodiment further includes determining the stage of a lung cancer lesion by a non-invasive method, For example, the size of a tumor can be determined by non-invasive methods.

[0072] In certain embodiments, methods herein for detecting SNVs can be used to direct a therapeutic regimen. Therapies are available and under development that target specific mutations associated with ADC and SCC (Nature Review Cancer. 14:535-551 (2014). For example, detection of an EGFR mutation at L858R or T790M can be informative for selecting a therapy. Erlotinib, gefitinib, afatinib, AZK9291, CO-1686, and HM61713 are current therapies approved in the U.S. or in clinical trials, that target specific EGFR mutations. In another example, a G12D, G12C, or G12V mutation in KRAS can be used to direct an individual to a therapy of a combination of Selumetinib plus docetaxel. As another example, a mutation of V600E in BRAF can be used to direct a subject to a treatment of Vemurafenib, dabrafenib, and trametinib.

[0073] A sample analyzed in methods of the present invention, in certain illustrative embodiments, is a blood sample, or a fraction thereof. Methods provided herein, in certain embodiments, are specially adapted for amplifying DNA fragments, especially tumor DNA fragments that are found in circulating tumor DNA (ctDNA). Such fragments are typically about 160 nucleotides in length.

[0074] It is known in the art that cell-free nucleic acid (cfNA), e.g cfDNA, can be released into the circulation via various forms of cell death such as apoptosis, necrosis, autophagy and necroptosis. The cfDNA, is fragmented and the size distribution of the fragments varies from 150-350 bp to >10000 bp. (see Kalnina et al. World J Gastroenterol. 2015 Nov. 7; 21(41): 11636-11653). For example the size distributions of plasma DNA fragments in hepatocellular carcinoma (HCC) patients spanned a range of 100-220 bp in length with a peak in count frequency at about 166 bp and the highest tumor DNA concentration in fragments of 150-180 bp in length (see: Jiang et al. Proc Natl Acad Sci USA 112:E1317-E1325).

[0075] In an illustrative embodiment the circulating tumor DNA (ctDNA) is isolated from blood using EDTA-2Na tube after removal of cellular debris and platelets by centrifugation. The plasma samples can be stored at −80° C. until the DNA is extracted using, for example, QIAamp DNA Mini Kit (Qiagen, Hilden, Germany), (e.g. Hamakawa et al., Br J Cancer. 2015; 112:352-356). Hamakava et al. reported median concentration of extracted cell free DNA of all samples 43.1 ng per ml plasma (range 9.5-1338 ng ml/) and a mutant fraction range of 0.001-77.8%, with a median of 0.90%.

[0076] In certain illustrative embodiments the sample is a tumor. Methods are known in the art for isolating nucleic acid from a tumor and for creating a nucleic acid library from such a DNA sample given the teachings here. Furthermore, given the teachings herein, a skilled artisan will recognize how to create a nucleic acid library appropriate for the methods herein from other samples such as other liquid samples where the DNA is free floating in addition to ctDNA samples.

[0077] Methods of the present invention in certain embodiments, typically include a step of generating and amplifying a nucleic acid library from the sample (i.e. library preparation). The nucleic acids from the sample during the library preparation step can have ligation adapters, often referred to as library tags or ligation adaptor tags (LTs), appended, where the ligation adapters contain a universal priming sequence, followed by a universal amplification. In an embodiment, this may be done using a standard protocol designed to create sequencing libraries after fragmentation. In an embodiment, the DNA sample can be blunt ended, and then an A can be added at the 3′ end. A Y-adaptor with a T-overhang can be added and ligated. In some embodiments, other sticky ends can be used other than an A or T overhang. In some embodiments, other adaptors can be added, for example looped ligation adaptors. In some embodiments, the adaptors may have tag designed for PCR amplification.

[0078] A number of the embodiments provided herein, include detecting the SNVs in a ctDNA sample. Such methods in illustrative embodiments, include an amplification step and a sequencing step (Sometimes referred to herein as a “ctDNA SNV amplification/sequencing workflow). In an illustrative example, a ctDNA amplification/sequencing workflow can include generating a set of amplicons by performing a multiplex amplification reaction on nucleic acids isolated from a sample of blood or a fraction thereof from an individual, such as an individual suspected of having a lung cancer, for example a squamous cell carcinoma, wherein each amplicon of the set of amplicons spans at least one single nucleotide variant loci of a set of single nucleotide variant loci, such as an SNV loci known to be associated with lung cancer; and

[0079] determining the sequence of at least a segment of at each amplicon of the set of amplicons, wherein the segment comprises a single nucleotide variant loci. In this way, this exemplary method determines the single nucleotide variants present in the sample.

[0080] Exemplary ctDNA SNV amplification/sequencing workflows in more detail can include forming an amplification reaction mixture by combining a polymerase, nucleotide triphosphates, nucleic acid fragments from a nucleic acid library generated from the sample, and a set of primers that each binds an effective distance from a single nucleotide variant loci, or a set of primer pairs that each span an effective region that includes a single nucleotide variant loci. The single nucleotide variant loci, in exemplary embodiments, is one known to be associated with lung cancer, for example lung adenocarcinoma and/or in especially illustrative embodiments squamous cell carcinoma. Then, subjecting the amplification reaction mixture to amplification conditions to generate a set of amplicons comprising at least one single nucleotide variant loci of a set of single nucleotide variant loci, preferably known to be associated with lung cancer; and

[0081] determining the sequence of at least a segment of each amplicon of the set of amplicons, wherein the segment comprises a single nucleotide variant loci.

[0082] The effective distance of binding of the primers can be within 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 20, 25, 30, 35, 40, 45, 50, 75, 100, 125, or 150 base pairs of a SNV loci. The effective range that a pair of primers spans typically includes an SNV and is typically 160 base pairs or less, and can be 150, 140, 130, 125, 100, 75, 50 or 25 base pairs or less. In other embodiments, the effective range that a pair of primers spans is 20, 25, 30, 40, 50, 60, 70, 75, 100, 110, 120, 125, 130, 140, or 150 nucleotides from an SNV loci on the low end of the range, and 25, 30, 40, 50, 60, 70, 75, 100, 110, 120, 125, 130, 140, or 150, 160, 170, 175, or 200 on the high end of the range.

[0083] Further details regarding methods of amplification that can be used in a ctDNA SNV amplification/sequencing workflow to detect SNVs for use in methods of the invention are provided in other sections of this specification.

[0084] SNV Calling Analytics

[0085] During performance of the methods provided herein, nucleic acid sequencing data is generated for amplicons created by the tiled multiplex PCR. Algorithm design tools are available that can be used and/or adapted to analyze this data to determine within certain confidence limits, whether a mutation, such as a SNV is present in a target gene, as illustrated in Example 1 herein.

[0086] Sequencing Reads can be demultiplexed using an in-house tool and mapped using the Burrows-Wheeler alignment software, Bwa mem function (BWA, Burrows-Wheeler Alignment Software (see Li H. and Durbin R. (2010) Fast and accurate long-read alignment with Burrows-Wheeler Transform. Bioinformatics, Epub. [PMID: 20080505]) on single end mode using pear merged reads to the hg19 genome. Amplification statistics QC can be performed by analyzing total reads, number of mapped reads, number of mapped reads on target, and number of reads counted.

[0087] In certain embodiments, any analytical method for detecting an SNV from nucleic acid sequencing data detection can be used with methods of the invention methods of the invention that include a step of detecting an SNV or determining whether an SNV is present. In certain illustrative embodiments, methods of the invention that utilize SNV METHOD 1 below are used. In other, even more illustrative embodiments, methods of the invention that include a step of detecting an SNV or determining whether an SNV is present at an SNV loci, utilize SNV METHOD 2 below.

[0088] SNV Method 1:

[0089] For this embodiment, a background error model is constructed using normal plasma samples, which were sequenced on the same sequencing run to account for run-specific artifacts. In certain embodiments, 5, 10, 15, 20, 25, 30, 40, 50, 100, 150, 200, 250, or more than 250 normal plasma samples are analyzed on the same sequencing run. In certain illustrative embodiments, 20, 25, 40, or 50 normal plasma samples are analyzed on the same sequencing run. Noisy positions with normal median variant allele frequency greater than a cutoff are removed. For example this cutoff in certain embodiments is >0.1%, 0.2%, 0.25%, 0.5%, 1%, 2%, 5%, or 10%. In certain illustrative embodiments noisy positions with normal medial variant allele frequency greater than 0.5% are removed. Outlier samples were iteratively removed from the model to account for noise and contamination. In certain embodiments, samples with a Z score of greater than 5, 6, 7, 8, 9, or 10 are removed from the data analysis. For each base substitution of every genomic loci, the depth of read weighted mean and standard deviation of the error are calculated. Tumor or cell-free plasma samples' positions with at least 5 variant reads and a Z-score of 10 against the background error model for example, can be called as a candidate mutation.

[0090] SNV Method 2:

[0091] For this embodiment Single Nucleotide Variants (SNVs) are determined using plasma ctDNA data. The PCR process is modeled as a stochastic process, estimating the parameters using a training set and making the final SNV calls for a separate testing set. The propagation of the error across multiple PCR cycles is determined, and the mean and the variance of the background error are calculated, and in illustrative embodiments, background error is differentiated from real mutations.

[0092] The following parameters are estimated for each base:

[0093] p=efficiency (probability that each read is replicated in each cycle)

[0094] p_e=error rate per cycle for mutation type e (probability that an error of type e occurs)

[0095] X₀=initial number of molecules

[0096] As a read is replicated over the course of PCR process, the more errors occur. Hence, the error profile of the reads is determined by the degrees of separation from the original read. We refer to a read as k^th generation if it has gone through k replications until it has been generated.

[0097] Let us define the following variables for each base:

[0098] X_ij=number of generation i reads generated in the PCR cycle j

[0099] Y_ij=total number of generation i reads at the end of cycle j

[0100] X_ij^e=number of generation i reads with mutation e generated in the PCR cycle j

[0101] Moreover, in addition to normal molecules X₀, if there are additional f_eX₀ molecules with the mutation e at the beginning of the PCR process (hence f_e/(1+fe) will be the fraction of mutated molecules in the initial mixture).

[0102] Given the total number of generation i−1 reads at cycle j−1, the number of generation i reads generated at cycle j has a binomial distribution with a sample size of and probability parameter of p. Hence, E(X_ij, |Y_i-1,j-1, p)=p Y_i-1,j-1 and Var(X_ij, |Y_i-1,j-1, p)=p(1−p) Y_i-1,j-1.

[0103] We also have Y_ij=Σ_k=i^jX_ik. Hence, by recursion, simulation or similar methods, we can determine E(X_ij). Similarly, we can determine Var(X_ij)=E(Var(X_ij, | p))+Var(E(X_ij, | p)) using the distribution of p.

[0104] finally, E(X_ij^e |Y_i-1,j-1, p_e)=p_e Y_i-1,j-1 and Var(X_ij^e |Y_i-1,j-1, p)=p_e (1−p_e) Y_i-1,j-1, and we can use these to compute E(X_ij^e) and Var(X_ij^e).

[0105] In certain embodiments, SNV Method 2 is performed as follows:

[0106] a) Estimate a PCR efficiency and a per cycle error rate using a training data set;

[0107] b) Estimate a number of starting molecules for the testing data set at each base using the distribution of the efficiency estimated in step (a);

[0108] c) If needed, update the estimate of the efficiency for the testing data set using the starting number of molecules estimated in step (b);

[0109] d) Estimate the mean and variance for the total number of molecules, background error molecules and real mutation molecules (for a search space consisting of an initial percentage of real mutation molecules) using testing set data and parameters estimated in steps (a), (b) and (c);

[0110] e) Fit a distribution to the number of total error molecules (background error and real mutation) in the total molecules, and calculate the likelihood for each real mutation percentage in the search space; and

[0111] f) Determine the most likely real mutation percentage and calculate the confidence using the data from in step (e).

[0112] A confidence cutoff can be used to identify an SNV at an SNV loci. For example, a 90%, 95%, 96%, 97%, 98%, or 99% confidence cutoff can be used to call an SNV.

[0113] Exemplary SNV METHOD 2 Algorithm

[0114] The algorithm starts by estimating the efficiency and error rate per cycle using the training set. Let n denote the total number of PCR cycles.

[0115] The number of reads R_b at each base b can be approximated by (1+p_b)ⁿX₀, where p_b is the efficiency at base b. Then (R_b/X₀)^1/n can be used to approximate 1+p_b. Then, we can determine the mean and the standard variation of p_b across all training samples, to estimate the parameters of the probability distribution (such as normal, beta, or similar distributions) for each base.

[0116] Similarly the number of error e reads R_b^e at each base b can be used to estimate p_e. After determining the mean and the standard deviation of the error rate across all training samples, we approximate its probability distribution (such as normal, beta, or similar distributions) whose parameters are estimated using this mean and standard deviation values.

[0117] Next, for the testing data, we estimate the initial starting copy at each base as

${\int }_0^1\frac{R_b}{{\left(1+p_b\right)}^n}f\left(p_b\right)dp_b$

where f(.) is an estimated distribution from the training set.

${\int }_0^1\frac{R_b}{{\left(1+p_b\right)}^n}f\left(p_b\right)dp_b$

where f(.) is an estimated distribution from the training set.

[0118] Hence, we have estimated the parameters that will be used in the stochastic process. Then, by using these estimates, we can estimate the mean and the variance of the molecules created at each cycle (note that we do this separately for normal molecules, error molecules, and mutation molecules).

[0119] Finally, by using a probabilistic method (such as maximum likelihood or similar methods), we can determine the best f_e value that fits the distribution of the error, mutation, and normal molecules the best. More specifically, we estimate the expected ratio of the error molecules to total molecules for various f_e values in the final reads, and determine the likelihood of our data for each of these values, and then select the value with the highest likelihood.

[0120] Primer tails can improve the detection of fragmented DNA from universally tagged libraries. If the library tag and the primer-tails contain a homologous sequence, hybridization can be improved (for example, melting temperature (Tm) is lowered) and primers can be extended if only a portion of the primer target sequence is in the sample DNA fragment. In some embodiments, 13 or more target specific base pairs may be used. In some embodiments, 10 to 12 target specific base pairs may be used. In some embodiments, 8 to 9 target specific base pairs may be used. In some embodiments, 6 to 7 target specific base pairs may be used.

[0121] In one embodiment, Libraries are generated from the samples above by ligating adaptors to the ends of DNA fragments in the samples, or to the ends of DNA fragments generated from DNA isolated from the samples. The fragments can then be amplified using PCR, for example, according to the following exemplary protocol:

[0122] 95° C., 2 min; 15×[95° C., 20 sec, 55° C., 20 sec, 68° C., 20 sec], 68° C. 2 min, 4° C. hold.

[0123] Many kits and methods are known in the art for generation of libraries of nucleic acids that include universal primer binding sites for subsequent amplification, for example clonal amplification, and for subsequence sequencing. To help facilitate ligation of adapters library preparation and amplification can include end repair and adenylation (i.e. A-tailing). Kits especially adapted for preparing libraries from small nucleic acid fragments, especially circulating free DNA, can be useful for practicing methods provided herein. For example, the NEXTflex Cell Free kits available from Bioo Scientific ( ) or the Natera Library Prep Kit (available from Natera, Inc. San Carlos, Calif.). However, such kits would typically be modified to include adaptors that are customized for the amplification and sequencing steps of the methods provided herein. Adaptor ligation can be performed using commercially available kits such as the ligation kit found in the AGILENT SURESELECT kit (Agilent, CA).

[0124] Target regions of the nucleic acid library generated from DNA isolated from the sample, especially a circulating free DNA sample for the methods of the present invention, are then amplified. For this amplification, a series of primers or primer pairs, which can include between 5, 10, 15, 20, 25, 50, 100, 125, 150, 250, 500, 1000, 2500, 5000, 10,000, 20,000, 25,000, or 50,000 on the low end of the range and 15, 20, 25, 50, 100, 125, 150, 250, 500, 1000, 2500, 5000, 10,000, 20,000, 25,000, 50,000, 60,000, 75,000, or 100,000 primers on the upper end of the range, that each bind to one of a series of primer binding sites.

[0125] Primer designs can be generated with Primer3 (Untergrasser A, Cutcutache I, Koressaar T, Ye J, Faircloth B C, Remm M, Rozen S G (2012) “Primer3—new capabilities and interfaces.” Nucleic Acids Research 40(15):e115 and Koressaar T, Remm M (2007) “Enhancements and modifications of primer design program Primer3.” Bioinformatics 23(10):1289-91) source code available at primer3.sourceforge.net). Primer specificity can be evaluated by BLAST and added to existing primer design pipeline criteria:

[0126] Primer specificities can be determined using the BLASTn program from the ncbi-blast-2.2.29+package. The task option “blastn-short” can be used to map the primers against hg19 human genome. Primer designs can be determined as “specific” if the primer has less than 100 hits to the genome and the top hit is the target complementary primer binding region of the genome and is at least two scores higher than other hits (score is defined by BLASTn program). This can be done in order to have a unique hit to the genome and to not have many other hits throughout the genome.

[0127] The final selected primers can be visualized in IGV (James T. Robinson, Helga Thorvaldsdóttir, Wendy Winckler, Mitchell Guttman, Eric S. Lander, Gad Getz, Jill P. Mesirov. Integrative Genomics Viewer. Nature Biotechnology 29, 24-26 (2011)) and UCSC browser (Kent W J, Sugnet C W, Furey T S, Roskin K M, Pringle T H, Zahler A M, Haussler D. The human genome browser at UCSC. Genome Res. 2002 June; 12(6):996-1006) using bed files and coverage maps for validation.

[0128] Methods of the present invention, in certain embodiments, include forming an amplification reaction mixture. The reaction mixture typically is formed by combining a polymerase, nucleotide triphosphates, nucleic acid fragments from a nucleic acid library generated from the sample, a set of forward and reverse primers specific for target regions that contain SNVs. The reaction mixtures provided herein, themselves forming in illustrative embodiments, a separate aspect of the invention.

[0129] An amplification reaction mixture useful for the present invention includes components known in the art for nucleic acid amplification, especially for PCR amplification. For example, the reaction mixture typically includes nucleotide triphosphates, a polymerase, and magnesium. Polymerases that are useful for the present invention can include any polymerase that can be used in an amplification reaction especially those that are useful in PCR reactions. In certain embodiments, hot start Taq polymerases are especially useful. Amplification reaction mixtures useful for practicing the methods provided herein, such as AmpliTaq Gold master mix (Life Technologies, Carlsbad, Calif.), are available commercially.

[0130] Amplification (e.g. temperature cycling) conditions for PCR are well known in the art. The methods provided herein can include any PCR cycling conditions that result in amplification of target nucleic acids such as target nucleic acids from a library. Non-limiting exemplary cycling conditions are provided in the Examples section herein.

[0131] There are many workflows that are possible when conducting PCR; some workflows typical to the methods disclosed herein are provided herein. The steps outlined herein are not meant to exclude other possible steps nor does it imply that any of the steps described herein are required for the method to work properly. A large number of parameter variations or other modifications are known in the literature, and may be made without affecting the essence of the invention.

[0132] In certain embodiments of the method provided herein, at least a portion and in illustrative examples the entire sequence of an amplicon, such as an outer primer target amplicon, is determined. Methods for determining the sequence of an amplicon are known in the art. Any of the sequencing methods known in the art, e.g. Sanger sequencing, can be used for such sequence determination. In illustrative embodiments high throughput next-generation sequencing techniques (also referred to herein as massively parallel sequencing techniques) such as, but not limited to, those employed in MYSEQ (ILLUMINA), HISEQ (ILLUMINA), ION TORRENT (LIFE TECHNOLOGIES), GENOME ANALYZER ILX (ILLUMINA), GS FLEX+(ROCHE 454), can be used for sequencing the amplicons produced by the methods provided herein.

[0133] High throughput genetic sequencers are amenable to the use of barcoding (i.e., sample tagging with distinctive nucleic acid sequences) so as to identify specific samples from individuals thereby permitting the simultaneous analysis of multiple samples in a single run of the DNA sequencer. The number of times a given region of the genome in a library preparation (or other nucleic preparation of interest) is sequenced (number of reads) will be proportional to the number of copies of that sequence in the genome of interest (or expression level in the case of cDNA containing preparations). Biases in amplification efficiency can be taken into account in such quantitative determination.

[0134] Target Genes

[0135] Target genes of the present invention in exemplary embodiments, are cancer-related genes, and in many illustrative embodiments, lung cancer-related genes. A cancer-related gene (for example, a lung cancer-related gene or a lung SCC-related gene or a lung ADC-related gene) refers to a gene associated with an altered risk for a cancer (e.g. lung cancer or lung SCC or lung ADC, respectively) or an altered prognosis for a cancer. Exemplary cancer-related genes that promote cancer include oncogenes; genes that enhance cell proliferation, invasion, or metastasis; genes that inhibit apoptosis; and pro-angiogenesis genes. Cancer-related genes that inhibit cancer include, but are not limited to, tumor suppressor genes; genes that inhibit cell proliferation, invasion, or metastasis; genes that promote apoptosis; and anti-angiogenesis genes.

[0136] An embodiment of the mutation detection method begins with the selection of the region of the gene that becomes the target. The region with known mutations is used to develop primers for mPCR-NGS to amplify and detect the mutation.

[0137] Methods provided herein can be used to detect virtually any type of mutation, especially mutations known to be associated with cancer and most particularly the methods provided herein are directed to mutations, especially SNVs, associated with lung cancer, specifically adenocarcinoma and squamous cell carcinoma. Exemplary SNVs can be in one or more of the following genes: EGFR, FGFR1, FGFR2, ALK, MET, ROS1, NTRK1, RET, HER2, DDR2, PDGFRA, KRAS, NF1, BRAF, PIK3CA, MEK1, NOTCH1, MLL2, EZH2, TET2, DNMT3A, SOX2, MYC, KEAP1, CDKN2A, NRG1, TP53, LKB1, and PTEN, which have been identified in various lung cancer samples as being mutated, having increased copy numbers, or being fused to other genes and combinations thereof (Non-small-cell lung cancers: a heterogeneous set of diseases. Chen et al. Nat. Rev. Cancer. 2014 Aug. 14(8):535-551). In another example, the list of genes are those listed above, where SNVs have been reported, such as in the cited Chen et al. reference. In another embodiment, the SNVs can include SNVs found in one of the genes found in Table 19 herein. SNVs in the genes listed in Table 19 were analyzed in the experiment of Example 1. SNVs in these genes were detected in tumor samples matched to the ctDNA samples of Example 1. In some embodiments, SNVs that are analyzed in methods provided herein can include any of the genes listed in this paragraph above or any of the genes in Table 19 that are not listed above. Provided herein, are methods that use the specific determination of a particular SNV in a particular gene to direct a targeted drug therapy.

[0138] Amplification (e.g. PCR) Reaction Mixtures:

[0139] Methods of the present invention, in certain embodiments, include forming an amplification reaction mixture. The reaction mixture typically is formed by combining a polymerase, nucleotide triphosphates, nucleic acid fragments from a nucleic acid library generated from the sample, a series of forward target-specific outer primers and a first strand reverse outer universal primer. Another illustrative embodiment is a reaction mixture that includes forward target-specific inner primers instead of the forward target-specific outer primers and amplicons from a first PCR reaction using the outer primers, instead of nucleic acid fragments from the nucleic acid library. The reaction mixtures provided herein, themselves forming in illustrative embodiments, a separate aspect of the invention. In illustrative embodiments, the reaction mixtures are PCR reaction mixtures. PCR reaction mixtures typically include magnesium.

[0140] In some embodiments, the reaction mixture includes ethylenediaminetetraacetic acid (EDTA), magnesium, tetramethyl ammonium chloride (TMAC), or any combination thereof. In some embodiments, the concentration of TMAC is between 20 and 70 mM, inclusive. While not meant to be bound to any particular theory, it is believed that TMAC binds to DNA, stabilizes duplexes, increases primer specificity, and/or equalizes the melting temperatures of different primers. In some embodiments, TMAC increases the uniformity in the amount of amplified products for the different targets. In some embodiments, the concentration of magnesium (such as magnesium from magnesium chloride) is between 1 and 8 mM.

[0141] The large number of primers used for multiplex PCR of a large number of targets may chelate a lot of the magnesium (2 phosphates in the primers chelate 1 magnesium). For example, if enough primers are used such that the concentration of phosphate from the primers is −9 mM, then the primers may reduce the effective magnesium concentration by ˜4.5 mM. In some embodiments, EDTA is used to decrease the amount of magnesium available as a cofactor for the polymerase since high concentrations of magnesium can result in PCR errors, such as amplification of non-target loci. In some embodiments, the concentration of EDTA reduces the amount of available magnesium to between 1 and 5 mM (such as between 3 and 5 mM).

[0142] In some embodiments, the pH is between 7.5 and 8.5, such as between 7.5 and 8, 8 and 8.3, or 8.3 and 8.5, inclusive. In some embodiments, Tris is used at, for example, a concentration of between 10 and 100 mM, such as between 10 and 25 mM, 25 and 50 mM, 50 and 75 mM, or 25 and 75 mM, inclusive. In some embodiments, any of these concentrations of Tris are used at a pH between 7.5 and 8.5. In some embodiments, a combination of KCl and (NH₄)₂SO₄ is used, such as between 50 and 150 mM KCl and between 10 and 90 mM (NH₄)₂SO₄, inclusive. In some embodiments, the concentration of KCl is between 0 and 30 mM, between 50 and 100 mM, or between 100 and 150 mM, inclusive. In some embodiments, the concentration of (NH₄)₂SO₄ is between 10 and 50 mM, 50 and 90 mM, 10 and 20 mM, 20 and 40 mM, 40 and 60 mM, or 60 and 80 mM (NH₄)₂SO₄, inclusive. In some embodiments, the ammonium [NH₄⁺] concentration is between 0 and 160 mM, such as between 0 to 50, 50 to 100, or 100 to 160 mM, inclusive. In some embodiments, the sum of the potassium and ammonium concentration ([K⁺]+[NH₄⁺]) is between 0 and 160 mM, such as between 0 to 25, 25 to 50, 50 to 150, 50 to 75, 75 to 100, 100 to 125, or 125 to 160 mM, inclusive. An exemplary buffer with [K⁺]+[NH₄⁺]=120 mM is 20 mM KCl and 50 mM (NH₄)₂SO₄. In some embodiments, the buffer includes 25 to 75 mM Tris, pH 7.2 to 8, 0 to 50 mM KCl, 10 to 80 mM ammonium sulfate, and 3 to 6 mM magnesium, inclusive. In some embodiments, the buffer includes 25 to 75 mM Tris pH 7 to 8.5, 3 to 6 mM MgCl₂, 10 to 50 mM KCl, and 20 to 80 mM (NH₄)₂SO₄, inclusive. In some embodiments, 100 to 200 Units/mL of polymerase are used. In some embodiments, 100 mM KCl, 50 mM (NH₄)₂SO_{4, 3} mM MgCl₂, 7.5 nM of each primer in the library, 50 mM TMAC, and 7 ul DNA template in a 20 ul final volume at pH 8.1 is used.

[0143] In some embodiments, a crowding agent is used, such as polyethylene glycol (PEG, such as PEG 8,000) or glycerol. In some embodiments, the amount of PEG (such as PEG 8,000) is between 0.1 to 20%, such as between 0.5 to 15%, 1 to 10%, 2 to 8%, or 4 to 8%, inclusive. In some embodiments, the amount of glycerol is between 0.1 to 20%, such as between 0.5 to 15%, 1 to 10%, 2 to 8%, or 4 to 8%, inclusive. In some embodiments, a crowding agent allows either a low polymerase concentration and/or a shorter annealing time to be used. In some embodiments, a crowding agent improves the uniformity of the DOR and/or reduces dropouts (undetected alleles). Polymerases In some embodiments, a polymerase with proof-reading activity, a polymerase without (or with negligible) proof-reading activity, or a mixture of a polymerase with proof-reading activity and a polymerase without (or with negligible) proof-reading activity is used. In some embodiments, a hot start polymerase, a non-hot start polymerase, or a mixture of a hot start polymerase and a non-hot start polymerase is used. In some embodiments, a HotStarTaq DNA polymerase is used (see, for example, QIAGEN catalog No. 203203). In some embodiments, AmpliTaq Gold® DNA Polymerase is used. In some embodiments a PrimeSTAR GXL DNA polymerase, a high fidelity polymerase that provides efficient PCR amplification when there is excess template in the reaction mixture, and when amplifying long products, is used (Takara Clontech, Mountain View, Calif.). In some embodiments, KAPA Taq DNA Polymerase or KAPA Taq HotStart DNA Polymerase is used; they are based on the single-subunit, wild-type Taq DNA polymerase of the thermophilic bacterium Thermus aquaticus. KAPA Taq and KAPA Taq HotStart DNA Polymerase have 5′-3′ polymerase and 5′-3′ exonuclease activities, but no 3′ to 5′ exonuclease (proofreading) activity (see, for example, KAPA BIOSYSTEMS catalog No. BK1000). In some embodiments, Pfu DNA polymerase is used; it is a highly thermostable DNA polymerase from the hyperthermophilic archaeum Pyrococcus furiosus. The enzyme catalyzes the template-dependent polymerization of nucleotides into duplex DNA in the 5′→3′ direction. Pfu DNA Polymerase also exhibits 3′→5′ exonuclease (proofreading) activity that enables the polymerase to correct nucleotide incorporation errors. It has no 5′→3′ exonuclease activity (see, for example, Thermo Scientific catalog No. EP0501). In some embodiments Klentaq1 is used; it is a Klenow-fragment analog of Taq DNA polymerase, it has no exonuclease or endonuclease activity (see, for example, DNA POLYMERASE TECHNOLOGY, Inc, St. Louis, Mo., catalog No. 100). In some embodiments, the polymerase is a PHUSION DNA polymerase, such as PHUSION High Fidelity DNA polymerase (M0530S, New England BioLabs, Inc.) or PHUSION Hot Start Flex DNA polymerase (M0535S, New England BioLabs, Inc.). In some embodiments, the polymerase is a Q5® DNA Polymerase, such as Q5® High-Fidelity DNA Polymerase (M0491S, New England BioLabs, Inc.) or Q5® Hot Start High-Fidelity DNA Polymerase (M0493S, New England BioLabs, Inc.). In some embodiments, the polymerase is a T4 DNA polymerase (M0203S, New England BioLabs, Inc.).

[0144] In some embodiment, between 5 and 600 Units/mL (Units per 1 mL of reaction volume) of polymerase is used, such as between 5 to 100, 100 to 200, 200 to 300, 300 to 400, 400 to 500, or 500 to 600 Units/mL, inclusive.

[0145] In some embodiments, hot-start PCR is used to reduce or prevent polymerization prior to PCR thermocycling. Exemplary hot-start PCR methods include initial inhibition of the DNA polymerase, or physical separation of reaction components reaction until the reaction mixture reaches the higher temperatures. In some embodiments, slow release of magnesium is used. DNA polymerase requires magnesium ions for activity, so the magnesium is chemically separated from the reaction by binding to a chemical compound, and is released into the solution only at high temperature. In some embodiments, non-covalent binding of an inhibitor is used. In this method a peptide, antibody, or aptamer are non-covalently bound to the enzyme at low temperature and inhibit its activity. After incubation at elevated temperature, the inhibitor is released and the reaction starts. In some embodiments, a cold-sensitive Taq polymerase is used, such as a modified DNA polymerase with almost no activity at low temperature. In some embodiments, chemical modification is used. In this method, a molecule is covalently bound to the side chain of an amino acid in the active site of the DNA polymerase. The molecule is released from the enzyme by incubation of the reaction mixture at elevated temperature. Once the molecule is released, the enzyme is activated.

[0146] In some embodiments, the amount to template nucleic acids (such as an RNA or DNA sample) is between 20 and 5,000 ng, such as between 20 to 200, 200 to 400, 400 to 600, 600 to 1,000; 1,000 to 1,500; or 2,000 to 3,000 ng, inclusive.

[0147] In some embodiments a QIAGEN Multiplex PCR Kit is used (QIAGEN catalog No. 206143). For 100×50 μl multiplex PCR reactions, the kit includes 2×QIAGEN Multiplex PCR Master Mix (providing a final concentration of 3 mM MgCl₂, 3×0.85 ml), 5×Q-Solution (1×2.0 ml), and RNase-Free Water (2×1.7 ml). The QIAGEN Multiplex PCR Master Mix (MM) contains a combination of KCl and (NH₄)₂SO₄ as well as the PCR additive, Factor MP, which increases the local concentration of primers at the template. Factor MP stabilizes specifically bound primers, allowing efficient primer extension by HotStarTaq DNA Polymerase. HotStarTaq DNA Polymerase is a modified form of Taq DNA polymerase and has no polymerase activity at ambient temperatures. In some embodiments, HotStarTaq DNA Polymerase is activated by a 15-minute incubation at 95° C. which can be incorporated into any existing thermal-cycler program.

[0148] In some embodiments, 1×QIAGEN MM final concentration (the recommended concentration), 7.5 nM of each primer in the library, 50 mM TMAC, and 7 ul DNA template in a 20 ul final volume is used. In some embodiments, the PCR thermocycling conditions include 95° C. for 10 minutes (hot start); 20 cycles of 96° C. for 30 seconds; 65° C. for 15 minutes; and 72° C. for 30 seconds; followed by 72° C. for 2 minutes (final extension); and then a 4° C. hold.

[0149] In some embodiments, 2×QIAGEN MM final concentration (twice the recommended concentration), 2 nM of each primer in the library, 70 mM TMAC, and 7 ul DNA template in a 20 ul total volume is used. In some embodiments, up to 4 mM EDTA is also included. In some embodiments, the PCR thermocycling conditions include 95° C. for 10 minutes (hot start); 25 cycles of 96° C. for 30 seconds; 65° C. for 20, 25, 30, 45, 60, 120, or 180 minutes; and optionally 72° C. for 30 seconds); followed by 72° C. for 2 minutes (final extension); and then a 4° C. hold.

[0150] Another exemplary set of conditions includes a semi-nested PCR approach. The first PCR reaction uses 20 ul a reaction volume with 2×QIAGEN MM final concentration, 1.875 nM of each primer in the library (outer forward and reverse primers), and DNA template. Thermocycling parameters include 95° C. for 10 minutes; 25 cycles of 96° C. for 30 seconds, 65° C. for 1 minute, 58° C. for 6 minutes, 60° C. for 8 minutes, 65° C. for 4 minutes, and 72° C. for 30 seconds; and then 72° C. for 2 minutes, and then a 4° C. hold. Next, 2 ul of the resulting product, diluted 1:200, is used as input in a second PCR reaction. This reaction uses a 10 ul reaction volume with 1×QIAGEN MM final concentration, 20 nM of each inner forward primer, and 1 uM of reverse primer tag. Thermocycling parameters include 95° C. for 10 minutes; 15 cycles of 95° C. for 30 seconds, 65° C. for 1 minute, 60° C. for 5 minutes, 65° C. for 5 minutes, and 72° C. for 30 seconds; and then 72° C. for 2 minutes, and then a 4° C. hold. The annealing temperature can optionally be higher than the melting temperatures of some or all of the primers, as discussed herein (see U.S. patent application Ser. No. 14/918,544, filed Oct. 20, 2015, which is herein incorporated by reference in its entirety).

[0151] The melting temperature (T_m) is the temperature at which one-half (50%) of a DNA duplex of an oligonucleotide (such as a primer) and its perfect complement dissociates and becomes single strand DNA. The annealing temperature (TA) is the temperature one runs the PCR protocol at. For prior methods, it is usually 5° C. below the lowest T_m of the primers used, thus close to all possible duplexes are formed (such that essentially all the primer molecules bind the template nucleic acid). While this is highly efficient, at lower temperatures there are more unspecific reactions bound to occur. One consequence of having too low a TA is that primers may anneal to sequences other than the true target, as internal single-base mismatches or partial annealing may be tolerated. In some embodiments of the present inventions, the TA is higher than T_m, where at a given moment only a small fraction of the targets have a primer annealed (such as only ˜1-5%). If these get extended, they are removed from the equilibrium of annealing and dissociating primers and target (as extension increases T_m quickly to above 70° C.), and a new ˜1-5% of targets has primers. Thus, by giving the reaction a long time for annealing, one can get ˜100% of the targets copied per cycle.

[0152] In various embodiments, the annealing temperature is between 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13° C. and 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, or 15° C. on the high end of the range, greater than the melting temperature (such as the empirically measured or calculated T_m) of at least 25, 50, 60, 70, 75, 80, 90, 95, or 100% of the non-identical primers. In various embodiments, the annealing temperature is between 1 and 15° C. (such as between 1 to 10, 1 to 5, 1 to 3, 3 to 5, 5 to 10, 5 to 8, 8 to 10, 10 to 12, or 12 to 15° C., inclusive) greater than the melting temperature (such as the empirically measured or calculated T_m) of at least 25; 50; 75; 100; 300; 500; 750; 1,000; 2,000; 5,000; 7,500; 10,000; 15,000; 19,000; 20,000; 25,000; 27,000; 28,000; 30,000; 40,000; 50,000; 75,000; 100,000; or all of the non-identical primers. In various embodiments, the annealing temperature is between 1 and 15° C. (such as between 1 to 10, 1 to 5, 1 to 3, 3 to 5, 3 to 8, 5 to 10, 5 to 8, 8 to 10, 10 to 12, or 12 to 15° C., inclusive) greater than the melting temperature (such as the empirically measured or calculated T_m) of at least 25%, 50%, 60%, 70%, 75%, 80%, 90%, 95%, or all of the non-identical primers, and the length of the annealing step (per PCR cycle) is between 5 and 180 minutes, such as 15 and 120 minutes, 15 and 60 minutes, 15 and 45 minutes, or 20 and 60 minutes, inclusive.

[0153] In various embodiments, long annealing times (as discussed herein and exemplified in Example 12) and/or low primer concentrations are used. In fact, in certain embodiments, limiting primer concentrations and/or conditions are used. In various embodiments, the length of the annealing step is between 15, 20, 25, 30, 35, 40, 45, or 60 minutes on the low end of the range and 20, 25, 30, 35, 40, 45, 60, 120, or 180 minutes on the high end of the range. In various embodiments, the length of the annealing step (per PCR cycle) is between 30 and 180 minutes. For example, the annealing step can be between 30 and 60 minutes and the concentration of each primer can be less than 20, 15, 10, or 5 nM. In other embodiments the primer concentration is 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, or 25 nM on the low end of the range, and 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 20, 25, and 50 on the high end of the range.

[0154] At high level of multiplexing, the solution may become viscous due to the large amount of primers in solution. If the solution is too viscous, one can reduce the primer concentration to an amount that is still sufficient for the primers to bind the template DNA. In various embodiments, between 1,000 and 100,000 different primers are used and the concentration of each primer is less than 20 nM, such as less than 10 nM or between 1 and 10 nM, inclusive.

[0155] The following examples are put forth so as to provide those of ordinary skill in the art with a complete disclosure and description of how to use the embodiments provided herein, and are not intended to limit the scope of the disclosure nor are they intended to represent that the Examples below are all or the only experiments performed. Efforts have been made to ensure accuracy with respect to numbers used (e.g. amounts, temperature, etc.) but some experimental errors and deviations should be accounted for. Unless indicated otherwise, parts are parts by volume, and temperature is in degrees Centigrade. It should be understood that variations in the methods as described can be made without changing the fundamental aspects that the Examples are meant to illustrate.

[0156] A prior pilot study demonstrated the successful detection of cancer-relevant point mutations in the plasma of cancer patients. In that study, the mutation profile of 4 lung cancer tumors was determined by whole exome sequencing (WES) or Ampliseq (Life Technologies, Carlsbad, Calif.), and a subset of those mutations were successfully detected in the corresponding plasma samples using a multiplex PCR-Next-Generation Sequencing (mPCR-NGS) method. In this experiment, called TRACERx, the mPCR-NGS method was used to detect and track over time cancer-specific mutations in the plasma of cancer patients, and to evaluate the utility of the method in monitoring disease progression through treatment. The overall project design is shown in FIG. 1. The first phase of the project was the determination of the baseline mutation profile in the plasma of 50 treatment-naïve lung cancer patients. Purified genomic DNA samples from several tumor regions (2-7 regions per tumor), purified germline DNA samples, and intact plasma samples from 50 patients were obtained. The mutation profile of all of the tumor regions was previously determined by WES and AmpliSeq, and a subset of mutations per patient was analyzed by mPCR-NGS. Those mutations included both driver and passenger mutations and both clonal and sub-clonal mutations. Based on these data, we designed multiplex PCR assays, prepared primer pools (primers were obtained from IDT, Coralville, Iowa), QC'ed the primer pools, and optimized the mPCR protocol for each pool. Plasma cfDNA was purified, quantified, and converted into libraries. The libraries were then used as input into mPCR, and the products were sequenced and analyzed. A similar protocol was applied to the genomic DNA from tumor and matched normal samples.

[0157] Samples.

[0158] For each of the first 50 TRACERx patients, 4-5 ml of plasma obtained before tumor resection and prior to any therapy was isolated. Plasma samples were aliquoted in 2 ml tubes and shipped frozen on dry ice. Purified genomic DNA from up to 7 tumor subsections, from affected lymph nodes (where available), and from the white blood cell fraction (referred to as the matched normal) were purified and 500 ng purified DNA from each sample, normalized at 10 ng/μl, was analyzed. The purified DNA samples were frozen and shipped on dry ice.

[0159] SNV Information.

[0160] The mutation profile, including single nucleotide variants (SNVs) and copy number variants (CNVs), of each tumor subsection was determined by TRACERx using WES. The full mutation profile of each tumor was used to detect clonal structure and to reconstruct the phylogenetic tree of each tumor. PyClone (PyClone: statistical inference of clonal population structure in cancer. Roth et al, Nature Methods 11, 396-398 (2014)) was used to detect clonal structure. PyClone identifies a list of SNV subclones and calculates their cancer cell fraction. It also categorizes SNVs as either clonal or subclonal. The driver category of each SNV was determined and provided as the driver category (1-4, where 1 is most likely to be a driver mutation, and 4 is the least likely). For each patient, up to 108 SNVs, spanning all driver categories and including clonal and subclonal mutations, were analyzed. The detected allele fractions of each SNV in each tumor subsection, lymph node and matched normal DNA sample along with PyClone clonal/subclonal cluster information were compared.

[0161] Additional Information.

[0162] For each patient, the following information was available: tumor size (mm), tumor location (lung lobe), tumor stage, tumor pathological type, number of lymph nodes affected, vascular invasion status, as well as de-identified information on the collecting hospital.

[0163] Assay design. Natera's standard assay design pipeline was used to design Right and Left PCR primers for all given SNVs. A pair of Right and Left PCR primers targeting an SNV is defined as an assay for that particular SNV. Note that it is possible for one assay to cover more than 1 target SNV if they are in close proximity. For every pair of assays, the probability of forming a primer-dimer was calculated. The SNV allele fraction data in each tumor was used to reconstruct phylogenetic trees using Lichee (Fast and scalable inference of multi-sample cancer lineages. Popic et al. Genome Biol. 2015 May 6; 16:91). The list of assays for each sample were filtered to remove primers that are predicted to form primer dimers while giving strong priority to assays covering driver 1 and 2 SNVs. The remaining assays were used to build 5 balanced pools. All assays pooled together were compatible meaning there were no primers predicted to form primer-dimers in a pool. At each step, the assays were chosen such that assays covering driver 1 and 2 SNVs have the highest priority and for each patient the number of selected SNVs per branch was proportional to the total number of SNVs of that branch from the reconstructed phylogenetic tree. More specifically, we tried to have a uniform sampling of SNVs from branches in the reconstructed phylogeny tree, making sure selected assays provided good coverage of the reconstructed tree. The final design consisted of 972 assays, equally distributed among 5 pools, and containing 15-20 assays for each sample. The number of SNVs and the number of assays per sample by driver category are shown in FIG. 2. The genes in which the SNVs are found and the number of SNVs that were assayed per gene are found in FIG. 19.

[0164] Pool QC and Optimization.

[0165] The 972 primer pairs were obtained (IDT, Coralville, Iowa) in individual wells, desalted and normalized to 100 μM. The assays were pooled according to the pooling scheme, and each pool was used in a combined QC/optimization experiment. For the optimization experiment, several PCR parameters were varied and the effects on the sequencing performance, as well as the number of drop-out assays were evaluated from the sequence data. The PCR conditions that yielded the best percentage of on target reads, depth of read uniformity, and error rate were determined. Primers that were responsible for the majority of primer dimers were identified and removed from each pool (for each primer removed, its corresponding partner was also removed). Following this step, 908 total assays remained, equally distributed among the 5 pools.

[0166] DNA Extraction and QC.

[0167] All the plasma aliquots from each patient were pooled prior to cfDNA extraction, and the hemolysis grade of each pooled plasma sample was evaluated visually (no hemolysis, mild hemolysis or severe hemolysis). cfDNA was extracted using the Qiagen NA kit (Valencia, Calif.) following a protocol optimized for 5 ml of plasma. All cfDNA samples were QCed on Bioanalyzer High Sensitivity chips (Agilent, Santa Clara, Calif.). The same Bioanalyzer High Sensitivity runs were also used to quantify the cfDNA samples by interpolation of the mononucleosomal peak height on a calibration curve prepared from a pure cfDNA sample that was previously quantified. This was necessary because cfDNA sometimes contains an intact DNA fraction that overlaps with the high size marker on the chip, which makes quantification of the mononucleosomal peak unreliable. A representative subset of the purified genomic DNA samples (from tumor subsections, lymph nodes and white blood cells) was quantified using Nanodrops (Wilmington, Del.). All of the samples quantified were in the expected range (˜10 ng/μl).

[0168] cfDNA Library Preparation.

[0169] The entire cfDNA amount from each plasma sample was used as input into Library Prep using the Natera library prep kit and following the kit instructions. For two samples with extremely high cfDNA amounts, the input amount into Library Prep was restricted to ˜50,000 genome equivalents (165 ng). The libraries were amplified to plateau and then purified using Ampure beads (Beckman Coulter, Brea, Calif.) following the manufacturer's protocol. The purified libraries were QCed on the LabChip.

[0170] cfDNA Multiplex PCR and Sequencing.

[0171] The library material from each plasma sample was used as input into multiplex PCR (mPCR) using the relevant assay pool and an optimized plasma mPCR protocol. The protocol utilized an annealing time of 15 minutes at a temperature of 60 C or 62.5 C, which was above the Tm of the primers. The Tms of the primers using theoretical calculations was 53 to 59 C. A 10 nM primer concentration was used. The mPCR products were barcoded in a separate PCR step, and the barcoded PCR products were pooled according to the assay pooling information (see section above) into 5 pools. The pools were purified using Ampure beads following the manufacturer's protocol, QCed on a Bioanalyzer DNA1000 chip (Agilent, Santa Clara, Calif.), and quantified using the Qubit dsDNA Broad Range kit (Thermo Fisher Scientific, Waltham, Mass.). Each pool contained libraries prepared as disclosed above, from 10 cancer patient plasma samples and 20 negative controls (prepared from cfDNA extracted from presumed healthy volunteers). The negative control samples were obtained following the necessary regulatory procedures. Each pool was sequenced on a separate HiSeq 2500 Rapid run (IIlumina, San Diego, Calif.) with 50 cycle paired end single index reads.

[0172] gDNA Multiplex PCR and Sequencing.

[0173] The genomic DNA samples were used as input into a similar mPCR using the relevant assay pools and an optimized genomic mPCR protocol. The mPCR products were barcoded in a separate PCR step, and all the barcoded products were combined into one pool. The pool was purified using Ampure beads following the manufacturer's protocol, QCed on a Bioanalyzer DNA1000 chip, and quantified using the Qubit dsDNA Broad Range kit. The pool was sequenced on a single HiSeq2500 Rapid run with 50 cycle single end single index reads.

[0174] FIG. 20 is a table showing detailed results of the analysis and detailed information regarding the samples that were analyzed in this study.

[0175] cfDNA Extraction and Analysis.

[0176] The distribution of cfDNA concentrations for the 50 plasma samples (FIG. 3) followed the expected distribution based on 5 ml of plasma (median of 2,200 genome copy equivalents per ml of plasma). The cfDNA concentrations, the hemolysis grade (visually estimated) and the qualitative evaluation of the cfDNA size profile (visually estimated from the Bioanalyzer traces) are shown in tabular form in FIG. 16

[0177] Cfdna Analysis.

[0178] The purified cfDNA concentration, plasma hemolysis grade and cfDNA profile are shown in FIG. 16. cfDNA concentration refers to the mononucleosomal peak only, and was determined from the mononucleosomal peak height using a calibration curve. Genome copy equivalents were calculated using a 3.3 pg/genome conversion factor; 40 μl purified cfDNA is used as input into Library Prep; green highlights: for those samples, the input into library prep was restricted to 50,000 genome equivalents. cfDNA size profile: 1: most of the cfDNA is in the mononucleosomal peak; 2: most of the cfDNA is in the mononucleosomal peak, but other sizes are seen; 3: a large peak of intact DNA (>1,000 bp) is seen along with the mononucleosomal peak and some higher molecular weight peaks. Hemolysis was estimated visually based on the plasma color. 0: no hemolysis (yellow plasma); 1: mild hemolysis (faint pink plasma); 2: severe hemolysis (bright pink or red plasma).

[0179] VAF Analysis in Tumor Subsections.

[0180] The sequence data from each of the tumor subsections was analyzed to determine the variant allele frequency (VAF) of each SNV in each tumor subsection, lymph node and matched normal sample. This data was compared with matched data provided separately from a different test site using different test methods, such as whole genome sequencing and exome sequencing. For most samples, the previously determined tissue VAF values from each tumor subsection closely matched the newly derived tissue VAF values (FIG. 4). However, there were a large number of samples in which significant discrepancies were seen (FIG. 5). Three types of discrepancies were observed: (i) for one or two subsections, all the VAFs are 0 or close to 0 in the previous analysis, but are non-zero (and span the range of VAFs seen in other subsections of the same sample) in the mPCR-NGS analysis (e.g.: LTX041, LTX111); (ii) for several assays, the VAFs are 0 in the mPCR-NGS analysis, but are non-zero (and span the range of VAFs seen in other subsections of the same sample) in the previous analysis, and no clustering by subsection was seen with this discrepancy mode (e.g.: LTX093, LTX074); (iii) for several assays or regions, none of the assays failed but concordance between VAFs obtained in the two analyses was generally poor (e.g.: LTX063, LTX059).

[0181] We also identified 16 somatic SNVs from tissue samples which were not reported in TRACERx SNV calls. Among these new somatic SNVs, 7 were called in their corresponding plasma cfDNA as well. Please see the list in FIG. 18.

[0182] One sample (U_LTX206, with 19 assays) failed sequencing and was removed from the analysis. 889 assays covering 911 SNVs were analyzed. Assays with a depth of read of less than 1,000 were considered failed, and their corresponding SNVs were marked as “no call”. In total 21 “no call” SNVs were removed from the analysis; 890 total SNVs were analyzed.

[0183] Each run belonged to one assay pool and contained 10 cancer samples as well as 20 control samples. The set of SNVs covered by assays in a pool are considered as target SNVs for the associated run. To make an SNV call at a specific position of a cancer sample, first a background error model for that position was built. The error model was constructed based on the 20 negative samples and the remaining cancer samples (8 or 9) that were not expected to contain an SNV at that position, based on the information provided. Positions with VAF >20% were excluded from the background error model. A positive plasma SNV call was made if the confidence for that mutation in the corresponding plasma sample passed our confidence threshold of 95 to 98%.

[0184] The overall SNV detection rate in plasma is 35.5% (310 out of 890), similar to a prior pilot study. While the algorithm made most confident true positive calls, the number of false positive calls are at an acceptable number (<0.25%). The average mutant allele frequency for the SNVs detected with high confidence is 0.875%, ranging from 0.011% to 13.93%. A sample was considered as ‘detected in plasma’ if at least one SNV expected to be present in that sample was confidently detected in plasma. Using this definition, the overall sample detection rate in plasma was 69% (34 out of 49 samples), and for those, and the average number of SNVs detected in plasma was 9.1 (ranging from 1 to 19).The number of SNVs detected in plasma for each sample is shown in tabular form in FIG. 17.

[0185] Analysis of SNVs that were not Detected in Plasma.

[0186] Several lines of evidence support the conclusion that the failure to detect >60% (580 out of 911) of the expected SNVs in the plasma is due to the fact that there is not enough evidence of presence for those mutations in the cfDNA sample, as opposed to some failure of the mPCR-NGS method: The depth of read (DOR) distribution is similar for the assays that detected the expected plasma SNV and the ones that didn't detect the expected SNV (FIG. 6a) (average DOR 45,551 for assays that detected the expected SNV vs 45,133 for the ones that didn't). This suggests assays corresponding to false negative SNV calls are as efficient as the ones for true positive calls. Furthermore, despite the high DOR at the target SNV position, the number of mutant reads is almost negligible. In fact, 36% of them have 0 mutant reads, 75% of them have more than 5 mutant reads, and the remaining 25% false negative calls have VAF <0.1%.

[0187] Several factors that influence plasma SNV detectability have been evaluated. The cfDNA amount and the tumor staging information, tumor size and the SNV frequencies in tumor subsections were determined in separate locations.

[0188] Histological Type.

[0189] The most important predictor of whether a particular tumor was detected in the plasma appeared to be histological type: 100% of the squamous cell carcinoma (SQCC) tumors were detected in plasma, whereas only 50% (15/29) of the adenocarcinoma (ADC) tumors were detected in plasma in this study (FIG. 7). Moreover, the average number of SNVs detected per sample was 12.7 (median=13) for SQCC and 2.6 (median=1) for ADC. There was only one carcinosarcoma tumor and one adenosquamous tumor in this cohort, so no conclusions about their general detectability in plasma could be derived about those tumor types at this time.

[0190] Tumor Stage and Size.

[0191] Tumor stage and size were some of the most important factors identified that influence the number of SNVs detected in the corresponding plasma sample (FIG. 8). Stage 1a tumors had the lowest chance of having at least one SNV detected, as well as the lowest success rate of detecting SNVs in the plasma. The VAF distribution for the SNVs that were detected from stage 1a tumors was also lower than for the rest of the tumors (FIG. 9). As tumor size and stage are correlated, a similar trend was seen with tumor size. As this was not due to assay failure or sensitivity limits (see below), the most likely explanation is that such tumors tend to not have cfDNA present in the plasma in quantities that are detectable in the plasma volumes used in this study. The effect of tumor stage and size on the number of SNVs detected in ctDNA varied between ADC and SQCC samples. The ADC samples were more dependent on these factors with a general trend of far fewer SNVs detected in ctDNA than were detected in ctDNA of the SCC samples. In fact, SNVs were detected in the ctDNA of all of the SQCC samples regardless of stage: Three SNVs were detected in the ctDNA of one of the SCC samples and at least 5 SNVs detected in the ctDNA of the remainder of the SCC samples (FIG. 15). In fact, between 3 and 19 SNVs were detected in the ctDNA of SCC samples. In 6 ADC samples that were stage 1a, an SNV was only detected in one of the ctDNA samples, and in that sample only a single SNV was detected. In none of the stage 1a ADC samples were more than 1 SNV detected in the ctDNA. In stage 1b ADC samples, less than 5 SNVs were identified in all but two samples, with 7 SNVs identified in one of the stage 1b ADC samples and 18 SNVs identified in one of the stage 1b samples.

[0192] Tumor VAF and Clonality.

[0193] The clonality ratio was calculated for each mutation as (number of sub-sections of a tumor where the mutation is detected)/(total number of sub-sections of that tumor analyzed). Mutations that were observed in all analyzed tumor sections are considered ‘clonal’, all others are considered ‘sub-clonal’. The VAF of SNVs detected in plasma correlates with the ‘clonality’ of the mutations, with more clonal mutations being responsible for the highest plasma VAF values (FIGS. 9 and 12); similarly, SNVs present in multiple tumor sub-sections tend to be responsible for higher plasma VAFs in the corresponding plasma samples. In addition to the clonality ratio, the clonal status of each SNV was categorized by PyCloneCluster based on WES data from the tumor tissue. Clonal SNVs tended to have higher VAFs (FIGS. 13 and 14).

[0194] cfDNA Input and Tumor VAF.

[0195] There was no correlation between the amount of cfDNA and the number and proportion of SNVs detected in the plasma samples. The number of SNVs detected in plasma is not predicted by the cfDNA input amount; however, all samples with high input (>25,000 copies) had at least one SNV detected in plasma (FIG. 10). The plasma SNV VAF also correlates with the tumor SNV VAF(FIG. 11).

[0196] Multivariate Analysis.

[0197] A regression analysis was performed to determine the variables that can be used to predict our detection of mutations. More specifically, a 0/1 response variable was used to annotate the mutations we called as present or not. The following independent variables were included in our model:

• • [0198] 1. tumor VAF
  • [0199] 2. PyClone cluster (categorical variable)
  • [0200] 3. cancer stage (categorical variable)
  • [0201] 4. size of the tumor
  • [0202] 5. input DNA amount
  • [0203] 6. pathological type (categorical variable)
  • [0204] 7. number of affected lymph nodes
  • [0205] 8. vascular invasion (categorical variable)
  • [0206] 9. affected lobe
  • [0207] A logistic regression showed that the following variables had statistically significant association with the detection of a mutation (with p-values <5%):
  • [0208] 1. tumor VAF (p=4.3e-6)
  • [0209] 2. PyClone cluster (p=1.6e-4)
  • [0210] 3. size of the tumor (p=3.5e-4)
  • [0211] 4. pathological type (p=8.3e-30)

[0212] Conclusions.

[0213] We demonstrate in this example the successful detection of lung cancer-related SNVs in plasma samples from patients with lung cancer. Using a custom multiplex PCR panel tailored for this sample cohort, SNVs with variant allele fraction as low as 0.01% were detected. Of the tested SNVs, 35% were detected in the plasma samples and 67% of the samples analyzed had at least one plasma SNV detected. We also identified some of the factors that contribute to the successful detection of plasma SNV. These include tumor type, tumor stage, tumor size, SNV allele frequency in tumor and, to a lesser extent, amount of cfDNA analyzed. The finding that not all SNVs were detected in plasma, and that not all samples have SNVs detectable in the plasma, does not appear to be due to assay or protocol limitations, as those assays were functional (as evidenced by their sequencing depth of read) and their limit of detection was sufficient to detect any SNVs, should they be present in the cfDNA sample. Rather, the failure to detect those SNVs was likely due to the fact that they are not present in the sample. Samples from low grade tumors and small tumors were more likely to have limited amounts of circulating tumor DNA. Similarly, SNVs that were present at low allele frequency in the tumor were less likely to be present in the plasma. However, even tumors of high grade and relatively large size can have no SNVs detected in the plasma. It is possible that other biological reasons are responsible for this (such as amount of ctDNA shedding from the tumor) and that analyzing more SNVs per sample will increase the chance of detecting some. 0.4 mM dNTPs (see FIG. 12-3C).

[0214] Those skilled in the art can devise many modifications and other embodiments within the scope and spirit of the presently disclosed inventions. Indeed, variations in the materials, methods, drawings, experiments examples and embodiments described may be made by skilled artisans without changing the fundamental aspects of the disclosed inventions. Any of the disclosed embodiments can be used in combination with any other disclosed embodiment.

[0215] The disclosed embodiments, examples and experiments are not intended to limit the scope of the disclosure nor to represent that the experiments below are all or the only experiments performed. Efforts have been made to ensure accuracy with respect to numbers used (e.g., amounts, temperature, etc.) but some experimental errors and deviations should be accounted for. It should be understood that variations in the methods as described may be made without changing the fundamental aspects that the experiments are meant to illustrate.

[0216] Natera's bespoke multiplex PCR (mPCR) protocol is designed to estimate plasma ctDNA level by tracking a set of patient-specific mutations identified from tumor tissue sample(s). Given a patient-specific mutation profile we designed custom mPCR panels that can be applied to time-series plasma samples of the corresponding patient.

[0217] SNV targets. The mutation profile, which included single nucleotide variants (SNVs) for each tumor subsection, was determined based on analyses of tumor sequencing. The full mutation profile of each tumor was used to reconstruct the phylogenetic tree of each tumor. PyClone (Roth, et al. (2014). PyClone: Statistical inference of clonal population structure in cancer. Nature Methods 11: 396-398) was used to identify clusters of SNVs, and calculate their cancer-cell fraction. This was used to categorize SNVs as either clonal or subclonal. The driver category of each SNV was determined (1-4, where 1 was most likely to be a driver mutation, and 4 was the least likely).

[0218] Assay Design.

[0219] Natera's standard assay design pipeline was used to design PCR primers for all given SNVs with following parameters:

• • [0220] Optimal melting temperature [Tm] 56° C., allowed range, 53° C.-59° C.
  • [0221] Amplicon length, 50-70 bp
  • [0222] GCcontent, 30-70%

[0223] We refer to a pair of Right and Left PCR primers targeting a SNV as an assay for that particular SNV. It is possible for one assay to cover more than 1 target SNV, if they were in close proximity. For every pair of assays, the probability of forming primer-dimer was calculated using thermodynamic approach (SantaLucia J R (1998) “A unified view of polymer, dumbbell and oligonucleotide DNA nearest-neighbor thermodynamics”, Proc Natl Acad Sci 95:1460-65) to estimate the stability of the primer pair's joint hybridization structure. Assays were pooled together to minimize the number of primers with a high probability of forming primer-dimer in the same pool. For each patient, assays were prioritized such that, 1) assays covering driver SNVs had highest priority, and 2) there was uniform sampling of the phylogenetic tree.

[0224] Pool QC and Optimization.

[0225] The primers were ordered from IDT in individual wells, on desalted and normalized to 100 uM. The assays were pooled at Natera according to the pooling scheme, to create assay pools where each primer was at 250 nM in water. Each pool was used in a combined QC/optimization experiment. For the optimization experiment, PCR parameters (primer concentration and annealing temperature) were varied and the effects on the percentage of on-target reads, depth-of-read uniformity (measured as the ratio of the 80th percentile/20th percentile), and the number of drop-out assays (defined as assays with <1,000 reads) were evaluated from the sequencing data. The PCR conditions that yield the best percentage of on-target reads, depth-of-read uniformity, and the lowest number of drop-outs were determined. For all pools, the optimal conditions were 10 nM primers and 60° C. or 62.5° C. annealing temperature.

[0226] Primers that were responsible for the majority of primer dimers were identified and removed from each pool (for each primer removed, its corresponding partner was also removed).

[0227] DNA Extraction and QC.

[0228] Plasma aliquots from each patient were pooled prior to cfDNA extraction, and the hemolysis grade of each pooled plasma sample was evaluated visually and noted (no hemolysis, mild hemolysis, or severe hemolysis). cfDNA was extracted at Natera using the Qiagen NA kit following a protocol optimized for 5 ml of plasma. All cfDNA samples were QCed on Bioanalyzer High Sensitivity chips. The same Bioanalyzer High Sensitivity runs were also used to also quantify the cfDNA samples by interpolation of the mononucleosomal peak height on a calibration curve prepared from a pure cfDNA sample that was quantified previously. This is necessary because cfDNA sometimes contains an intact DNA fraction that overlaps with the high size marker on the chip, making quantification of the mononucleosomal peak unreliable.

[0229] Genomic DNA samples (from tumor subsections, lymph nodes, and white blood cells) were quantified on the Nanodrop.

[0230] cfDNA Library Preparation.

[0231] The entire cfDNA amount from each plasma sample was used as input into Library Prep using the Natera library prep kit and following the kit instructions. For two samples with extremely high cfDNA amounts, the input amount into Library Prep was restricted to −50,000 genome equivalents (165 ng). In brief, 40 ul of DNA extracted from plasma, which is present in fragments of mononucleosomal and polynucleosomal length, were end repaired and A-tailed, and Natera custom adapters ligated. The libraries were amplified for 15 cycles to plateau and then purified using Ampure beads following the manufacturer's protocol. The purified libraries were QCed on the LabChip.

[0232] cfDNA Multiplex PCR and Sequencing.

[0233] The library material from each plasma sample was used as input into multiplex PCR using the relevant assay pool and an optimized plasma mPCR protocol. The PCR composition was: 1× in-house PCR master mix, 10 nM primers, 3 uL cfDNA library (corresponding to −600 ng DNA), in 10 uL total reaction volume. The thermocycling conditions were: 95° C., 10 minutes; 10 cycles of (95° C., 30 seconds; 60° C. or 62.5° C., 15 minutes; 72° C., 30 seconds); 72° C., 2 minutes, 4° C. hold.

[0234] The mPCR products were barcoded in a separate PCR step, and the barcoded PCR products were pooled according to the assay pooling information.

[0235] The pools were purified using Ampure beads following the manufacturer's protocol, QCed on a Bioanalyzer DNA1000 chip, and quantified using the Qubit dsDNA Broad Range kit. Each pool contained barcoded mPCR products of 10 cancer plasma libraries and 20 negative controls (prepared from cfDNA extracted from healthy volunteers). The negative control samples were obtained following the necessary regulatory procedures. Each pool was sequenced on a separate HiSeq2500 Rapid runs with 50 cycle paired end single index reads.

[0236] Genomic DNA Multiplex PCR and Sequencing.

[0237] The genomic DNA samples (gDNA) were used as input into a similar mPCR using the relevant assay pools and an optimized genomic mPCR protocol; 50 ng gDNA was used as input. The mPCR products were barcoded in a separate PCR step, and all the barcoded products were combined into one pool. The pool was purified using Ampure beads following the manufacturer's protocol, QCed on a Bioanalyzer DNA1000 chip, and quantified using the Qubit dsDNA Broad Range kit. The pool was sequenced on a single HiSeq2500 Rapid run with 50 cycle single end single index reads.

[0238] Bioinformatics Pipeline.

[0239] Paired-end reads were mapped to the hg19 reference genome with Novoalign v2.08.02, and sorted and indexed using SAMtools (Li H.*, Handsaker B.*, Wysoker A., Fennell T., Ruan J., Homer N., Marth G., Abecasis G., Durbin R. and 1000 Genome Project Data Processing Subgroup (2009) The Sequence alignment/map (SAM) format and SAMtools. Bioinformatics, 25, 2078-9). All the paired-end reads were merged using Pear (J. Zhang, K. Kobert, T. Flouri, A. Stamatakis. PEAR: A fast and accurate Illumina Paired-End reAd mergeR. Bioinformatics 30(5): 614-620, 2014) (using default parameters). Since all amplicons are less than 70 bases long, with paired 50 bp reads generated by Illumina HiSeq 2500 all on-target reads were merged with the minimum of 30 bp overlap. Unassembled Reads are off-target and were filtered at this step. Amplicons were designed such that the target SNV positions were located in the overlapping region. Bases that did not match in forward and reverse reads or that have Phred quality score less than 20 were filtered out to minimize sequencing errors in subsequent steps. Merged reads with mapping quality higher than 30 and at most one mismatch under the sequence of primers were marked as on-target. Targets with less than 1000 reads were considered failed and were filtered from further analyses. Quality control (QC) was performed using an in-house Java program checking for a wide list of statistics per sample that included total numbers of reads, mapped reads, on-target reads, number of failed targets, and average error rate. A sample with less than 90% mapped reads and more than 3 failed targets did not pass, and needed to be resequenced.

[0240] Statistical Model.

[0241] The PCR process was modeled as a stochastic process, estimating the error parameters using a set of 29 control plasma samples and making the final SNV calls on the target cancer samples. For each target SNV, we built a target-specific background error model by estimating the following parameters from the control samples.

• • [0242] PCR efficiency (p): Probability that each molecule is replicated in a PCR cycle.
  • [0243] Error rate (p_e): Error rate per cycle for mutation type e (e.g wildtype allele A to mutant allele G).
  • [0244] Initial number of molecules (X₀)

[0245] The target-specific error propagation model was used to characterize the distribution of error molecules. As a molecule is replicated over the course of PCR process, more errors occur. If an error occurs in cycle i and there are X_i wildtype molecules in the system, that error molecule is duplicated in next cycle with probability p and new error molecules are produced from wildtype background molecules according to a binomial process B(X_i, p_e). Using a recursive relation, we computed the mean and variance of number of total molecules X_n and number of error molecules E_n after n PCR cycles as shown in FIG. 21.

[0246] Algorithm Steps:

• • [0247] a. Estimating the PCR efficiency and per cycle error rate using the normal control samples.
  • [0248] b. Using the efficiency estimate, compute the starting number of molecules in the test set.
  • [0249] c. Use this starting copy number and the prior efficiency distribution from the training set to estimate the PCR efficiency in the test sample.
  • [0250] d. For a range of potential real mutant fraction values 0 between 0 and 1 (we used 0.15 as upper bound), we estimate the mean and variance for the total number of molecules, background error molecules and real mutation molecules using the error propagation model described in last paragraph and parameters estimated in steps a-c.
  • [0251] e. Use the mean and variance estimated in step d to compute the likelihood L(θ) for each potential real mutant fraction. Select the value of θ that maximizes this likelihood, (denoted by {circumflex over (θ)}N_MLE) and compute the confidence score

$\left(\mathrm{as}\frac{L\left({\hat{\theta }}_{\mathrm{MLE}}\right)}{L\left(0\right)+L\left({\hat{\theta }}_{\mathrm{MLE}}\right)}\right).$

• • [0252] f. Call a mutation if the confidence score is ≥≥95% for transitions and ≥98% for transversions.