Enhancing the accuracy of next-generation sequencing for detecting rare and subclonal mutations.

Mutations, the fuel of evolution, are first manifested as rare DNA changes within a population of cells. Although next-generation sequencing (NGS) technologies have revolutionized the study of genomic variation between species and individual organisms, most have limited ability to accurately detect and quantify rare variants among the different genome copies in heterogeneous mixtures of cells or molecules. We describe the technical challenges in characterizing subclonal variants using conventional NGS protocols and the recent development of error correction strategies, both computational and experimental, including consensus sequencing of single DNA molecules. We also highlight major applications for low-frequency mutation detection in science and medicine, describe emerging methodologies and provide our vision for the future of DNA sequencing.

PolyG-DS: An ultrasensitive polyguanine tract-profiling method to detect clonal expansions and trace cell lineage.

Polyguanine tracts (PolyGs) are short guanine homopolymer repeats that are prone to accumulating mutations when cells divide. This feature makes them especially suitable for cell lineage tracing, which has been exploited to detect and characterize precancerous and cancerous somatic evolution. PolyG genotyping, however, is challenging because of the inherent biochemical difficulties in amplifying and sequencing repetitive regions. To overcome this limitation, we developed PolyG-DS, a next-generation sequencing (NGS) method that combines the error-correction capabilities of duplex sequencing (DS) with enrichment of PolyG loci using CRISPR-Cas9-targeted genomic fragmentation. PolyG-DS markedly reduces technical artifacts by comparing the sequences derived from the complementary strands of each original DNA molecule. We demonstrate that PolyG-DS genotyping is accurate, reproducible, and highly sensitive, enabling the detection of low-frequency alleles (<0.01) in spike-in samples using a panel of only 19 PolyG markers. PolyG-DS replicated prior results based on PolyG fragment length analysis by capillary electrophoresis, and exhibited higher sensitivity for identifying clonal expansions in the nondysplastic colon of patients with ulcerative colitis. We illustrate the utility of this method for resolving the phylogenetic relationship among precancerous lesions in ulcerative colitis and for tracing the metastatic dissemination of ovarian cancer. PolyG-DS enables the study of tumor evolution without prior knowledge of tumor driver mutations and provides a tool to perform cost-effective and easily scalable ultra-accurate NGS-based PolyG genotyping for multiple applications in biology, genetics, and cancer research.

A high-resolution landscape of mutations in the BCL6 super-enhancer in normal human B cells.

The super-enhancers (SEs) of lineage-specific genes in B cells are off-target sites of somatic hypermutation. However, the inability to detect sufficient numbers of mutations in normal human B cells has precluded the generation of a high-resolution mutational landscape of SEs. Here we captured and sequenced 12 B cell SEs at single-nucleotide resolution from 10 healthy individuals across diverse ethnicities. We detected a total of approximately 9,000 subclonal mutations (allele frequencies <0.1%); of these, approximately 8,000 are present in the BCL6 SE alone. Within the BCL6 SE, we identified 3 regions of clustered mutations in which the mutation frequency is ∼7 × 10-4 Mutational spectra show a predominance of C > T/G > A and A > G/T > C substitutions, consistent with the activities of activation-induced-cytidine deaminase (AID) and the A-T mutator, DNA polymerase η, respectively, in mutagenesis in normal B cells. Analyses of mutational signatures further corroborate the participation of these factors in this process. Single base substitution signatures SBS85, SBS37, and SBS39 were found in the BCL6 SE. While SBS85 is a denoted signature of AID in lymphoid cells, the etiologies of SBS37 and SBS39 are unknown. Our analysis suggests the contribution of error-prone DNA polymerases to the latter signatures. The high-resolution mutation landscape has enabled accurate profiling of subclonal mutations in B cell SEs in normal individuals. By virtue of the fact that subclonal SE mutations are clonally expanded in B cell lymphomas, our studies also offer the potential for early detection of neoplastic alterations.

Extensive subclonal mutational diversity in human colorectal cancer and its significance.

Human colorectal cancers (CRCs) contain both clonal and subclonal mutations. Clonal driver mutations are positively selected, present in most cells, and drive malignant progression. Subclonal mutations are randomly dispersed throughout the genome, providing a vast reservoir of mutant cells that can expand, repopulate the tumor, and result in the rapid emergence of resistance, as well as being a major contributor to tumor heterogeneity. Here, we apply duplex sequencing (DS) methodology to quantify subclonal mutations in CRC tumor with unprecedented depth (104) and accuracy (<10-7). We measured mutation frequencies in genes encoding replicative DNA polymerases and in genes frequently mutated in CRC, and found an unexpectedly high effective mutation rate, 7.1 × 10-7. The curve of subclonal mutation accumulation as a function of sequencing depth, using DNA obtained from 5 different tumors, is in accord with a neutral model of tumor evolution. We present a theoretical approach to model neutral evolution independent of the infinite-sites assumption (which states that a particular mutation arises only in one tumor cell at any given time). Our analysis indicates that the infinite-sites assumption is not applicable once the number of tumor cells exceeds the reciprocal of the mutation rate, a circumstance relevant to even the smallest clinically diagnosable tumor. Our methods allow accurate estimation of the total mutation burden in clinical cancers. Our results indicate that no DNA locus is wild type in every malignant cell within a tumor at the time of diagnosis (probability of all cells being wild type, 10-308).

Accurate RNA consensus sequencing for high-fidelity detection of transcriptional mutagenesis-induced epimutations.

Transcriptional mutagenesis (TM) due to misincorporation during RNA transcription can result in mutant RNAs, or epimutations, that generate proteins with altered properties. TM has long been hypothesized to play a role in aging, cancer, and viral and bacterial evolution. However, inadequate methodologies have limited progress in elucidating a causal association. We present a high-throughput, highly accurate RNA sequencing method to measure epimutations with single-molecule sensitivity. Accurate RNA consensus sequencing (ARC-seq) uniquely combines RNA barcoding and generation of multiple cDNA copies per RNA molecule to eliminate errors introduced during cDNA synthesis, PCR, and sequencing. The stringency of ARC-seq can be scaled to accommodate the quality of input RNAs. We apply ARC-seq to directly assess transcriptome-wide epimutations resulting from RNA polymerase mutants and oxidative stress.

Mutational spectra of aflatoxin B1 in vivo establish biomarkers of exposure for human hepatocellular carcinoma.

Aflatoxin B1 (AFB1) and/or hepatitis B and C viruses are risk factors for human hepatocellular carcinoma (HCC). Available evidence supports the interpretation that formation of AFB1-DNA adducts in hepatocytes seeds a population of mutations, mainly G:C→T:A, and viral processes synergize to accelerate tumorigenesis, perhaps via inflammation. Responding to a need for early-onset evidence predicting disease development, highly accurate duplex sequencing was used to monitor acquisition of high-resolution mutational spectra (HRMS) during the process of hepatocarcinogenesis. Four-day-old male mice were treated with AFB1 using a regimen that induced HCC within 72 wk. For analysis, livers were separated into tumor and adjacent cellular fractions. HRMS of cells surrounding the tumors revealed predominantly G:C→T:A mutations characteristic of AFB1 exposure. Importantly, 25% of all mutations were G→T in one trinucleotide context (CGC; the underlined G is the position of the mutation), which is also a hotspot mutation in human liver tumors whose incidence correlates with AFB1 exposure. The technology proved sufficiently sensitive that the same distinctive spectrum was detected as early as 10 wk after dosing, well before evidence of neoplasia. Additionally, analysis of tumor tissue revealed a more complex pattern than observed in surrounding hepatocytes; tumor HRMS were a composite of the 10-wk spectrum and a more heterogeneous set of mutations that emerged during tumor outgrowth. We propose that the 10-wk HRMS reflects a short-term mutational response to AFB1, and, as such, is an early detection metric for AFB1-induced liver cancer in this mouse model that will be a useful tool to reconstruct the molecular etiology of human hepatocarcinogenesis.

Why Cockayne syndrome patients do not get cancer despite their DNA repair deficiency.

Cockayne syndrome (CS) and xeroderma pigmentosum (XP) are human photosensitive diseases with mutations in the nucleotide excision repair (NER) pathway, which repairs DNA damage from UV exposure. CS is mutated in the transcription-coupled repair (TCR) branch of the NER pathway and exhibits developmental and neurological pathologies. The XP-C group of XP patients have mutations in the global genome repair (GGR) branch of the NER pathway and have a very high incidence of UV-induced skin cancer. Cultured cells from both diseases have similar sensitivity to UV-induced cytotoxicity, but CS patients have never been reported to develop cancer, although they often exhibit photosensitivity. Because cancers are associated with increased mutations, especially when initiated by DNA damage, we examined UV-induced mutagenesis in both XP-C and CS cells, using duplex sequencing for high-sensitivity mutation detection. Duplex sequencing detects rare mutagenic events, independent of selection and in multiple loci, enabling examination of all mutations rather than just those that confer major changes to a specific protein. We found telomerase-positive normal and CS-B cells had increased background mutation frequencies that decreased upon irradiation, purging the population of subclonal variants. Primary XP-C cells had increased UV-induced mutation frequencies compared with normal cells, consistent with their GGR deficiency. CS cells, in contrast, had normal levels of mutagenesis despite their TCR deficiency. The lack of elevated UV-induced mutagenesis in CS cells reveals that their TCR deficiency, although increasing cytotoxicity, is not mutagenic. Therefore the absence of cancer in CS patients results from the absence of UV-induced mutagenesis rather than from enhanced lethality.

Ultra-deep sequencing detects ovarian cancer cells in peritoneal fluid and reveals somatic TP53 mutations in noncancerous tissues.

Current sequencing methods are error-prone, which precludes the identification of low frequency mutations for early cancer detection. Duplex sequencing is a sequencing technology that decreases errors by scoring mutations present only in both strands of DNA. Our aim was to determine whether duplex sequencing could detect extremely rare cancer cells present in peritoneal fluid from women with high-grade serous ovarian carcinomas (HGSOCs). These aggressive cancers are typically diagnosed at a late stage and are characterized by TP53 mutations and peritoneal dissemination. We used duplex sequencing to analyze TP53 mutations in 17 peritoneal fluid samples from women with HGSOC and 20 from women without cancer. The tumor TP53 mutation was detected in 94% (16/17) of peritoneal fluid samples from women with HGSOC (frequency as low as 1 mutant per 24,736 normal genomes). Additionally, we detected extremely low frequency TP53 mutations (median mutant fraction 1/13,139) in peritoneal fluid from nearly all patients with and without cancer (35/37). These mutations were mostly deleterious, clustered in hotspots, increased with age, and were more abundant in women with cancer than in controls. The total burden of TP53 mutations in peritoneal fluid distinguished cancers from controls with 82% sensitivity (14/17) and 90% specificity (18/20). Age-associated, low frequency TP53 mutations were also found in 100% of peripheral blood samples from 15 women with and without ovarian cancer (none with hematologic disorder). Our results demonstrate the ability of duplex sequencing to detect rare cancer cells and provide evidence of widespread, low frequency, age-associated somatic TP53 mutation in noncancerous tissue.

Sequencing small genomic targets with high efficiency and extreme accuracy.

The detection of minority variants in mixed samples requires methods for enrichment and accurate sequencing of small genomic intervals. We describe an efficient approach based on sequential rounds of hybridization with biotinylated oligonucleotides that enables more than 1-million-fold enrichment of genomic regions of interest. In conjunction with error-correcting double-stranded molecular tags, our approach enables the quantification of mutations in individual DNA molecules.

Frameshift mutagenesis and microsatellite instability induced by human alkyladenine DNA glycosylase.

Human alkyladenine DNA glycosylase (hAAG) excises alkylated purines, hypoxanthine, and etheno bases from DNA to form abasic (AP) sites. Surprisingly, elevated expression of hAAG increases spontaneous frameshift mutagenesis. By random mutagenesis of eight active site residues, we isolated hAAG-Y127I/H136L double mutant that induces even higher rates of frameshift mutation than does the wild-type hAAG; the Y127I mutation accounts for the majority of the hAAG-Y127I/H136L-induced mutator phenotype. The hAAG-Y127I/H136L and hAAG-Y127I mutants increased the rate of spontaneous frameshifts by up to 120-fold in S. cerevisiae and also induced high rates of microsatellite instability (MSI) in human cells. hAAG and its mutants bind DNA containing one and two base-pair loops with significant affinity, thus shielding them from mismatch repair; the strength of such binding correlates with their ability to induce the mutator phenotype. This study provides important insights into the mechanism of hAAG-induced genomic instability.

Ultra-sensitive sequencing reveals an age-related increase in somatic mitochondrial mutations that are inconsistent with oxidative damage.

Mitochondrial DNA (mtDNA) is believed to be highly vulnerable to age-associated damage and mutagenesis by reactive oxygen species (ROS). However, somatic mtDNA mutations have historically been difficult to study because of technical limitations in accurately quantifying rare mtDNA mutations. We have applied the highly sensitive Duplex Sequencing methodology, which can detect a single mutation among >10(7) wild type molecules, to sequence mtDNA purified from human brain tissue from both young and old individuals with unprecedented accuracy. We find that the frequency of point mutations increases ~5-fold over the course of 80 years of life. Overall, the mutation spectra of both groups are comprised predominantly of transition mutations, consistent with misincorporation by DNA polymerase γ or deamination of cytidine and adenosine as the primary mutagenic events in mtDNA. Surprisingly, G → T mutations, considered the hallmark of oxidative damage to DNA, do not significantly increase with age. We observe a non-uniform, age-independent distribution of mutations in mtDNA, with the D-loop exhibiting a significantly higher mutation frequency than the rest of the genome. The coding regions, but not the D-loop, exhibit a pronounced asymmetric accumulation of mutations between the two strands, with G → A and T → C mutations occurring more often on the light strand than the heavy strand. The patterns and biases we observe in our data closely mirror the mutational spectrum which has been reported in studies of human populations and closely related species. Overall our results argue against oxidative damage being a major driver of aging and suggest that replication errors by DNA polymerase γ and/or spontaneous base hydrolysis are responsible for the bulk of accumulating point mutations in mtDNA.

Mitochondrial point mutations do not limit the natural lifespan of mice.

Whether mitochondrial mutations cause mammalian aging, or are merely correlated with it, is an area of intense debate. Here, we use a new, highly sensitive assay to redefine the relationship between mitochondrial mutations and age. We measured the in vivo rate of change of the mitochondrial genome at a single-base pair level in mice, and we demonstrate that the mutation frequency in mouse mitochondria is more than ten times lower than previously reported. Although we observed an 11-fold increase in mitochondrial point mutations with age, we report that a mitochondrial mutator mouse was able to sustain a 500-fold higher mutation burden than normal mice, without any obvious features of rapidly accelerated aging. Thus, our results strongly indicate that mitochondrial mutations do not limit the lifespan of wild-type mice.

Sphingosine, a modulator of human translesion DNA polymerase activity.

Translesion (TLS) DNA polymerases are specialized, error-prone enzymes that synthesize DNA across bulky, replication-stalling DNA adducts. In so doing, they facilitate the progression of DNA synthesis and promote cell proliferation. To potentiate the effect of cancer chemotherapeutic regimens, we sought to identify inhibitors of TLS DNA polymerases. We screened five libraries of ∼ 3000 small molecules, including one comprising ∼ 600 nucleoside analogs, for their effect on primer extension activity of DNA polymerase η (Pol η). We serendipitously identified sphingosine, a lipid-signaling molecule that robustly stimulates the activity of Pol η by ∼ 100-fold at low micromolar concentrations but inhibits it at higher concentrations. This effect is specific to the Y-family DNA polymerases, Pols η, κ, and ι. The addition of a single phosphate group on sphingosine completely abrogates this effect. Likewise, the inclusion of other sphingolipids, including ceramide and sphingomyelin to extension reactions does not elicit this response. Sphingosine increases the rate of correct and incorrect nucleotide incorporation while having no effect on polymerase processivity. Endogenous Pol η activity is modulated similarly as the recombinant enzyme. Importantly, sphingosine-treated cells exhibit increased lesion bypass activity, and sphingosine tethered to membrane lipids mimics the effects of free sphingosine. Our studies have uncovered sphingosine as a modulator of TLS DNA polymerase activity; this property of sphingosine may be associated with its known role as a signaling molecule in regulating cell proliferation in response to cellular stress.

Detection of ultra-rare mutations by next-generation sequencing.

Next-generation DNA sequencing promises to revolutionize clinical medicine and basic research. However, while this technology has the capacity to generate hundreds of billions of nucleotides of DNA sequence in a single experiment, the error rate of ~1% results in hundreds of millions of sequencing mistakes. These scattered errors can be tolerated in some applications but become extremely problematic when "deep sequencing" genetically heterogeneous mixtures, such as tumors or mixed microbial populations. To overcome limitations in sequencing accuracy, we have developed a method termed Duplex Sequencing. This approach greatly reduces errors by independently tagging and sequencing each of the two strands of a DNA duplex. As the two strands are complementary, true mutations are found at the same position in both strands. In contrast, PCR or sequencing errors result in mutations in only one strand and can thus be discounted as technical error. We determine that Duplex Sequencing has a theoretical background error rate of less than one artifactual mutation per billion nucleotides sequenced. In addition, we establish that detection of mutations present in only one of the two strands of duplex DNA can be used to identify sites of DNA damage. We apply the method to directly assess the frequency and pattern of random mutations in mitochondrial DNA from human cells.

A substitution in the fingers domain of DNA polymerase δ reduces fidelity by altering nucleotide discrimination in the catalytic site.

DNA polymerase δ (Pol δ) is one of the major replicative DNA polymerases in eukaryotic cells, catalyzing lagging strand synthesis as well as playing a role in many DNA repair pathways. The catalytic site for polymerization consists of a palm domain and mobile fingers domain that opens and closes each catalytic cycle. We explored the effect of amino acid substitutions in a region of the highly conserved sequence motif B in the fingers domain on replication fidelity. A novel substitution, A699Q, results in a marked increase in mutation rate at the yeast CAN1 locus, and is synthetic lethal with both proofreading deficiency and mismatch repair deficiency. Modeling the A699Q mutation onto the crystal structure of Saccharomyces cerevisiae Pol δ template reveals four potential contacts for A699Q but not for A699. We substituted alanine for each of these residues and determined that an interaction with multiple residues of the N-terminal domain is responsible for the mutator phenotype. The corresponding mutation in purified human Pol δ results in a similar 30-fold increase in mutation frequency when copying gapped DNA templates. Sequence analysis indicates that the most characteristic mutation is a guanine-to-adenine (G to A) transition. The increase in deoxythymidine 5'-triphosphate-G mispairs was confirmed by performing steady state single nucleotide addition studies. Our combined data support a model in which the Ala-to-Gln substitution in the fingers domain of Pol δ results in an interaction with the N-terminal domain that affects the base selectivity of the enzyme.

Do mutator mutations fuel tumorigenesis?

The mutator phenotype hypothesis proposes that the mutation rate of normal cells is insufficient to account for the large number of mutations found in human cancers. Consequently, human tumors exhibit an elevated mutation rate that increases the likelihood of a tumor acquiring advantageous mutations. The hypothesis predicts that tumors are composed of cells harboring hundreds of thousands of mutations, as opposed to a small number of specific driver mutations, and that malignant cells within a tumor therefore constitute a highly heterogeneous population. As a result, drugs targeting specific mutated driver genes or even pathways of mutated driver genes will have only limited anticancer potential. In addition, because the tumor is composed of such a diverse cell population, tumor cells harboring drug-resistant mutations will exist prior to the administration of any chemotherapeutic agent. We present recent evidence in support of the mutator phenotype hypothesis, major arguments against this concept, and discuss the clinical consequences of tumor evolution fueled by an elevated mutation rate. We also consider the therapeutic possibility of altering the rate of mutation accumulation. Most significantly, we contend that there is a need to fundamentally reconsider current approaches to personalized cancer therapy. We propose that targeting cellular pathways that alter the rate of mutation accumulation in tumors will ultimately prove more effective than attempting to identify and target mutant driver genes or driver pathways.

DNA polymerases and human disease.

The human genome encodes at least 14 DNA-dependent DNA polymerases--a surprisingly large number. These include the more abundant, high-fidelity enzymes that replicate the bulk of genomic DNA, together with eight or more specialized DNA polymerases that have been discovered in the past decade. Although the roles of the newly recognized polymerases are still being defined, one of their crucial functions is to allow synthesis past DNA damage that blocks replication-fork progression. We explore the reasons that might justify the need for so many DNA polymerases, describe their function and mode of regulation, and finally consider links between mutations in DNA polymerases and human disease.

The Werner syndrome exonuclease facilitates DNA degradation and high fidelity DNA polymerization by human DNA polymerase δ.

DNA Polymerase δ (Pol δ) and the Werner syndrome protein, WRN, are involved in maintaining cellular genomic stability. Pol δ synthesizes the lagging strand during replication of genomic DNA and also functions in the synthesis steps of DNA repair and recombination. WRN is a member of the RecQ helicase family, loss of which results in the premature aging and cancer-prone disorder, Werner syndrome. Both Pol δ and WRN encode 3' → 5' DNA exonuclease activities. Pol δ exonuclease removes 3'-terminal mismatched nucleotides incorporated during replication to ensure high fidelity DNA synthesis. WRN exonuclease degrades DNA containing alternate secondary structures to prevent formation and enable resolution of stalled replication forks. We now observe that similarly to WRN, Pol δ degrades alternate DNA structures including bubbles, four-way junctions, and D-loops. Moreover, WRN and Pol δ form a complex with enhanced ability to hydrolyze these structures. We also present evidence that WRN can proofread for Pol δ; WRN excises 3'-terminal mismatches to enable primer extension by Pol δ. Consistent with our in vitro observations, we show that WRN contributes to the maintenance of DNA synthesis fidelity in vivo. Cells expressing limiting amounts (∼10% of normal) of WRN have elevated mutation frequencies compared with wild-type cells. Together, our data highlight the importance of WRN exonuclease activity and its cooperativity with Pol δ in preserving genome stability, which is compromised by the loss of WRN in Werner syndrome.

A random mutation capture assay to detect genomic point mutations in mouse tissue.

Herein, a detailed protocol for a random mutation capture (RMC) assay to measure nuclear point mutation frequency in mouse tissue is described. This protocol is a simplified version of the original method developed for human tissue that is easier to perform, yet retains a high sensitivity of detection. In contrast to assays relying on phenotypic selection of reporter genes in transgenic mice, the RMC assay allows direct detection of mutations in endogenous genes in any mouse strain. Measuring mutation frequency within an intron of a transcribed gene, we show this assay to be highly reproducible. We analyzed mutation frequencies from the liver tissue of animals with a mutation within the intrinsic exonuclease domains of the two major DNA polymerases, δ and ε. These mice exhibited significantly higher mutation frequencies than did wild-type animals. A comparison with a previous analysis of these genotypes in Big Blue mice revealed the RMC assay to be more sensitive than the Big Blue assay for this application. As RMC does not require analysis of a particular gene, simultaneous analysis of mutation frequency at multiple genetic loci is feasible. This assay provides a versatile alternative to transgenic mouse models for the study of mutagenesis in vivo.