Subclonal phylogenetic structures in cancer revealed by ultra-deep sequencing

Abstract

During the clonal expansion of cancer from an ancestral cell with an initiating oncogenic mutation to symptomatic neoplasm, the occurrence of somatic mutations (both driver and passenger) can be used to track the on-going evolution of the neoplasm. All subclones within a cancer are phylogenetically related, with the prevalence of each subclone determined by its evolutionary fitness and the timing of its origin relative to other subclones. Recently developed massively parallel sequencing platforms promise the ability to detect rare subclones of genetic variants without a priori knowledge of the mutations involved. We used ultra-deep pyrosequencing to investigate intraclonal diversification at the Ig heavy chain locus in 22 patients with B-cell chronic lymphocytic leukemia. Analysis of a non-polymorphic control locus revealed artifactual insertions and deletions resulting from sequencing errors and base substitutions caused by polymerase misincorporation during PCR amplification. We developed an algorithm to differentiate genuine haplotypes of somatic hypermutations from such artifacts. This proved capable of detecting multiple rare subclones with frequencies as low as 1 in 5000 copies and allowed the characterization of phylogenetic interrelationships among subclones within each patient. This study demonstrates the potential for ultra-deep resequencing to recapitulate the dynamics of clonal evolution in cancer cell populations.

Fig. 1.

Ultra-deep pyrosequencing of the IGH locus in patients with CLL. (A) Nested PCR to generate amplicons for resequencing used published consensus primers for the IGH locus in the first round. The second round of PCR used internal primers, specific to the patient's VDJ rearrangement, with a 5′ 3-bp barcode and forward (F-454) and reverse (R-454) sequencing primers. (B) Histogram of the sequence length of the ≈385,000 reads. (C) Histogram of the number of reads per patient, with the 2 sample codes (PD2087a and PD2106a) indicating the patients for whom phylogenetic trees were generated in Fig. 4.

Fig. 2.

Sequencing errors in the control locus. (A) Histogram showing the distribution and type of high-quality (Phred ≥ 20 for insertions and deletions; base height drop ≥ 0.75 peak height for deletions) sequencing errors in the forward reads from the control amplicon, together with the locations of homopolymer tracts (orange). The base marked with the asterisk actually had 922 deletions, beyond the range of the graph. (B) Observed (bars) and fitted theoretical (line) frequency distribution of T>C; A>G errors in the control locus. The number of substitutions at each T or A nucleotide >2 bases outside homopolymer tracts and primer sequence in the amplicon was counted across the 12,700 reads. The Luria–Delbrück distribution (line) was fitted to these observed frequencies, allowing estimation of the misincorporation rate of the polymerase. (C) Observed (bars) and fitted theoretical (line) frequency distribution of C>T; G>A errors in the control locus. The number of substitutions at each C or G nucleotide >2 bases outside homopolymer tracts and primer sequence in the amplicon was counted across the 12,700 reads. The Luria-Delbrück distribution (line) was fitted to these observed frequencies, allowing estimation of the misincorporation rate of the polymerase.

Fig. 3.

Identification of rare subclones at the IGH locus in patients with CLL. (A) Pyrosequencing traces from the dominant clone (ii) and 2 related subclones (i and iii) in a patient with CLL show closely linked base substitutions compared with the dominant clone: T>A; C>T; T>C in subclone 1 and C>G; T>C; A>G; A>C in subclone 2. The dominant clone has several mutations compared with the germ line, marked with an asterisk, and several of the variants in the subclones are the germ line sequence (e.g., T>A and T>C in subclone 1). The 2 subclones share the T>C variant, suggesting their clonal relationship. (B) Histogram showing the number of observed clones per sample, with the 2 sample codes (PD2087a and PD2106a) indicating the patients for whom phylogenetic trees were generated in Fig. 4. (C) Histogram showing the number of bases different between each subclone and the dominant clone for that patient.

Fig. 4.

Trees showing the phylogenetic interrelationships among the subclones for 2 patients, (A) PD2087a and (B) PD2106a. The trees were fitted using unrooted parsimony methods, and the length of each branch is proportional to the number of varying bases (evolutionary distance). The number shown beside each intermediate branch is the percentage support across 1000 bootstrap samples.