Whole-exome sequencing of circulating tumor cells provides a window into metastatic prostate cancer

Abstract

Comprehensive analyses of cancer genomes promise to inform prognoses and precise cancer treatments. A major barrier, however, is inaccessibility of metastatic tissue. A potential solution is to characterize circulating tumor cells (CTCs), but this requires overcoming the challenges of isolating rare cells and sequencing low-input material. Here we report an integrated process to isolate, qualify and sequence whole exomes of CTCs with high fidelity using a census-based sequencing strategy. Power calculations suggest that mapping of >99.995% of the standard exome is possible in CTCs. We validated our process in two patients with prostate cancer, including one for whom we sequenced CTCs, a lymph node metastasis and nine cores of the primary tumor. Fifty-one of 73 CTC mutations (70%) were present in matched tissue. Moreover, we identified 10 early trunk and 56 metastatic trunk mutations in the non-CTC tumor samples and found 90% and 73% of these mutations, respectively, in CTC exomes. This study establishes a foundation for CTC genomics in the clinic.

Figure 1

Experimental process for sequencing of CTCs. (a) Schematic of workflow for the enrichment, isolation and sequencing of CTCs. (b) Sample micrographs of CTCs isolated in nanowells are shown with matched transmitted light (T.L.) and immunophenotyping for EpCAM, CD45, and DAPI by epifluorescence. Scale bar denotes 50 µm. (c) Scatter plot of the number of CTCs enumerated versus levels of PSA from 51 blood samples from 36 prostate cancer patients (Supplementary Table 1) screened using the MagSweeper for enrichment. CTC numbers in blood correlated with PSA levels (p = 0.004; Spearman, two-tailed). (d) Scatter plot of the percentage of target bases covered > 20× from whole exome sequencing (WES) versus the autocorrelation coefficient (Online Methods) calculated from low pass whole genome sequencing (WGS) over chromosome 1 for patient CRPC_36 (p < 0.0001; Pearson, two-tailed). WES yielded 124 ± 12× mean target coverage (Supplementary Table 2). WGS yielded mean coverage over chromosome 1 between 0.0003× and 0.03×, with a median of 0.017×. (e) Genome-wide read densities (1 Mb bins) from low pass WGS of CTC libraries from four different patients (CRPC_10, CRPC_12, CRPC_35, CRPC_36). Examples of three quality libraries and one poor library are shown per patient. The log of the inverse correlation coefficient was used to select high-quality libraries, with a cut off of −1.8 used here.

Figure 2

Census-based variant calling from whole exome sequencing of CTCs from patient CRPC_36. (a) Characterization of allelic coverage in each CTC sequencing library from the same patient, compared to those libraries combined and primary tumor, as determined by 22,054 germline heterozygous SNP sites; an allele was scored as covered if there were ≥ 3 total reads of the particular allele(s). For reference, the autocorrelation coefficient is plotted below all CTC libraries except for three CTC libraries (n.d., not determined) that had insufficient low pass WGS coverage but passed quality control prior to exome sequencing based on visual inspection of genome-wide read densities (Supplementary Fig. 6). Coverage of the alternate allele (either alternate alone or both alleles) at germline heterozygous SNPs was correlated with the autocorrelation metric for individual CTC libraries (p < 0.0001; Spearman, two-tailed). When the individual CTC libraries were combined (“combined CTCs”), 99.995% of sites were covered by both alleles, similar to bulk sequencing of the primary tumor. (b) Estimation of false positive rate / Mb among 19 independent CTC libraries after requiring the variant to be observed in at least N independent CTC libraries (Supplementary Fig. 10). Grey dashed line indicates the reported mutation rate in bulk tumor sequencing of treated prostate cancer (~2 / Mb); black arrow head indicates the false positive rate / Mb observed for a single CTC library. Inset shows sensitivity versus false positive rate / Mb as a function of the required number N of independent observations of the variant. (c) The number of SSNVs called in total among 19 CTC libraries (73) and those that were validated as being present in matched tumor tissue (51) are shown. (d) Relative sensitivity to call CTC SSNVs (fraction of the total number called using 19 CTC libraries) as a function of the number of libraries sequenced, ranked in order by the autocorrelation coefficient (blue bars). A sustained improvement in sensitivity was observed. Additionally, considering only the 51 CTC SSNVs also observed in bulk whole exome sequencing of matched tumor tissue, we observed a very similar increase in sensitivity for each additional library sequenced (grey bars).

Figure 3

Comparison of mutation pattern across CTCs, primary cores and metastasized tumor from patient CRPC_36. (a) FDG-PET and bone scans show widespread metastatic disease. FDG-PET Maximum Intensity Projection (MIP) image (top left) and axial FDG-PET slice (bottom left) demonstrate multifocal FDG-avid skeletal metastases throughout the axial and appendicular skeleton as well as bilateral cervical, left supraclavicular (arrow), retroperitoneal and bilateral common iliac metastatic lymphadenopathy. Bone scan demonstrates widespread bone metastasis. (b) Venn diagram representing mutations called in the CTCs and metastasis. Of note, 51% of mutations in the metastasis were called in CTCs. (c) Hierarchical clustering using the Jaccard index for mutations called across the nine primary cores, metastasized tumor and CTCs (when observed in ≥3 out of 19 single CTCs). Only sites in the exomes that were considered to be powered for mutation calling, as described in Online Methods, were included in this analysis. Shading of green represents presence in CTCs and at least one other sample (dark green) or not present in CTCs (light green). Genes highlighted indicate non-synonymous mutations present in >2 patients from a previous sequencing study in prostate cancer. Of note, one of the cores included regions of both Gleason 3 and Gleason 5 cancer. (d) Dendrogram representing hierarchical clustering by the Jaccard index, and timeline of sample acquisition. SSNVs detected in all individual cores of tissue (early trunk), or in all cores that belong to only one of the two branchpoints of the clustering dendrogram are listed. Non-synonymous mutations are highlighted in bold with “*”. The areas shaded in pink represent the pathology blocks from which cores of tissue were obtained (drawn to scale). The dotted lines represent the area with histological presence of tumor within each block. The sites from which the individual cores of tissue were obtained for sequencing are displayed in colors corresponding to the cluster dendrogram and Fig. 3c. The regions of the prostate from which the pathology blocks were retrieved are schematized. (e) The number of mutations found in the metastasis and at least one core of the primary tumor (metastatic trunk), the early trunk mutations, and the overlap of these with CTC mutations are shown, excluding sites that were consistently underpowered in greater than half of the samples. The one early trunk mutation not detected in CTCs (N ≥ 3) was observed in 2 CTCs.