Systematic evaluation of signal-to-noise ratio in variant detection from single cell genome multiple displacement amplification and exome sequencing

Anita T. Simonsen, Marcus C. Hansen, Eigil Kjeldsen, Peter L. Møller, Johnny J. Hindkjær, Peter Hokland, Anni Aggerholm

The current literature on single cell genomic analyses on the DNA level is conflicting regarding requirements for cell quality, amplification success rates, allelic dropouts and resolution, lacking a systematic comparison of multiple cell input down to the single cell. We hypothesized that such a correlation assay would provide an approach to address the latter issues, utilizing the leukemic cell line OCI-AML3 with a known set of genetic aberrations. By analyzing single and multiple cell replicates (2 to 50 cells) purified by micromanipulation and serial dilution we stringently assessed the signal-to-noise ratio (SNR) from single as well as a discrete number of cells based on a multiple displacement amplification method, with whole exome sequencing as signal readout. In this setting, known OCI-AML3 mutations as well as large copy number alterations could be identified, adding to the current knowledge of cytogenetic status. The presence of DNMT3A R882C, NPM1 W288 fs and NRAS Q61L was consistent, in spite of uneven allelic read depths. In contrast, at the level of single cells, we observed that one-third to half of all variants were not reproduced in the replicate sample, and this allelic mismatch displayed an exponential function of cell input. Large signature duplications were discernible from 5 cells, whereas deletions were visible down to the single cell. Thus, even under highly optimized conditions, single cell whole genome amplification and interpretation must be taken with considerable caution, given that allelic change is frequent and displays low SNR. Allelic noise is rapidly alleviated with increased cell input, and the SNR is doubled from 2 to 50 cells. In conclusion, we demonstrate noisy allele distributions, when analyzing genetic aberrations within single cells relative to multiple cells. Based on the presented data we recommend that single cell analyses should include replicate cell dilution assays for a given setup for relative assessment of procedure-specific SNR to ensure that the resolution supports the specific hypotheses.