Sequencing thousands of single-cell genomes with combinatorial indexing

Abstract

Single-cell genome sequencing has proven valuable for the detection of somatic variation, particularly in the context of tumor evolution. Current technologies suffer from high library construction costs, which restrict the number of cells that can be assessed and thus impose limitations on the ability to measure heterogeneity within a tissue. Here, we present single-cell combinatorial indexed sequencing (SCI-seq) as a means of simultaneously generating thousands of low-pass single-cell libraries for detection of somatic copy-number variants. We constructed libraries for 16,698 single cells from a combination of cultured cell lines, primate frontal cortex tissue and two human adenocarcinomas, and obtained a detailed assessment of subclonal variation within a pancreatic tumor.

Figure 1. Single cell combinatorial indexing with nucleosome depletion

(a) Single cell combinatorial indexing workflow. (b) Phase contrast images of intact nuclei generated by standard isolation followed by nucleosome depletion using Lithium Assisted Nucleosome Depletion (LAND) or crosslinking and SDS treatment (xSDS). Scale bar: 100 µm. (c) Nucleosome depletion produces genome-wide uniform coverage that is not restricted to sites of chromatin accessibility.

Figure 2. Comparison of LAND and xSDS nucleosome depletion methods with SCI-seq

(a) Complexity for one of six LAND SCI-seq preparations on GM12878. Right panel, histogram showing distribution of read counts. Dashed line represents single cell read cutoff. (b) As in (a) but for xSDS nucleosome depletion for one of three PCR plates. (c) Left, model built on downsampled reads for the GM12878 xSDS preparation and used to predict the full depth of coverage. Right, Projections for one of the LAND preparations and the full xSDS preparation. Shading represents standard deviation over multiple models. Points represent actual depth of sequencing. (d) Coverage uniformity scores for SCI-seq using LAND or xSDS and for quasi-random priming (QRP) and degenerate oligonucleotide PCR (DOP). (e) Summary of the percentage of cells showing aneuploidy at the chromosome arm level across all preparations with and without imposing a variance filter. (f) Karyotyping results of 50 GM12878 cells. (g–h) Summary of windowed copy number calls and clustering of GM12878 single cells produced using LAND (g) or xSDS (h). Top represents a chromosome-arm scale summary of gain or loss frequency for all cells, bottom is the clustered profile for cells that contain at least one CNV call.

Figure 3. Somatic CNVs in the Rhesus brain

(a) Three single cell examples showing copy number variants, and one representative euploid cell for the SCI-seq preparation (HMM). (b) Frequency of aneuploidy as determined by each of the methods with and without filtering.

Figure 4. SCI-seq analysis of a stage III human Pancreatic Ductal Adenocarcinoma (PDAC)

(a) SCI-seq library complexity. Right panel, histogram showing distribution of read counts. Dashed line represents single cell read cutoff. (b) Breakpoint calls (top) and breakpoint window matrix of log2 sequence depth ratio. (c) Principle component analysis and k-means clustering on breakpoint matrix. (d) 100 kbp resolution CNV calling on aggregated cells from each cluster. (e) Cluster specific CNVs and CEBPA amplification present in all clusters (k4 shown).