Massively parallel digital transcriptional profiling of single cells

Key results

1. 01

   ShownGel-bead-in-emulsion (GEM) encapsulation across 8 parallel channels recovers ∼50% of loaded cells with low multiplet and barcode cross-talk rates; multiplets rise linearly with recovered cells, consistent with Poisson loading.

   ∼80% single-bead GEM fill; ∼100,000 GEMs per channel per ∼6-min run; ∼50% capture across four cell types at loads of ∼1,000 to ∼23,000 cells; 293T/3T3 mix: 1,012 cell-containing GEMs (482 human, 538 mouse), 1.6% inferred multiplets, 0.9% mean cross-species unique molecular identifier (UMI) counts; multiplets linear over 1,200 to 9,500 recovered cells. · Droplet-based platform enables barcoding of…; Technical demonstration with cell lines and…; Discussion

2. 02

   ShownCell-line and spike-in tests quantify sensitivity: UMI counts track input External RNA Controls Consortium (ERCC) molecule numbers with little GC or length bias, technical noise is estimated at ∼50% of total variance, and 293T cells at 1% of a Jurkat mix are recovered at the expected ratio.

   At 100k reads per cell, median ∼4,500 genes and ∼27,000 UMIs per human 293T or mouse 3T3 cell; ERCC mean UMIs vs expected molecules r=0.96; conversion 6.7–8.1% (ERCC Mix1/Mix2, 1:50) and 2.5 to 25.5% by droplet digital PCR for eight Jurkat genes; squared coefficient of variation ∼7% (ERCC) vs ∼11–19% (cells); 1:99 293T:Jurkat mix split at the expected ratio. · Technical demonstration with cell lines and…; Detection of individual populations in mixed…; ERCC assay; ddPCR assay

3. 03

   ShownAcross ∼68k fresh peripheral blood mononuclear cells (PBMCs) from one donor, unsupervised clustering resolved major immune lineages at expected ratios plus a rare megakaryocyte cluster, and a cryopreserved aliquot of the same donor's cells gave closely matching profiles.

   ∼20k reads per cell gave median ∼525 genes and 1,300 UMIs; k-means on 50 principal components yielded 10 clusters (>80% T, ∼6% natural killer, ∼6% B, ∼7% myeloid; megakaryocytes 0.3%); cells matched to 11 purified-subset reference profiles; frozen ∼3k-cell library vs fresh: mean expression r=0.96, gene/UMI counts P=0.8 and 0.1. · Subpopulation discovery from a large immune…; Single-cell RNA profiling of cryopreserved PBMCs; PCA and tSNE analysis of PBMCs; Cell classification analysis using purified…

4. 04

   ShownA Gibbs-sampled two-genome mixture model on single-nucleotide variants (SNVs) called de novo from transcripts assigns cells to individuals without prior genotypes, reliably down to a minor fraction of 3%.

   In silico Donor B/C PBMC mixes (12 ratios, 0 to 50%; 6,000 cells; ∼15k reads, ∼50 SNVs per cell): >95% sensitivity and positive predictive value down to a 3% minor fraction, none detected below 3%; physical 50:50, 90:10 and 99:1 mixes gave minor fractions of 43%, 12% and not detected (expected 50%, 10%, 1%); 293T/Jurkat SNV and marker calls 99% concordant. · Detection of individual populations in mixed…; Genotype-based method to detect individual cell…; SNV-based genotype assignment; Tables

5. 05

   SuggestedIn bone marrow from two acute myeloid leukemia (AML) transplant recipients, genotype groups agreed with clinical chimerism tests, and expression clusters revealed blasts, immature erythroids and granulocyte precursors consistent with one patient's relapse.

   AML027 post-transplant: genotype groups of 13.8% and 86.2%, the larger with 97% host overlap (clinical assay: 12% donor); AML035 post-transplant: one group with 78% host overlap, inferred all-donor, clinically confirmed (baseline overlap ∼98±0.3% same person, ∼73±2% different people); AML027 >80% erythroid pre-transplant, ∼20% immature granulocytes post-transplant. · Single-cell analysis of transplant bone marrow…; Discussion; Genotype comparison with the pure sample; Tables