On the design of CRISPR-based single-cell molecular screens

Abstract

Several groups recently coupled CRISPR perturbations and single-cell RNA-seq for pooled genetic screens. We demonstrate that vector designs of these studies are susceptible to ∼50% swapping of guide RNA-barcode associations because of lentiviral template switching. We optimized a published alternative, CROP-seq, in which the guide RNA also serves as the barcode, and here confirm that this strategy performs robustly and doubled the rate at which guides are assigned to cells to 94%.

Figure 1

Template switching during lentiviral packaging decreases the sensitivity of designs relying on cis-pairing of sgRNAs and distal barcodes. A) Generalized schematic of vectors that rely on cis pairing of sgRNAs and barcodes such as Peturb-seq, CRISP-seq, and MOSAIC-seq. A barcode is linked to an sgRNA separated by a distance of 2.4 kb or more depending on the specific design. This barcode is expressed as part of the Pol II transcript and sequenced as a proxy for the guide sequence. B) Generalized schematic of CROP-seq approach. One copy of the guide is cloned into the 3′ LTR of the vector. The guide sequence itself is therefore transcribed as part of both the Pol II transcript and can therefore be sequenced directly. A second copy of the guide expression cassette is produced in the 5′ LTR during lentivirus positive strand synthesis prior to integration. C) Schematic of experiment performed to quantify template switching rate at 2.4 kb separation between sequences. Distinguishing bases (3 bp differences) in GFP and BFP are separated from their respective barcodes by 2.4 kb. D) Cells were transduced with GFP or BFP virus separately or a virus generated from a mix of GFP/BFP produced from individual or combined lentiviral packaging. As an additional control, cells transduced with GFP or BFP only virus were mixed prior to sorting. Cells were sorted on GFP and BFP and the percent GFP and BFP barcodes in each sample is shown as a table. Note that in a mix of two plasmids only approximately half of all chimeric products are detectable due to homozygous virions (see Online Methods). E) Plot of sum of squared errors of observed data vs. expected values at various swap rates using the collected fraction GFP and BFP barcodes in the green and blue sorted samples (n = 4 measurements), assuming a relative proportion of 61.7% GFP+ cells as determined from FACS (see Supplementary Figure 4 for and methods for details). F) Transcription factor pilot screen from the arrayed screen of Adamson et al. (used here as a gold standard performed with arrayed lentivirus production) subjected to simulation of progressively higher fractions of target assignment swapping. Number of differentially expressed genes across the target label at FDR of 5% is plotted at each swap rate for 10 samplings per swap rate (n = 5321 cells used in tests). 0.5 corresponds to the 50% swap rate determined via FACS.

Figure 2

CROP-Seq screen of tumor suppressors with high capture rate by PCR enrichment, and assessment of alternate sgRNA placement within a pol II 3′UTR. A) Determination of the most likely multiplicity of infection and capture rate of barcoded transcripts in CROP-seq screen based on a generative model. B) tSNE embedding of a doxorubicin treated sample with colors corresponding to cells with guides to TP53, cells that contain non-targeting controls (NTC), cells containing guides to non-TP53 targets, and cells that are unassigned (n = 5879 cells). All vectors have been modified to contain a CRISPRi-optimized backbone. pHAGE-scKO places the sgRNA within a Pol II 3′ UTR and fails to knockdown mCherry.C) CDKN1A and TP53I3 expression in cells expressing either non-targeting controls or guides to TP53. Cells with TP53 guides are further stratified into cells inside and outside of the TP53 enriched cluster from panel 2B. Values capped at a minimum value of 0.1 to facilitate plotting. D) CRISPRi knock-down of mCherry in MCF10A and K562 cells not expressing a guide (- control), KHH30 (+ control), CROP-seq, and pHAGE-scKO design. All vectors have been modified to contain a CRISPRi optimized backbone. pHAGE-scKO places the sgRNA within a Pol II 3'UTR and performs poorly.