CRISPR-SURF: discovering regulatory elements by deconvolution of CRISPR tiling screen data

Reporting Summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.

To the Editor –

Tiling screens that use CRISPR-Cas technologies provide a powerful approach for the mapping of regulatory elements to phenotypes of interest^1–6. Here we present CRISPR screening uncharacterized region function (CRISPR-SURF), a deconvolution framework that can be used to identify functional regulatory regions in the genome from data generated by CRISPR-Cas nuclease, CRISPR interference (CRISPRi), or CRISPR activation (CRISPRa) tiling screens. CRISPR-SURF can be run as a stand-alone command line utility (https://github.com/pinellolab/CRISPR-SURF) or as a web application (http://crisprsurf.pinellolab.org/) (Supplementary Note 1).

The methodology underlying the CRISPR-SURF framework leverages the concept that single guide RNAs (sgRNAs) represent a functional readout for base pairs within the perturbation range. This range depends on the CRISPR screening approach used: CRISPR-Cas nucleases introduce insertion and deletion (indel) mutations of varying lengths (typically < 30 bp, although potentially varying with cell type), whereas CRISPRi and CRISPRa strategies may remodel chromatin structure across hundreds of nucleotides. Importantly, each CRISPR technology offers its own advantage: CRISPRi and CRISPRa strategies increase the likelihood of detecting regulatory elements, given their larger perturbation ranges, whereas CRISPR-Cas nucleases provide higher resolution on the boundaries of regulatory elements, given their sharper perturbation windows. Because each sgRNA perturbs variable-size regions around its target site, the sgRNA data from CRISPR tiling screens can be seen as imprecise measurements of an underlying genomic regulatory signal. To address this variable, we model these imprecise measurements by means of a convolution operation that accounts for the perturbation profiles associated with different CRISPR technologies.

CRISPR-SURF deconvolves tiling screen data to find the genomic regulatory signal that best explains the observed sgRNA scores given the perturbation profile and sgRNA spacing (Fig. 1). The CRISPR-SURF framework accounts for overlapping perturbation profiles between neighboring sgRNAs and leverages shared information to infer the underlying genomic regulatory signal even from noisy measurements. The exact sgRNA targeting coordinates are also taken into account, thus allowing for location-dependent statistical tests with a power that reflects the local density of sgRNAs in a region. This enables CRISPR-SURF to estimate perturbation-specific and position-specific statistical power for CRISPR tiling screens (Supplementary Note 2).

Fig. 1 |. CRISPR-SURF deconvolution framework.

An illustration of the deconvolution based on sgRNA targeting positions, different perturbation profiles (CRISPRi/a and CRISPR-Cas), and enrichment scores.

We evaluated the performance of CRISPR-SURF by using three published CRISPR tiling screens spanning CRISPR-Cas9¹, CRISPRi², and CRISPRa³ modalities. For all three datasets, CRISPR-SURF reliably identified all of the experimentally validated regulatory elements. CRISPR-SURF further identified potentially novel regulatory regions supported by both chromatin accessibility and epigenetic marks (Supplementary Notes 3 and 4, Supplementary Figs. 1–3). We elaborate on key differences between CRISPR-SURF and the analysis methods used in these previous studies in Supplementary Notes 5 and 6.

Furthermore, we carried out two matched CRISPR tiling screens using CRISPR-Cas9 (SpCas9) and CRISPRi (dCas9-KRAB) on the BCL11A locus (Supplementary Note 7) and found that significant regions identified within previously validated functional enhancers^1,7,8 were narrower in the CRISPR- Cas9 screen than in the CRISPRi screen, consistent with the narrower perturbation profiles of CRISPR-Cas9 indel mutations compared with those of CRISPRi epigenetic modifications (Supplementary Fig. 4). In summary, CRISPR-SURF leverages the broad CRISPRi and CRISPRa perturbation profile for efficient enhancer discovery and the narrow CRISPR-Cas perturbation profile for high-resolution mapping of critical elements within enhancers.