Tissue Culture

HAP1 cells were purchased from Horizon Discovery and cultured in Iscove’s modified Dulbecco’s medium with L-glutamine and 25 mM HEPES (GIBCO). The HAP1 cell line was derived from the near-haploid KBM7 line (male cells of chronic myelogenous leukemia origin) by the introduction of induced pluripotent stem cell factors. Despite the cell line’s male origin, HAP1 cells no longer hold a Y chromosome.³⁶ HEK293T cells were purchased from ATCC and cultured in Dulbecco’s modified Eagle’s medium with high glucose and sodium pyruvate (LifeTechnologies). Both media were supplemented with 10% fetal bovine serum (Rocky Mountain Biologicals) and 1% penicillin-streptomycin (GIBCO) and grown with 5% CO₂ at 37°C.

gRNA Library Design

To generate a list of gRNAs, we identified all 20 bp protospacers followed by a 5′-NGG PAM sequence from chrX: 133,507,694–133,713,798 (UCSC Genome Browser build hg19). We then excluded protospacers that had a perfect sequence match elsewhere in the genome and scored the remaining gRNAs for both on-target and off-target activity. We considered off-target sequences that had five or fewer mismatches to the putative gRNA and calculated an aggregate off-target score by using the method of Hsu et al.³⁷ In addition, we scored each site for on-target efficiency.³⁸ We matched final deletion pairs by using spacers that did not contain BsmBI restriction sites, were not predicted to have off-target hits in other 6TG resistance genes or in KBM7 essential genes (the HAP1 parental cell line), were greater than 25 bp apart, were further than 50 bp from an exon, and passed on-target (above 10) and off-target (above 25) thresholds. Contrastingly, the library of individual gRNAs included all of the spacers targeting the same region, excluding those predicted to have 2,000 or more off-targets or to have off-targets with four or fewer mismatches within the targeted HPRT1 region.

Building the Library of gRNA Pairs

This library cloning method was developed in parallel with similar recently published methods³⁹ and was modified from the GeCKO individual-gRNA cloning scheme.^{35, 40} First, the lentiGuide-Puro backbone (Addgene 52963) was digested with BsmBI (FastDigest Esp3I, Thermo Fisher Scientific) and gel purified. The paired spacers (flanked with lentiGuide-Puro overlap sequences) were synthesized twice on a microarray (CustomArray) such that each pairing was represented in both possible orders (Figure S1).

To ensure quality of array synthesis, we amplified 1 ng of the oligonucleotide (oligo) pool with Kapa HiFi Hotstart ReadyMix (KHF, Kapa Biosystems) and ran it on a gel to confirm that the oligos were of the expected 108 bp length. After PCR purification with Agencourt AMPure XP beads (Beckman Coulter), the amplicon was cloned into lentiGuide-Puro with In-Fusion HD Cloning Plus (Clontech) and transformed into Stable Competent E. coli (NEB C3040H) for minimizing repeat-based recombination of the lentivirus. This ensuing library (lentiGuide-Puro-2×Spacers) now contained each pair of spacers but was still missing the additional gRNA backbone and PolIII promoter.

We next cloned in the additional gRNA backbone and H1 promoter between each spacer pairing to enable expression of the two independent gRNAs. We ordered the gRNA-backbone-H1-promoter fragment as a gBlock (IDT) with flanking BsmBI sites to allow ligation into the BsmBI-digested lentiGuide-Puro-2×Spacers library. The gBlock and the lentiGuide-Puro-2×Spacers were each digested with BsmBI, purified, ligated together with Quick Ligase (NEB M2200S), and transformed into Stable Competent E. coli for the generation of a final lentiGuide-Puro-2×gRNA library.

To prevent bottlenecking of the library, we performed these cloning steps with enough replicates at high efficiency to maintain a minimum of 20× average library coverage (in relation to the expected library complexity). Sequencing of the lentiGuide-Puro-2×gRNA library revealed 97.8% retention of diversity from the designed paired spacers. However, 16% of library reads held unprogrammed, interswapped pairs. 88.5% of these swaps were seen only in a single read, implying that a more likely cause was template switching during either PCR or cluster generation. For all experimental analyses, only reads of gRNA pairs that perfectly matched programmed pairs were considered.

Building the Library of Individual gRNAs

The spacers of this library were similarly synthesized on an array, amplified, and purified as above. The lentiGuide-Puro backbone was linearized as above, and the library was cloned into it with NEBuilder HiFi DNA Assembly Master Mix (NEB). This plasmid was transformed into Stable Competent E. coli, generating enough transformants for 30× average coverage. This method produced 98.5% retention of complexity from the designed array.

Lentiviral Library Production, Delivery, and 6-Thioguanine Selection

We produced lentivirus with Lipofectamine 3000 (Life Technologies) to transfect HEK293T cells with the lentiviral vector libraries made above and third-generation packaging plasmids (pMDLg/pRRE Addgene 12251, pRSV-Rev Addgene 12253, and pMD2.G Addgene 12259). Supernatant was collected 72 hr after transfection, centrifuged at 300 rcf for 5 min for the removal of cell debris, and passed through a 0.45 μm syringe filter.

For the creation of a monoclonal HAP1 cell line stably expressing Cas9, HAP1 cells were transduced with lentivirus produced with lentiCas9-Blast (Addgene 52962), selected with 5 μg/mL Blasticidin (Thermo Fisher Scientific), and single-cell sorted via fluorescence-activated cell sorting (FACS).

HAP1-Cas9-Blast monoclonal cells were plated to be at 30% confluency on the day of lentiviral gRNA-pair transduction. For transduction, 5% of the recipient cells’ media was replaced with filtered virus, limiting the MOI to <0.3. Media were changed after 24 hr, and selection for transduced cells began 48 hr after transduction. Puromycin was added at 2 μg/mL for 2 days for the assessment of the percentage of cells transduced, and then cells were maintained in 1 μg/mL for 5 more days.

After puromycin treatment, an initial population of cells was collected. Selection for loss of HPRT function was performed by application of 5 uM 6TG to the remaining cells at <50% confluency for 7 days. An additional concern was that minor gene-expression changes caused by ScanDel-mediated mutations in regulatory elements might not be strong enough to confer resistance. To mitigate this, we used the lowest dosage of 6TG that completed HAP1 selection after 7 days. 6TG concentrations of 6–60 μM are reported in the literature to achieve effective selection in this time frame, depending on cell type.^{41, 42} We tested our monoclonal HAP1-lenti-Cas9-Blast line at concentrations just below this range (1, 2.5, and 5 μM 6TG). After 7 days, the 5 μM treatment had no readily identifiable surviving cells, whereas the 2.5 μM treatment retained a sparse population, and the 1 μM treatment produced appreciably more outgrowing colonies. On the basis of these results, we proceeded with selections by using 6TG at 5 μM for 7 days. Enough cells were transduced and sampled at each time point to maintain a minimum 2,000× average coverage of the library in each population.

Sequencing of the baseline (i.e., pre-6TG) population revealed that 98.4% diversity of the lentiGuide-Puro-2×gRNA library was preserved from replicate 1, and replicate 2 retained 78.8%. Because our deletions highly overlapped, we proceeded with replicate 2 because all base pairs were interrogated despite the lower diversity. We observed 95.6% retention of programmed library diversity in replicate 1 of the individual-gRNA plasmid library and 71.2% of replicate 2.

Interswapped gRNA pairs were observed in 35.5% of reads from the baseline pre-6TG sample. This is an increase from the 16% observed in reads from the lentiGuide-Puro-2×gRNA plasmid library. This suggests additional template switching during the library’s amplification from gDNA, which would require more cycles of PCR. However, because we directly sequenced each gRNA spacer as a readout instead of using barcoded libraries³⁰ and only took exact sequence matches, this did not pose a problem.

gRNA Library Amplification and Sequencing from HAP1 Cells

gDNA was extracted from the cells sampled before and after 6TG selection with the DNeasy Blood & Tissue kit (QIAGEN). KHF was used for all amplification steps. The libraries were initially amplified from a minimum of 6 μg of gDNA divided across thirty 50 μL reactions, ensuring sampling of about two million haploid genome equivalents at each time point. We performed two additional PCRs, and AMPure bead purification between each reaction, to add sequencing adapters and sample indices to the amplicon. We optimized amplification conditions by qPCR to minimize overamplification of the construct.

Sequencing was performed on an Illumina MiSeq with a 50-cycle kit. Read 1 and the Illumina index read were used for sequencing the two gRNAs in the paired-gRNA construct before paired-end turnaround, and read 2 was used for sequencing the 9 bp sample index.

Calculation of a Selection Score Assignment per Base Pair

Custom Python scripts counted tallies of gRNAs (for experiments with the library of individual gRNAs) or gRNA pairs before and after selection. These counts were normalized to the total number of reads per sample. We calculated an enrichment ratio for each gRNA pair by dividing its normalized read count after selection by its read count before selection. A selection score is the log₁₀ of the enrichment ratio: log₁₀(after/before). If a gRNA or gRNA pair was absent before selection, it was excluded from further analysis. Any gRNA pairs that used a self-paired gRNA with an independent selection ratio > 0 were also excluded from further analysis.

If a gRNA pair is absent after 6TG selection, its calculated selection score will be a negative number that is relatively large in magnitude and somewhat arbitrarily determined by the number of pre-selection reads. Thus, to limit the contribution of these scores to average measurements derived from many independent deletions, we set a minimum selection score equal to the middle of the bimodal distribution between the positively and negatively selected deletions of each replicate (Figure S2). For example, in ScanDel replicate 1, if the log₁₀ value of a selection score was less than −0.35, that gRNA pair’s score was set to −0.35. We assigned each individual base pair a per-base-pair selection score by taking the mean of all deletions programmed to cover that base pair. The per-base-pair score was normalized to the median score for all positive scores in that replicate. We averaged the per base-pair selection score of each replicate to get the final selection score per base pair. Per-base-pair scores were uploaded as a bedgraph for visualization on the UCSC Genome Browser.

For the individual-gRNA mutagenesis screen, we calculated selection scores per base pair similarly by assuming that a 10 bp deletion was made by each gRNA queried. If a base pair was scored at the minimum negative threshold in one screen, it was given that value for the consensus selection score of the two replicates.

Bulk ATAC-Seq of HAP1 Cells

Two biological replicates were separately maintained (on 10 cm dishes and split 1:10 three times per week) and processed separately. Chromatin accessibility in the HAP1 cell line was profiled with the ATAC-seq (assay for transposase-accessible chromatin with high-throughput sequencing) protocol⁴³ with slight modifications. The media for 10 cm plates of confluent HAP1 cells were aspirated and replaced with 2 mL of ice-cold lysis buffer (CLB+; made as described in Buenrostro et al.⁴³ but supplemented with protease inhibitors [Sigma cat. no. P8340]). Cells were incubated on ice for 10 min in CLB+ and then were dislodged with a cell scraper and transferred to a 15 mL conical tube and pelleted at 500 rcf for 5 min at 4°C. Nuclei were re-suspended in 1 mL of CLB+ and counted on a hemocytometer. 50,000 nuclei in 22.5 μL of CLB+ were combined with 2.5 μL of TDE1 enzyme and 25 μL of TD buffer (Illumina). Tagmentation conditions were as described in Buenrostro et al.⁴³ (37°C for 30 min). After MinElute purification into 10 μL EB buffer (QIAGEN), 5 μL of tagmented DNA was amplified in 25 μL reactions for 12 cycles with NEBNext Master Mix (NEB). Reactions were monitored with SYBR Green for ensuring that samples were not overamplified. PCR products were cleaned once with a QiaQuick PCR Cleanup Kit (QIAGEN) and once with 1× AMPure beads (Agencourt). The quality of the library was assessed on a 6% TBE gel, and the yield was measured by a Qubit (1.0) fluorometer (Invitrogen).

Samples were sequenced on two paired-end Illumina NextSeq 500 runs. Read lengths were 2 × 75 bp for the first run and 2 × 151 bp for the second run, so the second run was truncated to 75 bp. Sequencing reads were also trimmed for readthrough of adaptor sequences and quality with Trimmomatic⁴⁴ (“NexteraPE-PE.fa:2:30:10:1:true TRAILING:3 SLIDINGWINDOW:4:10 MINLEN:20” parameters) and then mapped to the 1000 Genomes integrated reference genome “hs37d5” with bowtie2⁴⁵ and the “-X 2000 -3 1” parameters. Only properly paired and uniquely mapped reads with a mapping quality above 10 were retained (“samtools -f3 -F12 -q10”). Reads mapping to the mitochondrial genome and non-chromosomal contigs were also filtered out. In addition, duplicate reads were removed with Picard. After checking quality-control metrics on the individual replicates, we combined reads from the two libraries for downstream analysis. Hypersensitive sites were called (at a 1% false-discovery rate) with the Hotspot algorithm.⁴⁶

Validation and Direct Genotyping of Positive Signal from the Screens

gRNA pairs that drove the ScanDel signal surrounding HPRT1’s first exon were cloned into simple lentiGuide-Puro-2×gRNA libraries. The transcription start site (TSS) ScanDel validation library contained four pairs, and the intron 1 library contained five (Tables S1 and S2). For the TSS sub-library validating the individual-gRNA screen, ten gRNAs were cloned into lentiGuide-Puro (Table S3). These constructs were lentivirally delivered to HAP1-Cas9-Blast cells and selected with 6TG, and gDNA was extracted as described above.

Because the expected deletions could remove up to 3 kb, the loci were sequenced with a Pacific Biosciences RSII (University of Washington PacBio Sequencing Services, P6C4 chemistry, RSII platform). To prepare libraries for PacBio sequencing, we amplified the TSS- or intron 1-targeted regions from 800 ng of gDNA each by using four 50 μL KHF reactions with primers adding sample indices and SbfI or NotI cut sites. The purified amplicons (Zymo Research DNA Clean & Concentrator-5) were digested with SbfI-HF (NEB) and NotI-HF (NEB), leaving sticky ends. 5′-phosphorylated SMRT-bell hairpin oligos (IDT) containing the PacBio priming site, hairpin-forming sequence, and resulting sticky ends for either SbfI or NotI were annealed by heating to 85°C and snap frozen in 10 mM Tris 8.5, 0.1 mM EDTA, and 100 mM NaCl. These were ligated at 10× molar excess to the digested amplicons, destroying the restriction site once attached. For the removal of undigested amplicons and primers, this ligation was performed in the presence of further SbfI and NotI and was followed by treatment with Exo7 (Affymetrix) and Exo3 (Enzymatics).

Only reads with over five circular consensus sequence passes and containing the expected first twelve 5′ and 3′ base pairs of the amplicon were used for further analysis. Reads positive for complex inversions (≥100 bp) were removed from the library by the Waterman-Eggert algorithm with match, mismatch, gap open, and gap extend scores of 2, 10, 10, and 5, respectively.⁴⁷ The resulting reads were then aligned to the amplicon reference with the NEEDLEALL⁴⁸ aligner with a gap-open penalty of 10 and a gap-extension penalty of 0.5. Insertions were required to start within a window of 5 bp up- or downstream of the putative cut site. Deletions were required to either start or end within the same 10 bp window or span the window. Reads that carried the same edit pattern were collapsed into haplotypes, and figures were generated with a custom D3 script.

Comparing Deletion Rates of U6-H1 and U6-U6

We chose two protospacers to program a 365 bp deletion within the second intron of HPRT1 and cloned their spacers into a U6-H1 construct and a U6-U6 construct (Figure S3). Virus was produced and delivered to cells, which were selected with puromycin, and gDNA was extracted as described above. The locus was amplified in four successive rounds of nested PCR. The first reaction was only three cycles and included a forward primer with a 10 bp unique molecular index (UMI). The second reaction amplified any UMI-tagged fragments. The third and fourth reactions added sample indices and Illumina flow-cell adapters. The products were cleaned by AMPure between each reaction at a concentration that would lose primer dimer but retain the smaller deletion-holding fragments and were sequenced on a MiSeq. Any reads that contained the same UMI or edit pattern were collapsed by custom scripts, and their alignments were visualized with the same D3 script as above.

Development of ScanDel

In genome-wide CRISPR/Cas9 screens, a gRNA library is lentivirally delivered to a large pool of cells at a low MOI, such that each infected cell is likely to receive only one gRNA.^{40, 49, 50} With the goal of perturbing the function of the targeted locus, each gRNA induces NHEJ-mediated indels centered at the Cas9-mediated cleavage position within the target sequence. However, given the small and variable length of indels, the robustness of perturbation is inherently limited, particularly when non-coding sequences in which frameshifts are irrelevant are being targeted. To instead program a kilobase-scale deletion in each cell, we devised the following approach (Figure 1). First, gRNA pairs are designed to program specific deletions (each gRNA specifies one of the deletion’s boundaries; Figure 1A), and the corresponding pairs of 20 bp spacers are synthesized in cis on a microarray (Figure 1B). Second, the paired spacers are inserted into the lentiGuide-Puro plasmid between the U6 promoter and the gRNA backbone. Third, a second gRNA backbone and a second RNA polymerase (Pol) III promoter (H1 or U6) are inserted between the paired spacers. Fourth, libraries of gRNA pairs are lentivirally delivered to a large pool of cells at a low MOI, such that each cell receives a pair of gRNAs that program a single deletion (Figure 1C). Finally, analogous to conventional genome-wide CRISPR/Cas9 screens, deep sequencing of the integrated gRNA pairs is used as a surrogate measure of the prevalence of each programmed deletion in a population of cells (e.g., before and after the cells have been subjected to functional selection), thus capturing the phenotypic consequences of individual deletions.

As an initial test of our paired guide system, we compared the efficacy of using two different promoters for the two guides (a U6-H1 system) with that of using two copies of the same promoter (U6-U6). We tested these lentiviral gRNA-pair expression constructs by targeting the same genomic site for deletion with each system (Figure S3). We performed PCR amplification of the site with UMIs in order to minimize biases related to amplicon size (see Material and Methods). The U6-H1 system induced more programmed deletions than the U6-U6 system (20% versus 10% of reads from cells 1 week after transduction). The U6-H1 system has several advantages (e.g., it avoids recombination between the two U6 promoters during cloning and has a unique primer design for deep sequencing of each gRNA), and we therefore proceeded with it.

An important caveat for ScanDel, in relation to conventional gRNA cell-based screens, is that deletions programmed by gRNA pairs occur only in a minority of cells;^{51, 52} the other major outcomes are small NHEJ-mediated indels at one or both gRNA-targeted sites. For example, in our test of the U6-H1 system, the programmed deletion was found in 32% of cells that had any edit, whereas the remaining edited cells were mutated at one or both gRNA-targeted sites but retained the intervening sequence. Although this complicates interpretation, the problem can be overcome by a robust functional assay in conjunction with multiple, independent gRNA pairs that query the same genomic region, as well as by the inclusion of unpaired gRNA controls to ensure that observed effects do not occur with the individual gRNAs that make up each pair (but rather are dependent on the presence of both gRNAs).

Application of ScanDel to Survey the 206 kb Region Surrounding HPRT1

With the goal of investigating the potential of non-coding mutations to compromise its function, we applied ScanDel to a 206 kb X chromosome region centered on HPRT1 (Figures 1A and 2A). We designed pairs of gRNAs that programmed deletions tiling across the 206 kb region, including tiles that overlapped HPRT1 exons, in order to allow coding regions to serve as positive controls. Given that deletion length has been shown to affect deletion rate,⁵¹ deletions were programmed to be consistently either ∼1 or ∼2 kb in length (Figure 1A). This design resulted in 4,342 programmed deletions that tiled across the region, and they collectively covered each base pair a median of 27 times (Figure 2B). Testing each base pair with numerous independently programmed tiling deletions is expected to reduce noise and also increase resolution (given that all successfully made deletions tiling a critical regulatory element should exhibit positive selection). However, to guard against the possibility that the effects of individual gRNAs could confound analysis (e.g., via off-target mutations or on-target small ∼10 bp indels), we also included all spacers in the library as pairs with themselves (“self-pairs”; Figure 1B, inset; Figure S1). Additionally, we included 330 negative-control gRNA pairs not expected to survive 6TG selection, because they program deletions in non-genic regions far from HPRT1 or use spacers made of random sequence not present in the reference genome (hg19).

The gRNA-pair library was array synthesized, cloned, and delivered via lentiviral infection to HAP1 cells in replicate (Figures 1B and 1C). Cell populations were sampled before and after 1 week of the 5 μM 6TG selection, and PCR amplification and deep sequencing of gRNA pairs were used to quantify abundance at each time point. The functional selection score was calculated as the log₁₀ ratio of normalized read counts after selection to those before 6TG treatment (“selection score” as log₁₀(after/before 6TG)). Positively scoring self-paired spacers were flagged, and gRNA pairs that used these flagged spacers were excluded from further analysis (11% of pairs in replicate 1 and 3% of pairs in replicate 2). To integrate signal from overlapping programmed deletions, we calculated a “per-base-pair” metric as the mean of selection scores of all deletions overlapping a given base (Figure 2D and Material and Methods). This per-base-pair score across the HPRT1 locus was well correlated between biological replicates (Pearson: 0.708; Figure S4). Importantly, none of the negative-control gRNA pairs that were sampled in each of the two replicates were positively selected in both experiments (Figure S5).

Crucially, all nine HPRT1 exons exhibited strong functional scores, confirming the sensitivity of ScanDel as applied here to detect sequences essential to HPRT1 function (Figure S6). However, all reproducibly positive non-coding signal across the 206 kb region was immediately proximal to an HPRT1 exon. This result suggests that no distal regulatory element in the 206 kb region is essential to HPRT1 expression in HAP1 cells.

Near exons, non-coding regions exhibiting positive signal did so even when deletions that also overlapped the exons themselves were excluded from the analysis (Figure S6D). This suggested the presence of essential, proximal regulatory sequences. We noted that the positively scoring regions immediately upstream and downstream of the first exon overlapped a region of open chromatin identified by ATAC-seq in HAP1 cells, supporting the region’s role in gene regulation (Figure 2C and Figures S6 and S7). Together, these observations motivated us to attempt validation experiments for this region with the goal of directly confirming which deletions of putative regulatory elements were impairing HPRT1 function (Figures 3A and 3E).

Direct Genotyping of Deletions That Survive Functional Selection

With the goal of validating the positive signal upstream of the first exon, we repeated the experiment with a small pool of four gRNA pairs targeting the putative HPRT1 promoter (Figure 3B). We then amplified 3 kb of this region by PCR and performed long-read sequencing of the amplicons (Pacific Biosciences). As expected, before 6TG selection, the programmed deletions were all well represented in the population, although deletions with boundaries deviating from Cas9 cut sites (i.e., “unprogrammed”) were also detected (Figure 3C). However, after selection with 6TG, deletions with unprogrammed boundaries predominated, including those unseen before 6TG and those that extended beyond the TSS (Figure 3D). The fact that these initially rare deletions were strongly selected (whereas 2 kb promoter deletions that did not cross the TSS were not) suggests that even relatively proximal sequences upstream of the HPRT1 TSS are not strictly essential for expression. On the basis of the results of these validation experiments, we conclude that only a narrow window of non-coding sequence immediately upstream of the TSS and 5′ UTR is required for HPRT1 expression.

We next sought to validate the positive signal downstream of the first exon. To do so, we again repeated the experiment with a small pool of just five gRNA pairs targeting the first ∼2.7 kb of intron 1 (Figure 3F). We then amplified the region and again performed long-read sequencing of the amplicons (Pacific Biosciences). As with the promoter, the programmed deletions were all well represented before 6TG selection, although deletions with unprogrammed boundaries were also detected at a low rate (Figure 3G). After selection, deletions with unprogrammed boundaries predominated again, particularly those that extended into the first exon, thereby disrupting coding sequences (Figure 3H). A low rate of non-exonic deletions survived after 6TG, but these were present at the same level as unedited reads, implying that there could be some other explanation for 6TG resistance in these cells. Thus, as with the promoter, the positive signals that we originally observed for deletions in the first intron were most likely a result of the positive selection of rare “on-target-but-with-incorrect-boundaries” deletions that extended into the first HPRT1 exon.

An Individual-gRNA Screen of the Same Region for Comparison with ScanDel

We next compared our ScanDel results against a more conventional screen relying on only individual gRNAs^{16, 17, 18, 19, 21} (Figure 2E). For this, we cloned a second lentiviral library consisting of 12,151 individual gRNAs targeting the same 206 kb region and assayed HPRT function in HAP1 cells as previously. Under the assumption that each individual gRNA potentially disrupts a ∼10 bp region, this experiment at best interrogates ∼70% of bases within the 206 kb region as a result of the sparsity of PAM sites (by comparison, ScanDel covers each base pair in the entire locus at a median ∼27-fold redundancy). 86% of exon-targeting gRNAs were positively selected, and exonic selection scores were well correlated between biological replicates (Pearson: 0.781). Of 612 negative-control gRNAs, none that were sampled in each replicate were positively selected in both experiments (Figure S8). In non-coding sequence, scores were poorly correlated between biological replicates, and there was a paucity of reproducible, positively selected signal (Pearson: 0.156; Figure S9).

Notably, we did observe a greater proportion of positively scoring gRNAs in the vicinity of exons—i.e., whereas only 2% of intergenic gRNAs were positively selected, 7.5% of deep intronic (>2 kb away from an exon boundary) and 20.5% of proximal intronic (<2 kb from an exon boundary) gRNAs were positively selected (Figure 4A). Given our earlier observation with ScanDel of rare “on-target-but-with-incorrect-boundaries” deletions that were confounding when targeting near exon boundaries, we next performed similar validation experiments on individual gRNAs that targeted non-coding sequences near exons (Figure 4B). We chose ten gRNAs in the HPRT1 promoter region (Figure 4C) and repeated the individual-gRNA experiment with a small pool of just these ten gRNAs, again by using long reads (Pacific Biosciences) to sequence the locus before (Figure 4D) and after (Figure 4E) 6TG selection. Similar to our results with ScanDel in this region, the only mutations that survived 6TG selection were initially rare deletions whose boundaries extended past the TSS and into the 5′ UTR and/or coding sequence (Figure 4D). This result strongly underscores that caution should be exercised in the interpretation of results from CRISPR-based screens of non-coding regions, whether performed with individual gRNAs or gRNA pairs, and the importance of sequencing-based validation of edited regions in the context of such screens.