A Genome-wide Framework for Mapping Gene Regulation via Cellular Genetic Screens

May combine two claims.

20 genes were linked with two candidate enhancers, and 6 genes with three or four, located within the same super-enhancer.

“More specifically, 20 genes were linked with two candidate enhancers, and 6 genes were linked with three or four candidate enhancers, that were located within the same super-enhancer (Table S4C).”

May combine two claims.

Of 359 positive control TSSs, 35 reduced expression of 1+ additional genes, yielding 45 apparent promoter-promoter relationships.

“Of our 359 “positive control” TSSs whose targeting successfully repressed the expected gene in both experiments, 35 reduced expression of 1+ additional genes (45 apparent promoter-promoter relationships in total).”

May combine two claims.

In the scaled experiment, 207,324 single cells were profiled and a median of 28 ± 15.3 gRNAs per cell was identified.

“We profiled the transcriptomes of 207,324 single cells and identified a median of 28 ± 15.3 gRNAs per cell (Figure 3C).”

Probably not a substantial claim: may not pass S3 (load-bearing).

24 candidate enhancers are paired with multiple target genes.

“Of note, 24 candidate enhancers are paired with multiple target genes (Figure S2D);”

May combine two claims.

The remaining 30 apparent promoter-promoter relationships are largely not enriched for proximity to affected genes, with a median distance of 405.3 kb.

“However, these 30 are largely not enriched for proximity to affected genes (Figure S5A; median distance of 405.3 kb, similarly restricted to upstream tests), in contrast with enhancer-gene pairs (median distance = 24.1 kb).”

May combine two claims.

The high-confidence pairs involved 441 candidate enhancers and impacted 353 target genes.

“involving 441 candidate enhancers (Figure S2C) and impacting expression of 353 target genes.”

May combine two claims.

At a 10% empirical FDR, 97% (369 of 381) of TSS-targeting positive controls repressed their associated genes, as did all β-globin LCR controls.

“Using the NTCs to set a more inclusive empirical FDR of 10%, 97% (369 of 381) of TSS-targeting positive controls repressed their associated genes, as did all β-globin LCR controls.”

May combine two claims.

Homozygous deletion lines for 3 enhancers all had expected direction and magnitude of effect, with stronger effect sizes than CRISPRi.

“These homozygously deleted lines all had the expected and magnitude of direction of effect (Figure 4E–4G), indeed with stronger effect sizes than seen by CRISPRi perturbation in the scRNA-seq screen (effect size with deletion: e-NMU = −100%, e-CITED2 = −57%, e-GLUL = 67%; Table S3B).”

May combine two claims.

Pairs identified in both experiments had stronger effect sizes, better correlated effect sizes, and involved more highly expressed genes than pairs identified in only one experiment.

“The pairs identified in both experiments had stronger effect sizes (median 25% versus 13% repression), better correlated effect sizes (Spearman’s rho for % repression: 0.82 versus 0.13; Figure S2A), and involved more highly expressed genes (median0.90 versus 0.63 UMIs per cell), than pairs identified in only one experiment.”

May combine two claims.

Paired candidate enhancers were enriched for high ChIP-seq peak strength for H3K27ac, p300, and lineage-specific TFs (GATA1, GATA2, SMAD1, TAL1, CCNT2), whereas RNA Pol II and H3K4me1 were not associated.

“We found paired candidate enhancers to be enriched for high chromatin immunoprecipitation sequencing (ChIP-seq) peak strength (based on average enrichment in ChIP-seq peak region) for enhancer-associated histone modifications (H3K27ac, logistic regression p value = 4e 5, candidate enhancers in the top quintile were 1.4-fold more likely to be paired than those in the bottom quintile), certain co-a”

May combine two claims.

At a 3.5% empirical FDR threshold, 94% (357 of 381) of TSS-targeting positive controls repressed their associated genes, as did all β-globin LCR controls.

“At this threshold, 94% (357 of 381) of TSS-targeting positive controls repressed their associated genes, as did all β-globin LCR controls (examples shown in Figure 2F).”

May combine two claims.

For 377 re-targeted candidate enhancers, 84 were directed at the same candidate enhancers by original and new gRNA pairs, a 2.4-fold enrichment (hypergeometric p=4e-33).

“Of these, 84 were directed at the same candidate enhancers, a highly significant overlap (hypergeometric test of overrepresentation p value 4e–33; 2.4-fold enriched over expectation).”

May combine two claims.

105 of 978 candidate enhancers were identified as paired in both experiments, a 3.3-fold enrichment over expectation (hypergeometric p=7e-45).

“Of these, 105 were identified in both experiments (hypergeometric test of overrepresentation p value 7e–45; 3.3-fold enriched over expectation).”

May combine two claims.

High-confidence candidate enhancers were enriched for strong peaks in 169 K562 ChIP-seq datasets, with 87 significantly enriched.

“we tested if the 441 high-confidence candidate enhancers were enriched for strong peaks in 169 K562 ChIP-seq datasets (ENCODE Project Consortium, 2012). We identified 87 that were significantly enriched (threshold of an adjusted p value <0.005)”

• cites ENCODE Project Consortium, 2012Not checked yet

  Not checked in the free run.

May combine two claims.

Of high-confidence pairs re-tested, 13/16 matched direction and magnitude, while 3/16 failed to validate.

“In summary, of the high confidence pairs that we re-tested by singleton CRISPRi and/or singleton CRISPR-mediated deletion, 13/16 matched with respect to both their direction and magnitude of effect size, whereas 3/16 failed to validate.”

May combine two claims.

Neither the absolute distance to center of DHS peak nor the absolute distance to center of overlapping p300 peak correlated with enhancer-gene pair effect size (Pearson's r 0.02 and 0.07, respectively).

“However, neither the absolute-distance-to-center-of-DHS-peak (Pearson’s r:0.02) nor the absolute-distance-to-center-of-overlapping-p300-peak correlated with effect size (Pearson’s r: 0.07).”

May combine two claims.

In scRNA-seq data, about 70% of gRNAs carried a 1 bp deletion, about 8% a 2 bp deletion, and about 2% a 3 bp deletion.

“In the scRNA-seq data, ~70% of gRNA carried a 1 bp deletion, ~8% carried a 2 bp deletion, and ~2% carried a 3 bp deletion (Figure S7A).”

Probably not a substantial claim: may not pass S3 (load-bearing).

Some genes with roles in inflammation are among the paired target genes, including NMU and IL6.

“although some genes with roles in inflammation are among our paired target genes (e.g., NMU, IL6)”

May combine two claims.

In the low-versus-high MOI comparison, the repression ratio distributions were centered at one, indicating largely equivalent repression, but a left tail contained tests with reduced estimated efficacy in high MOI experiments.

“Despite the distributions of the ratios described above being centered at one, which indicates largely equivalent repression in high and low MOI experiments, there was a left tail, representing a smaller number of tests with reduced estimated efficacy in the high MOI experiments.”

May combine two claims.

For 3 bp deletions, the ratio of reduction to full-length gRNA reduction was 0.908 for high-confidence pairs and 0.644 for TSS controls.

“However, for 3 bp deletions (very rare), the ratio was 0.908 (high confidence pair) or 0.644 (TSS control).”

May combine two claims.

1 bp deletions did not correlate with significant disruption of CRISPRi efficacy in scRNA-seq data; the ratio of 1 bp deletion reduction to full-length gRNA reduction was 1.01 for high-confidence enhancer-gene pairs and 0.958 for TSS controls.

“Fortunately, 1 bp deletions did not correlate with significant disruption of CRISPRi efficacy in the scRNA-seq data. (1 bp deletion % reduction) / (full length gRNA reduction) ratio was 1.01 (high confidence enhancer-gene pair) or 0.958 (TSS control).”

Probably not a substantial claim: may not pass S3 (load-bearing).

Three of the 381 TSS controls fell into the category of highly expressed in few cells.

“Three of our 381 TSS controls fell into this category.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Targeting the TSS of CITED2 did not seem to successfully repress CITED2's expression, potentially due to inaccuracy of one of two technical replicates.

“To note, targeting the TSS of CITED2 did not seem to successfully repress CITED2’s expression, though this is potentially due to inaccuracy of 1 of 2 technical replicates for this sample.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Most indels in the gRNA backbone were small deletions in bases 8 to 14, so the reported rates in Figure S7A were restricted to those 7 bp.

“However, it should be noted that the overwhelming majority of all indels were small deletions observed in bases 8 to 14 or so; thus, rates provided in Figure S7A are limited to these 7 bp.”

Probably not a substantial claim: may not pass S3 (load-bearing).

gRNA-groups with an impacted spacer were rare among the 145 significant enhancer-gene pairs.

“gRNA-groups with an impacted spacer were rare in our 145 significant enhancer-gene pairs.”

May combine two claims.

For 2 bp deletions, the ratio of reduction to full-length gRNA reduction was 0.959 for high-confidence pairs and 0.806 for TSS controls.

“For 2 bp deletions, this ratio was also not extreme (0.959 (high confidence pair) or 0.806 (TSS control)).”

May combine two claims.

In Sanger sequencing of final gRNA plasmid libraries and in the 8-15 bp downstream of the spacer, about 80% of gRNAs had a small insertion or deletion, mostly 1 bp deletions, between the spacer and the R primer.

“we identified that ~80% of gRNAs harbored a small insertion or deletion (vast majority 1 bp deletions, Figure S7A) in between the spacer and the R primer 5-acttgctaTGCTGTTTCCAGC-3 used in the initial amplification of spacer-oligos.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Scaling the expression values by the total number of cells observed per gRNA group did not appreciably change the overall distributions of the ratios.

“We note that this scaling does not appreciably impact the overall distributions in this case.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

UMI counts provide a cleaner distribution than read counts and were used in all calculations.

“we find that UMI counts provide a much cleaner distribution than read counts and have used UMI counts in all calculations.”

May combine two claims.

In the pilot experiment, NTCs showed inflation when tested against genes highly impacted by perturbing gRNAs, due to subtle nonrandom associations of gRNA-groups across cells, potentially from slight bottlenecking at transduction.

“In the pilot experiment, we identified inflation of NTCs when testing them against genes highly impacted by perturbing-gRNA in our library (for example, NTCs associated with targets of our TSS and globin LCR controls). This was due to subtle yet detectable nonrandom associations of gRNA-groups with other gRNA-groups across cells, potentially due to slight bottlenecking at the transduction level”

May combine two claims.

In the at-scale single-cell dataset, 8 TSS controls and 6 candidate enhancers were enriched in specific tSNE clusters, with odds ratio > 5 and adjusted P < 0.01.

“8 TSS controls and 6 candidate enhancers were enriched within specific clusters (odds ratio > 5, adjusted P-value < 0.01).”

May combine two claims.

Of 470 enhancer-gene pairs, only 2 had a significant association with dCas9-BFP-KRAB count and 10 with guide count at adjusted P < 0.01.

“Of the 470 enhancer-gene pairs, only 2 and 10 had any significant (adjusted P-value <0.01; 7 and 27 for adjusted P-value < 0.05) association with dCas9-BFP-KRAB count or guide count respectively”

May combine two claims.

15 of the 45 apparent promoter-promoter relationships involved overlapping promoters, so the CRISPRi effect is likely direct.

“15 of these 45 involved overlapping promoters (TSSs within 1 kb), such that the observed effect of CRISPRi is likely direct.”

May combine two claims.

One recent study used scRNA-seq for CRISPR perturbations of 71 candidate regulatory elements on ~100 genes in seven genomic regions, but its power and scope was limited by low MOI and a gRNA barcoding strategy with ~50% template switching.

“Of note, one recent study used scRNA-seq as a readout for the effects of CRISPR-based perturbations of 71 candidate regulatory elements on ~100 genes in seven genomic regions (Xie et al., 2017). However, its power and scope was limited by a low MOI (Figure 1B) and a gRNA barcoding strategy that suffers from a ~50% rate of template switching (Hill et al., 2018; Xie et al., 2018).”

• cites Xie et al., 2017 sourceNot checked yet

  Not checked in the free run.

• cites Hill et al., 2018 sourceNot checked yet

  Not checked in the free run.

• cites Xie et al., 2018 sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Prior cell-based CRISPR screens introduce one CRISPR perturbation per cell, with gRNAs usually at low MOI.

“Second, each cell is a vehicle for one CRISPR-mediated perturbation, with the specificity-conferring guide-RNAs (gRNAs) usually introduced via lentivirus at a low multiplicity of infection (MOI).”

Probably not a substantial claim: may not pass S1 (asserted).

The remaining 30 apparent promoter-promoter relationships are more likely consequent to trans effects of repressing the primary target.

“We therefore hypothesize that these are more likely consequent to trans effects of repressing the primary target of these TSS-targeting gRNAs.”

May combine two claims.

The screen's resolution is not constrained by linkage disequilibrium and is not limited to sites with common genetic variants.

“unlike eQTL studies, the resolution of our screen is not constrained by linkage disequilibrium, nor is it limited to studying sites in which common genetic variants happen to exist.”

Probably not a substantial claim: may not pass S2 (checkable).

These observations underscore the difficulty of the prediction task.

“These observations underscore the difficulty of the prediction task,”

Probably not a substantial claim: may not pass S3 (load-bearing).

The method was revised to the one defined above after discussion and review of eQTL literature.

“However, upon further discussion and review of the eQTL literature, we revised our method to the one defined above.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

In the pilot library, initial FlashFry quality annotations failed to label a small number of protospacers with perfect repeat off-targets, allowing their inclusion (81 of 2,238 spacers; only 9 gRNA-groups with both spacers affected).

“Our initial FlashFry quality annotations when designing the pilot experiment did not label a small number of protospacers with perfect repeat off-targets, permitting their inclusion in our library (81 of 2,238 spacers ordered in the pilot library; only 9 gRNA-groups with both spacers affected).”

Probably not a substantial claim: may not pass S3 (load-bearing).

HEK293T cells were authenticated by visual inspection.

“HEK293Ts were authenticated by visual inspection.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Motifs seen in the bottom quartile of promoter frequency were excluded from TF pair analysis due to insufficient power.

“Motifs in the bottom quartile of how often seen in a promoter were excluded for lack of power.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The original method for calling the 145 enhancer-gene pairs from the pilot dataset used a universal P-value cutoff corresponding to a 3.5% empirical FDR.

“This original threshold corresponded to a 3.5% empirical FDR rate, as defined above.”

Classified as probably not a substantial claim.

To avoid this problem, outlier genes expressed in <20,000 cells in either the high-MOI 47,650-cell dataset and/or the scaled 207,324-cell dataset, and with log10(total UMIs / cells with a UMI) > 0.2 greater than predicted by a spline fit, were excluded (35 genes total).

“we excluded outlier genes that were expressed in < 20,000 cells in either the high-MOI 47,650-cell dataset and/or the scaled 207,324-cell dataset, and with log10(total UMIs / cells with a UMI) > 0.2 greater than predicted by a spline fit generated via smooth.spline() with spar = 0.85 to limit overfitting (35 genes total).”

Classified as probably not a substantial claim.

ChIP-seq datasets in the bottom quartile of overlap with paired enhancers or promoters were excluded for power, leaving 168 TFe and 166 TFp datasets.

“ChIP-seq datasets that that were in the bottom quartile of how-often-overlapping with a paired enhancer or promoter were excluded for power (leaving 168 TFe and 166 TFp).”

Probably not a substantial claim: may not pass S3 (load-bearing).

The marks used to define candidate enhancers may exclude some classes of distal regulatory elements.

“the marks we used to define candidate enhancers may be excluding some classes of distal regulatory elements.”

May combine two claims.

8% of remaining gRNAs harbor 2 bp deletions that also largely do not affect CRISPRi efficacy.

“8% of the remaining gRNA harbor 2 bp deletions that also largely do not affect CRISPRi efficacy.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Three replication attempts matched direction and magnitude of effect but were not significant, potentially due to lack of power because only two RNA replicates were sequenced per sample.

“The 3 that matched direction and magnitude of effect but were not significant in a test of differential expression potentially were not detectable due to lack of power, as we sequenced only two RNA replicates per sample.”

Classified as probably not a substantial claim. May combine two claims.

To exclude this inflation in the pilot dataset, Fisher's exact test was used to identify nonrandom assortment between an NTC and a perturbing gRNA (adjusted P < 0.01 and odds ratio > 1), and tests of an NTC against a gene within 1 Mb of that gRNA's group were excluded.

“we used Fisher’s exact test to identify when an NTC was nonrandomly assorted with a perturbing-gRNA (adjusted P-value < 0.01 & odds ratio > 1). Then, any test of an NTC against a gene within 1 Mb of that gRNA’s gRNA-group was excluded from further analytical steps.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Libraries were sequenced to ~20% sequencing saturation.

“All libraries were sequenced to ~20% sequencing saturation.”

May combine two claims.

The appropriate effect sizes can be obtained with the method using coefficients output by VGAM, as verified by power simulations.

“We verified data from our power simulations that the appropriate effect sizes can be obtained with this method using the coefficients output by VGAM.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Fifteen gRNAs targeting the HBE1 TSS and HS1-4 of the Globin LCR were chosen as validated from prior studies.

“15 gRNAs targeting the HBE1 TSS, and HS1–4 of the Globin LCR were chosen as validated from Klann et al. (2017) and Xie et al. (2017).”

• cites Klann et al. (2017) sourceNot checked yet

  Not checked in the free run.

• cites Xie et al. (2017) sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

A singleton dataset for chr6:34191315-34191338 (paired with HMGA1 in pilot) did not reproduce between pilot and at-scale screen and was not included in the high-confidence enhancer-gene pair set.

“we additionally generated singleton datasets for chr6:34191315–34191338 (paired with HMGA1 in the pilot screen), but did not include this in analysis as it did not reproduce between the pilot and at-scale screen, and thus was not part of our high confidence enhancer-gene pair set.”

The quote only partly states this claim.

The monoclonal lines used for bulk RNA sequencing were validated to carry complete e-NMU deletions with no wild-type alleles remaining, using two rounds of genotyping PCR.

“Clones that harbored full deletions with no remaining wild-type alleles were submitted to bulkRNA-sequencing (Figures 4E–4G).”

• Also stated in: § METHOD DETAILS · plain

  “Second, to confirm homozygosity, primers internal to the deleted region were used to identify candidates that still harbored wild-type alleles (Figure S4A; Table S3D).”

Probably not a substantial claim: may not pass S3 (load-bearing).

A limitation of enhancer-gene pair screening as implemented here is the resolution of CRISPRi.

“A limitation of enhancer-gene pair screening as implemented here relates to the resolution of CRISPRi.”

Probably not a substantial claim: may not pass S3 (load-bearing).

K562 cells were authenticated by bulk/single-cell RNA-seq and visual inspection.

“K562s were authenticated by bulk/single-cell RNA-seq and visual inspection.”

Classified as probably not a substantial claim.

The 128 DHSs selected again by the logistic regression model were not removed because doing so would bias remaining DHSs away from the same annotations.

“Note that we did not remove the 128 DHSs that had been selected again by the logistic regression model, because doing so would bias our remaining DHSs away from the same annotations.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

PCR was monitored by qPCR to avoid overamplification, and each reaction was stopped immediately before saturation.

“PCR was monitored by qPCR to avoid overamplification, and each reaction was stopped immediately before it reached saturation.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Prior work has found shadow enhancers that mask the effects of perturbing individual enhancers.

“previous studies have identified shadow enhancers acting to mask the effects of perturbing individual enhancers (Hong et al., 2008)”

• cites Hong et al., 2008 sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

The small insertions/deletions were inferred to result from slippage of the KHF polymerase copying the secondary structure of the first stem extension loop added as part of the sgOPTI backbone.

“We inferred that this is due to slippage of the KHF polymerase as it copies the secondary structure of the first stem extension loop added as part of the more stable sgOPTI backbone.”

Probably not a substantial claim: may not pass S3 (load-bearing).

PCRs for e-NMU genotype enrichment were monitored with SybrGreen on a qPCR and stopped before overamplification to prevent excessive PCR bias.

“PCRs were spiked with SybrGreen, and monitored on a qPCR to allow removal before overamplification to prevent excessive PCR biases.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Not all enhancers may be susceptible to dCas9-KRAB perturbation.

“not all enhancers may be susceptible to dCas9-KRAB perturbation”

Classified as probably not a substantial claim. May combine two claims.

To account for this, an additional term log10(total_umis) was added to both full and reduced models and all size factors were set to 1.

“To account for this, we added an additional term to both the full and reduced model “~ log10(total_umis)” and set all size factors to 1.”

Probably not a substantial claim: may not pass S3 (load-bearing).

For distance-to-TSS analysis, only enhancers located upstream of the target gene were included, because gene body length would confound distances for downstream enhancer-gene pairs.

“To note, in Figure 6A and to calculate the median distance, we have only used enhancers that are upstream of the target gene, as the length of the gene body would confound distance-to-TSS measurements for downstream enhancer-gene pairs.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

K562 cells are a pseudotriploid ENCODE Tier I erythroleukemia cell line derived from a 53-year-old female with chronic myelogenous leukemia.

“K562s cells are a pseudotriploid ENCODE Tier I erythroleukemia cell line derived from a female (age 53) with chronic myelogenous leukemia (Zhou et al., 2017).”

• cites Zhou et al., 2017Not checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

Likely chimeric reads were detected and removed to reduce noise in gRNA-cell assignments.

“Likely chimeric reads are detected and removed to reduce noise in the assignments as previously described.”

Classified as probably not a substantial claim.

For the scaled experiment, the number of guide RNAs observed in a cell, the percentage of mitochondrial transcripts, and the prep batch were regressed out in differential expression testing.

“we regressed out the number of guide RNAs observed in a cell (as a proxy for the number of integrants), the percentage of total transcripts observed that are mitochondrial, and the prep batch”

May combine two claims.

Because the logistic regression classifier is biased toward annotations used to select the initially targeted candidate enhancers, submodular subset selection was used to include DHSs optimized for a diversity of epigenomic features.

“Because the logistic regression classifier is biased toward the annotations that were used to select the initially targeted candidate enhancers (Figure 2A), we additionally used submodular subset selection to include DHSs optimized for a diversity of epigenomic features (Wei et al., 2015).”

• cites Wei et al., 2015Not checked yet

  Not checked in the free run.

Classified as probably not a substantial claim. May combine two claims. The quote only partly states this claim.

gRNA-cell assignments used thresholds for minimum total reads and UMI counts for a gRNA-cell pair and for the proportion of CROP-seq transcript reads accounted for by each gRNA to distinguish noise from real assignments.

“We utilized thresholds to set minimum acceptable values for the total reads for a gRNA-cell pair and for the proportion of all CROP-seq transcript reads accounted for by each gRNA observed in a cell to distinguish noise from real assignments (Hill et al., 2018).”

• cites Hill et al., 2018 sourceNot checked yet

  Not checked in the free run.

Probably not a substantial claim: may not pass S3 (load-bearing).

The off-target annotation error was fixed for evaluating gRNA quality in the scaled experiment.

“This error was fixed for evaluating gRNA quality in the scaled experiment.”