CRISPR screens are feasible in TP53 wild-type cells.

A recent study by Haapaniemi et al (2018) reported that intact p53 signaling hampers CRISPR-based functional genomic screens. Brown et al report good performance of genome-scale screens in TP53 wild-type cells and reiterate best practices for CRISPR screening.

A recent study in Nature Medicine (Haapaniemi et al, 2018) caused concern amongst researchers and investors alike after reporting that genome editing with CRISPR may be less efficient in human cells with intact p53 signalling.

The conclusion by Haapaniemi et al that the p53‐mediated DNA damage response (DDR) hampers CRISPR‐based functional genomic dropout screens inspired us to investigate this phenomenon more closely. When performing a genome‐scale CRISPR screen in RPE1 cells, harbouring wild‐type TP53, Haapaniemi et al reported both a failure to detect essential genes and the enrichment of single‐guide RNAs (sgRNAs) targeting TP53, CDKN1A (encoding p21) and RB1 (pRb). This enrichment of sgRNAs targeting the p53 axis was not observed in a p53‐deficient RPE1 cell line, where essential genes were depleted as expected. From this, together with a number of follow‐up experiments, the authors conclude that the p53 pathway is activated by CRISPR‐induced double‐strand breaks (DSBs) leading to a cell cycle arrest. Therefore, this presents an obstacle to precision genome editing as well as functional genomic screening in TP53 wild‐type cells. This hypothesis would be consistent with previous research showing that a single DSB is sufficient for inducing a prolonged p53‐dependent arrest in normal human fibroblasts (Di Leonardo et al, 1994; Linke et al, 1996). It is also in line with the complementary findings by Ihry et al (2018) reporting that the p53‐dependent response to Cas9 toxicity reduces the efficiency of human pluripotent stem cell (hPSC) genome engineering and indicating that TP53 status should be monitored in hPSC before and especially after engineering.

We and others have performed genome‐wide screens in TP53 wild‐type and p53‐deficient cell lines, including RPE1 cells (Hart et al, 2015, 2017; Zimmermann et al, 2018). In agreement with Haapaniemi and colleagues, sgRNAs targeting TP53, CDKN1A, RB1 and other p53 pathway genes became enriched over the course of the RPE1 screen, while sgRNAs targeting the negative p53 regulators MDM2 and MDM4 were depleted (Fig 1A). Similar observations were also made in other TP53 wild‐type cell lines, including HCT116 colon cancer cells and a glioblastoma cell line, whereas HeLa cells (non‐functional p53 due to expression of the E6 viral oncogene) and DLD1 cells (oncogenic TP53 S241F/− mutation Ahmed et al, 2013; Sur et al, 2009) do not show this effect. While the assertion by Haapaniemi et al is that DSBs induced by Cas9 editing drive p53 activation in the wild‐type background, this is not directly measured in their (or our) experiments. In fact, similar enrichments have been observed in large‐scale RNAi experiments, which do not rely on the induction of DSBs (Giacomelli et al, 2018).

Figure 1

Performance of CRISPR–Cas9 functional genomic screens with respect to status

(A) Genome‐wide CRISPR–Cas9 screens were performed in five established human cell lines. Heatmap shows mean log2‐fold change of all sgRNAs targeting the indicated p53 pathway genes. (B) Mean log2‐fold change scatterplots for all 17,236 genes targeted by the TKOv1 library for screens described in (A). Bottom triangle indicates Spearman correlation between screen pairs. Red: cell lines with non‐functional p53. Green: cell lines with functional p53. (C) Mean log2‐fold changes of all sgRNAs per gene are shown for the results of a CRISPR screen with wild‐type RPE1 cells. Gene sets comprising gold‐standard essential genes (essentials), ribosome or proteasome subunits, or gold‐standard non‐essential genes (non‐essentials) are highlighted. (D) Ratios of CERES scores for reference gene sets between wild‐type and mutant/non‐functional cell lines across 356 cancer cell lines in the 18Q4 Avana (DepMap) dataset for which functional status was available in Giacomelli et al (***FDR < 0.001, **FDR < 0.01, Wilcoxon rank sum test with multiple testing correction). (E) Precision–recall curves based on the mean log2‐fold change of sgRNAs using gold‐standard essential and non‐essential genes across five RPE1 wild‐type genome‐wide CRISPR screens and two RPE1 ‐deficient (“null”) genome‐wide CRISPR screens. The CRISPR libraries and source of the screens are indicated. Also see main text for details and references.

Haapaniemi et al also report that sgRNAs targeting a set of ribosomal genes, largely expected to be essential for cell growth, failed to deplete in the wild‐type TP53 RPE1 background, whereas they dropped out as expected in their p53‐deficient RPE1 cells. From this, the authors conclude that a transient cell cycle arrest mediated by p53 explains the failure to detect essential genes in the CRISPR screen. In contrast, we find that gene‐level dropout profiles are highly correlated between all pairs of cell lines (P < 2.2 × 10−16), regardless of their p53 status (Fig 1B), arguing against p53 hampering the depletion of essential genes.

We and others have developed methods to evaluate the performance of functional genomics screens (Hart et al, 2014, 2015, 2017), which have been used to quality‐control different sgRNA libraries (Sanson et al, 2018). Applying these to a published RPE1 screen (Hart et al, 2015), we find that sgRNAs targeting essential genes such as the ribosome or proteasome subunits, or “gold‐standard essential genes” (Hart et al, 2017), deplete as expected (Fig 1C) and similar to cell lines with non‐functional TP53. These results demonstrate that screens in immortalized and cancer cell lines with wild‐type TP53, including RPE1, still identify essential genes targeted by the depleted sgRNAs.

To ensure that our results were not unique to a small or biased dataset, or to a specific CRISPR library, we examined the dropout profiles of essential gene sets across a compendium of 517 genome‐wide CRISPR screens in various cancer cell lines (Meyers et al, 2017) (DepMap18Q4 version; https://depmap.org/portal/) or 14 genome‐wide screens in AML cell lines (Wang et al, 2017). We found negligible differences in the fold‐change distributions of sgRNAs targeting these gene sets between TP53 wild‐type and different categories of TP53 mutant cell lines in the DepMap screens (Fig 1D, data not shown for AML cell lines). Precision–recall curves based on the gold‐standard gene sets show very high performance (i.e. high precision and recall and therefore a large area under the curve, AUC) across all DepMap screens, with a mean AUC of 0.99, regardless of p53 status. Critically, the DepMap screens were performed using the Avana gRNA library (Doench et al, 2016), indicating that our findings are not specific to our own CRISPR libraries. Similarly, Ihry et al (2019) reported successful genome‐scale CRISPR screens in a TP53 wild‐type human pluripotent stem cell line, despite observing considerable p53‐mediated Cas9 toxicity in these cells (Ihry et al, 2018), which is in agreement with data from our own laboratory in a comparable system (Mair et al, 2019). In summary, while the induction of a p53 response in TP53 wild‐type cells is not unexpected, it does not hamper the screenability of TP53 wild‐type cells in general.

Finally, we obtained the raw data from Haapaniemi et al, allowing us to generate precision–recall curves using the gold‐standard essential gene set (Hart et al, 2017) to assess screen performance. We incorporated the same analyses of four additional published or unpublished RPE1 genome‐wide loss‐of‐function CRISPR screens performed in three different laboratories, including one screen carried out in a TP53‐deficient background (Zimmermann et al, 2018) (Fig 1E). Our original RPE1 screen (red) and the other TP53 wild‐type and TP53‐deficient RPE1 screens show high performance, indicating that, regardless of TP53 status, recovery of the expected essential genes is feasible in RPE1 cells. In contrast, the screen in TP53‐deficient RPE1 cells from Haapaniemi et al shows moderate performance, while the screen in TP53 wild‐type RPE1 cells failed to identify gold‐standard essential genes with any precision. The authors argue that this is due to p53 signalling; however, that conclusion is inconsistent with the RPE1 TP53 wild‐type screens from other groups. In our experience, such underperformance is more likely due to poor editing efficiency. Haapaniemi et al selected their cells using functional editing of the HPRT1 gene, which does not require high or sustained Cas9 activity, and they did not directly report editing efficiency of their cell line after the initial selection process.

Haapaniemi et al and Ihry et al have reported an important phenomenon in the context of CRISPR–Cas9 genome editing, and we echo support for monitoring TP53 status, especially with respect to cell engineering for clinical applications. However, we suggest key steps that should not be ignored when publishing any CRISPR screen results:

Prior to performing a CRISPR screen, editing efficiency in the cell line system to be screened should be assessed using accepted protocols such as the SURVEYOR assay or editing tests with essential genes. Low editing efficiency can easily yield results for positive selection screens, but is not compatible with measuring sgRNA dropout in pooled screens.

Comprehensive assessment of screen performance should be conducted using established benchmarks and metrics, such as reference standard essential gene sets for pooled dropout screens.

Observations should be validated in independent cell lines or clones, preferably with orthogonal methods or through genetic “rescue” experiments to establish the robustness of the reported findings.

The application of gene editing technologies to genome‐wide functional genomics provides an incredible opportunity to accelerate functional annotation of the human genome with high precision and accuracy. While we do not disagree that CRISPR may evoke a p53 response in certain genotypes, examination of hundreds of published genome‐wide CRISPR screens conducted independently by multiple laboratories in different geographic locations showed that major concerns are not warranted when performing CRISPR screens in wild‐type TP53 cells. Therefore, we would like to reinforce the notion of existing functional genomics standards to quality‐control genome‐scale screening data in order to avoid some of the pitfalls that were discovered during the early years of RNA interference screens.

Author contributions

KRB and BM initiated the study and wrote the manuscript, KRB performed the data analysis, MS provided critical insight and interpretation to the work and assisted with editing the manuscript, and JM supervised the work, provided critical insight and helped edit the manuscript.

Dataset EV1

Click here for additional data file.

Review Process File

Click here for additional data file.