Clustered regularly interspaced short palindromic repeats screens in pediatric tumours: A review

Abstract

In recent years, the discovery and development of clustered regularly interspaced short palindromic repeats (CRISPR) technology has revolutionized and accelerated functional genetic screening in cancer research. In this review, we discuss different methods of executing CRISPR screens, with a focus on pediatric tumour entities. Historically, these tumours were thought to resemble their adult counterparts. However, the new era of genomic research and the extensive use of gene editing has identified pediatric tumours as distinct entities with drastically different development and presentation. Here we provide an overview of CRISPR screens performed in pediatric tumour models and highlight unique considerations for pediatric tumour screens. In particular, the results from CRISPR screens combining treatment with genetic knockouts can push treatment for pediatric patients. We conclude by discussing the potential of CRISPR genetic screening to unravel pediatric tumour biology and identify new treatment options.

INTRODUCTION

The prognosis for children and adolescents diagnosed with cancer has rapidly improved over the past decades in western countries, as five-year survival rates have increased from less than 50% before the 1970s to approximately 82% in 2018. However, some specific pediatric tumours like (relapsed) acute myeloid leukaemia (AML), rhabdomyosarcoma, neuroblastoma, and high-grade brain tumours still carry a poor prognosis. For decades, childhood malignancies have been treated with anti-cancer therapies that are effective against adult tumours. However, most pediatric tumours differ from similar tumour types in adults as evidenced by a different and often low mutational burden, a relatively high prevalence of tumour-driving fusion genes, and a distorted epigenetic landscape. As a result of these differences, anti-cancer therapies that are effective against adult tumours often fail in treating childhood malignancies. A direct comparison of driver mutations in pediatric tumours with driver mutations in tumours found in the adult population shows that more than half of the mutations do not overlap. While there are similarities in altered protein pathways between adult and pediatric tumours, the cell of origin is hypothesized to be different and at a different stage of development. This could be a possible explanation of why clinical adaptation of adult treatment regimens is insufficient for pediatric patients. As such, the field of pediatric oncology research rooted itself as an established and separate scientific discipline.

The recent era of molecular research taught us more about the basis of tumorigenesis, which triggered researchers to develop targeted therapies that avoid treatment-related side effects. This shift in mindset went together with the idea that cytotoxic screening approaches could be partly replaced by targeted genetic screenings. RNA interference (RNAi) techniques effectuate a knockdown of gene expression on the RNA level but are unable to generate a genetic knockout. Other precise genetic editing techniques such as zinc-finger nucleases and transcription activator-like effector nucleases TALENs relied on DNA protein interactions for selectivity and required extensive engineering for new targets. Meanwhile, the clustered regularly interspaced short palindromic repeats (CRISPR) technique uses a endonuclease enzyme combined with an RNA molecule to generate insertions or deletions in the DNA. These new techniques sparked the development of new whole genome, kinome, transcriptome, or other ‘ome’-wide RNAi and CRISPR screens, which have contributed greatly to our understanding of tumour biology and treatment of both adult and pediatric cancers.

The possibilities of CRISPR genetic screening are nearly endless, not only for understanding intricate tumour biology mechanisms but also for discovering new therapeutic possibilities. The cell model used for these types of investigations is of high importance. During the 20th century, in vitro cancer cell lines were either artificially immortalized or the offspring of a highly mutated subpopulation of cancer cells. Only in recent years, scientists have been able to develop cancer cell cultures that maintain the same cellular state and heterogeneity as the original tumour, as evidenced by (single cell) RNA sequencing and methylation profiling. The use of such so-called primary cancer cell cultures, maintained in custom enriched serum-free medium, generates more clinically relevant results in CRISPR genetic screening. However, these models are more difficult to genetically engineer thus limiting CRISPR screening efficacy in these cells. As such, most publications investigating pediatric tumours have utilized conventional serum-cultured cell lines, which are known to reflect the more aggressive cancers. A final caveat of this model system is the fact that CRISPR screens are performed in tumour cells without the appropriate micro-environment that also includes interactions with other cell types, such as immune cells, mesenchymal stem cells or extra-cellular matrix molecules. Such co-culture systems have been extensively studied but the use of these systems for CRISPR genetic screens is hindered by scale problems.

In this review, we show multiple ways of performing CRISPR screens in pediatric tumour models to produce data with a specific research question in mind, aiming to explain the benefits and drawbacks of each method. Next, we discuss the new treatment possibilities for pediatric cancer patients found through these CRISPR screens. And finally, we discuss the benefits and drawbacks of the Dependency Map project of the Broad institute and go into detail on the uses of such a large systematic setup. As such, this review provides an introduction to CRISPR screens and how such screens have been used in pediatric cancer research.

MAIN BODY

Application of CRISPR-Cas system in the context of genetic screening

The CRISPR-Cas system consists of two parts: CRISPR and Cas. CRISPR are short RNA sequences that, in the context of genetic engineering, are used to guide the Cas enzyme to a sequence of interest in the genome. These synthetic RNA molecules are ∼110 nucleotides in length and consist of a trans-activating RNA fused to a CRISPR RNA (crRNA), with the latter being variable depending on the targeted region of the genome. The Cas protein is an endonuclease and in the context of genetic screening, the most widely used variant of this protein is Streptococcus pyogenes Cas9. Cas9 scans the DNA for the protospacer adjacent motif (PAM) and when it has found this motif, the protein will check for complementarity to crRNA sequence of 20 nt before making a double-strand break (DSB) 3–4 nucleotides upstream of the PAM sequence. This DSB will be repaired by the DNA repair machinery of the cell, either via homology-directed repair or, more often, by non-homologous end-joining . This latter repair mechanism may generate deletions or insertions (indels) after which the Cas enzyme will continue binding and cutting until the sgRNA target is altered to a point where there is no complementarity of the sgRNA to the DNA anymore, or the PAM motif is lost. The effect of this genetic editing depends on the location of the sequence the sgRNA is complementary to: the indels need to result in a frameshift or a non-functional protein in order to generate a genetic knockout.

In addition to CRISPR knockout screens, other types of CRISPR screens, such as activation or repression genetic screens have been developed. In these types of screens, the selectivity of the CRISPR-Cas system is used to direct a nuclease-dead Cas enzyme to specific regions in the DNA. For inhibitory CRISPR (CRISPRi) screens, the Cas enzyme is combined with an inhibitory transcription factor such as KRAB to sterically hinder gene transcription, effectively generating a gene knockdown. While not as widely used as RNA-mediated gene inhibition, the recent improvements in CRISPRi screening make this a useful tool for the genetic knockdown. Combining the same dead-Cas9 nuclease with transcription activating domains such as VP64 or variations on this idea such as dCas9-VPR or dCas9-SunTag have been developed to induce gene expression. Another variant of genetic screening is base editor mediated screening, which employs a cytosine base editor to alter start codons, splice sites or insert a stop codon by altering a single base. This has been used successfully to interrogate clinically relevant single nucleotide variants and tumorigenic mutations. Besides, base editing can be used to generate isogenic (patient-derived) cell lines in which a single mutation is edited, such as the restoration of the H3K27M mutation in diffuse midline gliomas. This, combined with a genome-wide sgRNA library, could provide useful mechanistic insights into how the mutation at hand alters genetic dependencies, in particular for pediatric tumours with a low mutational burden. For a more in-depth review of the possibilities of the CRISPR-Cas system in cancer modelling, we refer to the paper by Yin et al.

In CRISPR-Cas genetic screens, a large collection of sgRNAs and a Cas protein are introduced into a (cancer) cell line protein to identify the role of genes in response to selection pressure. Since most publications discussed in this review use CRISPR genetic knock-out viability screens, we here elaborate on the many experimental considerations underlying the design of a CRISPR screen (Figure ).

[IMAGE OMITTED. SEE PDF]

The first step in setting up a CRISPR genetic screen is to decide which cell line model to use. Next, the decision has to be made whether to use a pooled or an arrayed format. In an arrayed setting, one gene per well is perturbed using one or multiple sgRNAs, whereas in a pooled CRISPR screen all cells are cultured in bulk. For the latter, the system is dependent on the assessment of relative sgRNA abundance in samples from different timepoints or conditions as determined by next-generation sequencing. In an arrayed set-up, however, the readout can be assessed per well, which negates the need for sequencing to determine which sgRNA induced the phenotype of interest. This set up also allows for more intricate readouts such as adherence or differentiation assessments but limits the number of genetic perturbations that can be studied simultaneously as often arrayed high throughput (image) analysis is difficult due to batch effects and noise limitations.

Next is the use of an appropriate sgRNA library for the research question at hand. These libraries are often (commercially) available and can contain several thousands of sgRNAs, with multiple sgRNAs targeting the same gene to ensure the loss of a functional protein in addition to providing a level of redundancy. More specific sub-libraries enriched for a group of genes of interest can streamline the validation process compared to genome-wide knockout libraries of ∼130.000 sgRNAs. While one can easily develop sgRNAs individually, these commercially available sgRNA libraries often use extensively validated sgRNA sequences, which were confirmed to induce loss of protein expression after gene editing.

A final decision is the duration of your screen, which also affects the control group required for your research question. A simple essentiality screen, during which the effect of loss of gene expression over time is assessed, uses an early time point as a control. An investigation into the genes involved in treatment response, on the other hand (e.g., a synthetic lethality screen), requires an untreated control group besides an early timepoint for sgRNA library representation to exclude those genes that are essential for cell survival regardless of treatment (Figure ). Indeed, the addition of therapy requires multiple decisions on duration, dosing and concentration of treatment. In particular, the duration of treatment determines the outcome: treatment for the entire duration of the screen emphasizes the role of genes involved in treatment resistance, whereas a short treatment at the start of the screen indicates possible mechanisms to improve the initial treatment response.

[IMAGE OMITTED. SEE PDF]

While further developing an experimental screening setup, it is vitally important to choose the ideal Cas and sgRNA system. As the Cas enzyme needs to make multiple cuts to generate indels that impair gene expression, the type and timing of the enzyme are of importance. With the use of a constitutively active Cas9, one ensures that there is both time and abundance of Cas enzyme to ensure a genetic knockout is made. However, the effect of a given knockout may be under- or overestimated as the timing of this system is inflexible. Alternatively, an inducible Cas system allows researchers to answer a variety of biological and therapeutic questions by inducing Cas9 expression at different timepoints throughout the experiment. Next, the expression of sgRNAs can be secured using either a method of transfection, so the sgRNAs are only temporarily present in the cell, or transduction, which integrates the sgRNA genetic code into the genome, ensuring stable expression. The former complicates downstream analysis as genomic DNA sequencing needs to be performed to identify which genes are lost or enriched but can be used in cell lines particularly sensitive to genet editing. Comparatively, sgRNA transduction allows for more time for Cas9 to make the required edits that confer a functional knockout of the protein of interest and simplifies downstream analysis, as integrated sgRNA-encoding sequences can easily be identified and quantified by NGS. An alternative to both these methods is the use of ribonucleoproteins (RNPs), where the Cas9 enzyme and sgRNA molecule are electroporated into the cell of interest as a readily active complex. This is a particularly good alternative for small sets of genes in cell cultures that are hard to transduce, such as many patient-derived cultures, but comes with the drawback of transfer inconsistencies and difficulties of scale.

Pooled CRISPR screens require DNA isolation at the end of the experiment to identify the effect of genetic perturbations, followed by polymerase chain reaction (PCR) to select genomic regions where the sgRNA is incorporated. In the case of sgRNA transfection or RNP-mediated genetic editing, genomic sequencing needs to be performed. As most screens in this review used transduction methods, we will go into further detail on analysis methods for these types of screens. After PCR, next-generation sequencing provides read counts for each sgRNA and many algorithms have been developed to study differences in the abundance of sgRNAs between groups. First, the reads of each guide are counted in both the experimental and control group, which is done using standard count software such as DESeq2⁴⁵ or edgeR whereas other algorithms work from raw read counts. From these counts, fold changes are calculated to represent changes in relative abundance which is used as the basis for most algorithms. In the most simple algorithm, the next step is ranking these fold changes, on a per guide level or averaged per gene. Other algorithms such as MAGeCK or CERES take additional steps to improve hit selection, such as using the read counts of negative and positive controls as baselines to determine cut-off values for p-value and other (statistical) measures and taking copy number variations or variable sgRNA efficiencies into account. For an extensive review of different analysis methods of (pooled) CRISPR screens, we refer to Bock et al.

To determine whether cells that harbour a certain sgRNA are lost or enriched, compared to the rest of the pool, CRISPR screens analyses should take so-called “essential genes” into consideration, as the knockout of these genes is considered consistently lethal. Many publications have developed sets of genes that can be used for this purpose. The reads from sgRNAs targeting these genes can be used as a scale next to reads of non-targeting sgRNAs to define the range of reads of interest, often integrated into analysis algorithms such as MAGeCK.

An important aspect of CRISPR screening is maintaining sufficient statistical power to ensure that the phenotype observedby loss of gene expression can be traced back to the causative gene, taking into account the desired effect size. A first step is the design of sgRNAs, ensuring that they target a functional part of the protein, followed by the number of sgRNAs per gene and finally the robustness of the screen. This robustness is affected by the multiplicity of infection (MOI), the number of cells for each sgRNA and the size of the effect of genetic knockout on phenotype. The MOI is a Poisson distribution of chance that a viral particle infects one or more cells, so an MOI below 0.3 has a very small percentage of cells infected with two or more viral particles. The robustness of the screen is also affected by the number of cells with the same sgRNA; maintaining the coverage, for example, the total number of cells representing each sgRNA at X times, throughout DNA isolation, PCR and next-generation sequencing is required. Finally, the response of the cell model to the loss of a particular gene needs to be large enough to induce a difference in sgRNA reads compared to the control.

After the design and optimization of a CRISPR screen, the effect of genetic perturbations on the phenotype requires extensive validation. These validations should consist of a repeat knockout experiment with other sgRNAs targeting the same gene to confirm the phenotype observed in the CRISPR screen. Next, the effect of the sgRNA guided cut by the Cas enzyme needs to be determined: whether the sequence of the gene is altered or any other form of regulation occurs. While disregarded in many publications, a final step in validation would be the restoration of the wild-type sequence that would rescue the phenotype. However, most researchers choose to continue with pharmacological methods such as small molecule inhibitors that replicate the phenotype, to provide a clinical translation.

CRISPR screens in pediatric tumour models

For this review we selected CRISPR screens performed in pediatric tumour cell lines, using an algorithm of “CRISPR” and “screen” mentioned in either title or abstract and selected those publications using human pediatric tumour models in PubMed. Most publications are based on experiments performed in serum-cultured cell lines, but some publications used primary patient-derived cell lines for CRISPR screens. Some of the CRISPR screens included in this review have used published data from the DepMap consortium or were a part of this consortium.

The majority of CRISPR screens were performed in leukaemia or neuroblastoma tumour cells, which does not reflect the incidence distribution of pediatric cancers, as CNS tumours are nearly as common as leukaemias in children. This lack of representation is mostly due to the availability of material – which is restricted in CNS tumours – and research evolution as leukaemia research was historically a driver of pediatric cancer research. In order to provide an integral overview of CRISPR genetic screens performed in pediatric tumours, we here discuss publications based on tumour type, going into detail on the method of screening and highlighting some of the results. For each tumour type, a table is included to provide an integral overview of all CRISPR screens performed in this pediatric tumour type.

CRISPR screens in leukaemia

We found twenty-one publications that performed CRISPR screens in leukaemia cell lines, with most of them investigating AML (see Table ). While this type of pediatric leukaemia is not the most common, it has a worse prognosis than acute lymphoblastic leukaemia (ALL) due to more common relapses and therefore new treatment strategies are needed. These screens all used established leukaemia cell lines, with many using MOLM-13, an AML cell line with an internal tandem duplication mutation in the FLT3 gene causing constitutive activation of this kinase (FLT3^ITD). This mutation is present in a third of AML patients and can be targeted using small molecule inhibitors, albeit with limited clinical use due to the development of treatment resistance.

TABLE 1 Overview of clustered regularly interspaced short palindromic repeats (CRISPR) knockout screens performed in pediatric leukaemia models. Here, targets are validated hits. The treatment represents the selection condition used compared to no treatment control. The analysis describes the algorithm used for mapping and hit calling of the CRISPR screen. SL KO screen = synthetic lethality knockout screen; ES screen = essentiality knockout screen; LoA = loss of adhesion. *No raw data publicly available

Leukaemia	Type	Treatment	Library (#genes)	Analysis	Target(s)	Author
ALK+ ALCL	Constitutively active Cas9 SL KO screen	ALK inhibitor Crizotinib	GeCKOv2 (19,050)	MAGeCK	PTPN1, PTPN2	Atabay
FLT3ITD AML	Constitutively active Cas9 SL KO screen	HSP90 inhibitor LAM-003	GeCKOv2 (19,050)	Custom	KDM6A	Beeharry*
pre-B ALL	Inducible Cas9 SL KO screen	Asparaginase	Sabatini Kinases (5070)	MAGeCK	BTK	Butler
AML	Constitutively active Cas9 ES screen	–	Custom: autophagy-related genes (198)	Unknown	ATG3	Baker
t(9;11) MLL rearranged AML	Constitutively active Cas9 SL KO screen	LSD1 inhibitor OG-86	GeCKOv2 (19,050)	MAGeCK	mTORC1	Deb
B cell lymphoma	Constitutively active Cas9 LoA KO screen	–	Brunello kinome (6,104)(p9)	DeSEQ2 + MAGecK		De Rooij
MLL-AF4 AML	Constitutively active Cas9 ES screen	–	GeCKOv2 (19,050)	RIGER	ENL	Erb
FLT3ITD AML	Constitutively active Cas9 SL KO screen	FLT3ITD inhibitor Quizartinib	Sanger v1 (18,010)	MAGeCK	GLS	Gallipolli
MV4;11 AML	Constitutively active Cas9 ES screen	-	Achilles Avana (18,343)	CERES (DepMap 17Q4)	ATM	Guerra
Asparaginase resistant T-ALL	Constitutively active Cas9 SL KO screen	Asparaginase	GeCKOv2 (19,050)	MAGeCK	NDK2, LGR6	Hinze
FLT3ITD AML	Constitutively active Cas9 SL KO screen	FLT3ITD inhibitor Quizartinib	GeCKOv2 (19,050)	Custom	SPRY3, GSK3	Hou
ALL	Constitutively active Cas9 SL KO screen	Panobinostat	GeCKOv2 (19,050)	MAGeCK	SIRT1	Jiang
AML	Constitutively active Cas9 ES screen	–	Custom: sphingolipid metabolism genes	–	KDSR	Liu
FLT3ITD AML	Constitutively active Cas9 SL KO screen	Venetoclax	Brunello lentiCRISPRv2 (19,114)	MAGeCK	TP53, BAX, PMAIP1, TFDP1	Nechiporuk
T-ALL	Constitutively active Cas9 ES screen	–	Achilles Avana (18,343)(p4)	CERES (DepMAP 19Q4)	SHMT1	Pikman
Venetoclax resistant AML	Constitutively active Cas9 SL KO screen	Venetoclax	Genome-wide Toronto KO v1 -90K (17,661)	MAGeCK	RBFA, MRPL17, MRPL54, DAP3	Sharon *
AML	Constitutively active Cas9 ES screen	–	Sanger v1 (18,010)	MAGeCK	NMNAT1	Shi
FLT3ITD AML	Constitutively active Cas9 ES screen	–	Custom: kinase (482)	unknown	SIK3	Tarumoto
AML	Constitutively active Cas9 ES screen	–	Sanger v1 (18,010)	MAGeCK	KAT2A	Tzelepis
FLT3ITD AML	Constitutively active Cas9 ES screen	–	GeCKOv2 (19,050)	DeSEQ2	miR-150, miR-155, miR-182	Wallace
MLL-AF9, FLT3ITD AML	Constitutively active Cas9 ES screen	–	Custom: ribosomal proteins (490)	unknown	RBM39	Wang

A synthetic lethality CRISPR knockout screen compares the effect of a certain gene knockout in the presence of a therapeutic compound or vehicle to identify genes involved in treatment response. Gallipolli et al. set up such a screen to identify genes involved in resistance to FLT3 inhibitors using the FLT3 inhibitor sensitive cell line MOLM-13. The authors used a genome-wide CRISPR knockout library and compared the effect of a short treatment window of 72 h at an IC50 concentration followed by 14 days of cell growth without treatment. In this setup it is expected that half of the clones sensitive to FLT3 inhibition would be removed from the pool, thus providing information on the remaining resistant clones; information that can be used to improve the initial response to FLT3 inhibition. Another approach to synthetic lethality CRISPR screening uses a different timeframe: Hou et al. treated another FLT3 mutant cell line, MV4-11, for 14 days at a concentration that reduced cell viability to 50% at 14 days. The remaining clones thus have lost genes that induce treatment resistance, which can provide treatment options for FLT3 inhibition-resistant patients. In a comparison of the top 100 most depleted genes In these screens, the hit investigated by Gallipolli et al. (the gene glutaminase involved in the metabolism of the cell) was confirmed in the screen by Hou et al., suggesting that this gene is quickly and steadily lethal for leukemic cells. Despite this common hit, most (98/100) other genes were not common between the two screens, suggesting that the two methods are biologically different approaches.

Besides viability, a plethora of readouts exists for CRISPR screens, depending on the research question at hand. For example, de Rooij et al. investigated integrin-mediated adhesion of B-cell lymphomas by using a cell model that grows in suspension but will adhere to fibronectin upon treatment. In their setup, a Namalwa cell line that constitutively expressed Cas9 was transduced with a kinome sgRNA library and using two different treatments adherence was induced. As cell adherence occurs quickly, adherent cells were separated from non- or weakly adherent cells after 30 min of incubation on fibronectin-coated plates. Following data analysis, the authors identified known controllers of B cell receptor-mediated integrin adhesion and suggested new proteins involved in this process. Of note is the limited effect observed on adhesion: deconvolution of sgRNA read counts to gene level was required for the identification of gene enrichment, suggesting that the difference between adherent and non-adherent cell populations was relatively small. For future CRISPR screens, a longer timeframe of adherence may be used or increasing the number of sgRNAs per gene in order to increase the screen robustness.

CRISPR screens in neuroblastoma

In addition to leukaemia, several publications describe the use of CRISPR genetic screens in neuroblastoma (see Table ). Neuroblastoma is a pediatric embryonal cancer that originates from precursor cells of the sympathetic nervous system. Neuroblastoma is the most common solid tumour in children and carries a good prognosis for early-stage disease. However, most patients present with advanced-stage disease, which has a mere survival rate of 40% due to treatment resistance and lack of successful targeted agents.

TABLE 2 Overview of clustered regularly interspaced short palindromic repeats (CRISPR) knockout or activation screens performed in pediatric neuroblastoma models. Here, targets are validated hits. Treatment represents the drug added to the experimental arm of the CRISPR screen. Analysis describes the algorithm used for mapping and hit calling of CRISPR screen. SL KO screen = synthetic lethality knockout screen; ES screen = essentiality knockout screen. Unless stated, screens are pooled CRISPR screens

Neuroblastoma	Type	Treatment	Library (#genes)	Analysis	Target(s)	Author
Neuroblastoma	Constitutively active Cas9 SL KO screen	PIM kinase inhibitor AZD1208	GeCKOv2 A (19,050)	DEseq2	NF1	Brunen
MYCN amplified neuroblastoma	Constitutively active Cas9 ES screen	–	Achilles Avana (18,343)	CERES (DepMap 17Q4)	EZH2	Chen
MYCN amplified neuroblastoma	Constitutively active Cas9 ES screen	–	Achilles Avana (18,343)	CERES (DepMap 17Q4)	HAND2	Durbin
Neuroblastoma	Constitutively active Cas9 rescue screen	BET inhibitors JQ1 and I-BET151	Avana v4 (18,675)	DEseq2	PI3K	Iniguez
Neuroblastoma	Constitutively active Cas9 differentiation screen	–	Custom (288)	n.a. - arrayed screen	PHF20	Long
Neuroblastoma	Constitutively active Cas9 reporter screen	-	GeCKOv2 (19,050)	DEseq2	N4BP1	Spel
Neuroblastoma	CRISPR SAM screen	ALK inhibitors brigatinib and ceritinib	SAMv1 (23,430)		PIM1	Trigg

In previous research, the proviral insertion in murine (PIM) kinase family was identified as a possible new treatment for neuroblastoma: in multiple in vitro and in vivo models targeting one of the PIM, kinase proteins would reduce tumour growth. In addition, PIM kinase protein expression is predictive of clinical outcome, but despite these observations treatment of patients with PIM kinase inhibitors had only some clinical success. This prompted Brunen et al. to unravel the biology of PIM kinases in neuroblastoma, with a genome-wide CRISPR knockout screen in a neuroblastoma cell line sensitive to PIM kinase inhibition. To investigate resistance mechanisms against PIM kinase inhibition, the authors used a long timeframe (30 days) and a high drug concentration of PIM kinase inhibitor AZD1208, postulating that this would induce treatment resistance. Analysis of their results suggested regulation of the mTOR pathway as one mechanism of resistance to PIM kinase inhibition, which they confirmed in multiple cell lines: loss of the upstream mTOR regulator NF1 reduced sensitivity to PIM kinase inhibition. The authors continued to validate these results in a subcutaneous xenograft model, where NF1 wildtype tumours responded to PIM kinase inhibition, but the loss of NF1 expression negated this response. Based on these results, PIM kinase inhibition may be an appropriate treatment for neuroblastoma patients with NF1-expressing tumours.

A different approach to study treatment resistance is to use a CRISPR activation (CRISPRa) screen, as done by Trigg et al. who used this technique to identify genes and pathways that confer resistance to ALK inhibitors in neuroblastoma. Such small molecule inhibitors are currently assessed in clinical trials for ALK-mutated neuroblastoma, but outcomes may be improved by counteracting the observed treatment resistance by targeting rescue mechanisms with a second drug or other treatment modality. Therefore, the authors used two different ALK inhibitors at two concentrations using two neuroblastoma cell lines bearing individual ALK mutations and cultured the cells for 14 days. Resistance genes were validated if the gene was significantly upregulated in all conditions: three genes met these criteria: PIM1, MET and SAGE1. As PIM1 was recently shown to be an interesting target in neuroblastoma, the authors continued to validate this finding and confirmed the synergy between PIM1 and ALK inhibitors in in vivo experiments. Interestingly, ALK is also one of the top genes depleted in the screen performed in an ALK mutated cell line by Brunen et al., confirming the synergy between ALK and PIM inhibitors in an independent experiment.

Besides the identification of putative targets for the treatment of cancer patients, CRISPR screens can also be used to unravel the underlying tumour biology, for example by using reporter cell lines. While antibody therapy for neuroblastoma is successful, other immunotherapeutic approaches often rely on immunogenicity signals such as MHC class I protein expression, which is low in neuroblastoma. To study this, Spel et al. developed a reporter cell line for NFκB activity, which is known to regulate MHC-I expression. With a genome-wide sgRNA library, the authors used a fluorescence-activated cell sort based on MHC-I antibody staining and GFP signal indicating NFκB activity to isolate two groups: NFκB^negMHC-I^neg and NFκB^posMHC-I^pos. By identifying sgRNAs enriched in the MHC-I^pos group, genes involved in the downregulation of MHC-1 expression were discovered: nine genes involved in the regulation of NFκB mediated downregulation of MHC-I expression were identified. Follow-up experiments focused on suppressors of NFκB and showed that Nedd 4 binding protein was involved in the regulation of both canonical and non-canonical NFκB pathways. This NFκB regulator and others are negatively correlated with survival in stage four neuroblastoma patients, although the authors did not investigate this correlation further. Loss of Nedd4 gene expression does not affect cell viability, as evidenced by an insignificant change of reads in the screen performed by Brunen et al. However, targeting this gene could enhance immunotherapy for neuroblastoma patients.

CRISPR screens in other types of pediatric cancer

Besides leukaemia and neuroblastoma, pediatric patients can present with other types of tumours, such as soft tissue tumours or brain tumours. There are many different (sub) types of these classes of cancer, but brain tumours and soft tissue tumours constitute the majority of them. One of the more commonly known types of soft tissue tumours is rhabdomyosarcoma, which affects the skeletal muscles and has a relatively good prognosis when patients present with localized disease. Although very rare, another type of pediatric tumour is rhabdoid tumours, often found in the kidneys or as an atypical teratoid rhabdoid tumour (AT/RT) in the brain, both with limited treatment options. In contrast to AT/RT, the most common malignant brain tumour in pediatric patients is medulloblastoma and can be, in most cases, fully excised. However, adjuvant disease management is required, which often leads to major side-effects. Additionally, refractory patients have limited treatment options. Most of the CRISPR screens in these tumour types focus therefore on finding new treatment options (see Table ).

TABLE 3 Overview of clustered regularly interspaced short palindromic repeats (CRISPR) knockout screens performed in other pediatric tumour models. Here, targets are validated hits. Treatment represents the drug added to the experimental arm of the CRISPR screen. Analysis describes the algorithm used for mapping and hit calling of CRISPR screen. SL KO screen = synthetic lethality knockout screen; ES screen = essentiality knockout screen. Unless stated, screens are pooled CRISPR screens. *No raw data publicly available

Tumour type	Type	Treatment	Library (#genes)	Analysis	Target(s)	Author
MYC amplified medulloblastoma	Constitutively active Cas9 ES screen	–	Avana (18,454)	CERES (DepMap)	CCND2, CCND3, BCL2L1, NEUROG1	Bandopadhayay
Synovial sarcoma	Constitutively active Cas9 ES screen	–	Custom: Chromatin regulators (1387)	Unknown	BRD9	Brien
Medulloblastoma	Constitutively active Cas9 SL KO screen + CRISPRi screen	CDK4/6 inhibitor abemaciclib	GeCKOv2 (19,050)	MAGeCK	RPL10	Daggubatti
Osteosarcoma	Constitutively active Cas9 SL KO screen	Doxorubicin	Genome-wide unspecified (20,914)	DESeq2	RAD18	Du
Renal medullary carcinoma	Constitutively active Cas9 ES screen	–	DCT v1.0 (435)	RIGER	PSMB2 PSMB5 PSMD1 PSMD2 and more	Hong
Sarcoma	Constitutively active Cas9 ES screen	–	DCT v1.0 (435)	RIGER	CDK4, XPO1	Hong
Diffuse midline glioma	Inducible Cas9 SL KO screen	AURKA inhibitor phthalazinone pyrazole	Sabatini kinome library (550)	MAGeCK	PLK-1	Metselaar
Rhabdoid tumours	Constitutively active Cas9 ES screen	–	GeCKOv2 (19,050)	CERES (DepMap 19Q1)	SHP2	Oberlick
Rhabdomyosarcoma	Constitutively active Cas9 ES + differentiation screen	–	Custom: HDAC (10)	Unknown	HDAC3	Phelps*
Rhabdoid tumour	Constitutively active Cas9 arrayed cell proliferation screen	–	Custom: kinase (800)	-	PLK4	Sredni
Ewing sarcoma	Constitutively active Cas9 ES screen		GeCKOv2 (19,050)	ATARiS (Achilles 3.3.8)	MDM2, MDM4	Stolte
Ewing sarcoma	Constitutively active Cas9 SL KO screen	LSD1 inhibitor SP-2509	Avana (18,333)	PoolQ	MRPL45, CYC1, UQCRFS1	Tokarsky
Medulloblastoma	Constitutively active Cas9 ES screen	–	Custom: druggable library (9,260)	Custom	CDK7	Veo
Rhabdoid tumours	Constitutively active Cas9 ES screen	–	GeCKOv2 (19,050)	CERES (DepMap 19Q1)	BRD9	Wang
Osteosarcoma	Constitutively active Cas9 reporter screen	–	GeCKOv2 (19,050)	MAGeCK	KLF11	Wang(p11)
Rhabdomyosarcoma	Constitutively active Cas9 ES screen	–	Custom: iExCN targets (110)	PERL	RIPK2	Xu

In order to identify new treatment options for pediatric rhabdoid tumours, Sredni et al. developed an arrayed CRISPR screen using a rhabdoid tumour cell line, MOM, knocking out 160 kinase genes with multiple sgRNAs targeting the same kinase per well. The authors used a colony formation assay to determine the effect of kinase knockout on clonogenicity: after selection for Cas9 and sgRNA-containing cells, an equal number of cells were plated and the time to confluency was used as a determinant for hit selection. Based on this colony formation assay and follow-up experiments, the authors identified Polo kinase 4 (PLK4) as a potential new therapeutic target for pediatric rhabdoid tumours.

Primary pediatric patient-derived tumour models from uncommon tumours are very scarce, not only due to prevalence but also due to the vicious cycle generated by this scarcity, to establish optimal protocols for tissue retrieval, cell line development and maintenance. Hong et al. were able to generate a primary cell line from a patient with a rare, undifferentiated sarcoma. The authors used this cell line to perform not only CRISPR but also RNAi screening as well as a large-scale drug screen, using a commercially available small molecule library. Using this trifecta of high throughput screening methods, they showed that the patient-derived cell line was sensitive to cyclin-dependent kinase 4 (CDK4)- and exportin 1 (XPO1) inhibition, both in vitro and in a subcutaneous xenograft mouse model, showing that generation of primary cell lines derived from rare pediatric tumours can provide clinicians with more than anecdotal evidence for disease treatment.

A patient-derived tumour model derived from AT/RT in custom enriched culturing conditions was used by Metselaar et al. where they used three different cell lines to investigate synthetic lethality with an Aurora kinase A (AURKA) inhibitor, phtalazinone pyrazole. The authors aimed to identify true synergistic hits from genetic knockouts to which the tumour cell could easily adapt, by collecting DNA at multiple time points during the screen. The synergistic hit followed up by the authors (polo-like kinase 1 [PLK1]) was present in all timepoints, but genes related to AURKA inhibition, for example, AURKB and BUB1B, were a hit after 14 days of treatment but not after 21 days, suggesting resistance or at least circumvention of this pathway.

Other publications combined different types of high throughput screening with CRISPR screening. In order to find targets with treatment options available, Veo et al. used a limited CRISPR library targeting only druggable genes (n = 1095). With this approach, the authors successfully identified cyclin-dependent kinase 7 (CDK7) as a new target in MYC-amplified medulloblastoma, in addition to confirming previously identified drug targets such as EZH2 and WEE1. The authors continued to investigate the mechanism of interaction between CDK7 and MYC: as CDK7 is a regulator of transcription initiation, inhibition of this protein alters RNA polymerase II transcriptional elongation and changes MYC-DNA associations. Indeed, a core set of DNA repair genes are deregulated upon CDK7 inhibition, which could be exploited by combining CDK7 inhibition with irradiation for MYC amplified medulloblastoma.

Similarly, Bandopadhayay et al. combined gene expression data and an open-reading frame (ORF) rescue screen with a CRISPR genetic knockout screen. First the genes that were essential for their cell lines were identified based on the CRISPR genetic knockout screen. Next, RNA sequencing of BETi-treated cell lines was used to identify which genes were regulated by treatment. Finally, the ORF rescue screen allowed making a selection of those genes whose overexpression would counteract the effect of treatment, that is, those genes that conferred treatment resistance. By performing these types of screening in combination, the authors robustly identified genes involved in the response to BET inhibitors for the treatment of medulloblastoma and determined that cyclin D2 is involved in resistance to BET inhibition. Whereas an ORF rescue screen is based on overexpression, Daggubatti et al. combined a genome-wide CRISPR knockout screen with a nuclease dead Cas9-KRAB screen that uses the selectivity of sgRNA and Cas9 protein to sterically inhibit gene transcription, thus providing genetic knock-down instead of knock-out data. By performing these screens in combination with the CDK4/6 inhibitor abemaciclib, the authors identified RPL10, a ribosomal protein, as a driver of CDK4/6 inhibitor resistance in hedgehog-associated medulloblastoma. This shows that combining different types of screens can identify new treatment possibilities for pediatric tumours: in particular when phenotype rescue screens and knock-out screens are combined, valuable information on resistance mechanisms arising in patients can be retrieved.

CRISPR screens on large scale: the DepMAP project

In an attempt to streamline the process, minimize cost and increase robustness of results, the Broad Institute has developed a standardized method to perform CRISPR knockout screens on a large scale, using genome-wide sgRNA libraries in 1840 cell lines as of June 2022. The results of these CRISPR screens provide valuable information on gene essentiality in different tumour types. In addition, information on gene expression, copy number variation and genetic mutations have been determined and all cell lines used have been well annotated. This data is publicly available and the consortium has developed a user-friendly bioinformatic analysis platform to visualize genetic dependencies in cancer cell lines (). Until now the DepMap consortium has performed CRISPR screens in 82 pediatric cell lines.

The pipeline described above has several benefits and the effect of scale is present at nearly every step. As the cost of acquisition, development and optimization is reduced, additional quality control and validations can be performed for every screen. Biological assessments during the CRISPR screen, such as confirmation of Cas9 activity, growth rate and gene expression analysis of each cell line and corrections for biological parameters in the downstream analysis of reads generates an accurate and robust screening setup. With such a standardized pipeline, the data can be analyzed across many cell lines and across tumour types. In addition, the DepMAP consortium has subjected one-third of these cell lines to drug sensitivity assays using a small molecule panel containing nearly 5000 drugs. The drug sensitivity data in combination with CRISPR genetic screens provide useful insights into treatment-related tumour biology, as genetic dependencies inspired or confirmed current clinical trials, for example, venetoclax treatment for neuroblastoma patients (ClinicalTrial.gov NCT04029688). As CRISPR screens are often used as a starting point for further investigations, the availability of genetic perturbation data in many cell line models is a valuable tool for many researchers.

While these efforts may initially present as the most optimal way of performing CRISPR screens, there are also many benefits to performing CRISPR screens outside of this consortium. A particular benefit of individual CRISPR screens is the use of unique cell lines, often patient-derived, that can be used to model a particular (subtype of) disease and often require delicate cell culture conditions. This would allow for patient-specific or (rare) disease-specific assessment of gene dependencies and provide a way to individualize therapies. Additional biologically relevant readouts can be added to CRISPR screens such as the use of a reporter cell line in a functional screen or using a clonogenic readout instead of cell viability. For clinical translation, reduced clonogenicity may represent stable disease, which can improve outcomes despite not being cytotoxic in vitro. Many other treatment-related questions cannot be answered using gene essentiality data as provided by the DepMAP project. As the screens performed by the DepMAP project are gene essentiality screens, the addition of treatment to perform a synthetic lethality or treatment resistance screen needs to be done outside this consortium. Indeed, with the addition of treatment, many different research questions can be answered by varying concentration, dosing and duration. As such, these investigations into early and later onset effects of genetic knock-out in combination with treatment, provide new ideas on treatment optimization. And finally, as highlighted above, the combination of multiple high-throughput screens such as genetic knock-down or drug screens with CRISPR genetic screens streamlines the bench-to-bed process.

Previous research has shown that more than 50% of driver genes common in pediatric tumours do not overlap with driver genes commonly seen in adult tumours. We can investigate if genetic dependencies are similarly different using DepMAP data. A comparison of the top 100 genetic dependencies in pediatric tumours compared to adult tumours shows no overlap of pathways enriched in this gene selection. It shows that pediatric tumours are dependent on DNA repair, telomere maintenance and epigenetic regulation pathways, whereas dependencies on receptor signalling pathways such as EGFR and transcription factors like YAP and NFκB are enriched in adult tumours.

Concluding, the DepMAP project can be seen as a valuable tool to discover (pan) cancer vulnerabilities but should not discourage other researchers to develop and use large-scale CRISPR screens, given the high variability of CRISPR screen results.

DISCUSSION

This review summarizes the use of CRISPR genetic screening in pediatric tumour cell lines, showing that performing this type of investigation is feasible, informative, and often provides clinically relevant results. CRISPR screens can not only yield new insights into tumour biology but can also be combined with various treatment modalities to study their mechanism or investigate treatment resistance. In this way, researchers may improve their understanding of tumour biology, as well as the response to treatment, providing future pediatric patients with more personalized or better treatment options. Current approaches as reviewed here have multiple issues that may need to be considered before translating laboratory results to new treatment modalities for clinical use.

Pediatric tumour models lack cell diversity: most are heavily selected clonal cell lines cultured in highly artificial environments. In addition, the availability of tumour models used for CRISPR screens does not reflect tumour incidence nor disease mortality. CRISPR screens in malignant brain tumours, for example, have been performed rarely, while this group constitutes 30% of cancers. Moreover, cell lines often consist of a single cell type, whereas the tumour microenvironment is multicellular and very complex. Overcoming the limitations set by CRISPR screens in cultured tumour cells in the future could be the use of xenograft or genetically engineered mouse models to perform these screens in vivo. This type of screen, in particular for leukaemia or (non-metastatic) solid tumours, overcomes many of the drawbacks of in vitro CRISPR screens. Another tool would be the use of spontaneous tumour-developing mouse models in combination with CRISPR to study tumour biology with an intact immune system. This type of direct in vivo CRISPR screen has been performed in other disease models already, as reviewed by Kuhn et al. Developing and using more and better models may be a leap forward in harnessing the power of CRISPR screens for transforming patient care.

One downside of using CRISPR knockout screening to study (tumour) cell biology is the mechanism in which CRISPR screens generate a loss of gene expression. Inactivation of genes in patients occurs in several ways: either by chromosomal rearrangements or losses, missense and frameshift mutations or epigenetic dysregulation. Pediatric tumours have a low mutational burden compared to adult tumours and most tumorigenic changes are mediated through epigenetic mutations. Comparing this to CRISPR-mediated gene knockout, which results in homozygous mutations inducing loss-of-function, does not reflect many of the mutations found in pediatric tumours. While investigating haploinsufficiency is not (yet) possible using CRISPR knock-out screens, knockdown screens using RNAi or CRISPRi methods can be more similar to the effect of treatment or loss of gene expression in patients. Using RNAi to validate CRISPR knockout results would overcome this downside and even better would be the combination of both a CRISPR and RNAi screen. Another aspect to consider for future pediatric CRISPR screens is the use of prime editing enzymes, which do not rely on indels created after DNA repair of DSBs to alter gene expression but can insert an early stop codon to remove gene expression. This approach could prove to be more precise, reliable and less destructive than conventional spCas9 mediated genetic knockout screens, possibly allowing cell models with limited DNA repair capabilities to be used in high throughput genetic screens.

An important part of CRISPR screens is the validation and reproducibility of results. There is a myriad of variables, such as sgRNA integration, Cas9 expression and efficiency and DNA repair mechanisms in the used cell model that may have a large impact on downstream calculations, thus affecting the reproducibility of the screen itself. A thorough analysis of Cas9-mediated mutations has shown that large deletions, complex chromosomal rearrangements and retrotransposons can occur because of DSBs generated by Cas9, thus overestimating or even replacing the effect of loss of gene expression. Subsequent validation of CRISPR screen results is of great importance, as not only the effect loss of expression needs to be confirmed but preferably the observed phenotype needs to be rescued upon re-expression of the gene of interest, showing that the observed effect is due to the gene of interest and not due to off-target effects of CRISPR-Cas mediated DSBs. Alternatively, whole genome sequencing of single-cell clones can be used to exclude large re-arrangements and other mutations. Translating these results for clinical applications is another difficult aspect: selection of a potent and precise treatment modality, albeit small molecule inhibitors or ligand traps, with good pharmacokinetics and limited side effects in patients can be a time-consuming process. Indeed, the validation and translation process of CRISPR screen results as a whole is time-consuming. As such, by making raw CRISPR screen data publicly accessible, for transparency and reproducibility, other researchers can use the same CRISPR screen data to perform their own validation experiments on other targets not investigated by the original authors. This is part of the reasoning behind the DepMAP project and should be required for publication as part of open access initiatives.

Other studies have shown that pediatric tumours are often stalled in their developmental programs related to their cell of origin, which could be translated to a developmental program specific sgRNA library to use in pediatric tumour models. In particular for transductions of tumour source material or intricate organoid model systems, where cell numbers are often limited, a smaller sgRNA library would be of great benefit to study genetic dependencies.

Many attempts have been made to establish a database of so-called ubiquitous essential genes: a set of genes whose knockout negatively affects cell growth, regardless of the cell model used. There are, however, a few caveats: the cell models used for essential gene identification are cultured in an artificial environment, where the composition of the medium does not reflect the tumour (micro) environment. Secondly, CRISPR screening has only recently been used in pediatric tumour models, the essential genes in such databases are based on (mostly) adult cell lines and therefore a poor representation. Dismissing these genes in initial analyses can hide interesting therapeutic targets. Instead, essential gene databases should be used as part of quality control of the performed CRISPR screen. Finally, known essential genes can still be an interesting target for clinical translation, as side-effects may be manageable in patients or even accepted if tumour growth is markedly reduced, as may be the case for WEE1 inhibitors for treatment of diffuse midline gliomas or PLK1 inhibitors in AML.

Despite the rarity of pediatric cancers compared to adult cancers, they present a unique challenge. Not only do pediatric tumours often have a limited number of mutations in their genome but pediatric patients also respond differently to treatment than adults. The recent development of more accurate pediatric tumour models reflecting their unique tumour micro-environment ensures that the information gathered from CRISPR genetic screens is more valuable compared to serum-cultured cell lines. Even though some publications did not find new (combinational) treatments, many publications provided new insights into the underlying tumour biology. In conclusion, this review summarizes the different methods of CRISPR screens and the many new therapeutic targets that can be used for clinical trials for pediatric cancer patients, providing ample rationale to design and perform CRISPR screens using (unique) pediatric tumour models.

AUTHOR CONTRIBUTIONS

Madelaine van Mackelenbergh prepared the figures and tables. Madelaine van Mackelenbergh, Michaël Meel and Esther Hulleman wrote the manuscript. All authors (Madelaine van Mackelenbergh, Michaël Meel, Dennis Metselaar, Esther Hulleman) edited and finalized the manuscript.

ACKNOWLEDGEMENTS

The authors would like to acknowledge Joshua Goulding, from the University of Gent (Gent, Belgium), for his initial work on this manuscript.

FUNDING

This work was supported by Stichting Semmy.

CONFLICT OF INTEREST

The authors declare no conflict of interest. The paper was handled by editors and has undergone a rigorous peer-review process. Dr. Esther Hulleman was not involved in the journal's review of/or decisions related to this manuscript.

DATA AVAILABILITY STATEMENT

Data sharing is not applicable to this article as no new data were created or analyzed in this study.

ETHICAL APPROVAL

Not applicable.

[Correction added on April 3, 2023 after first online publication: the Author Contributions, Acknowledgments, Funding, Conflicts of Interest, Data Availability Statement and Ethical approval has been updated]

References

Schulpen M, Visser O, Reedijk AMJ, et al. Significant improvement in survival of advanced stage childhood and young adolescent cancer in the Netherlands since the 1990s. Eur J Cancer Oxf Engl. 2021;157:81‐93. doi: [DOI: https://dx.doi.org/10.1016/j.ejca.2021.08.001]