Edited by: Jie Sun, Zhejiang University, China
Reviewed by: Guangchuan Wang, Center for Excellence in Molecular Cell Science (CAS), China; Kilian Schober, University Hospital Erlangen, Germany
*Correspondence: Namshik Han,
This article was submitted to Cancer Immunity and Immunotherapy, a section of the journal Frontiers in Immunology
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Cancer immunotherapy targets the interplay between immune and cancer cells. In particular, interactions between cytotoxic T lymphocytes (CTLs) and cancer cells, such as PD-1 (
Tumor immunotherapy invigorates the immune system to fight against cancer. It impedes interactions between immune cells, most notably cytotoxic T lymphocytes (CTLs), and cancer cells (
Genome-wide clustered regularly interspaced short palindromic repeats (CRISPR) screens have been increasingly used for the systematic discovery of targets for cancer therapy. The most notable examples come for targeted therapy, where a large consortium called the Cancer Dependency Map performed 1076 genome-wide CRISPR screens in 908 cell lines (
However, such CRISPR screening approaches have three main limitations. First, most studies did not focus on intercellular interactions, which underlie the mechanism of action for immunotherapy drugs, instead focused on single genes. Focusing on a single gene is suboptimal, as its protein product can interact with multiple partners (
Intercellular interactions are potent targets for cancer immunotherapy.
In this study, we introduced intercellular CRISPR screens, a computational approach for the evaluation of intercellular interactions as IO drug targets. We employed intercellular CRISPR screening for the analysis of two genome-wide CRISPR datasets, an immune cell and a cancer cell dataset. Through integrating these datasets, we calculated the “intercellular normZ score” for each intercellular interaction and quantified its potential as a target for immunotherapy. As a proof of concept, we applied the intercellular CRISPR screen to two publicly available genome-wide CRISPR screening datasets, analyzing cancer cell and CTL interactions (
The list of drugs, along with their development status (‘groups’), anatomical therapeutic chemical (ATC) codes, and target information was collected from DrugBank 5.1.8 (
Two resources were used to classify cancer drugs as IO and non-IO drugs. The first resource was the ‘Cancer immunotherapy’ category (accession number DBCAT005215) from DrugBank. The second was the IO Landscape, developed by the Cancer Research Institute to catalog the developmental status of IO drugs (
Among the ‘Cancer immunotherapy’ drugs from DrugBank, a few drugs such as trastuzumab, a
Drug target information was collected from DrugBank. We used the classification term ‘targets’ and excluded ‘enzymes’, ‘transporters’, and ‘carriers’. As a result, 282 targets for 185 approved cancer drugs were identified.
The targets were classified using ChEMBL 29 (
Drugs were categorized based on their target classification. If a drug had multiple targets, it was considered to belong to all classes into which its targets were classified.
We set five criteria for selecting CRISPR screening datasets in immune and cancer cells; (1) Genome-scale, and (2) publicly available datasets that were (3) obtained while the immune and cancer cells were interacting with each other. (4) The type of immune and cancer cells should be similar enough between the datasets and (5) transcriptome of the edited cell types should be available.
Genome-wide pooled CRISPR screens of CTLs were obtained from Dong et al. (
Overview of methods.
The normalized read count matrix was downloaded from
In the study by Dong et al. (
A genome-wide pooled CRISPR screen in two mouse TNBC cell lines, 4T1 and EMT6, was conducted by Lawson et al. (
To identify genes that were expressed in the mouse TNBC cell lines 4T1 and EMT6, bulk RNA sequencing data were obtained from
Intercellular interaction data were collected from two sources: ConnectomeDB2020 (
We performed differential analysis using the drugZ algorithm (
For the TNBC CRISPR screening dataset, we used drugZ algorithm to calculate fold changes of genes co-cultured with CTLs compared to monocultured TNBC cells. As a result, genes whose knockout help TNBC cells escape from CTL-mediated killing obtained positive normZ scores. Since knockouts of these genes prevent CTLs from killing TNBC cells, they may be components of CTL-mediated cancer cell removal. For the CTL CRISPR screening dataset, we calculated the fold changes of genes in cell libraries against the tumor samples in order to give negative scores to CTL-suppressive genes. Genes whose knockout increased CTL infiltration into tumor tissue were enriched in the tumor sample, and therefore given negative scores. These genes may prevent CTL infiltration in their normal state. As a result, positive normZ scores were given to the genes that activate CTL function, while CTL-suppressive genes had negative normZ scores for both CRISPR screening datasets.
We combined two genome-wide CRISPR screening datasets to calculate the intercellular normZ scores. Before combining the CTL CRISPR and TNBC CRISPR screens, we set the normZ scores of unexpressed genes to zero for each cell type to reduce noise resulting from the high-throughput pooled screening procedure (
After zero out the normZ scores of unexpressed genes for each cell type, the normZ score of a ligand from one cell type and the corresponding receptor’s normZ score from the other cell type were summed to calculate the cell-line-specific intercellular normZ score (
Because normZ scores follow the standard normal distribution (
Other than the proposed method in 2.2.3, we devised two alternative methods to calculate the intercellular normZ scores. The first alternative method is ‘Strict’, which explicitly requires normZ scores to have the same sign when summed. If two normZ scores with different signs have to be summed, the ‘Strict’ method sets the result as zero instead of summing the scores. On the other hand, the proposed method in 2.2.3 does not put any constraints on the signs when two normZ scores are added, hence we named it as ‘Tolerant’. The second alternative method is ‘Composite’, a hybrid of ‘Strict’ and ‘Tolerant’. When the gene-level normZ scores of CTL and TNBC screens are summed, ‘Composite’ requires them to have the same sign. However, the intercellular normZ scores for 4T1-CTL and EMT6-CTL are not required to have the same sign to calculate the final score because cancer cell lines are highly heterogenous and they may not show the same immunological effect. In total, we used three different methods (‘Strict’, ‘Tolerant’, and ‘Composite’) to calculate intercellular normZ scores.
To benchmark the performance of intercellular CRISPR screens over using either of the screening data alone, a gold-standard dataset was obtained. Approved immunotherapy drugs that modulate CTLs were used as the gold-standard dataset to discover novel immunotherapeutic targets using CTL-related CRISPR screening data. We downloaded IO Landscape data and filtered drugs whose clinical stage was ‘Approved’ (
The size of the gold-standard dataset was still limited to quantitatively evaluating and comparing the methods. Therefore, we collected a silver standard dataset composed of well-known immunomodulators as potential immunotherapeutic targets. These immunomodulators include cytokines, and co-stimulatory and co-inhibitory molecules since they compose the majority of targets of T cell modulators and other modulators (
Intercellular CRISPR screen evaluates each interaction, whereas CTL CRISPR and TNBC CRISPR screens evaluate each gene. To compare the intercellular CRISPR screen with the CTL and TNBC CRISPR screens, we aggregated interaction-level intercellular normZ scores into gene-level scores (
We used the area under the receiver operating characteristic curve (AUROC), precision, recall, and F1 scores to evaluate the ability of CRISPR screens to discover the silver standard dataset. For the AUROC, we used the normZ scores. For precision, recall, and F1 score, we used FDR < 5% to classify normZ scores as CTL-activating, CTL-suppressive, and unknown. Genes/interactions with FDR ≥ 5% were classified as unknown. The remaining genes/interactions with positive and negative normZ scores were classified as CTL-activating and CTL-suppressive, respectively.
Because we have three classes, CTL-activating, CTL-suppressive, and unknown, we used an averaging scheme to evaluate performances. There are two averaging schemes in multiclass classifications: micro- and macro-averaging. Micro-averaging weighs each instance equally. On the other hand, macro-averaging regards each class as equally important, giving more weight to instances from minor classes. Since the silver standard dataset was highly imbalanced, we adopted the macro-averaging scheme. The performance for the unknown class was not evaluated since it may include potential immunotherapeutic targets that activate or suppress CTL functions.
To discover novel immunotherapeutic targets, we focused on intercellular interactions instead of single genes. We propose the use of intercellular CRISPR screens as a pipeline to discover potentially therapeutic interactions between immune and cancer cells. As a proof of concept, we used an intercellular CRISPR screen to identify the interactions that affect CTLs in TNBC (
NormZ scores from the intercellular, CTL, and TNBC CRISPR screens. Intercellular normZ scores were calculated by the ‘Tolerant’ method.
The intercellular CRISPR screen evaluates each interaction, whereas CTL and TNBC CRISPR screen evaluates a single gene. To directly compare intercellular CRISPR screen against CTL and TNBC ones, we aggregated intercellular normZ scores involving the identical gene to obtain gene-level intercellular normZ scores. We used FDR < 5% to predict genes as CTL-activating and CTL-suppressive based on the normZ scores. The results for gene-level comparisons showed a similar tendency (
To benchmark the use of intercellular CRISPR screens (‘Tolerant’ method) relative to the use of either CRISPR screen alone, we obtained a gold-standard dataset containing data for 38 intercellular interactions targeted by approved IO drugs or phase III clinical trial drug candidates for immunotherapy.
In summary, the intercellular CRISPR screen identified two intercellular interactions,
We performed three quantitative evaluations to compare the performances of different CRISPR screens. First, we evaluated the performance of intercellular CRISPR screens calculated by the ‘Tolerant’ method. Next, we compared the ‘Tolerant’ method with two alternative methods, ‘Strict’ and ‘Composite’, to identify the best way to calculate intercellular normZ scores. Lastly, we compared the performance of intercellular CRISPR screens with CTL and TNBC CRISPR screens to estimate the degree of performance enhancement. Because there are few IO drugs, the gold-standard dataset may not be appropriate for the quantitative evaluation. Therefore, we used a silver standard dataset containing cytokines and co-stimulatory molecules as potential immunotherapeutic targets. Cytokines and co-stimulatory molecules were selected since they compose the majority of targets of immunomodulatory drugs (
The confusion matrix of the ‘Tolerant’ method is shown in
The performance of the intercellular CRISPR screen (‘Tolerant’ method).
We evaluated two alternative methods to calculate the intercellular normZ scores (
Next, we compared the performance of the intercellular CRISPR screen relative to the CTL and TNBC CRISPR screens. We calculated the AUROC, precision, recall, and F1 scores for each CRISPR screen (
Based on the intercellular normZ scores, we identified potential IO targets with previously unknown effects. We investigated 21 interactions with FDR < 5% (
Among the seven interactions identified, none was identifiable when CTL CRISPR screen was used only. On the other hand,
Highly ranked intercellular interactions for immunotherapeutic targets.
Rank | CTL gene | TNBC gene | CTL | 4T1 | EMT6 | Intercellular | Supporting literature | ||||
---|---|---|---|---|---|---|---|---|---|---|---|
normZ score | FDR | normZ score | FDR | normZ score | FDR | normZ score | FDR | ||||
9 |
|
|
1.33 | 0.546 | 4.28 | 0.003 | 3.81 | 0.011 | 5.375 | 4.62e-5 | |
10 |
|
|
1.19 | 0.546 | 2.49 | 0.552 | 5.05 | 7.5e-5 | 4.96 | 3.40e-4 | ( |
11 |
|
|
0.51 | 0.546 | 2.49 | 0.552 | 5.05 | 7.5e-5 | 4.28 | 0.003 | ( |
30 |
|
|
1.19 | 0.546 | 0.96 | 0.552 | 4.74 | 2.53e-4 | 4.04 | 0.004 | ( |
56 |
|
|
-2.72 | 0.613 | -0.88 | 0.543 | -1.04 | 0.548 | -3.68 | 0.094 | ( |
64 |
|
|
-3.28 | 0.178 | 0.36* | 0.543 | -0.02* | 0.544 | -3.28 | 0.313 | ( |
67 |
|
|
-2.52 | 0.613 | -0.91 | 0.543 | -0.42 | 0.548 | -3.185 | 0.388 | ( |
None of the genes were identifiable from CTL CRISPR screens since their FDRs are higher than 0.05. Even though Calr, Tnfrsf1a, and Tnfrsf1b were statistically significant from TNBC CRISPR screens (FDR < 0.05), Bst2, Ccl4, and Fn1 were not. * Ccl4 was not expressed in 4T1 and EMT6 TNBC cell lines. All the other genes were expressed in the corresponding cell types.
In summary, our results suggested seven intercellular interactions as immunotherapeutic targets for TNBC. Among these, four may activate CTL function, for which agonists can be investigated as IO drugs. The remaining three were suggested to suppress CTL function, for which antagonists may be investigated to treat TNBC.
In this study, we showed the utility of intercellular CRISPR screens for the discovery of immune–cancer cell interactions as IO targets, relative to focusing on single genes. Intercellular CRISPR screens integrate two CRISPR screens, one for immune and one for cancer cells, which are obtained when both cells interact with each other. Our results showed that CTL and TNBC CRISPR screens were complementary, and the intercellular CRISPR screen outperformed individual screens.
Although our results successfully identified approved IO targets and known immunomodulators, they have three limitations. First, the experimental settings for the CTL and TNBC CRISPR screens were different. The CTL and TNBC CRISPR screens were used
Despite these limitations, the intercellular CRISPR screening method identified nine (
Considering the aforementioned advantages, intercellular CRISPR screens can be used to evaluate interactions between several cancer and immune cell types such as natural killer cells or macrophages (
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/
SY and WH conceptualized the study. SY developed and evaluated the proposed method along with WH under the supervision of NH and DL. All authors contributed to writing, reading, and have approved the manuscript.
The work described and publication of this article were supported by the Bio-Synergy Research Project (NRF-2012M3A9C4048758) of the Ministry of Science and ICT through the National Research Foundation. NH and WH were funded by LifeArc.
NH is a cofounder of KURE.ai and CardiaTec Biosciences and an advisor at Biorelate, Promatix, Standigm, VeraVerse, and Cellaster.
The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
The Supplementary Material for this article can be found online at:
CRISPR, clustered regularly interspaced short palindromic repeats; CTL, cytotoxic T lymphocyte; TNBC, triple-negative breast cancer; IO, immuno-oncology; ATC, anatomical therapeutic chemical; sgRNA, single guide RNA; TIL, tumor-infiltrating lymphocyte; KO, knockout; CTRL, control; diff, differential analysis; exp, expression; comb, combination; FDR, false discovery rate; AUROC, area under the receiver operating characteristic curve.