Science & Nature
Data availability
The list of cell surface proteins was obtained from the in silico human surfaceome database at http://wlab.ethz.ch/surfaceome/ (ref. 23). The tumor-normal single-cell meta-atlas is available at https://cellatlas.kaist.ac.kr/cart. Raw and processed datasets used in this study are deposited at the Zenodo repository (https://doi.org/10.5281/zenodo.7416669)50. Ovarian cancer single-cell transcriptome data generated in this work are available on the Gene Expression Omnibus with accession number GSE192898 (ref. 51). Source data are provided with this paper.
Code availability
Source codes for CAR target antigen identification, named PCASA (prioritization of combinatorial cancer-associated surface antigens), are available on GitHub at https://github.com/kaistomics/PCASA (ref. 52).
References
Sterner, R. C. & Sterner, R. M. CAR-T cell therapy: current limitations and potential strategies. Blood Cancer J. 11, 69 (2021).
Marofi, F. et al. CAR T cells in solid tumors: challenges and opportunities. Stem Cell Res. Ther. 12, 81 (2021).
Taylor, A. M. et al. Genomic and functional approaches to understanding cancer aneuploidy. Cancer Cell 33, 676–689 (2018).
Fan, J. et al. Linking transcriptional and genetic tumor heterogeneity through allele analysis of single-cell RNA-seq data. Genome Res. 28, 1217–1227 (2018).
Patel, A. P. et al. Single-cell RNA-seq highlights intratumoral heterogeneity in primary glioblastoma. Science 344, 1396–1401 (2014).
Gao, R. et al. Delineating copy number and clonal substructure in human tumors from single-cell transcriptomes. Nat. Biotechnol. 39, 599–608 (2021).
Pasquini, G., Rojo Arias, J. E., Schäfer, P. & Busskamp, V. Automated methods for cell type annotation on scRNA-seq data. Comput. Struct. Biotechnol. J. 19, 961–969 (2021).
Parker, K. R. et al. Single-cell analyses identify brain mural cells expressing CD19 as potential off-tumor targets for CAR-T immunotherapies. Cell 183, 126–142 (2020).
Jing, Y. et al. Expression of chimeric antigen receptor therapy targets detected by single-cell sequencing of normal cells may contribute to off-tumor toxicity. Cancer Cell 39, 1558–1559 (2021).
Wilkie, S. et al. Dual targeting of ErbB2 and MUC1 in breast cancer using chimeric antigen receptors engineered to provide complementary signaling. J. Clin. Immunol. 32, 1059–1070 (2012).
Hege, K. M. et al. Safety, tumor trafficking and immunogenicity of chimeric antigen receptor (CAR)-T cells specific for TAG-72 in colorectal cancer. J. Immunother. Cancer 5, 22 (2017).
Yang, Y. et al. Bispecific CAR T cells against EpCAM and inducible ICAM-1 overcome antigen heterogeneity and generate superior antitumor responses. Cancer Immunol. Res. 9, 1158–1174 (2021).
Davies, D. M. & Maher, J. Gated chimeric antigen receptor T-cells: the next logical step in reducing toxicity? Transl. Cancer Res. 5, S61–S65 (2016).
Cho, J. H., Collins, J. J. & Wong, W. W. Universal chimeric antigen receptors for multiplexed and logical control of T cell responses. Cell 173, 1426–1438 (2018).
Han, X., Wang, Y., Wei, J. & Han, W. Multi-antigen-targeted chimeric antigen receptor T cells for cancer therapy. J. Hematol. Oncol. 12, 128 (2019).
Hu, Z. et al. The cancer surfaceome atlas integrates genomic, functional and drug response data to identify actionable targets. Nat. Cancer 2, 1406–1422 (2021).
Orentas, R. J. et al. Paired expression analysis of tumor cell surface antigens. Front .Oncol. 7, 173 (2017).
MacKay, M. et al. The therapeutic landscape for cells engineered with chimeric antigen receptors. Nat. Biotechnol. 38, 233–244 (2020).
Dannenfelser, R. et al. Discriminatory power of combinatorial antigen recognition in cancer T cell therapies. Cell Syst. 11, 215–228 (2020).
Aran, D. et al. Reference-based analysis of lung single-cell sequencing reveals a transitional profibrotic macrophage. Nat. Immunol. 20, 163–172 (2019).
Hie, B., Cho, H., DeMeo, B., Bryson, B. & Berger, B. Geometric sketching compactly summarizes the single-cell transcriptomic landscape. Cell Syst 8, 483–493 (2019).
Park, J. E. et al. A cell atlas of human thymic development defines T cell repertoire formation. Science 367, eaay3224 (2020).
Bausch-Fluck, D. et al. The in silico human surfaceome. Proc. Natl Acad. Sci. USA 115, E10988–E10997 (2018).
Selvaraju, R. R. et al. Grad-CAM: visual explanations from deep networks via gradient-based localization. Proc. IEEE International Conference on Computer Vision https://ieeexplore.ieee.org/document/8237336 (2017).
Chattopadhay, A., Sarkar, A., Howlader, P. & Balasubramanian, V. N. Grad-CAM++: generalized gradient-based visual explanations for deep convolutional networks. in: 2018 IEEE Winter Conference on Applications of Computer Vision (WACV). https://doi.org/10.1109/WACV.2018.00097 (2018).
Cancer Genome Atlas Research Networket al. The Cancer Genome Atlas Pan-Cancer analysis project. Nat. Genet. 45, 1113–1120 (2013).
Lonsdale, J. et al. The Genotype-Tissue Expression (GTEx) project. Nat. Genet. 45, 580–585 (2013).
Stoeckius, M. et al. Simultaneous epitope and transcriptome measurement in single cells. Nat. Methods 14, 865–868 (2017).
Nakamura, K. et al. CD24 expression is a marker for predicting clinical outcome and regulates the epithelial-mesenchymal transition in ovarian cancer via both the Akt and ERK pathways. Oncol. Rep. 37, 3189–3200 (2017).
Yang, S. et al. Decontamination of ambient RNA in single-cell RNA-seq with DecontX. Genome Biol. 21, 57 (2020).
Tiernan, J. P. et al. Carcinoembryonic antigen is the preferred biomarker for in vivo colorectal cancer targeting. Br. J. Cancer 108, 662–667 (2013).
Zhang, J. et al. Carbonic anhydrase IV inhibits colon cancer development by inhibiting the Wnt signalling pathway through targeting the WTAP–WT1–TBL1 axis. Gut 65, 1482–1493 (2016).
Kharchenko, P. V., Silberstein, L. & Scadden, D. T. Bayesian approach to single-cell differential expression analysis. Nat. Methods 11, 740–742 (2014).
Qiu, P. Embracing the dropouts in single-cell RNA-seq analysis. Nat. Commun. 11, 1169 (2020).
Zheng, G. X. Y. et al. Massively parallel digital transcriptional profiling of single cells. Nat. Commun. 8, 14049 (2017).
T, S. et al. Comprehensive integration of single-cell data. Cell 177, 1888–1902 (2019).
Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol. 19, 15 (2018).
Martens, J. H. A. & Stunnenberg, H. G. BLUEPRINT: mapping human blood cell epigenomes. Haematologica 98, 1487–1489 (2013).
Davis, C. A. et al. The Encyclopedia of DNA Elements (ENCODE): data portal update. Nucleic Acids Res. 46, D794–D801 (2018).
Mabbott, N. A., Baillie, J. K., Brown, H., Freeman, T. C. & Hume, D. A. An expression atlas of human primary cells: inference of gene function from coexpression networks. BMC Genomics 14, 632 (2013).
Karlsson, M. et al. A single-cell type transcriptomics map of human tissues. Sci. Adv. 7, eabh2169 (2021).
Polański, K. et al. BBKNN: fast batch alignment of single cell transcriptomes. Bioinformatics 36, 964–965 (2020).
Tweedie, S. et al. Genenames.org: the HGNC and VGNC resources in 2021. Nucleic Acids Res. 49, D939–D946 (2021).
GTEx Consortium. The Genotype-Tissue Expression (GTEx) project. Nat. Genet. 45, 580 (2013).
Jiang, L. et al. A quantitative proteome map of the human body. Cell 183, 269–283 (2020).
Breiman, L. Random forests. Machine Learning 45, 5–32 (2001).
Sing, T., Sander, O., Beerenwinkel, N. & Lengauer, T. ROCR: visualizing classifier performance in R. Bioinformatics 21, 3940–3941 (2005).
Abadi, M. et al. TensorFlow: a system for large-scale machine learning. Proc. 12th USENIX Symposium on Operating Systems Design and Implementation. https://www.usenix.org/system/files/conference/osdi16/osdi16-abadi.pdf (2016).
Kingma, D. P. & Ba, J. Adam: a method for stochastic optimization. Proc. 3rd International Conference on Learning Representations. https://doi.org/10.48550/arXiv.1412.6980 (2014).
Kwon, J. Integrative single-cell mapping of combinatorial target antigens for logical CAR switches. https://doi.org/10.5281/zenodo.7416669 (2022).
An, H. J. Single cell 5′ RNA sequencing of 9 Korean ovarian cancer patients. Gene Expression Omnibus. https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?&acc=gse192898 (2023).
omics lab @ KAIST. kaistomics/PCASA: PCASA. https://doi.org/10.5281/zenodo.7514043 (2023).
Acknowledgements
This work was supported by the Bio & Medical Technology Development Program of the National Research Foundation, funded by the Ministry of Science & ICT (NRF-2017M3A9A7050612, NRF-2019M3A9B6064688, NRF-2019M3A9B6064691 and NRF-2021M3A9I402444711), and also supported by the Korea Health Technology R&D Project of the Korea Health Industry Development Institute, funded by the Ministry of Health & Welfare (HI16C1559).
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Nature Biotechnology thanks Raphael Gottardo, Peter Linsley and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data
Extended Data Fig. 1 UMAP representation of the normal cell atlas overlaid with tumor cells and tumor-infiltrating normal cells from selected tumor datasets.
. a, Same plot as Fig. 1e but with cell-type annotation instead of tissue origin information. b, Same plot as Fig. 1e but showing representative cells selected during tumor cellpreserving subsampling by geometric sketching.
Extended Data Fig. 2 Tumor versus normal differences in ECFs and expression levels.
Differences of ECFs (a) and expression levels (b) between tumor and normal cellsfor surfaceome genes ordered by the FI values from the RF model in selected cancer types.
Extended Data Fig. 3 Comparison of cancer types with expression level and FI for surfaceome genes.
a, Clustering heatmap displaying the expression level of surface antigens according to the cancer type. Five major clusters (C1 ~ C5) of genes were denoted. b, Comparison of the different cancer types based on the correlation of the FI values from the cancer type-specific RF model.
Extended Data Fig. 4 Tumor-specific ECF values of subtype marker genes in BRCA samples for which clinical subtype information was available.
The boxplot shows the ECF status of three markers, containing ERBB2, ESR1, and PGR, for all subtypes (ERpositive: n = 11, HER2-positive: n = 8, HER2 and ER double positive: n = 4, TNBC: n = 12): TNBC, Triple-negative breast cancer; ‘+‘ means ‘positive’. Average values of ECFs for ERBB2, ESR1, and PGR, were 0.33 ± 0.15 (s.d.), 0.68 ± 0.15 (s.d), and 0.29 ± 0.18 (s.d.) in ER-positive subtype while HER2-positive subtype showed 0.86 ± 0.19 (s.d.), 0.18 ± 0.3 (s.d.), and 0.01 ± 0.01 (s.d.), respectively. In the case of HER2+ ER+, average values were 0.47 ± 0.19 (s.d.), 0.25 ± 0.23 (s.d.), and 0.06 ± 0.05 (s.d.), and the TNBC subtype showed 0.24 ± 0.14 (s.d), 0.05 ± 0.12 (s.d.), and 0.05 ± 0.15 (s.d.), for ERBB2, ESR1, and PGR, respectively. Data in the box plot represent the first quartile (25%), median (center), and third quartile (75%) with minimum and maximum values. Black points indicate the outliers.
Extended Data Fig. 5 Correlation between the CNN weight and ECF of each antigen combination in PAAD, LIHC, NSCLC, CRC, and BRCA.
For each of the pairs, the GradCAM weight returned by the CNN model was mapped to the ECF among tumor cells (blue) and normal cells (red). The Pearson correlation coefficient and two-sided P values are shown at the top. Translucent bands around the regression lines represent the confidence interval (CI) of 95%.
Extended Data Fig. 6 Comparison of tumor versus normal ECFs according to the CNN weight in PAAD, LIHC, NSCLC, CRC, and BRCA.
Tumor and normal ECFs of the combinations were ordered by their CNN weight. The combinations meeting the ECF criteria (> 70% tumor and < 10% normal) were highlighted in color.
Extended Data Fig. 7 Sample-wise co-expression of selected combinations in OV.
Expression correlations of the OV gene pairs were computed by using pan-cancer TCGA samples (n = 11,768) and GTEx normal samples (n = 17,382). The Pearson correlation coefficient was used.
Extended Data Fig. 8 Across-sample expression patterns of selected logical CAR switches in PAAD, LIHC, NSCLC, CRC, and BRCA.
Shown are the top 10 gene combinations with the largest CNN weight for each cancer type and each logic switch (AND, OR, and NOT). The heatmap intensity scales with the combinatorial tumor (left) and normal (right) ECF values.
Extended Data Fig. 9 Additional single-cell epitope analysis results for EPCAM and CD24 in ovarian cancer.
a, Distribution of EPCAM and CD24 protein expression levels broken down by annotated cell type. Log2 fold changes comparing CITE-seq ADT signals from targeting antibodies versus isotype controls were computed and are displayed in each cell type. b, Clustering and cell-type mapping of tumor and tumor-resident normal cells based on the CITE-seq transcriptome. c, Single-cell maps, as in (b), overlaid with mRNA expression levels retrieved from CITEseq. d, Single-cell maps, as in (b), overlaid with protein expression levels retrieved from CITE-seq. e, Single-cell maps, as in (b), overlaid with the mRNA expression status indicating cells with both EPCAM and CD24 (brown), only EPCAM (light blue), only CD24 (light red), or none (gray). f, Single-cell maps, as in (b), overlaid with the epitope expression status indicating cells with both EPCAM and CD24 (brown), only EPCAM (light blue), only CD24 (light red), or none (gray).
Extended Data Fig. 10 Single-cell epitope analysis of validation targets in colorectal cancer.
a, Clustering and cell-type mapping of the tumor and tumor-infiltrating normal cells based on the CITE-seq transcriptome of our pooled CRC samples. b, Single-cell maps, as in (a), overlaid with mRNA expression levels retrieved from the CITE-seq experiments. c, Single-cell maps, as in (b), overlaid with protein expression levels retrieved from the CITE-seq experiments. d, Distribution of protein expression levels broken down by annotated cell type. Log2 fold changes comparing CITE-seq ADT signals from targeting antibodies versus isotype controls were computed and are displayed for each protein in each cell type. e, Cell-by-cell protein expression patterns in the tumor and tumor-resident normal cells for CEACAM5 versus EPCAM, CEACAM5 versus CA4, CEACAM5 versus CPM, and CEACAM5 versus VSIG2. Cells enclosed in the marked zones indicate the expression of both proteins for the AND gating and only one protein (CEACAM5) for the NOT gating.
Supplementary information
Reporting Summary
Supplementary Tables 1–9
Supplementary Table 1: Summary of our cancer cell atlas. Supplementary Table 2: Status of cell types for tumor datasets integrated with normal cell atlas. Supplementary Table 3: Summary of selected tumor datasets of six cancer types. Supplementary Table 4: Performance of RF cell classifier for each cancer type. Supplementary Table 5: RF FI for top 100 genes for each cancer type. Supplementary Table 6: CNN weights calculated by Grad-CAM for each cancer type. Supplementary Table 7: Performance of CNN cell classifier for each cancer type. Supplementary Table 8: Sample-wise co-expression for OV target combinations. Supplementary Table 9: Normal ECF distribution by cell types for OV target combinations.
Source data
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Kwon, J., Kang, J., Jo, A. et al. Single-cell mapping of combinatorial target antigens for CAR switches using logic gates.
Nat Biotechnol (2023). https://doi.org/10.1038/s41587-023-01686-y
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41587-023-01686-y
Read More
Joonha Kwon