Genome Med. 2022;14(1):111.

Transcriptional signatures of the BCL2 family for individualized acute myeloid leukaemia treatment.

, , , , , , , ,

제1저자 이찬섭, 이선영 · 교신저자 윤성수, 윤홍석, 고영일

PMID 36171613 PMC9520894 DOI 10.1186/s13073-022-01115-w IF 10.8 혈액종양내과

소속 (13)

Cancer Research Institute, Seoul National University College of Medicine, Seoul, Republic of Korea.
Center for Medical Innovation, Seoul National University Hospital, Seoul, Republic of Korea.
Department of Genomic Medicine, Seoul National University Hospital, Seoul, Republic of Korea.
Center for Precision Medicine, Seoul National University Hospital, Seoul, Republic of Korea.
Division of Hematology and Medical Oncology, Department of Internal Medicine, Seoul National University Hospital, Seoul, Republic of Korea.
Department of Genomic Medicine, Seoul National University Hospital, Seoul, Republic of Korea. entropy.yun@gmail.com.
Center for Precision Medicine, Seoul National University Hospital, Seoul, Republic of Korea. entropy.yun@gmail.com.
Cancer Research Institute, Seoul National University College of Medicine, Seoul, Republic of Korea. snuhgo01@gmail.com.
Center for Medical Innovation, Seoul National University Hospital, Seoul, Republic of Korea. snuhgo01@gmail.com.
Division of Hematology and Medical Oncology, Department of Internal Medicine, Seoul National University Hospital, Seoul, Republic of Korea. snuhgo01@gmail.com.
Cancer Research Institute, Seoul National University College of Medicine, Seoul, Republic of Korea. ssysmc@snu.ac.kr.
Center for Medical Innovation, Seoul National University Hospital, Seoul, Republic of Korea. ssysmc@snu.ac.kr.
Division of Hematology and Medical Oncology, Department of Internal Medicine, Seoul National University Hospital, Seoul, Republic of Korea. ssysmc@snu.ac.kr.

Abstract

BACKGROUND Although anti-apoptotic proteins of the B-cell lymphoma-2 (BCL2) family have been utilized as therapeutic targets in acute myeloid leukaemia (AML), their complicated regulatory networks make individualized therapy difficult. This study aimed to discover the transcriptional signatures of BCL2 family genes that reflect regulatory dynamics, which can guide individualized therapeutic strategies.

METHODS From three AML RNA-seq cohorts (BeatAML, LeuceGene, and TCGA; n = 451, 437, and 179, respectively), we constructed the BCL2 family signatures (BFSigs) by applying an innovative gene-set selection method reflecting biological knowledge followed by non-negative matrix factorization (NMF). To demonstrate the significance of the BFSigs, we conducted modelling to predict response to BCL2 family inhibitors, clustering, and functional enrichment analysis. Cross-platform validity of BFSigs was also confirmed using NanoString technology in a separate cohort of 47 patients.

RESULTS We established BFSigs labeled as the BCL2, MCL1/BCL2, and BFL1/MCL1 signatures that identify key anti-apoptotic proteins. Unsupervised clustering based on BFSig information consistently classified AML patients into three robust subtypes across different AML cohorts, implying the existence of biological entities revealed by the BFSig approach. Interestingly, each subtype has distinct enrichment patterns of major cancer pathways, including MAPK and mTORC1, which propose subtype-specific combination treatment with apoptosis modulating drugs. The BFSig-based classifier also predicted response to venetoclax with remarkable performance (area under the ROC curve, AUROC = 0.874), which was well-validated in an independent cohort (AUROC = 0.950). Lastly, we successfully confirmed the validity of BFSigs using NanoString technology.

CONCLUSIONS This study proposes BFSigs as a biomarker for the effective selection of apoptosis targeting treatments and cancer pathways to co-target in AML.

Keywords

Acute myeloid leukaemiaB-cell lymphoma-2 familyTranscriptional signatures

Figures

**Fig. 1** Flowchart of the study. From AML RNA-seq datasets (BeatAML, LeuceGene, and TCGA), gene optimization is independently conducted for capturing regulation factors of the BCL2 family. Afterward, the BCL2 family signatures (BFSigs) are calculated using the selected genes. Using the signatures, three novel subtypes are identified and functionally characterized. A classifier for predicting venetoclax response is developed and validated. Additionally, drug response analysis reveals the signature-based subtype-specific drug sensitivity. Finally, the validity of our selected genes is confirmed in a custom NanoString panel. *For external validation of BeatAML results, the BFSigs were re-extracted after batch effect correction

**Fig. 2** Identification of BCL2 family-based acute myeloid leukaemia (AML) subtypes. A Profiles of BCL2 family signatures calculated using optimized genes in each RNA-seq dataset (BeatAML, LeuceGene, and TCGA). These datasets show three distinct clusters annotated as BCL2, MCL1/BCL2, and BFL1/MCL1 signature subtypes. Columns are clustered using hierarchical clustering with average distance. B Sample proportion of these subtypes. C Comparison of BCL2 family signatures between venetoclax response groups in BeatAML (81 sensitive and 72 resistant) and LeuceGene (20 sensitive and 3 resistant). P-values are calculated by Welch’s t-test. * < 0.05, ** < 0.01, ns > 0.10. D A profile of BCL2 family signatures in the NanoString dataset. The samples are also divided into three clusters resulting from RNA-seq datasets

**Fig. 3** Concordance of BCL2 family signatures between acute myeloid leukaemia (AML) datasets. Weight of optimized genes in the definition of the BCL2 family signatures (BCL2, MCL1/BCL2, and BFL1/MCL1 signature). BCL2, MCL1, and BFL1 are marked in cyan, magenta, and yellow, respectively. Some determinant components of the signatures are marked in black. Four AML datasets show the consistent weight of optimized genes. Each weight of genes is normalized to sum 1. Correlation coefficients are calculated using Spearman’s rho

**Fig. 4** Functional analysis of BCL2 family-based acute myeloid leukaemia (AML) subtypes. Gene set enrichment analysis (GSEA) from the comparison between one subtype and the others identifies enriched gene sets in each subtype. Enrichment patterns are consistent across three AML datasets. NES indicates a normalized enrichment score. The gene set of the MAPK pathway is from the GO database. The others are from the hallmark database

**Fig. 5** Association between BFSigs and drug responses. Comparison of drug responses between the subtypes in BeatAML and Tavor datasets. The lower the y-axis value, the more sensitive to the drugs. P-values above the panel and between the box plots are calculated by Kruskal-Wallis test and Wilcoxon rank-sum test, respectively. * < 0.05, ** < 0.01, *** < 0.001, ns > 0.05

**Fig. 6** Prediction response to venetoclax. A Probability of sensitivity to venetoclax calculated from BCL2 family signature-based logistic regression model in training set (BeatAML; 81 sensitive and 72 resistant) and external validation set (LeuceGene and Tavor; 20+32 sensitive and 3+2 resistant). The sensitivity probabilities of BeatAML represent the average probability from the 10-times repeated training-testing scheme. Those of LeuceGene and Tavor are calculated from the whole BeatAML-based classifier. P-values are calculated using Wilcoxon rank-sum test by comparing the probability rank between the response groups. B Comparison of prediction performance between venetoclax response classifiers. The black bar indicates the BCL2 family signature-based logistic regression model. The dark grey bars indicate logistic regression models using the original expression of five BCL family genes (BCL2+MCL1+BFL1+BCLXL+BCLW), three BCL2 family genes (BCL2+MCL1+BFL1), and top differentially expressed genes (DEGs), respectively. The light grey bars indicate machine learning-based models using total genes or pre-collected genes related to BCL2 family regulation. The used machine learning methods are support vector machine, Lasso, and random forest (RF). Error bar indicates 95% confidence interval (CI). P-values are calculated compared with the signature model using DeLong’s test. * < 0.05, ** < 0.01, *** < 0.001

그림 출처: PubMed Central 공개 논문 (PMC9520894)

Conflict of interest statement

The authors declare that they have no competing interests.

그 밖의 정보