Ma Yushu, Shih Chien-Hung, Cheng Jinxiong, Chen Hsiao-Chun, Wang Li-Ju, Tan Yanhao, Zhang Yuan, Brown Daniel D, Oesterreich Steffi, Lee Adrian V, Chiu Yu-Chiao, Chen Yu-Chih
UPMC Hillman Cancer Center, University of Pittsburgh, 5115 Centre Ave, Pittsburgh, Pennsylvania 15232, United States.
Department of Computational and Systems Biology, University of Pittsburgh, 3420 Forbes Avenue, Pittsburgh, Pennsylvania 15260, United States.
Anal Chem. 2025 Mar 18;97(10):5498-5506. doi: 10.1021/acs.analchem.4c05138. Epub 2025 Mar 4.
Therapy resistance in breast cancer is increasingly attributed to polyploid giant cancer cells (PGCCs), which arise through whole genome doubling and exhibit heightened resilience to standard treatments. Characterized by enlarged nuclei and increased DNA content, these cells tend to be dormant under therapeutic stress, driving disease relapse. Despite their critical role in resistance, strategies to effectively target PGCCs are limited, largely due to the lack of high-throughput methods for assessing their viability. Traditional assays lack the sensitivity needed to detect PGCC-specific elimination, prompting the development of novel approaches. To address this challenge, we developed a high-throughput single-cell morphological analysis workflow designed to differentiate compounds that selectively inhibit non-PGCCs, PGCCs, or both. Using this method, we screened a library of 2726 FDA Phase 1-approved drugs, identifying promising anti-PGCC candidates, including proteasome inhibitors, FOXM1, CHK, and macrocyclic lactones. Notably, RNA-Seq analysis of cells treated with the macrocyclic lactone Pyronaridine revealed AXL inhibition as a potential strategy for targeting PGCCs. Although our single-cell morphological analysis pipeline is powerful, empirical testing of all existing compounds is impractical and inefficient. To overcome this limitation, we trained a machine learning model to predict anti-PGCC efficacy , integrating chemical fingerprints and compound descriptions from prior publications and databases. The model demonstrated a high correlation with experimental outcomes and predicted efficacious compounds in an expanded library of over 6,000 drugs. Among the top-ranked predictions, we experimentally validated five compounds as potent PGCC inhibitors using cell lines and patient-derived models. These findings underscore the synergistic potential of integrating high-throughput empirical screening with machine learning-based virtual screening to accelerate the discovery of novel therapies, particularly for targeting therapy-resistant PGCCs in breast cancer.
乳腺癌的治疗耐药性越来越多地归因于多倍体巨癌细胞（PGCCs），这些细胞通过全基因组加倍产生，对标准治疗表现出更强的抵抗力。这些细胞以细胞核增大和DNA含量增加为特征，在治疗压力下往往处于休眠状态，导致疾病复发。尽管它们在耐药性中起关键作用，但有效靶向PGCCs的策略有限，主要是由于缺乏评估其活力的高通量方法。传统检测方法缺乏检测PGCC特异性消除所需的灵敏度，这促使了新方法的开发。为应对这一挑战，我们开发了一种高通量单细胞形态分析工作流程，旨在区分选择性抑制非PGCCs、PGCCs或两者的化合物。使用这种方法，我们筛选了一个包含2726种美国食品药品监督管理局（FDA）批准的1期药物的文库，确定了有前景的抗PGCC候选药物，包括蛋白酶体抑制剂、FOXM1、CHK和大环内酯类。值得注意的是，对用大环内酯类药物磷酸咯萘啶处理的细胞进行的RNA测序分析显示，抑制AXL是靶向PGCCs的一种潜在策略。尽管我们的单细胞形态分析流程很强大，但对所有现有化合物进行实证测试既不切实际也效率低下。为克服这一限制，我们训练了一个机器学习模型来预测抗PGCC疗效，整合了来自先前出版物和数据库的化学指纹和化合物描述。该模型与实验结果显示出高度相关性，并在一个超过6000种药物的扩展文库中预测了有效化合物。在排名靠前的预测结果中，我们使用细胞系和患者来源模型通过实验验证了五种化合物是有效的PGCC抑制剂。这些发现强调了将高通量实证筛选与基于机器学习的虚拟筛选相结合的协同潜力，以加速新型疗法的发现，特别是针对乳腺癌中对治疗耐药的PGCCs。
