PO.CL01.12 · 临床研究
整合bulk、单细胞和空间转录组学的多组学分析确定头颈部涎腺腺样囊性癌的稳健预后生物标志物
Multi-omics integration of bulk, single-cell, and spatial transcriptomics identifies robust prognostic biomarkers in head and neck salivary adenoid cystic carcinoma
作者与单位 Authors & Affiliations
摘要 Abstract
中文摘要
头颈部腺样囊性癌（ACC）是一种罕见的恶性肿瘤，具有矛盾的临床病程——生长缓慢但高度侵袭性，治疗选择有限，且缺乏经过验证的分子预后生物标志物。为解决这一未被满足的需求，我们实施了一项全面的多组学策略，整合了来自八个不同解剖亚部位ACC肿瘤的bulk RNA测序（n=20）、单细胞RNA测序（n=24）和高分辨率空间转录组学（4个Visium HD样本）数据。临床元数据能够将患者分层为"不良"（<2年生存）和"良好"（>5年生存）预后组。在缺乏明确死因数据的情况下，我们开发了一种基于机器学习的分类器来定义转录组预后亚组，揭示了与临床结局相符的生物学一致性簇。我们的生物标志物发现流程涵盖三个关键阶段：
1. 跨模态差异表达和通路分析：我们识别了在bulk、单细胞和空间模态中区分不良与良好预后肿瘤的保守基因表达特征和失调通路。高风险肿瘤在共有的细胞类型和解剖区域中一致地表现出致癌信号（如MYC、NOTCH）、免疫抑制和基质激活程序的富集。
2. 空间解析的细胞生态系统图谱：单细胞与空间转录组学的整合能够精确定位与不良预后相关的恶性细胞状态和免疫龛。空间分析揭示了以致癌活性升高、免疫排斥和基质重塑为特征的瘤内"热点"。经空间邻近性验证的配体-受体相互作用网络揭示了驱动肿瘤进展的关键信号轴（如CXCL12-CXCR4、TGFB1-TGFBR2）。
3. 预后生物标志物面板的开发：我们构建了一个在所有模态中可重现的机器学习衍生多基因特征。与现有ACC基因集相比，该面板展现出更优的风险分层性能，其预后准确性独立于临床特征。重要的是，该生物标志物面板可通过bulk RNA分析实现临床转化，为患者分层和治疗决策提供即时效用。总之，我们的整合多组学方法揭示了ACC不良预后背后稳健的分子程序和空间定义的细胞生态系统。由此产生的生物标志物面板为精准预后预测提供了强大工具，并为这一具有挑战性的恶性肿瘤的靶向治疗开发奠定了基础。
查看英文原文 English abstract
Adenoid Cystic Carcinoma (ACC) of the head and neck is a rare malignancy with a paradoxical clinical course, slow-growing yet highly invasive, with limited therapeutic options and no validated molecular prognostic biomarkers. To address this unmet need, we implemented a comprehensive multi-omics strategy integrating bulk RNA sequencing (n=20), single-cell RNA sequencing (n=24), and high-resolution spatial transcriptomics (4 Visium HD samples) from ACC tumors spanning eight distinct anatomical subsites. Clinical metadata enabled stratification into “poor” (<2-year survival) and “good” (>5-year survival) prognosis groups. In the absence of definitive cause-of-death data, we developed a machine learning-based classifier to define transcriptomic prognosis subgroups, revealing biologically coherent clusters aligned with clinical outcomes.Our biomarker discovery pipeline encompassed three key phases:
1.Cross-Modality Differential Expression and Pathway Profiling: We identified conserved gene expression signatures and dysregulated pathways distinguishing poor from good prognosis tumors across bulk, single-cell, and spatial modalities. High-risk tumors exhibited consistent enrichment of oncogenic signaling (e.g., MYC, NOTCH), immune suppression, and stromal activation programs across shared cell types and anatomical regions.
2. Spatially Resolved Cellular Ecosystem Mapping: Integration of single-cell and spatial transcriptomics enabled precise localization of malignant cell states and immune niches associated with poor prognosis. Spatial analyses revealed intratumoral “hotspots” characterized by elevated oncogenic activity, immune exclusion, and stromal remodeling. Ligand-receptor interaction networks, validated by spatial proximity, uncovered key signaling axes (e.g., CXCL12-CXCR4, TGFB1-TGFBR2) driving tumor progression.
3. Development of a Prognostic Biomarker Panel: We constructed a machine learning-derived multi-gene signature reproducible across all modalities. This panel demonstrated superior risk-stratification performance compared with existing ACC gene sets, with prognostic accuracy independent of clinical features. Importantly, the biomarker panel is amenable to clinical translation via bulk RNA profiling, offering immediate utility for patient stratification and therapeutic decision-making. In summary, our integrative multi-omics approach reveals robust molecular programs and spatially defined cellular ecosystems underlying poor prognosis in ACC. The resulting biomarker panel offers a powerful tool for precision prognostication and lays the foundation for targeted therapeutic development in this challenging malignancy.
利益披露 Disclosure
G. Bijukumar, None..
J. S. Edwards, None..
V. Palanisamy, None.