Transcriptome-based macroautophagy and chaperone-mediated autophagy states reveal context-dependent vulnerabilities across human cancers

Macroautophagy (MA) and chaperone-mediated autophagy (CMA) are lysosomal degradation pathways with context-dependent roles in cancer. However, how MA- and CMA-associated transcriptional states jointly relate to cancer molecular features and drug responses remains unclear. Here, we developed a transcriptome-based framework integrating MA- and CMA-associated gene signatures to define relative MA and CMA states across cancer types. These states were associated with distinct patterns of genomic instability, oncogenic signaling, immune features, and pharmacogenomic profiles. In pooled cancer cell-line analyses, MAhigh CMAhigh states were associated with relative resistance to EGFR tyrosine kinase inhibitors (EGFR-TKIs), whereas MAhigh CMAlow states showed greater sensitivity, nominating an autophagy-associated pharmacological pattern for mechanistic investigation. Using KRAS-mutant pancreatic cancer cells as an autophagy-dependent, therapy-resistant model, we found that LAMP2A depletion enhanced sensitivity to EGFR-TKIs in an MA-dependent manner. All-trans retinoic acid (ATRA), a pleiotropic retinoid that modulated CMA-related activity in this system, phenocopied key effects of genetic CMA suppression and potentiated sensitivity to EGFR-TKIs. CMA suppression was associated with increased autophagic flux, TFEB nuclear localization, and ULK1 phosphorylation changes consistent with MA activation. Moreover, transcriptomic analysis reveals that CMA suppression sensitizes cancer cells to EGFR-TKIs at least partially through downregulation of SEMA6D. SEMA6D depletion enhanced autophagic flux, increased lysosomal capacity, and partially contributed to the response to combined EGFR-TKI and ATRA treatment. In PANC-1 xenografts, ATRA potentiated EGFR-TKI-mediated tumor suppression without significant toxicity. Together, these findings establish a transcriptome-based MA-CMA framework for prioritizing context-dependent autophagy-associated vulnerabilities and provide focused mechanistic support for MA-CMA crosstalk in KRAS-mutant pancreatic cancer models.