Back

A Pan-Cancer Analysis of Lesion-Level Treatment Response to Extend the 'Seed and Soil' Paradigm

Dilley, I.; Zhou, J.; Li, Q.; Cao, Y.

2025-09-18 oncology
10.1101/2025.09.16.25335893 medRxiv
Show abstract

BackgroundThe classical "seed and soil" hypothesis suggests that metastatic spread is shaped by tumor-intrinsic traits ("seeds") and the organ-specific microenvironment ("soil"). We expand this concept to explain lesion-level therapeutic responses and phenotypic variability across metastatic cancers. MethodsWe analyzed 55,220 lesions from 6,087 patients enrolled in 20 clinical trials across six cancer types. Using nonlinear mixed-effects modeling, we estimated lesion-specific parameters: regression rate (kkill), progression rate (kge), and resistant fraction (Fx). Multivariable Cox models, adjusted for cancer type, treatment modality, and clinical covariates, were used to assess organ-specific response patterns. Two key microenvironmental features--Vascular Perfusion and Leakiness Index (VaPLI) and tissue immune tolerance--were evaluated as predictors of lesion-level phenotypes. ResultsTreatment response dynamics varied significantly across metastatic sites, even within the same cancer type and therapy. Lesion-level responses reflected a strong seed-soil interaction, influenced by treatment modality. Liver metastases showed high initial regression but rapid progression, while bone lesions, especially in prostate cancer, exhibited more stable responses. VaPLI and immune tolerance status were significant predictors of lesion behavior. ConclusionsOur findings support an expanded seed-soil framework that incorporates physiological and therapeutic context, enabling site-aware treatment strategies and refined patient selection for metastatic cancer therapies.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.