Circulating tumor DNA and Response Evaluation Criteria In Solid Tumors: ctDNA-RECIST proof-of-concept in HER2-positive metastatic breast cancer

TL;DR

This study explores how combining tumor DNA measurements with traditional imaging methods can better track cancer treatment responses in breast cancer patients.

Contribution

The study introduces a new algorithm (cEoT) that improves treatment response prediction using ctDNA data in HER2-positive metastatic breast cancer.

Findings

ctDNA-RECIST (cRECIST) showed deeper responses than traditional RECIST in 27 patients with progressive disease.

A personalized cEoT algorithm significantly improved prediction of progression-free survival compared to ctDNA alone.

ctDNA waving patterns suggest complex tumor dynamics that require advanced analytical approaches.

Abstract

Response Evaluation Criteria In Solid Tumors (RECIST 1.1) and circulating tumor DNA (ctDNA) recapitulate and anticipate response to treatment, respectively. However, ctDNA-RECIST (cRECIST) and ctDNA-guided End of Treatment (cEoT) are not applied routinely.

To provide proof-of-concept for RECIST1.1/cRECIST integration, HER2-positive metastatic breast cancer patients (n = 50) were enrolled in the multi-center prospective GIM21 study to receive Trastuzumab-emtansine (T-DM1). CT scans (113 tumor lesions) were longitudinally assessed for classical Objective Responses (ORs: progressive disease/stable disease/partial response/complete response; PD/SD/PR/CR) applying default RECIST 1.1 cut-offs (SD/PD ≥ 20%; SD/PR ≤ 30%). Likewise, bespoke NGS/dPCR (78 genomic alterations; 466 time points) were converted into ctDNA-Objective Responses (cORs: cPD/cSD/cPR/cCR) exploring wide cPD/cSD/cCR cut-off ranges, both default (RECIST 1.1-like) and alternative.

Whichever the cut-off, cORs were much deeper than ORs, leading to RECIST 1.1/cRECIST divergence in 27 cPD-positive patients. Moreover, due to complex ctDNA trajectories (multiple successive ctDNA increases/decreases, termed ctDNA waving), cPD (the earliest ctDNA increase) correlated with outcome in broad patient subsets but not individual patients. To deconvolute ctDNA waving, cPD was combined with three-point ctDNA Trends (Tr), resulting in a personalized cEoT clinical algorithm that, once retrofitted to the 27 cPD-positive patient dataset, aligned with PFS much better than cPD (cEoT/PFS vs cPD/PFS linear regression: R2 = 0.85 vs 0.35).

Even in difficult ctDNA scenarios, the cEoT algorithm may help to: (a) predict treatment efficacy during drug development, (b) adaptively randomize for patient-specific, timely treatment switch in clinical trials, and (c) prevent premature treatment withdrawal in long-responders. Future randomized studies are warranted for cRECIST/RECIST 1.1 integration/personalization in different tumors/settings.

NCT05735392.

The online version contains supplementary material available at 10.1186/s13046-025-03605-2.

Figures