Circulating Proteins for Pretreatment Risk Stratification in Image-guided Liver Therapy

See also the article by Nardone et al.

Shakthi Kumaran Ramasamy, MD, is a clinical research associate in the Department of Interventional Radiology at the Baptist Health Miami Cardiac and Vascular Institute. He completed a postdoctoral fellowship at the Stanford School of Medicine. Dr Ramasamy specializes in interventional radiology research, with a focus on interventional oncology. He serves as a chair of the Society of Interventional Radiology Resident and Fellow Section Research and Innovation Committee 2026–2027 and is a member of the RadioGraphics 2026–2027 trainee editorial board.

Ripal T. Gandhi, MD, FSIR, FSVM, is a vascular interventional radiologist and interventional oncologist at Baptist Health Miami Cardiac and Vascular Institute and Herbert Wertheim Cancer Institute. Dr Gandhi specializes in vascular and interventional radiology, with a focus on interventional oncology and advanced, image-guided therapies for the treatment of cancer and complex vascular conditions. His most recent research interests include yttrium 90 radioembolization, treatment of locally advanced pancreatic cancer, and management of pulmonary embolism and vascular disease.

Thermal ablation and interstitial brachytherapy (iBT) are viable treatment options for patients with hepatocellular carcinoma (HCC) who cannot undergo surgery, achieving local tumor control with limited morbidity (1). Imaging criteria and clinical guidelines are used to determine whether an HCC lesion can be treated based on its size, location, and proximity to organs at risk, but they do not predict patient outcomes. If two morphologically indistinguishable lesions can have very different clinical courses, anatomic assessment alone cannot capture the biologic risk that drives these outcomes. These outcomes are influenced by systemic tumor-host biology, and blood-based markers are increasingly being investigated for their ability to identify and correlate with these processes (2–4).

The study by Nardone et al (5) published in Radiology: Imaging Cancer addresses this gap by asking a single question: Can one circulating signature predict outcomes across multiple tumor types? ESTIMATE is a prospective, nonrandomized controlled trial of liver-directed ablation (DRKS registration no. 00010587). The authors performed a proteomic substudy of ESTIMATE in 103 patients treated with iBT. The cohort included 67 patients with primary liver cancers (51 HCC, 16 intrahepatic cholangiocarcinoma) and 36 with liver metastases (23 colorectal, 13 breast). The median follow-up was 24 months, with a median progression-free survival of 12 months and overall survival of 30 months. At the follow-up cutoff, 52% of patients had died. Baseline ethylenediaminetetraacetic acid (ie, EDTA) plasma collected the day before treatment was assayed from 1 µL of sample using the 92-protein Olink Target 96 Immuno-Oncology panel (Olink Proteomics).

FASLG and CD83 levels were higher in the long overall survival group, whereas interleukin (IL)–8, IL-1α, IL-15, chemokine (C-C) motif ligand 20 (CCL20), CAIX, and IL-6 levels were higher in the short overall survival group. In patients with metastatic breast cancer, CCL20 and IL-6 expression was associated with poor prognosis. In patients with metastatic colorectal cancer, IL-6, IL-8, and CAIX expression were associated with poor prognosis, as were IL-15 expression in patients with intrahepatic cholangiocarcinoma and CCL20 expression in patients with HCC. A combination of IL-8, IL-15, and CCL20 was associated with worse clinical outcomes in the entire cohort, and these three proteins were combined into a prognostic model. After bootstrap correction for overfitting, the model showed moderate discrimination for overall survival but weak discrimination for progression-free survival.

The study differs from the existing literature, which has focused on HCC and a single marker class within one treatment setting: plasma angiogenic and inflammatory proteins after resection (3), circulating tumor cells after surgery or before transplant ablation (2,4), and cellular or cytokine immune profiling after transarterial chemoembolization (TACE) (6,7). The findings by Nardone et al make a strong argument for a shared host-response axis across tumor types treated with the same local therapy, and this signal did not depend on the presence of extrahepatic disease. Individual circulating proteins can carry causal effects shared across anatomically distinct cancers, many of which are immunoregulatory (8).

IL-8 has previously been linked to worse outcomes after HCC resection (3) and TACE (7) and has been strongly associated with aggressive disease characterized by angiogenesis, immune suppression, and treatment resistance. IL-15 is an immunostimulatory, natural killer cell–activating cytokine that acts as a mediator of tumor control rather than progression in preclinical models (9). The findings by Nardone et al suggest that elevated IL-15 levels reflect a perturbed, tumor-associated inflammatory milieu rather than serve as a driver of disease. Elevated CCL20 levels have been independently linked to HCC and shown to add diagnostic value to α-fetoprotein (10), consistent with the study’s finding that CCL20 was prognostic specifically in the HCC and breast cancer subgroups.

Nardone et al studied patients treated with iBT, so whether the same baseline proteins predict outcomes after thermal ablation remains untested. Different liver-directed modalities elicit distinct immune responses. In a longitudinal profiling study, TACE produced a stronger early proinflammatory response than iBT, with marked increases in CD8+ T cells and CTLA-4 expression, whereas brachytherapy induced a more gradual, monocyte-predominant response with little change in CTLA-4 (6). These differences could alter the prognostic significance of elevated IL-8, IL-15, and CCL20 levels. It also remains unknown whether a pretreatment inflammation-based blood test can be applied consistently across the full range of liver-directed therapies, including TACE, yttrium 90 radioembolization, stereotactic body radiation therapy, histotripsy, and various thermal and nonthermal ablation techniques, each of which engages different inflammatory pathways. Testing the panel across treatment types, rather than only after brachytherapy, would therefore be valuable.

The findings should be interpreted as prognostic associations rather than validated predictive biomarkers or evidence of causation. No single protein remained significant after correction for multiple testing, none remained independently significant after adjustment for tumor type and treatment response, and a conservative least absolute shrinkage and selection operator criterion selected no proteins. When evaluated for their ability to improve treatment decision-making, the panel showed no clear benefit over treating all patients alike. The study demonstrates that higher levels of these proteins are associated with poor prognosis but does not show that the test can identify which patients specifically benefit from iBT. Because the study lacked a randomized comparison group, the data cannot distinguish patients with poor prognosis because of their underlying disease biology from those with poor prognosis despite, or even because of, the treatment received. Elevated IL-8, IL-15, and CCL20 levels more likely reflect an aggressive inflammatory state of the tumor and host rather than cause it, and the possibility that an unmeasured confounder explains both the elevated marker levels and the poor outcomes cannot be excluded.

The study provides an important starting point and underscores the need for external testing in independent cohorts. Using prespecified thresholds and testing the markers in laboratory models will help determine whether they are causes or consequences of aggressive disease. Another practical question is whether the full 92-protein platform is necessary or whether a focused assay measuring only the three-marker signature would suffice. Although the broad panel was essential for biomarker discovery, clinical implementation would likely rely on a standardized, low-cost assay measuring only these three proteins. As noted by Nardone et al, prospective confirmation will be needed to demonstrate that such a compact assay reproduces the performance of the full panel.

The panel consists entirely of immune- and inflammation-related proteins. Standard blood tumor markers such as α-fetoprotein, carbohydrate antigen 19-9, and carcinoembryonic antigen, which are part of routine cancer care, were not included despite their established role in predicting outcomes and monitoring disease in these patients. Adding these familiar markers to the three-protein signature, while expanding the search beyond the 92 proteins on the Olink platform, could yield a more comprehensive biomarker panel. The same markers may also prove valuable after treatment. Repeating the inflammatory blood test over time could help monitor treatment response and detect recurrence earlier, rather than only stratifying patients by pretreatment risk. As discussed by the authors, integrating proteomic, radiomic, functional imaging, and genomic data using machine learning approaches could ultimately provide a more comprehensive characterization of primary and metastatic liver malignancies.

If validated, the clinical value of this three-protein panel would be straightforward. A simple pretreatment blood test could provide an early indication of prognosis after liver ablation, complementing imaging assessment of whether a lesion is technically amenable to treatment. Such information could help identify which patients are most likely to benefit from local therapy. A patient with a high-risk blood marker profile raises a clinical question: not only how best to treat the lesion, but whether local ablation alone is sufficient or whether surgery, alternative local-regional therapies, and/or systemic therapy would provide greater benefit. Conversely, a patient whose tumor appears ideal for ablation but whose blood markers indicate aggressive disease may require a more comprehensive treatment strategy.

Disclosures of conflicts of interest

S.K.R. is a member of the RadioGraphics trainee editorial board. Please see ICMJE form(s) for other author conflicts of interest. These have been provided as supplemental materials.