Can stromal FAP-α predict response to lenvatinib plus PD-1 inhibitor in HCC?

FAP-α separates responders from non-responders

The paper asks whether a protein measured in the tumor's supporting tissue can flag, before treatment, which hepatocellular carcinoma (HCC) patients will fail lenvatinib plus a PD-1 inhibitor. Its answer is that stromal FAP-α does this better than the immune cell markers the authors also tested. FAP-α marks cancer-associated fibroblasts, the stromal cells that build and remodel the matrix around tumor and immune cells.

The strongest single number comes from the largest validation set: 54 patients with stage IB-IIA HCC who received three cycles of neoadjuvant lenvatinib plus tislelizumab. Scoring the percentage of FAP-α-positive cells on whole-slide immunohistochemistry images separated responders from non-responders with 80% sensitivity and 71.4% specificity. Among patients above the optimal threshold, 85.7% did not respond.

Two conditions frame that number. Response was defined by mRECIST as complete or partial response versus stable or progressive disease, so the endpoint is radiological response, not survival. And the 54 patients were a distinct group from the 10-patient discovery cohort profiled spatially and the 22-patient immunofluorescence cohort, so the three sets test related but not identical questions.

Bulk tissue averages away the stromal signal

Prior biomarker work in this setting relied on bulk tissue profiling, which averages molecular signals across tumor cells, immune cells and stroma. The authors argue this masks region-specific interactions that determine whether a combination of an anti-angiogenic tyrosine kinase inhibitor and a PD-1 inhibitor works.

To get around that, the discovery cohort of 10 patients — five responders and five non-responders — was analyzed with GeoMx digital spatial profiling. Regions of interest were chosen on slides, then split by a mask segmentation strategy into epithelial cell-enriched areas (PanCK-positive) and immune cell-enriched areas (CD45-positive). After quality control, 93 areas of interest from these 10 patients entered downstream analysis, each quantified for 47 proteins.

This design is what makes a stromal marker visible. FAP-α elevation in non-responders was attributed mainly to the stromal compartment, a distinction that a bulk measurement of the whole biopsy could not have drawn.

Tumor and immune niches carry different proteins

The spatial data first had to show that the segmentation was meaningful. Epithelial-enriched areas were high in tumor-associated markers including PanCK, NY-ESO-1 and ARG1, while immune-enriched areas were high in CD45, CD3 and CD163, all at p < 0.001. Clustering separated the two region types across patients and within individual patients.

The paper also tested whether physical proximity tracks molecular similarity. Immune-enriched areas were the most internally correlated, followed by epithelial-enriched areas. Epithelial and immune areas taken from the same region of interest correlated more strongly than corresponding areas from different regions (p < 0.001), which the authors read as evidence that local microenvironmental crosstalk shapes molecular phenotypes.

That reading is the authors' interpretation of a correlation, not a demonstrated causal mechanism. What the data do establish is that the platform resolves two functionally distinct niches inside the same tumor, which is the precondition for calling FAP-α a stroma-specific signal.

FAP-α sits with M2 macrophages, not T cells

In immune-enriched regions, non-responders showed higher expression of proteins tied to M2 polarization of tumor-associated macrophages — ARG1, CD14, CD68 and CD163 — and lower CD45, CD3 and CD8. Single-sample gene set enrichment analysis agreed, showing M2 macrophage enrichment alongside reduced T cell abundance in non-responders.

FAP-α fit this pattern spatially. It correlated positively with CD163 and with the immune checkpoints PD-L1 and FOXP3, and negatively with CD8, a T cell marker. The authors propose that FAP-α-high fibroblasts and M2 macrophages cooperate to build a barrier that keeps effector T cells out.

One detail complicates a simple story: fibroblasts were more abundant in responders, yet fibroblast-associated proteins diverged. Fibronectin and SMA were high in responders, whereas FAP-α was high in non-responders. The authors interpret this as different fibroblast states rather than different fibroblast quantities, an explanation the protein data support but do not prove.

PD-1 on immune cells, not PD-L1 on tumor cells

The same spatial comparison produced a second, separate observation about who responds. Responders had substantially higher PD-1 expression in immune-enriched regions than non-responders, while PD-L1 in epithelial-enriched regions did not differ significantly between the groups.

This matters because PD-L1 on tumor cells is the more familiar biomarker. Here the discriminating signal was PD-1 on immune cells, which the authors read as evidence that pre-existing immune activation predicts benefit from the PD-1 inhibitor. Responders also showed higher STING, Tim-3 and VISTA in these regions.

These are region-level protein comparisons in a 10-patient discovery cohort, so they are hypothesis-generating rather than validated cutoffs. The paper's validated claim rests on FAP-α, which was carried forward into the two larger cohorts.

Immunofluorescence and IHC confirm the stromal signal

Validation cohort 1 used multiplex immunofluorescence on 22 patients treated with the same regimen, quantifying FAP-α fibroblasts, CD163+ M2 macrophages, CD8+ T cells and CD20+ B cells. FAP-α was significantly higher in non-responders, and the other markers did not differ substantially between groups. The elevation was again concentrated in the stromal compartment.

Validation cohort 2 moved to immunohistochemistry, a routine and cheaper stain, on 54 patients. Percentage, density and count of FAP-α-positive cells were all significantly higher in non-responders, and receiver operating characteristic analysis produced the 80% sensitivity and 71.4% specificity reported above.

The progression from spatial profiling to a standard stain is the paper's practical argument: the marker does not require the discovery platform to be measured. What the paper does not report is how the optimal threshold was chosen or how it would hold up if fixed in advance on a new cohort.

Public data link FAP-α to progression and survival

Beyond treatment response, the authors analyzed 254 HCC samples from TCGA with RNA-seq and clinical data. FAP-α mRNA rose with HCC progression and was associated with shorter overall survival.

The paper reports no significant correlation between FAP-α and disease-free survival, and offers an explanation: disease-free survival reflects early postoperative recurrence, which is strongly influenced by surgical thoroughness and adjuvant therapy, whereas FAP-α acts mainly through immunosuppression and stromal remodeling. That is the authors' reasoning, not a tested mediation.

These public-data results concern FAP-α mRNA in a different cohort from the protein-based treatment cohorts. They support a prognostic role but should not be read as extending the 80%/71.4% predictive performance beyond the 54 patients in which it was measured.

A stromal marker for stratification, with caveats

The decision-relevant comparison in this paper is between marker types, not between drugs. In the same patients, a stromal FAP-α measurement separated responders from non-responders, whereas CD163, CD8 and CD20 did not differ substantially in the immunofluorescence cohort, and PD-L1 in tumor regions did not differ in the discovery cohort. If the result replicates, the practical implication is that a routine FAP-α stain could flag likely non-responders before lenvatinib plus PD-1 inhibitor therapy, which is a different test from scoring PD-L1 on tumor cells.

Several limits sit close to that claim. The discovery cohort was small at 10 patients, and the authors note that HBV infection was significantly associated with poor response while other clinical variables were not, leaving clinical-context effects to be verified in larger samples. The DSP platform quantifies areas of interest of roughly 0.5–2 mm² and cannot resolve single cells, so the fine-scale interaction between FAP-α-positive fibroblasts and neighboring immune cells remains unresolved. The 47-protein panel lacked a comprehensive set of angiogenesis proteins, and the study had no longitudinal samples.

The priority the evidence supports is therefore specific: among the markers measured here, stromal FAP-α is the one worth testing prospectively as a pre-treatment stratification tool for this combination, with the 85.7% non-response rate above threshold as the number to challenge in an independent, pre-specified cohort.

Sources used in this answer