Preclinical identification and clinical validation of dynamic biomarkers for trifluridine/tipiracil with or without bevacizumab in metastatic colorectal cancer

Abstract

Background

Biomarkers predicting response to trifluridine/tipiracil (FTD/TPI) with or without bevacizumab (BEV) in refractory metastatic colorectal cancer (mCRC) remain unclear. We identified candidate biomarkers in preclinical models and evaluated their clinical relevance.

Materials and methods

Candidate biomarkers were identified through preclinical transcriptomic analyses. Serum interleukin (IL)-8, epiregulin (EREG), amphiregulin (AREG), interferon-inducible protein 16 (IFI16), and vascular endothelial growth factor A (VEGF-A) were measured in two cohorts treated with FTD/TPI+BEV or FTD/TPI alone. Samples were collected at baseline, before cycle 2, and at progression. Biomarker changes were categorized as increased or decreased. Propensity score matching was carried out, and survival outcomes were compared according to biomarker dynamics.

Results

Among 112 patients, 34 matched pairs were analyzed after propensity score matching. FTD/TPI+BEV yielded longer overall survival (OS) than FTD/TPI alone [10.5 versus 6.2 months; hazard ratio (HR) 0.53]. Increased IL-8 at cycle 2 and elevated IFI16 at progression were associated with poorer prognosis, whereas decreased AREG at cycle 2 predicted improved survival. In patients with increased IL-8, OS was longer with FTD/TPI+BEV than with FTD/TPI alone (12.8 versus 5.8 months; HR 0.43). Early EREG decrease was associated with better OS only in the FTD/TPI+BEV group. No significant interaction was observed between treatment and biomarker dynamics.

Conclusion

On-treatment changes in IL-8, AREG, and EREG may serve as prognostic—and potentially predictive—biomarkers of FTD/TPI efficacy in mCRC. These findings support further evaluation of biomarker-guided treatment strategies.

Highlights

• •FTD/TPI+BEV produced a clinically meaningful survival advantage over FTD/TPI alone.
• •Early IL-8 elevation and IFI16 upregulation emerged as strong indicators of poor prognosis.
• •Early AREG reduction consistently identified patients with improved survival outcomes.
• •EREG decline uniquely signaled greater benefit from FTD/TPI+BEV therapy.
• •Dynamic biomarker profiling shows clear potential for guiding personalized mCRC treatment.

Introduction

Trifluridine/tipiracil (FTD/TPI) is an oral anticancer agent used in patients with refractory metastatic colorectal cancer (mCRC).^1,2 We previously identified candidate biomarkers for FTD/TPI efficacy through a combination of preclinical and translational research approaches.³ Recently, the addition of bevacizumab (BEV) to FTD/TPI has been shown to improve overall survival (OS) in this patient population, as demonstrated in randomized clinical trials.⁴ Although several potential biomarkers for FTD/TPI efficacy have been reported, including our own previous pharmacogenetic studies.^5,6 none have yet been adopted in routine clinical practice. Moreover, biomarker studies involving the FTD/TPI plus BEV are scarce, and the mechanisms contributing to its clinical benefit remain to be clarified.

In this study, we aimed to evaluate the clinical relevance of previously identified biomarkers in patients with refractory mCRC treated with either FTD/TPI plus BEV or FTD/TPI alone. To achieve this, we conducted in silico and in vitro analyses to identify candidate markers, followed by a translational validation study using clinical samples.

This integrated approach, combining bioinformatic analysis with experimental and clinical validation, allows for a more robust and mechanistically grounded evaluation of biomarkers than conventional retrospective analyses. Moreover, such a strategy may help uncover previously unrecognized therapeutic mechanisms and facilitate the development of novel treatment strategies.

Materials and methods

Study design and patients

This study consists of two steps (Supplementary Figure S1, available at https://doi.org/10.1016/j.esmogo.2026.100323). In the preclinical phase (STEP I), candidate biomarkers were selected through transcriptomic analysis. In the clinical phase (STEP II), we evaluated the predictive value of these biomarkers in patients with mCRC who received FTD/TPI+BEV (cohort A) or FTD/TPI alone (cohort B). To address baseline imbalances between the two cohorts, propensity score matching (PSM) was carried out. Propensity scores were estimated using logistic regression including sex, age, Eastern Cooperative Oncology Group performance status, primary tumor location, and presence of liver metastasis as covariates. Patients were matched 1 : 1 using nearest-neighbor matching without replacement (caliper = 0.2). Blood samples were prospectively collected at baseline (BL), before the second cycle (2nd), and at progressive disease (PD) as part of routine clinical practice and institutional translational research protocol. The present biomarker analysis was conducted retrospectively using these samples.

STEP I: Preclinical investigation

Gene expression and database analysis

Human colon cancer HT-29 cells were obtained and maintained as described previously.⁷ Cells were either left untreated or treated with 10 μM of FTD for 48 hours (2 days) and 72 hours (3 days). Total RNA was extracted using RNeasy mini kit (Qiagen, Hilden, Germany), and cDNA microarray analysis (Affymetrix HG U133 plus2.0, Santa Clara, CA) was conducted as described previously⁷ to identify genes that were upregulated or down-regulated by more than twofold after treatment with FTD (10 μM, 72 hours). We next focused on secreted factors that exhibit low baseline expression in normal human serum according to the GeneCards database (https://www.genecards.org/) and high expression in colon cancer tissues based on the Oncomine database (https://www.oncomine.org/) as candidate biomarkers for subsequent clinical validation study. Through this process, we identified interleukin (IL)-8, vascular endothelial growth factor A (VEGF-A), epiregulin (EREG), and amphiregulin (AREG) as candidate biomarkers for FTD/TPI efficacy.³ These markers were not predictive in the regorafenib control cohort. In the present study, we additionally included AREG as a new candidate marker.

Establishment of FTD-resistant cell lines and identification of drug resistance-related genes

We established an FTD-resistant HT-29 cell line (HT-29/FTD-R) by continuously treating HT-29 cells with 0.5 μM FTD for 90 days. The sensitivity of HT-29 and HT-29/FTD-R cells to FTD was assessed using the thiazolyl blue tetrazolium bromide assay (Sigma, St Louis, MO, USA), as described previously.⁸ Total RNA was extracted from HT-29 and HT-29/FTD-R cells, and cDNA microarray analysis was carried out as described above.

STEP II: Clinical validation

Sample collection and biomarker measurement

Peripheral blood samples were collected from 112 patients with mCRC (cohort A, n = 34; cohort B, n = 78). After PSM, 34 matched pairs were included in the analysis. Serum was isolated and stored at −80°C until analysis. Levels of IL-8, VEGF-A, EREG, AREG, and IFI16 were measured using ELISA kits (R&D Systems, Minneapolis, MN) per the manufacturer’s protocols. Changes in biomarker levels were classified as either ‘increased’ or ‘decreased’ compared with baseline.

Treatment and evaluation

All patients had received prior standard chemotherapy including 5-fluorouracil, oxaliplatin, irinotecan, BEV, and anti-epidermal growth factor receptor (EGFR) antibodies if their tumor was RAS wild-type. Tumor response was assessed per RECIST v1.1.

FTD/TPI (Taiho Pharmaceutical Co., Tokyo, Japan) was administered at 35 mg/m² twice daily on days 1-5 and 8-12 in a 28-day cycle, with or without BEV (5 mg/kg) on days 1 and 15. Dose modifications were made according to toxicities following standard protocols. Treatment continued until disease progression or unacceptable toxicity. Adverse events were graded according to the Common Terminology Criteria for Adverse Events, version 4.0. The study was approved by the Institutional Review Board of the Cancer Institute Hospital and conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all patients.

Statistical analysis

The primary endpoint was OS; secondary endpoints included progression-free survival (PFS), and disease control rate (DCR). OS was measured from the start of treatment to death from any cause. PFS was defined as the time from treatment initiation to progression or death. Patients without events were censored at the date of last follow-up. The Kaplan–Meier method and log-rank test were used to compare survival curves. Associations between biomarker dynamics and outcomes were evaluated using Fisher’s exact test and Cox proportional hazards models. All statistical analyses were carried out using SPSS software (version 29.0; IBM Corp., Armonk, NY, USA). A P value of < 0.05 was considered statistically significant.

Results

Preclinical identification of candidate biomarkers

According to recent analyses, drug resistance develops in two distinct phases: an early induced-resistance phase characterized by a plastic phenotype, and a later advanced stable-resistance phase that often arises from genetic alterations.⁹ To identify candidate biomarkers associated with FTD resistance, we searched for (i) the genes induced by FTD treatment and (ii) the genes overexpressed in stable FTD-resistant cells using preclinical model of human colon cancer HT-29 cells (Supplementary Figure S2A, available at https://doi.org/10.1016/j.esmogo.2026.100323). When HT-29 cells were exposed to 0.3∼30 μM of FTD, cell proliferation was suppressed at 2, 3, and 5 days after the drug exposure (Supplementary Figure S2B, available at https://doi.org/10.1016/j.esmogo.2026.100323). Based on these findings, cells were treated with FTD for 48 or 72 hours, and transcriptome analyses were carried out (Figure 1A).

Figure 1.

Candidate molecular changes related to early response to FTD and to stable FTD resistance revealed by preclinical analyses with colorectal cancer (CRC) cells. (A) Preclinical study design to identify genes associated with the early-phase response to FTD in human CRC cells. (B) Heatmap showing gene expression changes of representative soluble factors that are selectively in CRCs and markedly altered following FTD treatment. Circulating DNA microarray analysis of HT-29 cells demonstrated substantial changes in the RNA expression of IL-8, VEGF-A, EREG, and AREG after FTD exposure. (C) Stable FTD resistance in HT-29/FTD-R cells compared with parental HT-29 cells. (D) Preclinical study design to identify genes associated with stable FTD resistance in human CRC cells. (E) Overexpression of soluble factors in FTD-resistant HT-29/FTD-R cells. AREG, amphiregulin; EREG, epiregulin; FTD, trifluridine; FTD-R, trifluridine resistant; IL-8, interleukin 8; VEGF-A, vascular endothelial growth factor A.

We particularly focused on soluble factors whose expressions changed by more than twofold after FTD treatment in HT-29 cells and that are known to be highly expressed in colorectal cancers but low in normal human serum, according to the GeneCards and Oncomine databases. This approach identified four cytokines (VEGF-A, EREG, AREG, and IL-8) as candidate markers for FTD response (Figure 1B).

In parallel, we established stable FTD-resistant HT-29/FTD-R cells by continuously exposing HT-29 cells to a sublethal concentration of FTD (Figure 1C). We then conducted transcriptome profiling to extract candidate soluble factors that are markedly upregulated in HT-29/FTD-R cells (Figure 1D). This analysis revealed that several interferon-responsive genes were upregulated in the resistant cells. Among them, we focused on interferon gamma inducible protein 16 (IFI16), whose expression was elevated by >10-fold compared with parental HT-29 cells (Figure 1D), for further clinical validation.

Patient characteristics

A total of 112 patients were included in this study (Supplementary Table S1, available at https://doi.org/10.1016/j.esmogo.2026.100323): 34 in cohort A (FTD/TPI+BEV) and 78 in cohort B (FTD/TPI alone). To minimize baseline imbalances, PSM was carried out, resulting in 34 matched pairs. After PSM, baseline characteristics were generally well balanced, except for differences in adjuvant therapy history and the number of prior chemotherapy regimens (Table 1).

Table 1.

Baseline patients and tumor characteristics after PSM

Cohort	Cohort A (FTD/TPI + BEV, N = 34)		Cohort B (FTD/TPI, N = 34)		P valuea
	N	%	N	%
Sex					0.81
Male	15	44.1	14	41.2
Female	19	55.9	20	58.8
Age, years					0.19
Median (range)	55 (30-77)		61 (34-77)
≤65	23	67.6	20	58.8
>65	11	32.4	14	41.2
ECOG performance status					0.09
ECOG 0	29	85.3	23	67.6
ECOG 1	5	14.7	11	32.4
Primary tumor site					0.25
Right	6	17.6	10	29.4
Left	28	82.4	24	70.6
Liver metastasis					0.36
Yes	29	85.3	26	76.5
No	5	14.7	8	23.5
Lung metastasis					0.80
Yes	22	64.7	23	67.6
No	12	35.3	11	32.4
Lymph node metastasis					0.33
Yes	14	41.2	18	52.9
No	20	58.8	16	47.1
Number of metastases					1.0
<2	6	17.6	6	17.6
≥2	28	82.4	28	82.4
Primary tumor resected					0.23
Yes	25	73.5	29	85.3
No	9	26.5	5	14.7
Adjuvant history					0.005
Yes	6	17.6	17	50
No	28	82.4	17	50
No. of prior chemo					<0.001
≤3	31	91.2	16	47.1
>3	3	8.8	18	52.9
Prior anti-VEGF					1.0
Yes	34	100	33	97.1
No	0	0	1	2.9
Prior anti-EGFR					0.015
Yes	11	32.4	21	61.8
No	23	67.6	13	38.2
KRAS status					0.23
Wild-type	15	44.1	20	58.8
Mutant	19	55.9	14	41.2

BEV, bevacizumab; ECOG, Eastern Cooperative Oncology Group; EGFR, endothelial growth factor receptor; FTD/TPI, trifluridine/tipiracil; PSM, propensity score matching; VEGF, vascular endothelial growth factor.

^aP value was based on chi-square test, or the Mann–Whitney test when appropriate. Statistically significant P values (<0.05) are shown in bold.

Treatment efficacy and survival outcomes

After PSM, cohort A exhibited significantly longer OS than cohort B [10.5 versus 6.2 months; hazard ratio (HR) 0.53, 95% confidence interval (CI) 0.30-0.93, P = 0.026]. DCR was also higher in cohort A than in cohort B (58.8% versus 34.4%, P = 0.047). PFS was numerically longer in cohort A (3.1 versus 2.0 months; HR 0.77, 95% CI 0.47-1.27, P = 0.297), though not statistically significant.

Biomarker dynamics and survival

Serum levels of IL-8, AREG, EREG, VEGF-A, and IFI16 were measured at three time points: BL, 2nd, and PD. For IFI16, measurements were carried out at BL and PD only, as this marker was specifically evaluated as a resistance-associated biomarker. Although our primary analyses focused on on-treatment biomarker dynamics, we also explored associations between baseline biomarker levels and disease control (Supplementary Table S2, available at https://doi.org/10.1016/j.esmogo.2026.100323). At baseline, IL-8, VEGF, and IFI16 levels were not significantly correlated with disease control, whereas lower AREG and EREG levels showed a trend toward improved response in the overall cohort. These associations were less attenuated when analyzed separately within the FTD/TPI and FTD/TPI+BEV groups.

Changes in biomarker levels were categorized as either increased or decreased from BL, and their association with survival outcomes was evaluated (Table 2). Increased IL-8 at 2nd was associated with shorter PFS and showed a trend toward shorter OS in cohort B and the entire population (Figure 2A). Decreased AREG at 2nd was correlated with longer PFS and OS, with a marked benefit in OS particularly in cohort A and the entire population (Figure 2B). Furthermore, increased IFI16 at PD was significantly associated with shorter OS across both cohorts and the overall population, consistent with preclinical findings (Figure 2C).

Table 2.

Biomarker dynamics and survival

BM	Changes (BL-2nd)	All patients (N = 68)						Cohort A (FTD/TPI + BEV, N = 34)						Cohort B (FTD/TPI, N = 34)
		DC (%)	P valuea	PFSc	P valueb	OSc	P valueb	DC (%)	P valuea	PFS	P valueb	OS	P valueb	DC (%)	P valuea	PFS	P valueb	OS	P valueb
IL-8	Inc	15/46	<0.001	2.0	0.002	8.0	0.11	9/20	0.011	2.2	0.051	12.8	0.90	6/26	0.011	2.0	0.027	5.8	0.12
	Dec	16/18		4.2		10.1		11/12		4.1		10.1		5/6		4.8		9.7
AREG	Inc	12/30	0.21	2.0	0.014	5.8	<0.001	6/10	1.0	3.1	0.29	8.0	<0.001	6/10	0.70	2.0	0.084	5.6	0.24
	Dec	19/34		3.4		12.3		14/22		3.5		14.2		6/20		3.0		9.7
EREG	Inc	16/30	0.46	3.0	0.34	8.2	0.85	8/16	0.14	2.3	0.91	8.1	0.28	8/14	0.027	3.6	0.096	8.2	0.29
	Dec	15/34		3.0		9.7		12/16		3.8		12.8		3/18		2.0		6.2
VEGF	Inc	14/30	0.79	2.3	0.37	9.8	0.51	12/20	1.0	3.0	0.19	12.0	0.97	2/10	0.43	2.0	0.78	8.5	0.98
	Dec	17/34		3.3		8.2		8/12		3.5		10.5		9/22		2.3		6.2
	Changes (BL-PD)
IFI-16	Inc	16/37	0.75	2.3	0.25	7.4	0.004	8/15	0.70	3.1	0.31	9.1	0.034	8/22	0.68	2.0	0.99	5.8	0.12
	Dec	10/21		3.4		12.8		8/12		3.5		14.2		2/9		3.0		9.8

2nd, before second cycle; AREG, amphiregulin; BL, baseline; BM, biomarker; Dec, decreased; EREG, epiregulin; IFI16, interferon-inducible protein 16; IL-8, interleukin 8; Inc, increased; PD, progressive disease; OS, overall survival; PFS, progression-free survival; VEGF-A, vascular endothelial growth factor A.

^aP values were calculated using Fisher’s exact test for the percentage of patients showing changes from baseline to the second cycle (BL-2nd) for IL-8, AREG, EREG, and VEGF-A, whereas changes from baseline to progressive disease (BL-PD) were evaluated for IFI16, as IFI16 was measured at BL and PD only.

^bP value was based on log-rank test for PFS and OS. Statistically significant P values (<0.05) are shown in bold.

^cPFS and OS are reported as months.

Figure 2.

Overall survival stratified by protein changes (decreased versus increased). (A) OS according to IL-8 changes at 2nd in all patients, (B) OS according to AREG changes at 2nd in all patients, and (C) OS according to IFI16 change at PD in all patients. 2nd, before second cycle; AREG, amphiregulin; CI, confidence interval; HR, hazard ratio; IFI16, interferon-inducible protein 16; IL-8, interleukin 8; OS, overall survival; PD, progressive disease.

In addition to survival outcomes, biomarker dynamics were also associated with DCR. Patients with decreased IL-8 consistently showed higher DCR compared with those with increased IL-8, both in the FTD/TPI monotherapy and FTD/TPI+BEV cohorts. Similarly, in the FTD/TPI-monotherapy cohort, patients with early EREG increase achieved higher DCR, whereas this association was not observed with the FTD/TPI+BEV cohort. By contrast, AREG, VEGF, and IFI16 dynamics did not demonstrate a significant impact on DCR. These findings suggest that IL-8 and EREG dynamics may serve not only as prognostic markers of OS but also as early indicators of disease control, further supporting their clinical utility.

In multivariable analysis, both decreased AREG and grade ≥3 neutropenia emerged as significant predictors of improved outcomes. Specifically, decreased AREG was associated with better PFS (HR 0.50, 95% CI 0.29-0.87) and OS (HR 0.29, 95% CI 0.15-0.56), whereas grade ≥3 neutropenia similarly predicted favorable PFS (HR 0.53, 95% CI 0.30-0.93) and OS (HR 0.31, 95% CI 0.16-0.60). Treatment regimen itself was not an independent prognostic factor for OS (Supplementary Table S3, available at https://doi.org/10.1016/j.esmogo.2026.100323). Prior anti-EGFR exposure was not associated with OS in univariable analysis (P = 0.678), and additional adjustment for prior anti-EGFR exposure did not materially alter the multivariable results.

Subgroup analyses showed that in patients with increased IL-8 at 2nd, FTD/TPI+BEV conferred a significantly longer OS compared with FTD/TPI alone (12.8 versus 5.8 months; HR 0.43, 95% CI 0.21-0.89) (Figure 3A). Similar trends were observed in patients with decreased AREG (14.2 versus 9.7 months; HR 0.40, 95% CI 0.17-0.95) or EREG (12.8 versus 6.2 months; HR 0.27, 95% CI 0.11-0.65) at 2nd (Figure 3B and C). However, no significant interaction was observed between treatment regimen and biomarker dynamics, suggesting these biomarkers may serve as prognostic rather than predictive indicators.

Figure 3.

Impact of FTD/TPI+BEV versus FTD/TPI on OS based on protein changes. (A) OS in patients with increased (left) and decreased (right) IL-8, (B) OS in patients with decreased (left) and increased (right) AREG, and (C) OS in patients with decreased (left) and increased (right) EREG. P values for interaction between treatment regimen and biomarker dynamics are shown in each panel. AREG, amphiregulin; BEV, bevacizumab; EREG, epiregulin; FTD/TPI, trifluridine/tipiracil; IL-8, interleukin 8; OS, overall survival.

Discussion

This study evaluated the clinical significance of serum biomarkers initially identified in our preclinical and translational research presented at ASCO-GI 2018, which focused on patients treated with FTD/TPI monotherapy without BEV.³ In that exploratory analysis, these biomarkers did not demonstrate predictive value for regorafenib treatment. In the present study, we expanded our investigation to patients receiving FTD/TPI with or without BEV, enabling assessment of biomarker utility across different treatment regimens.

Our results demonstrate that dynamic changes in IL-8, AREG, and IFI16 during treatment are associated with survival outcomes largely independent of BEV administration. Decreased AREG at cycle 2 and increased IFI16 at disease progression were consistently associated with OS across both cohorts. In addition, grade ≥3 neutropenia was associated with improved prognosis regardless of treatment regimen, consistent with previous observations in fluoropyrimidine-based therapies.¹⁰

The biological plausibility of these findings is supported by the known mechanisms of FTD/TPI. FTD is a nucleoside analog that induces replication stress and DNA damage through incorporation into DNA, thereby activating DNA damage response (DDR) and inflammatory signaling pathways.^11,12 IL-8, a pro-inflammatory chemokine implicated in tumor progression and treatment resistance,13, 14, 15, 16, 17 can be induced through at least two upstream mechanisms during FTD/TPI therapy: (i) DDR-mediated innate immune activation involving DNA sensors such as IFI16, and (ii) EGFR signaling, which transcriptionally upregulates IL-8, whereas EGFR inhibition suppresses IL-8 and promotes a more favorable immune milieu.^18,19 Consistent with this, we previously demonstrated that circulating IL-8 serves as a prognostic biomarker for anti-angiogenic therapy efficacy irrespective of treatment line in mCRC.²⁰

In the current cohort, an early on-treatment decline in IL-8 was associated with favorable outcomes, consistent with effective tumor control and attenuation of inflammatory–angiogenic signaling.^21,22 BEV-mediated vascular normalization may further reduce hypoxia-driven IL-8 induction, reinforcing IL-8 decline in responders. Conversely, on-treatment IL-8 elevation likely reflects adaptive resistance. Thus, IL-8 dynamics may function as an on-treatment pharmacodynamic marker integrating DDR- and EGFR-derived signals, aligning with our proposed mechanistic framework (Figure 4).

Figure 4.

Proposed mechanism of action and the role of on-treatment biomarkers in patients receiving FTD/TPI ± BEV. Trifluridine (FTD), the active component of FTD/TPI, is incorporated into tumor-cell DNA, causing replication stress and double-strand breaks that trigger DNA damage response (DDR) pathways. DDR activation promotes innate immune signaling, partly mediated through interferon-inducible protein 16 (IFI16), leading to inflammatory cytokine production, including interleukin (IL)-8. In parallel, epidermal growth factor receptor (EGFR) signaling is activated via its ligands amphiregulin (AREG) and epiregulin (EREG), contributing to both IL-8 transcription and DNA repair through homologous recombination and nucleotide excision repair pathways. Early upregulation of AREG and EREG may represent an adaptive tumor response that enhances EGFR-mediated DNA repair and treatment resistance. Conversely, tumors failing to up-regulate AREG, and to a lesser extent EREG, may be more vulnerable to FTD-induced cytotoxicity. IL-8 levels reflect combined inputs from DDR-driven innate immune activation and EGFR signaling, serving as a dynamic pharmacodynamic marker of tumor adaptation. Bevacizumab (BEV) may further modify these pathways by normalizing tumor vasculature, reducing hypoxia-induced IL-8 production, and attenuating vascular endothelial growth factor-driven compensatory angiogenesis, particularly in patients with early EREG decrease. HR, homologous recombination; MMP, matrix metalloproteinase; NER, nuclear excision repair.

Decreased circulating AREG at cycle 2 was consistently associated with improved OS and remained an independent prognostic factor in multivariable analysis, regardless of BEV use. AREG is a high-affinity EGFR ligand that induces sustained EGFR activation, promoting tumor proliferation and therapeutic resistance.23, 24, 25 Suppression of AREG may therefore reflect impaired EGFR-driven survival signaling, providing a strong biological rationale for its prognostic value independent of anti-angiogenic therapy.

Our prior preclinical studies and the ASCO-GI 2018 FTD/TPI-monotherapy cohort suggested that early EREG induction during FTD/TPI treatment was associated with therapeutic benefit, possibly reflecting tumor sensitivity to FTD-induced stress.³ In the present study, a similar but non-significant trend was observed in the FTD/TPI-monotherapy cohort. Notably, among patients with early EREG decrease, OS was substantially improved when BEV was added compared with FTD/TPI alone. This finding suggests that BEV may compensate for the unfavorable prognostic implications of EREG reduction, potentially through attenuation of VEGF-driven escape mechanisms. Although formal interaction testing did not reach statistical significance, these data suggest a regimen-dependent effect of EREG dynamics that warrants further investigation.

AREG and EREG are EGFR ligands produced by epithelial cells, tumor cells, and tumor-associated immune cells, and their circulating levels may reflect composite signals derived from tumor burden and microenvironmental activation.^26,27 Whereas AREG induces sustained EGFR signaling, EREG triggers a more transient and regulated response.²⁸ The combination of AREG suppression and EREG induction may therefore represent a molecular signature of FTD/TPI sensitivity. Conversely, tumors lacking EREG induction may shift toward VEGF-driven angiogenesis, providing a biological explanation for the preferential benefit of BEV in this subset. Such effects may be mediated through EGFR–VEGF pathway crosstalk and adaptive vascular remodeling, whereby BEV suppresses compensatory angiogenic signaling.29, 30, 31, 32

These findings support a mechanistic link between FTD-induced DNA damage and EGFR signaling via AREG/EREG modulation (Figure 4). Incorporation of FTD induces replication stress and double-strand breaks, activating DDR pathways, whereas EGFR signaling has been shown to support DNA repair through homologous recombination and nucleotide excision repair.33, 34, 35 Thus, early upregulation of AREG and EREG may reflect an adaptive response that enhances DNA repair and promotes treatment resistance, whereas tumors unable to activate these ligands—particularly AREG—may be more vulnerable to FTD-induced cytotoxicity.

IFI16, a DNA sensor and interferon-inducible gene, was markedly upregulated in FTD-resistant cells in our preclinical model and was clinically associated with poorer survival when elevated at disease progression. IFI16 mediates innate immune sensing of cytoplasmic DNA and activation of inflammasome and type I interferon pathways, which may contribute to immune modulation and treatment resistance.36, 37, 38, 39 These findings suggest that IFI16 elevation at progression reflects acquired resistance rather than early treatment response.

VEGF, a central driver of tumor angiogenesis, is the direct target of BEV, which was co-administered in cohort A. Clinically, the addition of BEV to FTD/TPI prolonged OS compared with FTD/TPI monotherapy. However, in our biomarker analyses, circulating VEGF levels themselves did not significantly correlate with survival. This apparent discrepancy may be explained by the effective neutralization of VEGF activity by BEV, thereby limiting the ability of serum VEGF levels to reflect treatment impact. Furthermore, complex crosstalk between VEGF signaling and EGFR ligands may influence the tumor microenvironment and therapeutic response, which may be better captured by downstream biomarkers such as IL-8, AREG, and EREG.^21,22,29,30 Accordingly, dynamic changes in these biomarkers may better reflect the biological impact of treatment than VEGF levels alone.

Although BEV improved OS and disease control numerically, treatment regimen was not independently prognostic in multivariable analysis. Subgroup analyses suggested that patients with increased IL-8 or decreased AREG derived greater benefit from FTD/TPI plus BEV, and EREG dynamics may identify another subgroup enriched for BEV responsiveness. The interaction test for AREG dynamics showed a borderline signal (P for interaction = 0.09) but did not reach statistical significance, reinforcing the primarily prognostic nature of these biomarkers. In the matched cohort, prior anti-EGFR exposure remained more frequent in the FTD/TPI-monotherapy group; however, it was not associated with OS and did not materially alter the multivariable results, supporting the robustness of our OS findings. Regorafenib data were not included in the present analysis, as AREG and IFI16 levels were not measured in that cohort and prior results demonstrated that IL-8, EREG, and VEGF lacked predictive value for regorafenib therapy.³ Accordingly, the current analysis focused on FTD/TPI with or without BEV.

Clinically, these findings support monitoring dynamic changes in serum IL-8, AREG, and IFI16 as on-treatment biomarkers in refractory mCRC. Although these biomarkers predominantly act as prognostic indicators, EREG dynamics may also provide insight into patient subsets that preferentially benefit from BEV-containing regimens. Given the feasibility of ELISA-based measurement, integration of these biomarkers into clinical practice may enable biomarker-guided treatment strategies.

The PFS observed in our cohort was shorter than that reported in the phase III SUNLIGHT trial.⁴ This difference likely reflects variations in patient population and study design, as our study represents a single-center real-world cohort including a higher proportion of heavily pretreated patients. In addition, differences in radiologic assessment intervals and treatment practices in routine clinical settings may attenuate measurable PFS differences. Importantly, despite the modest PFS separation, the consistent OS benefit and improved DCR observed in our study support the clinical activity of FTD/TPI plus BEV in real-world practice.

Several limitations should be acknowledged. This was a single-center study with a limited sample size, particularly after PSM, and residual confounding cannot be fully excluded. In addition, the relatively modest PFS difference compared with randomized clinical trials may reflect real-world patient heterogeneity and assessment variability. Furthermore, comprehensive molecular data including NRAS, BRAF, and MMR status were not uniformly available, as extended molecular testing was not routinely carried out during the study period. Although supported by preclinical data, in vivo functional validation remains necessary. Larger prospective studies are required to confirm these findings and clarify the mechanistic basis of biomarker dynamics.

Conclusion

Dynamic changes in serum IL-8, AREG, and IFI16 are significantly associated with treatment outcomes in patients with mCRC receiving FTD/TPI with or without BEV. These biomarkers primarily demonstrate prognostic value, as no statistically significant interaction with BEV use was observed. They reflect key biological processes involving DDR, inflammation, and EGFR signaling. Their dynamic, on-treatment nature supports their role as pharmacodynamic markers and warrants further evaluation in biomarker-guided therapeutic strategies.