The research team, led by investigators at Beijing University of Chinese Medicine and Peking University Third Hospital, systematically searched PubMed, Embase, and the Cochrane Library through 6 November 2025. From an initial haul of 1,180 records, they identified ten randomized controlled trials encompassing 3,959 patients with metastatic colorectal cancer who had progressed after second-line therapy. Each trial compared one of the three oral agents—alone or, in some cases, TAS-102 combined with bevacizumab—against placebo or best supportive care. Risk of bias was assessed with the Cochrane RoB-2 tool, and the protocol was prospectively registered with PROSPERO.

Because no trial has ever randomized patients directly between regorafenib, fruquintinib, and TAS-102, the researchers turned to network meta-analysis, a statistical framework that uses a common comparator—in this case placebo—to make indirect comparisons across trials. They fitted a Bayesian random-effects consistency model with non-informative normal priors, ran four parallel Markov chains through 20,000 adaptation iterations and 50,000 sampling iterations, and confirmed convergence using the Gelman-Rubin diagnostic. The resulting network was star-shaped, anchored on placebo, with no closed loops among the active treatments, which meant formal inconsistency testing was impossible; instead, the team relied on transitivity checks comparing trial characteristics and potential effect modifiers.

The headline finding for the overall population was straightforward: every active treatment significantly prolonged overall survival compared with placebo. Fruquintinib produced a hazard ratio of 0.66 (95% credible interval 0.53–0.84), regorafenib 0.69 (0.51–0.88), TAS-102 0.69 (0.55–0.85), and TAS-102 plus bevacizumab 0.41 (0.28–0.59). The combination of TAS-102 with bevacizumab ranked first by SUCRA—the surface under the cumulative ranking curve—while the three single agents showed broadly similar efficacy, with indirect differences among them not consistently statistically significant. Progression-free survival followed the same pattern, with all experimental arms outperforming placebo and the doublet again delivering the largest benefit.

One indirect comparison did stand out: fruquintinib showed a more favorable progression-free survival than TAS-102, with a hazard ratio of 0.65 (95% CrI 0.45–0.96), placing it highest among the monotherapies. The authors caution, however, that this estimate rests entirely on indirect evidence in the absence of head-to-head trials and should be interpreted cautiously. Heterogeneity also varied considerably across the network: it was essentially absent for fruquintinib versus placebo (I² ≈ 0%), moderate for TAS-102 (OS I² = 21.8%), and high for regorafenib (OS I² = 67.5%, PFS I² = 83.1%), prompting the use of random-effects models throughout.

The most novel contribution of the analysis lies in its KRAS stratification. During transitivity checks, KRAS mutational status emerged as the most plausible source of the between-study heterogeneity, driven largely by differences in baseline KRAS prevalence—Asian trials such as CONCUR and TERRA enrolled 34% and 37% KRAS-mutant patients respectively, while the remaining trials ranged from 40% to 70%. When the team restricted the analysis to trials reporting KRAS-subgroup overall survival data, heterogeneity collapsed dramatically in the wild-type subgroup, with I² falling to 20% or below for every comparison, and the three agents ranked almost side by side with no meaningful differences between them.

In the KRAS-mutant subgroup, the picture shifted subtly. Fruquintinib and TAS-102 each produced numerically favorable overall survival estimates versus placebo—hazard ratios of 0.71 (95% CrI 0.51–1.01) and 0.74 (95% CrI 0.50–0.99) respectively—whereas regorafenib did not reach statistical significance, with a hazard ratio of 0.81 (95% CrI 0.52–1.19). The exploratory ranking placed fruquintinib first, TAS-102 second, and regorafenib last. The authors are emphatic, however, that these differences in statistical significance versus placebo do not demonstrate superiority of fruquintinib or TAS-102 over regorafenib, and that the rank probabilities are hypothesis-generating rather than evidence of a predictive effect of KRAS status.

The mechanistic backdrop helps explain why KRAS status might matter. KRAS is mutated in approximately 40% of colorectal cancers, driving constitutive RAS signaling and unrestrained cell growth. Prior laboratory work has suggested that exon-2 KRAS mutations, particularly G12C and G12A, maintain elevated BRAF-ERK-ELK phosphorylation and blunt regorafenib-induced apoptosis and vessel shutdown, while a recently described KRAS/ERK2–TRIM21–c-Myc/ENO2 glycolytic axis promotes regorafenib resistance. Notably, among the three TAS-102 trials, the Japanese phase 2 study by Yoshino and colleagues showed a markedly stronger benefit in KRAS-mutant patients (hazard ratio 0.44) than in wild-type disease (0.70), a pattern that largely accounts for the residual heterogeneity in that treatment node.

The combination of TAS-102 with bevacizumab emerged as the standout regimen overall. Mechanistically, the pairing is plausible: trifluridine inhibits thymidylate synthase and, through its metabolite FTY, becomes incorporated into DNA causing miscoding, while tipiracil protects trifluridine from degradation by blocking thymidine phosphorylase; bevacizumab, meanwhile, suppresses VEGF-driven angiogenesis and remodels the tumor microenvironment. The authors suggest these mechanisms may act synergistically, and a separate analysis indicated the combination conferred a particularly pronounced survival benefit in RAS-mutant patients (hazard ratio 0.38, 95% CrI 0.19–0.79) with no significant advantage over monotherapy in the wild-type population—though only one study contributed to that subgroup node, precluding a stable network estimate.

The study has clear limitations that the authors acknowledge candidly. No head-to-head randomized comparisons exist among the three agents, the star-shaped network precluded formal inconsistency testing, and individual-participant data were unavailable, so a formal treatment-by-KRAS interaction test could not be performed. Comparative safety was not formally meta-analyzed because safety definitions and reporting thresholds differed substantially across trials, and progression-free survival subgroup analysis was abandoned due to insufficient sample size and high heterogeneity (I² ranging from 77.7% to 92.2%). Publication-bias testing was deemed unreliable with fewer than ten trials per outcome. The authors conclude that for KRAS-wild-type patients, standard third-line regimens or TAS-102 plus bevacizumab are reasonable, while for KRAS-mutant patients fruquintinib may represent a sensible later-line option—pending confirmation from prospective, KRAS-stratified or head-to-head trials that the field has yet to deliver.

Subject of Research: Comparative efficacy of third-line oral therapies for KRAS-mutant metastatic colorectal cancer

Article Title: Comparative efficacy of fruquintinib, regorafenib or TAS-102 in patients with KRAS-mutant metastatic colorectal cancer: a systematic review and network meta-analysis

Article References: Zhou, Y., Zhao, S., Zheng, X., Wang, Y., Wang, M., Chen, S., Wu, Z., Zang, D., & Si, W. (2026). Comparative efficacy of fruquintinib, regorafenib or TAS-102 in patients with KRAS-mutant metastatic colorectal cancer: a systematic review and network meta-analysis. Frontiers in Oncology, 16, Article 1948508. https://doi.org/10.3389/fonc.2026.1948508

Image Credits: AI Generated

DOI: 10.3389/fonc.2026.1948508

Keywords: fruquintinib, regorafenib, TAS-102, KRAS mutation, metastatic colorectal cancer, network meta-analysis, overall survival, progression-free survival, bevacizumab, third-line therapy, precision oncology, Bayesian analysis

The logic of the new study rests on a vulnerability embedded in the biology of cancer itself: the DNA damage checkpoint. When anticancer drugs inflict lesions on DNA, healthy cells respond by halting the cell cycle through checkpoint pathways, buying time for repair before replication resumes. Two major checkpoints govern this process. The G1 checkpoint depends on the ATM–p53–p21 axis, while the G2/M checkpoint is controlled by the ataxia telangiectasia and Rad3-related kinase, or ATR, acting through Chk1 and Cdc25. Because malignant cells frequently carry p53 mutations and cannot arrest in G1, they become unusually dependent on the ATR-driven G2/M checkpoint to survive genotoxic stress. Normal cells, by contrast, retain a functional ATM–p53–p21 pathway. This asymmetry creates a therapeutic window: blocking ATR should selectively cripple cancer cells while sparing healthy tissue.

AZD6738, developed as an oral ATR inhibitor, does exactly that. By suppressing ATR-dependent phosphorylation—most notably Chk1 phosphorylation at serine 345—it disables the G2/M checkpoint that resistant tumors rely upon. The drug has already attracted attention across oncology, with Phase II trials suggesting clinical activity in combination with the immunotherapy durvalumab in previously treated advanced gastric cancer, particularly in patients lacking ATM expression, and in non-small cell lung cancer. Emerging data also support combinations with the PARP inhibitor olaparib in triple-negative breast cancer, with durvalumab in melanoma, and with radiotherapy. But in colorectal cancer, preclinical work on ATR inhibition had remained sparse. The Nagoya team had previously shown that AZD6738 potentiates both 5-FU and trifluridine in treatment-naïve colorectal cancer cells. The critical open question was whether that synergy survives the acquisition of 5-FU resistance—the setting in which trifluridine is actually given to patients.

To answer it, the researchers built a rigorous experimental system using two genetically defined human colorectal cancer cell lines, HCT116 and DLD-1, alongside 5-FU–resistant derivatives established through more than 100 passages of stepwise drug escalation. Both resistant sublines, HCT116/5FUR and DLD-1/5FUR, exhibited roughly 100-fold higher IC50 values for 5-FU than their parental counterparts, confirming profound acquired resistance. All cell lines were authenticated by short tandem repeat profiling and confirmed free of mycoplasma contamination. The team then designed experiments around a deliberately conservative dosing strategy: AZD6738 was used at 0.5 micromolar, a concentration shown to have no effect on proliferation after 72 hours in any of the four cell lines, so that any enhancement of cell killing could be attributed to genuine potentiation rather than simple additive toxicity.

The in vitro results were striking. As expected, 5-FU and its combination with AZD6738 lost all effectiveness in the resistant sublines—the combination that worked in parental cells was completely abrogated by acquired resistance, echoing earlier findings with Chk1 inhibitors. Trifluridine, however, told a different story. Although trifluridine alone was modestly weakened in the resistant cells, adding noncytotoxic AZD6738 significantly reduced cell viability across a wide dose range, from 1 to 1000 micromolar. Combination index analysis confirmed true synergy, with values of 0.735 in HCT116/5FUR and 0.418 in DLD-1/5FUR, where values below 1 indicate synergistic interaction. In time-course assays, the combination suppressed proliferation significantly at 72 hours in both resistant lines—p = 6.0 × 10⁻⁵ for HCT116/5FUR and p = 4.0 × 10⁻⁷ for DLD-1/5FUR—and restored a level of growth inhibition comparable to what trifluridine achieved in the parental, drug-sensitive cells. DLD-1/5FUR, the most resistant line, saw its trifluridine IC50 collapse from 155 micromolar to just 3.5 micromolar in the presence of AZD6738.

The molecular fingerprints of the combination revealed its mechanism. Western blotting showed that trifluridine alone produced little induction of γH2AX, a marker of DNA double-strand break signaling, but cotreatment with AZD6738 triggered a marked surge. Cleaved caspase-3, the executioner enzyme of apoptosis, rose far more strongly with the combination than with trifluridine alone, indicating that stalled checkpoint signaling had collapsed into programmed cell death. Meanwhile, the trifluridine-induced phosphorylation of Chk1 at Ser345—the very signal AZD6738 is designed to block—was suppressed by the drug, confirming on-target activity. Flow cytometry added a cell-cycle dimension: trifluridine alone drove a pronounced G2/M arrest, effectively locking damaged cells in a state of suspended repair, while the combination relieved that arrest, pushing cells past the checkpoint and into catastrophic division with unrepaired DNA damage.

Crucially, the synergy translated into living tumors. In a mouse xenograft model bearing DLD-1/5FUR tumors, neither TAS-102 alone nor AZD6738 alone significantly shrank tumors compared with controls. But the combination significantly suppressed tumor growth over the two-week treatment period, and at necropsy only the combination group showed a significant reduction in tumor weight relative to all other groups, including TAS-102 monotherapy (p = 0.039). Immunohistochemistry on excised tumors corroborated the mechanism in vivo: 25.8 percent of tumor cell nuclei stained positive for γH2AX in the combination group, compared with only 4.1 percent in the TAS-102 group. Equally important, safety signals were reassuring. Body weight did not differ among groups, and organ weights and blood counts showed no exacerbation of the hematologic toxicity that TAS-102 alone can induce, suggesting the addition of AZD6738 did not compound harm in this preclinical model.

The pharmacological reasoning behind the finding illuminates why trifluridine, unlike 5-FU, retains its potency in resistant tumors. Both drugs inhibit thymidylate synthase, an enzyme essential for DNA synthesis, and both can be incorporated into DNA and RNA. However, overexpression of thymidylate synthase is a major driver of 5-FU resistance, neutralizing the TS-inhibitory arm of both drugs. Trifluridine’s distinguishing feature is its exceptionally high rate of direct incorporation into DNA during S phase in place of thymidine, and earlier work by Emura and colleagues demonstrated that trifluridine retains substantial cytotoxicity in 5-FU–resistant colorectal cancer cells precisely because its activity depends largely on DNA incorporation rather than TS inhibition. That massive misincorporation generates replication stress, which the ATR–Chk1 axis manages at the G2/M checkpoint—making trifluridine and ATR inhibition a mechanistically rational pairing even after 5-FU resistance has taken hold.

The authors are candid about the study’s limitations. Mechanistic conclusions rest solely on pharmacological ATR inhibition without genetic validation; normal colon epithelial cells were not evaluated, leaving the therapeutic window uncertain; and only a single concentration of AZD6738 was tested, so broader dose–response studies will be needed. The roles of p53 and ATM also remain unresolved, as synergy appeared in both p53 wild-type and p53-mutant models, and no direct experiments interrogated these determinants. Nevertheless, the clinical relevance of the setting is hard to overstate: trifluridine is administered as TAS-102, known commercially as Lonsurf, almost exclusively after failure of fluoropyrimidine-based first- and second-line regimens, meaning most patients receiving it already harbor 5-FU resistance. By providing proof of concept that pharmacological ATR inhibition can restore—and even rescue—the antitumor activity of trifluridine in that refractory context, the study lays the groundwork for mechanistic follow-up and, ultimately, translational trials of a ceralasertib–TAS-102 combination in patients whose options are running out.

Subject of Research: ATR inhibition with AZD6738 to enhance trifluridine efficacy in 5-FU–resistant colorectal cancer

Article Title: AZD6738 (Ceralasertib) Enhances Trifluridine's Antitumor Effect in 5‐FU–Resistant Colorectal Cancer Cells

Article References: Uehara, S., Suzuki, T., Kato, J., Asai, H., Kato, A., Harata, S., Ushigome, H., Yamakawa, Y., Hirokawa, T., Takahashi, H., Matsuo, Y., & Takiguchi, S. (2026). AZD6738 (Ceralasertib) Enhances Trifluridine's Antitumor Effect in 5‐ FU –Resistant Colorectal Cancer Cells. Cancer Reports, 9(9), Article e70668. https://doi.org/10.1002/cnr2.70668

Image Credits: AI Generated

DOI: 10.1002/cnr2.70668

Keywords: AZD6738, ceralasertib, trifluridine, TAS-102, colorectal cancer, 5-FU resistance, ATR inhibitor, DNA damage checkpoint, Chk1, G2/M checkpoint, drug resistance, preclinical study