Thawing “cold” tumors: The NRF2–COX2–PGE2 axis as a pathway to precision immunotherapy in HCC

The introduction of immune checkpoint inhibitors (ICIs) has fundamentally reshaped the treatment paradigm for advanced hepatocellular carcinoma (HCC). In particular, the combination of programmed death-ligand 1 (PD-L1) blockade with anti–vascular endothelial growth factor (VEGF) therapy—atezolizumab plus bevacizumab (Atez/Bev)—has established a new first-line standard of care.1 Nevertheless, a substantial proportion of patients exhibit primary resistance, and many initial responders eventually develop acquired resistance, underscoring the urgent need to clarify the biological basis of immunotherapy failure in HCC and to develop biomarkers for mechanism-based patient stratification. These challenges are increasingly relevant in the context of a shifting etiological landscape. Non-viral HCC is rapidly rising worldwide, with metabolic dysfunction–associated steatotic liver disease (MASLD) emerging as a major underlying cause. Recent studies suggest that HCC arising in MASLD may be less responsive to immunotherapy,2 highlighting the importance of elucidating etiology-linked pathways that shape the tumor immune microenvironment for the development of precision immunotherapy strategies in HCC. In this issue of Hepatology, Yamamoto et al.3 address these unmet needs through an elegant, integrative approach combining mouse models, single-cell and spatial transcriptomics, and large clinical cohorts. To investigate mechanisms of resistance to combined immune-oncology (IO)–based PD-L1 blockade and VEGF inhibition (IO–VEGF therapy), the authors employed a genetically heterogeneous immunocompetent mouse model of HCC that generates multiple tumor lesions with distinct molecular features within the same host. Using this system, IO–VEGF therapy preferentially eliminated tumors with high cytotoxic T-cell infiltration and immune-activated transcriptional profiles (immune-hot), whereas lesions with limited immune infiltration persisted and expanded (immune-cold). Importantly, this shift reflected selective outgrowth of immune-evasive lesions rather than phenotypic conversion, providing a biological framework for primary resistance in HCC. Within this context, the authors identified activation of the NRF2 pathway as a defining feature of immune-evasive tumor clones. NRF2 is a master regulator of cellular stress responses,4 and activating alterations in the NRF2 pathway—including mutations in NFE2L2 or its negative regulator KEAP1—have been reported in a subset of HCCs.5 Complementary experimental analyses further supported a role for NRF2 activation in enforcing immune exclusion and resistance to IO–VEGF therapy, thereby positioning NRF2 not merely as a metabolic stress sensor, but as a determinant of tumor-intrinsic immune evasion and primary resistance to immunotherapy in HCC. Consistent with the murine findings, analysis of human cohorts revealed that tumors with high NRF2 activity exhibited immune-cold phenotypes and poor clinical responses to Atez/Bev treatment. Importantly, NRF2 activation was not restricted to tumors harboring NFE2L2 or KEAP1 mutations, but was frequently linked to accumulation of p62/SQSTM1, which stabilizes NRF2 through KEAP1 sequestration. Prior studies demonstrated that metabolic stress in MASLD promotes p62 accumulation and hepatocarcinogenesis by rewiring stress-response pathways.6,7 Building on this framework, the present study shows that the p62–NRF2 axis is enriched in non-viral HCC, particularly in MASLD-associated tumors, linking metabolic stress to tumor-intrinsic immune evasion and reduced immunotherapy responsiveness. These findings suggest that etiology-associated differences in immunotherapy efficacy may reflect activation of specific stress-response pathways rather than etiology itself. Notably, these data also indicate that NRF2-driven stress-response programs are dynamic rather than static. Analysis of paired pre-treatment and post-treatment human tumor biopsies revealed reinforcement of NRF2-associated signatures during therapy, implicating this pathway not only in primary resistance but also in the development of acquired resistance. Mechanistically, the authors show that NRF2 promotes immune exclusion through transcriptional activation of the cyclooxygenase-2 (COX2)–prostaglandin E2 (PGE2) pathway. NRF2 activation induces COX2 expression via ATF1/ATF2-dependent transcriptional programs, leading to increased intratumoral PGE2 production. This program was observed in murine models and recapitulated in human HCC cohorts, in which high NRF2 activity was associated with elevated COX2/PGE2 signatures, immune-cold tumor microenvironments, and poor responses to Atez/Bev therapy. PGE2 has emerged as a critical immunomodulatory lipid mediator in cancer. While historically linked to dendritic cell regulation and myeloid-driven immunosuppression, 2 independent studies published in 2024 demonstrated that PGE2 directly constrains antitumor immunity by limiting the expansion and effector differentiation of tumor-infiltrating stem-like CD8+ T cells through disruption of IL-2 signaling and mitochondrial fitness.8,9 In line with these findings, NRF2-high tumors displayed enhanced COX2–PGE2 signaling accompanied by marked exclusion of functional CD8+ T cells, establishing this pathway as a mechanistic link between tumor-intrinsic stress responses and immune escape in HCC. These mechanistic insights have direct therapeutic implications. In preclinical models, pharmacologic COX2 inhibition attenuated PGE2 signaling and remodeled the tumor immune microenvironment toward immune-hot states, restoring sensitivity to IO–VEGF therapy. Retrospective human analyses further showed improved outcomes in patients receiving concomitant COX2 inhibitors, such as celecoxib, during IO–VEGF treatment. Moreover, elevated circulating PGE2 levels were associated with inferior responses and shorter overall survival, supporting serum PGE2, together with NRF2-related and COX2-related transcriptional signatures, as candidate biomarkers for identifying patients with immune-cold tumors unlikely to benefit from IO–VEGF therapy alone. Collectively, these findings support a precision immunotherapy framework in HCC, guided by tumor-intrinsic stress-response programs rather than clinical stage or etiology alone (Figure 1).FIGURE 1: The NRF2–COX2–PGE2 axis as a molecular driver of immune exclusion and a therapeutic target in HCC. In NRF2-activated HCC (left), driven by genetic alterations or metabolic stress such as MASLD-mediated p62 accumulation, activation of the NRF2–COX2–PGE2 pathway promotes an immune-cold tumor microenvironment. Elevated PGE2 levels exert potent immunosuppressive effects, limiting effective antitumor immunity and conferring resistance to IO–VEGF therapy (eg, atezolizumab plus bevacizumab). Circulating PGE2 may serve as a non-invasive, blood-based biomarker indicative of this immune-resistant state. Conversely (right), pharmacological inhibition of COX2 disrupts this suppressive axis, converting the tumor microenvironment from immune-cold to immune-hot. This transition sensitizes tumors to IO–VEGF therapy and provides a conceptual framework for precision immunotherapy in HCC. Figure created with BioRender.com. Abbreviations: COX2, cyclooxygenase-2; IO–VEGF, immune-oncology (IO)–vascular endothelial growth factor (VEGF); MASLD, metabolic dysfunction–associated steatotic liver disease; PGE2, prostaglandin E2.Several considerations remain when interpreting these findings. Because COX2 is a well-established regulator of angiogenesis, the NRF2–COX2–PGE2 axis identified here may be particularly relevant in the setting of combined immune checkpoint blockade and VEGF inhibition, potentially explaining its strong association with resistance to IO–VEGF therapy. Whether this pathway similarly influences resistance to other immunotherapeutic platforms, such as CTLA-4–based IO–IO regimens, remains an important question for future studies. In addition, the biomarker potential of circulating PGE2, as well as the efficacy and safety of COX2 inhibition in immunotherapy settings, will need to be clarified in prospective studies. Finally, how NRF2–COX2–PGE2 signaling is activated or reinforced during the emergence of acquired resistance warrants further investigation. Nonetheless, this work represents a clear conceptual advance by linking tumor-intrinsic stress-response programs to immune exclusion and therapeutic resistance in HCC. Importantly, precision immunotherapy should be dynamic rather than static, as stress-response pathways such as the NRF2–COX2–PGE2 axis can be reinforced during treatment, highlighting circulating PGE2 as a potential real-time biomarker of the evolving tumor immune microenvironment. Moreover, in light of recent evidence implicating COX2–PGE2 signaling in fibrogenesis in MASLD,10 the present study extends the relevance of this pathway beyond immunotherapy resistance to the formation of a tumor-promoting niche across multiple stages of hepatocarcinogenesis. In this broader context, targeting the NRF2–COX2–PGE2 axis—including through COX2 inhibition—may offer a therapeutically attractive strategy, particularly given that COX2 inhibitors are already widely used in clinical practice, enabling the potential for drug repositioning. In summary, this study identifies the NRF2–COX2–PGE2 axis as a key tumor-intrinsic mechanism underlying immune exclusion and resistance to immunotherapy in HCC. By linking cellular stress responses to immune-cold tumor states, the work provides a mechanistic basis for treatment resistance and points toward rational combination strategies to improve immunotherapy outcomes in HCC.