Unmasking immune checkpoint resistance in esophageal squamous cell carcinoma: Insights into the tumor microenvironment and biomarker landscape

Abstract

Esophageal squamous cell carcinoma (ESCC) remains a daunting global health concern. It is marked by aggressive progression and poor survival. While immunotherapy has emerged as a promising treatment modality, both primary and acquired resistance continue to limit its clinical impact, leaving many patients without durable benefits (e.g., CheckMate-648, ESCORT-1^st). This review explains resistance mechanisms and suggests new strategies to improve outcomes. These mechanisms include immunosuppressive cells (Treg cells, myeloid-derived suppressor cells), inhibitory cytokines, molecular alterations involving programmed death 1/programmed death-ligand 1 signaling, and impaired antigen presentation. We also highlight key clinical trials—for example, CheckMate-648 and ESCORT-1^st—that reveal both the potential and pitfalls of current immune checkpoint blockade strategies, underscoring the need for robust predictive biomarkers. Moreover, we examine cutting-edge tactics to overcome resistance, including combination regimens, tumor microenvironment remodeling, and tailored treatment approaches rooted in the patient’s unique genomic and immunologic landscape.

Key Words: Esophageal squamous cell carcinoma; Immunotherapy resistance; Tumor microenvironment; Immune checkpoint blockade; Biomarkers

Core Tip: Esophageal squamous cell carcinoma (ESCC) poses significant clinical challenges due to its aggressive progression and poor survival outcomes. While immunotherapy offers promise, resistance—both primary and acquired—remains a major hurdle, limiting its efficacy. This review explores the mechanisms of immunotherapy resistance in ESCC, including immunosuppressive cells, inhibitory cytokines, and molecular alterations in programmed death 1/programmed death-ligand 1 signaling. We also examine emerging strategies to overcome resistance, such as combination therapies and tumor microenvironment remodeling, emphasizing the need for predictive biomarkers to improve patient outcomes.

INTRODUCTION

Esophageal cancer, characterized by aggressive progression and poor prognosis, poses a significant global health challenge. According to GLOBOCAN 2022 estimates (511054 new cases and 445391 deaths in 2022), esophageal cancer ranks as the 11^th most common cancer and the 7^th leading cause of cancer-related death worldwide[1]. In the United States, 2001140 new cancer cases and 611720 deaths are projected in 2024, reflecting rising incidence in several major cancers and persistent disparities[2]. Distinct geographic variations in esophageal cancer incidence exist, particularly high in Eastern Asia and Eastern Africa, highlighting diverse etiologies across regions. Recent insights emphasize the importance of targeting not only tumor cells but also their surrounding microenvironment, including stromal and immune components, to overcome therapy resistance[3]. Notably, two major histologic subtypes—esophageal squamous cell carcinoma (ESCC) and adenocarcinoma—exhibit substantial epidemiological differences. ESCC, predominant globally, is strongly linked to lifestyle factors such as tobacco smoking and alcohol consumption.

Despite advancements in multimodal treatment such as surgery, chemotherapy, radiotherapy, and immunotherapy, the clinical outcomes of ESCC patients remain suboptimal. Recently, immune checkpoint inhibitors (ICIs) targeting programmed death 1 (PD-1)/programmed death-ligand 1 (PD-L1) have become integral in ESCC management[4,5]. Landmark trials, including CheckMate-648, ESCORT-1^st, JUPITER-06, RATIONALE-306 and GEMSTONE-304 have collectively confirmed the survival benefits of combining ICIs with chemotherapy[6-11]. For instance, ESCORT-NEO/NCCES01 demonstrated significantly improved pathological complete response rates with camrelizumab plus chemotherapy compared to chemotherapy alone in the neoadjuvant setting[12]. Similarly, first-line trials such as JUPITER-06 (toripalimab) and GEMSTONE-304 (sugemalimab) established significant enhancements in overall survival and progression-free survival, emphasizing immunotherapy's pivotal role in ESCC treatment[6,8].

However, the clinical effectiveness of immunotherapy in ESCC remains constrained by various resistance mechanisms. A substantial proportion of patients experience either primary or acquired resistance, diminishing long-term therapeutic benefits. Complex factors contribute to this resistance, including immunosuppressive cellular components within the tumor microenvironment (TME), notably regulatory T cells (Tregs) and myeloid-derived suppressor cells (MDSCs). Additional resistance mechanisms encompass molecular alterations affecting antigen presentation, immune checkpoint signaling dysregulation, and inhibitory cytokine production[13,14]. These multifaceted mechanisms underscore the urgent need for reliable biomarkers to predict immunotherapy responsiveness.

Addressing these challenges requires innovative strategies. Recent research emphasizes exploring combination therapies, integrating ICIs with chemotherapy, targeted therapies, and other immunomodulators to overcome immunotherapy resistance[14]. Emerging predictive biomarkers, such as immune-related gene signatures and proteomic profiles, hold promise for improved patient stratification and individualized therapeutic approaches[15,16]. Notably, G protein-coupled receptor 84 (GPR84), predominantly expressed on MDSCs, has been identified as a negative predictor for anti-PD-1 therapy efficacy in ESCC patients. High levels of GPR84⁺ MDSCs correlate with reduced CD8⁺ T cell infiltration and poorer overall survival, underscoring its potential as both a prognostic biomarker and a therapeutic target[17].

Collectively, elucidating resistance mechanisms and identifying effective predictive biomarkers remain critical unmet needs. This review synthesizes recent advances, provides a comprehensive analysis of resistance mechanisms, and highlights promising therapeutic strategies aimed at overcoming immunotherapy resistance, ultimately seeking transformative improvements in clinical outcomes for ESCC patients.

MOLECULAR MECHANISMS OF IMMUNOTHERAPY RESISTANCE IN ESCC

Immunosuppressive ESCC microenvironment

Recent evidence indicates that multiple immunosuppressive factors and cell populations within the ESCC microenvironment interact to inhibit effective antitumor immune responses. At the cytokine level, interleukin (IL)-10 is particularly notable for its dual role. While it can inhibit certain proinflammatory mediators (e.g., IL-6, IL-12/23) and block excessive myeloid or Treg accumulation, it may also potentiate CD8⁺ T-cell effector functions. This is achieved by enhancing interferon-gamma (IFN-γ) production, increasing major histocompatibility complex (MHC) expression, and mitigating terminal exhaustion[18-20]. Nevertheless, under particular conditions, macrophage-derived IL-10 can limit dendritic cell (DC) activity. This suggests that IL-10’s ultimate impact depends heavily on the overall immunological and therapeutic context[21]. IL-6, in contrast, more consistently drives tumor-cancer-associated fibroblast (CAF) interactions and is frequently associated with poorer survival outcomes, suggesting that strategies targeting IL-6 signaling could help disrupt a tumor-promoting microenvironment[22]. IL-1β has also emerged as a key mediator of immune escape in ESCC through ferroptosis resistance: Blocking IL-1β sensitizes tumor cells to CD8⁺ T-cell-mediated cytotoxicity, thereby synergizing with PD-1 blockade[23].

Beyond these soluble mediators, transforming-growth-factor-beta (TGF-β) exerts wide-ranging immunosuppressive and pro-tumorigenic effects in ESCC by impairing T-cell and natural killer (NK)-cell function, inducing Treg differentiation, and promoting invasion and metastasis[24,25]. The pleiotropic nature of TGF-β poses a challenge for clinical intervention, as globally blocking TGF-β may also remove its normal tissue homeostasis benefits. At the cellular level, CAFs serve as central modulators of the ESCC microenvironment through their capacity to secrete growth factors such as fibroblast growth factor 2 (FGF2), IL-6, and TGF-β[26,27]. They also remodel the extracellular matrix, a process that can hinder T-cell infiltration and enhance immune evasion[27,28]. Notably, certain subpopulations of CAFs may help activate T cells, underscoring the functional heterogeneity that complicates simple “CAF-depletion” strategies.

MDSCs promote immune tolerance through various mechanisms, including nutrient depletion and stabilization of PD-L1 expression. Specifically, GPR84 overexpression on MDSCs inhibits lysosomal degradation of PD-L1, leading to sustained PD-L1 Levels on the cell surface. This stabilization impairs CD8⁺ T cell function and contributes to resistance against anti-PD-1 therapies[29]. In parallel, Treg cells curtail cytotoxic T lymphocyte activity through multiple pathways, including trogocytosis of CD80/CD86 on antigen-presenting cells and the release of immunosuppressive cytokines[30-32]. Finally, tumor-associated macrophages (TAMs)—particularly M2-like phenotypes—facilitate immune escape through TGF-β1 secretion, PD-L1 upregulation, and direct interference with T-cell trafficking[33-36]. Taken together, these overlapping immunosuppressive signals and cell subsets form a multilayered barrier to immune-mediated tumor eradication (Figure 1). A deeper understanding of these interactions will be critical for designing precise combination therapies aimed at disarming immunosuppressive networks, thereby improving the efficacy of existing immunotherapeutic approaches in ESCC.

Figure 1 Immunosuppressive micro-environment in esophageal squamous cell carcinoma. The diagram highlights the dominant suppressor populations and cytokines that shape the esophageal squamous cell carcinoma tumor micro-environment. Treg, myeloid-derived suppressor cell, tumor-associated macrophage, dendritic cell and cancer-associated fibroblast are shown interacting with effector T cells and natural killer cells. Colored dots denote key soluble mediators [blue = transforming-growth-factor-beta, purple = interleukin (IL)-10, orange = interferon-gamma, grey = IL-6]. Solid and dashed arrows indicate predominant inhibitory or stimulatory signals affecting cytotoxicity and antigen presentation. The schematic inset at the top illustrates how dense stroma and aberrant vasculature further impede immune-cell trafficking. IL: Interleukin; MHC: Major histocompatibility complex; PD-L1: Programmed death-ligand 1; TGF-β: Transforming-growth-factor-beta; IFN-γ: Interferon-gamma.

Checkpoint and exhaustion pathways

Immune evasion in ESCC arises from a constellation of regulatory pathways and molecular alterations that collectively limit the effectiveness of immunotherapeutic interventions. Central to this process are well-characterized checkpoints such as PD-1/PD-L1 and cytotoxic T-lymphocyte-associated protein 4 (CTLA-4), which orchestrate inhibitory signals not only in T cells but also in other immune cells including NK cells and macrophages[37-39]. PD-L1 engagement with PD-1, for instance, leads to the formation of suppressive microclusters that recruit phosphatases like SHP2, thereby attenuating TCR- and CD28-mediated costimulatory signals[40]. PD-L2, although overlapping in function with PD-L1, exhibits distinct expression patterns—particularly under low-antigen conditions—and further fortifies immunosuppression[41,42]. Meanwhile, CTLA-4 exerts inhibitory effects earlier in T-cell activation by outcompeting CD28 for B7 Ligands. This mechanism plays a particularly important role in regulatory T cells (Tregs), where CTLA-4 helps maintain immune tolerance—but in cancer, this can inadvertently protect tumor cells[43,44]. Importantly, immune suppression in the TME is not limited to checkpoint pathways alone. Tumor cells and stromal components cooperatively drive CD8⁺ T-cell dysfunction, often pushing them into an exhausted state through persistent antigen exposure and metabolic stress. Recent evidence from breast cancer, for example, has revealed novel exhaustion markers in CD8⁺ T cells, highlighting how the tumor milieu actively reprograms antitumor immunity[45]. These insights not only clarify mechanisms of immune evasion but also point toward innovative therapeutic strategies. Among them, next-generation chimeric antigen receptor (CAR) T-cell therapies are being designed to overcome the hostile metabolic conditions of solid tumors—such as nutrient deprivation and lactic acid accumulation—that blunt T-cell cytotoxicity[46].

In parallel, T-cell exhaustion is a major driver of immune escape in ESCC. Chronic antigen exposure induces sustained expression of inhibitory receptor proteins such as PD-1, LAG-3, and others, leading to functional impairment in proliferation, cytokine production, and cytotoxic activity[47-50]. Although PD-1 blockade can partially restore exhausted T-cell function, these effects are frequently transient. Emerging evidence points to epigenetic reprogramming as a key contributor to the stable maintenance of T-cell exhaustion. This includes mechanisms such as DNA methylation changes and the transcription factor TOX, which encodes the thymocyte selection-associated high mobility group box protein and helps enforce the exhausted state[48,51,52]. Nevertheless, a subset of exhausted T cells with progenitor-like properties retains a degree of proliferative capacity and can respond more effectively to ICIs[53,54]. Additional factors in ESCC, such as long non-coding RNA (lncRNA) FOXP4-AS1 or fibroblast-derived FGF2, further exacerbate T-cell exhaustion by stabilizing PD-L1 expression or upregulating SPRY1, respectively, underscoring the multifactorial nature of this process[55].

Reduced tumor antigen presentation represents another crucial mechanism of immune evasion. Multiple lines of evidence demonstrate that the downregulation of MHC class I molecules—whether through B2M or human leukocyte antigen (HLA) mutations, epigenetic silencing by Polycomb complexes, or disruptions in mitochondrial metabolism—compromises effective T-cell recognition[56-59]. Additionally, embryonic transcription factors such as DUX4 can override IFN-γ-mediated MHC-I upregulation, further compromising antigen presentation[60]. These multifaceted pathways collectively diminish the tumor’s visibility to cytotoxic lymphocytes and highlight the importance of restoring antigen presentation for successful immunotherapy.

Beyond T-cell dynamics and antigen presentation, ESCC cells can actively alter their immunophenotype to evade immune surveillance. Elevated CD276 (B7-H3) expression, for example, drives the formation of neutrophil extracellular traps via the CXCL1-CXCR2 axis, which in turn inhibits NK cell function and promotes immune escape[61]. The Hippo-YAP pathway similarly facilitates immune escape by upregulating CD24, generating a "do-not-eat-me" signal that impedes macrophage phagocytosis[62]. These findings highlight that ESCC cells exploit multiple parallel strategies—ranging from checkpoint molecule manipulation to metabolic reprogramming and interference with innate immune effectors.

Finally, additional molecular factors can further reinforce immune evasion. STC1, a less conventional “phagocytosis checkpoint”, limits calreticulin exposure on the tumor cell surface, thereby preventing effective macrophage and DC-mediated antigen uptake[63]. Moreover, certain lncRNAs modulate immune-related gene expression via epigenetic modifications such as mixed-lineage leukemia-1-dependent tri-methylation of histone H3K4me3 changes, further weakening tumor immunogenicity[64]. Taken together, these findings underscore the complex network of immunosuppressive pathways in ESCC, suggesting that future therapeutic approaches will likely require a combination of targeted interventions to reactivate T cells, restore antigen presentation, and dismantle tumor-intrinsic immune evasion strategies (Figure 2).

Figure 2 Integrated mechanisms of immune evasion in esophageal squamous cell carcinoma. This overview assembles the main, temporally and spatially coordinated, escape pathways employed by esophageal squamous cell carcinoma (ESCC). (1) Loss of tumor antigen visibility through B2M/human leukocyte antigen mutations, polycomb-mediated epigenetic silencing and mitochondrial dysfunction that down-regulate major histocompatibility complex-I; (2) FOXP4-AS1 and interferon-gamma-driven programmed death-ligand 1/programmed death-ligand 2 up-regulation recruits SHP-2 via PD-1 and LAG-3 to dampen TCR/CD28 signaling; (3) Tumour-intrinsic programmes—Hippo-YAP-induced CD24, CD276-CXCL1-CXCR2-mediated neutrophil extracellular traps, and STC1-mediated calreticulin masking—reconfigure “don’t-eat-me” or pro-inflammatory cues to alter immune phenotype; and (4) Cancer-associated fibroblast-derived fibroblast growth factor 2 elevates SPRY1 in CD8⁺ T cells, deepening functional exhaustion. Together these layers converge to reduce antigen presentation, blunt cytotoxic function and remodel innate immunity, thereby enabling durable immune escape in ESCC. HLA: Human leukocyte antigen; IFN-γ: Interferon-gamma PD-L1: Programmed death-ligand 1; PD-L2: Programmed death-ligand 2; FGF2: Fibroblast growth factor 2; MHC: Major histocompatibility complex; NK: Natural killer.

CLINICAL ADVANCES IN IMMUNOTHERAPY FOR ESCC

Clinical outcomes of immunotherapy and mechanisms of resistance

Checkpoint blockade has permeated every treatment window of ESCC, but the magnitude and durability of benefit are tightly linked to clinical context. In the adjuvant setting, the CheckMate 577 trial showed that one year of nivolumab nearly doubled median disease-free survival compared to placebo (22.4 months vs 11.0 months; HR = 0.69; P < 0.001). This provided the first strong evidence that immunotherapy can eliminate micrometastatic disease in ESCC[65]. This and other pivotal trials are summarized in Table 1. Yet the study deliberately excluded patients who achieved a pathologic complete response (pCR) to chemoradiotherapy, leaving unresolved whether adjuvant therapy is necessary for the very subgroup most likely to be cured by surgery alone. Longer follow-up is also required to confirm that disease-free survival improvements translate into overall survival benefits, especially as many patients relapse outside the thorax, where local control is optimal. The neoadjuvant setting is evolving even faster. The phase III ESCORTNEO trial established that adding camrelizumab to platinumtaxane chemotherapy significantly increases pCR in resectable, locally advanced tumors without delaying surgery or enhancing perioperative morbidity[12]. Smaller phase Ib/II trials have further expanded the possibilities: Two cycles of PD-L1 blockade with adebrelimab as monotherapy resulted in respectable major pathological responses with minimal high-grade toxicity[66], while toripalimab combined with chemoradiation achieved a remarkable 79% major response and 47% pCR, albeit with predictable cytopenias[67]. Collectively, these data argue that immune priming before resection is feasible and that chemotherapy—or radiation—augments, rather than duplicates, checkpoint activity. They also hint that an inflamed baseline microenvironment predicts deeper responses, whereas CAFrich tumors fare worse, a signal that could steer future patient selection.

Table 1 Summary of clinical studies on immunotherapy for esophageal squamous cell carcinoma.

Line of therapy	Study	Therapy/intervention	Sample size	Study design	Primary outcomes	Key findings	Treatment combination (if any)
Neoadjuvant	ChiCTR2000040034[14]	Camrelizumab + nabpaclitaxel + cisplatin; camrelizumab + paclitaxel + cisplatin; paclitaxel + cisplatin	391 (132/130/129)	Multicenter, randomized, open-label, 3arm trial	pCR rate; EFS	EFS: PCR: 28.0% (camrelizumab + nabpaclitaxel) & 15.4% (camrelizumab + paclitaxel) vs 4.7% (paclitaxel); both P < 0.01; grade ≥ 3 TRAEs: 29%-34%; EFS immature	PD-1 inhibitor + doublet chemotherapy
	NCT04389177[108]	Pembrolizumab + paclitaxel + Cisplatin (neoadjuvant; adjuvant PD-1 if non-pCR)	47	Single-arm, single-center, open-label, phase II	MPR; safety	Results pending (protocol)	PD-1 inhibitor + doublet chemotherapy
	NCT04215471[73]	Adebrelimab	30	Phase Ib, single-arm	Safety, feasibility, pCR	pCR: 8%; major pathologic response: 24%; 2-year OS: 92%; acceptable safety profile with no grade ≥ 3 adverse events	PD-L1 inhibitor monotherapy
	ChiCTR2100045104[109]	Toripalimab + paclitaxel/carboplatin + radiotherapy	23	Phase Ib, single-arm	MPR & pCR	pCR: 55%; 2-year PFS: 63.8%; 2-year OS: 78%; manageable safety profile	PD-1 inhibitor + doublet chemotherapy + radiotherapy
Adjuvant	NCT02743494[72]	Nivolumab vs placebo	794 (532 nivolumab, 262 placebo)	Phase III, randomized, double-blind	DFS	Median DFS significantly improved with nivolumab (22.4 months vs 11.0 months, HR = 0.69, P < 0.001). Acceptable safety profile	PD-1 inhibitor monotherapy
Firstline	NCT03748134[9]	Sintilimab + cisplatin + paclitaxel (or cisplatin + 5-FU)	659 (327/332)	Multicenter, randomized, double-blind, phase III	OS & PFS	OS: 16.7 months vs 12.5 months (HR = 0.63, P < 0.001); PFS: 7.2 months vs 5.7 months (HR = 0.56, P < 0.001); grade ≥ 3 TRAEs: 60% vs 55%	PD-1 inhibitor + doublet chemotherapy
	NCT03829969[8]	Toripalimab + paclitaxel + cisplatin	514 (257/arm)	Multicenter, randomized, double-blind, placebo-controlled, phase III	PFS; OS	PFS: HR = 0.58 (95%CI = 0.46-0.74; P < 0.0001); OS: HR = 0.58 (95%CI = 0.43-0.78; P = 0.0004); grade ≥ 3 TEAEs: Similar between arms	PD-1 inhibitor + doublet chemotherapy
	NCT04187352[10]	Sugemalimab + cisplatin + 5 fluorouracil	540 (360/180)	Multicenter, randomized, double-blind, phase III	PFS; OS; ORR	PFS: 6.2 months vs 5.4 months (HR = 0.67, P = 0.0002); OS: 15.3 months vs 11.5 months (HR = 0.70, P = 0.0076); ORR: 60.1% vs 45.2%; grade ≥ 3 TRAEs: 51.3% vs 48.4%	PD-L1 inhibitor + doublet chemotherapy
	NCT03143153[11]	Nivolumab + chemo/nivolumab + ipilimumab vs chemo	970 (321/325/324)	Phase III, randomized, open-label, 3-arm	OS, PFS	OS: 15.4 months (nivolumab + chemo) vs 10.7 months (chemo); HR = 0.54, P < 0.001; OS: 13.7 months (nivolumab + ipilimumab) vs 10.7 months; HR = 0.64; PFS and ORR also improved	PD-1 inhibitor ± CTLA-4 inhibitor + chemo
	NCT03691090[12]	Camrelizumab + paclitaxel + cisplatin vs paclitaxel + cisplatin	596 (298/298)	Phase III, randomized, open-label	OS	OS: 15.3 months vs 12.0 months; HR = 0.70, P < 0.001; PFS and ORR also improved; acceptable safety profile	PD-1 inhibitor + doublet chemotherapy
Second-line	NCT02564263[6]	Pembrolizumab	387	Phase III, randomized, controlled trial	OS; HRQoL	No significant difference in HRQoL between pembrolizumab and chemotherapy; stable global health, symptom scores in both groups	PD-1 inhibitor monotherapy
	NCT02569242[7]	Nivolumab vs chemotherapy	419 (210 nivolumab, 209 chemotherapy)	Phase III, randomized, open-label	OS	Median OS significantly improved with nivolumab (10.9 months vs 8.4 months, HR = 0.77, P = 0.019). Favorable safety profile compared to chemotherapy	PD-1 inhibitor monotherapy
	NCT03852251[90]	Cadonilimab (anti-PD-1/CTLA-4 bispecific antibody)	22 (ESCC cohort)	Phase Ib/II, multicenter, open-label	ORR, safety	ORR: 18.2% in ESCC cohort; manageable safety profile	PD-1/CTLA-4 inhibitor monotherapy
	NCT03736863[95]	Camrelizumab + apatinib	52	Single-arm, open-label	ORR	ORR: 34.6%; manageable safety profile	PD-1 inhibitor + VEGFR2 inhibitor
	NCT03736863 (rechallenge)[96]	Camrelizumab + apatinib	49	Single-arm, open-label	ORR	ORR: 10.2%; DCR: 69%; median PFS: 4.6 months; manageable grade ≥ 3 AEs in 35%	PD-1 inhibitor + VEGFR2 inhibitor

pCR: Pathologic complete response; EFS: Event-free survival; TRAE: Treatment-related adverse event; MPR: Major pathologic response; OS: Overall survival; PFS: Progression-free survival; DFS: Disease-free survival; ORR: Overall response rate; HRQoL: Health-related quality-of-life; PD-1: Programmed death-1; ESCC: Esophageal squamous cell carcinoma; PD-L1: Programmed death-ligand 1; CTLA-4: Cytotoxic T-lymphocyte-associated protein 4.

For patients with unresectable or metastatic disease, four recent phase III trials converge on a clear conclusion: Combining an anti-PD-1 or anti-PD-L1 antibody with platinumbased doublets reduces the risk of death by approximately onethird compared with chemotherapy alone. One-third achieved an initial. Notably, the benefit extends across PD-L1 subgroups, suggesting that chemotherapy-induced antigen release and dendriticcell maturation can compensate for a “cold” baseline immune phenotype. Toxicity remains manageable—grade ≥ 3 immune events hover below 15%—making these regimens broadly accessible even in resourcelimited settings. Earlyphase monotherapy data with SHR1210 remind us, however, that higher tumor mutational burden and PD-L1 staining of at least 5% enrich for response, reinforcing the idea that intrinsic immunogenicity still matters when cytotoxics are removed from the equation[68].

Once platinum failure occurs, checkpoint monotherapy yields more modest gains. Pembrolizumab improved overall survival over investigator’s-choice chemotherapy in the KEYNOTE181 trial, but only in tumors with a combined positive score ≥ 10—roughly onethird of screened patients[4]. Nivolumab in ATTRACTION3 extended survival irrespective of PD-L1 status, yet the absolute median benefit was only about 2.5 months[5]. These outcomes highlight a biological ceiling for singleagent activity in lateline settings and underscore the importance of earlier intervention or rational combinations, for example, with T-cell immunoreceptor with Ig and ITIM domains (TIGIT), LAG-3, or vascular endothelial growth-factor receptor 2 (VEGFR2) blockade, to break primary resistance.

In summary, the emerging evidence points to two guiding principles. First, the earlier PD-1/PD-L1 inhibition is embedded in the multimodality sequence, the deeper and more durable the response—likely because the tumor burden is lower, antigenic diversity is preserved, and the immune repertoire is less exhausted. Second, synergy with chemotherapy (and sometimes radiotherapy) appears to expand efficacy beyond strictly biomarkerdefined subsets, albeit at the cost of overlapping hematologic toxicity that must be managed. Future advances will depend on integrating multiomics biomarkers to identify patients who can safely omit chemotherapy, refining microenvironmentmodulating strategies for noninflamed tumors, and elucidating mechanisms of acquired resistance under prolonged immune pressure. Only then can the substantial, yet still fragile, benefits delivered by checkpoint blockade be translated into uniformly durable control of ESCC.

Clinical experience makes it increasingly clear that resistance, rather than response, is the dominant trajectory for checkpoint blockade in ESCC. A large realworld analysis spanning more than a thousand gastrointestinalcancer patients highlights the problem starkly: Barely onethird achieved an initial objective response to PD-1/PD-L1 inhibition. Almost half of those responders relapsed within two years, the vast majority with oligoprogression confined to a few lymph nodes or visceral sites[69]. This pattern is clinically telling. Localized escape is associated with far better postprogression survival than diffuse polymetastatic spread, implying that timely focal radiotherapy or surgery could meaningfully prolong the life of an otherwise effective systemic regimen. Yet it also underscores how quickly tumor-immune coevolution erodes initial gains.

Correlative studies have begun to reveal the molecular mechanisms underlying such escape. Refractory tumors that remain PD-L1 positive often upregulate alternative exhaustion checkpoints—most notably LAG-3 and the immunometabolic enzyme indoleamine 2,3-dioxygenase 1 (IDO1)—while blood-based assays have linked high baseline CXCL10 to longer survival and elevated IL-2Rα or IL-6 to early failure[70]. These findings suggest that once a tumor survives first-line PD-1 blockade, it does so not by switching off the pathway, but by building redundant inhibitory loops and inflammatory barriers. Consequently, rational combinations involving LAG-3, TIGIT, or IDO1 antagonists in combination with ongoing PD-1 therapy hold more mechanistic promise than merely substituting one PD-1 antibody for another. More broadly, early trials of nivolumab and pembrolizumab across solid tumors have confirmed that durable benefits are seen in roughly a quarter of heavily pre-treated patients, and no single baseline marker—PD-L1 included—reliably predicts who will belong to that fortunate minority[71]. Dynamic biomarkers, whether through serial chemokine measurements or on-treatment biopsies, are garnering increasing attention as a means to detect imminent failure before radiographic progression, thus opening a window for preemptive therapeutic switches.

Taken together, these clinical and translational signals converge on a pragmatic strategy: Treat early, monitor frequently, and intervene locally when progression is limited, all while layering additional immunomodulators that neutralize emergent escape nodes. Until multiomic signatures can prospectively identify durable responders, such adaptive, combinationoriented tactics remain the most realistic path to converting immunotherapy from a transient reprieve into sustained disease control for patients with ESCC.

Clinical biomarkers of immunotherapy resistance

Identifying who will, and will not, benefit from immunecheckpoint blockade in ESCC has proved far more complex than testing a single marker, yet several converging lines of evidence are beginning to outline a workable framework. Early pantumor analyses, such as KEYNOTE028, suggested that three variables—tumormutational burden (TMB), PD-L1 expression, and a Tcellinflamed geneexpression profile—each enrich for response, but only when they coincide do objective responses become truly frequent[72]. In ESCC specifically, immunohistochemical surveys confirm that PD-L1 positivity in both tumor cells and infiltrating immune cells tracks with robust CD8⁺ infiltration and, somewhat counterintuitively, with better overall survival after surgery, implying that PD-L1 may be more a surrogate of preexisting antitumor immunity than a strict brake on it[73]. This nuance is reinforced by mechanistic work showing that regression under PD-1 blockade requires CD8⁺ T cells poised at the invasive margin and held in check by the PD-1/PD-L1 axis—a setting in which PD-L1 is a marker of vulnerability rather than resistance[74].

Still, PD-L1 alone is far from sufficient. Genomic studies in lung and pancancer cohorts demonstrate that a high nonsynonymous mutation load or a copynumberadjusted TMB correlates with longer survival on checkpoint therapy, yet the cutpoints that matter vary by histology, and in ESCC copynumber instability itself can confound TMB estimates[75-77]. Epigenetic context adds another twist: Widespread DNA demethylation in highly proliferative, aneuploid tumors represses entire immunomodulatory pathways, attenuating benefit even when mutations are abundant[78]. In such cases methylation loss seems to outperform TMB as a negative predictor, highlighting the value of layered genomic and epigenomic profiling.

At the protein level, circulating angiogenic factors such as IL-8, TIE2, and hepatocyte growth factor (HGF) act as harbingers of poor outcome on checkpoint monotherapy, but they also flag patients who may benefit from adding an antiangiogenic agent, a hypothesis now supported by improved survival in combination trials[15]. Within the tumor microenvironment, co-expression patterns are equally informative: Simultaneous upregulation of PD-L1 with T-cell immunoglobulin and mucin-domain 3 (TIM-3) or TIGIT identifies patients with significantly shorter survival, likely reflecting convergent exhaustion programs on CD8⁺ tumorinfiltrating lymphocytes[79]. These observations dovetail with translational data from refractory cohorts where LAG-3 and IDO1 rise in PD-L1-positive tumors, providing a mechanistic rationale for emerging dualcheckpoint or metabolic combinations.

Neoadjuvant datasets offer a different vantage point. Sevengene signatures derived from pretreatment biopsies now predict which patients will achieve major pathological response after chemoimmunotherapy, and mesenchymal markers such as pleckstrin-2 and interferon-α-inducible protein 6 flag a stromalrich, immuneexcluded phenotype that correlates with failure of neoadjuvant PD-1 blockade[80,81]. Notably, these mesenchymal tumors show increased Treg and type-2 helper T cell infiltration with a concomitant absence of B cells and effector T cells, reiterating that cellular context outweighs any single checkpoint molecule.

Taken together, the field is moving away from binary biomarkers toward composite, contextspecific scores that integrate mutational load, immune infiltration, epigenetic state, and soluble mediators. The practical challenge is less about discovering additional markers than about harmonizing thresholds across platforms and treatment settings and, critically, validating that biomarkerguided decisions actually improve patient outcomes. Until then, clinicians must interpret any single assay with caution and favor multidimensional panels that capture the intricate balance between tumor antigenicity and microenvironmental suppression, the true fulcrum of resistance to immunotherapy in ESCC.

Clinical interventions to overcome immunotherapy resistance in ESCC

Combination immunotherapy strategies: Efforts to overcome immune resistance in ESCC often rely on PD-1 blockade. Combining it with agents that target additional immunosuppressive circuits has shown promising results. Dual checkpoint inhibition is the most mature of these strategies. Early data with the bispecific antibody cadonilimab, which targets both PD-1 and CTLA-4, show responses in roughly one-fifth of heavily pre-treated ESCC cases while maintaining a safety profile closer to singleagent therapy than to conventional ipilimumab combinations[82]. Mechanistic work helps explain why: CTLA-4 removal not only expands a progenitorlike pool of exhausted CD8⁺ T cells that can be rejuvenated by PD-1 blockade but also strips regulatory T cells of their ability to steal CD80/86 from antigenpresenting cells, thereby freeing additional PD-L1 for interception[31,83]. Similar logic underpins PD-1 + LAG-3 regimens. Transcriptomic profiling of tumors exposed to relatlimab plus nivolumab reveals that CD8⁺ T cells can sustain an “exhausted yet cytotoxic” hybrid state—marked by TOX, PR/SET domain 1, and BATF—if LAG-3 signaling is suppressed, resulting in clonal expansion and improved disease control[84]. These findings suggest that exhaustion is not an irreversible fate but a tunable equilibrium whose effector side can be weighted by multicheckpoint therapy.

Checkpoint combinations alone, however, are unlikely to suffice for the onethird of patients whose tumors lack baseline immune infiltration. Here, coupling immunotherapy to cytotoxic or targeted modalities becomes critical. Mathematical modeling and early clinical experience concur that radiotherapy given concurrently with PD-1 or in the periradiation window for CTLA-4 or TIGIT blockade maximizes antigen release, dendriticcell activation, and systemic Tcell trafficking[85]. Chemotherapy partnerships are already standard in the firstline metastatic setting, but targeted agents may push response rates even higher. Fibroblast growth-factor receptor (FGFR) inhibitors exemplify this potential: Erdafitinib kills FGFR-dependent tumor cells, broadens the Tcell receptor repertoire through necrotic antigen release, and—when combined with PD-1 blockade—narrows that repertoire into a focused, highaffinity antitumor response while simultaneously depleting TAMs[86]. Early futibatinib plus pembrolizumab data reinforce these principles in the clinic, with objective response rates climbing toward 70% when chemotherapy is also onboard.

Normalizing the tumor vasculature adds yet another layer. In the CAP02 program, camrelizumab in combination with the VEGFR2 inhibitor apatinib achieved a 35% response rate in platinumrefractory disease and even coaxed modest tumor control when reintroduced after prior checkpoint failure, hinting that VEGF blockade can reprime an immuneexcluded microenvironment[87,88]. Serum IL8, TIE2, and HGF levels—proxies for angiogenic drive—track with poorer outcomes on monotherapy and may therefore identify patients who stand to gain most from the addition of an anti-angiogenic agent.

Collectively, these vignettes argue that combination therapy should be chosen to match a tumor’s dominant escape route: Multiple checkpoint antibodies for Tcell-rich yet functionally inert tumors, radiotherapy or chemotherapy to inflame “cold” lesions, and pathwaytargeted or antiangiogenic drugs to dismantle stromal and metabolic shields. The common denominator is that each partner both relieves a discrete immunosuppressive mechanism and amplifies antigenspecific Tcell activity rather than merely adding independent cytotoxic pressure.

Beyond checkpoint pairing and pathway-directed regimens, modern immuno-oncology is expanding to include cellular engineering and microenvironmental modulation. In melanoma, circulating biomarkers—such as soluble PD-L1, cytokine profiles, and lymphocyte subsets—have shown potential in predicting responses to PD-1/PD-L1 blockade, and may ultimately inform immunotherapy stratification strategies in ESCC[89]. Likewise, next-generation CAR T-cell therapies are being designed to resist hostile metabolic conditions in solid tumors by enhancing mitochondrial function and adapting to lactic acidosis, offering a path forward in otherwise unresponsive tumors[46]. In gastric cancer, recent work underscores the value of tailoring PD-1/PD-L1-based therapies to immune subtypes, highlighting the promise of combinatorial checkpoint strategies guided by TME profiling[90]. Furthermore, disrupting small extracellular vesicle-mediated crosstalk between tumor cells and TAMs has emerged as a novel strategy to reprogram immunosuppressive macrophages and boost antitumor immunity[91].

As biomarker refinement proceeds, the field’s challenge is less about inventing new agents than about deploying existing ones in the right sequence, dose, and patient subset to convert transient tumor control into durable remission.

Novel targeted immunotherapy strategies: Efforts to outflank resistance in ESCC have begun to look well beyond the classical PD-1 axis and focus instead on nodal suppressors that either fortify the tumor microenvironment or blunt innate immune priming. One of the most compelling examples is the MDSC receptor GPR84. In both human specimens and orthotopic mouse models, GPR84 is virtually restricted to MDSCs, where it prevents lysosomal degradation of PD-L1 and thereby protects tumors from CD8⁺ mediated killing. Pharmacologic antagonism of GPR84, when layered onto PD-1 blockade, reverses that protection and restores cytotoxic Tcell activity, offering a clinically actionable means of dismantling myeloidbased resistance—an avenue that conventional checkpoint inhibitors leave untouched[17].

SMARCAL1, a DNA translocase of the SWI/SNF family, facilitates tumor immune evasion through two distinct mechanisms. First, it suppresses the cGAS-STING pathway by limiting endogenous DNA damage, thereby reducing innate immune signaling. Second, SMARCAL1 collaborates with the transcription factor JUN to maintain chromatin accessibility at the PD-L1 promoter region, promoting PD-L1 expression. Consequently, loss of SMARCAL1 Leads to increased activation of innate immune responses and decreased PD-L1 expression, enhancing the tumor's sensitivity to immune checkpoint blockade therapies. These findings suggest that SMARCAL1 represents a promising therapeutic target to overcome resistance to immunotherapy in cancers such as ESCC[92,93]. Although these data are from melanoma models, they may apply to ESCC. The mechanism appears agnostic to tissue type and may be particularly relevant to ESCC, where frequent genomic instability could be harnessed—rather than suppressed—if SMARCAL1 were inhibited. To validate this hypothesis, future studies should employ patient-derived xenograft models that better recapitulate the tumor biology of ESCC. Xenograft systems have been widely used to investigate cancer biology in vivo, including studies exploring microRNA-based therapeutic strategies in non-small cell lung cancer[94].

Checkpoint innovation is also progressing toward targets that sit upstream or parallel to PD-1. TIGIT has advanced furthest, propelled by a clear mechanistic rationale: PD-1’s intracellular tail dephosphorylates the costimulatory receptor CD226, whereas TIGIT prevents CD226 engagement altogether by monopolising its ligand, CD155[95]. Preclinical studies further reveal that Fcγreceptor engagement by antiTIGIT antibodies can reprogram TAMs and DCs, converting an exhausted CD8⁺ population into a memorylike pool capable of durable control[96,97]. Together, these observations underscore that TIGIT blockade is not merely additive but rewires both innate and adaptive compartments.

Finally, a growing body of transcriptomic and multiplexedimmunofluorescence data ties coexpression of PD-L1 with TIM3 or TIGIT on CD8⁺ tumorinfiltrating lymphocytes to particularly poor survival in ESCC, implying that these checkpoints operate cooperatively in the very cells most needed for tumor rejection[79]. That convergence argues for combination regimens that include TIM3 or TIGIT antagonists, especially in patients whose tumors display the doublepositive signature at baseline. The current generation of trials—some pairing antiTIGIT with chemotherapy, others evaluating bispecific antibodies that engage multiple receptors simultaneously—will tell whether such precision targeting can transform biological insight into consistent clinical gain. What is already evident is that breaking resistance in ESCC will require dismantling the layered suppressive architecture one node at a time, and these new targets offer the first credible blueprint for doing so.

TME modulation: Reprogramming the tumor microenvironment is emerging as a practical way to convert immunologically “cold” ESCC into responsive disease. One angle is to amplify endogenous effector cells rather than simply unleash them. Pegilodecakin is an IL-10 variant modified with polyethylene glycol to enhance its stability and delivery. It stimulates interferon-γ and granzyme B production, expands novel CD8⁺ T cell clones, and increases PD-1⁺ LAG-3⁺ populations that can be rescued by checkpoint blockade—especially when combined with anti-PD-1 therapy[98]. A second, complementary tactic is to dismantle physical and metabolic barriers erected by CAFs. Inhibition of NADPH oxidase 4 with setanaxib “normalizes” CAFs, reverses the CD8⁺cell exclusion phenotype, and restores sensitivity to PD-1 blockade[99]. Senolytic clearance of aging CAFs with venetoclax similarly reinvigorates cytotoxic T cell activity and deepens responses in otherwise refractory, stroma-rich tumors[100]. Beyond lymphocytes and fibroblasts, engineered immunocytokines such as PD-1IL2v colocalize an IL2 variant to PD-1⁺ T cells, fostering a stem-like CD8⁺ pool and simultaneously re-educating TAMs; when paired with anti-PD-L1, this approach has achieved durable regressions in models that resist conventional checkpoints[101]. Taken together, these studies illustrate that microenvironmental intervention is not a single maneuver but a spectrum of strategies—cytokine amplification, stromal normalization, and myeloid reprogramming—that, when integrated with existing checkpoints, can erode the layered defenses that underpin immunotherapy resistance in ESCC.

Personalized immunotherapy: Personalizing immunotherapy for ESCC is moving from singlemarker triage toward composite genomic and epigenomic stratification. A notable example is the JUPITER-06 cohort, where whole-exome sequencing of 486 tumors was used to integrate refined tumor mutational burden with HLA diversity and key driver mutations like PIK3CA and TET2. This led to the development of the Esophageal Genome Immuno-Oncology Classification, which stratifies patients into three groups. Importantly, only those with immunogenic features or lacking high-risk oncogenic alterations benefited from chemoimmunotherapy[77]. In other words, the therapeutic dividend depends on the balance between antigenicity and oncogenic immune evasion, rather than on mutational load alone. Another caution comes from epigenetics: Genomewide hypomethylation, especially in latereplicating domains that harbour many immunomodulatory genes, can transcriptionally silence antigenpresentation pathways and blunt checkpoint efficacy even in tumors with abundant mutations[78]. Across multiple data sets, loss of methylation outperformed raw mutation count in predicting resistance, underscoring that “how the genome is packaged” may outweigh “how much it is mutated”. Together, these findings argue that truly individualized immunotherapy for ESCC will require multilayered profiling—combining refined mutational indices, drivergene status, and methylome context—to identify patients whose tumors are both visible to the immune system and not hardwired for immune escape.

CONCLUSION

Despite the expanding application of ICIs in ESCC, several obstacles limit efficacy. From the immunoediting perspective, the tumor undergoes stages of elimination, equilibrium, and escape, culminating in a highly heterogeneous microenvironment. As a result, therapies targeting a single molecule often fail to address the myriad resistance mechanisms at play. Although “counter-immunoediting therapy” proposes an individualized approach based on immunoediting stages, practical implementation remains challenging: Accurately identifying each patient’s immune phase and deploying multiple “normalization” interventions in a coordinated manner is complex and resource-intensive[102]. Moreover, current immunotherapeutic approaches—especially immune checkpoint blockade—exhibit considerable variability in response across patient subgroups. Their efficacy is often limited to individuals with high PD-L1 expression or pre-existing immune infiltration, while patients with “immune-cold” tumors derive minimal benefit. Additionally, immune-related adverse events such as pneumonitis, colitis, or endocrinopathies can compromise tolerability, particularly in older adults or those with autoimmune predispositions. These limitations underscore the need to better define the indications, stratify responders, and manage toxicity profiles in clinical practice. Another critical challenge arises from the biophysical properties of the TME, such as matrix stiffness, interstitial fluid pressure, solid stress, and vascular abnormalities. These factors limit immune cell infiltration into tumor tissue and are not easily reversed by conventional immunotherapies alone[103]. Although emerging strategies—such as physical methods to disrupt the extracellular matrix or normalize fluid dynamics—offer promise, translating these into safe, clinically viable protocols requires deeper mechanistic insights and validation in large-scale trials. In parallel, the quest for reliable predictive biomarkers illustrates both progress and complexity. While PD-L1 expression, TMB, and immune cell infiltration can each hint at potential response, none fully captures the intricacies of ESCC’s immune landscape[104,105]. This shortfall has driven the development of multidimensional frameworks—such as the Esophageal Genome Immuno-Oncology Classification—that integrate genomic, epigenetic, and immunological indicators. Although such systems show promise for guiding patient-specific therapies, their clinical implementation is often hampered by high costs, assay variability, and limited availability in routine settings. Ensuring the clinical utility of these biomarkers will require not only biological validation but also improvements in affordability, standardization, and platform compatibility across institutions. Incorporating emerging biotechnologies into ESCC immunotherapy research could help overcome current limitations. For example, CRISPR/Cas9 gene editing can be used to modify T cells, enhancing their cytotoxicity, persistence, and resistance to exhaustion, thereby supporting personalized adoptive therapies[106]. In addition, nanomedicine platforms allow targeted delivery of agents such as checkpoint inhibitors, STING agonists, or RNA-based drugs into the TME, improving therapeutic precision while reducing systemic toxicity. Although these technologies remain at the preclinical stage in ESCC, combining them with spatially resolved immune profiling may enable modulation of hard-to-target tumor components and broaden the clinical benefit of immunotherapy[107]. Moving forward, surmounting these clinical challenges will likely require a robust, multidisciplinary collaboration. Incorporating molecular, epigenetic, and microenvironmental data sets within unified treatment algorithms can inform combination therapies tailored to each patient’s unique tumor biology. Moreover, large-scale prospective trials and real-world evidence are essential to validate these integrative models. It is equally important that future studies include multi-center, multi-ethnic patient cohorts to enhance the generalizability and global relevance of findings. To improve translational relevance, future studies should adopt preclinical tools such as patient-derived xenografts, spatial transcriptomics, and multiplex immune phenotyping to assess treatment efficacy in biologically representative systems. In parallel, detailed experimental design frameworks—defining relevant models, patient stratification strategies, and evaluation endpoints—should be embedded in early-phase studies to guide clinical translation. Only by addressing the multi-layered nature of ESCC immune resistance—from biophysical barriers to complex biomarker profiles—can immunotherapy be refined to deliver durable survival benefits for the broader patient population. Future studies should test integrated biomarker-guided and microenvironment-modulating regimens in prospective trials.

ACKNOWLEDGEMENTS

We express our gratitude to Dr. Tan for assisting in the preparation of the manuscript.