Integrating Three Functional Subsets of T Cell Immunity Into an Integrative Model for Cancer Immunotherapy

Abstract

Current cancer immunotherapy relies heavily on tumor-Ag specific T cells (TASTs). While checkpoint blockade has redefined the therapeutic landscape of oncology, this single-mechanism strategy shows limitations from stochastic de novo priming and terminal exhaustion. High-dimensional single-cell data reveal the tumor microenvironment as a heterogeneous immune ecosystem where virus-specific T cells (VSTs) and bystander T cells often predominate over TASTs. We propose an integrative model built on three functional subsets: 1) classical TASTs; 2) VSTs acting as TASTs via viral etiology or molecular mimicry; and 3) bystander T cells representing a tumor-independent compartment. Characterizing these subsets by ontogeny and transcriptional programs suggests their potential utility as predictive biomarkers for checkpoint inhibitor responses and as distinct platforms for adoptive cell transfer strategies. This explains resistance mechanisms in immunologically cold tumors and guides mechanistically distinct therapeutic approaches—from classical priming to in situ viral activation and off-the-shelf cellular products.

INTRODUCTION

Cancer immunotherapy operates on a foundational principle: T cells detect and eliminate malignant cells through TCR recognition of tumor-derived peptides on MHC molecules (1). This concept has driven immunology research for decades. Clinical translation through immune checkpoint inhibitors (ICIs) targeting PD-1 and CTLA-4 has substantially altered treatment for high mutational burden tumors, including melanoma and non-small-cell lung cancer (2,3). These interventions assume the primary anti-tumor effectors are TASTs recognizing tumor-associated Ags (TAAs) or neoantigens from somatic mutations.

This classical perspective, while validated in specific contexts, represents an incomplete view. Although effective in select populations, it depends on a single mechanism challenged by limitations. TASTs targeting TAAs are compromised by central tolerance, leaving low-affinity clones (4,5). TASTs targeting neoantigens face constraints from rare TCR precursors, rapid terminal exhaustion (Tex-term), and tumor immunoediting (6,7). Only a minority mount durable TAST responses, leaving most tumors “cold” despite lymphocytic infiltration (8).

High-dimensional single-cell technologies have reshaped understanding of tumor-infiltrating lymphocytes (TILs) (9,10). These datasets show classical TASTs often constitute a minority, outnumbered by Ag-experienced memory T cells from prior viral exposures (11,12,13). Many exhibit cytotoxic capacity, maintained responsiveness, or tissue-resident features—yet their functional roles remain inadequately integrated into anti-tumor immunity models (14,15).

We propose a three-subset integrative model categorizing T cells by ontogeny and reactivity mode:

• 1) Classical TASTs: de novo primed cells targeting TAAs or neoantigens, constrained by tolerance or exhaustion susceptibility.

• 2) Cross-reactive VSTs: viral-memory cells with potential for repurposed for tumor recognition through established viral etiology or hypothesized molecular mimicry, which may leveraging memory inflation for sustained function (16).

• 3) Bystander memory T cells: tumor-independent viral-memory cells lacking direct tumor Ag recognition but possessing latent capacity for tumor microenvironment (TME) remodeling through cytokine-driven activation (17).

This accommodates TIL diversity observed in high-resolution studies and illuminates tumor immune control mechanisms outside classical neoantigen recognition (18). It provides a mechanistic foundation for next-generation strategies that reinforce, recruit, or repurpose each subset according to patient-specific contexts. By moving beyond single-mechanism models toward comprehensive T-cell ecology, this establishes a conceptual blueprint for precise, durable, and broadly effective cancer immunotherapies (Fig. 1).

Figure 1. Functional and molecular characteristics of tumor-infiltrating T cell subsets. Three functionally distinct T cell subsets in tumors. Tumor-Ag specific T cells (left) express exhaustion markers (CXCR13⁺, CD39⁺, NR4A^hi, TOX^hi, TCF-1^low) with MHC-I and PD-L1/PD-L2 recognition via TCR and PD-1. Virus-Ag specific T cells (right) display effector memory phenotype (CD39⁺, CX3CR1⁺, KLRG1⁺) with cytotoxic function (granzyme B, perforin). Bystander T cells (bottom) adopt memory-like state, responding to proinflammatory cytokines (IL-12, IL-15, IL-18) and type I IFN.

Importantly, these modes are not mutually exclusive. Individual tumors typically contain mixtures of all three subsets in varying proportions, and a single patient may concurrently harbor exhausted neoantigen-specific TASTs (mode I), human papillomavirus (HPV)-reactive T cells (mode II), and cytomegalovirus (CMV)-specific bystander memory T cells (mode III). The value of this model lies in distinguishing functional mechanisms rather than imposing rigid categorical boundaries.

THE CURRENT TAST-CENTRIC MODEL AND ITS CONSTRAINTS

Cancer immunotherapy principles lie in reinvigorating pre-existing endogenous TASTs (19). This approach has proven valuable but faces fundamental limitations. The central premise assumes spontaneous de novo priming against tumor-unique Ags falling into two categories with distinct constraints.

TAAs such as MAGE family, NY-ESO-1, or carcinoembryonic Ag represent self-proteins aberrantly expressed by cancer cells (20). Because these are fundamentally “self” Ags, high-affinity clones are eliminated during thymic development through central tolerance (21). Surviving clones undergo peripheral tolerance inducing anergy or deletion (22). The TAA-specific repertoire comprises low-avidity clones with limited killing capacity, constraining immune responses even with checkpoint inhibition (23).

Neoantigens from somatic mutations offer mechanistically superior targets as “non-self,” bypassing central tolerance (24). However, neoantigen-specific response generation represents a stochastic event dependent on random intersection of mutation with appropriate TCR in the naive repertoire. This probabilistic process is influenced by HLA haplotype; an immunogenic epitope in one patient may fail MHC binding in another (25). In low tumor mutational burden (TMB) tumors—pancreatic adenocarcinoma, microsatellite stable colorectal cancer—this intersection may not occur, yielding tumors devoid of functional TASTs (26). This quantitative deficit drives primary checkpoint blockade resistance with insufficient therapeutic substrate (27,28).

Even with successful de novo TAST initiation, progressive functional decline occurs through T cell exhaustion (29). Upon TME entry, TASTs encounter chronic Ag stimulation with immunosuppressive signals including TGF-β, IL-10, and metabolic starvation (30,31). This drives progressive differentiation with sequential effector function loss. While early progenitor exhausted T cells (Tpex) retain stem-like proliferative properties, express T cell factor-1 (TCF-1), and demonstrate PD-1 blockade responsiveness (32), continuous stimulation drives Tex-term.

Tex-term is not transient dysfunction but a distinct lineage with extensive epigenetic remodeling (33). Assay for Transposase-Accessible Chromatin using sequencing reveals terminally exhausted cells acquire fixed alterations at genomic loci controlled by TOX and NR4A transcription factors (34). Once established, cells lose proliferative and cytotoxic capacity and become refractory to reinvigoration (35). In patients with chronic tumors, TASTs may exist physically but remain functionally irretrievable, leading to acquired checkpoint blockade resistance (36).

The TAST-centric model is vulnerable to tumor evasion through immunoediting (37). Under TAST response pressure, tumor cells losing Ag expression or downregulating MHC class I gain survival advantage (38). This “Ag escape” represents a common relapse mechanism. Somatic mutations in beta-2 microglobulin or JAK1/JAK2 pathways lead to complete MHC class I loss or IFN-γ insensitivity, rendering tumors invisible to CD8⁺ T cells (39). Because existing treatment strategies rely almost exclusively on TCR-mediated recognition, such evasion nullifies the therapeutic strategy. Single-mechanism reliance creates concentrated vulnerability, suggesting need for broader approaches incorporating alternative recognition and activation mechanisms.

THE THREE-SUBSET INTEGRATIVE ARCHITECTURE

Constructing a comprehensive therapeutic model requires considering T cell populations beyond de novo tumor specificity (Fig. 2). The TME is populated by diverse T cells, many representing bystanders with specificities for EBV, CMV, and influenza viruses (11,13). High-resolution studies using single-cell transcriptomics and mass cytometry reveal these exist in complex functional states rather than “immunological noise.” Classifying the intratumoral T-cell landscape into three populations by developmental ontogeny and functional reactivity integrates the host's complete immunological history (Table 1).

Figure 2. Comparison of single-subset versus three-functional subset models of tumor immunity. Single-subset model (left) demonstrates naive T cells undergoing stochastic de novo priming to tumor neoantigens, expanding as TASTs, then succumbing to exhaustion (PD-1⁺, LAG-3⁺, TIM-3⁺) or Ag escape. Three-functional subset model (right) proposes three distinct populations: subset I (classical exhausted TASTs from naive origin), subset II (cross-reactive VSTs with effector memory phenotype: CX3CR1⁺, KLRG1⁺, granzyme B^hi), and subset III (bystander tissue-resident memory VSTs: CD103⁺, CD69⁺). Subset III promotes TME remodeling via proinflammatory cytokines (IL-15, IL-12, IL-18), type I IFN (IFN-γ, TNF-α), increased MHC-I expression, and immune cell recruitment (macrophages, NK cells).

Table 1. Comparative characteristics of three tumor-infiltrating T cell subsets.

	Mode I: classical TAST	Mode II: cross-reactive VSTs	Mode III: bystander VSTs
Definition	Tumor-Ag-specific T cell	Virus-specific T cells repurposed to function as TASTs	Viral-memory T cells representing a tumor-independent compartment
Origin	Naïve T cell precursors	Systemic viral memory pool	Viral memory cells
Ag recognition	TCR-mediated recognition of tumor neoantigens or TAAs	TCR-mediated recognition of tumor Ags via viral etiology or molecular mimicry	None (TCR-independent)
Key phenotype	CD39+, PD-1, TIM-3, LAG-3, CXCL13	CD39+, CX3CR1, KLRG1	CD39−, tissue residency (CD103, CD69)
Functional state	Tex-term	Effector-memory	Dormant/tissue resident
Primary activation mechanism	TCR-mediated signaling	TCR-mediated signaling	Innately driven
Effector function	Direct cytotoxicity (limited by exhaustion depth)	Direct cytotoxicity (high capacity)	TME remodeling, bystander killing

Mode I: the classical TAST

Classical TASTs represent cells undergoing de novo priming against tumor Ags (40). These originate from naive precursors primed against private neoantigens or TAAs following tumorigenesis, typically in tumor-draining lymph nodes (41,42). Developmental ontogeny influences functional trajectory: deriving from naive precursors encountering chronic Ag within tumor context confers particular exhaustion susceptibility (43).

Functionally, classical TASTs demonstrate characteristic multidimensional phenotype driven by chronic TCR stimulation (44). They express high CD39 ectonucleotidase alongside inhibitory receptors (Table 1) (11,44). Single-cell transcriptomic analyses identify chemokine CXCL13 as particularly informative (45). CXCL13, typically associated with B-cell recruitment in germinal centers, is upregulated by CD8⁺ TASTs, suggesting roles in orchestrating tertiary lymphoid structure (TLS) formation within tumors (46,47). CD39⁺PD-1^hiCXCL13⁺ cell presence emerged as a biomarker for spontaneous anti-tumor responses (48). This phenotype links chronic Ag engagement—often accompanied by exhaustion programs—to B cell recruitment and potential intratumoral niche formation supporting ongoing responses.

However, despite central roles in current immunotherapy, Mode I reliance faces substantial challenges. Cell generation requires sufficiently high mutational threshold, and once generated, they demonstrate marked propensity for Tex-term-associated epigenetic modifications (49,50). These constraints limit patient fractions benefiting from classical checkpoint blockade.

Mode II: the cross-reactive VST

Mode II comprises two distinct scenarios with differing levels of evidence. In virus-associated cancers (e.g., HPV⁺ cervical/head-neck, EBV⁺ nasopharyngeal, HBV⁺ hepatocellular), viral Ags are well-established drive authentic tumor-specific responses. In contrast, emerging hypothesis suggests molecular mimicry-driven responses in non-viral tumors, for which prevalence and contribution remain to be systematically defined. Cross-reactive VSTs represent convergence of pathogen immunity and cancer surveillance (51,52).

These are pre-existing viral memory origin cells repurposed to recognize tumor cells. Tumor recognition occurs through two primary mechanisms.

In virus-associated malignancies—EBV-driven nasopharyngeal carcinoma, HPV-driven head and neck cancers, Merkel cell polyomavirus-driven Merkel cell carcinoma—viral epitopes are constitutively expressed by tumor cells (53,54,55). VSTs naturally function as tumor-specific cells targeting foreign viral Ags on cancer cells.

For cancers without viral etiology, molecular mimicry or “heterologous immunity” becomes relevant (56). T cell cross-reactivity allows individual TCRs to recognize multiple structurally related pMHC complexes (57). This raises the possibility that cells initially primed against viral peptides (e.g., CMV pp65) may recognize structurally homologous tumor-expressed neoantigens (58,59,60). Unlike classical TASTs, these originate from systemic memory pools (central memory or effector memory) and inherit “memory inflation” properties, particularly well-characterized for CMV where specific clones undergo progressive expansion and long-term persistence (61,62).

Cross-reactive VSTs often express CD39 from tumor Ag engagement but retain high cytotoxic potential markers including CX3CR1, KLRG1, and elevated granzyme B. Their unique effector-memory developmental heritage may render them more resilient to exhaustion and apoptosis compared to naive-derived cells, potentially functioning as more durable effectors with superior longevity and killing capacity (55). This population suggests infection history substantially shapes anti-tumor immune potential, providing a pre-existing cellular reservoir for therapeutic leverage.

Mode III: the bystander VST

Bystander VSTs represent a “latent reservoir” of immune potential. These are pre-existing viral memory cells specific for CMV, influenza, or EBV that infiltrate tumors via inflammation-associated chemokine gradients including CXCL9 and CXCL10, yet cognate viral Ags are absent within tumors (63). Previously viewed as irrelevant, they are now understood to acquire a unique, TME-conditioned bystander state.

While these cells do not engage tumors via TCR, they are functionally active (64). They function through TCR-independent activation driven by inflammatory cytokine milieu. The TME environment, enriched for IL-15, IL-12, IL-18, and type I IFNs, triggers these memory cells to upregulate granzyme B and produce substantial IFN-γ and TNF-α through JAK-STAT pathways (65). This cytokine production substantially remodels the TME, increasing MHC class I expression on adjacent tumor cells (potentially counteracting immunoediting) and recruiting NK cells and macrophages (66).

Interestingly, while bystander T cells often express PD-1, this signifies tissue residency or recent cytokine activation rather than terminal dysfunction (67). Many display tissue-resident memory phenotype characterized by CD103 and CD69 expression, suggesting they represent pre-existing tissue surveillance networks co-opted by tumor environment (11,44,68). These cells may represent dormant cellular resources that, if appropriately triggered, could substantially alter tumor immune dynamics.

HIGH-DIMENSIONAL DECONVOLUTION OF T CELL ONTOGENY

Achieving therapeutic precision requires accurate identification and quantification of these three subsets within patient specimens. Traditional functional assays based on IFN-γ release or bulk RNA sequencing prove insufficient because markers like PD-1 and IFN-γ are expressed by all three groups (69). A combinatorial biomarker strategy integrating mass cytometry, single-cell RNA sequencing with paired TCR sequencing, and spatial transcriptomics is needed to deconvolute the TME by functional states and developmental ontogeny.

CD39 as a primary activity discriminator

The primary discriminator for T cell activity status is ectonucleotidase CD39 (ENTPD1), serving as robust surrogate for chronic Ag stimulation (70). Studies in lung and colorectal cancer demonstrate CD39 is highly expressed on CD8⁺ TILs undergoing repeated cognate Ag stimulation. Mechanistically, CD39 expression is driven by chronic TCR signaling transcription factors, particularly NFAT and AP-1 (11). High CD39 identifies actively signaling T cells against tumor Ags, encompassing classical TASTs (mode I) and cross-reactive VSTs (mode II). In Merkel cell carcinoma, MCPyV-specific CD8⁺ T cells displayed distinct CD39⁺CLA⁺ and CD39⁺CD103⁺ phenotypes tracking with clinical outcomes, illustrating how CD39-based stratification resolves functionally meaningful tumor-specific subsets (55).

CD39⁺ TILs are functionally distinct from CD39⁻ population, substantially enriched for bystander specificities (11,44). CD39 absence serves as the primary mode III hallmark. While bystander cells may express PD-1 (likely driven by TGF-β signaling or innate cytokine activation), CD39 absence confirms they are not chronically Ag-engaged (71). This binary classification allows functional TIL compartment stratification ex vivo without requiring prior tumor Ag knowledge.

Nevertheless, for molecular mimicry–driven mode II responses in non-viral tumors and for mode III bystander populations, current marker associations remain largely inferential, suggested primarily by viral immunology rather than validated by systematic Ag-specific profiling within tumors. Future studies integrating single-cell TCR sequencing with functional epitope mapping will be required to rigorously test and refine these proposed marker signatures.

Distinguishing ontogeny: separating mode I from mode II

Distinguishing Ag-engaged populations—separating de novo TASTs of mode I from cross-reactive VSTs of mode II—requires interrogating T-cell developmental ontogeny. Definitive identification is achieved through single-cell TCR sequencing coupled with epitope mapping. Sequencing CD39⁺ TIL TCR repertoires and comparing against comprehensive viral epitope databases such as VDJdb or McPAS-TCR identifies clones with clear viral origins (72,73).

Without TCR sequencing capabilities, phenotypic surrogate markers provide valuable insight. Mode I cells from naive precursors typically lack CX3CR1 and KLRG1, reflecting differentiation trajectory arrested by exhaustion program. Mode II cells from memory populations retain these effector-memory markers alongside elevated T-bet and granzyme B (74). Additional markers including CD27 and CD28 expression patterns provide further discrimination, as memory-derived cells show differential expression versus exhausted naive-origin cells (75).

Spatial profiling and functional cartography

Beyond cell-surface phenotyping and transcriptional profiling, spatial analysis technologies reveal critical localization pattern differences. Spatial transcriptomics approaches suggest terminally exhausted TASTs preferentially localize deep within tumor cell nests, maintaining direct physical contact with malignant cells (76). Bystander T cells are often spatially excluded from tumor core and found within stroma or tertiary lymphoid structures (77).

This “functional cartography” maps not merely presence and phenotype but precise spatial localization and interaction networks within tumor architecture (77). Such spatial information provides critical context for understanding how different subsets influence tumor biology and respond to therapeutic interventions. Integrating phenotypic, transcriptional, and spatial analytical layers enables comprehensive diagnostic models assessing not merely abundance but functional competence, ontogenetic provenance, and anticipated therapeutic responsiveness.

CLINICAL IMPLICATIONS: PREDICTING ICI RESPONSE AND DESIGNING ADOPTIVE CELL TRANSFER (ACT) STRATEGIES

Uniform immunotherapy strategies applied to all patients overlook distinct architectural features of individual immune responses. Rather than relying on abstract classifications, the three-subset model is best understood through its practical clinical implications: 1) predicting ICI outcomes based on subset composition and 2) designing mechanistically distinct ACT products. This approach transitions oncology from empirical approaches toward rational therapeutic design where treatment selection is guided by specific structural characteristics and vulnerabilities of patient T cell infiltrate (Fig. 3).

Figure 3. Diagnostic stratification algorithm and potential intervention strategies. Patient biopsies can be stratified by CD39 status. CD39-positive cases undergo viral homology/TCR assessment, identifying mode I (classical TAST) or mode II (cross-reactive VST). CD39-negative cases indicate mode III (bystander VST). Mode I (“hot but exhausted,” high TMB) receives anti-PD-1 monotherapy (progenitor TCF1⁺ cells) or combination therapy with anti-LAG3/TIGIT (terminal TOX^hi TCF1^– cells). Mode II (“weak de novo, strong viral memory”) employs mimicry vaccination with viral peptides to amplify systemic viral memory. Mode III (“cold/ignorant,” abundant CD39^–PD-1⁺ bystander T cells) uses ACT with VST infusion plus in situ viral vaccination via intratumoral injection for TME remodeling.

Mode I-dominant tumors: ICI responsiveness and reinforcement strategies

Patients with mode I-dominant responses are characterized by high TMB and substantial exhausted cells (Table 1) (78). This phenotype predicts ICI responsiveness because PD-1 blockade prevents Tpex transition into terminally exhausted states, sustaining proliferative and functional capacity of tumor-specific responses (79). The presence of TCF-1⁺ progenitor exhausted cells within the CD39⁺ compartment serves as a critical predictive biomarker for durable checkpoint blockade benefit (32).

However, efficacy depends critically on exhaustion state reversibility. If biomarker profiling indicates preponderance of terminally exhausted cells characterized by TOX^hiTCF1⁻ phenotype, standard checkpoint inhibitors yield diminishing returns (32). In such cases, combination therapies targeting alternative inhibitory checkpoints (LAG-3, TIGIT) may provide additional benefit (80).

For mode I-dominant tumors, ACT strategies focus on expanding and reinvigorating classical TASTs. TIL therapy involves ex vivo expansion of tumor-reactive lymphocytes followed by reinfusion after lymphodepletion. Recent advances include TCR-engineered T cells targeting defined neoantigens identified through whole-exome sequencing and immunopeptidomics (81).

The key challenge remains manufacturing scalability and identifying targetable neoantigens. Personalized neoantigen vaccines combined with ACT represent an emerging approach, priming and expanding neoantigen-specific responses before cellular transfer (81). Metabolic interventions aimed at revitalizing mitochondrial function in exhausted cells—glutamine supplementation, pyruvate delivery, or glycolytic enzyme inhibition—represent complementary strategies for enhancing ACT product quality (82).

Therapeutic strategies inducing or enhancing TLS formation represent intriguing possibilities for sustaining mode I responses. CXCL13 delivery through intratumoral injection or viral vector-mediated expression could create intratumoral niches supporting ongoing T cell priming, B-cell recruitment, and sustained responses (83). Such approaches might prove valuable where spontaneous TLS formation has not occurred despite TAST presence.

Mode II enhancement: leveraging molecular mimicry for ICI and ACT

Mode II-dominant responses occur in virus-associated malignancies or tumors with structural mimicry between neoantigens and viral epitopes. These patients may show paradoxical ICI responses: CD39⁺ TILs are present, but they derive from memory rather than naive origins. The critical predictive distinction lies in phenotypic profiling: CD39⁺CX3CR1⁺KLRG1⁺ cells suggest mode II dominance with potentially greater ICI durability due to memory-derived resilience against Tex-term (74).

For low TMB tumors where mode I is weak or absent, mode II may represent the primary mechanism of spontaneous anti-tumor immunity. Identifying cross-reactive VST clones through TCR sequencing or bioinformatic structural modeling predicts which patients harbor latent anti-tumor potential recruitable through heterologous vaccination strategies (84).

Mode II enables new treatment options in ACT: off-the-shelf banked VSTs from healthy donors matched by HLA. Unlike personalized TIL or TCR-engineered products requiring patient-specific manufacturing, VST banks allow immediate product availability. For virus-associated malignancies, this is straightforward—EBV-specific or CMV-specific T cells directly target tumors expressing viral Ags (85,86).

For non-viral cancers, heterologous prime-boost vaccination becomes viable. This exploits “Original Antigenic Sin” for therapeutic benefit: vaccinating with viral peptides (structural mimics) bypasses kinetic bottlenecks of priming rare naive T cells against neoantigens. Instead, it amplifies robust systemic viral memory pools, recruiting high-affinity effector cells to tumor sites where they recognize neoantigens as structural variants (87).

Because memory-derived cells are epigenetically distinct from exhausted naive-origin cells, they mount rapid, high-magnitude cytotoxic responses with reduced TME suppression susceptibility. This represents a conceptual shift from personalized neoantigen vaccines—facing manufacturing time, cost, and immunogenicity challenges—toward off-the-shelf viral peptide vaccines rationally matched to patient viral exposure history and tumor mutational landscape (88).

Practical implementation requires comprehensive databases linking TCR sequences, viral epitopes, and tumor neoantigens, plus structural modeling tools predicting cross-reactive recognition. Initial validation studies might focus on tumors where viral mimicry has been documented (e.g., specific KRAS mutations mimicking bacterial peptides) before broader applications.

Mode III activation: repurposing bystanders to overcome ICI resistance

Mode III-dominant tumors represent immunologically “cold” phenotypes lacking substantial mode I and II responses yet harboring abundant CD39⁻PD-1⁺ TILs (89). This paradoxical phenotype—PD-1 expression without CD39—explains a critical ICI resistance mechanism: PD-1 in bystanders reflects tissue residency or cytokine activation rather than exhaustion. Checkpoint blockade fails because there are insufficient tumor-reactive cells to reinvigorate; the abundant PD-1⁺ population is functionally dormant, not exhausted (11,44).

This mechanistic insight explains why PD-1 expression alone poorly predicts ICI response. Comprehensive profiling distinguishing tumor-reactive exhausted from bystander resident populations improves ICI response prediction and identifies patients requiring alternative strategies beyond checkpoint blockade.

Mode III offers a powerful ACT alternative: repurposing bystander VSTs as programmable biological tools for TME remodeling. The therapeutic rationale converts abundant tumor-resident bystanders from passive observers into active immunity participants through in situ activation.

The strategy involves infusing off-the-shelf banked VSTs (e.g., HLA-matched CMV-specific cells from healthy donors) followed by intratumoral cognate viral peptide injection (90,91). This approach decouples T cells from tumor Ag recognition requirements, using them as cytokine factories to remodel the microenvironment and recruit endogenous responses.

Infused VSTs, encountering cognate Ag within tumors, undergo local activation and proliferation, producing inflammatory cytokines reshaping immune landscape (90). This can overcome traditional TIL or CAR-T limitations including tumor Ag expression requirements and patient-specific cell generation challenges. Released IFN-γ polarizes tumor-associated macrophages from immunosuppressive M2 toward pro-inflammatory M1 phenotype, fundamentally altering TME from immunosuppressive to immunostimulatory.

Furthermore, this activation induces “bystander killing” mechanisms where activated T cells eliminate adjacent Ag-negative tumor cells through death receptor-mediated pathways including Fas/FasL or TRAIL-mediated apoptosis. Such mechanisms offer potential solutions to Ag escape as they don't require specific tumor Ag recognition. This may shift immunologically cold TME to a more inflamed, immunostimulatory state without requiring tumor-specific T cells.

Intratumoral immunodominant viral peptide or stimulator of IFN genes agonist administration induces Ag-specific reawakening of dormant populations (92). This initiates local inflammatory cascades with activated bystanders releasing substantial IFN-γ. Combined with checkpoint blockade, this two-step approach—first heating cold tumors via bystander activation, then sustaining responses via PD-1 blockade—represents rational sequential combination therapy.

Safety considerations

Intratumoral administration of viral peptides to elicit bystander VST activation may be associated with safety liabilities, largely attributable to non-specific, cytokine-mediated immune activation (93). Potential concerns include cytokine release syndrome from IFN-γ, TNF-α, fand IL-12 surges; off-target inflammation if peptides drain to systemic circulation; and autoimmune activation if mimicry epitopes cross-react with self-Ags. While these risks have not yet been fully defined, they highlight the need for careful therapeutic design and monitoring. Several mitigation strategies merit further investigation. These include restricting peptide exposure to the tumor bed using hydrogel-based slow-release formulations, fractionating doses to blunt acute cytokine surges, and applying predictive safety measures such as baseline cytokine profiling with early monitoring of IL-6 and C-reactive protein. In addition, incorporating controllable safety elements—such as inducible suicide genes in transferred VSTs or reversible small-molecule adjuvants—may allow immune activation to be modulated if toxicity emerges.

Additionally, how bystander populations interface with Tregs within TME remains important: Do activated bystanders compete with tumor-specific cells for metabolic resources like glucose, glutamine, and IL-2? (94). Could excessive bystander activation dampen de novo tumor-specific responses in resource-limited microenvironments? These competitive dynamics and resource allocation questions remain poorly understood but could prove critical for rational combination therapy design.

UNRESOLVED QUESTIONS AND FUTURE DIRECTIONS

T cell immunity in cancer is evolving beyond an exclusive emphasis on TASTs. Integrative single-cell studies point to a three-subset model that organizes intratumoral T cells by developmental origin and effector potential: classical TASTs, cross-reactive VSTs, and bystander VSTs. This conceptual structure helps interpret heterogeneous immunotherapy outcomes and informs the design of future treatment strategies. Despite this progress, important gaps remain that will be critical to resolve for clinical implementation.

Breadth of bystander specificities

The full breadth of bystander specificities beyond common herpesviruses remains incompletely characterized (11,12). Whether T cells specific for commensal or other environmental Ags perform comparable TME roles needs establishment (95). Recent evidence suggests gut microbiome influences immunotherapy responses, potentially through cross-reactive T cell mechanisms (96). Comprehensive bystander repertoire mapping—including bacterial, fungal, and other non-viral pathogen specificities—represents an important investigation priority.

Validation of molecular mimicry

Molecular mimicry extent and functional significance in human tumors requires rigorous validation. While cross-reactivity between viral epitopes and tumor neoantigens has been demonstrated in isolated cases, comprehensive understanding of natural prevalence and contribution to spontaneous anti-tumor immunity remains absent (97). High-throughput TCR screening technologies applied to comprehensive pathogen-derived and tumor-derived peptide libraries are urgently needed to establish true mode II response frequency and therapeutic potential. Advanced computational tools are required for accurate structural mimicry prediction from peptide-MHC binding properties and TCR interaction profiles. The clinical relevance of molecular mimicry in non-viral solid tumors remains largely conceptual. Future studies must quantify mimicry-driven T cell frequency, correlate with clinical outcomes, and validate epitope cross-reactivity experimentally.

Integration with other immune populations

The TME contains numerous other immune cell types interacting with these T cell subsets. CD4⁺ T helper cells, B cells within TLS, NK cells, and various myeloid populations contribute to anti-tumor immunity or immunosuppression. Understanding how three T cell modes interact with these populations will be essential for comprehensive strategies. Activated bystander T cells may recruit NK cells through chemokine production or influence B cell responses through cytokine signaling (98).

Clinical implementation challenges

Practical clinical implementation challenges must be addressed. The proposed diagnostic algorithm (Fig. 3) represents an idealized framework requiring technical advances for clinical deployment. High-dimensional profiling technologies including mass cytometry and single-cell sequencing remain expensive and technically demanding, limiting routine clinical deployment. Single-cell TCR sequencing with epitope mapping, while conceptually definitive, remains impractical for routine use. Simplified, cost-effective diagnostic platforms are needed for reliably stratifying patients by subset architecture. Standardized biomarker panel and reference dataset development will be essential for translating research tools to clinical diagnostics.

CONCLUDING REMARKS

The therapeutic landscape of cancer immunotherapy stands to benefit from this expanded integrative model. Moving beyond single-mechanism tumor-Ag specific immunity, this review delineates how viral memory T cells—whether functioning as tumor-reactive cells through mimicry or as bystander cells capable of TME remodeling—contribute to complex TIL ecology.

The ultimate goal is developing intelligent, sequential combination therapies leveraging entire immune infiltrates within individual tumors. Future clinical workflows may incorporate comprehensive tumor specimen profiling to determine 'Subset Status'—relative abundance and functional state of classical TASTs, cross-reactive VSTs, and bystander populations. Based on functional mapping, clinicians could select precise intervention sequences: perhaps in situ viral vaccination activating cold tumors dominated by bystander cells, followed by heterologous prime-boost vaccination recruiting and expanding cross-reactive memory populations, finally consolidated with checkpoint blockade sustaining responses by preventing tumor-reactive cell exhaustion.

Such precision immunotherapy approaches would represent substantial advances over current empirical treatment strategies. Leveraging all three T-cell subsets has potential to broaden immunotherapy therapeutic reach, including to patients with treatment-refractory cold tumors. This requires continued investigation into mechanisms governing each mode, sophisticated diagnostic tool development for patient stratification, and carefully designed clinical trials testing subset-specific therapeutic interventions.

The path forward requires integrating multiple disciplines: single-cell biology for comprehensive cellular characterization, structural immunology for understanding cross-reactivity, computational modeling for predicting mimicry events, and clinical oncology for translating insights into patient benefit. Despite remaining challenges, the three-subset model supplies a conceptual anchor for this integrative approach and may support the development of more effective and widely applicable cancer immunotherapies.

Abbreviations

ACT

adoptive cell transfer

CMV

cytomegalovirus

EBV

Epstein-Barr virus

HPV

human papillomavirus

ICI

immune checkpoint inhibitor

TAA

tumor-associated Ag

TAST

tumor-antigen specific T cell

TCF-1

T cell factor-1

Tex-term

terminal exhaustion

TIL

tumor-infiltrating lymphocyte

TLS

tertiary lymphoid structure

TMB

tumor mutational burden

TME

tumor microenvironment

Tpex

progenitor exhausted T cells

VST

virus-specific T cell