Predictors of sensitivity to immune therapies in classic Hodgkin lymphoma

Visual Abstract

Abstract

Immune checkpoint blockade, particularly programmed cell death protein 1 inhibition, has redefined the management of classic Hodgkin lymphoma (cHL), achieving unprecedented efficacy in relapsed/refractory settings. Yet, durable benefit is not universal, because mechanisms of primary and acquired resistance remain incompletely understood. This review integrates current knowledge on predictors of sensitivity to immune therapies in cHL across clinical, biological, and technological dimensions. Established predictors, including disease burden, previous treatment exposure, CD30 intensity, programmed death-ligand 1 (PD-L1)/PD-L2 copy number gains, and loss of major histocompatibility complex expression, offer valuable but incomplete prognostic information. Tumor microenvironmental features such as macrophage polarization, T-cell exhaustion, and immune spatial organization further refine response prediction, whereas circulating biomarkers such as soluble PD-L1, circulating tumor DNA kinetics, and cytokine profiles provide noninvasive insights. Molecular and cellular pathways underlying resistance encompass genetic and epigenetic alterations, immune editing, and adaptive checkpoint upregulation. Emerging predictive frameworks, spanning multiomics and spatial profiling, radiomics, artificial intelligence, and microbiome-host cross talk, promise to enhance precision in patient stratification. Finally, the review outlines key challenges and research priorities for translating these multidimensional biomarkers into clinical trials and practice. A unified predictive framework integrating clinical, molecular, and computational indicators may ultimately enable personalized immunotherapy and overcome resistance in cHL.

Introduction

Classic Hodgkin lymphoma (cHL) is an uncommon lymphoid malignancy, accounting for ∼10% of all lymphomas, affecting ∼85 000 individuals worldwide each year, with a peak incidence among adolescents and young adults.¹ Since the advent of radiation therapy and combination chemotherapy in the 1960s, treatment outcomes have markedly improved, with current 5-year overall survival (OS) rates nearing 90%.² Continued advances in the biological understanding of HL have led to the development of targeted and immune-based therapies, further enhancing patient outcomes (Table 1).

Table 1.

Immune therapies used in the management of cHL

Therapy	Class/target	Clinical setting in article	References
Nivolumab	Anti–PD-1 monoclonal antibody	R/R cHL after ASCT failure; PD-1 inhibitor backbone in CheckMate-205	3,4
N + AVD	Anti–PD-1 + doxorubicin/vinblastine/dacarbazine	Frontline therapy in advanced-stage cHL, improved 2-year PFS vs BV-AVD (SWOG S1826)	5
Pembrolizumab	Anti–PD-1 monoclonal antibody	R/R cHL vs BV (KEYNOTE-204)	6,7
Pembrolizumab + AVD	Anti–PD-1 + doxorubicin/vinblastine/dacarbazine	Frontline chemoimmunotherapy	6
Sintilimab	Anti–PD-1 monoclonal antibody	R/R cHL	8
BV	Anti-CD30 ADC	R/R cHL	9, 10, 11
BV + AVD	Anti-CD30 ADC + doxorubicin/vinblastine/dacarbazine	Frontline regimen in advanced-stage cHL; improved outcomes vs ABVD but with added toxicity (ECHELON-1)	9,11
AFM13	Bispecific CD30/CD16A (CD30×CD16A) innate cell engager	R/R cHL; CD30-directed immunotherapy engaging NK cells via CD16A	12
Camrelizumab	Humanized high-affinity IgG4 monoclonal antibody against PD-1	R/R cHL	13
Sintilimab	Monoclonal antibody against PD-1 and its ligands	R/R cHL	14
Tislelizumab	Anti–PD-1 monoclonal antibody engineered to reduce Fcγ receptor binding	R/R cHL	15
Anti-CD30 CAR T-cell therapy	Anti-CD30 CAR T-cell therapy	R/R cHL	16, 17, 18

ADC, antibody-drug conjugate; CAR, chimeric antigen receptor; Fcγ, Fc gamma; IgG4, immunoglobulin G4.

cHL is uniquely sensitive to immune therapy due to 9p24.1 amplification and JAK-STAT–driven programmed death-ligand 1 (PD-L1)/PD-L2 overexpression on Hodgkin-Reed Sternberg (HRS) cells, enabling robust activity of programmed cell death protein 1 (PD-1) blockade.^3,19,20 For decades, cHL was managed with ABVD/BEACOPP (doxorubicin, bleomycin, vinblastine, dacarbazine/bleomycin, etoposide, doxorubicin, cyclophosphamide, vincristine, procarbazine, and prednisone) with or without radiotherapy and positron emission tomography (PET)–adapted refinements.21, 22, 23 The advent of brentuximab vedotin (BV) and PD-1 blockade first proved highly effective in relapsed/refractory (R/R) disease and then moved earlier, after autologous stem cell transplantation (ASCT) consolidation, salvage before ASCT, and now frontline, redefining care.^9,10,23,24 In advanced-stage disease, BV-AVD (adriamycin [doxorubicin], vinblastine, dacarbazine) improved outcomes over ABVD (ECHELON-1) but with added toxicity.¹¹ In the front line, nivolumab plus AVD improved 2-year progression-free survival (PFS) vs BV-AVD in SWOG S1826, establishing an emerging standard, whereas early-stage ICI-based approaches (eg, NIVAHL[Nivolumab and AVD in Early-Stage Unfavorable Classic Hodgkin Lymphoma trial]) achieved excellent outcomes.^5,25 Beyond frontline use, nivolumab has demonstrated durable activity in R/R cHL, both as monotherapy and in combination regimens.²⁶ In R/R disease, pembrolizumab outperformed BV in KEYNOTE-204 and PD-1–based combinations yield high complete response (CR) rates and durable PFS.^6,7

In addition to immune checkpoint inhibition and antibody-drug conjugates, several other immunotherapeutic strategies are actively being explored in cHL. CD30-directed chimeric antigen receptor T-cell therapies have demonstrated encouraging activity in heavily pretreated R/R cHL in early-phase trials, with overall response rates exceeding 60% in some series, although durability of response and cytokine-mediated toxicities remain challenges.^27,28 Adoptive natural killer (NK) cell–based approaches, including CD30-targeted NK-engaging bispecific antibodies such as AFM13, have also shown preliminary clinical activity with favorable safety profiles.^28,29

Despite remarkable advances with immune therapies, not all patients with cHL achieve durable remission. A subset exhibits primary resistance or relapses after an initial response, highlighting heterogeneity in treatment sensitivity.³⁰ Understanding the biological and clinical predictors of response is therefore essential for optimizing patient selection and guiding individualized therapy. Factors such as tumor PD-L1/PD-L2 expression, 9p24.1 amplification, tumor microenvironment (TME) composition, immune-cell exhaustion signatures, and early clearance of circulating tumor DNA (ctDNA) have emerged as potential correlates of sensitivity.31, 32, 33, 34, 35 Conversely, genetic and epigenetic mechanisms driving immune evasion, such as JAK/STAT pathway dysregulation, β2-microglobulin (B2M) loss, and reduced antigen presentation, may underlie resistance to immune checkpoint inhibitors.^34,36, 37, 38 Identifying and validating such biomarkers will guide rational regimens to overcome resistance.

This review synthesizes evidence across clinical, biological, and technological domains to define predictors of immune therapy response in cHL. Given the limited number of prospectively validated predictive biomarkers across immune-based therapies in cHL, we distinguish levels of evidence throughout this review. Established biomarkers are defined as features supported by clinical outcome correlations in patients treated with immune-based therapies, whereas emerging biomarkers are supported by correlative analyses, strong mechanistic rationale, or consistent associations in cHL biology independent of treatment context. When applicable, we specify whether evidence derives from studies of PD-1–directed therapy or from broader investigations of HL immunobiology.

Biological background

cHL is defined by a striking dissociation between tumor cell burden and microenvironment: rare malignant HRS cells are embedded in a quantitatively dominant TME, which can account for ∼99% of the cellular infiltrate.39, 40, 41 The TME consists of stromal elements surrounded by a diverse infiltrate of nonneoplastic immune cells, including CD4⁺ and CD8⁺ T cells, B cells, eosinophils, mast cells, and tumor-associated macrophages (TAMs). Among these populations, CD4⁺ T cells are numerically dominant; helper T (Th) cells and FOXP3⁺ regulatory T cells (Tregs) cluster around HRS cells and organize locally immunosuppressive niches.42, 43, 44 HRS cells actively educate this infiltrate via secretion of cytokines/chemokines and modulation of surface molecules,⁴⁵ shaping a lymphoma-specific ecosystem that promotes immune evasion and provides survival and growth signals.^42,43,46 Key soluble mediators include CCL5, CCL17 thymus and activation-regulated chemokine (TARC), CCL20, and CCL22, which recruit Th cell and Treg subsets, as well as interleukin-5 (IL-5), IL-6, IL-10, IL-13, and transforming growth factor β, which drive eosinophil, mast-cell and macrophage accumulation, and skewing toward an M2, immunosuppressive phenotype.47, 48, 49, 50, 51

High-resolution spatial and single-cell studies have refined this view and directly linked TME architecture to immune escape.³⁹ Multiparameter imaging and topological analysis have shown that PD-L1⁺ TAMs often outnumber PD-L1⁺ HRS cells and colocalize with PD-1⁺ CD4⁺ T cells, creating PD-1/PD-L1–rich niches that may be critical targets of checkpoint blockade.^39,52 These data help explain the remarkable efficacy of PD-1 inhibitors in cHL despite frequent loss of major histocompatibility complex class I (MHC-I) due to B2M alterations. Time-of-flight mass cytometry further demonstrates a TME enriched for Th1-polarized CD4⁺ effector cells expressing PD-1, alongside Th1-skewed Tregs that lack PD-1, consistent with an exhausted yet immunosuppressive T-cell compartment.⁵³ Single-cell RNA sequencing coupled with spatial profiling has identified disease-defining T-cell subsets, including type 1 regulatory cells with high Lymphocyte-activation gene 3 (LAG-3) expression that cluster around HRS cells lacking MHC-II, providing a mechanistic link between loss of antigen presentation and recruitment of immunosuppressive LAG-3⁺ T cells.54, 55, 56 Additional spatial interactions, such as CXCR5⁺ HRS cells engaged with CXCL13⁺ TAMs and CXCR5⁺ nonmalignant B cells, have been incorporated into prognostic spatial scores that predict outcome in R/R cHL after ASCT.^54,56

Despite the striking clinical efficacy of PD-1 blockade in cHL, key aspects of its underlying immunobiology remain incompletely understood. Although high PD-L1 expression on HRS cells and TAMs clearly contributes to profound immune suppression within the TME, the precise mechanisms by which PD-1 inhibition restores effective antitumor immunity are still debated.⁵⁷ In particular, the frequent loss of MHC-I expression and variable impairment of MHC-II antigen presentation challenge classical models of neoantigen-driven CD8⁺ T-cell reinvigoration.^58,59 These observations suggest that PD-1 blockade in cHL may rely on alternative or complementary mechanisms, including restoration of CD4⁺ T-cell function, modulation of PD-1–expressing innate immune cells such as macrophages and NK cells, or noncognate immune interactions within PD-L1–rich cellular niches.^60,61 The relative contributions of these pathways, as well as the role of tumor-derived vs viral antigens in shaping response, remain important areas for future investigation.

The immunobiology of cHL is further shaped by age and Epstein-Barr virus (EBV) status. Gene expression profiling and TME deconvolution reveal that eosinophil, B-cell and mast-cell signatures are more prominent in younger patients, whereas macrophage and stromal signatures are enriched in older adults, prompting development of age-specific TME-based prognostic models.^56,62 Pediatric and adolescent/young adult cHL shows enrichment of M1-polarized macrophages and cytotoxic/Th1 cells, particularly in EBV⁺ cases, which may contribute to relatively favorable outcomes in some pediatric EBV-associated tumors.^44,63, 64, 65 By contrast, adult EBV⁺ cHL is characterized by accumulation of FOXP3⁺ Tregs and immunosuppressive cytokines, suggesting an aged TME with attenuated antitumor immunity that may underlie poorer prognosis in older patients with EBV⁺ status.^44,63, 64, 65, 66 Recent single-cell work spanning the age spectrum has shown that mononuclear phagocytes (dendritic cells and monocytes) expressing PD-L1 and T-cell immunoglobulin and mucin domain-containing protein 3 (TIM-3) cluster near HRS cells, with immune checkpoint expression in these populations increasing with age, again pointing to age-dependent myeloid biology and checkpoint ligand availability in the TME.⁶⁷

EBV⁺ cHL differs from EBV⁻ disease through both tumor-intrinsic and microenvironmental mechanisms. Viral latency IIa proteins expressed by HRS cells (eg, LMP1/LMP2A/EBNA1) provide potential antigenic targets, and impaired responses to LMP1 epitopes have been documented in patients with HL, consistent with active immune evasion. In parallel, EBV can amplify immune suppression by increasing PD-L1 expression (beyond 9p24.1 amplification effects) and promoting an immunoregulatory microenvironment enriched for checkpoint signaling and regulatory cytokines/cells.68, 69, 70

At the genomic level, HRS cells harbor recurrent alterations that promote immune escape through coordinated disruption of antigen presentation and amplification of inhibitory immune signaling. One of the most characteristic features of cHL is overexpression of PD-L1 and PD-L2, most commonly driven by copy number gains or focal amplification of chromosome 9p24.1 amplification, detected in ∼70% to 75% of cases. These alterations frequently coamplify JAK2, further enhancing JAK-STAT–dependent PD-L1 expression and establishing a dominant PD-1–mediated inhibitory axis within the TME.^35,59,67,71, 72, 73, 74, 75, 76

Defects in antigen presentation represent a second, complementary mechanism of immune evasion. Loss of MHC-I expression is common in cHL and is most frequently caused by inactivating alterations of B2M, which are often biallelic and reported in ∼33% to 70% of cases. These lesions are sufficient to abrogate surface MHC-I expression and impair CD8⁺ T-cell–mediated tumor recognition, particularly in EBV⁻ disease, in which B2M mutations are enriched and associated with inferior PFS.^19,58,59,74,77, 78, 79, 80, 81, 82, 83, 84, 85 Additional recurrent truncating mutations, missense variants, and structural alterations affecting HLA-B and other MHC-I components further contribute to this phenotype.^38,73

MHC-II expression is also frequently disrupted in cHL. Lack of surface MHC-II expression is observed in ∼41% of patients, with aberrant cytoplasmic localization reported in an additional subset, indicating impaired antigen presentation despite retained protein expression.^59,86 Genomic rearrangements and structural variants involving CIITA (class II transactivator), the master regulator of MHC-II transcription, are detected in ∼9% to 16% of cases and account for only a subset of MHC-II–negative tumors.^35,59,73,75 This indicates that the remaining proportion of MHC-II loss arises through nongenetic mechanisms, including transcriptional repression, epigenetic silencing, and defects in antigen loading machinery. Supporting this, even in tumors retaining MHC-II expression, loss of HLA-DM results in persistent surface presentation of CLIP (class II–associated invariant chain peptide), which occupies the MHC-II peptide-binding groove and prevents loading of tumor-derived peptides, thereby limiting productive CD4⁺ T-cell priming.^87,88

These antigen presentation defects coexist with extensive expression of inhibitory ligands on HRS cells, including PD-L1, PD-L2, CD200, HLA-G, and HLA-E, as well as abundant PD-L1 expression on TAMs, in part, through trogocytosis. Together, these features generate a PD-1/PD-L1–rich and profoundly immunosuppressive microenvironment.81, 82, 83, 84 Clinically, responses to PD-1 blockade correlate with MHC-II, but not MHC-I, expression on HRS cells, supporting an important role for CD4⁺ T-cell–associated immunity (and potentially other effector populations) in mediating therapeutic benefit from PD-1 blockade in cHL.⁸⁹ Despite this strong clinical correlation, direct confirmation that CD4⁺ T cells recognize HRS cell-derived tumor antigens in an MHC-II–restricted manner and mediate direct cytotoxic killing remains limited. Existing studies (including microenvironmental profiling and T-cell clonality/expansion analyses) are consistent with antigen-driven CD4⁺ responses, but definitive antigen-specific assays are technically challenging in cHL because HRS cells are rare.^52,53,74 Accordingly, whether CD4⁺ T cells act primarily through direct MHC-II–restricted tumor recognition vs indirect mechanisms (eg, cytokine production and microenvironmental modulation) remains an important unresolved question. PD-1 blockade may also interfere with prosurvival reverse signaling through PD-L1 in HRS cells, further contributing to treatment responses.^90,91

From an immunoediting perspective, immune escape in cHL reflects coordinated hide and defend strategies along the elimination-equilibrium-escape continuum. HRS cells hide by disrupting antigen presentation through loss of MHC-I and MHC-II, whereas simultaneously defending themselves by upregulating inhibitory ligands that suppress residual immune effector function.⁹² Although MHC-II expression has been reported in specific contexts to transmit inhibitory signals to exhausted LAG-3⁺ tumor-infiltrating lymphocytes, particularly in EBV⁺ or mixed-cellularity subtypes, the overall biological and clinical evidence indicates that MHC-II⁺ HRS cells remain more immunogenic. Any inhibitory signaling mediated by MHC-II appears context-dependent and outweighed by enhanced CD4⁺ T-cell–mediated antitumor activity, particularly in the setting of PD-1 blockade.93, 94, 95

Established predictors of response to immune therapy

Table 2 lists the established predictors of response to immune therapy in cHL. It summarizes clinicopathologic, tumor-intrinsic, and microenvironmental features in cHL.

Table 2.

Established predictors of response to immune therapy in cHL

Predictor/biomarker	Level/category	Association with response/outcome	References
Lower baseline disease burden (stage, TMTV)	Clinical/imaging	Lower stage, lower TMB generally associated with better PFS/OS and more favorable biology; high burden frequently linked to poorer outcomes with PD-1 inhibitors	96, 97, 98
Fewer previous lines of therapy/no previous ASCT failure	Clinical	Less heavily pretreated disease and preserved chemosensitivity predict higher response rates to PD-1 blockade and other immune strategies	4,99,100
Favorable performance status, limited comorbidities	Clinical	Good ECOG and absence of major organ dysfunction associate with better tolerability and higher likelihood of meaningful benefit from ICIs	101
IPS components (stage IV, leukocytosis, anemia, LMR, lymphopenia)	Clinical/inflammatory	Stage IV disease, leukocytosis, anemia, and low LMR predict inferior PFS; lymphopenia and male sex predict inferior OS; high-risk IPS features overlap with features of more aggressive disease biology that responds less well to immunotherapy	31,96
Early PET/CT metabolic response (PET negativity, deep remission)	Imaging	Rapid achievement of PET negativity and durable metabolic responses after salvage or chemoimmunotherapy correlate with higher sensitivity to immune therapies; persistent PET avidity or short remission interval suggests immune-refractory phenotype	97,102
9p24.1 gain/amplification with high PD-L1/PD-L2 expression	Pathologic/genomic	Defines cHL biology; higher 9p24.1 amplification and PD-L1 expression correlate with better responses to PD-1 blockade, whereas being adverse in chemotherapy-only cohorts	19,58,89
MHC-II expression on HRS cells	Pathologic/immunophenotypic	Retained/strong MHC-II expression is associated with higher response rates to nivolumab and improved outcomes after PD-1 blockade, supporting CD4+ T-cell–dependent mechanisms	54, 55, 56,89
MHC-I expression on HRS cells	Pathologic/immunophenotypic	MHC-I presence is associated with better outcomes after polychemotherapy; its role for PD-1 blockade is more complex, but profound MHC-I loss is generally linked to immune escape and poorer prognosis	73,74,77
CD30 expression intensity	Pathologic	Universal CD30 expression is a disease hallmark; whereas usually not graded, it enables CD30-directed agents (BV, AFM13) that can synergize with ICIs; not a finely quantitative predictor but a prerequisite biomarker	12,46
TAM density (CD68+/CD163+)∗	TME	High numbers of TAMs, especially CD163+ M2-skewed cells, consistently predict treatment failure and inferior survival, indicating a more resistant microenvironment	59,103,104
T-cell exhaustion signatures (PD-1, LAG-3, TIM-3, TIGIT coexpression)∗	TME	High expression and coexpression of inhibitory receptors on tumor-infiltrating T cells mark functionally exhausted populations and associate with worse outcomes; also highlight potential targets for combination checkpoint blockade	105,106
Spatial organization (RHL-4S, PD-1+ T cells and PD-L1+ TAMs in HRS niches)∗	TME/spatial	Adverse spatial scores (eg, RHL-4S) with close proximity of suppressive immune cells to HRS cells strongly predict post-ASCT failure-free survival; spatially “cold/suppressive” architectures are less responsive to therapy	52,54,56
TME B-cell content∗	TME	Low infiltrating B-cell content on whole-slide analysis correlates with adverse outcomes in advanced cHL treated with BEACOPP and reflects an immunologically “poor” milieu	107
Baseline ctDNA level†	Circulating	Baseline ctDNA correlates with tumor burden; in patients treated with anti–PD-1 therapy, higher baseline ctDNA in responders and dynamic reductions are predictive of benefit	8,76,108
Early ctDNA decline/clearance (eg, ≥40% drop; MRD negativity)†	Circulating	Early and profound ctDNA declines during therapy, and MRD negativity by ultrasensitive assays, are strongly associated with durable responses and superior PFS, including in PD-1–based regimens	6,8,108, 109, 110
Cytokine signatures (IL-6, IL-2R, IL-10, TNF-α, TARC)†	Circulating	Elevated pretreatment cytokines generally predict early relapse and inferior survival; dynamic changes in cytokines (eg, TARC decline) can mirror response to therapy	49,111,112
EV-miRNAs (eg, EV-miR-127-3p, EV-let-7a-5p) combined with serum TARC†	Circulating	EV-miRNA levels fall with complete metabolic response and rise at relapse; when combined with TARC they improve discrimination of PET-positive vs PET-negative disease and may anticipate benefit from immunotherapy	113, 114, 115

ECOG, Eastern Cooperative Oncology Group; IL-2R, IL-2 receptor; LMR, lymphocyte-to-monocyte ratio; MRD, minimal residual disease; RHL-4S, relapsed Hodgkin lymphoma 4-spatial; TARC, thymus and activation-regulated chemokine; TMB, tumor mutational burden; TNF-α, tumor necrosis factor α.

^∗Together, these TME-derived predictors build a comprehensive framework in which cellular abundance, immune checkpoint expression, spatial organization, and high-dimensional immune profiling converge to explain interpatient heterogeneity in treatment sensitivity and resistance. Because these biomarkers progress toward clinical translation, they promise to refine risk stratification and guide individualized therapeutic strategies in cHL.

^†Overall, these data support a model in which ctDNA dynamics, mutational profiles, cytokine signatures, and EV-miRNA/TARC panels act as circulating surrogates of tumor burden and immune-tumor interaction that can refine risk stratification, anticipate benefit from PD-1–based strategies, and flag emerging resistance in cHL.

Clinical predictors

Baseline clinical variables still play a central role in predicting outcomes in cHL, including in patients who may later receive immune checkpoint blockade.¹¹⁶ Recent PET/computed tomography (CT)–staged analyses highlight that traditional International Prognostic Score (IPS) components retain prognostic value in the modern era. In a cohort of 274 patients treated with PET/CT-guided ABVD, Cellini et al⁹⁶ identified stage IV disease, leucocytosis, anemia, and low lymphocyte-to-monocyte ratio as significant predictors of inferior PFS, whereas male sex, stage IV disease, and lymphopenia significantly predicted inferior OS. Importantly, not all IPS variables remained independently prognostic, and incorporating lymphocyte-to-monocyte ratio with only the significant IPS items produced a simplified model that better discriminated high-risk patients. Although this study was conducted in the chemotherapy setting, its findings reinforce a consistent clinical theme across the immunotherapy literature: patients with higher baseline disease burden, systemic inflammatory features, or cytopenias reflecting compromised host immunity tend to experience more aggressive disease biology overall. These same features have repeatedly been associated with poorer outcomes in R/R cohorts receiving PD-1 inhibitors, supporting their continued relevance as baseline clinical predictors of immunotherapy response.^96,116

Clinical factors remain highly informative in anticipating which patients will derive meaningful benefit from PD-1 blockade or other immune-based strategies in cHL. Patients with fewer previous lines of therapy, preserved chemosensitivity, and no history of ASCT failure typically demonstrate higher responsiveness to ICIs,^4,99,100 reflecting a less treatment-exhausted disease biology. Early metabolic responses, particularly rapid PET negativity and deeper, durable remissions after salvage therapy, correlate with increased sensitivity to immune therapies, whereas persistent PET avidity or short remission intervals often mark a more immune-refractory phenotype.⁹⁷ In addition, baseline patient status plays a nontrivial role: good performance status, limited comorbidities, and absence of treatment-limiting frailty are associated with superior outcomes, whereas poor Eastern Cooperative Oncology Group performance status scores and organ dysfunction predict reduced tolerance and diminished responsiveness to immune-based regimens.^101,117,118

With the increasing use of PD-1 blockade in the frontline setting, optimal management of R/R HL after previous immunotherapy exposure remains uncertain. Emerging evidence suggests that retreatment or rechallenge with immune checkpoint inhibitors (ICIs) may retain substantial clinical activity in selected patients. In a recent systematic review and meta-analysis¹¹⁹ including 207 patients with R/R cHL who underwent ICI rechallenge after previous discontinuation, the pooled overall response rate was 63% with a CR rate of 43% and a median PFS of 13.3 months, outcomes comparable with uninterrupted ICI therapy. Importantly, most patients rechallenged had achieved an initial response to PD-1 blockade, supporting previous depth and durability of response as key clinical predictors of benefit from retreatment. ICI rechallenge was generally feasible, with grade 3 to 5 adverse events occurring in approximately one-third of patients, underscoring the need for careful patient selection, particularly in those who discontinued initial therapy due to immune-related toxicity. Collectively, these data suggest that patients with previous sensitivity to PD-1 blockade may remain immunotherapy-responsive upon retreatment, whereas truly refractory disease likely reflects underlying immune escape mechanisms requiring alternative or combinatorial strategies.¹¹⁹

Pathologic and immunophenotypic predictors

cHL is defined by rare HRS cells with uniform, universal CD30 expression in the absence of typical B-cell markers, a phenotype that has been successfully exploited therapeutically with CD30-directed agents such as BV and CD30-targeted cellular/bispecific approaches rather than graded as a quantitative predictive biomarker.^12,29 The most mature link between HRS cell genomics and treatment response concerns 9p copy number alterations: recurrent gains/amplifications at 9p24.1 amplification drive increased PD-L1/PD-L2 expression on HRS cells and form the basis for prognostic and predictive biomarker evaluations in the setting of standard chemotherapy and PD-1 checkpoint inhibition.^77,89 In parallel, alterations in antigen presentation machinery are emerging as key correlates of outcome: MHC-I expression on HRS cells has been associated with treatment outcomes after polychemotherapy,⁵⁸ and combined assessment of PD-L1 and MHC-II expression on HRS cells correlated with the efficacy of single-agent nivolumab in CheckMate-205, supporting a CD4⁺ T-cell–dependent mechanism of action for PD-1 blockade in cHL.⁸⁹ Complementing these clinical correlates, single-cell and spatial studies have shown that MHC-II–negative HRS cells are specifically associated with an immunosuppressive T-cell milieu enriched in LAG-3⁺ type 1 regulatory–like cells and IL-6–driven cross talk, further linking loss of MHC-II to an immune-evasive microenvironment with potential relevance for resistance to immunotherapies.⁵⁵

TME features

It should be noted that much of the current understanding of the cHL TME derives from studies of disease biology independent of treatment context. Although these observations provide a critical framework for understanding immune escape, their direct relevance to response or resistance to immune-based therapies, has only been partially explored. When available, we highlight data from immunotherapy-treated cohorts; however, for many TME features, including macrophage polarization states, T-cell exhaustion programs, and spatial immune organization, prospective interrogation in patients receiving immune therapies remains an important unmet need.

An expanding body of translational research highlights that the cellular composition, functional state, and spatial architecture of the TME play central roles in determining clinical outcomes and therapeutic responsiveness in cHL. Among these features, TAMs have emerged as one of the most robust and consistently validated prognostic biomarkers.¹²⁰ Early work demonstrated that high densities of CD68⁺ macrophages strongly correlate with primary treatment failure and inferior survival, a finding subsequently reinforced across multiple cohorts where CD163⁺ macrophages, often representing a more immunosuppressive, M2-skewed phenotype, similarly marked poor-risk disease.^103,120,121 These macrophage subsets support an immunosuppressive niche through cytokine production, antigen presentation defects, and direct support of HRS cell survival, making their abundance a surrogate for a more therapy-resistant microenvironment.^103,104

In parallel, profiling of T-cell states within the TME has revealed that immune exhaustion signatures are highly informative. Expression of inhibitory receptors such as PD-1, LAG-3, TIM-3, and T-cell immunoreceptor with Ig and ITIM domains (TIGIT) on tumor-infiltrating lymphocytes reflects progressive functional impairment and contributes to immune escape.^105,106 Multiplex immunohistochemistry (IHC) studies show that coexpression of these checkpoints, particularly PD-1 with LAG-3 or TIM-3, identifies T-cell populations with diminished effector capacity and associates with worse outcomes.¹⁰⁶ These exhaustion markers not only map the degree of T-cell dysfunction but also provide biologic rationale for checkpoint blockade combinations beyond PD-1 inhibition (Figure 1).

Figure 1.

Spatial organization of the cHL TME around a HRS cell, resistance to immunotherapies, and poor response predictors. The cHL TME with macrophage markers (CD68, CD163) and T-cell exhaustion markers (PD-1, LAG-3). CD68⁺ macrophages (TAMs, M1) and CD163⁺ macrophages (M2-skewed TAMs) often closest to the HRS cells form part of the tight pericellular ring. PD-1 T-cell exhausted (CD4⁺ > CD8⁺ T cells) are positioned at moderate proximity to the HRS cells, whereas LAG-3⁺ T cells (deeply exhausted T cells) form a slightly more peripheral layer, depending on the exhaustion gradient. They frequently coexpress on the same T-cell subsets. Coexpression of PD-1 and LAG-3 identifies populations most strongly associated with adverse outcomes. Microphotographs on the right of the schema show immunohistochemical expression of CD163⁺ macrophages (top) and LAG-3⁺ T cells (bottom) around a HRS cell (immunohistochemistry, hematoxylin counterstain, original magnification ×40). 9p24.1 amplification in the context of MHC loss: amplification drives high PD-L1/PD-L2 expression; when combined with MHC-I loss, associated with particularly poor outcomes, indicating synergistic “hide and defend” immune evasion and potential resistance to PD-1 blockade. Constitutive JAK-STAT activation: promotes PD-L1 overexpression and secretion of chemokines such as CCL17, attracting Tregs and reinforcing an M2-skewed TME; supports immune suppression and resistance. High PD-L1 expression on HRS cells and TAMs: creates PD-1/PD-L1–rich niches that inhibit T-cell and NK cell function; although PD-L1 predicts benefit from PD-1 blockade, extreme ligand abundance within a profoundly suppressive milieu may also mark dependence on this axis and adaptive resistance. Macrophage-dominated, M2-skewed TME: high CD68⁺/CD163⁺ macrophage infiltrates support tumor survival, secrete suppressive cytokines, and correlate with treatment failure; such TMEs may be less amenable to monotherapy ICIs. Exhausted T-cell infiltrates with high PD-1/LAG-3/TIM-3/TIGIT: deeply exhausted effector T cells show impaired function; persistent or dominant exhaustion signatures suggest that PD-1 monotherapy may be insufficient and that alternative/checkpoint coblockade is needed. LAG-3⁺ Tr1-like cells clustering around MHC-II–negative HRS cells: spatial coupling of MHC-II–negative HRS with LAG-3⁺ Tr1 cells creates highly suppressive niches that limit effective antitumor immunity and may underpin resistance to PD-1 blockade. Spatial organization (RHL-4S, PD-1⁺ T cells, and PD-L1⁺ TAMs in HRS niches): adverse spatial scores (eg, RHL-4S) with close proximity of suppressive immune cells to HRS cells strongly predict post-ASCT failure-free survival; spatially “cold/suppressive” architectures are less responsive to therapy. Tr1, type 1 regulatory.

Beyond individual cellular subsets, the spatial relationships between HRS cells and surrounding immune populations have emerged as crucial determinants of disease behavior. High-resolution imaging mass cytometry has demonstrated that the physical proximity of suppressive cells, such as PD-1⁺ CD4⁺ T cells, CD68⁺ macrophages, and CXCR5⁺ B cells, to HRS cells predicts post-ASCT failure-free survival, culminating in the Relapsed Hodgkin Lymphoma 4-Spatial (RHL-4S) spatial score.⁵⁶ This model underscores that prognostically relevant biology is not merely defined by cellular abundance but by coordinated spatial ecosystems that either restrain or facilitate tumor survival. These spatial signatures offer a new frontier of biomarker development with direct ties to the mechanisms of immune evasion.

Complementing these insights, advanced multiplex immunophenotyping, flow cytometry, and whole-slide digital pathology have further refined the understanding of TME composition, revealing distinct immune-cell communities that stratify patients into biologically meaningful risk groups.^67,107 Such multidimensional analyses consistently show that tumors characterized by macrophage dominance, exhausted T-cell infiltrates, and suppressive spatial niches are more likely to resist conventional therapy and may respond differently to immune-based treatments.^67,107

Circulating biomarkers

Circulating biomarkers are emerging as a key complement to PET/CT in predicting response and resistance to immune-based therapies in cHL. Among them, ctDNA is the most advanced: multiple studies show that ctDNA levels mirror HRS cell genetics, correlate with metabolic tumor volume, and fall rapidly in patients with favorable clinical courses, whereas persistent ctDNA or copy number alterations mark a higher risk of relapse.^{76,108,122,123} In a comprehensive ctDNA sequencing study, pretreatment and on-treatment ctDNA levels refined risk prediction and minimal residual disease detection with ultrasensitive phased variant enrichment and detection sequencing; the authors concluded that ctDNA can complement interim PET and guide the selection of appropriate therapies in cHL.^76,109 More specifically in the immunotherapy setting, Shi et al showed that baseline ctDNA was significantly higher in responders to anti–PD-1 therapy, that a ≥40% decrease in ctDNA from baseline identified patients with better outcomes, and that mutations in B2M, TNFRSF14, and KDM2B were associated with acquired resistance to this treatment.⁸ Spina et al further demonstrated that longitudinal ctDNA profiling tracks clonal evolution under chemotherapy and nivolumab, with larger early ctDNA declines in patients achieving CRs and less pronounced drops (<2-log) in those who ultimately relapsed, even when interim PET/CT was sometimes judged negative.¹⁰⁸ In patients treated front line with pembrolizumab plus chemotherapy, ctDNA clearance after 2 cycles and at end of treatment was strongly associated with superior PFS, and patients who remained ctDNA-negative despite residual imaging abnormalities did not relapse during follow-up, underscoring ctDNA’s potential to disentangle true progression from residual imaging signals in the immunotherapy era.⁶ In R/R cHL, integrating baseline ctDNA genotyping and serial quantification with interim PET improved early identification of high-risk patients treated with BEGEV (bendamustine, gemcitabine, and vinorelbine), leading the authors to suggest that such patients might benefit from an early change to immunotherapy.¹¹⁰

The prognostic relevance of ctDNA has recently been further strengthened by analyses from the phase 3 SWOG S1826 trial in newly diagnosed advanced-stage cHL treated with nivolumab-AVD or BV-AVD. Using a targeted sequencing approach to quantify molecular tumor burden, ctDNA positivity at cycle 3 day 1 was independently associated with inferior PFS and outperformed interim PET in risk stratification across both treatment arms. Importantly, among patients with ctDNA positivity, the magnitude of early molecular tumor burden decline further delineated outcomes, with only minor reductions identifying a high-risk subgroup. At end of therapy, ctDNA positivity remained strongly prognostic, and integration of ctDNA with PET refined risk assessment, identifying patients with excellent vs dismal outcomes.¹²⁴

Beyond ctDNA, other circulating markers, such as elevated pretreatment cytokines (eg, IL-6, IL-2R, IL-10, and tumor necrosis factor α) that independently predict early relapse and inferior survival,^49,111,112 and extracellular vesicle–associated microRNAs (EV-miRs) whose levels fall with complete metabolic response (CMR) and rise again at relapse, also correlate closely with PET status; combining EV-miR-127-3p or EV-let-7a-5p with serum TARC markedly improved the accuracy and negative predictive value for distinguishing PET-positive from PET-negative disease.113, 114, 115

Novel and emerging predictive approaches

Table 3 shows novel and emerging predictive approaches in cHL highlighting emerging biomarkers and analytic approaches.

Table 3.

Novel and emerging predictive approaches in cHL

Approach	Modality/examples	Predictive application/readout	References
Bulk gene expression–based models (eg, RHL30)	FFPE-based expression panels (RHL30 in R/R cHL; pediatric TME signatures)	Distinguish inflammatory vs immunosuppressive programs; prognosticate outcomes after salvage therapy and frontline chemotherapy; inform who may benefit from intensified or immune-based approaches	62,125, 126, 127
scRNA-seq	scRNA-seq of HRS and TME cells	Defines disease-defining T-cell subsets (eg, LAG-3+ Tr1), links MHC-II loss to recruitment of suppressive subsets, identifies age- and EBV-dependent TME programs; supports resistance/sensitivity maps	54,55,128
Time-of-flight mass cytometry and imaging mass cytometry	High-dimensional protein profiling and spatial phenotyping	Characterizes exhausted CD4+ effector/Treg compartments, PD-L1+ myeloid cells; defines PD-1/PD-L1–rich niches and spatial ecosystems predictive of post-ASCT outcomes (eg, RHL-4S score)	52, 53, 54,56
Spatial transcriptomics/multiplex IF	Spatially resolved transcriptional and protein maps	Integrates HRS genotype (eg, 9p24.1, MHC-II loss) with spatial T-cell/myeloid niches; generates spatial signatures that stratify risk and predict response to immune therapies	54,56,128
Integrated multiomic profiling (genomic + transcriptomic + epigenomic + proteomic)∗	WGS/WES, scRNA-seq, ATAC-seq/methylation, proteomics, ctDNA	Aims to delineate HRS/TME clones associated with durable response vs resistance; integrates ctDNA genotyping, MRD, cytokines (TARC, IL-10, soluble CD163) and TMTV to build composite predictive scores	35,76,109,129
Radiomics on FDG PET/CT†	High-dimensional shape/texture features; first-, second-, higher-order metrics	Captures intratumoral heterogeneity and dissemination patterns not evident visually; radiomic signatures may predict immunotherapy outcomes better than DS alone	98,130,131
TMTV and TLG†	Semiautomated PET segmentation to derive TMTV/tTLG	High baseline TMTV/tTLG consistently associated with inferior PFS/OS; potential integration with ICI-based strategies to identify high-risk patients needing escalation or novel combinations	98,131,132
Dynamic PET metrics (ΔSUVmax, early response quantification)†	Quantitative change in SUVmax/other PET metrics after early cycles	Early SUVmax reductions outperform visual DS in aggressive lymphomas; may provide a quantitative metric of early chemoimmunotherapy response in cHL	102,133,134
AI/ML models based on PET/CT	ML algorithms using TMTV, texture, dissemination (Dmax), clinical covariates	Predict 2-year EFS and 5-year PFS/OS; suggest personalized treatment adaptation; may refine risk even in patients with interim PET negativity	135, 136, 137
Deep-learning tools for automated DS	CNNs and automated SUV quantification (liver/mediastinal reference)	Standardize PET response criteria across centers; reduce interobserver variability in DS assignment, thereby improving the robustness of PET-based predictive models	138

ATAC-seq, assay for transposase-accessible chromatin with sequencing; CNNs, convolutional neural networks; Dmax, maximum tumor dissemination; EFS, event-free survival; FFPE, formali fixed paraffin embedded; IF, immunofluorescence; ML, machine learning; MRD, minimal residual disease; RHL-4S, relapsed Hodgkin lymphoma 4-spatial; scRNA-seq, single-cell RNA sequencing; TLG, total lesion glycolysis; Tr1, type 1 regulatory; tTLG, total tumor lesion glycolisis; WES, whole-exome sequencing; WGS, whole-genome sequencing.

^∗Integrating these epigenetic layers with transcriptomic, proteomic, and spatial single-cell data is expected to refine multiomic signatures that distinguish durable responders from patients harboring treatment-resistant HRS/TME clones.

^†Quantitative PET tumor burden metrics (eg, MTV and TLG) may provide complementary prognostic information during PD-1 therapy by capturing overall changes in metabolic burden beyond single-lesion DS, and early reductions in these volumetric parameters have been associated with subsequent response in small immunotherapy-treated HL cohorts.¹³⁹

Multiomic profiling

Multiomic profiling in cHL aims to integrate information from HRS cells and the TME across tissue and blood to derive robust predictors of treatment response. Bulk and tissue-based gene expression models such as RHL30 in R/R cHL and pediatric TME-based signatures already capture distinct inflammatory vs immunosuppressive programs with prognostic impact after salvage therapy and frontline chemotherapy.125, 126, 127^,140 Deep phenotyping with single-cell RNA sequencing, time-of-flight mass cytometry, imaging mass cytometry, and multicolor immunofluorescence links HRS genotypes (eg, 9p24.1/PD-L1, MHC-II loss) to specific T-cell and macrophage niches and spatial architectures that track with post-ASCT outcome, effectively providing “spatial transcriptomic” surrogates of resistant ecosystems.54, 55, 56^,77,89 In parallel, serum cytokine/chemokine panels (eg, TARC, IL-10, soluble CD163) and other soluble proteins, together with ctDNA-based genotyping and minimal residual disease tracking, offer proteomic and genomic readouts that correlate with tumor burden, total metabolic tumor volume (TMTV), and early molecular response.¹²⁹ Systematic DNA methylation and chromatin-accessibility mapping in cHL is still emerging.141, 142, 143

Radiomic and imaging biomarkers

Radiomics and advanced PET-derived imaging biomarkers are rapidly emerging as powerful tools for quantifying tumor biology and predicting immunotherapy outcomes in lymphoma. Radiomics enables extraction of high-dimensional quantitative features, ranging from semantic descriptors of size and shape to agnostic first-, second-, and higher-order texture metrics, that capture intratumoral heterogeneity not discernible by visual inspection.^129,130

Quantitative PET/CT-derived imaging biomarkers provide a complementary layer of prognostic information beyond conventional visual Deauville scoring (DS) and may help refine prediction of immune responsiveness in cHL. Standardized uptake value (SUV) metrics, particularly SUVmax, offer a simple measure of metabolic aggressiveness and have been explored to distinguish indolent from aggressive lymphoma and to flag histologic transformation, although baseline SUV alone has important technical and biological limitations.^133,134,144 More integrative volumetric indices, such MTV and total lesion glycolysis, better capture overall metabolic tumor burden; high baseline TMTV total lesion glycolysis have consistently been associated with inferior PFS and OS across multiple lymphoma subtypes, including HL; although its prognostic impact appears attenuated in patients treated with escalated BEACOPP. Despite variability in segmentation methods and cut-off definitions, high TMTV consistently identifies patients with biologically aggressive disease, supporting the integration of PET-derived tumor burden metrics with clinical risk factors in future cHL-tailored response-adaptation strategies.^98,131,132 Dynamic PET metrics, including early changes in SUVmax after a few cycles of chemoimmunotherapy, have shown superior discriminatory power over purely visual DS in aggressive lymphomas,¹⁰² supporting the concept that early metabolic response integrates both tumor-intrinsic chemosensitivity and evolving host-tumor interactions.

In patients with cHL treated with immunotherapies, fluorodeoxyglucose (FDG)–PET interpretation requires immunotherapy-specific context because immune activation can transiently increase lesion size and/or FDG uptake (tumor flare/pseudoprogression) and can even generate new FDG-avid lesions despite clinical improvement.^145,146 This creates a risk of misclassifying early scans as progression if conventional Lugano criteria are applied without confirmation. Accordingly, lymphoma-adapted immunotherapy response frameworks such as LYRIC introduce an indeterminate response category and recommend multidisciplinary correlation and repeat imaging (or biopsy when feasible) to distinguish inflammation-related changes from true progression.^139,147

AI and computational models

Artificial intelligence (AI) and computational imaging models are emerging research tools for risk stratification and response assessment in HL. Most approaches build on FDG PET/CT by automating or augmenting quantitative biomarkers such as MTV and lesion dissemination. In early-stage HL, baseline MTV, derived from semiautomated segmentation, has been shown to predict 5-year PFS and OS, improving baseline risk classification beyond traditional clinical indices.¹³⁵ In HL, pretreatment PET/CT-based machine learning models combining TMTV, dissemination metrics, and radiomic features have achieved strong discriminatory performance for 2-year event-free survival, suggesting a role for individualized treatment adaptation.¹³⁶ Lesion dissemination, quantified as the maximum distance between the 2 farthest PET-avid lesions (Dmax), also provides independent prognostic information, even in patients with interim PET negativity treated with ABVD-like regimens.¹³⁷ For response assessment, deep-learning models that automatically quantify liver and mediastinal SUV metrics enable more reproducible DS calculation, potentially standardizing PET-based response criteria across centers.¹³⁸

Recent machine learning frameworks integrate clinical variables, genomic features, and imaging-derived biomarkers to model immunotherapy outcomes, outperforming single biomarkers such as PD-L1 or tumor mutational burden alone.²⁷ Notably, AI systems such as SCORPIO and LORIS have demonstrated improved discrimination of responders vs nonresponders to PD-1 blockade across multiple tumor types, with area-under-the-curve values exceeding those of traditional biomarkers.^148,149 Imaging-based AI approaches, including radiomics, aim to capture immune-related tumor features and metabolic heterogeneity that may reflect immune activation or resistance.²⁷ However, a major limitation remains the lack of robust external validation and lymphoma-specific data sets, and most models currently demonstrate reduced performance outside their development cohorts.²⁷

Conclusions

Established predictors, such as baseline disease burden, previous treatment intensity, and clinical inflammatory indices, remain central to risk stratification. These clinical variables are complemented by pathologic and immunophenotypic correlates, including PD-L1 expression, MHC-I/II status, and 9p24.1 copy number alterations, that directly inform the likelihood of response to PD-1 blockade. Increasingly, the TME itself has emerged as a critical determinant: macrophage polarization, T-cell exhaustion signatures, and suppressive spatial niches consistently associate with treatment failure, whereas high-resolution spatial and single-cell technologies now delineate prognostically significant immune ecosystems that shape therapeutic outcomes. Circulating biomarkers add a dynamic, minimally invasive layer to this predictive framework.

Despite this progress, several biological mechanisms continue to complicate effective immunosurveillance and facilitate immune escape. Loss of antigen presentation through B2M mutations, CIITA abnormalities, or defective HLA-DM loading, combined with pervasive PD-L1 expression on both HRS cells and TAMs, reinforces profound local immunosuppression. These lesions converge to shape primary or acquired resistance and underscore the need for rationally designed combination therapies that simultaneously address multiple resistance nodes. Furthermore, there is a need to validate the molecular and cellular pathways that potentially underlie resistance (Table 4; which focuses on putative mechanisms of resistance and immune escape) and which promise to improve the accuracy of patient stratification.

Table 4.

Predictors and mechanisms of resistance to immune therapies in cHL that need to be validated

Predictor/mechanism	Level (clinical/molecular/TME/circulating)	Proposed mechanism of resistance/adverse association	References
High baseline disease burden, advanced stage (eg, stage IV), high TMTV	Clinical/imaging	Reflects aggressive biology and extensive immunosuppressive niches; consistently associated with inferior PFS/OS and likely reduced durability of ICI responses	96,98
Multiple previous therapies and ASCT failure	Clinical	Treatment-exhausted disease with selection of resistant clones and damaged host immunity; associated with lower response rates and shorter duration of benefit from PD-1 blockade	4
Short remission interval/refractory to previous lines	Clinical	Indicates intrinsically resistant disease biology that often overlaps with immune escape mechanisms and poor responsiveness to ICIs	30,97
Poor performance status and frailty	Clinical	Limits ability to deliver optimal therapy; may reflect systemic inflammation and compromised immunity, reducing ICI effectiveness	101,117
B2M mutations and loss of MHC-I	Tumor genomic/antigen presentation	B2M loss leads to absent MHC-I, preventing CD8+ T-cell recognition; associated with shorter PFS and a “hide” strategy of immune escape; may blunt responses to ICIs that rely on CD8+ effectors	73,77,101
CIITA rearrangements and loss/mislocalization of MHC-II	Tumor genomic/antigen presentation	CIITA lesions reduce MHC-II expression; MHC-II loss or cytoplasmic mislocalization impairs CD4+ T-cell help and cytotoxic CD4+ responses; independent adverse prognostic factor and linked to immunosuppressive T-cell niches	55,59,86,88
HLA-DM loss and CLIP-restricted antigen loading	Tumor antigen processing	Defective antigen loading leads to presentation of CLIP instead of tumor peptides, severely limiting effective CD4+ T-cell priming despite preserved MHC-II surface expression	87,88
CD58 deletions or inactivating mutations	Tumor genomic/NK–T-cell interactions	Loss of CD58 reduces susceptibility to NK cell–mediated killing and alters interactions with CD4+ T cells, enabling immune escape	73,78
9p24.1 amplification in the context of MHC loss	Tumor genomic/checkpoint ligand	Amplification drives high PD-L1/PD-L2 expression; when combined with MHC-I loss, associated with particularly poor outcomes, indicating synergistic “hide and defend” immune evasion and potential resistance to PD-1 blockade	19,77,89
Constitutive JAK-STAT activation (eg, JAK2 coamplification, IL-4R mutations)	Tumor signaling	Promotes PD-L1 overexpression and secretion of chemokines such as CCL17, attracting Tregs and reinforcing an M2-skewed TME; supports immune suppression and resistance	19,35,38,76
High PD-L1 expression on HRS cells and TAMs (including via trogocytosis)	TME/checkpoint ligand	Creates PD-1/PD-L1–rich niches that inhibit T-cell and NK cell function; whereas PD-L1 predicts benefit from PD-1 blockade, extreme ligand abundance within a profoundly suppressive milieu may also mark dependence on this axis and adaptive resistance	52,82, 83, 84
Macrophage-dominated, M2-skewed TME	TME	High CD68+/CD163+ macrophage infiltrates support tumor survival, secrete suppressive cytokines, and correlate with treatment failure; such TMEs may be less amenable to monotherapy ICIs	48,103,104,120
Exhausted T-cell infiltrates with high PD-1/LAG-3/TIM-3/TIGIT	TME	Deeply exhausted effector T cells show impaired function; persistent or dominant exhaustion signatures suggest that PD-1 monotherapy may be insufficient and that alternative/checkpoint coblockade is needed	55,93,106
LAG-3+ Tr1-like cells clustering around MHC-II-negative HRS cells	TME/spatial	Spatial coupling of MHC-II–negative HRS with LAG-3+ Tr1 cells creates highly suppressive niches that limit effective antitumor immunity and may underpin resistance to PD-1 blockade	54,55,95
Age-dependent myeloid and EBV-related immunosuppression	TME/clinical	Older EBV+ cHL shows enriched FOXP3+ Tregs and immunosuppressive cytokines; age-related accumulation of PD-L1+/TIM-3+ mononuclear phagocytes further increases checkpoint ligand availability and may reduce ICI efficacy	41,63,66,67
Persistent ctDNA despite therapy (lack of early decline/clearance)	Circulating	Failure to achieve substantial ctDNA reduction or MRD negativity indicates resistant clones and predicts relapse or nonresponse to chemoimmunotherapy, even when PET appears improved	8,108, 109, 110
ctDNA mutations in B2M, TNFRSF14, KDM2B under anti–PD-1 therapy	Circulating/genomic	These mutations have been linked to acquired resistance to PD-1 blockade in cHL, reflecting evolution of immune-evasive subclones during treatment	8,33
Elevated and persistent inflammatory cytokines (IL-6, IL-10, TNF-α, etc)	Circulating	Chronic systemic inflammation and immune dysregulation correlate with inferior survival and may represent a host-level barrier to durable immune surveillance under ICIs	49,50,111
Adverse PET dynamics (high baseline SUV/TMTV, poor early ΔSUVmax)	Imaging	High metabolic burden and inadequate early metabolic response (limited ΔSUVmax) mark aggressive, less chemotherapy-sensitive and likely less ICI-sensitive disease	98,102,131

IL-4R, IL-4 receptor; MRD, minimal residual disease; TNF-α, tumor necrosis factor α; TNFRSF14, tumor necrosis factor receptor superfamily member 14; Tr1, type 1 regulatory.

Conflict-of-interest disclosure: A.C. is a member of the World Health Organization International Agency for Research on Cancer (IARC) Monograph Working Group on Biological Agents, contributing to the IARC Monograph on the Evaluation of Carcinogenic Risks to Humans. Biological Agents volume 100b, A Review of Human Carcinogens. The remaining authors declare no competing financial interests.