Types and Locations of Cells That Respond to PD-1/PD-L1 Blockade

During acute antigen exposure, such as acute infections and vaccinations (Figure2A, left panel), naïve CD8⁺ T cells clonally expand and differentiate into two distinct subsets of effector cells capable of cytotoxicity and cytokine production, known as KLRG1^hi CD127^lo terminal effector T (T_EFF) cells and KLRG1^lo CD127^hi memory precursor T (T_MP) cells.^[42^] Following antigen clearance and resolution of inflammation, over 90% of T_EFF cells undergo apoptosis to avoid excessive immune response,^[43^] while T_MP cells give rise to long‐lived memory T (T_MEM) cells, establishing immune memory.^[42^]

Under circumstances with persistent antigen stimulation, such as chronic infections or tumors (Figure 2A, right panel), the main effector immune response is generated by a distinct subset of T_EX cells. T_EX cells represent a functionally impaired lineage that is transcriptionally and epigenetically distinct from T_EFF cells and T_MEM cells.^[44, 45, 46^] They display crippled effector functions, progressive loss of proliferative capability, and stepwise upregulation of IR expression. T_EX cells acquire a fixed chromatin state during exhaustion known as “epigenetic scarring” early after T cell activation, which renders them hyporesponsive even after antigen elimination.^[44, 45, 46^] The compromised yet critical immune response mounted by T_EX cells represents a delicate tipping point in balancing disease control and tissue damage during the persistent disease course.^[47, 48, 49^] However, T‐cell exhaustion is also one of the major mechanisms employed by tumors to evade immune surveillance.

T_EX cells maintain a hierarchal organization, consisting of stem‐like progenitor exhausted (T_EX^prog) cells, intermediate/transitory exhausted T (T_EX^int/trans) cells, and terminal exhausted (T_EX^term) cells (Figure 2B). T_EX^prog cells express higher levels of stem/memory‐related genes and exhibit greater persistence and recall response compared to T_EX^term cells.^[21, 23, 30, 50, 51, 52, 53^] They are also more polyfunctional regarding cytokine production, which corresponds to their lower degree of differentiation, while T_EX^term cells show a superior ability to degranulate, produce granzymes and perforin, and perform in vitro cytolysis.^[21, 23, 30, 50, 52, 53^] T_EX^prog cells also contain heterogeneous subsets with different levels of proliferative and metabolic activity.^[34, 54^] T_EX^int/trans cells show an intermediate level of stem/memory‐like and exhaustion features and can transition between T_EX^prog cells and T_EX^term cells.^[37, 55, 56^] Importantly, they display superior effector functions than both subsets.^[37, 55, 56^] Markers used to delineate these T_EX subsets vary across studies (Table1). Generally, T_EX^prog cells express higher levels of T‐bet,^[23^] TCF1,^[50, 51^] SLAMF6/Ly108 in mice (surface surrogates of TCF1), and CXCR5.^[29, 30, 40, 57^] This hierarchal relationship was initially described in mouse models of lymphocytic choriomeningitis virus (LCMV) infection and later confirmed in mouse cancer models and patients with various cancer types.^[29, 57, 58, 59, 60, 61, 62, 63, 64, 65^] Most studies propose a linear model in which T_EX^prog cells self‐renew and differentiate into T_EX^term cells through an intermediate/transitory state. Along the differentiation process, stem‐like features gradually diminish, and exhaustion features become more prominent. Notably, some studies suggest a branched trajectory in which either T_EX^int/trans cells^[55^] or T_EX^prog cells^[54, 56^] could differentiate into a more cytotoxic subset. This cytotoxic T_EX subset shares transcriptional, epigenetic, and proteomic features with T_EFF cells, and shows superior granzyme B production and direct killing capacity in vitro compared to the T_EX subsets described earlier.^[54, 55, 56, 66^] Further studies are needed to explore the relationship between these two developmental paths.

In the context of immune memory, it was conventionally believed that chronic infection and cancer preclude the formation of T_MEM cells^[67, 68^] until the recent discovery of tumor‐specific memory T (T_TSM) cells in tumor‐draining lymph nodes (TdLNs) (Figure 2A, right panel). T_TSM cells have not committed to the exhaustion lineage but could differentiate into and sustain the T_EX population in TdLNs and the TME.^[40, 69^] They show robust antigen‐independent self‐renewal and antigen‐dependent recall capabilities comparable to T_MEM cells generated during acute viral infection and superior to T_EX cells.^[40, 69^] Additionally, a subset of T_EX cells has been observed to possess significant stem/memory‐like characteristics.^[67^] Their ability to proliferate and produce inflammatory cytokines is only slightly weaker than stem cell‐like memory T (T_SCM) cells and central memory T (T_CM) cells.^[67^] T_EX cells also express markers of conventional circulating memory T cells (T_SCM: CCR7⁺ CD45RO^‐ CD95⁺, T_CM: CCR7⁺ CD45RO⁺ CD95⁺).^[67^] Thus, even under chronic antigen exposure, certain T cells, whether exhausted or not, show memory features and may be considered memory cells.

In addition to circulating memory cells that recirculate between lymph organs and peripheral tissues, tissue‐resident memory T (T_RM) cells that primarily patrol peripheral sites also contribute to anti‐tumor immune memory (Figure 2A, right panel). T_RM cells express tissue retention markers CD49a, CD69, and CD103 while lacking the expression of lymph‐homing receptor CD62L and the transcription factor (TF) TCF1,^[70^] distinguishing them from the aforementioned T_TSM and T_EX^prog cells. T_RM cells contribute substantially to the intra‐tumoral CD8⁺ T cell population (CD103⁺, ≈70%^[57, 71, 72^]; CD69⁺ CD103⁺, ≈30%^[73, 74, 75^]), and are enriched with tumor reactivity.^[55, 71, 76, 77^] Notably, intra‐tumoral T_RM cells are not spared from exhaustion. They express intermediate levels of exhaustion‐associated markers and TFs^[62, 71, 77, 78^] and are functionally impaired compared to T_RM cells in healthy tissues.^[79^] The relationship between T_RM cells and T_EX cells, including whether T_RM cells display an exhausted scenario similar to circulating memory T cells and whether they also contribute to the reported T_EX cells, requires further investigation.^[61, 65^]

Among T_EX subpopulations, PD‐1/PD‐L1 blockade preferentially expands T_EX^prog cells, which corresponds to their higher proliferative ability (Figure3A). Transfer experiments have shown that only T_EX^prog cells can expand during PD‐1/PD‐L1 blockade, and only the transfusion of T_EX^prog cells could enhance disease control.^[21, 30, 52, 54^] Beltra JC, et al. demonstrated that although the intermediate subset shows the most significant expansion during PD‐1 blockade, truly proliferating responders are the progenitor cells.^[37^] Consistently, CD62L⁺ T_EX^prog cells, which represent a quiescent T_EX^prog subpopulation with high proliferative potential, exhibit the most significant proliferation in response to PD‐1/PD‐L1 blockade, exceeding CD62L^‐ T_EX^prog cells (an actively proliferating subpopulation) and T_EX^term cells.^[54^] These observations suggest that less differentiated cells expand more greatly during PD‐1/PD‐L1 blockade compared to their more differentiated counterparts, leading to an increase in the latter.

In addition to promoting the proliferation of stem‐like T_EX cells (Figure 3B, left panel), PD‐1/PD‐L1 blockade promotes their terminal exhaustion and may generate a more cytotoxic subset (Figure 3B, left panel). PD‐1 blockade has been shown to increase the conversion of virus‐specific CXCR5⁺ T_EX^prog cells to CXCR5^‐ T_EX^term cells (CXCR5^‐/CXCR5⁺ ratio: 3.2 vs. 35.2) in a transfer experiment in chronically infected mice.^[30^] The increased terminal conversion has also been observed in tumors, where PD‐1 blockade increases the proportion of TCF1⁺ Tim3^‐ T_EX^prog cells converting to the TCF1^‐ Tim3⁺ T_EX^term counterpart (57% vs. 67%).^[80^] Furthermore, PD‐1/PD‐L1 blockade may affect the differentiation path of T_EX cells (Figure 3B, left panel). Anti‐PD‐1 treatment may skew the differentiation of T_EX^prog cells toward a more cytotoxic terminal state, generating a subset of CXCL13⁺ “better effectors”.^[85^] This phenomenon has only been observed in responsive patients, indicating its importance for the efficacy of PD‐1/PD‐L1 blockade.^[85^] However, it is reported that only when combined with cis IL‐2R agonism can PD‐1 blockade enhance the effector function of TCF1⁺ T_EX^prog cells.^[86^] Although this discrepancy may be attributed to different tumor models and experimental settings, it implies the need for further confirmation of the effects of PD‐1/PD‐L1 blockade and influencing factors.

The specific T_EX subset that shows the most significant accumulation in response to PD‐1/PD‐L1 blockade can vary across studies and cancer types. Liu B, et al. compared the pre‐ and post‐treatment CD8⁺ T cells based on single‐cell RNA sequencing (scRNA‐seq) datasets from different tumor types and demonstrated that PD‐1/PD‐L1 blockade induces increased CXCL13⁺ T_EX^prog cells in responsive melanoma and non‐small cell lung cancer (NSCLC), but not in responsive skin basal cell carcinoma (BCC) and squamous cell carcinoma (SCC), where T_EX^term cells accounted for the majority of post‐treatment tumor‐infiltrating T cells.^[39^] This variable pattern may be attributed to different levels of immunosuppression in the TME. When immunosuppression is low (as assessed by the expression scores of IRs), PD‐1/PD‐L1 blockade is more likely to induce an increase in progenitor cells, whereas it induces more terminally differentiated cells when immunosuppression is high.^[39^] Consistent with this, in chronic infections, which are less immunosuppressive and have a less exhausted pool of antigen‐specific T cells compared to tumors, T_EX^prog cells are the subset that displays the most significant increase in proportion and number during PD‐1/PD‐L1 blockade.^[34, 37^] Similar results are observed in tumors with low‐level immunosuppression, such as melanoma and NSCLC.^[39^] Furthermore, different treatments may contribute to these variable patterns. It was observed in three breast cancer cohorts that patients receiving chemotherapy and anti‐PD‐1 treatment showed an increased proportion of T_EX^prog cells, while patients receiving single‐agent anti‐PD‐1 treatment showed scarce and unchanged T_EX^prog cells in post‐treatment responsive tumors.^[87^] This could be explained by the augmented T cell priming and generation of T_EX^prog cells induced by enhanced tumor immunogenicity through increased antigen expression^[88^] and release^[89^] as well as enhanced major histocompatibility complex (MHC)‐I‐mediated antigen presentation.^[90^]

Although T_EX^prog cells are crucial for sustaining anti‐tumor immunity and mediating response toward PD‐1/PD‐L1 blockade, defective terminal differentiation of stem‐like CD8⁺ T cells may limit endogenous anti‐tumor immunity^[91^] and the efficacy of tumor vaccine.^[92^] Thus, it is essential to find the optimal balance between sustainability and magnitude of anti‐tumor response mediated by less‐ and more‐differentiated T_EX cells, respectively. Furthermore, exploring the differential effects of other therapeutic agents and endogenous factors on different subsets is an important area of investigation.

Recent studies have emphasized the importance of circulating and tissue‐resident memory responses in effective anti‐tumor immunity,^[93^] both of which could be enhanced by immune checkpoint blockade (ICB)^[94, 95^] (Figure 3B, left panel).

Circulating memory T cells possess the greatest potential in responding to PD‐1/PD‐L1 blockade. Huang Q, et al. reported that cells that expand most significantly during treatment are the non‐exhausted T_TSM cells, which show greater responsiveness than T_EX^prog cells, the focus of previous studies.^[40, 69^] When an equal number of TCF1⁺ TOX^‐ T_TSM cells and TCF1⁺ TOX⁺ T_EX^prog cells were transferred into tumor‐bearing mice receiving anti‐PD‐L1 treatment, TCF1⁺ TOX^‐ T_TSM cells showed the most pronounced expansion.^[40^] Moreover, within the second tumor transplanted into mice that had received PD‐L1 blockade and adoptive transfer of either T_TSM cells or T_EX^prog cells, T_TSM cells exhibited a more significant expansion than T_EX^prog cells when compared to pre‐existing tumor‐infiltrating T cells.^[40^] On the other hand, some T_EX^prog cells, as mentioned earlier, are highly similar to T_MEM cells.^[67^] However, further exploration is needed to investigate which subset of T_EX cells can be considered bona fide memory cells to accurize our understanding of the memory response toward cancer.

T_RM cells, which express PD‐1 and become dysfunctional in the TME,^[62, 71, 77, 78, 79^] may also be reinvigorated by PD‐1/PD‐L1 blockade. Accumulation of CD103⁺ CD8⁺ T cells has been observed following PD‐1/PD‐L1 blockade in melanoma,^[74^] NSCLC,^[71, 78^] and metastatic vaginal melanoma patients^[96^] as well as in murine liver cancer,^[97^] esophageal cancer,^[98^] and breast cancer models.^[73^] CD8⁺ T cells responsive to neoadjuvant PD‐1/PD‐L1 blockade have been reported to express a tissue‐resident memory gene program,^[78, 95^] which is further enriched during treatment.^[95^] The predictive value of the frequency or density of T_RM cells and T_RM signature also suggests the responsiveness of T_RM cells.^[71, 73, 74, 99, 100^] Notably, increased T_RM cells may have two possible origins: 1) tissue‐resident cells that are already present in the local microenvironment before tumor development, and 2) circulating cells that infiltrate the TME and subsequently differentiate into T_RM cells.^[94, 101^] Similar to circulating memory T cells, T_RM cells are heterogeneous, displaying different levels of immune checkpoint expression and differentiation.^[72, 78, 102, 103^] In ovarian cancer patients, T_RM‐like tumor‐infiltrating lymphocytes (TILs) follow a differentiation trajectory from stem‐like to terminally exhausted cells.^[76^] Consistently, Blimp1^lo Id3^hi CD8⁺ T cells and Blimp1^hi Id3^lo CD8⁺ T cells with tissue‐resident gene signatures respectively share transcriptomic similarities with memory‐like cells (e.g., T_CM, T_MP, and T_EX^prog cells) and terminally differentiated effector‐like cells (e.g., T_EM, T_EFF, and T_EX^term cells).^[104^] Therefore, it is likely that T_RM subsets with relatively more stem/memory‐like characteristics may have a higher responsiveness to PD‐1/PD‐L1 blockade. However, current understanding has been limited regarding the exact origin of intra‐tumoral T_RM cells, their heterogeneity, and how they may respond to ICB, suggesting the need for further investigation.

T_EX^term cells, which constitute the largest proportion of intra‐tumoral tumor‐reactive T cells, can hardly be reinvigorated.^[45, 105, 106, 107, 108^] Meanwhile, T_TSM cells and T_EX^prog cells, major responders to PD‐1/PD‐L1 blockade, are strictly^[40, 69^] and preferentially resident in LNs,^[56, 59, 61^] respectively.

Notably, certain LN‐related cellular structures within the TME are associated with the localization of T_EX^prog cells, supporting the notion of preferential lymph residency of stem‐like T cells. T_EX^prog cells are reported to co‐localize with antigen‐presenting cells (APCs) in the TME.^[63, 81, 109^] Such APC niches resemble the T‐cell zone in LNs, where naïve T cells are primed by dendritic cells (DCs).^[66^] Many studies have further characterized APCs in such cellular structures. MregDCs, which express high levels of maturation‐related, costimulatory, and MHC‐II molecules, have been found to enrich near CD8⁺ T_EX^prog cells in human liver cancer^[85^] and lung cancer.^[110^] They enable the optimal priming of CD8⁺ T cells and express high levels of cytokines that promote the survival of CD8⁺ T cells.^[85^] Another study reported the presence of type 1 conventional dendritic cells (cDC1s) near TCF1⁺ stem‐like CD8⁺ T cells.^[111^] These cDC1s also express high levels of MHC‐II molecules and secrete T cell‐chemo‐attractants CXCL9 and CXCL10, promoting T‐cell response.^[111^] In addition to APCs, the optimal priming of CD8⁺ T cells within LNs requires help from CD4⁺ T cells.^[112^] Correspondingly, CD4⁺ T cells have been reported as critical components of stem‐like T cell‐related structures. In hepatocellular carcinoma patients receiving anti‐PD‐1 treatment, CXCL13⁺ helper T (T_H) cells are arranged in close proximity with CD8⁺ T_EX^prog cells and mregDCs, especially in responsive patients.^[85^] They can sustain the survival of CD8⁺ T_EX cells and produce chemo‐attractants for T_EX^prog cells.^[85^] A recent study suggested that intra‐tumoral immune cell triads involving CD4⁺ T cells, CD8⁺ T cells, and DCs are indispensable for effective ICB and T cell transfer therapy.^[113^] CD4⁺ T cells within the triad reprogram CD8⁺ T cells toward a progenitor‐like state, alleviating exhaustion and enhancing cytotoxicity of intra‐tumoral CD8⁺ T cells.^[113^] Nevertheless, it is also reported that the proliferation and phenotype of T_EX^prog cells remain unaffected in the absence of CD4⁺ T cell help.^[114^] These divergent findings suggest the potential heterogeneity of intra‐tumoral T‐cell zone‐like structures. CCL19⁺ fibroblasts are recognized as markers of some APC niches with putative interactions with CD4⁺ T cells and B cells.^[110^] Similarly, fibroblastic reticular cells are critical to the organization and function of the T‐cell zone.^[115^]

Tertiary lymphoid structures (TLS) are B‐cell follicle‐like immune cell aggregates found in tumors.^[116^] Intra‐tumoral CD8⁺ T cells are frequently located inside or surrounding TLS,^[59, 117, 118^] especially the stem‐like TCF1⁺ CD8⁺ T cells.^[81^] Mechanistically, tumor‐reactive CD8⁺ T cells express TLS‐forming CXCL13,^[87^] and T_EX^prog cells are characterized by the expression of the CXCL13 receptor, CXCR5.^[116^] As tumor‐reactive CD8⁺ T cells become exhausted, they upregulate CXCL13 expression and initiate TLS formation. Reciprocally, TLS recruits and maintains CXCR5⁺ T_EX^prog cells, forming a positive feedforward loop.

Overall, studies have provided preliminary insights into the cellular crosstalk within these cellular structures responsible for the presence of T_EX^prog cells. Factors facilitating the formation of these structures and how cellular interactions promote the generation and maintenance of stem‐like T cells are areas worthy of investigation.

Surgical removal of TdLNs or pharmacologically blocking T cell egression from TdLNs abolishes the efficacy of PD‐1/PD‐L1 blockade, even in mouse tumor models that are sensitive to immunotherapy.^[119, 120, 121^] Interestingly, the diminished immune activation and therapeutic efficacy can be reproduced by the adoptive transfer of lymph‐resident T_TSM cells.^[119^] Selective administration of anti‐PD‐1/PD‐L1 agents to TdLNs induces immune reprogramming in the TME and tumor control comparable to systemic administration.^[122^] In three different mouse models of melanoma and breast cancer, intradermal administration targeting TdLNs results in higher intra‐tumoral concentrations of anti‐PD‐1 antibodies, elevated T cell responses in both primary and secondary tumors, and enhanced tumor growth control than systemic administration.^[123^] Similarly, both intraperitoneal and intranodal injection of anti‐PD‐L1 antibodies effectively suppress tumor growth, whereas intra‐tumoral injection has shown limited efficacy.^[40^] The importance of LNs in generating a systemic anti‐tumor immune response has also been proposed in other anti‐tumor therapeutic modalities, including chemotherapy‐induced immunogenic cell death and antigen spread^[124^] and radiotherapy‐induced abscopal effects.^[125, 126^]

Specifically, DCs in TdLNs emerge as critical mediators of LN‐dependent immune activation.^[127, 128^] As demonstrated by longitudinal sampling and scRNA‐seq, cDC1s maintain a reservoir of TCF1⁺ CD8⁺ T cells within the LNs, and the frequency of cDC1s correlates with the strength of anti‐tumor immune response.^[121^] Although both macrophages and DCs express high levels of PD‐L1 in LNs, only the depletion of DCs or selective deletion of PD‐L1 in DCs significantly impairs the efficacy of anti‐PD‐L1 treatment.^[122, 129^]

These findings further underscore LNs as important immune base stations in mounting effective anti‐tumor immune responses and response to PD‐1/PD‐L1 blockade.

The immune activation detected in peripheral blood can predict treatment response in patients receiving ICB. The peripheral T cell proliferative response early after treatment, as reflected by the proportion of Ki67⁺ CD8⁺ T cells,^[130^] TCR diversity and T cell clone size,^[131^] and the number of neoantigen‐reactive CD8⁺ T cells,^[132^] correlates with patients’ response to PD‐1 blockade. Gene signatures of clones that exhibit parallel expansion in both the blood and TME are associated with an improved response to anti‐PD‐1 treatment.^[102^]

Neoantigen‐specific CD8⁺ T cells are enriched in the peripheral circulation during ICB,^[133, 134^] and T cell proliferation dynamics in the peripheral blood have been documented. In a patient who experienced a strong cellular and pathological response to neoadjuvant immunotherapy, the abundance of treatment‐responsive T cells peaked two weeks after treatment initiation in circulation and gradually decreased to an undetectable level, possibly due to T cell homing to malignant tissues where they exert their cytotoxic functions.^[95^] A similar dynamic has been observed in an NSCLC patient who experienced a pathological complete response after neoadjuvant anti‐PD‐1 treatment, with a transient peripheral T‐cell expansion.^[78^] The proportion of common TCR clonotypes between the TME and peripheral blood also significantly increases two to four weeks after the initiation of neoadjuvant anti‐PD‐1 treatment, and the proportion of shared clonotypes correlates with the patient's pathological response.^[135^]

These observations suggest that the peripheral blood may serve as an important route for the anti‐tumor immune response induced by PD‐1/PD‐L1 blockade to reach the TME.

The origin of intra‐tumoral treatment‐expanded T cells can be attributed to two possible sources: 1) local expansion of pre‐existing T cells and 2) replenishment with peripheral T cells (Figure 3B, left panel).

After labeling peripheral T cells with 5‐Bromo‐2‐deoxyUridine (BrdU), accumulation of BrdU⁺ T cells within the TME has been observed after anti‐PD‐L1 treatment,^[40^] suggesting the infiltration of T cells from the peripheral blood into the TME during ICB. The aforementioned TCR overlap between the TME and peripheral blood^[35, 78, 95^] also implies that these intra‐tumoral treatment‐expanded T cells are blood‐related. From a dynamic perspective, novel TCR clonotypes emerge and the TCR repertoire expands during PD‐1/PD‐L1 blockade,^[120^] indicating the mobilization of T cells from outside the TME. Tracing the dynamic of every single T cell clonotype, Yost KE, et al. discovered that the majority of post‐treatment exhausted (84%) and expanded (68%) T cells have novel TCR clonotypes absent from pre‐treatment samples.^[35^] Similarly, post‐treatment T cell clones are predominantly composed of novel T cell clonotypes in two cohorts of BCC patients.^[39^] In some patients, the ratio of novel clonotypes to re‐emerged clonotypes reaches ≈90% and novel T_EX clones account for ≈99% of total T_EX clones.^[39^] The emergence of novel TCR clonotypes during PD‐1/PD‐L1 blockade is termed “clonal replacement” (Figure 3B, left panel). Conversely, it is reported in several cancer types that the majority of treatment‐expanded/post‐treatment exhausted T cells are derived from clonotypes present in pre‐treatment samples.^[39, 95, 136^] In a study by Magen A, et al.,^[68^] the proportion of TCRs shared between pre‐ and post‐treatment tumors was higher than that between post‐treatment tumors and either peripheral blood or LNs, suggesting a predominant local response.^[85^] The expansion of pre‐existing TCR clonotypes induced by PD‐1/PD‐L1 blockade is termed “clonal revival”^[39^] (Figure 3B, left panel).

Heterogeneity of immunological background may underlie the variable contributions of clonal replacement and clonal revival in different studies. The frequency of re‐emerged (pre‐existing) T_EX clones positively correlates with the number of baseline T_EX clonotypes in NSCLC, melanoma, BCC, and SCC.^[39^] A meta‐analysis of public scRNA‐seq datasets from two NSCLC, three breast cancer, a BCC, and a SCC cohorts also demonstrated that a low level of baseline infiltration correlates with a high abundance of post‐treatment new clones.^[87^] PD‐1/PD‐L1 blockade primarily enhances the expansion and infiltration of T cells derived from pre‐existing clonotypes in “hot” tumors, which already contain a large pool of tumor‐reactive T cells before treatment, whereas it boosts the process of novel T cell clonotypes in “cold” tumors lacking pre‐existing responding cells.^[137^] Additionally, it has been proposed that longer intervals between treatment initiation and biopsy correlate with the emergence of novel clonotypes.^[136^] It could be explained by that pre‐existing clonotypes expand immediately after anti‐PD‐1 treatment, while novel clonotypes take longer to arise and infiltrate the tumor,^[138^] which aligns with the increase in peripheral tumor‐reactive CD8⁺ T cells two to four weeks after treatment initiation.^[78, 87, 95, 135^] However, a follow‐up analysis of NSCLC patients receiving biopsy at different post‐treatment time points showed that it is the baseline clonotype abundance rather than the sampling time gap that underlies the frequency of novel clonotypes.^[87^] Generally, current evidence suggests that baseline infiltration may be a dominant determinant of the clonal dynamics of T cells upon PD‐1/PD‐L1 blockade, with a lower pre‐treatment T cell abundance linking to a higher proportion of novel T cell clonotypes. Given limited samples available for analysis, future studies are expected to confirm the hypothesis and explore confounding factors such as the sampling time.

Overall, both systemic and local immune responses could be provoked by PD‐1/PD‐L1 blockade, albeit with different relative contributions. It is important to note that studies based on TCR analysis alone do not provide definitive evidence for the exact source of intra‐tumoral treatment‐expanded T cells, as T cells with pre‐existing TCR clonotypes can either expand locally or be replenished from peripheral blood. Further investigation applying robust cell tracing analysis is expected to chart T cell dynamics during treatment and evaluate the relative contributions of local expansion and peripheral replenishment. Another question that arises is the exact origin(s) of the replenished T cells from the peripheral blood. Based on the studies mentioned above, it is reasonable to conclude that these responding cells primarily, if not exclusively, originate from LNs. However, contributions from other possible sources of T cells, such as the spleen, thymus, bone marrow, and nearby healthy tissue, cannot be ruled out.

As CD8⁺ T cells are widely recognized as the main fighters against cancer, most research discussed in previous sections has focused on CD8⁺ T cells or T cells in general. However, the importance of CD4⁺ T cells in anti‐tumor immunity and their role in the response to PD‐1/PD‐L1 blockade has been increasingly recognized.^[139^]

Among classical T_H cells, type 1 helper T (T_H1) cells have received significant research interest in cancer.^[112, 140^] T_H1 cells, characterized by the TF T‐bet, can secrete IFNγ and TNFα, thereby activating other immune cells, such as natural killer (NK) cells and cytotoxic CD8⁺ T cells.^[141^] Follicular helper T (T_FH) cells play a role in both humoral immunity via stimulating the proliferation and differentiation of B cells and cellular immunity via enhancing the function of CD8⁺ T cells through IL‐21 secretion.^[141^] T_FH cells are considered memory‐like. They can effectively self‐renew to maintain homeostasis without antigen stimulation^[142, 143^] and are preferentially sustained under chronic antigen exposure.^[61, 143, 144, 145^] In recent years, cytotoxic CD4⁺ T cells capable of secreting granzymes and perforin have been discovered in mice and humans. As they express key features of different T_H cell subsets, such as T_H1 cells^[146^] and T_FH cells,^[147, 148, 149, 150^] further exploration is required to determine the developmental pathway and reliable markers of this specific cell subset and its relationship with conventional T_H classification.

Resembling CD8⁺ T cells, CD4⁺ helper T cells and cytotoxic T cells become exhausted during chronic antigen exposure, as manifested by impaired proliferation capacity, inhibited cell cycle, co‐expression of multiple IRs, and loss of helper and cytotoxic functions.^[77, 151, 152, 153, 154^] PD‐1/PD‐L1 blockade induces expansion of T_H1 cells,^[155, 156, 157^] T_FH cells,^[35, 41, 120, 136, 158, 159, 160^] CXCL13⁺ T_H cells,^[85^] and cytotoxic CD4⁺ T cells.^[155, 161^] It also strengthens T‐bet expression and restores T_H1 prototypical cytokine production of IFNγ and TNFα.^[154, 155, 156, 162^] Restored T_FH functions have also been reported by PD‐1/PD‐L1 blockade.^[163, 164^] Additionally, gene signatures of CXCL13⁺ CD4⁺ T_H1‐like cells^[87, 165, 166^] and cytotoxic CD4⁺ T cells^[161, 167^] are associated with the response toward PD‐1/PD‐L1 blockade.

Recent research has revealed the heterogeneity of CD4⁺ T cell subsets in chronic LCMV infection, which is in stark similarity to that of CD8⁺ T_EX cells.^[51, 168, 169^] CD4⁺ progenitor/memory‐like T cells can self‐renew, differentiate into both T_FH cells and other effector CD4⁺ T cells, and sustain the anti‐viral immune response. Furthermore, in a mouse mesothelioma model, CD4⁺ T_EX^prog cells are reported to be transcriptionally similar to CD8⁺ T_EX^prog cells and increase during PD‐1/PD‐L1 blockade.^[122^] Notably, a subset of non‐cytotoxic T_FH cells also exhibit a stem‐like progenitor exhaustion phenotype.^[41^] They can differentiate into T_FH‐like cytotoxic CD4⁺ T cells, which bear a terminally differentiated exhaustion phenotype. PD‐1 blockade can activate non‐cytotoxic progenitor T_FH cells but not T_FH‐like cytotoxic cells. While there is an accumulation of correlative evidence, unlike in the case of CD8⁺ T cells, robust evidence confirming which CD4⁺ T cell subset specifically responds to PD‐1/PD‐L1 blockade is still lacking.

Another major population of intra‐tumoral CD4⁺ T cells is regulatory T (T_REG) cells. They play a crucial role in preventing excess immune responses while also limiting anti‐tumor immunity.^[170, 171^] Similar to other T cell subsets, PD‐1 expression can be induced on T_REG cells upon activation,^[172^] raising the question of whether and how PD‐1/PD‐L1 blockade affects T_REG cells. Current studies have yielded mixed results. PD‐1 blockade has been found to impair the functional maturation of T_REG cells, leading to enhanced tumor control,^[173^] while it has also been reported to enhance the proliferation and suppressive functions of T_REG cells, which result in hyper‐progressive diseases.^[174, 175, 176^] The proportion of T_REG cells can either increase,^[156^] decrease,^[39, 177^] or remain stable^[178^] during treatment. Notably, it is recently proposed that the increase in T_REG cells under PD‐1/PD‐L1 blockade may be a secondary effect of increased secretion of IL‐2 by CD8+ T cells upon treatment.^[179^] A more in‐depth study of the response pattern of T_REG cells is expected.

In general, current studies have provided abundant yet preliminary phenotypic data suggesting an exhaustion‐like state acquired by CD4⁺ T cells and their response to PD‐1/PD‐L1 blockade. Further investigation can be approached from two main perspectives. First, the research outlines and methodologies used in studies of CD8⁺ T cells can be applied to CD4⁺ T cells to investigate their behavior and the underlying molecular mechanisms. Second, special attention should be given to the unique features of CD4⁺ T cells, such as their unique cell clustering and biology.

Although non‐T immunocytes are commonly known as sources of PD‐L1, they can express PD‐1 and may respond to PD‐1/PD‐L1 blockade. For example, targeted deletion of PD‐1 in myeloid progenitor cells induces more differentiated macrophages and DCs and promotes their differentiation into effector cells capable of phagocytosis and antigen presentation.^[180^] This effect is at least partially mediated by activating ERK1/2 and mTOR pathways and rewiring cholesterol metabolism.^[180^] Over 50% of tumor‐associated macrophages express PD‐1,^[181^] which inhibits their phagocytic activity against cancer cells and induces an immunosuppressive M2‐like program.^[181, 182, 183^] Correspondingly, PD‐1/PD‐L1 blockade promotes M1 polarization, improving tumor control and survival.^[181, 182, 183^] PD‐1 blockade increases IL‐10 secretion by tumor‐infiltrating dendritic cells (TIDCs),^[184, 185^] which blunts their antigen‐presenting function.^[186^] The PD‐1 expression level on TIDCs is further augmented by the induced IL‐10 secretion, forming an immunosuppressive feedforward loop and resulting in PD‐1 blockade resistance.^[184, 185^] PD‐1 expression on B cells is induced upon B‐cell receptor triggering, and it could identify a subset of regulatory B cells capable of producing IL‐10.^[187, 188^] Disrupting this axis on B cells enhances their activation, proliferation, and production of inflammatory cytokines.^[187^] Although NK cells do not intrinsically express PD‐1, they can trogocytose PD‐1 from leukemia cells via SLAM receptors when activated.^[189^] This acquisition of PD‐1 can suppress their maturation, proliferation, degranulation, and cytokine production and promote their apoptosis, leading to impaired tumor control in mouse models and poorer patient outcomes.^[190, 191^] Blockade of this axis induces a strong NK cell‐mediated immune response, which is critical to successful immunotherapy.^[190, 191^] Expansion of NK cells following immunotherapy has been observed in the TME, peripheral blood, and LNs in NSCLC patients.^[157^]

PD‐1 is expressed across a broad range of tumor cells.^[192^] However, the effects of PD‐1 on tumor cells are not yet fully understood. It has been observed in different tumor cell types that the intrinsic activity of PD‐1 inhibits the proliferation and survival of tumor cells through the activation of AKT and ERK1/2 pathways.^[193^] Blocking this axis promotes tumor growth, which may mediate treatment resistance.^[193^] However, contrasting results have also been reported that the PD‐1 receptor expressed on tumor cells promotes their growth via activating mTOR signaling.^[194, 195^] Targeted antibodies reduce tumor growth in an immune‐independent way.^[194, 195^] These findings create a “tumor cell‐intrinsic PD‐1/PD‐L1 paradox”, which may be explained by distinct signaling pathways utilized by different tumors.

Overall, there is increasing evidence supporting the existence of PD‐1/PD‐L1 blockade responders beyond T cells. However, most studies to date have only reported phenotypic, functional, and/or quantitative changes in these cells during PD‐1/PD‐L1 blockade. The correlative and implicative nature of current evidence calls for a more in‐depth characterization of the response of these non‐T cell responders. Notably, it is crucial to determine whether the observed differences are a direct effect of PD‐1/PD‐L1 blockade on these cells or a secondary effect resulting from the impact of PD‐1/PD‐L1 blockade on other components. Additionally, it is important to determine the relative contribution of responses produced by different cell types to the efficacy of anti‐PD‐1/PD‐L1 treatment.

The identification of stem‐like T_EX^prog cells and T_TSM cells in LNs as major responders to PD‐1/PD‐L1 blockade explains the suboptimal performance of current TME‐based predictors^[10^] and offers insights into identifying better predictors (Figure4A, left panel). Factors that restrain the ability of LNs to generate an effective anti‐tumor immune response, such as a high density of PD‐1/PD‐L1 interaction^[119, 122^] and the presence of metastasized malignant cells,^[66^] correlate with patients’ response to ICB. In addition, T cell proliferative response early after treatment in the peripheral blood can help distinguish responsive patients.^[102, 130, 132^] Plausible parameters include the number/proportion of antigen‐specific T cells, the expression level of proliferation markers, and changes in the TCR repertoire diversity. In a neoadjuvant setting, the frequency of activated blood CD8⁺ T cells in the blood, particularly the less‐differentiated PD‐1⁺ KLRG1^‐ CD8⁺ T cells, is strongly associated with the patients’ pathological response.^[95^] Given that T cell local expansion also accounts for a substantial proportion of cellular response to PD‐1/PD‐L1 blockade, it is also worth considering intra‐tumoral progenitor‐related features for response prediction. For example, in esophageal cancer patients receiving neoadjuvant PD‐1 blockade, the T_EX^prog cell signature score has shown higher predictive sensitivity and specificity for treatment response compared to PD‐L1 expression.^[267^]

As discussed in the previous section, many emerging immunomodulatory therapies are likely to synergize with PD‐1/PD‐L1 blockade via expanding responders or rewiring non‐responders (Figure 4A, middle panel). Promising examples include therapies targeting additional IRs, co‐stimulatory molecules, and cytokines, cancer vaccines, gene therapy, and therapies promoting T cell homing and infiltration. The knowledge gained from understanding the identity of PD‐1/PD‐L1 blockade responders and their response dynamics can serve as a clue for identifying optimal combinations.

Targeted delivery to responders in a tissue‐specific and/or cell‐specific manner is promising to improve the performance of anti‐PD‐1/PD‐L1 agents (Figure 4A, right panel). Preclinical studies have shown that intradermal administration and low‐dose intrapleural administration (1/80 of systemic dose) enable the preferential accumulation of ICIs in TdLNs compared to systemic administration, and have demonstrated comparable anti‐tumor effects in three different subcutaneous tumor models.^[122, 123^] Similar TdLN‐targeting effect and TME reprogramming can also be achieved by nanoneedle‐based transdermal administration of anti‐PD‐1 antibodies in a mouse melanoma model.^[268^] Currently, both intravenous and subcutaneous administration have been approved for different ICIs, with equivocal safety and efficacy. The feasibility of these local delivery strategies applied in animal studies in clinical settings remains to be investigated. Nanotechnology represents a useful platform enabling the selective accumulation of ICIs in the TME. For instance, pH‐responsive^[269^] and ROS‐responsive^[270^] nanocarriers of anti‐PD‐L1 therapeutics induced selectively higher concentrations in the TME and better therapeutic efficacy. Chimeric antigen receptor (CAR) T cells can also be engineered to secrete PD‐1/PD‐L1 blocking molecules at tumor sites.^[271, 272, 273, 274^] T cell‐based delivery has demonstrated superior anti‐tumor efficacy and safety than combined cell transfer and checkpoint blockade.^[271, 272, 273, 274^] Whether these TME‐targeted delivery techniques can be extrapolated into the targeted delivery of ICIs to LNs requires further exploration. On the other hand, to avoid the potentially immunosuppressive response mediated by additional responders, such as T_REG^[174^] and tumor cells,^[193^] cell‐specific agents targeting the PD‐1 receptor on a specific cell subset have the potential to enhance treatment efficacy. In the context of autoimmune diseases, researchers have engineered a bispecific molecule consisting of a PD‐1 agonist and a recombinant TCR.^[275^] It remains inactive in its soluble form, capable of activating the PD‐1 receptor on activated T cells localized to pancreatic β cells and inducing localized immune suppression without impairing systemic immunity in type I diabetes patients.^[275^] Antibody‐modified nanoparticles are ideal tools for cell type‐specific delivery. In vivo targeted reprogramming of CAR T cells could be achieved through CD3‐targeting nanoparticles.^[276^] CD8 T cell‐targeted nanoparticles encapsulating immunostimulatory agents demonstrate higher efficacy in tumor growth control and prolonging survival.^[277^] Notably, targeting functional markers, such as PD‐1, can also be achieved.^[277^] These discoveries have the potential to be applied in targeting specific responders of immunotherapies, calling for future research endeavors.

Utilizing cells with stem‐ and memory‐like properties holds the potential to enhance the persistence and efficacy of cell therapy. The proportion of CD39^‐ CD69^‐ stem‐like TILs in the TIL products correlates with complete cancer regression and TIL persistence in melanoma patients. Transfer of CD39^‐ CD69^‐ cells into tumor‐bearing mice yielded significantly better tumor control than transferring CD39⁺ CD69⁺ cells, even at a tenfold concentration.^[82^] Similarly, characteristics of less‐differentiated CAR T cell products positively correlate with long‐term anti‐tumor activity and clinical response.^[278^] Stem‐like CAR T cells have shown better leukemia control while reducing adverse events.^[279^]

Strategies such as the in vitro induction or selection of stem‐like cell products^[280^] can be employed to improve current cell therapy (Figure 4B). Given that LNs harbor T_TSM cells and T_EX^prog cells, cell products derived from cells sorted from LNs may outperform current cell products derived from either tumor (e.g., TIL therapy) or peripheral blood (e.g., CAR immune cell therapy, TCR‐T cell therapy). In addition, concurrent use of PD‐1/PD‐L1 blocking agents and reprogramming exhaustion‐related epigenetic and metabolic shifts may antagonize T cell exhaustion and maintain the progenitor phenotype in vivo.^[281^]

By analyzing the influence of TdLN dissection,^[40, 66^] blocking cell migration from LNs pharmacologically,^[119, 122^] and TdLN‐targeted ICB^[40, 122, 123^] on the therapeutic efficacy of ICIs, multiple research teams have demonstrated the importance of TdLNs as the basis of anti‐tumor immune response. Consistently, LN‐damaging strategies that aim at lowering the risk for metastases, such as surgical removal^[282^] or irradiation,^[283, 284^] impair anti‐tumor immunity (Figure 4C, left panel). The radiotherapy‐induced abscopal effect, which refers to the anti‐tumor effect observed in distant, non‐irradiated tumors due to systemic immune awakening, is also reduced by LN irradiation.^[126, 285^] Therefore, the management of TdLNs requires a balance between lowering the risk of metastasis and preserving their potential to generate anti‐tumor immune response (Figure 4C, left panel).

Immunotherapy before surgical removal of TdLNs is capable of pre‐stimulating anti‐tumor immune response and preserves enough numbers of activated T cells to function against remaining tumors after LN removal (Figure 4C, right panel). Preoperative immunotherapy resulted in a higher number of related immune cells in resected tumor tissue and patients’ blood, which persisted till at least one month post‐surgery.^[66^] This approach, mostly combined with concurrent chemotherapy or chemoradiotherapy, is being investigated in various cancer types, and preliminary results are promising.^[286^] Sequencing lymphatic‐preserving radiation with immunotherapy is another feasible approach to address this challenge. Sequential LN‐sparing radiation and ICB induce an optimal and durable treatment response that controls both the primary and metastatic diseases and prevents tumor rechallenge.^[125, 284, 287^] Although LN‐sparing radiation may increase the regional recurrence rate (Figure 4C, right panel), receiving subsequent surgical resection or including sentinel LNs into the irradiation field may reverse this effect, enabling both the activation of T cell response and the elimination of metastatic tumor cells.^[283^] In addition to pre‐stimulating anti‐tumor immunity before LN damage by radiation or surgery, cancer vaccination capable of activating remaining LNs or adoptive cell therapy to directly provide tumor‐specific T cells are also promising to complement immune response.

The dose of radiation plays a critical role in its immunological effects. Low‐dose radiotherapy (LDRT) can reprogram the TME, turning an immunologically “cold” tumor “hot”.^[288^] Whether LDRT of LNs can also activate anti‐tumor immunity in LNs or rescue the ability of metastasized LNs (MetLNs) to generate immune response is worth investigating (Figure 4C, right panel). Preliminary clinical results of combined ICB and neoadjuvant LDRT delivered only to the primary tumor and metastatic LNs in head and neck squamous cell carcinoma are promising,^[289, 290^] although the specific contribution of LN‐directed LDRT needs further exploration. MetLNs display a larger fraction occupied by T_EX^term cells, indicating a failure of T_EX^prog cells to become activated and differentiate into T_EX^term cells. They also have a higher number and activity of T_REG cells after ICI treatment than non‐metastasized LNs.^[66^] This suggests that the immune activation induced by PD‐1/PD‐L1 blockade is impaired in MetLNs, and selective surgical removal or irradiation strategies may be finetuned based on the status of LNs (Figure 4C, right panel).

Overall, further investigation is needed to test the viability of these theoretical conjectures in animal models and humans. It is also important to determine whether these preclinical findings can be extrapolated to the clinical setting.

Lu Gan is an associate professor at the Research Laboratory of Emergency Medicine, West China Hospital, Sichuan University. Her research interest mainly focuses on exosome‐mediated cellular crosstalk in various pathological conditions.

Jianxin Xue is a professor at the Division of Thoracic Tumor Multimodality Treatment, West China Hospital, Sichuan University. His research interest mainly focuses on cancer immunotherapy and radiotherapy, particularly their mechanistic interactions and related combination therapies.