TIGIT and PD-L1 co-blockade promotes clonal expansion of multipotent, non-exhausted antitumor T cells by facilitating co-stimulation

Nutsch, Katherine; Banta, Karl L.; Wu, Thomas D.; Tran, Charles W.; Mittman, Stephanie; Duong, Ellen; Nabet, Barzin Y.; Qu, Yan; Williams, Katherine; Müller, Sören; Patil, Namrata S.; Chiang, Eugene Y.; Mellman, Ira

doi:10.1038/s43018-024-00870-6

Download PDF

Article
Open access
Published: 16 December 2024

TIGIT and PD-L1 co-blockade promotes clonal expansion of multipotent, non-exhausted antitumor T cells by facilitating co-stimulation

Nature Cancer volume 5, pages 1834–1851 (2024)Cite this article

11k Accesses
23 Altmetric
Metrics details

Subjects

A Publisher Correction to this article was published on 17 January 2025

This article has been updated

Abstract

Blockade of immune checkpoints PD-1 and TIGIT has demonstrated activity in mouse tumor models and human patients with cancer. Although these coinhibitory receptors can restrict signaling in CD8⁺ T cells by regulating their associated co-stimulatory receptors CD28 and CD226, the functional consequences of combining PD-1 and TIGIT blockade remain poorly characterized. In mouse tumor models, we show that combination blockade elicited CD226-driven clonal expansion of tumor antigen-specific CD8⁺ T cells. The expanded clones emerged from a population of stem-like cells in draining lymph nodes, entering the blood as a previously unidentified single-phenotype, multiclonal population. Upon reaching the tumor, these transiting cells expanded further and differentiated into effector or exhausted T cells, with combination blockade restricting entry into the exhaustion pathway by favoring co-stimulation. Thus, PD-1 and TIGIT inhibition helps shape the repertoire of tumor-reactive CD8⁺ T cells in draining lymph nodes and determines their immunological fate in the tumor to enhance therapeutic benefit. Analysis of clinical trial samples suggests a similar mechanism may also occur in patients with cancer.

The immune checkpoint B7x expands tumor-infiltrating Tregs and promotes resistance to anti-CTLA-4 therapy

Article Open access 06 May 2022

IFNγ signaling in cytotoxic T cells restricts anti-tumor responses by inhibiting the maintenance and diversity of intra-tumoral stem-like T cells

Article Open access 19 January 2023

CD25-T_reg-depleting antibodies preserving IL-2 signaling on effector T cells enhance effector activation and antitumor immunity

Article 09 November 2020

Main

Immunotherapies targeting the PD-1–PD-L1 pathway have shown promise, but only ~30% of patients achieve durable responses, necessitating new strategies such as combinations targeting multiple or novel immune checkpoint receptors¹. TIGIT (T cell immunoreceptor with Ig and immunoreceptor tyrosine-based inhibitory domains) has garnered attention due to efficacy in early clinical trials using blocking antibodies against both TIGIT and PD-L1 (ref. ²). Recent analysis of the randomized phase 2 CITYSCAPE trial (NCT03563716) evaluating atezolizumab versus anti-TIGIT monoclonal antibody tiragolumab plus atezolizumab in non-small cell lung cancer (NSCLC)³ revealed that high baseline intratumoral macrophages and regulatory T (T_reg) cells were associated with clinical benefit⁴. Although these results suggest that TIGIT–PD-L1 co-blockade reprograms the tumor microenvironment (TME), high levels of CD8⁺ effector T (T_eff) cells were also associated with response.

In CD8⁺ tumor-infiltrating lymphocytes (TILs), TIGIT and PD-1 expression are highly correlated⁵. Whereas PD-1 primarily regulates co-stimulation by CD28, TIGIT and PD-1 together regulate the function of CD226, the activating counterreceptor to TIGIT⁶. Coexpression may define distinct populations of ‘stem cell-like memory (T_scm)’ cells^7,8,9,10 believed to be primary targets of PD-1–PD-L1 blockade in both antitumor and antiviral immunity^11,12,13. Blocking PD-1 signaling may differentiate these progenitors into T cells with cytolytic effector activity against tumor cells, perhaps via a transient population of T precursor exhausted cells (T_pex)^{14,15,16,17,18}. Thus, PD-1 expression may reflect T cell activation status in addition to denoting exhaustion or commitment to exhaustion. It remains uncertain whether T_pex cells give rise to only exhausted T (T_ex) cells in the tumor, whether commitment to the T_ex pathway begins in draining lymph nodes (dLNs) or whether T_ex, T_eff and T effector memory (T_em) cells originate from separate precursors either before or following tumor arrival. The extent to which checkpoint blockade, especially the role of TIGIT blockade alone or in combination with PD-1 blockade, reprograms CD8⁺ T cells already committed to the exhaustion pathway or discourages developmental commitment to exhaustion also remains a key unknown¹⁹.

To inform these questions, we utilized a multicompartment, multi-omics single-cell approach, analyzing over 245,000 T cells. We examined not only the features of CD8⁺ T cells in dLNs and tumors, as has been carried out previously, but also in the blood. Sampling these three critical tissue compartments facilitated insight into the spatial and temporal effects of TIGIT and PD-1 blockade on T cell fate decisions.

Results

Combination treatment requires trafficking of lymphocytes from dLNs to tumor

The coordinate regulation of T cell co-stimulation by PD-1 and TIGIT suggests that both receptors may activate T cells at the same steps and anatomical sites⁹. Using the CT26 syngeneic mouse tumor model, we evaluated the role of dLNs in TIGIT blockade by restricting trafficking of T cells with FTY720, an inhibitor of T cell egress from lymphoid organs²⁰. Consistent with previous observations⁵, the combination of anti-TIGIT with anti-PD-L1 demonstrated therapeutic efficacy, whereas anti-PD-L1 or anti-TIGIT monotherapies had only limited impact on tumor growth (Fig. 1a and Extended Data Fig. 1a). FTY720 reduced the activity of both single-agent anti-TIGIT and TIGIT–PD-L1 co-blockade (Fig. 1a). Similar results were observed in the EO771 tumor model (Extended Data Fig. 1b).

**Fig. 1: Blockade of T cell egress from dLNs reduces efficacy of anti-PD-1 and anti-TIGIT treatment in mouse CT26 tumor model.**

Treatment with anti-PD-L1, anti-TIGIT, or both did not affect total numbers of CD8⁺ T cells, CD4⁺ T cells or T_reg cells in CT26 dLNs or tumor, either with or without FTY720 treatment (Extended Data Fig. 1c). We therefore asked whether checkpoint blockade and FTY720 affected the abundance or distribution of tumor antigen-specific CD8⁺ T cells, identified using tetramers that bind T cell receptors (TCRs) specific for gp70, a tumor-associated, immunodominant retroviral antigen expressed by CT26 cells²¹ (Extended Data Fig. 2a). Anti-TIGIT in combination with anti-PD-L1 increased the fraction of gp70⁺CD8⁺ T cells (P = 0.0138) in dLNs, whereas anti-PD-L1 or anti-TIGIT alone had little effect (Fig. 1b). The addition of FTY720 before combination treatment further increased the frequency of gp70⁺CD8⁺ T cells in dLNs (P = 0.0472), likely reflecting their accumulation in dLNs by preventing T cell egress.

In blood, numbers of gp70⁺CD8⁺ T cells were significantly increased with anti-TIGIT (P = 0.0134) or combination treatment (P < 0.001), but not in FTY720-treated animals (Fig. 1b). In tumors, only the combination of anti-TIGIT and anti-PD-L1 significantly increased the fraction of gp70⁺CD8⁺ T cells (P = 0.0098) (Fig. 1b). As trafficking via blood was blocked, at least some expansion of intratumoral T cells was likely to have occurred locally. Although FTY720-treated mice exhibited a trend toward increased gp70⁺CD8⁺ T cells in tumors, these presumably locally expanded cells were unable to control tumor growth.

We next asked whether antitumor efficacy relied on the continuous recruitment of newly generated T cells from dLNs and blood. Early administration of FTY720 blocked combination efficacy, whereas delaying the blockade of T cell egress until 7 days after combination treatment resulted in only slight impairment in antitumor efficacy (Fig. 1c). Thus, the efficacy of combination checkpoint blockade depends on the induction of tumor-specific CD8⁺ T cells in dLNs that then traffic to and infiltrate tumors via the circulation. Once the newly mobilized T cells seeded tumors, they were sufficient to sustain therapeutic benefit in response to TIGIT–PD-L1 co-blockade.

CD226 has a role in tumor-specific CD8⁺ T cell differentiation

As human TILs in NSCLC differentially express CD226 and CD28 in various CD8⁺ T cell clusters, combination treatment may be required to optimally activate the entire tumor-reactive TIL repertoire⁶. To evaluate the role of CD226 on tumor-specific CD8⁺ T cell subsets, we segregated gp70⁺CD8⁺ T cells based on CD226 expression. Anti-TIGIT alone or in combination with anti-PD-L1 increased the frequency of CD226⁺gp70⁺CD8⁺ T cells in both dLNs and tumors, even with FTY720 treatment (Fig. 2a). Following combination blockade, CD226⁺gp70⁺CD8⁺ T cells were substantially more proliferative (Ki67⁺), but only in dLNs (Fig. 2b,c and Extended Data Fig. 3a). CD226^–gp70⁺CD8⁺ T cell proliferation was not affected by any treatment. Few CD226⁺gp70⁺CD8⁺ T cells in dLNs were naive compared to the CD226^– fraction (Fig. 2b and Extended Data Fig. 3b); only combination treatment increased the frequency of CD226⁺gp70⁺CD8⁺ T cells with a T_eff or T_em phenotype, whereas no effects were observed in the CD226^– population (Fig. 2b and Extended Data Fig. 3c).

**Fig. 2: CD226 expression is a determinant of tumor-specific CD8⁺ T cell differentiation state.**

To further elucidate the effects of checkpoint blockade on activation and differentiation, we measured various markers of T cell states. Slamf6 and TCF1 coexpression delineate T_scm or T_pex cells^9,19. In dLNs, the frequency of these cells in the CD226⁺ fraction was not affected by any treatment, but anti-TIGIT alone or in combination with anti-PD-L1 significantly reduced frequencies in the CD226^– subset (Fig. 2b and Extended Data Fig. 3d; P = 0.0014). By contrast, in tumors, anti-TIGIT and combination treatment increased frequencies of Slamf6⁺TCF1⁺ cells in both CD226⁺ and CD226⁻ subsets (Fig. 2c and Extended Data Fig. 3d; P = 0.0014).

As T cells differentiate from the T_scm or T_pex state, they express immune checkpoints such as Tim3. Combination treatment as well as anti-TIGIT alone increased the frequencies of both CD226⁺ and CD226^–TCF1⁺Tim3⁺gp70⁺CD8⁺ T cells in tumor, whereas effects in the dLNs were limited to the CD226⁺ subset; FTY720 largely abolished these effects (Fig. 2b,c; see Extended Data Fig. 3e for statistics). As T cells further differentiate, they lose expression of TCF1 although transcription of the Tcf7 gene seems to precede the loss of the TCF1 protein itself (compare to Fig. 3b). In the dLNs, a significant increase in the frequency of TCF1^– tumor-specific CD8⁺ T cells was seen in the CD226⁺ fraction with anti-TIGIT or combination treatment; no effect was detected in CD226^– cells (Fig. 2b and Extended Data Fig. 3f).

TOX is a key transcriptional regulator of exhaustion programming and differentiation toward terminal exhaustion^22,23. Treatment with either anti-TIGIT alone or anti-TIGIT plus anti-PD-L1 markedly decreased TOX expression in CD226⁺ but not CD226^–gp70⁺CD8⁺ T cells in dLNs, while decreased TOX expression was seen in both CD226⁺ and CD226^– fractions in tumors; FTY720 seemed to diminish the combination effect on TOX expression in some cases (Fig. 2b,c; see Extended Data Fig. 3g for statistics).

Similar pharmacodynamic effects with combination treatment were seen in the EO771 model, with combination treatment increasing the frequency of CD8⁺ T cells in tumors, promoting CD226 expression on tumor CD8⁺ T cells, and increasing the TCF1⁺Tim3⁺ phenotype while reducing TOX⁺ frequencies (Extended Data Fig. 3h–m).

To assess the effector state of TILs responding to checkpoint blockade, we measured production of the proinflammatory effector cytokines interferon (IFN)γ and tumor necrosis factor (TNF). Single-agent anti-TIGIT and combination treatment increased dual production of IFNγ and TNF in the CD226⁺ fraction of intratumoral CD8⁺ T cells relative to the CD226^– fraction, with FTY720 eliminating this effect, suggesting that T cells derived from the periphery might possess superior effector function (Fig. 2d,e). Assessment of cytokine production by tumor-specific TILs was not possible due to downregulation of TCR upon in vitro stimulation; the effects were observed in PD-1⁺ TILs, indicating that activated or antigen-experienced TILs may be functionally impacted by combination treatment (Fig. 2e).

As anti-TIGIT plus anti-PD-L1 seemed to have more pronounced effects on CD8⁺ T cells expressing CD226, particularly in dLNs, we concurrently treated mice receiving the combination with CD226-blocking monoclonal antibody. As we could not segregate gp70-specific CD8⁺ T cells on the basis of CD226 expression in the presence of the blocking monoclonal antibody, we examined total gp70⁺ cells and could not discern effects on Slamf6⁺TCF1⁺ cells (Fig. 2f and Extended Data Fig. 3n); however, anti-CD226 monoclonal antibody impaired the ability of combination treatment to increase the frequency of TCF1⁺Tim3⁺ tumor-specific CD8⁺ T cells in dLNs and tumors (Fig. 2f and Extended Data Fig. 3o). CD226 blockade also showed a trend toward reducing the ability of combination treatment to drive differentiation to a T_eff/T_em phenotype (Fig. 2f and Extended Data Fig. 3p). Anti-CD226 monoclonal antibody prevented the reduction in TOX-expressing cells in dLNs and to a greater extent in tumors (Fig. 2f and Extended Data Fig. 3q; P = 0.025 and 0.009 respectively).

Taken together, addition of anti-TIGIT to PD-1–PD-L1 blockade initiated distinct differentiation pathways of T_scm or T_pex cells in dLNs in a CD226-dependent fashion. These cells were further expanded in the tumor and developed into polyfunctional effectors. Similarly, the combination prevented upregulation of TOX characteristic of T_pex and T_ex differentiation, again in a CD226-dependent manner.

TIGIT and PD-L1 co-blockade expands different CD8⁺ T cell states

We further examined how co-blockade affects the generation, phenotype and trajectory of tumor-specific T cells using a multi-omics single-cell approach, performing single-cell RNA sequencing (scRNA-seq) and TCR sequencing (scTCR-seq) on T cells from tumors, dLNs and blood. These assays were supplemented by antibody-derived tag sequencing (ADT-seq) with tetramers against gp70 and cellular indexing of transcriptomes and epitopes (CITE-seq) using a panel of 18 cell surface proteins.

Gene expression profiles of a large dataset of 245,675 T cells yielded 24 distinct clusters (Extended Data Fig. 4a), with contributions across treatment groups (Extended Data Fig. 4b), but with some clusters appearing selectively localized to dLNs, blood or tumors (Extended Data Fig. 4c). Effector status, as indicated by granzyme B expression, was confined primarily to CD8⁺ T cells that showed clonal expansion and high ADT counts, a measure of the number of gp70 tetramers bound (Extended Data Fig. 4d–g). CITE-seq provided a complementary characterization of T cell differentiation, effector and memory states based on surface marker expression (Extended Data Fig. 4h).

We obtained greater resolution of CD8⁺ T cell phenotypes by reanalyzing the T cells with high CD8a expression. These 155,496 CD8⁺ T cells comprise one of the largest datasets used for this type of analysis, enabling higher resolution clustering and unprecedented insight into the responses of CD8⁺ T cells to checkpoint inhibition. Twenty distinct CD8⁺ clusters were identified (Fig. 3a,b, Extended Data Fig. 5 and Supplementary Table 1), with contributions consistent across individual mice (Extended Data Fig. 6a), experimental groups (Extended Data Fig. 6b), and clusters associating with specific tissues. Clonal expansion and ADT counts were increased in non-Ccr7 clusters (Extended Data Fig. 6c).

**Fig. 3: Single-cell RNA sequencing identifies distinct CD8⁺ T cell clusters in CT26 tumors, dLNs and blood.**

The clusters exhibited various phenotypes (Fig. 3a):

(a)
four Ccr7 clusters (‘Ccr7.1-4’) characterized by Ccr7, a marker expressed by naive, T_scm and central memory (T_cm) cells but low in cytotoxic CD8⁺ T_eff and T_em cells²⁴, as well as genes associated with T_scm cells such as Sell, Lef1 and Tcf7 (ref. ¹⁸) and also high expression of ribosomal proteins;
(b)
a distinct cluster (‘Early’) characterized by expression of Cd69 and other markers of early T cell activation;
(c)
a distinct ‘Slamf6’ cluster marked by high Slamf6 and Tcf7 expression representative of a T_scm population;
(d)
three Ifit clusters (‘Ifit.1-3’) with hallmarks of interferon response genes indicating activated T cells;
(e)
two Ccl5 clusters (‘Ccl5.1-2’) marked by this chemokine that can exert chemotactic effects on T cells and is associated with CD8⁺ T cell infiltration into tumors²⁵;
(f)
two Cytotox clusters (‘Cytotox.1-2’) exhibiting hallmarks of cytotoxic gene expression as well as genes associated with exhaustion such as Tox and checkpoint inhibitory checkpoint receptors;
(g)
three Cyt/Mit clusters (‘Cyt/Mit.1-3’) that represent proliferating cytotoxic cells;
(h)
two mitotic clusters (‘Mitotic.1-2’) expressing genes associated with mitosis but not genes associated with effector function; and
(i)
two clusters representing dying cells (‘Dying.1-2’).

The Ccl5 clusters shared expression of a number of genes associated with the Cytotox or Cyt/Mit clusters, but lacked properties of exhaustion. CITE-seq protein analysis corroborated the categorization by gene expression (Extended Data Fig. 6d–f). Both scRNA-seq and CITE-seq showed that CD226 expression was most characteristic of Ccl5.1 T cells. CD28 showed some overlapping expression with CD226 but also marked a few distinct clusters, consistent with our previous findings for human NSCLC TILs⁶ (Extended Data Fig. 6d–f). The Ccl5.1 cluster is of particular interest in that it was the only major non-naive cell state found in the blood.

Comparison of our clusters with reference gene signatures from published datasets^16,26,27,28 showed general concordance albeit with more granularity due to the larger sample set used here (Fig. 3c). Of relevance, our Ccl5, Ifit.3 and Cytotox clusters shared strong similarities with the ‘better effectors’ described by others in response to a combination of anti-PD-1 therapy with interleukin (IL)-2 agonists²⁷; however, our Ccl5.1 cluster also corresponded with the ‘Stem-like cluster’ in that study and with the ‘Transitory T_ex cluster’ by Huang and colleagues¹⁶.

We used spliced and unspliced messenger RNA counts to estimate RNA velocities and infer differentiation trajectories. Although the directionality of cell traffic often cannot be assigned confidently from velocity-based trajectories²⁹, visualization results from Li and colleagues using photoactivation have established the in vivo migration of T cells into and out of tumors²⁴. By assigning our clusters to the Li et al. gene expression signatures (Extended Data Fig. 7a), we could ascertain directionality in our analysis. Using control-treated tumor-bearing mice as a reference, RNA velocity patterns differed between dLNs and tumors (Fig. 3d). In dLNs, a major trajectory originated from Early and Ccr7 clusters and yielded Slamf6 cells, which then differentiated into Ifit or Ccl5 cells. In tumors, differentiation progressed from Ccl5 cells through Cytotox cells to Cyt/Mit cells. From there, a second differentiation pathway generated mitotic cells. RNA velocity patterns were similar across treatment groups, indicating that differentiation pathways were not fundamentally affected by the various treatments (Extended Data Fig. 7b,c).

Combination treatment expands Ccl5⁺ tumor-specific CD8⁺ T cells

We then applied our scTCR-seq data to characterize T cells by the expansion of their parent clone, which revealed notable differences across treatment groups, especially when using ADT-seq counts to distinguish gp70⁺ from gp70⁻ cells (Extended Data Fig. 6c). As shown in Fig. 4, cells in dLNs were predominantly singletons (having only one cell expressing a given TCR clonotype) across each cluster, but showed evidence of clonal expansion in the Slamf6 and Ccl5.1 clusters following combination treatment. In contrast, cells in tumor were almost exclusively expanded clones. Although clones were specific to individual mice, these results were not attributable to any single mouse (Extended Data Fig. 8).

Fig. 4: Treatment effects on CD8⁺ T cell cluster and clonal composition in CT26 tumor-bearing mice, and stacked bar graphs of CD8⁺ T cell cluster composition in each tissue under each treatment condition.

On day 0 and 7, T cells in the blood across all treatment groups were largely gp70^– (Fig. 4, bars facing left) and mostly in the immature Ccr7 clusters. In contrast, gp70⁺ T cells in blood (Fig. 4, bars facing right) appeared only on day 7 and were specific to the Ccl5.1 cluster and to the anti-TIGIT and combination treatments. Expanded gp70^– T cells in the Ccl5.1 cluster were observed in the blood as well; these cells could represent bystanders or clones specific to tumor antigens other than gp70. Absolute cell numbers of Ccl5.1 cells in blood were low, perhaps reflecting their transient residence in blood. The appearance of nearly all T cells in blood was blocked by FTY720 treatment, indicating that continuous egress from dLNs is required to supply blood T cells. The origin of Ccl5.1 cells from dLNs was also supported by the accumulation of clonally expanded Ccl5.1 cells in dLNs with FTY720. Tumors, unlike the dLNs or blood, contained relatively large numbers of both clonally expanded gp70⁻ and gp70⁺ TILs in all treatment groups; however, in mice treated with both anti-PD-L1 and anti-TIGIT, this increase was relatively pronounced for gp70⁺ T cells in the Ifit and Ccl5.2 clusters (Fig. 4). The increase in gp70⁺ clones in these clusters was selectively decreased by FTY720 treatment, strongly suggesting that these cells derived proximately from the blood-borne Ccl5.1 population. In FTY720-treated mice, gp70⁺ clones expanded in the Cytotox and Cyt/Mit clusters, indicating that these may derive from pre-existing clones in tumors and expand and differentiate intratumorally in response to combined PD-L1–TIGIT blockade.

Thus, in response to combination treatment, tumor antigen-specific (and possibly also nonspecific) clonotypes expand in the dLNs, exit as Ccl5.1 cells into the blood and continue to expand after arrival in the tumor.

Co-blockade of PD-L1 and TIGIT focuses TCR clonal diversity

We next compared the degree of clonal expansion in dLNs, blood and tumors on day 7 post-treatment, characterizing each clone by its majority cluster at each site (Fig. 5a and Extended Data Fig. 9a). Inhibiting both PD-L1 and TIGIT elicited notable coordinated clonal dynamics. Although only a few clones exhibited large expansions, they did so in each of the three tissue compartments (Fig. 5a). In dLNs, expansion occurred mostly in Ifit.3 and Ccl5.1 cells, while in the tumors, Ccl5.2 cells were preferentially expanded. Combination treatment also resulted in correlated expansion in the blood (Fig. 5a, circle diameters, and Extended Data Fig. 9a). Expansion occurred with single-agent treatment (to a greater extent following anti-TIGIT alone) but was mostly limited to dLNs or tumors and not correlated between the two tissues.

**Fig. 5: Effects of anti-PD-L1, anti-TIGIT and combination treatment on clonal diversity and dual expansion in dLNs and tumors.**

In the presence of FTY720, many clones exhibited dual expansion in dLNs and tumors, but with relatively limited expansion in the blood, suggesting that these dual-expanded clones arose independently in dLNs and tumors.

The most highly expanded clones following combination treatment were gp70⁺, indicated by a high ADT count (Fig. 5b). The relatively few dual-expanded clones in the single-agent treatment groups had low or undetectable gp70 ADT counts, suggesting that they were either ‘bystander’ nontumor-reactive T clones^25,30 or specific for other tumor-associated antigens. In the presence of FTY720, high gp70 ADT counts were also detectable in some dual-expanded clones, as expected if these cells represented pre-existing clones already present in dLNs and tumors before treatment.

As the scatter plots (Fig. 5a,b) depict only the primary cluster type for each clone, we evaluated the composition of the 30 most expanded clones for each treatment group in tumors and matched them to dLNs and blood to study the distribution of individual clones across T cell clusters (Fig. 5c). The largest clones in tumors had measurable counterparts in dLNs but only following combination treatment. In dLNs, these clones consisted predominantly of the Ifit.3, Ccl5.1 and Cytotox.1 populations. The same expanded TCR clones were also found in the blood, again contained almost exclusively in the Ccl5.1 population. FTY720 treatment prevented the appearance of this population.

The picture was quite different with single-agent treatments. CD8⁺ T cells in the tumor following anti-PD-L1 had largely the same composition as the control group, comprised primarily of Cytotox.2 and Cyt/Mit clusters. Expansion of the Cytotox.2 cluster was more pronounced than with other treatments, suggesting that anti-PD-L1 drives T cell differentiation toward this specific state in tumors. Anti-TIGIT, in contrast, promoted a shift in the tumor toward the Ccl5.2 cluster. With single-agent treatment, none of the largest clones in tumor had appreciable counterparts in dLNs or blood.

When we examined clonal expansion separately in each tissue compartment, each treatment had distinct effects on T cell differentiation (Extended Data Fig. 9b,c). In dLNs, anti-TIGIT and combination treatment, but not anti-PD-L1 alone, caused expansion of Ccl5.1 T cells and, to a lesser extent, Mitotic clusters. FTY720 treatment shifted the intralymphatic composition to almost exclusively Ccl5.1, suggesting that these cells accumulated in dLNs as their egress into blood was inhibited. Combination treatment, with or without FTY720, resulted in reduced proportions of the Slamf6 cluster in dLNs, especially in the most expanded clones, suggesting that the Slamf6 (putative T_scm) cluster might be the source from which Ccl5.1 T cells are mobilized.

Anti-PD-L1 and anti-TIGIT differentially reshape CD8⁺ T cell trajectories

We next probed the lineage relationships across CD8⁺ T cell clusters following various treatments. Although we previously generated cellular trajectories using RNA velocity (Fig. 3d), it is apparent that individual clones exhibit complex and distinct expansion behaviors. scTCR-seq unambiguously identifies lineages of T cells, which provides a complementary approach to infer kinetics and differentiation based on the co-occurrence of phenotypes in individual clonotypes within and across tissue compartments. We analyzed co-occurrences of cell phenotypes by tabulating numbers of intraclonal pairs over all clonotypes, plotting only pairs between different clusters (Fig. 6a–c), and interpreting such co-occurrences as steps in differentiation.

**Fig. 6: CD8⁺ T cell cluster relationships within and across tissues of CT26 tumor-bearing mice following anti-TIGIT and/or anti-PD-L1 treatment.**

In dLNs (Fig. 6a), control mice exhibited a predominant differentiation of Slamf6 to the Cytotox.1 phenotype, with little connection to other populations as illustrated by the absence of additional intercluster links. With single-agent treatment, increased differentiation from Slamf6 to the Ccl5.1 phenotype was observed, but with anti-TIGIT further increased differentiation of Ccl5.1 cells into Cytotox.1 and Mitotic.1 cells. Combination treatment produced an even more complex pattern of differentiation, with Ccl5.1 cells also differentiating to Ifit.3 cells, and those co-occurrences being shared across cytotoxic and mitotic clusters. FTY720 treatment resulted in most Slamf6 cells differentiating to Ccl5.1, but then a sharp reduction in Ccl5.1 cells differentiating to other clusters, as indicated by the absence of intercluster links. Intraclonal pairs in dLNs consisted of primarily gp70⁺ specificities across treatment groups, and some gp70^– with control or single-agent treatment. Thus, although Slamf6 (T_scm) cells differentiated to cell states other than Ccl5.1 in dLN, only the Ccl5.1 population entered the blood, seeding tumors with new CD8⁺ T cells.

Co-occurrence profiles were different in tumors compared to dLNs (Fig. 6b). Intraclonal pairs in control tumors showed an origin from the Cytotox.2 phenotype to the Cyt/Mit.1 and Cyt/Mit.2 phenotypes. Anti-PD-L1 had a similar pattern, but with additional co-occurrence of Cytotox.2 with the Ifit.3 and Cytotox.1 clusters. In sharp contrast, anti-TIGIT exhibited an expansion of clones with Ccl5.2 cells that differentiated to Cyt/Mit.2, Cyt/Mit.1 and Cytotox.2 cells; these clones were largely gp70⁻ (blue lines), consistent with the largest clonotypes in that group being gp70⁻ (Fig. 5c). Combination treatment resembled anti-TIGIT monotherapy in terms of Ccl5.2 expansion, but those Ccl5.2 cells differentiated primarily to Cytotox.1 cells. FTY720 treatment produced a complex pattern of co-occurrences among Ccl5.2, Cytotox.1, Cytotox.2, Cyt/Mit.1 and Cyt/Mit.2 clusters, revealing the extent of differentiation within tumors. In contrast with anti-TIGIT treatment, the vast majority of intraclonal pairs in tumors with combination treatment were gp70⁺.

We then tabulated intraclonal pairs from across tissues to determine migration relationships, plotting only co-occurrences between different tissues, but otherwise showing co-occurrences between both same and different clusters (Fig. 6c). In contrast to single-agent therapy, the anti-PD-L1–TIGIT combination facilitated migration of Ccl5.1 cells from dLNs to Ccl5.2 cells in tumor, presumably through blood Ccl5.1 cells, but with co-occurrences from dLNs to blood less apparent because of its relatively low degree of clonal expansion in both compartments (Fig. 4). Co-occurrences were also seen from Ccl5.1 cells in dLNs to Cytotox.1 and Cytotox.2 clusters in tumors, but these were presumably attributable to intratumor differentiation (Fig. 6b). The co-occurrences between Ccl5.1 in dLNs and Ccl5.2 in tumors were also observed in the presence of FTY720, with an absence of blood involvement, indicating that combination treatment may act on pre-existing TILs in tumors that had progenitors remaining in the dLNs.

To visualize these differentiation and migration patterns in the context of gene expression, we projected the co-occurrence data onto our previously computed Uniform Manifold Approximation and Projections (UMAPs). From these plots (Fig. 6d,e), it is apparent that Slamf6 cells (putative T_scm) in dLNs serve as progenitors for Cytotox.1 cells in control and anti-PD-L1 treated mice and for Ccl5.1 cells in other treated mice. These Ccl5.1 cells then migrate into blood, with more frequent migration occurring with anti-TIGIT and combination-treated mice in gp70⁺ clones (Fig. 6d) than gp70⁻ clones (Fig. 6e). In these groups, and especially with combination treatment, the migration links revealed a convergence of multiple clusters from dLNs onto Ccl5.1 cells in blood, and then a divergence from these cells into multiple clusters in tumors. With anti-TIGIT, and to a greater extent with combination therapy, gp70⁺ Ccl5.1 cells in blood then migrated into tumor where they seemed to give rise to the Ccl5.2 phenotype. Ccl5.2 cells differentiated into Cytotox.2 cells, which then differentiated into other cytotoxic and mitotic (precursor exhausted) phenotypes. Differentiation from gp70⁺ Cytotox.2 cells to other phenotypes was greater for anti-PD-L1 and FTY720 treatment, compared to anti-TIGIT and combination treatment. These results suggest that anti-TIGIT and especially combination treatment promote an immune response characterized by an influx of tumor-specific Ccl5.1 T cells, whereas anti-PD-L1 and FTY720 treatment exhibit primarily the differentiation of Cytotox.2 T cells already existing in the tumor.

CD8⁺ T cell cluster association with response to tiragolumab plus atezolizumab

To explore whether these observations inform the clinical setting, we analyzed scRNA-seq data of peripheral blood T cells from a phase 1b study of patients with NSCLC treated with the combination of tiragolumab plus atezolizumab (T + A) (GO30103)³¹. We mapped human CD8⁺ T cells onto the nearest mouse reference CD8⁺ T cell cluster (Extended Data Fig. 10a,b). Patients with a clinical response, evaluated as either complete response (CR) or partial response (PR), compared to nonresponders (stable disease (SD) or progressive disease (PD)), had an increased frequency of CD8⁺ T cells mapping to the Ccl5.1 and Ccl5.2 clusters and a decreased frequency mapping to Ccr7.3 and Ccr7.4 clusters (Extended Data Fig. 10c), consistent with Ccl5 clusters in our mouse models predominating with effective treatment.

To address whether gene signatures derived from the mouse CD8⁺ T cell clusters associated with improved overall survival (OS), we analyzed bulk RNA sequencing (RNA-seq) data from baseline tumor samples from patients in CITYSCAPE³. The top 20 differentially expressed signature genes for each mouse CD8⁺ T cell cluster were used to derive orthologous human gene signature ‘scores’ in CITYSCAPE samples (Supplementary Table 2), which compared patients treated with T + A or placebo plus atezolizumab (P + A). Ccr7.3, Slamf6, Ifit.1, Ifit.2, Ifit.3, Ccl5.2 and Cytotox.2 gene signature scores were significantly higher in CR and PR responders compared to SD and PD nonresponders (Fig. 7a). While all CD8⁺ T cell cluster signatures trended with favorable OS hazard ratios (HRs) in patients treated with T + A compared to P + A (Extended Data Fig. 10d), high expression of Ccr7.3, Slamf6 and Ccl5.1 gene scores associated with significantly improved HRs for OS (HR 0.44 (95% CI 0.22–0.91; P = 0.028), 0.46 (95% CI 0.22–0.95; P = 0.036) and 0.45 (95% CI 0.22–0.90; P = 0.025), respectively), as did low expression of Cytotox.1 and Cyt/Mit.2 (OS HR 0.46 (95% CI 0.23–0.90; P = 0.023) and HR 0.48 (95% CI 0.23–0.98; P = 0.045), respectively). Dichotomization of patients on the basis of high or low cluster gene signature score and by treatment showed that high expression of the Ccl5.2 gene signature trended with increased OS with T + A but not P + A (Extended Data Fig. 10e).

**Fig. 7: Associations of human CD8⁺ T cell clusters and gene signatures with clinical response to tiragolumab plus atezolizumab.**

Gene signatures predominantly associated with response to T + A were characterized by high expression of chemokines or chemokine receptors. We focused on CXCR3, CXCR6 and CCL5, genes that were among the most highly expressed in each of the clusters. High expression of each of these individual genes was associated with response in patients treated with T + A (Fig. 7b) and high expression of CCL5 or CXCR3 was individually associated with favorable OS HR in T + A compared to P + A, outperforming CD8A (OS HR 0.32 (95% CI 0.14–0.73; P = 0.006), 0.41 (95% CI 0.18–0.94; P = 0.035) and 0.43 (95% CI 0.20–0.91; P = 0.027), respectively) (Fig. 7c). CXCR3, CXCR6 and CCL5 were associated with improved OS for T + A, again outperforming CD8A (Fig. 7d).

A composite gene signature score consisted of the average expression of CCL5, CXCR3 and CXCR6 was significantly higher (P = 0.012) in responder CITYSCAPE patients compared to nonresponders (Fig. 7e). A high gene signature score was associated with favorable OS HR in patients treated with T + A (HR 0.43; P = 0.035) compared to P + A, whereas a low signature score did not associate significantly with OS benefit (HR 0.70; P = 0.277) (Fig. 7f). Segregation of patients on the basis of high or low gene signature scores showed that those treated with T + A who had high gene score expression had improved OS compared to patients with a low gene signature (Fig. 7g). The composite gene signature score was also associated with improved progression-free survival and OS in the phase 3 OAK study (NCT02008227) of atezolizumab monotherapy in patients with locally advanced or metastatic previously treated NSCLC³² (Extended Data Fig. 10f).

Thus, our analysis of patients treated with T + A largely recapitulates the findings of anti-TIGIT plus anti-PD-L1 in our mouse tumor studies, providing translational evidence that the events observed in dLNs of tumor-bearing mice may also be detected in human blood and tumors. Furthermore, our study suggests that CD8⁺ T cell quality, as represented by newly arrived cells from dLNs, rather than the mere presence of CD8⁺ T cells in the tumors supplied by the periphery at steady state³³, may be more predictive of response and clinical benefit.

Discussion

Despite the profound influence of checkpoint inhibitors on oncology practice and our understanding of tumor immunity, one important unknown is whether these inhibitors work primarily in dLNs or at the tumor site, and on which populations of cells. By considering T cells not only in dLNs and tumors but also in the blood, we demonstrate that PD-1 and TIGIT act to direct T cell fate at both anatomical sites, with activation, expansion and differentiation beginning in dLNs, but with final determination of progression to T_eff or T_ex cells occurring in tumors, challenging the notion that the trajectory to exhaustion is established at or near the time of priming^9,19,34.

After mobilization with combination therapy, the blood compartment exhibited predominantly a single CD8⁺ T cell population (Ccl5.1 cluster) comprising TCR clonotypes that had expanded in dLNs and that were found in the tumor. Of note, in the tumor these clonotypes were distributed among multiple T cell states. Trajectory analysis based on RNA velocity and lineage tracing of TCR clonotypes suggest that peripheral blood Ccl5.1 cells differentiated into the closely related Ccl5.2 population after reaching the tumor. Thus, polyclonal Ccl5.1 cells can be considered as ‘transit cells’ whose main function is to convey newly expanded T cells to the tumor.

Once in the tumor, the transit cell progeny (Ccl5.2) differentiated along the T_ex or T_eff cell pathways, a decision that seems to be influenced or determined by the degree of co-stimulation available via CD28 and CD226 co-stimulatory receptors. Indeed, CD226 signaling was required to block the expression of TOX in dLNs, and especially in the tumor where nearly 80% of tumor antigen-specific CD8⁺ T cells were otherwise TOX⁺. The prevention of coinhibitory receptor suppression of co-stimulatory receptor signaling by anti-PD-L1 and anti-TIGIT may explain how combination therapy directs differentiation away from the exhaustion pathway. This mechanism is consistent with observations that CD28 and CD226 signaling is under the control of the PD-1 and TIGIT coinhibitory receptors⁶, thus providing an attractive functional link between checkpoint inhibition and the accumulation of T_ex cells in the tumor. It seems likely that dendritic cells (DCs) present in the dLNs and tumors help to determine fate decisions between T_eff and T_ex cells, as DCs present both antigen and co-stimulatory ligands, consistent with recent work^35,36; however, the role of DCs in this proposed mechanism remains to be determined.

Notably, the effects of PD-1 and TIGIT inhibition seem to be distinct. While both facilitated the differentiation of tumor-specific T cell trajectories from the T_scm (Slamf6 cluster) compartment to the Ccl5.1 transit cell population in dLNs, anti-PD-L1 treatment showed differentiation also to the Cytotox.1 phenotype, whereas anti-TIGIT and combination treatment showed a second stage of extensive differentiation from the Ccl5.1 phenotype to other phenotypes. TIGIT blockade, alone or especially in combination with anti-PD-L1, produced far more emigration of tumor antigen-specific Ccl5.1 T cells into the blood than did anti-PD-L1 alone. Once in the tumor, anti-PD-L1 monotherapy showed differentiation mainly of the gp70⁺ Ccl5.2 T cells to Cytotox.2 and then to the more exhausted Cyt/Mit phenotypes, whereas anti-TIGIT monotherapy showed differentiation mainly of the gp70^– T cells also toward exhausted phenotypes. That anti-TIGIT therapy alone seemed to preferentially affect the gp70^– population may be a factor in its relative therapeutic ineffectiveness. Combination therapy showed a coordinated infiltration of gp70⁺ tumor-specific T cells from the blood and less exhaustion of T cells in the tumor, suggesting replenishment by newly arriving T cells into the tumor.

T_scm or T_pex cells have been proposed as targets for PD-1–PD-L1-targeted immunotherapies^11,14,16,17, so they are likely also targets for PD-L1–TIGIT combination. Both are presumed to be precursor populations, consistent with our results, but it is difficult to precisely map our subpopulations to these designations. Nevertheless, our scRNA-seq study has greater resolution relative to previous studies, given its large number of cells studied across a range of effective and ineffective treatments. T_scm cells were originally defined as a CXCR5⁺/TCF1⁺/Slamf6⁺ self-renewing compartment present in dLNs that give rise to all subsequent T cells¹⁹. T_pex are generally defined as cells that have at least some of these features (Slamf6 and TCF1) and also some, but not all, features of exhausted cells; evidence indicates that they are along a continuum of precursors of terminally differentiated T_ex cells^18,19,34. These two populations are often invoked interchangeably. Our evidence suggests that the anti-PD-L1–anti-TIGIT combination works on a precursor population, likely defined by our Slamf6⁺ cluster in dLNs and subsequently the Ccl5.1 and Ccl5.2 clusters in the tumor. The effect of combination treatment, however, enables these populations to give rise to T_eff, not just T_ex cells, and it is these T_eff cells that seem to correlate with effective treatment. Were it possible to conduct the experiments for longer periods, it seems likely that combination treatment would also favor the differentiation of T_em cells in addition to T_eff cells. At this point, there is no single marker that unequivocally defines the Ccl5.1 or Ccl5.2 clusters, precluding experimental validation through methods such as in vivo adoptive cell transfer. Although it will ultimately be important to agree upon a common lexicon, our finding that the anti-PD-L1–anti-TIGIT combination influences CD8⁺ T cell trajectories in a manner dependent on co-stimulatory receptor signaling can be viewed within any of the existing frameworks.

It is noteworthy that features of the T cell populations observed for combination treatment in mice seem to have counterparts in human patients with cancer who respond to combination treatment with T + A. Specifically, markers associated with Ccl5 clusters in mice, which represent newly expanded T cell clones trafficking from dLNs to tumors, were found to be associated with clinical benefit. If the differentiation trajectories influenced by blockade of PD-1–PD-L1 and TIGIT observed in mouse tumor models are also recapitulated in human patients with cancer, then more persistent and durable responses with better survival outcomes may be attained by focusing our therapeutic efforts on generating tumor-reactive effector cells that are either resistant to exhaustion programming or replacements for terminally exhausted cells. Of note, combination of PD-1 blockade with immunostimulatory cytokines such as IL-2 (refs. ^27,37), blockade of immunosuppressive cytokines such as TGF-β³⁸ or co-stimulatory (for example 4-1BB) agonists¹⁸ may also skew T_scm/T_pex differentiation trajectories toward effector and cytotoxic states and/or away from exhaustion. As combination of anti-TIGIT with anti-PD-L1 has the additional mechanism of action of reshaping the TME⁴, higher quality antitumor T cells generated in response to combination treatment will be able to exert their effector function in a less suppressive, more permissive environment. Leveraging combination therapy strategies such as anti-TIGIT with anti-PD-L1 that drive both mechanisms may potentially bring improved clinical benefit for more patients beyond anti-PD-(L)1 alone.

Methods

Ethics statement

All experimental animal studies were conducted under the approval of the Institutional Animal Care and Use Committees (IACUCs) of Genentech Lab Animal Research and were performed in an Association for the Assessment and Accreditation of Laboratory Animal Care-accredited facility. Study design, patient cohort and response assessment for clinical trials GO30103 (NCT02794571)³¹ and CITYSCAPE (NCT03563716)³ have been previously described⁴, with trial protocols approved by the institutional review board or ethics committee at each participating center and complying with Good Clinical Practice guidelines, and studies performed in accordance with the principles of the Declaration of Helsinki, the International Council for Harmonization guidelines for Good Clinical Practice and country-specific laws and regulations, as noted in the originally published clinical trials^3,31. GO30103 (ref. ³¹) was conducted at 13 sites in six countries (Australia, Canada, France, South Korea, Spain and the United States). CITYSCAPE³ was conducted in 41 clinical centers across France, Serbia, South Korea, Spain, Taiwan and the United States. All patients provided written informed consent. An internal monitoring committee reviewed available safety data periodically to make recommendations regarding study conduct to ensure the safety of patients enrolled in the study, as noted in the originally published clinical trials^3,31.

Mice

BALB/c or C57BL/6 mice were purchased from the Charles River or Jackson Laboratories. Mice were housed under specific-pathogen-free conditions at the Genentech animal facility. Mice were maintained in accordance with the Guide for the Care and Use of Laboratory Animals (National Research Council, 2011). Genentech is an American Association of Laboratory Animal Care-accredited facility and all animal activities in this research study were conducted under protocols approved by the Genentech IACUC. Mice were housed in individually ventilated cages within animal rooms maintained under a 14 h–10 h light–dark cycle. Animal rooms were temperature and humidity controlled between 68 and 79 °F (20.0–26.1 °C) and from 30% to 70%, respectively, with 10–15 room air exchanges per hour. Female mice (aged 6–8 weeks) that seemed healthy and free from obvious abnormalities were used for the study.

Cell lines

CT26 (ATCC-CRL2638) and EO771 (ATCC-CRL3461) cell lines (obtained ATCC with corresponding certificates of analysis) were maintained at a dedicated internal cell line facility and tested to be mycoplasma free. CT26 or EO771 cells were cultured in RPMI 1640 medium supplemented with 10% FBS and 100 U ml⁻¹ penicillin and 100 mg ml⁻¹ streptomycin, and grown in a 37 °C humidified, 5% CO₂ incubator.

Syngeneic tumor studies

CT26 tumor studies were performed by inoculating age-matched 6–8-week-old BALB/c female mice with a subcutaneous injection of 0.1 × 10⁶ CT26 cells in 100 µl Hank’s balanced salt solution (HBSS) and Matrigel (BD Biosciences). EO771 tumor studies were performed by inoculating age-matched 6–8-week-old C57BL/6 female mice with an injection into the fifth mammary fat pad of 0.1 × 10⁶ EO771 cells in 100 µl HBSS + Matrigel. Once tumors achieved a mean volume of 150–200 mm³, animals were apportioned into treatment groups and treated with isotype control (anti-gp120 mIgG2a), 10 mg kg⁻¹; anti-PD-L1.mIgG2a LALAPG monoclonal antibody (clone 6E11), 10 mg kg⁻¹ followed by 5 mg kg⁻¹; anti-TIGIT.mIgG2a monoclonal antibody (clone 10A7), 10 mg kg⁻¹; or TIGIT.mIgG2a.LALAPG, 10 mg kg⁻¹; and administered intravenously for the first dose and subsequently intraperitoneally. For the tracking of tumor volume, doses were given three times a week for 3 weeks. For single-cell analyses, the mIgG2a version of anti-TIGIT was used and three doses were given over the course of 1 week. To inhibit trafficking, FTY720 (Cayman Chemical Company, 1 mg kg⁻¹) was administered by daily oral gavage starting day −1 before indicated treatment, or where indicated, day 7 after treatment, and continued until end of study. Tumor volumes were measured and calculated twice per week using the modified ellipsoid formula 1/2 × (length × width²). For pharmacodynamic analyses, mice were killed on day 7 after initial treatment. Tumors were dissociated into single-cell suspensions by using gentleMACS dissociator (Miltenyi Biotec) and enzymatically digested in a buffer containing collagenase D (2 mg ml⁻¹) and DNase (40 U ml⁻¹, Roche). Single-cell suspensions of dLNs were obtained by mechanical dissociation through 40-µm cell strainers and performing red blood cell lysis as needed. Blood was obtained by terminal cardiac puncture and collected in lavender Microtainer Blood Collection Tubes (BD Biosciences, 365974) and subjected to red blood cell lysis. Data collection and analyses from mouse tumor studies were not performed blind to the conditions of the experiment. The maximum tumor size approved by the IACUC was 2,000 mm³. Animals bearing tumors exceeding 2,000 mm³ or showing ulceration were killed following protocols approved by the IACUC. Tumors were measured three times per week. In the case of tumors exceeding 2,000 mm³, tumor measurement was recorded before killing. To minimize the number of mice with tumors exceeding 2,000 mm³, mice were killed if tumors were measured at greater than 1,700 mm³ on any given day, as the tumor growth rate would make it highly likely for the tumor to exceed 2,000 mm³ by the next measurement; however, despite these measures, some tumors grew in excess of 2,000 mm³ between two measurements, as outlined here. In Fig. 1a, 8 mice were killed with tumors >1,700 mm³, 8 mice with tumors >2,000 mm³, 44 mice with ulcerations and 1 mouse for other reason, across the entire study. In Fig. 1c, 5 mice were killed with tumors >2,000 mm³, 11 mice before tumors reached 2,000 mm³ and 1 mouse with ulceration, across the entire study. Details for experimental groups and individual mice are provided in Fig. 1 Source Data. In Extended Data Fig. 1b, 34 mice were killed with tumors >1,700 mm³, 26 mice with tumors >2,000 mm³ and 8 mice with ulcerations, across the entire study. Details of the experimental groups and individual mice are provided in Extended Data Fig. 1 Source Data.

Flow cytometry and FACS sorting

Immune cell phenotyping by flow cytometry was performed on single-cell suspensions from mouse dLNs, tumors and blood obtained and described elsewhere. In brief, dead cells were excluded by using a fixable viability dye. Cell surface markers were stained on ice after tetramer staining. The FoxP3 nuclear staining buffer set (Invitrogen) was then performed using recommended manufacturer’s instructions to detect intracellular or nuclear staining. For intracellular cytokine detection, cells with stimulated for 4 h with Cell Stimulation Cocktail (Invitrogen, 00-4970-93) at 37 °C. After stimulation, cells were stained for surface markers and intracellular factors as described above. For obtaining cells for single-cell analysis, tumors and dLNs were processed into single-cell suspensions as described elsewhere and subjected first to tetramer staining, then surface markers and CITE-seq antibodies together. Processing of blood samples on day 0 before any treatment or on day 7 were first stained with hash-tagged antibodies, then stained with surface markers. Cells were purified by fluorescence-activated cell sorting (FACS) on a Becton Dickinson FACSAria Fusion cell sorter equipped with four lasers (405 nm, 488 nm, 561 nm and 638 nm). A 70-μm nozzle running at 70 psi and 90 kHz was used as the setup for each sort session. Before gating on fluorescence, single cells were gated using forward scatter (FSC-A) and side scatter (SSC-A) (for intact cells) and SSC-W/SSC-H and FSC-W/FSC-H (to ensure that only singlets were sorted). FACS gates were drawn to include only live single cells based on calcein blue AM+ and propidium iodide (Thermo Fisher Scientific). Antibodies used for flow cytometry, cell sorting by FACS or CITE-seq are shown in Supplementary Table 3. Samples were acquired on LSR-Fortessa or BD Symphony Instruments (BD Biosciences) using FACSDiva (v.8.0.1) or Cytek Aurora using SpectroFlo (v.3.0.3) and data were analyzed using FlowJo v.10 or higher version software (Tree Star).

Single-cell RNA-seq and TCR V(D)J clonotype profiling

Processing for single-cell expression (scRNA-seq) and T cell receptor V(D)J clonotypes (scTCR-seq) was carried out using the Chromium Single Cell 5′ Library and Gel Bead kit (10x Genomics), following the manufacturer’s instructions. T cells were isolated from tumors, dLNs and blood from 31 mice. Cell density and viability from each mouse tissue of FACS-sorted CD90⁺ T cells from tumors and blood, or CD90⁺CD44⁺ T cells from dLNs, were determined by hemacytometer. Approximately 6,000–10,000 cells per sample were used for the reverse transcription mastermix. A total of 14 cycles of PCR amplification was performed to obtain sufficient cDNA used for both RNA-seq library generation and TCR V(D)J targeted enrichment followed by V(D)J library generation after Gel Bead-in-Emulsion reverse transcription (GEM-RT) reaction and cleanup. TCR V(D)J enrichment was carried out as per the manufacturer’s user guide using a Chromium Single Cell V(D) J Enrichment kit, human T cell (10x Genomics). Libraries for RNA-seq and V(D)J were prepared following the manufacturer’s user guide (10x Genomics), then profiled using a Bioanalyzer High Sensitivity DNA kit (Agilent Technologies) and quantified with Qubit (Thermo Fisher Scientific). scRNA-seq libraries were sequenced in one lane of HiSeq4000 (Illumina). scTCR V(D)J libraries were tagged with a sample barcode for multiplexed pooling with other libraries, sequenced in both lanes of a HiSeq2500 machine (Illumina) using Rapid Run mode, and then demultiplexed. All sequencing was conducted according to the manufacturer’s specifications (10x Genomics). Detailed information on mice, tissue isolation and batching of samples is provided in Supplementary Table 4.

Pre-processing of single-cell data

Sequencing files from Illumina assays were run through Cell Ranger v.6.1.1 against a transcriptome derived from Ensembl v.2.2.0 for the mouse genome GRCm38. The combined matrix files from the filtered_feature_bc_matrix directory for the RNA and ADT libraries were divided into separate submatrices for each sample, based on 52,636 genes for expression, 6 tetramer barcodes for ADT counts, 24 antibody measurements for CITE-seq and 10 barcodes for multiplexing of the blood samples. Measurements corresponding to various alleles of T cell receptor genes (for example, Trbv1 through Trbv31) were combined into a single gene measurement (Trbv). As blood samples were pooled from several mice based on an encoding scheme that used two multiplex barcodes to identify each mouse, single cells were demultiplexed using the two multiplex barcodes with the highest counts. In cases of a tie for the second-highest multiplex count (4.6% of cells), those single cells could not be assigned to a particular mouse using this method. TCR sequence data from the filtered_contig_annotations.csv files were processed using a custom script that identified clones across multiple tissues in each mouse, based on identical sets of α and β sequences. To handle the blood cells that could not be assigned using the multiplex counts, blood cells with a TCR nucleotide sequence uniquely matching a cell from lymph node or tumor of a mouse in the pool were assigned to the corresponding mouse. ADT barcodes came from 12 distinct tetramers, of which 2 had gp70 antigens and the remaining 10 had a non-gp70 antigen (C28, UV or C142). A cell was assigned to an antigen based on its ADT barcode with the highest count, and was not assigned in cases of ties.

Integration of single-cell expression data

Analysis was performed in the statistical language R v.4.2.0 and with scripts written for Perl v.5.16.3. The single-cell unique molecular identifier (UMI) count matrix for each tumor and lymph node, and each pooled blood sample, was processed using scDblFinder v.1.12.0 to identify and remove doublets, or gel beads containing more than one cell. The remaining singlet count matrices were processed using Seurat v.4.1.1 using the SCTransform function (unless specified otherwise, Seurat functions were run using default parameters). All samples were merged into a single Seurat object, then subjected to a filtering process to remove anomalous or low-quality cells, where 10,584 genes were first identified as each being present in more than 1% of all cells, and then 245,675 of the 260,391 cells were retained because more than 99% of their UMI counts were represented by these genes. Counts of mitochondrial genes were not used for filtering, as such genes are present in T cells at the end of their lifespan due to apoptosis, and not necessarily an indicator of poor-quality cells.

Since the mice in this study were taken from batches on two different dates, we performed batch correction was performed using the Harmony package v.0.1.1 with the batch date as the controlling variable. We calculated principal-component analysis (PCA) cell embeddings following the procedure at https://cran.r-project.org/web/packages/harmony/vignettes/Seurat.html, where we processed the count matrix with the Seurat procedures NormalizeData; FindVariableFeatures using selection.method = ‘vst’ and nfeatures = 2,000; ScaleData and RunPCA on the variable genes with npcs = 30. The dataset was then processed with the procedure RunHarmony and the Seurat procedures RunUMAP and FindNeighbors on the Harmony reduction, and FindClusters to obtain 24 clusters of CD4⁺ and CD8⁺ T cell subtypes.

The reason that we made two calls to SCTransform is as follows. The first call was performed on individual samples before integrating them, a standard practice in Seurat protocols. The second call was required because we used Harmony, which excels at batch correction, rather than the Seurat integration procedure. Harmony requires a PCA, and this in turn requires finding variable genes and scaling the data, as described above. While SCTransform is essentially equivalent to NormalizeData, FindVariableGenes and ScaleData, we used these three steps separately as this is a recommended procedure for Harmony. Furthermore, we used the procedure FindVariableFeatures with the parameter selection.method = ‘vst’ because this is recommended in the above-referenced website and the SCTransform method does not allow for this option.

Isolation of CD8 expression data

To obtain better resolution and a clustering that was not affected by the CD4⁺ T cells, we determined the mean Cd4 and Cd8a expression of the 24 clusters, and isolated the 155,496 single cells belonging to the 16 clusters where Cd8a expression was predominant (Extended Data Fig. 4d). We then performed a re-clustering of that data using the Harmony reduction to yield 20 phenotypic CD8⁺ clusters, which represented a reformulation of the original clusters.

Correspondences with clusters from external single-cell datasets

For each correspondence with an external dataset, we obtained metadata containing cluster assignments and scRNA-seq data for constructing centroids for each cluster. Metadata was obtained either publicly or from personal communication with the authors. In the latter case, these metadata files are provided in the software package for this paper, as mentioned in Data Availability.

For Huang et al.¹⁶, we used metadata provided by the authors to us and scRNA-seq count data for the six samples referenced in the metadata from the NCBI Gene Expression Omnibus (GEO) for GSE180095, GSE122712, GSE152628 and GSE182509. For Daniel et al.²⁶, we used metadata from supplementary files and the scRNA-seq count data, both available at NCBI GEO for GSE188666. For Deak et al.²⁷, we used metadata provided by the authors to us, and scRNA-seq count data available at ArrayExpress for E-MTAB-11773. For Giles et al.²⁸, we used metadata from the Seurat object through the link provided in their paper under Data Availability and scRNA-seq count data available at NCBI GEO for GSE199563. For Li et al.²⁴, we used metadata (taking cluster assignments for celltype_cluster-2) and scRNA-seq normalized data from the Scanpy object available at ArrayExpress for E-MTAB-10176.

For datasets where raw counts were available, we generated normalized data by dividing each count by the total count for the cell and multiplying by 1 million to yield a value in transcripts per million (TPM) and adding 1 as a pseudo count. For GSE182509 within the analysis for Huang et al.¹⁶, where only processed data are provided, we reversed the authors’ square-root transformation and confirmed that the sum of those values for cells totaled 1,000, so that our normalization procedure would work. In that analysis, where samples came from different sources, normalization was performed before samples were limited to those genes in common across all datasets. For Li et al.²⁴, where only log-normalized data were available, we used that data. Then for cells belonging to each cluster, we generated a centroid for that cluster by computing a trimmed mean of the TPM for each gene, rejecting 10% of measurements from each end of the range. For datasets with raw counts, we converted centroids to a log scale by taking the logarithm base 2. These centroids were used as reference gene signatures to assign each cell from our dataset, where genes with zero expression across an entire sample were excluded, gene expression for each cell was converted to log₂(TPM + 1) and assignment was performed by the SingleR package in R, using default parameters. Assignments between the two clustering schemes were cross-tabulated and normalized by the total counts for each of our clusters.

Assignment of gp70 status

The single-cell ADT assay provided measures for each cell on its binding to two tetramers for gp70 antigens and ten for non-gp70 antigens (two for C28, five for UV and three for C142). To determine whether a cell was gp70⁺, we used the minimum value for the gp70 as a test statistic in a Poisson test where base rate was the maximum value for the non-gp70 antigens, using the poisson.test function in R. A cell was considered gp70⁺ if the one-sided P value with alt = ‘greater’ was less the 1 × 10^–6. A clone was considered gp70⁺ if any of its cells was gp70⁺.

Clonal co-occurrence analysis

Co-occurrence matrices were tabulated by summing intraclonal pairs across all clones. Specifically, for a given set of samples, each clone with n cells, where n > 1, contributed to the co-occurrence matrix with its outer product xxT, where the outer product represents the count of any two cluster–tissue pairs occurring in the same clone. For migration analysis, we performed the same computation, including all dLNs, blood day 7 and tumor samples for each experimental group and keeping track of intraclonal pairs for each combination of cluster and tissue. The resulting co-occurrence matrices were plotted using the chordDiagram function from the circlize package³⁹, v.0.4.15, in R, with the parameters transparency = 0.2 and reduce = 0. In the resulting plots, link widths are normalized by the total number of intraclonal pairs, which make up a full circumference. Same-cluster links or same-tissue links for the migration analysis, were hidden using the link.visible parameter.

In this co-occurrence analysis, clones contribute information according to their possible pairwise counts, so that singleton clones contribute no information and expanded clones contribute information according to the square of their size. For migration analysis in Fig. 6c, chord thicknesses are proportional to the square of the clone sizes between tissues. As effective clones are highly expanded in tumors but less expanded in dLNs and blood, lines may not be discernible between dLNs and blood while reasonable line thicknesses will be seen between blood and tumors or dLNs and tumors. We also characterized a clone as being gp70⁺ if any one of its cells was determined to be gp70⁺, although the largest clones also biased these counts according to the square of their size.

When projecting co-occurrence onto the UMAPs, such projections can be noisy because of the transitive nature of co-occurrence, where co-occurrence of cluster A and B and co-occurrence of clusters B and C necessarily implies co-occurrence of A and C. Therefore, to identify primary differentiation pathways, we applied a minimum spanning tree (MST) algorithm in R to the co-occurrence data within dLNs and within tumors, where links were processed in order from the largest to smallest count of intraclonal pairs, and retaining links only if they did not create a cycle in the graph with links previously kept. Co-occurrence links were plotted with the same relative thicknesses as shown in the circular co-occurrence plots of Fig. 6a–c, normalized to the total number of intraclonal pairs, but with a relative multiplier of 3 for the migration links, as they are relatively sparse.

RNA velocity analysis

The paired-end FASTQ files from each sample were mapped using kallisto bustools (v.0.46.1) to a transcriptome index from Ensembl v.90 on genome GRCm38. The transcriptome index was generated using kallisto with a read length of 90 nt and intronic sequences from BUSpaRse⁴⁰ (BUSpaRse: kallisto | bustools R utilities; R package v.1.6.1, https://github.com/BUStools/BUSpaRse). The resulting spliced and unspliced count matrices for each tissue sample from each mouse were filtered to correspond to the cells used in the Seurat-based analysis, and the Seurat-based UMAP coordinates for those cells were added to the data object. The cells for each tissue and experimental group were combined using the concatenate procedure with join = ‘outer’. The resulting object was processed by scvelo package v.0.2.4 within Python v.3.7.3, using the commands ‘pp.filter_and_normalize’, ‘pp.moments’, ‘tl.recover_dynamics’ and ‘tl.velocity’ with mode = ‘dynamical’. Velocity graphs were generated using the command ‘tl.velocity_graph’ and ‘pl.velocity_embedding_stream’, with the parameter arrow_size = 0.001 to hide arrows, which otherwise gave directions often inconsistent with one another and with empirically determined T cell behavior.

Projection of human CD8⁺ T cells from a Ph1b scRNA-seq dataset to a mouse reference

Human genes from the Ph1b GO30103 (NCT02794571) scRNA-seq data were first converted to their mouse orthologs using babelgene (v.22.9). Human genes without mouse orthologs or with mouse orthologs not present in the mouse scRNA-seq dataset were left unmodified without renaming. Human CD8⁺ T cells were then separated by patient and normalized with SCTransform in Seurat (v.4.2) using the default parameters. These samples were then integrated using reference-based integration to overcome the memory limits of canonical correlation analysis integration. The second patient in the dataset was chosen at random as the integration reference. After integration, transfer anchors were identified between the query human CD8⁺ T cell dataset and the mouse CD8⁺ T cell reference. The MapQuery function in Seurat was used to transfer cell type labels, integrate embeddings and to project the query data onto the reference UMAP.

Gene signature scores for CITYSCAPE

The top 20 differentially expressed genes in each of the mouse CD8⁺ T cell clusters identified from scRNA-seq were converted to their human orthologs using babelgene (v.22.9) in R (v.4.2.0). Mouse genes that did not have human orthologs or with human orthologs that were not present in the CITYSCAPE dataset were removed. The final curated table of signature genes used for analysis is shown in Supplementary Table 2.

Analysis of CITYSCAPE and OAK clinical trial data

CITYSCAPE (NCT03563716) is a phase 2 trial investigating tiragolumab with atezolizumab compared to placebo with atezolizumab in patients with locally advanced or metastatic NSCLC³. Patients were treated until disease progression or loss of clinical benefit. Patient tumor samples were submitted for RNA-seq and the average, log-normalized expression of the genes shown in Supplementary Table 2 or selected genes as indicated in the text was used to define gene signature scores. Objective response was categorized according to Response Evaluation Criteria in Solid Tumors (v.1.1). For Kaplan–Meier (KM) survival curves and HRs, patients in the CITYSCAPE trial were separated by treatment group and further subdivided by high or low expression of individual genes or gene signatures, where high or low is defined as greater than or equal to, or less than, the global median expression, respectively, of that gene or gene signature score. The survminer package (v.0.4.9), survival package (v.3.4-0) and R (v.4.2.0) were used to generate the KM plot. A log-rank test was used for statistical testing on the survival data. A Cox proportional hazards regression model was fit on gene or gene signature score high or low data and the HR and 95% CI for OS calculated and plotted for patients receiving tiragolumab with atezolizumab compared to patients receiving placebo with atezolizumab. Similarly, KM survival curves for progression-free survival and OS were generated for the phase 3 OAK study (NCT02008227) evaluating atezolizumab versus chemotherapy in PD-L1-positive previously treated patients with advanced or metastatic NSCLC³².

Statistics and reproducibility

Pharmacodynamic data were analyzed using GraphPad Prism software v.9 (GraphPad). Measures between two groups were performed with a Student’s t-test (two-tailed). Groups of three or more were analyzed by one-way or two-way analysis of variance (ANOVA) followed by Tukey’s post-testing for multiple comparisons, as appropriate. Reproducibility of studies is indicated by the number of independent experiments provided in figure legends. Data distribution was assumed to be normal but this was not formally tested. No statistical methods were used to predetermine sample size. Sample sizes were based on previous experience^5,6 and balanced animal welfare and statistical robustness. For syngeneic mouse tumor studies, animals whose tumors became ulcerated before progression or CR or at time of allocation into experimental groups were killed and removed from the study. For syngeneic mouse tumor studies, data collection and analysis were not performed blind to the conditions of the experiments. CITYSCAPE was a randomize, double-blinded, placebo-controlled study³; for data analysis; patients were separated on the basis of clinical response irrespective of treatment arm. All other statistical methods were performed as described in the figure legends corresponding to the data figure.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

FASTQ files containing raw sequencing reads for the scRNA-seq, scTCR-seq, ADT-seq and CITE-seq analyses have been deposited with the NCBI Short Read Archive under accession PRJNA911822. Processed output files from Cell Ranger and metadata with cluster assignments, clonotypes and ADT assignments have been deposited with the NCBI GEO under accession code GSE220901. Raw counts and metadata for reanalyzed scRNA-seq data from GO30103 (NCT02794571)^4,31 are available from the European Genome–Phenome Archive (EGA) under accession code EGAD50000000367 and at https://github.com/cwtran/nutsch_nature_cancer_2024/. Reanalyzed datasets from the CITYSCAPE (NCT03563716)^3,4 and OAK (NCT02008227)^4,32 clinical trials can be found under EGA accession codes EGAD50000000251 and EGAD50000000368, respectively. All data relating to mouse sequencing data at NCBI Short Read Archive and NCBI GEO and pseudoanonymized clinical trial data are available without restrictions. Due to the legal requirements for data sharing, users must agree to the Data Access Agreement detailed in the EGA entries above before they can access these human datasets. Data access requests are reviewed by the Genentech DevSci Data Access Committee (devsci-dac-d@gene.com).

Reanalysis of previously published datasets generated by others and publicly available was as follows: for Huang et al.¹⁶, we used metadata provided by the authors to us and scRNA-seq count data for the six samples referenced in the metadata from NCBI GEO for GSE180095, GSE122712, GSE152628 and GSE182509; for Daniel et al.²⁶, we used metadata from supplementary files and the scRNA-seq count data, both available at NCBI GEO for GSE188666; for Deak et al.²⁷, we used metadata provided by the authors to us, and scRNA-seq count data available at ArrayExpress for E-MTAB-11773; for Giles et al.²⁸, we used metadata from the Seurat object through the link (https://www.dropbox.com/scl/fo/wuwas5f1jeac3nuw46de4/ADc75QyDwGrS_7XklqP7PLY?rlkey=9wrgn38p3hherec19xxw2rrvs&e=1&st=wimjk55a&dl=0) provided in their paper under Data Availability and scRNA-seq count data available at NCBI GEO for GSE199563; for Li et al.²⁴, we used metadata (taking cluster assignments for celltype_cluster-2) and scRNA-seq normalized data from the Scanpy object available at ArrayExpress for E-MTAB-10176.

Source data for Figs. 1, 2a,d,e, 3b,c, 4, 5c and 6a–c and Extended Data Figs. 1b,c, 3a–q, 6a,e, 7a, 8 and 9b,c have been provided. Source data are provided with this paper. All other data supporting the findings of this study are available from the corresponding author on reasonable request.

Code availability

Computer code used to generate the single-cell analyses and figures in this paper are provided as a Supplementary File to the NCBI GEO accession code GSE220901. Code for reanalyzed datasets from the GO30103 (NCT02794571)^4,31, CITYSCAPE (NCT03563716)^3,4 and OAK (NCT02008227)^4,32 clinical trials are accessible at https://github.com/cwtran/nutsch_nature_cancer_2024/.

Change history

17 January 2025
A Correction to this paper has been published: https://doi.org/10.1038/s43018-025-00908-3

References

Mellman, I., Chen, D. S., Powles, T. & Turley, S. J. The cancer-immunity cycle: indication, genotype, and immunotype. Immunity 56, 2188–2205 (2023).
Article PubMed CAS Google Scholar
Chiang, E. Y. & Mellman, I. TIGIT–CD226–PVR axis: advancing immune checkpoint blockade for cancer immunotherapy. J. Immunother. Cancer 10, e004711 (2022).
Article PubMed PubMed Central Google Scholar
Cho, B. C. et al. Tiragolumab plus atezolizumab versus placebo plus atezolizumab as a first-line treatment for PD-L1-selected non-small-cell lung cancer (CITYSCAPE): primary and follow-up analyses of a randomised, double-blind, phase 2 study. Lancet Oncol. 23, 781–792 (2022).
Article PubMed CAS Google Scholar
Guan, X. et al. Anti-TIGIT antibody tiragolumab leverages myeloid cells and regulatory T cells to improve PD-L1 checkpoint blockade. Nature 627, 646–655 (2024).
Article PubMed PubMed Central CAS Google Scholar
Johnston, R. J. et al. The immunoreceptor TIGIT regulates antitumor and antiviral CD8⁺ T cell effector function. Cancer Cell 26, 923–937 (2014).
Article PubMed CAS Google Scholar
Banta, K. L. et al. Mechanistic convergence of the TIGIT and PD-1 inhibitory pathways necessitates co-blockade to optimize anti-tumor CD8⁺ T cell responses. Immunity 55, 512–526 (2022).
Article PubMed PubMed Central CAS Google Scholar
Gattinoni, L. et al. A human memory T cell subset with stem cell-like properties. Nat. Med. 17, 1290–1297 (2011).
Article PubMed PubMed Central CAS Google Scholar
Lugli, E., Galletti, G., Boi, S. K. & Youngblood, B. A. Stem, effector, and hybrid states of memory CD8⁺ T cells. Trends Immunol. 41, 17–28 (2020).
Article PubMed CAS Google Scholar
Zehn, D., Thimme, R., Lugli, E., Pereira de Almeida, G. & Oxenius, A. Stem-like precursors are the fount to sustain CD8⁺ T cell responses. Nat. Immunol. 23, 836–847 (2022).
Article PubMed CAS Google Scholar
Galletti, G. et al. Two subsets of stem-like CD8⁺ memory T cell progenitors with distinct fate commitments in humans. Nat. Immunol. 21, 1552–1562 (2020).
Article PubMed PubMed Central Google Scholar
Siddiqui, I. et al. Intratumoral Tcf1⁺PD-1⁺CD8⁺ T cells with stem-like properties promote tumor control in response to vaccination and checkpoint blockade immunotherapy. Immunity 50, 195–211.e110 (2019).
Article PubMed CAS Google Scholar
Im, S. J. et al. Defining CD8⁺ T cells that provide the proliferative burst after PD-1 therapy. Nature 537, 417–421 (2016).
Article PubMed PubMed Central CAS Google Scholar
Kurtulus, S. et al. Checkpoint blockade immunotherapy induces dynamic changes in PD-1⁻CD8⁺ tumor-infiltrating T cells. Immunity 50, 181–194.e186 (2019).
Article PubMed PubMed Central CAS Google Scholar
Kallies, A., Zehn, D. & Utzschneider, D. T. Precursor exhausted T cells: key to successful immunotherapy? Nat. Rev. Immunol. 20, 128–136 (2020).
Article PubMed CAS Google Scholar
Dammeijer, F. et al. The PD-1/PD-L1-checkpoint restrains T cell immunity in tumor-draining lymph nodes. Cancer Cell 38, 685–700 (2020).
Article PubMed CAS Google Scholar
Huang, Q. et al. The primordial differentiation of tumor-specific memory CD8⁺ T cells as bona fide responders to PD-1/PD-L1 blockade in draining lymph nodes. Cell 185, 1–18 (2022).
Article Google Scholar
Connolly, K. A. et al. A reservoir of stem-like CD8⁺ T cells in the tumor-draining lymph node preserves the ongoing antitumor immune response. Sci. Immunol. 6, eabg7836 (2021).
Article PubMed PubMed Central CAS Google Scholar
Barboy, O. et al. Modeling T cell temporal response to cancer immunotherapy rationalizes development of combinatorial treatment protocols. Nat. Cancer https://doi.org/10.1038/s43018-024-00734-z (2024).
Article PubMed Google Scholar
Giles, J. R., Globig, A.-M., Kaech, S. M. & Wherry, E. J. CD8⁺ T cells in the cancer-immunity cycle. Immunity 56, 2231–2253 (2023).
Article PubMed PubMed Central CAS Google Scholar
Pinschewer, D. D. et al. FTY720 immunosuppression impairs effector T cell peripheral homing without affecting induction, expansion, and memory. J. Immunol. 164, 5761–5770 (2000).
Article PubMed CAS Google Scholar
Huang, A. Y. et al. The immunodominant major histocompatibility complex class I-restricted antigen of a murine colon tumor derives from an endogenous retroviral gene product. Proc. Natl Acad. Sci. USA 93, 9730–9735 (1996).
Article PubMed PubMed Central CAS Google Scholar
Scott, A. C. et al. TOX is a critical regulator of tumour-specific T cell differentiation. Nature 571, 270–274 (2019).
Article PubMed PubMed Central CAS Google Scholar
Khan, O. et al. TOX transcriptionally and epigenetically programs CD8⁺ T cell exhaustion. Nature 571, 211–218 (2019).
Article PubMed PubMed Central CAS Google Scholar
Li, Z. et al. In vivo labeling reveals continuous trafficking of TCF-1⁺ T cells between tumor and lymphoid tissue. J. Exp. Med. 219, e20210749 (2022).
Article PubMed PubMed Central CAS Google Scholar
Meier, S. L., Satpathy, A. T. & Wells, D. K. Bystander T cells in cancer immunology and therapy. Nat. Cancer 3, 143–155 (2022).
Article PubMed Google Scholar
Daniel, B. et al. Divergent clonal differentiation trajectories of T cell exhaustion. Nat. Immunol. 23, 1614–1627 (2022).
Article PubMed PubMed Central CAS Google Scholar
Deak, L. C. et al. PD-1-cis IL-2R agonism yields better effectors from stem-like CD8⁺ T cells. Nature https://doi.org/10.1038/s41586-022-05192-0 (2022).
Article Google Scholar
Giles, J. R. et al. Shared and distinct biological circuits in effector, memory and exhausted CD8⁺ T cells revealed by temporal single-cell transcriptomics and epigenetics. Nat. Immunol. 23, 1600–1613 (2022).
Article PubMed PubMed Central CAS Google Scholar
Bergen, V., Soldatov, R. A., Kharchenko, P. V. & Theis, F. J. RNA velocity–current challenges and future perspectives. Mol. Syst. Biol. 17, e10282 (2021).
Article PubMed PubMed Central CAS Google Scholar
Simoni, Y. et al. Bystander CD8⁺ T cells are abundant and phenotypically distinct in human tumour infiltrates. Nature 557, 575–579 (2018).
Article PubMed CAS Google Scholar
Kim, T. W. et al. Anti-TIGIT antibody tiragolumab alone or with atezolizumab in patients with advanced solid tumors: a phase 1a/1b nonrandominzed controlled trial. JAMA Oncol. 9, 1574–1582 (2023).
Article PubMed PubMed Central Google Scholar
Rittmeyer, A. et al. Atezolizumab versus docetaxel in patients with previously treated non-small-cell lung cancer (OAK): a phase 3, open-label, multicentre randomised controlled trial. Lancet 389, 255–265 (2017).
Article PubMed Google Scholar
Tumeh, P. C. et al. PD-1 blockade induces responses by inhibiting adaptive immune resistance. Nature 515, 568–571 (2014).
Article PubMed PubMed Central CAS Google Scholar
Brunell, A. E., Lahesmaa, R., Autio, A. & Thotakura, A. K. Exhausted T cells hijacking the cancer-immunity cycle: assets and liabilities. Front. Immunol. 14, 1151632 (2023).
Article PubMed PubMed Central CAS Google Scholar
Pittet, M. J., Di Pilato, M., Garris, C. & Mempel, T. R. Dendritic cells as shepherds of T cell immunity in cancer. Immunity 56, 2218–2230 (2023).
Article PubMed PubMed Central CAS Google Scholar
Oh, S. A. et al. PD-L1 expression by dendritic cells is a key regulatory of T-cell immunity in cancer. Nat. Cancer 1, 681–691 (2020).
Article PubMed CAS Google Scholar
Hashimoto, M. et al. PD-1 combination therapy with IL-2 modifies CD8⁺ T cell exhaustion program. Nature https://doi.org/10.1038/s41586-022-05257-0 (2022).
Article PubMed PubMed Central Google Scholar
Castiglioni, A. et al. Combined PD-L1/TGFβ blockade allows expansion and differentiation of stem cell-like CD8 T cells in immune excluded tumors. Nat. Commun. 14, 4703 (2023).
Article PubMed PubMed Central CAS Google Scholar
Gu, Z., Gu, L., Eils, R., Schlesner, M. & Brors, B. circlize implements and enhances circular visualization in R. Bioinformatics 30, 2811–2812 (2014).
Article PubMed CAS Google Scholar
Moses, L. & Pachter, L. Museum of spatial transcriptomics. Nat. Methods 19, 534–546 (2022).
Article PubMed CAS Google Scholar

Download references

Acknowledgements

We thank the patients who kindly provided tumor samples for this study, as well as the investigators and staff involved in the GO30103, CITYSCAPE^3,4 and OAK studies. We thank Genentech’s FACS core facility for contributing their expertise and performing cell sorting. We thank the Genentech laboratory animal core groups for microinjection, animal care and genotyping support. We thank L. Adams for providing assistance with the pharmacodynamic studies. We thank R. Johnston, J. Grogan, A. Chitre and S. Oh for thoughtful discussions.

Author information

These authors contributed equally: Katherine Nutsch, Karl L. Banta, Thomas D. Wu.
These authors jointly supervised this work: Eugene Y. Chiang, Ira Mellman.

Authors and Affiliations

Genentech, South San Francisco, CA, USA
Katherine Nutsch, Karl L. Banta, Thomas D. Wu, Charles W. Tran, Stephanie Mittman, Ellen Duong, Barzin Y. Nabet, Yan Qu, Katherine Williams, Sören Müller, Namrata S. Patil, Eugene Y. Chiang & Ira Mellman

Authors

Katherine Nutsch
View author publications
You can also search for this author in PubMed Google Scholar
Karl L. Banta
View author publications
You can also search for this author in PubMed Google Scholar
Thomas D. Wu
View author publications
You can also search for this author in PubMed Google Scholar
Charles W. Tran
View author publications
You can also search for this author in PubMed Google Scholar
Stephanie Mittman
View author publications
You can also search for this author in PubMed Google Scholar
Ellen Duong
View author publications
You can also search for this author in PubMed Google Scholar
Barzin Y. Nabet
View author publications
You can also search for this author in PubMed Google Scholar
Yan Qu
View author publications
You can also search for this author in PubMed Google Scholar
Katherine Williams
View author publications
You can also search for this author in PubMed Google Scholar
Sören Müller
View author publications
You can also search for this author in PubMed Google Scholar
Namrata S. Patil
View author publications
You can also search for this author in PubMed Google Scholar
Eugene Y. Chiang
View author publications
You can also search for this author in PubMed Google Scholar
Ira Mellman
View author publications
You can also search for this author in PubMed Google Scholar

Contributions

K.N., K.L.B. and E.D. performed in vivo tumor studies, pharmacodynamic studies, bioinformatic studies and data analysis; T.D.W., K.W. and S. Müller performed mouse bioinformatic analyses; S. Mittman and Y.Q. managed and performed in vivo tumor studies; C.W.T., B.Y.N. and N.S.P. performed clinical trial bioinformatic analyses; K.N., K.L.B., E.Y.C. and I.M. conceived this work; E.Y.C. and I.M. supervised this work; and K.N., K.L.B., T.D.W., E.Y.C. and I.M. wrote the paper.

Corresponding authors

Correspondence to Eugene Y. Chiang or Ira Mellman.

Ethics declarations

Competing interests

The authors declare the following competing interests: all authors are employees of Genentech, a member of the Roche group, which develops and markets drugs for profit.

Peer review

Peer review information

Nature Cancer thanks Hazem Ghoneim and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Fig. 1 Experimental schemas, efficacy and pharmacodynamic data for CT26 and EO771 studies.

a, Experimental schemas. Tumor growth and PD/multi-omics analysis for CT26 and EO771 studies (left); delayed FTY720 treatment in CT26 model (right). b, c, Mouse EO771 orthotopic tumor model. b, FTY720 treatment impairs efficacy of anti-PD-1 and anti-TIGIT treatment in EO771 tumor model. C57BL/6 mice inoculated in mammary fat pad with EO771 tumor cells and treated with isotype control, anti-PD-L1, anti-TIGIT, or the combination of anti-PD-L1 and anti-TIGIT antibodies, with or without FTY720. Tumor growth was monitored and grouped analysis and growth curves for each individual animal are shown (n = 10 animals per group). Data shown are representative of two independent experiments. c, Total numbers of CD8⁺ T cells, CD4⁺ T cells or Tregs in dLN (top panels) or tumor (bottom panels) of CT26 tumor-bearing mice treated with isotype control, anti-PD-L1, anti-TIGIT or the combination, with or without FTY720. n = 5 animals per sample and treatment group, mean ± s.d. are represented by bars and whiskers. Data are representative of three independent experiments.

Source data

Extended Data Fig. 2 Flow cytometry gating strategies.

a, Representative dot plots for CD226 and gp70 tetramer co-expression on CD8⁺ T cells in dLN or tumor, or gp70 tetramer staining on CD8⁺ T cells in blood. b, Representative dot plots for CD226 and PD-1, Tim3 and PD-1 co-expression on CD8⁺ T cells in dLN. c, Representative dot plots for CD226 and LAG3 or Tim3 and PD-1 co-expression on CD8⁺ T cells in tumor. d, Representative dot plots for TCF1 and Tox co-expression on CD8⁺ T cells in dLN or tumor. e, Representative dot plots for IFN-γ and TNF intracellular staining in CD8⁺ T cells from tumor.

Extended Data Fig. 3 Treatment effects on tumor-specific CD8⁺ T cells.

a-g, Phenotyping of CD226⁺ or CD226⁻ gp70⁺ CD8⁺ T cells from dLN or tumor, as indicated, of CT26 tumor-bearing mice treated with anti-PD-L1, anti-TIGIT, or combination with or without FTY720. Frequencies of gp70⁺CD8⁺ T cells expressing Ki67 (a), naïve phenotype (b), Teff/Tem phenotype (c), Slamf6 and TCF1 co-expression (d), TCF1 and Tim3 co-expression (e), not expressing TCF1 (f) or expressing Tox (g). For a-g, data are represented as mean +/- SD with individual symbols representing individual mice (n = 5 animals per sample and treatment group) and are representative of three independent experiments. p-values are indicated where differences between two groups were determined to be statistically significant by ordinary one-way ANOVA with Tukey’s multiple comparisons test. h-m, anti-TIGIT plus anti-PD-L1 combination treatment increases frequencies of TCF1⁺Tim⁺ CD8⁺ T cells and decreases frequencies of Tox⁺ CD8⁺ T cells in EO771 tumor-bearing mice. Mice with established EO771 tumors were treated with isotype control Ab or anti-TIGIT combined with anti-PD-L1, then dLN and tumors were collected on day 7 post-treatment for phenotypic characterization of CD8⁺ T cells by flow cytometry. h, Frequencies of CD8⁺ T cells in dLN (left) or tumor (right) as percentage of CD45⁺ cells. i, Frequencies of CD8⁺ T cells expressing CD226. j, Frequencies of CD8⁺ T cells positively stained with tetramer against EO771-specific antigen p15E. k, Frequencies of p15E⁺CD8⁺ T cells expressing CD226. l, m, Frequencies of CD226⁺ (left) or CD226⁻ (right) p15E⁺CD8⁺ Teff/Tem cells co-expressing TCF1 and Tim3 (l) or Tox (m). For h-m, data are represented as mean +/- SD with individual symbols represent individual mice (n = 4 animals per sample and group); data are representative of one of two independent experiments. p-values are indicated where difference were determined by unpaired t-test to be statistically significant. n-q, Frequencies of gp70⁺CD8⁺ T cells co-expressing Slamf6 and TCF1 (n), co-expressing TCF1 and Tim3 (o), Teff/Tem phenotype (p) or Tox (q) in dLN (left) or tumor (right) of CT26 tumor-bearing mice treated with anti-PD-L1 plus anti-TIGIT combination or combo with anti-CD226 Ab. For n-q, data are represented as mean +/- SD with individual symbols representing individual mice (n = 5 animals per sample and treatment group) and are representative of three independent experiments. p-values are indicated where differences between two groups were determined to be statistically significant by ordinary one-way ANOVA with Tukey’s multiple comparisons test.

Source data

Extended Data Fig. 4 Single-cell RNA-seq, TCR-seq and CITE-seq on 245,675 total T cells pooled from CT26 tumor, dLN and blood from 31 mice.

a, UMAP showing T cell clusters representing CD8⁺ T cells, CD4⁺ T cells and regulatory T cells. b,c. Treatment group (Group) (b) or tissue source (c) of T cells. d-g. CD8a expression (d), granzyme B expression (e), clone size (f), gp70 ADT counts (g) projected on UMAP. h, CITE-seq relative expression levels for indicated markers are projected on UMAP.

Extended Data Fig. 5 CD8⁺ T cell marker expression.

Marker expression was measured by scRNA-seq and projected on UMAPs comprised of CD8⁺ T cells from CT26 tumor, dLN and blood. n = 155,496 cells in all plots.

Extended Data Fig. 6 Multiomic analysis of CD8⁺ T cells.

a, CD8⁺ T cell cluster composition by individual mice in each treatment group. Each color represents an individual mouse. Data show that no cluster came solely from a single animal or experimental group. b, Treatment group (Group) and tissue source projected on CD8⁺ T cell UMAP. c, Clone size and ADT counts projected on CD8⁺ T cell UMAP. In panels (b) and (c), n = 155,496 cells. d–f. CD8⁺ T cell marker expression measured by CITE-seq. d, Relative marker expression levels determined using CITE-seq antibodies projected on UMAPs comprised of CD8⁺ T cells from CT26 tumor, dLN and blood. e, Heatmap of relative CITE-seq marker expression levels in each CD8⁺ T cell cluster. f, Heatmap of relative CITE-seq marker expression levels in tissues under various treatment conditions.

Source data

Extended Data Fig. 7 Characterization of CD8⁺ T cell clusters.

a, Heatmap showing clusters defined using metadata from Li et al.²⁴, where CD8T1 are likely naïve, CD8T2 has Tscm/memory properties, CD8T4 has activated effector cell properties, and CD8T5 has an exhausted phenotype. CD8T1 and CD8T2 were described as T cells that recently entered the tumor, while CD8T4 and CD8T5 were cells resident in tumor. CD8T2 also have the capacity to recirculate from tumor to dLN. b, RNA velocity projections on UMAPs for each treatment group in dLN (top) and tumour (bottom). c, UMAPs showing cluster composition in dLN (Lymph), tumor and blood from CT26 tumor-bearing mice treated with isotype control, anti-PD-L1, anti-TIGIT, combination, or combination with FTY720.

Source data

Extended Data Fig. 8 Characterization of tumor-specific CD8⁺ T cell cluster composition.

CD8⁺ T cell cluster composition segregated by specificity for gp70, determined by comparing ADT counts for gp70 tetramers, requiring that they be higher than control tetramer count by a Poisson test with a one-sided p-value < 1×10^-6. The proportion of each cluster within total CD8⁺ T cells from the indicated tissue for each treatment group is shown. Each symbol represents an individual mouse; each color identifies the same mouse within the treatment group.

Source data

Extended Data Fig. 9 Clonal diversity of CD8⁺ T cells.

a, Scatterplots showing primary clusters of each individual clonotype in dLN and blood at day 7 (D7 Blood). n represents the total number of clones within a cluster in each scatterplot; n_B is the number of clones observed in blood at day 7; n_BB is the number of clones in blood at day 7 with clone size greater than 1. Color of circles denote cluster designation as shown in panels (b) and (c). b, c, Cluster composition for the 30 largest clonotypes from tumor (upper panels), D7 blood (middle panels) or dLN (lower panels) of CT26 tumor-bearing mice treated with isotype control, anti-PD-L1, anti-TIGIT, combination, or combination with FTY720, shown as (b) absolute numbers, or (c) normalized against the total number of cells for each individual clonotype (total cell number = 1). Cluster identity is indicated by color. In (b), clonotypes from individual mice within each treatment group are identified by the color legend at the top of the tumor bar graphs. Individual mice are labeled as S”group number”.”mouse number”. gp70⁺ clones are identified by black symbol at the top of the bar graphs.

Source data

Extended Data Fig. 10 Analysis of human CD8⁺ T cell clusters.

Association of human CD8⁺ T cell clusters corresponding to mouse reference with response in tiragolumab plus atezolizumab Ph1b and Ph2 NSCLC clinical trials. a. scRNA-seq of 144,413 human CD8⁺ T cells from blood of patients in a Ph1b NSCLC study of tiragolumab plus atezolizumab (T + A). Human genes were renamed to their mouse ortholog (if present) and gene expression was normalized before sample integration and projection onto a mouse CD8⁺ T cell reference UMAP. b. Predicted cell type score for mapped human CD8⁺ T cells for each assigned mouse CD8⁺ T cell reference cluster. c. Frequencies of human predicted clusters in patients with complete or partial response (CRPR) compared to stable or progressive disease (SDPD) on cycle 2 day 1 of treatment with T + A in the Ph1b study. Percent total was calculated as the percentage of the cluster in total CD8⁺ T cells for each patient. d. Forest plot comparing high or low expression of the top corresponding human 18-20 signature genes (signature gene score) from each mouse CD8⁺ T cell cluster or CD8A and their association with overall survival (OS) hazard ratio (HR) T + A or placebo plus atezolizumab (P + A) treatment groups in Ph2 CITYSCAPE. Mean HR with 95% confidence intervals and p-values are shown. e, Kaplan-Meier curves showing OS probability in P + A or T + A treatment groups dichotomized on the basis of high or low CD8⁺ T cell cluster gene scores from each reference cluster. p-value is from log-rank test with null hypothesis that there is no difference between the groups. f. Kaplan-Meier curves comparing progression-free survival (PFS, left) or OS (right) in patients from the phase 3 NSCLC OAK study who received atezolizumab monotherapy. Patients were dichotomized by median gene score calculated using the average expression of the CD8 gene panel comprised of CXCR3, CXCR6 and CCL5.

Supplementary information

Reporting Summary

Supplementary Tables 1–4

Supplementary Tables 1–4, each table in separate tab in workbook.

Source data

Source Data Fig. 1

Statistical source data.

Source Data Fig. 2

Statistical source data.

Source Data Fig. 3

Statistical source data.

Source Data Fig. 4

Statistical source data.

Source Data Fig. 5

Statistical source data.

Source Data Fig. 6

Statistical source data.

Source Data Extended Data Fig. 1

Statistical source data.

Source Data Extended Data Fig. 3

Statistical source data.

Source Data Extended Data Fig. 6

Statistical source data.

Source Data Extended Data Fig. 7

Statistical source data.

Source Data Extended Data Fig. 8

Statistical source data.

Source Data Extended Data Fig. 9

Statistical source data.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Cite this article

Nutsch, K., Banta, K.L., Wu, T.D. et al. TIGIT and PD-L1 co-blockade promotes clonal expansion of multipotent, non-exhausted antitumor T cells by facilitating co-stimulation. Nat Cancer 5, 1834–1851 (2024). https://doi.org/10.1038/s43018-024-00870-6

Download citation

Received: 01 April 2024
Accepted: 28 October 2024
Published: 16 December 2024
Issue Date: December 2024
DOI: https://doi.org/10.1038/s43018-024-00870-6

Subjects

Abstract

Similar content being viewed by others

Main

Results

Combination treatment requires trafficking of lymphocytes from dLNs to tumor

CD226 has a role in tumor-specific CD8+ T cell differentiation

TIGIT and PD-L1 co-blockade expands different CD8+ T cell states

Combination treatment expands Ccl5+ tumor-specific CD8+ T cells

Co-blockade of PD-L1 and TIGIT focuses TCR clonal diversity

Anti-PD-L1 and anti-TIGIT differentially reshape CD8+ T cell trajectories

CD8+ T cell cluster association with response to tiragolumab plus atezolizumab

Discussion

Methods

Ethics statement

Mice

Cell lines

Syngeneic tumor studies

Flow cytometry and FACS sorting

Single-cell RNA-seq and TCR V(D)J clonotype profiling

Pre-processing of single-cell data

Integration of single-cell expression data

Isolation of CD8 expression data

Correspondences with clusters from external single-cell datasets

Assignment of gp70 status

Clonal co-occurrence analysis

RNA velocity analysis

Projection of human CD8+ T cells from a Ph1b scRNA-seq dataset to a mouse reference

Gene signature scores for CITYSCAPE

Analysis of CITYSCAPE and OAK clinical trial data

Statistics and reproducibility

Reporting summary

Data availability

Code availability

Change history

17 January 2025

References

Acknowledgements

Author information

Authors and Affiliations

Contributions

Corresponding authors

Ethics declarations

Competing interests

Peer review

Peer review information

Additional information

Extended data

Supplementary information

Source data

Rights and permissions

About this article

Cite this article

Share this article

Search

Quick links

CD226 has a role in tumor-specific CD8⁺ T cell differentiation

TIGIT and PD-L1 co-blockade expands different CD8⁺ T cell states

Combination treatment expands Ccl5⁺ tumor-specific CD8⁺ T cells

Anti-PD-L1 and anti-TIGIT differentially reshape CD8⁺ T cell trajectories

CD8⁺ T cell cluster association with response to tiragolumab plus atezolizumab

Projection of human CD8⁺ T cells from a Ph1b scRNA-seq dataset to a mouse reference