PD-1 Blockade Promotes Epitope Spreading in Anticancer CD8⁺ T Cell Responses by Preventing Fratricidal Death of Subdominant Clones To Relieve Immunodomination

This article is featured in In This Issue, p.3009

A pivotal role in immune surveillance against spontaneously arising neoplastic cells and in controlling intracellular pathogens is played by CD8⁺ T cells (T_CD8). However, when the immune system fails to eradicate cancer or clear stubborn infections, prolonged antigenic stimulation may lead to T_CD8 functional impairments, including exhaustion and anergy (1–4). Exhausted or anergic T_CD8 are often unable to secrete effector cytokines or launch optimal proliferative and cytotoxic responses to cognate Ags, which may compromise host defense mechanisms, positive clinical outcomes, or even survival (5–7).

Of several coinhibitory molecules known to interfere with bona fide T_CD8 activation, programmed death-1 (PD-1, CD279) has emerged as a major mediator of exhaustion and anergy (8). PD-1 is a type I transmembrane protein expressed by cells of hematopoietic origin, including T cells (9, 10). TCR triggering drives the expression of PD-1 at both transcriptional and translational levels (11, 12), which subsides once the Ag source is removed. However, PD-1 remains upregulated when TCR engagement is sustained, for instance in individuals with high tumor burden. Once ligated, PD-1 is phosphorylated on its intracellular tyrosine residues, which in turn leads to enhanced recruitment of Src homology 2–containing tyrosine phosphatase (SHP)-1 and SHP-2 to PD-1’s immunoreceptor tyrosine-based switch motif (13), thus dampening signal transduction through PI3K and the TCR complex (10).

PD-1 binds to two distinct ligands, namely PD-1 ligand (PD-L)1 (also known as B7-H1 or CD274) (14, 15) and PD-L2 (also known as B7-DC or CD273) (16, 17). PD-L1 is expressed, constitutively or inducibly, by a variety of hematopoietic and nonhematopoietic cells (10), and in various types of cancer (18). In contrast, PD-L2 has a much more restricted expression pattern and is primarily found on activated macrophages and myeloid dendritic cells (DCs) (10).

“Checkpoint inhibitors” that block PD-1–PD-L1 interactions have shown promise in preclinical studies and in clinical trials for several types of cancer (19, 20). Unfortunately, however, not all malignancies respond favorably to PD-1–targeted therapies. This highlights the need to better understand how the PD-1–PD-L1 pathway operates to incapacitate antitumor responses (21, 22).

Blocking the PD-1–PD-L1 axis improves T_CD8 responses by relieving coinhibition during T_CD8 priming by APCs, by preventing the lysis of effector T_CD8 by PD-L1⁺ tumors or virus-infected cells, and by reinvigorating exhausted T_CD8 (23). Although these effects clearly benefit host defense mechanisms, many, if not most, studies to date have focused on the effects of PD-1 signaling or blockade on T_CD8 that recognize immunodominant determinants (IDDs) of tumor and viral Ags. Although logistically convenient and still valid, this approach may overlook a critical aspect of T_CD8 responses—that is, their epitope breadth. This is concerning in light of the growing appreciation that immunodominant T_CD8 may not necessarily be the most protective clones (24). Importantly, whether signaling through PD-1, or its blockade, broadens or narrows T_CD8 responses is essentially unexplored.

Complex Ags harbor thousands of potentially immunogenic peptides; however, only a “selected” few induce detectable T_CD8 responses of varying magnitude, which are reproducibly arranged in a hierarchical order. This intriguing phenomenon is called immunodominance (ID) (25). We and others have demonstrated that T_CD8 ID can be shaped by Ag dose and administration route (26, 27), T_CD8 priming pathway (i.e., direct priming versus cross-priming) (28) and the type of APCs involved (29), abundance of protein substrates (30), efficiency and kinetics of peptide liberation by standard proteasomes and immunoproteasomes (31, 32), degenerate selectivity of TAP for peptides (33), peptide binding affinity for MHC class I (MHC I) allomorphs (33, 34), presence and precursor frequency of cognate T_CD8 in one’s T cell repertoire (35), TCR structural diversity (for instance due to N-nucleotide addition within junctional sequences) (36, 37), selective suppression of T_CD8 responses by naturally occurring regulatory T (nTreg) cells (38), and immunomodulatory actions of certain intracellular enzymes such as IDO (39) and mammalian target of rapamycin (mTOR) (40). Additionally, immunodominant T_CD8 clones may outcompete subdominant clones for access to APCs (41) or even directly kill them, although the evidence for the latter scenario has been scarce.

It is important to note that the above factors and mechanisms contribute to but do not fully account for ID. In this work, we demonstrate, to our knowledge for the first time, that: 1) PD-1, unlike several other receptors implicated in T cell coinhibition or exhaustion, enforces ID disparities in T_CD8 responses to a clinically relevant oncoprotein; 2) blockade of PD-1–PD-L1 interactions increases the epitope breadth of tumor-specific T_CD8 responses, thus increasing the range of peptide epitopes that can be targeted by the host; and 3) treatment with anti–PD-1 prevents immunodomination otherwise exerted by immunodominant T_CD8 through a fratricidal mechanism. These findings shed new light on T_CD8 ID and also have clear implications for immunotherapy of cancer and potentially other conditions such as chronic viral diseases.

Wild-type (wt) female C57BL/6 (B6) mice were purchased from Charles River Canada (St. Constant, QC, Canada) and housed in our institutional barrier facility. Closely age-matched adult mice were used following an animal use protocol approved by the Western University Animal Use Subcommittee and the Canadian Council on Animal Care guidelines.

The mouse mastocytoma cell line P815 was grown in RPMI 1640 medium containing 10% heat-inactivated FBS, GlutaMAX-I, 0.1 mM MEM nonessential amino acids, 1 mM sodium pyruvate, and 50 μM 2-ME.

We and/or others have previously described the generation of several cell lines that enable in vivo monitoring of SV40 large tumor Ag (T Ag)-specific T_CD8 responses. C57SV cells are transformed fibroblasts on the B6 (H-2^b) background (42, 43), and KD2SV cells (H-2^d) are of kidney epithelial origin (40, 43, 44). The TAP1^−/− wt T Ag line was generated by transfecting primary mouse kidney cells from B6.129S2-Tap1^tm1Arp mice with pPVU0, a plasmid containing the intact SV40 early region (45). B6/K-TagI cells were derived from B6 primary kidney cells transformed with pLM506-G(DC-1), a plasmid that was designed to encode a T Ag mutant with alanine substitutions at positions N227, F408, and N493. These amino acids comprise critical MHC I anchor residues within T Ag’s sites II/III, IV, and V epitopes, respectively (46). B6/TpLM237-9Ab cells are B6 mouse embryonic fibroblasts transformed through expression of a site IV–loss variant of T Ag containing a deletion of residues 404–411 (47). All T Ag⁺ cell lines were maintained in DMEM supplemented with 10% FBS.

Peptides used in this investigation are listed in Table I. They were procured or synthesized, purified by HPLC, and analyzed by mass spectrometry at or under the supervision of the Research Technologies Branch, National Institute of Allergy and Infectious Diseases (Rockville, MD), to confirm a purity of >95%. Stock solutions were prepared at 1 mM in DMSO and stored at −30°C.

Monolayers of tumor cell lines were trypsinized after they reached 100% confluency. Cells were washed thoroughly and resuspended in sterile PBS. To prime B6 mice, 2 × 10⁷ tumor cells were injected i.p.

A recombinant vaccinia virus (rVV) expressing full-length SV40 large tumor Ag (rVV-FL T Ag) and an rVV expressing site I as a cytosolic minigene (rVV-I) were initially provided by Dr. S. Tevethia (Pennsylvania State University, Hershey, PA) and propagated in the thymidine kinase–deficient human osteosarcoma cell line 143B. To infect mice, 1 × 10⁶ PFUs of each rVV were injected i.p.

Bone marrow–derived DCs (BMDCs) were generated by culturing marrow cells with recombinant mouse GM-CSF and IL-4 as previously described (39), and matured using 100 ng/ml LPS during the final 16 h of the culture. Adherent and floating DCs were harvested and pulsed for 2 h at 37°C with synthetic peptides corresponding to T Ag’s site I and/or site IV (Table I) at a final concentration of 1 μM. Cells were then washed three times and resuspended in sterile PBS before 5–10 × 10⁵ BMDCs were injected into the tail vein of each mouse.

Two hours before inoculation with tumor cells, rVVs, or peptide-pulsed DCs, mice received 100 μg of an anti-mouse PD-1 mAb (clone RMP1-14) or a rat IgG2a isotype control (clone 2A3) i.p. Animals received two additional 100-μg doses of anti–PD-1 or isotype on days 3 and 6. Following an identical protocol, we treated separate cohorts of mice, where indicated, with an anti–PD-L1 mAb (clone 10F.9G2) or rat IgG2b (clone LTF-2), or with a combination of anti–PD-1 and anti–PD-L1 mAbs (or isotype controls). To inactivate nTreg cells, a 1-mg single i.p. injection of an anti-mouse CD25 mAb (clone PC-61.5.3) was given 3 d prior to priming with C57SV cells. Control animals received a rat IgG1 (clone HRPN). The above mAbs and isotype controls were all purchased from Bio X Cell (West Lebanon, NH).

Unless otherwise specified, mice were euthanized 9 d after priming with T Ag⁺ tumor cells or 7 d after infection with rVVs or immunization with peptide-pulsed BMDCs. These time points coincide with the peak of in vivo T_CD8 responses to T Ag– and rVV-derived epitopes (38, 48), and T_CD8 responses after DC vaccination are often detectable after 7 d.

Splenic cell preparations were depleted of erythrocytes before they were washed, filtered, and stained. In a limited number of experiments, peripheral blood was collected into heparinized microhematocrit capillary tubes, diluted 1:2 in sterile PBS, and overlaid on 500 μl of low-endotoxin (<0.12 endotoxin unit/ml) Ficoll-Paque Plus. Cells were spun at 400 × g for 30 min at room temperature, and PBMCs gathering at the plasma–Ficoll interface were gently harvested and washed. Peritoneal exudate cells were collected via peritoneal lavage in several experiments in which mice had received i.p. injections of tumor cells.

We have previously described MHC I tetramer reagents that enable sensitive quantitation of T Ag–specific T_CD8 responses (48). Splenocytes from naive and primed B6 mice were placed at a density of 2 × 10⁶ cells per well in a round-bottom microplate and exposed to 20 μl of the 2.4G2 hybridoma supernatant containing an anti-CD16/CD32 mAb on ice to prevent nonspecific, FcγR-mediated adherence of Abs. After 20 min, cells were washed and resuspended in 2% FBS containing a PE-conjugated anti-mouse CD8α mAb (clone 53-6.7) and allophycocyanin-conjugated H-2K^b/IV or H-2D^b/I tetramer, which were used at a 1:200 dilution. Cells were incubated for 15 min in dark at room temperature, washed twice, and immediately interrogated by flow cytometry. Naive B6 splenocytes served as a negative staining control to allow for proper gating. In several experiments, surface staining for CTLA-4, LAG-3, PD-1, TIM-3, and CD107a and intracellular staining for Ki-67 were conducted in conjunction with tetramer staining. To detect CD107a, splenocytes were simultaneously exposed to antigenic peptides and stained with an Alexa Fluor 647–conjugated rat anti-mouse CD107a mAb (clone 1D4B) in the presence of 1 μM monensin before they were subjected to tetramer staining. A BD FACSCanto II cytometer and FlowJo software (Tree Star, Ashland, OR) were employed for data acquisition and analysis, respectively.

Intracellular cytokine staining (ICS) for IFN-γ was performed using a standard protocol. Erythrocyte-depleted splenocytes, peritoneal exudate cells, or PBMCs were suspended in medium and seeded at 0.5–2 × 10⁶ cells per well of a round-bottom microplate. Cells were left untreated or stimulated with indicated T Ag⁺ tumor cells or synthetic peptides corresponding to T Ag–derived T_CD8 epitopes (Table I). Peptides were used at a final concentration of 500 nM. After 2 h incubation at 37°C, 10 μg/ml brefeldin A was added to each well to retain IFN-γ in the endoplasmic reticulum of activated T_CD8, and cultures were continued for an additional 3–4 h. Cells were subsequently washed, briefly incubated with anti-CD16/CD32, and stained for surface CD8α before they were washed, fixed with 1% paraformaldehyde, washed again, permeabilized with 0.1% saponin, and stained for intracellular IFN-γ. The frequency of IFN-γ⁺ cells was determined after live gating on CD8α⁺ events, which was used to also calculate Ag-specific T_CD8 numbers in each spleen.

All fluorochrome-labeled mAbs were from eBioscience or BD Biosciences except for anti-CD107a, which was purchased from BioLegend.

We have previously described a modified, “three-peak” version of in vivo cytotoxicity assays to study T_CD8 ID (37). Syngeneic splenocytes were split into three populations, each of which was pulsed for 1 h at 37°C with 1 μM of an irrelevant or cognate peptide and labeled with a given dose of CFSE as follows. Control target cells were coated with an H-2^b–binding immunodominant peptide of HSV-1, namely gB₄₉₈, and 0.025 μM CFSE (CFSE^low). Splenocytes that were pulsed with site II/III or site I and stained with 0.25 μM (CFSE^int) or 2 μM CFSE (CFSE^high), respectively, served as cognate target cells. Cells were washed and mixed in equal numbers before a total of 2 × 10⁷ cells in PBS were injected i.v. into naive and primed B6 mice. Four hours later, mice were sacrificed for their spleens, which were immediately homogenized and transferred onto ice. Up to 2 × 10³ CFSE^low events were acquired for each spleen, and the specific lysis of target cells was calculated using the following formula: percent specific killing = {1 − [(CFSE^int/high event number in T Ag–primed mouse ÷ CFSE^low event number in T Ag−primed mouse) ÷ (CFSE^int/high event number in naive mouse ÷ CFSE^low event number in naive mouse)]} × 100.

A CaspaTag pan-caspase in situ assay kit from Chemicon (catalog no. APT420) was used to identify T_CD8 undergoing apoptosis based on the ability of fluorochrome-labeled inhibitors of caspases (FLICA) to enter cells and covalently bind to, label, and irreversibly inhibit active caspases (49). The FLICA probe used in this kit was the green fluorescent, cell-permeable, nontoxic pan-caspase inhibitor FAM-VAD-FMK. Splenocytes from anti–PD-1- and isotype-treated animals were stained with an allophycocyanin-conjugated anti-CD8α mAb and PE-conjugated H-2K^b/IV or H-2D^b/I tetramer before they were exposed to the FLICA reagent as per the manufacturer’s instructions. The frequency of site I– and site IV–specific T_CD8 emitting a green signal was determined by flow cytometry.

Splenocytes from anti–PD-1- and isotype-treated mice were seeded at 5 × 10⁵ cells per well of a round-bottom microplate. Cells were stimulated with a 1:20 dilution of the 145-2C11 hybridoma supernatant containing ∼0.25 μg/ml of an anti-CD3ε mAb, with 5 μg/ml Con A, or with a combination of 50 ng/ml PMA and 500 ng/ml ionomycin. Plates were kept at 37°C and 6% CO₂ for 72 h, and cells were exposed to 1 μCi of tritiated thymidine ([³H]TdR) during the final 18 h of the incubation period. Cultures were harvested onto glass fiber filter mats, and [³H]TdR incorporation into replicating DNA was quantified by liquid scintillation counting.

Statistical comparisons were carried out with the aid of GraphPad Prism 6 software. We used parametric Student t tests or ANOVA with Holm–Sidak post hoc analysis as appropriate. A p value <0.05 was considered significant (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001).

Despite intense investigations on ID, whether a subdominant status within T_CD8 hierarchies can be due to exhaustion/anergy or at least correlated with coinhibitory proteins T_CD8 may express is unclear. To begin to address this neglected but important question, we examined the expression of classic coinhibitory molecules, namely CTLA-4, PD-1, LAG-3, and TIM-3, by T_CD8 in an in vivo model of ID in which the SV40-encoded T Ag is targeted.

Priming B6 mice with SV40-transformed, T Ag⁺ tumor cells generates T_CD8 clones that recognize four well-defined peptide epitopes of T Ag, termed sites I, II/III, IV, and V (Table I) (38, 43, 48). Site IV is an IDD by virtue of its ability to elicit a rigorous proliferative response, whereas sites I and II/III are considered subdominant determinants (SDDs), and site V–specific T_CD8 are “immunorecessive” and only measurable when recognition of the other epitopes is absent (48, 50). Therefore, T Ag–specific T_CD8 display the following rank order: site IV >> I ≥ II/III >> V.

Nine days after inoculation of B6 mice with C57SV cells, a syngeneic T Ag⁺ cancer cell line, T_CD8 specific for sites IV and I, respectively representing IDDs and SDDs, were readily detectable by tetramer staining in the spleens (Fig. 1A). Cytofluorimetric analysis of site I– and site IV–specific T_CD8 revealed no expression of CTLA-4, LAG-3, or TIM-3 (Fig. 1A, 1B). However, >80% of both T_CD8 populations stained positively for PD-1 (Fig. 1B). Interestingly, however, PD-1 expression on a per cell basis, as judged by the mean fluorescence intensity (MFI) of PD-1, was significantly and reproducibly higher among subdominant site I–specific T_CD8 present in the spleen (Fig. 1C) and within the peritoneal cavity (with an average geometric MFI of 904 and 630 for site I– and site IV–specific peritoneal T_CD8, respectively). Given the i.p. route of C57SV cell injection, peritoneal and splenic T_CD8 responses provide a picture of local and systemic reactivity to T Ag, respectively.

To explore whether the high expression level of PD-1 by site I–specific T_CD8 is biologically significant and could potentially mean higher susceptibility to PD-1 blockade, we administered three separate doses of a blocking anti–PD-1 mAb or isotype control to mice that were injected with C57SV cells. We found that unlike site IV–specific T_CD8 whose percentage increased only marginally in anti–PD-1-treated animals, the site I–specific T_CD8 population almost tripled in size (Fig. 2A). This was not a global effect because T cell proliferation in response to nonspecific stimuli that work by cross-linking TCRs (e.g., anti-CD3ε and Con A) or bypass TCR ligation (PMA plus ionomycin) was comparable in anti–PD-1- and isotype-treated mice (Fig. 2B).

Tetramer reagents identify T_CD8 bearing TCRs of unique specificity for a given peptide:MHC I complex without providing any information on their functional competence. It was thus pertinent to ascertain whether T Ag–specific T_CD8 whose frequency rose upon anti–PD-1 treatment retained their effector functions, including their ability to secrete cytokines. To this end, we performed ICS for IFN-γ, the prototypic effector cytokine of T_CD8, to enumerate functional T cells recognizing each of the four peptide epitopes of T Ag. Blocking PD-1 dramatically increased both the frequency and the absolute number of site I–specific, IFN-γ–producing T_CD8 in the spleen (Fig. 3A). A similar enhancement was observed, albeit to a slightly lesser degree, for T_CD8 targeting site II/III, the other SDD in the T Ag model (Fig. 3A). Consistent with tetramer staining findings (Fig. 2A), anti–PD-1 failed to elevate the percentage of site IV–specific, IFN-γ⁺ T_CD8. Interestingly, the MFI of IFN-γ remained unaltered across all four T_CD8 clones. To be exact, the mean MFI ± SEM of IFN-γ in isotype- and anti–PD-1-treated mice (n = 19 per group) was 1434 ± 171 versus 1453 ± 171 (site I), 1382 ± 206 versus 1268 ± 125 (site II/III), 1711 ± 152 versus 1504 ± 123 (site IV), and 1220 ± 130 versus 1031 ± 99 (site V), respectively. Therefore, PD-1 blockade raises the frequencies of T Ag–derived SDD-specific T_CD8 without changing their IFN-γ production capacity.

Our kinetic studies yielded two additional observations. First, the boosting effect of anti–PD-1 on site I manifested itself at a relatively late time point such that the curves corresponding to anti–PD-1 and isotype control segregated only after day 6 (Fig. 3B). This suggests that PD-1 blockade does not affect the early phase of naive T_CD8 activation. Second, in anti–PD-1-treated animals, the response against site I continued to rise after day 9, the expected peak of T Ag–specific T_CD8 responses (48), and was even more vigorous on day 12 (Fig. 3B). We also assessed memory T_CD8 responses to T Ag epitopes 3 wk after C57SV cell inoculation and demonstrated that the enhancing effect of anti–PD-1 on site I– and site II/III–specific but not site IV–specific T_CD8 was still evident (Supplemental Fig. 1).

Next, we compared the efficacy of anti–PD-1 and anti–PD-L1. Anti–PD-L1 was found to mimic the enhancing impact of anti–PD-1 on site I– and site II/III–specific T_CD8 (Fig. 3C). Furthermore, combining the two mAbs did not increase the frequency of subdominant T_CD8 beyond the levels achieved by either anti–PD-1 or anti–PD-L1 alone (Fig. 3C). These results indicate that: 1) PD-1–PD-L-1 interactions curb T_CD8 responses to SDDs, and 2) PD-1– and PD-L1–blocking mAbs should be equally effective in augmenting subdominant T_CD8 responses in similar experimental or therapeutic settings.

Finally, the preferential effect of anti–PD-1 on subdominant T_CD8 was not limited to the spleen and could also be observed in the peripheral blood (Fig. 3D). Collectively, the above results indicate that interfering with PD-1–PD-L-1 interactions results in a numerical increase in subdominant tumor Ag–specific T_CD8 that maintain their IFN-γ production capacity.

The ultimate role of T_CD8 in cancer immunity is to kill malignant cells displaying cognate peptide:MHC I complexes. Therefore, we employed an in vivo killing assay to extend our findings to this important function. CFSE-labeled syngeneic splenocytes were coated with synthetic peptides corresponding to T Ag epitopes or an IDD of HSV-1, gB₄₉₈, which was used as an irrelevant peptide. Control and cognate target splenocytes were mixed, injected i.v. into naive or C57SV-primed B6 mice, and tracked after 4 h. Although target cells remain intact in naive mice, site IV–coated cells are quickly and completely destroyed in primed animals regardless of the treatment they receive. As expected, gB₄₉₈-coated target cells were not removed from the spleens (Fig. 4A). Importantly, in comparison with the control cohort, anti–PD-1-treated mice exhibited substantially higher cytolytic activities against site I– and site II/III–pulsed splenocytes (Fig. 4). In separate experiments, we found that almost all IFN-γ⁺ T_CD8 and the majority of tetramer⁺ T_CD8 expressed the degranulation marker CD107a (data not shown). Furthermore, CD107a⁺ cell frequencies and CD107a expression levels were comparable in control and anti–PD-1-treated mice (data not shown). These findings demonstrate that PD-1 blockade gives rise to an enlarged pool of cytotoxic T lymphocytes capable of targeting the SDDs of T Ag.

Tumor cells that supply cognate peptide:MHC I complexes (signal 1) along with a requisite costimulatory signal (signal 2) can directly activate naive T_CD8. However, T_CD8 responses to many tumors, especially those of nonhematopoietic origin, which do not express adequate costimulatory molecules (e.g., CD80 and CD86), rely on cross-priming (51). In this pathway, professional APCs (pAPCs), such as DCs, acquire exogenous materials from tumor cells and process their proteins to form peptides, which are then complexed with MHC I and presented alongside costimulatory molecules to naive T_CD8. We previously documented the in vivo significance of cross-priming for antitumor and antiviral immunity (43). Cross-priming also operates in the context of therapeutic vaccination to provoke anticancer T_CD8 activation (52).

C57SV cells used in our immunization protocol should not directly prime naive T_CD8. This is because: 1) they are fibrosarcoma cells, not pAPCs; 2) they express neither costimulatory molecules such as CD40, CD80, CD86, and CD137L (Supplemental Fig. 2) nor MHC class II molecules (Supplemental Fig. 2) that would, at least in theory, enable them to recruit CD4⁺ T cell help; and 3) they are transformed with subgenomic fragments of SV40 (43); therefore, they do not release SV40 virions that could otherwise infect host pAPCs to trigger direct priming. Nevertheless, to more definitively evaluate the beneficial effect of PD-1 blockade on cross-primed anticancer T_CD8, we administered anti–PD-1 to separate cohorts of B6 mice that were immunized with two other cell lines. TAP1^−/− wt T Ag cells are syngeneic renal cells that cannot form peptide:MHC I complexes due to TAP mutation (45). KD2SV kidney epithelial cells are allogeneic (H-2^d) to B6 mice (H-2^b) and should not directly activate host T_CD8 according to the rule of MHC restriction (53). Inoculation of B6 mice with either TAP1^−/− wt T Ag or KD2SV cells resulted in detectable responses to sites I, II/III, and IV, of which only the site I–specific response was boosted by anti–PD-1 (Fig. 5). Consequently, IFN-γ production upon ex vivo exposure to C57SV cells, which was used as a rough indicator of the overall T Ag–specific response, was also elevated. In the case of immunization with KD2SV cells, the effect of anti–PD-1 was so impressive that site I– and site IV–specific T_CD8 became equidominant (Fig. 5B). This effect could also be recapitulated by using B6/K-TagI cells, which only express site I (46), in lieu of the corresponding peptide for ex vivo restimulation of splenocytes (Fig. 5B). Therefore, PD-1 blockade expands the site I–specific T_CD8 clonal size, which is commensurate with a more rigorous response to site I displayed by either APCs or tumor cells.

It is noteworthy that in vivo priming of B6 mice with KD2SV cells also generates T_CD8 alloreactivity to the mismatched MHC, which is independent of T Ag recognition. The allospecific response in this model can be detected after brief ex vivo stimulation of splenocytes with either H-2^d+T Ag⁺ cells (e.g., KD2SV) or H-2^d+T Ag⁻ cells (e.g., P815) (39, 40). In the experiment depicted in Fig. 5B, this response was also augmented upon treatment with anti–PD-1. To be precise, the mean frequencies (±SEM) of alloreactive IFN-γ⁺ T_CD8 recognizing KD2SV cells were 8.19 ± 0.62% in anti–PD-1-treated mice and 3.86 ± 0.82% in isotype-treated controls (p = 0.005, n = 4). Following a similar trend, alloreactive T_CD8 responding to P815 cells ex vivo comprised 3.23 ± 0.26% and 2.11 ± 0.46% of total splenic T_CD8 in anti–PD-1- and isotype control-treated animals, respectively (p = 0.043, n = 4). It needs to be re-emphasized that as demonstrated in Fig. 5B and consistent with our previous studies (38–40, 43), T_CD8 alloreactivity in this model does not alter the hierarchical pattern of syngeneic T_CD8 responses to T Ag–derived epitopes, which we measure as an indication of cross-priming.

Taken together, the above findings imply that PD-1 blockade offers a previously unappreciated benefit through boosting subdominant T_CD8 responses against many tumor cell types that are unable to activate naive T cells on their own.

In the next series of experiments, we sought to determine whether anti–PD-1 could also augment the response to site I in the context of therapeutic vaccination for cancer. B6 mice were immunized either with site I peptide-pulsed DCs (Fig. 6A) or with rVV-I (Fig. 6B), and treated with anti–PD-1 or isotype control. To our initial surprise, anti–PD-1 failed to increase the intensity of site I–specific T_CD8 response and their absolute numbers in both models as evidenced by comparable ex vivo reactivity of splenocytes from anti–PD-1- and isotype-treated animals to site I peptide (Fig. 6A, 6B and data not shown), B6/K-TagI cells, or C57SV cells (Fig. 6B and data not shown). Curiously, however, when we used rVV-FL T Ag as a vaccine, anti–PD-1 increased both the frequency and the absolute number of site I–specific, IFN-γ⁺ T_CD8 in the spleen (Fig. 6C and data not shown). In contrast, T_CD8 responses to other T Ag epitopes, including site IV, were not altered.

Data shown in Fig. 6 suggested that site IV–specific T_CD8 may “dominate” their site I–specific counterparts through a PD-1–dependent mechanism. Immunodomination is experimentally defined as augmented reactivity to SDDs when responses to IDDS are diminished. To demonstrate this phenomenon in a T Ag–based tumor system, we first inoculated mice with B6/K-TagI cells, which lack all T Ag epitopes but site I (46). This was confirmed by detection of a robust T_CD8 response against site I but no other epitopes, which could not be further strengthened upon anti–PD-1 treatment (Fig. 7A). We theorized that anti–PD-1 relieves immunodomination imposed by site IV–specific T_CD8. To test this hypothesis, we injected mice with B6/TpLM237-9Ab cells that are devoid of site IV only (47). As anticipated, the invigorating effect of anti–PD-1 on site I– and sties II/III–specific T_CD8 was abolished in this model (Fig. 7B).

To provide further evidence in support of the above theory, we next primed anti–PD-1- and isotype control–treated mice with peptide-pulsed BMDCs. A cohort of mice received mixed DC populations that were separately pulsed with site I and site IV peptides, and a parallel cohort received DCs that had been copulsed with both sites I and IV. We found treatment with anti–PD-1 to increase the site I–specific response, as judged by both ICS for IFN-γ and tetramer staining, only when copulsed DCs were used (Fig. 8). Therefore, in order for the boosting effect of anti–PD-1 to manifest itself, site I– and site IV–specific T_CD8 clones had to be in close proximity through engaging the same APCs. This is consistent with the hypothesis that PD-1 blockade increases the size of the site I–reactive clone by negating or mitigating the dominating behavior of site IV–specific T_CD8.

The mechanisms underlying immunodomination are not completely understood but may involve retarded subdominant T cell proliferation or their increased susceptibility to death. To find out which of these two scenarios could be reversed by the blockade of PD-1, we stained site I– and site IV–specific T_CD8 for intracellular Ki-67 and active caspases on day 9 after priming. Treatment with anti–PD-1 modestly decreased, rather than increased, the frequencies of tetramer-reactive cells that expressed the proliferation marker Ki-67 (Fig. 9A). Additionally, the MFI of Ki-67 was comparable in both treatment groups (Fig. 9A). However, site I–specific T_CD8 contained substantially more active caspases than did site IV–specific cells (Fig. 9B). Importantly, PD-1 blockade lowered the level of active caspases in site I– but not in site IV–specific T_CD8. Therefore, interfering with PD-1 triggering relieves immunodomination by preventing the lysis of subdominant T_CD8 as opposed to promoting their proliferative growth.

In this work, using a well-established model of antitumor T_CD8 ID, we have demonstrated a new role for PD-1–PD-L1 cross-talk in regulation of anticancer immunity. We found that PD-1–PD-L1 interaction is involved in a mechanism of immunodomination that selectively inhibits subdominant T_CD8 responses to SV40 T Ag. Therefore, interfering with this checkpoint pathway boosts SDD-specific T_CD8, which in turn increases the epitope breadth of the overall T_CD8 response to T Ag. This finding is important because SDDs are capable of conferring protective immunity in certain conditions (54, 55). They are “less visible” to the immune system and may thus escape central or peripheral tolerance mechanisms in mice (56, 57) and humans (58).

The model we employed is clinically relevant for multiple reasons. First, T Ag mediates neoplastic transformation of a variety of mammalian cell types (59). Second, SV40 T Ag is homologous to the BK virus T Ag detected in human kidneys (60). Third, a causal relationship has been recently established between the human Merkel cell polyomavirus large T Ag and a rare but aggressive type of cancer called Merkel cell carcinoma (61, 62). Of note, the presence of T Ag–specific T_CD8 in Merkel cell carcinoma tumors is correlated with better prognosis (63), and vaccination against a “cryptic” epitope of this Ag has shown promise in a preclinical study (64).

One needs to refrain from discounting the role of immunodominant T_CD8 in cancer, as their significance (or lack thereof) can be cancer type–specific even when dealing with the same tumor Ag. Additionally, ex vivo–expanded, IDD-specific T_CD8 can be helpful once included in immunotherapeutic protocols. Previous work on SV40 large T Ag supports the above notions. In SV11 mice, which develop autochthonous T Ag–driven choroid plexus papillomas inside brain ventricles (65), endogenous T_CD8 against T Ag epitopes are deleted due to negative selection in the thymus. However, adoptive transfer of naive B6 splenocytes into sublethally irradiated SV11 mice results in extended control of tumors, which is associated with in vivo priming of site IV–specific immunodominant T_CD8 among transferred splenocytes (66, 67) and the ability of splenocytes to produce IFN-γ (66). These observations simulate promising clinical results achieved in certain cancers after adoptive transfer of ex vivo–expanded autologous T_CD8, particularly when it is preceded by lymphodepletion (68, 69). Alternatively, in the transgenic adenocarcinoma of the mouse prostate (TRAMP) model (70), in which T Ag is expressed under the control of rat probasin promoter when male mice hit puberty, the response to site IV fizzles out with progression of the cancer. However, the immunorecessive site V–specific T_CD8 escape thymic detection and avoid peripheral tolerance (71), thus providing opportunities for therapeutic interventions targeting this epitope.

We have used several different adjuvant strategies in the past in an attempt to boost subdominant T_CD8 responses to T Ag, albeit to little avail. For example, when we inhibited IDO (39) or used an immunostimulatory dose of the mTOR inhibitor rapamycin (40), we could only boost the dominant response to site IV. Therefore, it appears that SDD-specific T_CD8 responses are refractory to many forms of treatment. As such, the findings of our current study are important because treatment with anti–PD-1, but not with several other checkpoint inhibitors (data not shown), could remarkably boost the responsiveness of subdominant T_CD8. Accordingly, we propose that the beneficial effect of anti–PD-1 or anti–PD-L1 Abs in certain cancers may be, at least partially, due to their ability to augment subdominant T cell responses. By the same token, this effect may be absent in cancers that do not respond favorably to PD-1 blockade (e.g., prostate cancer). In fact, anti–PD-1 monotherapy in TRAMP mice does not make T Ag–specific T_CD8 clones detectable, regardless of their hierarchical position (A. Memarnejadian and S.M. Mansour Haeryfar, unpublished observations). It is possible though that a combination of anti–PD-1 and other immuno-, radio-, or chemotherapeutic modalities may exhibit additive or synergistic benefits. When we combined nTreg cell depletion and anti–PD-1, we noticed a greater effect on site I– and site II/III–specific responses, but none on site IV (Supplemental Fig. 3A, 3B). The additive nature of this strategy suggests that PD-1 blockade and nTreg cell depletion work through different mechanisms. We were also pleasantly surprised by the sudden appearance of a modest but reproducible response to site V (Supplemental Fig. 3). Another tempting possibility that merits consideration is to combine PD-1–targeted therapies with approaches or agents that strengthen IDD-specific T_CD8, such as IDO and mTOR inhibitors (39, 40) among many others, in appropriate settings, to boost SDD- and IDD-specific clones alike.

From a mechanistic standpoint, several findings suggest that blocking PD-1–PD-L1 interactions could selectively boost subdominant clones by reversing immunodomination by site IV–specific T_CD8. First, the observed rise in site I–specific T_CD8 was numerical in essence, and these cells did not express higher levels of IFN-γ or CD107a, on a per cell basis, in anti–PD-1-treated animals. Second, although the beneficial effect of anti–PD-1 could also be demonstrated on memory T_CD8 (Supplemental Fig. 1), this was also seemingly numerical and not due to an early commitment to generate more memory T_CD8 precursors, which are phenotyped as CD127^highKLRG1^low cells (40) (Supplemental Fig. 4). Third, the enhancing effect of anti–PD-1 was noticeable relatively late in the course of our kinetic studies, not during the initial priming by APCs (Fig. 3B). Fourth, in DC- and rVV-based immunization settings, in which we targeted site I alone, anti–PD-1 failed to elevate the response (Fig. 6A, 6B). Therefore, to conclusively demonstrate the “anti-immunodomination” effect of PD-1 blockade, we used two different tumor cell lines expressing T Ags that lacked site IV. These experiments confirmed that PD-1 engagement indeed enforces site IV–mediated immunodomination. This conclusion was finally supported by our finding, in DC transfer experiments, that the selective enhancement of the site I–directed response was evident only when site I– and site IV–specific T_CD8 were physically adjacent and engaging the same APCs (Fig. 8). To shed more light on this phenomenon, we sought to rule in or rule out the possibilities that the numerical rise in site I–specific T_CD8 was owed to their increased proliferation or survival, scenarios that are not mutually exclusive. We found the frequency of Ki-67⁺ cells among site I– and site IV–specific T_CD8 to be moderately decreased, rather than increased. In contrast, only site I–specific T_CD8 contained high levels of active caspases, which could be lowered in anti–PD-1-treated mice. Therefore, PD-1 blockade prevents “lysis-dependent” immunodomiantion of site I–specific T cells, which is unlike other mechanisms of immunodomination reported to date (27, 28, 41).

We found that C57SV cells, the T Ag⁺ fibrosarcoma line that was frequently used in this investigation, do not express PD-L1 (Supplemental Fig. 2). Therefore, they should not be able to kill naive T_CD8 even in the unlikely event they might form stable and sustained immunological synapses with these cells. Moreover, if cross-presenting DCs had somehow selectively killed site I–specific T_CD8 through a PD-L1/PD-1–dependent mechanism, anti–PD-1 treatment should have enhanced the response to site I when it was presented alone and in the absence of other T Ag epitopes. Alternatively, cognate T_CD8 cannot remove APCs during the priming phase because they are not yet armed with a cytotoxic arsenal. In comparison, effector T_CD8 may “bite the hands that feed them” by eliminating APCs (72). Nevertheless, this possibility also seems remote. If PD-L1 and PD-1 were sufficiently expressed by T_CD8 and APCs, respectively, treatment with anti–PD-1 would likely have an indiscriminate effect on T_CD8. Because anti–PD-1 administration selectively diminishes the intracellular caspase content of site I–specific T_CD8 (Fig. 8B), we believe that blocking PD-1 prevents the fratricidal death of these cells.

It is currently unclear why subdominant site I–specific T_CD8 express a higher level of PD-1 to begin with. Future investigations will address whether site I–specific T cells have a lower activation threshold resulting in swift, robust PD-1 expression and/or a propensity to retain PD-1 on their surface for a longer period of time. It is plausible to assume a link between the kinetics and stability of peptide:MHC complex formation for various IDDs and SDDs in the T Ag recognition model and the strength/sustenance of TCR triggering, which could in turn control the intensity of PD-1 expression. Additionally, the balance/imbalance between a myriad of costimulatory and coinhibitory signals should influence PD-1 expression and functions. This possibility is not far-fetched in light of findings in other models that CD40 ligation inhibits PD-1 induction (73) and that the efficacy of PD-1–based checkpoint inhibitor therapy is CD28-dependent (23).

To summarize our findings, our work reveals that ID hierarchies of antitumor T_CD8 can be governed by PD-1–PD-L1 interactions. Blocking PD-1 broadens antitumor T_CD8 responses, thus providing the host with more target choices, some of which may not evade immune detection or paralyze T cells. This represents a previously unrecognized effect of PD-1–targeted therapies. Interfering with PD-1 engagement blocks lysis-dependent immunodomination of subdominant T_CD8. By lifting this pressure, PD-1–targeted therapies reinvigorate subdominant T_CD8 responses that can potentially contribute to antitumor immunity. These findings should be considered in PD-1–based immunotherapies and in rational vaccine design for cancer.

We thank members of the Haeryfar Laboratory for helpful discussions and Jeremy Haley for preparation of MHC tetramer reagents.

The authors have no financial conflicts of interest.