Heterologous Vaccination and Checkpoint Blockade Synergize To Induce Antileukemia Immunity

Patients with B cell acute lymphoblastic leukemia (B-ALL) harboring the BCR-ABL chromosomal translocation have very poor outcomes (1, 2). Current therapies for BCR-ABL⁺ B-ALL include cytotoxic chemotherapeutics, tyrosine kinase inhibitors, and bone marrow transplantation. These treatments are often transiently effective, indicating that new treatment options are urgently needed. One such option is immunotherapy. Recent work in cancers with frequent nonsynonymous mutations, such as melanomas, has demonstrated that immunotherapy involving neutralization of programmed death 1 (PD1) and CTLA4 (checkpoint blockade) is an effective treatment option (3, 4). It remains unclear whether immunotherapy involving checkpoint blockade strategies will also be effective in cancers with few nonsynonymous mutations, such as B-ALL (5).

To determine whether immunotherapy is an effective option for treating B-ALL, we used a syngeneic mouse model of BCR-ABL⁺ B-ALL to characterize the host immune response to this leukemia in immune-competent recipient animals (6–8). We previously demonstrated that the host adaptive immune system responds to BCR-ABL⁺ B-ALL (9). Although B-ALL cells have been shown to have low numbers of nonsynonymous mutations (5), the fusion between BCR and ABL does generate an MHC class II (MHC-II)–restricted peptide Ag that can be recognized by a small population of endogenous BCR-ABL peptide (BAp):I-A^b–specific T cells in mice (9). Transfer of BCR-ABL⁺ leukemic cells into C57BL/6 mice resulted in proliferation of BAp:I-A^b–specific T cells, although 50% of these cells differentiated into FOXP3⁺ regulatory T cells (Tregs) (10). Thus, T cells do respond to BCR-ABL⁺ leukemia in this mouse model, but the response was immunosuppressive in nature and detrimental to host survival. In this study, we address whether the immune response to leukemia could be modulated, thus making BCR-ABL⁺ B-ALL malleable to checkpoint blockade–based T cell immunotherapy.

C57BL/6 mice and Cdkn2a^−/− (strain 01XF6, B6, 129-Cdkn2a^tm1Cjs/Nci) (11) mice came from the National Cancer Institute. Foxp3-GFP (stock no. 006772) and Ifng^−/− (stock no. 002287) mice came from The Jackson Laboratory (Bar Harbor, ME). OT-I × Rag2^−/− mice were generated locally as previously described (12). Mice were housed at the University of Minnesota in specific pathogen-free conditions or biosafety level 2 facilities, and all experiments were approved by the Institutional Animal Care and Use Committee.

ActA–L. monocytogenes strain 1942 (from Dr. Sing Sing Way) expressing BAp from a plasmid was constructed as previously described (13–15).

L. monocytogenes expressing BAp (LM+BAp) (10⁷ CFU) was injected i.v. through the tail vein. Mice vaccinated with lymphocytic choriomeningitis virus (LCMV)–Armstrong received 2 × 10⁵ PFU i.p. at day 0. Vesicular stomatitis virus (VSV)–Indiana was used at 5 × 10⁵ PFU i.v. at day 0. At days 3 and 5, mice were injected i.v. with 200 μg BAp. Mice were harvested at indicated time points.

Cdkn2a^−/− mouse bone marrow cells were transduced with viral supernatant containing a BCR-ABL (P190)–IRES-GFP retrovirus (16) and cultured for adoptive transfer as previously described (7, 9).

Purified monomer was tetramerized with streptavidin-PE or streptavidin-allophycocyanin and cells were enriched as previously described (9, 17).

Unvaccinated mice were treated with 100 μg anti–PD ligand 1 (PDL1) and anti-CTLA i.p. every other day. Vaccinated mice received 200 μg anti-PDL1 and anti-CTLA i.p. twice per week. Mice treated with anti-CD40 received 200 μg i.p. every other day.

Abs for flow cytometry included CD3 PE, CD4 (RM4-5) PerCP-Cy5.5, CD8 (53-6.7) BV650, CD11c (N418) PE, FOXP3 (FJK16S) PE, CD80 (16-10A1) allophycocyanin, CD86 (GL1) PE-Cy7, CD19 BV605, B220 (RA3-6B2) Horizon V500, IFN-γ (XMG1.2) BV650, LAP (TW7-16B4) PE, TNF-α (MP6-XT22) BV421, IL-17A (TC11-18H10) Alexa Fluor 488, and PSGL1 (2PH1) BV421 purchased from BD Biosciences (San Jose, CA); NK1.1 (PK136), CD11b (M1/70), CD11c (N418), B220 (RA3-6B2), and F4/80 in allophycocyanin–eFluor 780, and PD1 (J43) FITC, CD73 eFluor 450, FR4 PE-Cy7, PDL1 PerCP–eFluor 710, MHC-II I-A^b eFluor 450, IL-10 (JESS-16E3) PE, granzyme B (NGZB) PE-Cy7, GARP (YGIC86) eFluor 450, and all ELISPOT Abs were purchased from eBioscience (San Diego, CA); and IgM F(ab′)₂ allophycocyanin was purchased from Jackson ImmunoResearch Laboratories (West Grove, PA). Rat IgG1 (HRPN) PerCP-Cy5.5 isotype and rat IgG2a (2A3) violet eFluor 450 isotype were purchased from Tonbo Biosciences (San Diego, CA). Cells from enriched fractions were analyzed on LSRII or Fortessa cytometers (BD Biosciences, San Jose, CA) and data were analyzed in FlowJo (Tree Star, Ashland, OR).

Standard normality tests suggested departures from normality, and thus nonparametric tests (Mann–Whitney U test for two groups, Kruskal–Wallis and Dunn tests for more than two groups) were used unless otherwise stated. Normality assessments, nonparametric tests, and survival analyses were done in GraphPad Prism (La Jolla, CA). For the Cox–Mantel tests, we report hazard ratios, which describe the multiplicative change in risk when moving from the baseline group to the treatment group. Principal component analysis (PCA) was conducted in R (prcomp function) (18). Linear regressions and correlation coefficients were estimated in GraphPad Prism and R.

Detailed descriptions of the PCA and corresponding linear regression are included below. We performed a PCA on the following five phenotype metrics collected on each mouse: PDL1, MHC-II, CD40, CD80, CD86. We added one to each metric, and then log transformed the resulting value so that our data met the PCA assumption of joint normality. Components were estimated using the prcomp function in the stats package in R.

First, pair plots of the raw manifest variables and log-transformed manifest variables were created. PCA is a method for reducing the dimension of a dataset by transforming an initial set of possibly correlated manifest variables into a set of new orthogonal variables, which are referred to as PCs. These components describe the multivariate correlation in the dataset. Although the method generates as many components as there are measured variables in the dataset, most of the variation can usually be captured with only a few components. Each component consists of a value that describes the proportion of variation in the original dataset explained by the component, and a set of loadings that describe the extent to which each manifest variable correlates with that component. Let X be an n × k matrix containing measurements of k different manifest variables for n sampled individuals. Estimation is obtained through an eigen decomposition of the square matrix X′X. Eigenvalues correspond to proportions of variance in the original dataset captured by each component, and eigenvectors describe correlations between each manifest variable and each component. Although the method constructs as many PCs as there are manifest variables in the dataset, interpretation is limited to those components that explain the preponderance of variation. The number of components to interpret is often determined using a scree plot, showing the proportion of variance explained by each component.

A scree plot for the components identified for the log-transformed immunogenicity phenotype metrics is shown in Fig. 1D. Based on the scree plot, we interpreted the first two components. Loading of each manifest variable on each component is shown in Table I.

We extracted PC scores for PC1 and PC2 for each mouse in our dataset. We used a linear regression model to relate these PC scores to percentage leukemic burden measured on these same mice. The regression model consisted of four terms: an intercept (β₀), main effects for each PC (β₁ and β₂), and an interaction term between the two PCs (β₃). Let y_i be the percentage leukemic burden in the ith studied mouse, let X_1,i be the ith mouse’s PC1 score, and let X_2,i be the ith mouse’s PC2 score. Then the regression model we fit can be written as: y_i = β₀ + β₁(X_1,i) + β(X_2,i) + β( X_1,i)( X_2,i) + ε_i. The model was fit using the lm function in R (1). An overall F test clearly suggested that at least some of the coefficients in the model differed significantly from 0 (F statistic = 23.76 on 3 and 26 df; p < 0.0001). Specifically, the model detected strong relationships between the first two PCs and leukemic burden, as well as a marginally significant interaction effect in our dataset.

In general, the intercept term corresponds to expected leukemic burden for mice with average scores on both PC1 and PC2. Specifically, under this model we expect that mice with average PC1 and PC2 scores have an average leukemic burden of 62.64%. For mice with an average score of PC2, but who are 1 unit above average on PC1, we expect an average leukemic burden of 52.64% (62.64 − 10.00%). For mice with an average score on PC1, but who are 1 unit about average on PC2, we expect an average leukemic burden of 43.34% (62.64 − 19.30%). For mice that are 1 unit above average on both PC1 and PC2, we expect an average leukemic burden of 27.38% (62.640−10.00−19.30−5.96%).

We previously showed that there was a higher fraction of live leukemic cells in the bone marrow and secondary lymphoid organs of OT-I × Rag2^−/− mice than in C57BL/6 mice. Furthermore, the range of leukemic burdens was quite broad in the C57BL/6 hosts (interquartile range [IQR] = 11–69%) but less so in the OT-I × Rag2^−/− hosts (IQR = 90–98%) (9). To understand why there was such a range in leukemic burden, we looked for characteristic differences in the leukemic cells from mice with low versus high leukemic burden. Because B cells can function as APCs, we examined the expression of MHC-II to stratify the leukemic burden based on expression of surface markers. MHC-II expression inversely correlated with leukemic burden (mice that had a low percentage of leukemic cells in the bone marrow had high MHC-II expression on the leukemic cells; Fig. 1A). We also examined the expression of the costimulatory molecules CD40, CD80, CD86, and PDL1. None of these costimulatory molecules individually correlated with leukemic burden. Therefore, we used PCA to identify whether there was an ensemble of costimulatory molecules that correlated with leukemic burden (Fig. 1B). The first component described a positive correlation between MHC-II, CD80, CD86, and PDL1. The second component was driven by a negative correlation between CD40 and PDL1 (Fig. 1B). These first two components described 77% of the variation in leukemic burden that we observed in mice (Fig. 1C, Table I). Mice that had low leukemic burden tended to have high scores for both PC1 and PC2 (Fig. 1D). Therefore, we used linear regression to examine relationships between the first two PC scores and leukemic burden (Fig. 1E). Both high PC1 and high PC2 scores were associated with significantly decreased leukemic burden (p < 0.001, Fig. 1E). A low PC1 score (score of −2) was predictive of high leukemic burden (Fig. 1E, left panel) regardless of PC2 score. In contrast, leukemias with higher PC1 scores (score of 0–2) showed dependence on PC2 in predicting leukemic burden (Fig. 1E). Taken together, this analysis supports the conclusion that robust Ag presentation combined with CD80/86 costimulation (PC1), as well as a high ratio of CD40/PDL1 (PC2), correlates with improved disease outcome.

Our PCA suggested that an ensemble of costimulatory molecules (CD80, CD86, PDL1, CD40) functioned as a cohesive unit to modulate antileukemia immunity. Nonetheless, it was possible that individual targeting of Ag presentation and costimulatory molecules might change the disease course. We tested whether Ab targeting of PDL1, CTLA4, or CD40 would be sufficient to change leukemia progression. Ab blockade of PDL1 and CTLA4 (either individually or in combination) led to a modest but significant increase in survival of leukemic mice (Fig. 2A–C). Additionally, treatment of leukemic mice with an anti-CD40 Ab that is characterized as an agonist also led to a modest yet significant increase in survival (19) (Fig. 2D). Thus, the components defined by our PCA do not individually identify very effective therapeutic targets in this model. These results support the concept that the molecules identified by our PCA are best addressed as an ensemble.

None of the immune checkpoint modulations we tested substantially improved survival of leukemic mice. However, we have previously shown that Ly6C⁺ BAp:I-A^b–specific T cells correlate with antileukemia immunity upon Treg depletion (9). We attempted to recreate an inflammatory environment to generate many Ly6C⁺ BAp:I-A^b–specific T cells. To do this, we infected mice with LM+BAp, which caused a 65-fold increase in BAp:I-A^b–specific T cell numbers. In parallel, we infected mice with either LCMV-Armstrong or VSV-Indiana and then delivered 200 μg BAp i.v. at 3 and 5 d postinfection. This allowed us to use the inflammation caused by acute viral infection to induce a strong BAp:I-A^b–specific CD4⁺ T cell response (termed LCMV+BAp or VSV+BAp). At peak infection LCMV+BAp caused a 74-fold proliferation of BAp:I-A^b–specific T cells, whereas VSV+BAp caused a 114-fold proliferation of BAp:I-A^b–specific T cells (Fig. 3A). These results show that BAp:I-A^b–specific T cell proliferation can be initiated by immunization. Additionally, we found that LCMV+BAp induced a high frequency of Ly6C⁺ memory BAp:I-A^b–specific T cells following leukemia rechallenge, whereas LM+BAp induced substantially fewer Ly6C⁺ memory BAp:I-A^b–specific T cells following leukemia rechallenge (Fig. 3B). Because our previous work showed that Ly6C was expressed on most BAp:I-A^b–specific T cells upon Treg depletion (which also resulted in significantly less leukemic burden and significantly longer survival of leukemic mice), we reasoned that acute viral infections that result in increased Ly6C expression might induce protective BAp-specific immunity.

Our PCA suggested that MHC-II expression, and thus Ag presentation, was important in describing the immune response to leukemia. We have identified one peptide from BCR-ABL (BAp) that is processed and presented on MHC-II in vivo (9). Thus, we hypothesized that immunization with BAp plus strong adjuvants might mediate protection from BCR-ABL⁺ B-ALL in mice. To test this, we infected mice with either LCMV-Armstrong (with and without BAp) or VSV-Indiana (with and without BAp) and rechallenged mice with BCR-ABL⁺ leukemia >40 d later, a memory time point when no acute inflammation remained (Fig. 3C). Mice that were infected with an acute viral pathogen plus BAp survived long-term. In contrast, mice that were infected with an acute viral pathogen in the absence of BAp succumbed to leukemia rapidly. The hazard ratio comparing all “+BAp” vaccinations to all “−BAp” vaccinations in Fig. 3C was 0.24 with a 95% CI from 0.12 to 0.46 (p < 0.0001). Thus, BAp-specific adaptive immunity confers long-term survival in this model. Because BAp only binds to MHC-II and not MHC class I (data not shown), our results support the conclusion that BAp:I-A^b–specific T cells are critical for protecting against BCR-ABL⁺ leukemia in our prophylactic vaccination studies.

In CD4⁺ T cells, IFN-γ is normally produced by Th1 cells, which can have a role in antitumor immunity (20). We have previously shown that in unvaccinated mice most BAp:I-A^b–specific T cells responding to BCR-ABL⁺ leukemia are Tregs and thus are likely not making IFN-γ. Consistent with this idea, we found that the ability of host T cells to make IFN-γ in unvaccinated mice did not affect survival following leukemia inoculation, because Ifng^−/− hosts succumbed to leukemia similarly to C57BL/6 hosts (Fig. 3D). However, we hypothesized that IFN-γ might play a role in the adaptive immune response to leukemia following prophylactic vaccination. To test this, we vaccinated Ifng^−/− mice with LCMV-Armstrong+BAp and rechallenged with leukemia at >40 d postvaccination. Vaccination of IFN-γ–deficient mice did not increase survival following challenge with leukemia, when compared with unvaccinated control mice. In contrast, vaccination of Ifng-replete mice resulted in significant long-term survival when compared with unvaccinated controls (Fig. 3E). Thus, IFN-γ production was one mechanism required for effective antileukemia immunity following prophylactic vaccination.

Immune memory is a critical component of prophylactic vaccination. To determine whether effective BAp:I-A^b–specific memory T cells were formed by vaccination, we infected mice with LM+BAp or LCMV+BAp and waited 40 d to enumerate BAp:I-A^b–specific memory T cells. We recovered significantly more memory BAp:I-A^b–specific T cells from LCMV+BAp–infected mice than from LM+BAp–infected mice (Fig. 4A). We then vaccinated mice with either LCMV+BAp or LM+BAp and rechallenged them by transferring 2500 leukemic cells into the mice 30 d later. Vaccination with LCMV+BAp, but not LM+BAp, led to a significant increase in the number of BAp:I-A^b–specific T cells following leukemia challenge and decreased leukemic burden (4-fold, Fig. 4B, 4C). Thus, the increase in BAp:I-A^b memory T cell numbers following LCMV+BAp but not LM+BAp vaccination correlated with disease outcome.

The quality and quantity of BAp:I-A^b-specific memory T cells was different comparing LM+BAp vaccination to LCMV+BAp vaccination. We observed that the IQR of leukemic burdens in the mice vaccinated with LCMV+BAp was broad (IQR = 1.4 × 10⁶, Fig. 4C), showing that protection mediated by LCMV+BAp vaccination was more effective in some mice than in others. We previously observed that Ly6C expression was increased on BAp:I-A^b–specific T cells when Tregs were depleted (9). Therefore, we examined whether Ly6C expression on BAp:I-A^b–specific T cells correlated with leukemic burden. Mice with high leukemic burden despite prophylactic vaccination (and thus considered “failed vaccinated mice”) had a significantly lower percentage of Ly6C⁺ BAp:I-A^b–specific T cells than did the “successfully vaccinated mice” (Fig. 4D). Additionally, significantly more BAp:I-A^b–specific T cells expressed Ly6C after LCMV+BAp vaccination (which lowered leukemic burden) than did LM+BAp (which had no effect on leukemic burden) (Fig. 3B). Importantly, the number of Ly6C⁺ BAp:I-A^b–specific T cells was inversely correlated with leukemic burden in LCMV+BAp–vaccinated mice (Fig. 4E). In contrast, leukemic burden did not correlate with total CD4⁺Ly6C⁺ cells in these mice (Fig. 4F). We also observed that Tregs made up a smaller portion of the BAp:I-A^b–specific T cell population in mice that were prophylactically vaccinated with LCMV+BAp than in unvaccinated mice (Fig. 4G). These results support a functional role for Ly6C⁺FOXP3⁻ BAp:I-A^b–specific T cells during the immune response to leukemia following prophylactic vaccination.

Our PCA suggests that high ratios of CD40/PDL1 and MHC-II/PDL1 may be predictive of low leukemic burden. We examined the leukemic cells from LCMV+BAp–vaccinated mice and LM+BAp–vaccinated mice. First, we found that leukemias in mice that were successfully vaccinated with LCMV+BAp had higher expression of CD40 and MHC-II than did their “failed vaccination” counterparts (Fig. 5A, 5B). Second, we found that CD40/PDL1 and MHC-II/PDL1 increased on LCMV+BAp–vaccinated mice (which was an effective vaccination regimen, Fig. 5C, 5D) but not significantly on LM+BAp–vaccinated mice (an ineffective vaccination regimen, Fig. 5E, 5F). Thus, prophylactic vaccination with acute viral pathogens plus BAp results in protection from leukemia and correlates with expression of the biomarkers that we previously demonstrated were linked to strong antileukemia immune responses (Fig. 1).

We hypothesized that a proinflammatory environment might counter leukemia-derived immune suppression while also inducing BAp-specific adaptive immunity, and thus inhibit leukemia progression. To test this hypothesis, we therapeutically vaccinated mice, which had established leukemia, using either homologous vaccinations with LCMV-Armstrong+BAp or heterologous vaccinations with LCMV-Armstrong+BAp, LM+BAp, and VSV-Indiana+BAp (Fig. 6A). Homologous vaccination with LCMV+BAp significantly prolonged survival, although all mice ultimately succumbed to leukemia. Heterologous vaccination should create a more robust proinflammatory response, because Abs created during the primary infection will not neutralize the secondary and tertiary infections. Indeed, heterologous vaccination was significantly more effective and led to long-term survival (more than twice the median untreated survival) in ∼10% of mice. Thus, repeated vaccination with heterologous agents was an effective treatment strategy in mice with BCR-ABL⁺ B-ALL.

The immune response to acute viral and bacterial infection is canonically proinflammatory. However, because mice with active leukemia have high doses of leukemia Ags during this proinflammatory state, this may cause chronic Ag stimulation, a situation where PDL1 signaling is highly expressed (21). Additionally, our initial findings showing that CD44 was not highly expressed on all BAp:I-A^b–specific T cells responding to leukemia suggest that BAp-specific T cell priming is not optimal (9). CTLA4 blocks interaction of CD28 with B7-1 and B7-2 molecules, thereby reducing T cell functionality (22–24). Thus, we hypothesized that therapeutically vaccinated mice that were treated with dual PDL1/CTLA4 checkpoint blockade might show improved survival. This treatment strategy led to a significant increase in survival beyond that seen for either PDL1+CTLA4 blockade (Fig. 2C) or therapeutic vaccination (Fig. 6A), with 31% of mice surviving long-term. Because this long-term survival is far past the time point when inflammation would remain from the therapeutic vaccination, it suggests that an adaptive immune response is mediating long-term survival.

To understand some of the mechanisms allowing effective therapeutic vaccination, we compared homologous, heterologous, and heterologous plus checkpoint blockade treatments and assessed BAp:I-A^b–specific T cell expansion and effector function. To do this, we inoculated mice with BCR-ABL⁺ leukemia and started therapeutic vaccination at the same time point. We then harvested the mice 21 d later and enumerated BAp:I-A^b–specific T cells. We found that all regimens induced robust proliferation (∼1250-fold over naive precursor numbers); however, there was no difference in the number of BAp:I-A^b–specific T cells recovered between any of the three treatment groups (Fig. 6B). Because total numbers of BAp:I-A^b–specific T cells did not help give insight into the mechanisms that allowed effective therapeutic vaccination, we examined the phenotype of the BAp:I-A^b–specific T cells. We reasoned that BAp:I-A^b–specific T cells should take on a more Th1-like phenotype in response to therapeutic vaccination, and that this phenotype should correlate with improved disease outcome. However, we found no significant increase in IFN-γ or TNF-α (two canonical Th1 cytokines) when comparing all three treatment groups. Interestingly, in mice that received heterologous vaccination plus checkpoint blockade, we found that a larger fraction of BAp:I-A^b–specific T cells produced more IL-10, granzyme B, and both IL-17 and granzyme B together (Fig. 7A, 7C–F), all of which have previously been associated with proinflammatory tumor clearance (25–28). Additionally, we found that PD1 expression on BAp:I-A^b–specific T cells positively correlated with leukemic burden in all therapeutically vaccinated mice (Fig. 7B), and that PD1 expression was lowest on these cells in heterologous vaccination plus checkpoint blockade–treated mice (Fig. 7A). Taken together, these results demonstrate that polyfunctional CD4⁺ leukemia-specific T cells produce a combination of IL-10, IL-17, and granzyme B, and this correlated with effective antileukemia adaptive immunity.

BCR-ABL⁺ B-ALL is only transiently responsive to current therapies (29), with a 5-y survival of ∼35% (30, 31). Given this low survival rate, additional therapies are needed. One approach that has not been well explored in this disease is checkpoint blockade–based immunotherapy. Checkpoint blockade works well in malignancies with many nonsynonymous mutations and can lead to improved long-term survival in patients with such cancers (4, 32–36). Presumably, this is because the increased number of nonsynonymous mutations allows for a concurrent increase in the number of neoantigen-specific T cells. Comparatively, in cancers with low numbers of nonsynonymous mutations, such as B-ALL (36), checkpoint blockade–based immunotherapy targeting the endogenous immune response is relatively unstudied. In fact, current paradigms suggest that cancers with low numbers of nonsynonymous mutations may not be effectively treated using anti-CTLA4 and anti-PD1 checkpoint blockade (35). In the present study, we show that an endogenous T cell response can be effective in controlling BCR-ABL⁺ B-ALL, but this requires both checkpoint blockade and an intensive heterologous vaccination strategy.

In this study, we found that a strong immune response correlated with decreased leukemic burden. MHC-II expression on leukemic cells correlated with disease outcome, which hinted that CD4⁺ T cells were important for antileukemia immunity. We have previously shown that the leukemia Ag BAp is presented on MHC-II (9). In contrast, BAp does not bind to MHC class I in C57BL/6 mice (data not shown). These findings, in combination with our studies showing that the presence of BAp during prophylactic vaccination is required for protection from leukemia (Fig. 3C), provide evidence supporting a role for MHC-II–mediated presentation of BAp in antileukemia immunity. Moreover, an effective antileukemia immune response requires IFN-γ and correlates with increased induction of Ly6C on the BAp:I-A^b–specific T cells (Fig. 3D, 3E).

Our PCA strongly suggested a role for targeting the immune checkpoint molecules PDL1 and CTLA4 as part of a therapeutic strategy for BCR-ABL⁺ B-ALL. Costimulatory and coinhibitory molecules play a role in cancer progression (4, 37–42). In our model we observed statistically significant increases in survival upon monotherapy with either anti-PDL1 or anti-CTLA4, or dual checkpoint blockade therapy with both anti-PDL1 and anti-CTLA4. However, despite achieving statistical significance, the effects of checkpoint blockade alone were modest (increased survival of 2–4 d), and thus possibly of limited biological impact. The limited impact of checkpoint blockade treatment alone in BCR-ABL⁺ leukemia fits prevailing concepts regarding checkpoint blockade. Current models suggest that cancers with low numbers of nonsynonymous mutations will not be susceptible to checkpoint blockade (35, 36). Our leukemia model likely has few nonsynonymous mutations, and checkpoint blockade is not very effective in this leukemia model. Thus, our observations are consistent with the idea that small numbers of nonsynonymous mutations result in poor anticancer responses following checkpoint blockade therapy.

Immune checkpoint blockade therapy alone was only minimally effective in treating leukemic mice in our model. Thus, we explored therapeutic vaccination immunotherapy. Two lines of evidence precipitated this strategy. First, previous reports show that therapeutic heterologous vaccination can be effective in other cancers, albeit those with higher mutation rates (43). Second, it was clear that MHC-II–mediated Ag presentation was important for leukemia outcome, and the pathogens used in our therapeutic vaccination scheme all induce MHC-II expression on APCs (44–46). When mice were therapeutically vaccinated with these MHC-II–inducing proinflammatory pathogens, we saw increased survival (Fig. 6). We used heterologous vaccination because this approach has previously been shown to be effective at inducing a robust T cell response (47–49). In this approach, we used multiple infectious adjuvants to generate a proinflammatory environment that should promote robust adaptive immune activation. Similar approaches have been used prophylactically (47) and therapeutically for cancer (43, 49). However, our study examines therapeutic heterologous vaccination in combination with checkpoint blockade specifically to target CD4⁺ T cells in cancer, an underexplored field.

We found that therapeutic vaccination synergized with anti-PDL1 and anti-CTLA4 therapies to improve long-term survival in mice with BCR-ABL⁺ leukemia. Thirty-one percent of the mice that received therapeutic heterologous vaccination in combination with anti-PDL1 and anti-CTLA4 checkpoint blockade exhibited long-term survival. In contrast, only 10% of mice treated with therapeutic heterologous vaccination alone survived long-term. Furthermore, no leukemia-bearing mice treated therapeutically with checkpoint blockade alone exhibited long-term survival. These results suggest that even malignancies with few nonsynonymous mutations (such as B-ALL) can be responsive to immunotherapies that classically work well only in malignancies with high levels of nonsynonymous mutations (35). Importantly, such results are contingent upon intensive therapeutic vaccination approaches. One possible explanation for the synergistic effect of vaccination plus checkpoint blockade is that leukemia-derived Ag is available for the entire duration of the therapeutic vaccination regimen. This chronic Ag stimulation may lead to continual high expression of PDL1 and CD80/86 on leukemic cells, which may explain the synergy between therapeutic vaccination (which is susceptible to inhibition by PDL1/PD1 and CD80/86/CTLA4 pathways) and dual checkpoint blockade (which inhibits those pathways). Finally and importantly, note that oncolytic viruses (which include VSV, used in our scheme) have been used for anticancer immunotherapy in the past (50) and are currently being used in clinical trials as a treatment option for cancer (51, 52). Thus, the approach taken to treat leukemia in our murine model is feasible to consider for human patients with BCR-ABL⁺ leukemia.

Our data provide initial mechanistic insights into how therapeutic vaccination therapy plus checkpoint blockade can lead to leukemia rejection by the C57BL/6 host. First, we saw a trend toward decreased PD1 expression on BAp:I-A^b–specific T cells that correlated significantly with decreased leukemic burden (Fig. 7A, 7B). Checkpoint blockade could interfere with this potential mechanism of tolerance induction. Second, during the therapeutic vaccination response, we saw that many BAp:I-A^b–specific T cells were polyfunctional (producing granzyme B and multiple cytokines such as IFN-γ, TNF-α, IL-10, and IL-17). Importantly, in the most effective vaccination regimen (therapeutic heterologous vaccination plus checkpoint blockade), we saw a significantly increased fraction of BAp:I-A^b–specific T cells that produced granzyme B, IL-10, and a combination of granzyme B plus IL-17. This observation demonstrates that effective therapeutic vaccination induces formation of polyfunctional leukemia-specific CD4⁺ T cells. Future studies are needed to delineate the importance of these cytokines and granzyme B expression. However, it is intriguing that although IL-10 is more typically associated with immunosuppression, previous literature supports a role for T cell–derived IL-10 in antitumor immunotherapy (25, 53, 54). Additionally, IL-17 and granzyme B have both been implicated in T cell responses to cancer (26, 28, 55). Thus, we envision two possibilities for how and when BAp:I-A^b–specific T cells might elicit antileukemia immunity after therapeutic vaccination. First, because the most effective therapeutic vaccination regimen we used (heterologous vaccination plus checkpoint blockade) yielded the greatest fraction of granzyme B–producing BAp:I-A^b–specific T cells, it is possible that these cells directly kill MHC-II⁺ BCR-ABL⁺ leukemic cells. Second, it is possible these polyfunctional BAp:I-A^b–specific T cells induce BAp:I-A^b–specific memory T cells, which may be required for long-term leukemia control. In support of this idea, Th17 cells responding to tumors in other models have a long lifespan, which may be associated with memory formation (28, 56). Therefore, our observations support the idea that polyfunctional BAp:I-A^b–specific T cells are induced by intensive therapeutic vaccination, and that these cells contribute to effective leukemia control.

The current paradigm suggests that neoantigen-specific T cells respond better to tumors because the repertoires of these cells have not been pruned by thymic central tolerance (36). This idea implies that cross-reactive T cells will respond poorly to tumors because the repertoires of these cells have been limited by thymic central tolerance. We have previously shown that BAp:I-A^b–specific T cells are cross-reactive with self-Ag and that the BAp:I-A^b–specific T cell repertoire is limited by thymic central tolerance (9). Nonetheless, we observed in this study that BAp-specific adaptive immunity is crucial for antileukemia immunity following prophylactic vaccination. Thus, our observations provide a counterpoint to the idea that neoantigen-specific T cells are a prerequisite for effective endogenous anticancer T cell responses (36). Taken broadly, our observations suggest that fusion proteins created by chromosomal translocations may be viable immunotherapy targets even when the fusions do not create neoantigens. This is particularly relevant because chromosomal translocations often result in “driver” mutations, thus leaving minimal opportunity for cancer immunoediting to occur.

Checkpoint blockade is thought to work best in tumors with high numbers of nonsynonymous mutations (35). Our results support this concept, as checkpoint blockade was only minimally effective in B-ALL, a leukemia that generally has lower numbers of nonsynonymous mutations (5). However, we also demonstrate that intensive heterologous vaccination synergizes with checkpoint blockade to unmask a strong immune response that is capable of controlling this highly aggressive and uniformly fatal form of leukemia in mice. In conclusion, our work establishes that immunotherapy approaches can induce long-term survival with B-ALL, even though mice with B-ALL are refractory to checkpoint blockade–based immunotherapy (36).

We thank Gregory Hubbard, Alyssa Kne, Christopher Reis, Amy Mack, and Emilea Sykes for assistance with mouse husbandry, Justin Taylor for experimental design, Markus Muschen for BCR-ABL–IRES-GFP retrovirus, Christine Henzler for statistical advice, and Lynn Heltemes-Harris, Casey Katerndahl, and Dan Kaplan for commentary on the manuscript. David Masopust and Marc Jenkins contributed ideas and reagents, particularly in support of the vaccination regimens.

R.T.W. is currently an employee at Amgen, Inc. The work in this manuscript was completed prior to his accepting a position there. His previous employer, Puma Biotechnology, has no interests in leukemia, and thus we do not think there is an inherent conflict of interest. The interaction with R.T.W. dates back to the time when he was an Asst. Prof. at St. Jude Children's Research Hospital (Memphis, TN). The other authors have no financial conflicts of interest.