Despite promising advances in breast cancer immunotherapy, augmenting T-cell infiltration has remained a significant challenge. Although neither individual vaccines nor immune checkpoint blockade (ICB) have had broad success as monotherapies, we hypothesized that targeted vaccination against an oncogenic driver in combination with ICB could direct and enable antitumor immunity in advanced cancers.
Our models of HER2+ breast cancer exhibit molecular signatures that are reflective of advanced human HER2+ breast cancer, with a small numbers of neoepitopes and elevated immunosuppressive markers. Using these, we vaccinated against the oncogenic HER2Δ16 isoform, a nondriver tumor-associated gene (GFP), and specific neoepitopes. We further tested the effect of vaccination or anti–PD-1, alone and in combination.
We found that only vaccination targeting HER2Δ16, a driver of oncogenicity and HER2-therapeutic resistance, could elicit significant antitumor responses, while vaccines targeting a nondriver tumor-specific antigen or tumor neoepitopes did not. Vaccine-induced HER2-specific CD8+ T cells were essential for responses, which were more effective early in tumor development. Long-term tumor control of advanced cancers occurred only when HER2Δ16 vaccination was combined with αPD-1. Single-cell RNA sequencing of tumor-infiltrating T cells revealed that while vaccination expanded CD8 T cells, only the combination of vaccine with αPD-1 induced functional gene expression signatures in those CD8 T cells. Furthermore, we show that expanded clones are HER2-reactive, conclusively demonstrating the efficacy of this vaccination strategy in targeting HER2.
Combining oncogenic driver targeted vaccines with selective ICB offers a rational paradigm for precision immunotherapy, which we are clinically evaluating in a phase II trial (NCT03632941).
This study demonstrates two fundamental tenets of immunotherapy: vaccines targeting any tumor antigen will not be as effective as those targeting true oncogenic drivers and neither the stimulation of tumor-specific T cells nor the blockade of a key immune checkpoint is enough to overcome the layers of immune suppression by itself. We provide single-cell genetic evidence that vaccination alone generates a population of CD8 T cells incapable of long-term tumor control due to the activation of numerous immune dysfunction pathways within the tumor. These promising studies have led to the initiation of a phase II clinical trial testing a novel HER2 vaccine in combination with pembrolizumab (NCT03632941) to determine whether this combination can elicit effective antitumor immunity while minimizing off-target immune responses in patients with advanced HER2+ breast cancer.
Immune checkpoint blockade (ICB) via mAbs targeting the PD-1/PD-L1 interaction has demonstrated a clinical benefit in patients with metastatic disease across many malignancies (1). However, these therapeutic responses are limited to a subset of patients, with ICB responsiveness being associated with the presence of effector tumor-infiltrating lymphocytes (TILs) or a high number of somatic mutations encoding for clonal neoepitopes (2–4). This dichotomy is exemplified in breast cancer, which is composed of multiple molecular subtypes within a single tissue environment that each exhibit differential responses to ICB (5). PD-L1 ICB has demonstrated antitumor activity in some triple-negative breast cancers (TNBCs), a subtype with elevated levels of p53 mutations, high TILs, and increased neoepitope burden (6, 7). This is in contrast to the hormone receptor (HR+) and HER2+ molecular subtypes of breast cancer, which have distinct oncogenic drivers, generally contain fewer neoepitopes and TILs, and have shown lower response rates to ICB than TNBC (5, 8). As HR+ and HER2+ subtypes are highly dependent upon the expression of their respective oncogenes, these subtypes present an opportunity to therapeutically target essential, nonmutated driver genes (9). Rather than relying on the presence of endogenous T-cell responses, this type of vaccine strategy could be used to actively induce and direct oncogene-specific adaptive immunity.
Although HER2 is expressed in many cell types, it is highly amplified in HER2+ breast cancer and is not just a marker of tumorigenesis but is recognized as an oncogenic driver (10). However, multiple reports have identified a particularly oncogenic isoform of HER2 that lacks exon 16 (HER2Δ16), which can constitutively dimerize to confer sustained cellular signaling (11–14). Although the importance of HER2Δ16 in patient samples is not clear, amplification of HER2 may result in elevated expression of this isoform sufficient to transform cells (13). Clinical use of HER2 mAbs has demonstrated the importance of engaging HER2-specific innate immune responses for antitumor efficacy (9, 15), while multiple clinical vaccine trials (using both viral vectors and peptides) have suggested the importance of eliciting HER2-directed T-cell responses in antitumor immunity (16–18). We hypothesized that immunologically targeting the oncogenic HER2Δ16 isoform may be a particularly attractive option due to its importance in maintaining tumor signaling. In our past studies, we found that a HER2-specific viral vaccine can enhance progression-free survival in patients with recurrent HER2+ metastatic breast cancer (18). Here, we altered our vaccine to target the HER2Δ16 isoform to determine whether this strategy could better elicit adaptive responses in comparison to targeting nondriver, tumor-specific genes or unique tumor neoepitopes. We further wanted to determine whether this vaccine would synergize with PD-1 ICB to allow for enhanced antitumor immunity in a model of advanced HER2+ breast cancer.
In these studies, we establish evidence of HER2Δ16 isoform expression in clinical samples of breast cancer and confirm that this isoform functions as a potent oncogenic driver to transform breast epithelial cells. Using a HER2Δ16 transgenic mouse model, we determined that the spontaneous tumors that formed were largely nonmutated HER2+ breast cancer with immunosuppressive molecular signatures, reflective of advanced human HER2+ breast cancer (14). We demonstrate that vaccination targeting the extracellular domain of HER2Δ16 was essential to elicit effective antitumor immunity, but that this effect was diminished in established tumors. Likewise, we found that established endogenous tumors failed to respond to αPD-1 monotherapy, mirroring early clinical findings (5). However, the combination of HER2Δ16 vaccination with αPD-1 treatment led to long-term tumor-free survival and the activation of HER2-specific T cells in the tumor microenvironment (TME) of established tumors without the induction of an exhaustion signature that was present when the vaccine was given alone. This supports an approach targeting oncogenes with vaccines combined with systemic ICB to offer an effective, scalable means to achieve precision immunotherapy in advanced cancers. These results supported the combination of αPD-1 with our previously published viral HER2 vaccine to validate these findings in patients with advanced breast cancer (NCT03632941, currently enrolling).
Materials and Methods
Cell lines and signaling assays
Tumor cell lines MM3MG, NMUMG, EPH4, 293T, and Jurkat were obtained from and maintained as recommended by the ATCC. Cells were modified by stable lentivirus transduction and selection for expression of indicated genes. CRISPR-Cas9 lentiviruses were used to knockout mouse PD-L1, checked by sequencing and flow cytometry. In signaling assays, 293T cells stably expressing dox-inducible HER2 isoforms were transfected with dual luciferase reporter constructs (Cignal Reporter Assay Kit, 336841, Qiagen) and harvested at 24 to 48 hours for luciferase activity. Each condition was plated in quadruplicate and GFP control vectors were used as negative controls.
Human HER2-transgenic mice (C57BL/6 background; kindly provided by Dr. Wei-Zen Wei, Wayne State University, Detroit, MI; ref. 19) were crossed with BALB/c mice (000651, The Jackson Laboratory) and MM3MG cells were implanted into the mammary fat pads (1 × 105 cells) of 6- to 10-week-old F1 generation mice. Tumor measurements were made using calipers and volumes calculated using the formula [v = width × width × (length/2)]. Vaccinations were given by footpad injection of 2.6 × 1010 adenoviral particles/mouse. Peptides for peptide vaccines were mixed 1:1 with montanide and given weekly, twice subcutaneously (100 μg/peptide) and once in the footpad (20 μg/peptide). The HER2Δ16 transgenic mouse model was generated by crossing two different strains of mice, TetO-HER2Δ16-IRES-EGFP (a kind gift from W.J. Muller; ref. 14) and MMTV-rtTA (a kind gift from Dr. Lewis A. Chodosh; ref. 14) or MMTV-tTA (a kind gift from K.-U. Wagner; ref. 21) to generate TET-ON or TET-OFF expression of HER2Δ16 to drive spontaneous tumor formation. In TET-ON mice, mice older than 6 weeks were put on a doxycycline-containing diet (200 mg/kg, Bio-Serv). Individual animals were either vaccinated 1 week postinduction (early prevention model) or randomly enrolled into a specific treatment group as soon as palpable breast tumors were detected (∼200 mm3) in any of the eight mammary fat pads. Animals were terminated once any single tumor volume reached >2,000 mm3. Mice treated with αPD-1 were given 200 μg i.p. weekly. All mouse experiments were done in accordance with Duke Institutional Animal Care and Use Committee–approved protocols.
RNA sequencing alignment, quantification, and neoantigen prediction
Library preparation and sequencing was performed by Admera. Libraries were created using SureSelect XT Mouse All Exon Kit (Agilent) according to the manufacturer's instructions. RNA expression from HER2Δ16 tumors was compared with RNA sequencing (RNA-seq) data from 175 other murine mammary and tumor samples listed in Supplementary Table S1A. All murine RNA-seq data have been deposited in the NCBI GEO accession number GSE148482. Matched normal DNA extracted from tail clippings was respectively derived. Bioinformatics prediction of neoantigens was performed as described previously (22, 23). Predicted neoantigens were filtered on expression in all replicates with >5 × read support.
Library preparation for single-cell RNA-seq
Tumors from treated transgenic mice were harvested and processed into single-cell suspension using Mouse Tumor Dissociation Kit (Miltenyi Biotec) following the manufacturer's protocol with recommendations for 10X Genomics platform use. Red blood cells were lysed with ACK lysing buffer (Thermo Fisher Scientific) for 5 minutes, and stained with Fixable Far Red Dead Cell Stain (Thermo Fisher Scientific) and antibodies for CD45 (30F11), CD8β (YTS156.7.7), CD4 (RM4-5; all BioLegend). Live, single, CD45+, CD4, and/or CD8+ cells from tumor suspension were sorted by FACS. 10X libraries were created using Chromium Single Cell 5′ Library Construction Kit (v1.0) following the manufacturer's protocol. A targeted cell recovery of 3,000 cells was used for each tumor sample. Both gene expression and V(D)J enrichment libraries were created for each sample. Generated cDNA and final GEX/TCR libraries were quality checked using an Agilent Bioanalyzer 2100 and submitted to MedGenome Inc for sequencing on a NovaSeq S4 instrument.
Single-cell RNA-seq analysis
Fastq files from 10X library sequencing were processed using the CellRanger pipeline (10x Genomics, version 2.1.0) by MedGenome Inc. Filtered gene-barcode matrices that contained only barcodes with unique molecular identifier (UMI) counts that passed the threshold for cell detection were used for further analysis. We followed Seurat's (version 3.1.2; ref. 24) data integration procedure (see Supplementary Methods for complete details), which is designed to integrate diverse single-cell datasets. Seurat returned a corrected data matrix for all datasets, which was used for subsequent analyses.
Expanded clone TCR reactivity test
Aligned sequences from expanded clones were extracted using CellRanger (10× Genomics; Supplementary Table S1E) cloned into vectors containing GFP. These constructs were stably expressed in NFAT-luciferase reporter Jurkat cells with mouse CD8 alpha and beta chains. Expression was verified by GFP and TCRβ (H57-597; BioLegend) staining. Various tumor cell lines expressing HER2 or not were plated overnight before Jurkat cells were added for 18 hours. After incubation, cells were lysed and measured using a Veritas microplate luminometer (Turner Biosystems). All samples run in quadruplicate.
Statistical analysis and data visualization
Data are presented as mean ± SEM. Tumor volumes, flow cytometry, ELISA, and ELISPOT data from experiments with three or more treatment groups were analyzed by one-way ANOVA with Bonferroni multiple comparisons test. Comparisons were made to untreated or control group unless otherwise indicated. A two-tailed, unpaired Student t test was used for experiments with only two groups. Group sizes for animal tumor growth experiments were determined on the basis of preliminary datasets. All subjects in spontaneous tumor experiments were randomized into a treatment or control group. Kaplan–Meier methods were used to generate time to event plots, and groups were compared using the log-rank test. Graphs were generated and statistical analysis was performed using Prism (GraphPad) and R. Script for R code for single-cell RNA-seq (scRNA-seq) can be found at https://github.com/cramanuj/Her2VAX. P values of 0.05 or less were considered statistically significant. Not all significant differences are shown in every graph. *, P < 0.05; **, P < 0.01; ***, P < 0.001.
HER2Δ16 expression in breast cancer and its role in transformation and resistance to targeted therapies
Although HER2Δ16 has been reported to be expressed in many breast cancer cell lines, we wanted to determine its expression in patient samples to assess its potential as a molecular driver of breast cancer. There is currently no antibody that can distinguish these two isoforms at the protein level. As such, we utilized isoform-specific primers to measure expression of HER2 and HER2Δ16 in 154 different breast cancer specimens and 14 normal tissue specimens by qRT-PCR. Surprisingly, we found that HER2Δ16 expression was not only elevated in HER2+ tissue, but was also elevated in tumor tissue expressing estrogen receptor (ER) and/or progesterone receptor (PR; Fig. 1A). HER2Δ16 accounted for approximately 10% of total HER2 transcripts in all of these samples (Fig. 1B).
To assess the oncogenic impact of HER2Δ16 on canonical cellular signaling pathways, we stably transduced cells with doxycycline-inducible expression of GFP, HER2WT, HER2Δ16, as well as a truncated form of HER2Δ16 that expresses only the extracellular portion (HER2Δ16-EC) and a HER2Δ16 with a point mutation in the ATP-binding site that renders the kinase domain inactive (HER2Δ16-KI; Supplementary Fig. S1A). Using these cells, we tested the impact of these genes on 43 different canonical signaling pathways using a luciferase tagged transcription factor reporter system. GFP and HER2WT did not induce any of the pathways, but HER2Δ16 stimulated high transcriptional activity of five different signaling pathways (MAPK/ERK, MAPK/JNK, PKC/Ca++, EGR, and Hedgehog; Fig. 1C). MAPK, PKC, and EGR pathways are all canonically expressed downstream of HER-family dimers (25–27), supporting the hypothesis that the HER2Δ16 isoform is functioning as a constitutively active homodimer (28). The Hedgehog pathway has not previously been associated with HER2 signaling but has been shown to play a role in other subtypes of breast cancer, particularly in the epithelial–mesenchymal transition (EMT) and may represent a novel downstream effect of HER2Δ16 expression (29, 30). Importantly, truncation or mutation of HER2Δ16 completely abrogates signaling through these pathways (Fig. 1C).
We next investigated whether HER2Δ16 expression transformed the nonmalignant murine mammary cell lines MM3MG and NMUMG. After identical stable transduction and selection, we found that HER2Δ16 had consistently lower expression than HER2WT on the surface of these cells despite identical promoters (Supplementary Fig. S1B). HER2Δ16 expression conferred significantly enhanced anchorage-independent growth in vitro, even with this reduced expression (Fig. 1D; Supplementary Fig. S1C). Furthermore, expression of HER2Δ16, but not HER2WT, allowed for robust tumor formation when these cells were implanted in mice (Fig. 1E). These results demonstrate that HER2Δ16 can effectively transform murine mammary cells, despite lower levels of expression.
Consistent with other studies, we observed that antibody-dependent cytotoxicity (ADCC) by the conventional HER2-targeting therapy trastuzumab and cell-mediated lysis by trastuzumab-DM1 (T-DM1) were reduced in HER2Δ16-expressing cells, compared with HER2-WT (Fig. 1F and G). Furthermore, HER2Δ16-expressing tumor cells were completely resistant to TDM1 treatment in vivo (Fig. 1H). This suggests that clinical HER2Δ16 expression could enhance tumor growth signaling and result in decreased efficacy of HER2-targeting therapies. Thus, our data support that HER2Δ16 expression may represent an important mechanism of resistance to these standard-of-care (SOC) therapies and critical target for immunotherapy in refractory HER2+ breast cancers (28, 31).
Endogenous model of HER2Δ16-driven breast cancer as a model of HER2+, immunosuppressive tumors with low tumor mutational burden
We have previously described a mouse model that utilizes a doxycycline-inducible promoter to drive expression of HER2Δ16 and EGFP. When bred to MMTV-tTA (21) or MMTV-rtTA mice (20), these inducible models express HER2Δ16 in basal and luminal mammary epithelial cells, inducing multifocal, HER2Δ16+ tumors in approximately 30% of MMTV-tTA mice and approximately 85% of MMTV-rtTA (Fig. 2A). Withdrawal of doxycycline results in loss of HER2Δ16 expression and complete tumor regression (Supplementary Fig. S1D), demonstrating the critical nature of HER2Δ16 in tumorigenesis. To evaluate the genetic identity of these tumors, RNA and DNA of four distinct tumors from different mammary glands in four different mice (total of 16 tumors) was sequenced and analyzed. These analyses revealed an elevated HER2-enriched subtype signature score (32) compared with other mouse models (Supplementary Table S1A), consistent with that found in human HER2+ breast cancer and Erbb2/neu mouse tumors, generated by expression of the rat form of HER2 that develop spontaneous mutations in the juxtamembrane region (Fig. 2B). In addition, there was a significantly enhanced immunosuppression score (22) and myeloid-derived suppressor cell (33) score, highlighting the immunosuppressive TME present in these mice compared with other mouse models.
As these tumors are all driven by expression of the same oncogene, we next evaluated the similarity between the spontaneous tumors formed within the same mouse compared with tumors formed in different mice. This analysis revealed that distinct tumors from the same mouse were more genetically similar to other tumors from that same mouse than to tumors that formed in a different mouse (Fig. 2C). As neoepitope burden has been shown to correlate with immunogenicity and responsiveness to immune therapies, we employed next-generation DNA-seq and RNA-seq to determine the tumor mutational burden (TMB) and neoepitope load in these spontaneous tumors (22). We found that the TMB averaged only 0.57 mutations/MB (Fig. 2D). This corresponded to a very small number of predicted neoepitopes, with only one neoepitope with strong predicted MHCI binding potential and six additional with moderate binding potential present between all 16 sequenced tumors (Supplementary Table S1B). Collectively, these studies indicate that HER2Δ16 is a critical oncogenic pathway and spontaneous tumors driven by HER2Δ16 are reflective of clinically advanced immunosuppressive HER2+ breast cancer.
Generation of antitumor immunity with kinase-inactivated HER2Δ16 viral vaccine
To direct immunity against HER2Δ16 without the risk of inducing an oncogenic signaling pathway, we generated an adenoviral vaccine encoding an inactivating mutation (K753A) in the kinase domain of HER2Δ16 (34). We found that a single vaccination of HER2 transgenic (HER2-Tg) mice with this vector elicited significant HER2-specific T-cell (Fig. 3A) and antibody responses (Fig. 3B). Moreover, these antibody responses were highly polyclonal, consisting of multiple isotypes, capable of stimulating ADCC and complement-dependent cell cytotoxicity (Fig. 3C–E). This HER2-specific immunogenicity was not limited to these HER2-Tg mice, but was also seen in all other strains of mice tested (Supplementary Fig. S2).
To determine whether these HER2Δ16-specific immune responses could inhibit tumor growth, we orthotopically implanted MM3MG cells expressing HER2Δ16 (Fig. 1E) and vaccinated 3 days after implantation. Ad-HER2Δ16-KI vaccinated mice had a significant antitumor response compared with an irrelevant Ad-GFP control vaccine (Fig. 3F). Notably, the suppression of tumor growth correlated with an induction of both HER2-specific T-cell and antibody responses (Fig. 3G and H). These studies validated that our vaccine could elicit significant HER2-specific immunity against HER2Δ16 expressing tumors and slow their growth. However, this implantable tumor model grew too rapidly for long-term studies.
Efficacy of targeting HER2Δ16 versus neoepitope peptides or nononcogenic cancer-specific antigen
To perform long-term immunologic studies, we returned to our inducible HER2Δ16 model. The presence of EGFP and development of specific neoepitopes also allowed us to evaluate whether immunologic targeting of HER2Δ16 using our Ad-HER2Δ16 vaccine would be sufficient to elicit effective antitumor immunity relative to a neoepitope peptide vaccine (Supplementary Table S1B) or an identical viral vaccine targeting a coexpressed nondriver tumor antigen (GFP). To further identify which domain of HER2Δ16 provided the best tumor suppression, we engineered additional HER2Δ16-specific vaccines targeting the intracellular (ICD) and extracellular domains (ECDs).
We first evaluated the ability of these vaccine regimens to suppress tumor development when given early, before the establishment of palpable tumors. In this setting, we found that both the Ad-HER2Δ16-KI and the Ad-HER2Δ16-ECD vaccines were highly effective at suppressing tumor formation (Fig. 4A). However, neither Ad-HER2Δ16-ICD nor Ad-GFP were effective at delaying the appearance of tumors, with tumor formation identical to untreated and irrelevant vaccine (Ad-LacZ) treated mice (Fig. 4A). As Ad-GFP elicited potent anti-GFP responses (Supplementary Fig. S3A–S3C), but none of the GFP-expressing tumors were delayed in growth, we conclude that targeting a nononcogenic tumor antigen is not an effective therapeutic strategy. Similarly, vaccines targeting only the ICD of HER2 did not have a protective effect (Fig. 4A). As an additional comparison of the impact of targeting a coexpressed nondriver tumor antigen like GFP in an implantable tumor model, we engineered our MM3MG-HER2Δ16–expressing cells to also express GFP and vaccinated mice after implantation of these cells. Again, the vaccine targeting HER2Δ16 significantly delayed tumor growth compared with one targeting GFP in this implantable tumor model (Supplementary Fig. S4A).
Vaccines targeting neoantigens have been shown to be effective in other tumor models (35). Neoepitopes identified when sequencing tumors were targeted with a peptide vaccine and compared with peptides from GFP and HER2 (Supplementary Table S1B). Vaccination against both neoepitope and GFP peptides once again had no effect, while HER2 peptides were able to delay tumor formation (Supplementary Fig. S4B).
To evaluate the role of each arm of the adaptive immune system in the protection elicited by vaccination, we depleted CD8 T cells or B cells and tracked the formation of tumors following vaccination with Ad-HER2Δ16-KI. Delay of tumor formation by vaccination was completely lost when CD8 T cells were depleted, while B-cell depletion had no impact on the time or proportion of mice that developed tumors (Fig. 4D). Although CD8 depletion was complete and sustained, it should be noted that despite initial B-cell depletion, the populations of B cells rebounded to normal levels even with prolonged weekly treatment of depleting antibodies (Supplementary Fig. S4C–S4E). Collectively, these data highlight the critical role that CD8 T cells play in the protection elicited by our vaccine.
To better model the impact of immunotherapy in patients with advanced HER2+ breast cancer, we evaluated the efficacy of vaccination in mice with an established tumor. Mice were placed on doxycycline and once tumors reached approximately 100 mm3, were randomized into different treatment arms. We found that while Ad-HER2Δ16-KI significantly increased survival of mice and generated robust anti-HER2 T-cell immunity, in this setting, all treated mice still ultimately succumbed to tumors (Fig. 4E and F). Consistent with our early vaccine studies, we found that vaccination targeting HER2Δ16 significantly enhanced responses compared to targeting the tumor antigen GFP (Fig. 4E). To determine whether earlier oncogene induction would elicit enhanced immune tolerance and alter the efficacy of vaccination, we repeated this experiment in MMTV-tTA mice, where expression is induced immediately after weaning by discontinuing doxycycline treatment. However, despite this early expression, we again observed the same pattern of efficacy (Supplementary Fig. S4F and S4G).
In summary, we found that vaccination is able protect 75% of mice from tumor formation when given prior to overt tumor formation and significantly slows tumor growth after measurable tumors have formed. However, vaccination alone was never able to achieve a complete cure. Thus, although these studies reveal that immunologic targeting of HER2Δ16 is effective at combating tumor formation and progression, they also highlight that the induction of tumor-specific T-cell responses alone is not enough to eradicate advanced breast cancer.
Role of the PD-1–PD-L1 axis in mediating local HER2Δ16 breast cancer immunosuppression
One possible explanation for the inability of our HER2-targeting vaccine to eradicate tumors is the regulatory effect of PD-1-PD-L1 signaling in the TME. To test this and establish the role of PD-1-PD-L1 immunosuppression in our HER2+ breast cancer models, we evaluated rtTA-HER2Δ16 tumors and found that both HER2+ tumor cells (Fig. 5A) and myeloid populations from the tumors including monocytes and neutrophils (Fig. 5B) expressed high levels of PD-L1. To determine the potential of this pathway in mediating antitumor immunosuppression, we first investigated the role of tumor cell PD-L1 expression. Using an implantable tumor model, we again observed elevated tumor-specific expression of PD-L1 (Fig. 5C). After ablation of tumor PD-L1 using CRISPR lentiviruses (Fig. 5C), we observed that PD-L1-KO cells grew much more poorly compared with controls (Fig. 5D). This growth difference, however, did not result in an enhancement of systemic HER2-specfic T-cell or B-cell responses (Fig. 5E and F), suggesting that this axis altered local T-cell responses. Consistent with these results, treatment of MM3MG-HER2Δ16 tumors with αPD-1 antibodies significantly slowed tumor growth and continued treatment resulted in tumor regression in most mice (Fig. 5G), but did not significantly enhance systemic HER2-specific T-cell responses or B-cell responses (Fig. 5H and I). Collectively, these results suggest that the PD-1-PD-L1 axis plays a critical immunosuppressive role in HER2Δ16-driven cancers and that its blockade is sufficient to cause significant tumor regression in an implantable tumor model of HER2+ cancer where significant anti-HER2 immune responses are present.
Synergistic impact of HER2Δ16 immunologic targeting enabled by PD-1 immune checkpoint blockade
Although αPD-1 demonstrated efficacy in an orthotopic model, we wanted to assess the impact of αPD-1 alone and in combination with HER2Δ16 vaccination in a more immunosuppressive context, as occurs in advanced HER2+ breast cancer. Using HER2Δ16 transgenic mice with established HER2+ tumors, we administered αPD-1 alone or in combination with Ad-HER2Δ16-KI. In contrast to our orthotopic HER2Δ16+ breast cancer model, we found that endogenous HER2+ breast cancers did not respond to αPD-1 as a single agent (Fig. 6A). Likewise, treatment with αPD-1 was insufficient to increase systemic anti-HER2 T-cell responses (Fig. 6B). However, the combination of αPD-1 and Ad-HER2Δ16-KI resulted in a significant enhancement in survival, with approximately 30% of mice exhibiting complete tumor regression and long-term (>150 days) tumor-free survival (Fig. 6A). Interestingly, the addition of αPD-1 treatment did not further enhance the systemic HER2-specific T-cell response above what was seen with vaccine alone (Fig. 6B). This suggested that the difference in antitumor efficacy of the combination therapy was again occurring within the TME.
To determine the effect of vaccination and ICB on the activity and clonal expansion of T cells in the TME, we utilized scRNA-seq to profile TILs from treated mice. Given the high rate of complete tumor regression and in an effort to minimize differences in tumor size, samples were collected within 3 weeks of enrolling in a given treatment arm. Tumors were digested and enriched for T cells prior to scRNA-seq for total gene expression and TCR genes. Overlaying the TCR libraries confirms our cluster identification of T cells (Fig. 6C; Supplementary Fig. S5). Furthermore, by identifying clones that have expanded (>10 with a shared TCR sequence), we see that nearly all expanded clones are contained within a single cluster (activated CD8 T cells). We focused our analysis on the gene expression of cells with a corresponding TCR sequence and utilized previously published gene signatures (Supplementary Table S1C) to examine the immunologic properties of cells from each treatment groups. Predictably, TILs from mice that had received the viral Ad-HER2Δ16 vaccine, regardless of the addition of αPD-1, showed significant expression of proinflammatory and type I IFN pathways (Fig. 6D). Strikingly, in mice that only received the vaccine, this robust immune activation was also coupled with a significant and profound increase in immune dysfunction pathways including anergy and coinhibitory molecule expression (Fig. 6D). Given this dichotomy, we generated an activation and exhaustion signature based on genes that were differentially expressed with known activation (TNFα, IFNγ) and exhaustion (Havcr2, Lag3 Tigit) markers (Fig. 6E; Supplementary Table S1D). These scores once again confirm greatly enhanced activation in both vaccine groups, with a correspondingly high exhaustion score in vaccine alone that is diminished when αPD-1 is added to the vaccine. Finally, a gene signature of IFNγ-associated genes has been published as a predictor of responsiveness to αPD-1 therapy in patients (36). Consistent with this report, our data confirm a high IFNγ gene signature score in vaccinated mice.
Interestingly, we saw evidence of expanded clones in unvaccinated mice, which led us to question the specificity of the TILs in these samples. With the paired TCR alpha and beta chain sequence data from our scRNA-seq (Supplementary Table S1E), we cloned the TCRs from the top expanded clone from mice that had received a vaccine or not into NFAT-luciferase reporter Jurkat cells. We then screened these TCRs using a tumor cell line created from an rtTA-HER2Δ16 spontaneous tumor that had retained the dox-dependent expression of HER2Δ16. Strikingly, we were able to demonstrate that the expanded clone from an Ad-HER2Δ16 vaccinated mouse was highly reactive to HER2 (Fig. 6F). We confirmed this reactivity in several other HER2Δ16-expressing cell lines (Supplementary Fig. S6). Collectively, these results demonstrate the ability of HER2Δ16 vaccination to not only induce systemic adaptive immune responses, but expand HER2-specific CD8 T cells that infiltrate into tumors. Thus, our model of established, immunosuppressive tumors (similar to those seen clinically) demonstrates that the addition of αPD-1 can effectively enable vaccinated induced HER2-specific T cells in the TME.
HER2+ breast cancer comprises approximately 15% to 20% of all breast cancer cases and is typified by amplification of HER2 expression (37). Here, we have demonstrated that the oncogenic HER2Δ16 comprises approximately 10% of HER2 transcripts but confers exceptionally enhanced oncogenic signaling to tumor cells and resistance to standard-of-care mAb therapies (Fig. 1). In patients with metastatic HER2+ breast cancer, acquired resistance to these is inevitable, and nearly all patients will die of disease progression even though tumors continue to express HER2 (38). Recent clinical trials utilizing αPD-1 in patients with breast cancer have revealed only modest response rates, with the highest responses being in TNBC. On the basis of these findings, we hypothesized that specific immunologic targeting of the HER2Δ16 isoform may offer an effective therapeutic strategy for metastatic and/or treatment-resistant HER2+ breast cancer, which could be rationally combined with PD-1/PD-L1 mAbs.
To explore this hypothesis, we generated a novel viral vaccine vector to elicit HER2-specific adaptive immune responses, without oncogenic signaling (critical for vaccine safety). Using an implantable model, we have shown that both our novel vaccine and αPD-1 ICB were highly effective, which was in contrast to the endogenous HER2Δ16+ breast cancer model, where neither single agent achieved complete tumor regression. Although we utilized HER2-Tg animals and orthotopic implantation into the mammary fat pad, implantation of tumor cells alone appeared able to break tolerance and stimulate an anti-HER2 response, possibly caused by cell death and release of damage-associated molecular patterns (DAMPs) during implantation, foreign residual FBS proteins from tissue culture, or from the rapid growth of these tumors. Conversely, we did not observe striking anti-HER2Δ16 immune responses in transgenic animals bearing endogenous tumors, indicating a level of tolerance that is more consistent with the muted response rates seen in patients (39). This aspect of the study supports the use of transgenic models for immunotherapy research and suggests that studies relying upon orthotopic implantation of tumor cells may exaggerate immune agent responsiveness.
Our study also directly tested the efficacy of a vaccine targeting an oncogene versus a nononcogenic tumor antigen (GFP). Importantly, we found that vaccination against HER2Δ16 or the ECD of this gene, was far more effective at protecting from tumor formation and eliciting significantly prolonged survival of tumor bearing mice (Fig. 4). This is in contrast to vaccination against a gene that was still specific to the tumor but did not play any role in transformation (GFP), which had no protective or therapeutic effect. These studies highlight the potential difference of directing immune responses against antigens that are credentialed tumor drivers, versus tumor-specific antigens or private neoepitopes that do not play a significant role in oncogenesis to increase the success and translatability of immunotherapies, particularly in cancers with a low TMB and few neoepitopes to target.
We also leveraged our inducible transgenic model to explore the impact of tumor progression and timing on the effectiveness of our vaccine strategy. Although the size and progression of tumors are generally thought to alter responsiveness to antitumor immunity (40), few studies have directly tested whether outcomes differ in an endogenous tumor setting after identical immunologic interventions. Our studies illustrate that vaccinations during early tumor development prior to any overt tumor formation are highly effective, but are drastically suppressed once a larger immunosuppressive TME has formed. This is significant for the clinical use of these therapies where early-phase clinical trials almost exclusively focus on late-stage metastatic patients, who have progressed on all conventional therapies. These data provide evidence that failure of immune therapies in this advanced patient population does not predict how effective that therapy may be in an earlier stage of the disease. Our studies also reveal that the PD-1–PD-L1 axis plays a critical role in this immunosuppressive phenotype in HER2Δ16-driven breast cancer. Notably, the use of αPD-1 in established disease was not effective at suppressing tumor growth (∼0% >150 day survival rate), despite the presence of TILs and expression of PD-1 and PD-L1 in various cell types within the TME, consistent with what has been reported clinically (Figs. 4A and 6A). This observation is somewhat at odds with the premise that tumors with high expression of PD-L1 should be more responsive to αPD-1/PD-L1 therapies. In bladder cancer, PD-L1 expression by nontumor cells in the TME is more associated with clinical response to PD-L1 inhibition, which highlights the need to better define biomarkers for these ICB therapies within specific cancer types (41, 42).
Cancer immunotherapy is in an era where monotherapies are often promising but subtherapeutic for most patients and the challenge remains to devise rational combinations that will maximize the benefit and curative potential. Much work has been done to test various tumor antigen targeting vaccine strategies and checkpoint inhibitors, with many showing evidence of CD8 T-cell activation (43–45). Consistent with these studies we show that our HER2Δ16 vaccine in combination with αPD-1 had a synergistic curative effect (Fig. 6A). Mechanistically, scRNA-seq analysis revealed that TILs from mice receiving vaccine alone exhibited a broad program of immune dysfunction including upregulation of anergy, coinhibitory molecule expression, and exhaustion pathways. It is important to note that these pathways were upregulated early, as the scRNA-seq samples for vaccine alone were harvested only 15 days postvaccination. Coinhibitory molecule expression is one of the markers of an exhausted CD8 T-cell population that distinguishes it from a function memory or effector CD8 T cells (46). Although exhaustion has been the main focus of ICB therapy in cancer, tolerance of T cells toward specific antigens, or T-cell anergy, may develop due to TCR stimulation without sufficient costimulatory signals or in the presence of inhibitory stimulation. T-cell anergy may develop in our model and in cancer more generally because coinhibitory signals prevail over costimulatory signals in the complex milieu of the TME (47). The induction of an exhaustion signature is consistent with a previous report investigating CD8 TILs in patients with carcinoma after PD-1 blockade (48). Overall, our data highlight and provide single cell evidence for why vaccines for cancer may not be potent enough to combat advanced disease as single agents.
The ability to sequence the paired TCRα/β chains from single cells provides an unprecedented opportunity to understand the specificity of cells that are generated by vaccination and infiltrating into the TME. We leveraged this technology by cloning of the expanded TCRα/β chains and subsequent in vitro assays to demonstrate that our vaccine not only expanded T cells, but the expanded clones are specific for HER2 epitopes (Fig. 6F). Other groups have utilized similar techniques to evaluate expansion of vaccine and tumor-specific T cells in PBMCs (49, 50), but here we show the expansion of vaccine-induced antigen-specific T cells taken directly from the tumor. Although this observation is exactly what we would predict based on the established vaccine paradigm and observations of PBMC responses reported by others, it is an important proof-of-concept experiment that elegantly demonstrates this foundational premise in a endogenous model of advanced cancer. It further provides a basis for investigation of the specificity of expanded T cells in tumors or infectious diseases, which may provide critical information about the accuracy of immunodominance predictions and the level of bystander “nonspecific” infiltration that occurs.
The success of PD-1/PD-L1 immune checkpoint blockade has renewed interest in clinical immunotherapy, spurred efforts to identify what defines a responsive patient population, and prompted the testing of combinatorial strategies to improve patient response rates. Most of these efforts have centered on pairing αPD-1/PD-L1 mAbs with standard-of-care therapies (radiation, chemotherapy, etc.) and different immune checkpoint blockade antibodies (i.e., CTLA4, TIM3, LAG3, etc.). In our study, we demonstrate that a vaccine targeting a nonmutated oncogene isoform (HER2Δ16) is effective in mobilizing antitumor T-cell responses that can then be effectively enabled by an αPD-1 mAb to elicit complete tumor regression and long-term tumor-free survival. As such, we believe that this approach offers a rational paradigm for precision immunity through the use of oncogene targeted vaccines in combination with selective ICB mAbs. Although our study has only tested the impact of this combination in HER2+ breast cancer, future preclinical studies may extend this framework to other cancers where HER2 is highly expressed (such as gastric cancer) or more critically, to other oncogene targets (such as EGFR, c-myc, etc.). Clinically, these promising studies have led to the initiation of a phase II clinical trial testing a similar novel HER2 vaccine (targeting the HER2 ECD in an alphaviral particle) in combination with pembrolizumab (NCT03632941) to determine whether this approach can elicit effective antitumor immunity while minimizing off-target immune responses in patients with advanced HER2+ breast cancer.
Disclosure of Potential Conflicts of Interest
L.A. Chodosh reports grants from Breast Cancer Research Foundation and NIH during the conduct of the study, as well as personal fees from Imerys Talc America (expert witness in cancer litigation), Sterigenics, Inc (consultant in cancer litigation), U.S. Department of Justice (expert witness in cancer litigation), Arnold & Porter, LLP (consultant in cancer litigation), Takeda, Inc (expert witness in cancer litigation), and Janssen Pharmaceuticals (consultant in cancer litigation) outside the submitted work. G. Broadwater reports grants from NIH during the conduct of the study. T. Hyslop reports grants from NCI during the conduct of the study, as well as personal fees from AbbVie outside the submitted work. C.M. Perou reports personal fees from Bioclassifier LLC (equity stock holder and consultant of BioClassifier LLC) outside the submitted work, as well as ownership of a patent for U.S. Patent No. 12,995,459 issued, licensed, and with royalties paid from Bioclassifier LLC. B.G. Vincent reports personal fees from GeneCentric Therapeutics outside the submitted work. J. Force reports Consulting from Exact Sciences, Pfizer, Myriad, and NanoString Biotechnologies. M.A. Morse reports grants from Merck, BMS, AstraZeneca, personal fees from Taiho, Exelixis, Bayer, Novartis, and Roche/Genentech, and grants and personal fees from Eisai and Ipsen outside the submitted work. H.K Lyerly reports AlphaVax (equity) outside the submitted work, as well as membership of the Board of Directors of OncoSec. Z.C. Hartman and H.K. Lyerly report a patent WO2017120576A1 on cancer vaccines and methods of delivery, owned by Duke University and optioned to license to a Duke spin-out (Replicate Bioscience, Inc.). No potential conflicts of interest were disclosed by the other authors.
E.J. Crosby: Conceptualization, data curation, formal analysis, funding acquisition, visualization, writing-original draft. C.R. Acharya: Formal analysis, visualization, writing-review and editing. A.-F. Haddad: Methodology, writing-review and editing. C.A. Rabiola: Investigation, writing-review and editing. G. Lei: Resources, investigation, writing-review and editing. J.-P. Wei: Investigation, writing-review and editing. X.-Y. Yang: Investigation, writing-review and editing. T. Wang: Investigation, writing-review and editing. C.-X. Liu: Investigation, writing-review and editing. K.U. Wagner: Resources, writing-review and editing. W.J. Muller: Resources, funding acquisition, writing-review and editing. L.A. Chodosh: Resources, writing-review and editing. G. Broadwater: Conceptualization, formal analysis. T. Hyslop: Conceptualization, formal analysis. J.H. Shepherd: Resources, data curation, formal analysis, writing-review and editing. D.P. Hollern: Resources, data curation, formal analysis. X. He: Resources, data curation, formal analysis. C.M. Perou: Resources, data curation, writing-review and editing. S. Chai: Resources, data curation, formal analysis. B.K. Ashby: Resources, data curation, formal analysis. B.G. Vincent: Resources, funding acquisition, writing-review and editing. J.C. Snyder: Resources, validation, writing-review and editing. J. Force: Writing-review and editing. M.A. Morse: Writing-review and editing. H.K. Lyerly: Supervision, funding acquisition, project administration, writing-review and editing. Z.C. Hartman: Conceptualization, supervision, funding acquisition, writing-original draft, project administration, writing-review and editing.
This work was supported by the Susan G. Komen Foundation (CCR14299200 to Z.C. Hartman; PDF17481657 to E.J. Crosby; CCR17483467 to B.G. Vincent), Merck Sharp-Dohme Investigator Initiated Grant (LKR160831 to Z.C. Hartman), Department of Defense (BC113107 and BC132150 to H.K. Lyerly), NCI (5K12CA100639-09 to H.K. Lyerly and Z.C. Hartman, 1R01CA238217-01A1 to Z.C. Hartman, 5K12-CA100639-10 to H.K. Lyerly and J.C. Snyder, K22CA212058 to J.C. Snyder), Cancer Research Society (No. 21068 to W.J. Muller), and Canadian Institutes of Health Research (FND-148373 to W.J. Muller).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.