Personalized Neoantigen Vaccine and Pembrolizumab in Advanced Hepatocellular Carcinoma: A Phase 1/2 Trial
![ReachMD Healthcare Image](https://cdn.axismeded.com/cache/13/85/13855c3dbe620bbdbb67312f57ca6c76.jpg)
07/11/2024
Safety, feasibility and clinical responses
The baseline demographic and clinical characteristics of the study population are shown in Table 1. The first and last patients were enrolled on June 16, 2020, and June 14, 2023, respectively. The trial is ongoing. The median number of vaccinations at the data cutoff date was 5 (range 1–18), and the median duration of treatment was 6.1 months. At the data cutoff date (August 18, 2023), 25 patients had discontinued the study therapy (Fig. 1). The most common reason for discontinuation was disease progression (n = 22). All 36 patients had their personalized vaccine product available for dosing at the time they were eligible to receive second-line therapy (Fig. 2a).
The CONSORT (Consolidated Standards of Reporting Trials) diagram shows the flow of patients as of August 18, 2023. SAE, severe adverse event.
a, Manufacturing process for GNOS-PV02 and clinical trial design. In patients without disease progression, (1) treatment with pembrolizumab may continue every 3 weeks (Q3w) for 2 years per label recommendation; (2) treatment with GNOS-PV02 + pIL12 may continue Q3w for four doses, followed by Q9w until year 2 (Y2) and Q12w beyond 2 years. b, Pie chart with the percentage ORR, CR, PR, SD and PD according to RECIST 1.1 (n = 36, mITT). c, Waterfall plot showing the best overall response achieved by the 34 evaluable patients of the GT-30 trial at the time of data cutoff (August 18, 2023). aPR patient with a primary liver lesion and two lung metastases who achieved secondary resectability owing to tumor shrinkage and remained tumor-free for 18.2 months after the first treatment dose. d, Spider plot showing changes in the target lesion from baseline for the 34 evaluable patients of the GT-30 trial at the time of data cutoff (August 18, 2023). aThe same as in c.
Treatment-related adverse events (TRAEs) observed by the cutoff date are listed in Table 2. Overall, the treatment was safe and well tolerated, with an adverse event profile similar to that reported for pembrolizumab monotherapy in HCC, except for an increase in local vaccine injection-site reactions. Low-grade TRAEs were observed in 27 patients (75.0%), and there were no grade ≥3 TRAEs. Three patients (8.3%) experienced an immune-related adverse event (irAE) requiring systemic steroid treatment (grade 2 nephritis, grade 2 pneumonitis and grade 2 hepatitis). One patient (2.8%) discontinued pembrolizumab owing to an adverse event, but no patients discontinued PTCV therapy because of an adverse event.
At the time of data analysis, 34 of the 36 patients had undergone at least one on-treatment restaging scan and were evaluable for response according to Response Evaluation Criteria in Solid Tumors (RECIST) 1.1. Two patients discontinued therapy owing to unrelated severe adverse events (one after the first dose, one after the third dose of therapy) and were deemed unevaluable, but both patients were included in the modified intention-to-treat (mITT) analysis as nonresponders. By investigator review, the objective response rate (ORR) (confirmed + unconfirmed, mITT) per RECIST 1.1 was 30.6% (11 of 36 patients), with 8.3% (3 of 36) of patients achieving a complete response (CR) and 22.2% (8 of 36) of patients achieving a partial response (PR). The disease control rate was 55.6% (20 of 36 patients) (Fig. 2b–d). Of the 11 patients with an objective response, 9 patients (including the 3 patients with CR) had their response confirmed at the next regularly scheduled on-treatment imaging scan. Two patients with PR had their target lesions continue to show further tumor reduction over subsequent imaging time point(s), confirming a durable response (−44% and −59%), but were categorized as unconfirmed PR owing to the emergence of new nontarget lesions (one of these patients is described as a case study in ‘Vaccine-induced immune editing leads to tumor escape’ below).
At the data cutoff, the median follow-up was 21.5 months. Initial response assessment was performed at 9 weeks; among patients who had a response, the median time to the response was 9.3 weeks (range 8–46 weeks). One patient with initially unresectable HCC achieved secondary resectability after five PTCV doses. The median progression-free survival (mPFS) was 4.2 months, and the median overall survival (mOS) was 19.9 months. The median duration of response was not reached. Clinical response (CR/PR versus stable disease (SD)/progressive disease (PD)) was significantly associated with survival (PFS and OS) (Extended Data Fig. 1). Example responses are shown in Extended Data Fig. 2a–d. Clinical results were generally consistent across sex, etiologic disease subgroups and time on first-line TKI treatment at baseline (Extended Data Fig. 3).
Anti-PD-1 agents have been studied extensively in both the first-line and second-line advanced HCC settings. Across registrational clinical trials of pembrolizumab, nivolumab, durvalumab and tislelizumab, the ORR ranged from 12% to 18% (refs. 4,5,6,7,8,9,10,27). For the protocol development, sample size estimation and prespecified statistical hypothesis testing, we used the comparator pembrolizumab ORR of 16.9% based on KN-240 (ref. 28), which was consistent with the 17.0% ORR for pembrolizumab monotherapy observed in the KN-224 phase 2 study9. The observed ORR of 30.6% (11 of 36 patients) achieved statistical significance with a one-sided P value of 0.031 (one-sided 90% confidence interval (CI), 20.4–100%) versus the prespecified historical control.
We performed circulating tumor DNA (ctDNA) analyses using day 0 (baseline), week 3 and week 9 on-treatment samples from 13 patients. A molecular response was defined as a >50% reduction in the ctDNA level from baseline22,29. Changes in ctDNA levels broadly tracked with magnetic resonance imaging scans in monitoring objective responses (CR and PR). As shown in Extended Data Fig. 4a, the difference in the percentage change in ctDNA levels between patients with disease control (CR, PR or SD) and those with PD reached significance at week 9 (P = 0.006). The ctDNA analysis detected a stronger molecular response relative to imaging in two patients (Extended Data Fig. 4b,c). Patient 9 (determined to have SD by imaging) had a 95% reduction in ctDNA level, and patient 19 (determined to have PR by imaging) had a 100% reduction in ctDNA level. Both patients had durable responses lasting for more than 12 months. A ctDNA decrease at week 9 was significantly associated with longer survival (P = 0.01) (Extended Data Fig. 4d).
Biomarker analysis of the observed clinical responses
We next evaluated potential biomarkers of the observed clinical responses. All patients enrolled in GT-30 had a low TMB (fewer than five mutations per megabase), with a median TMB of 2.0 mutations per megabase. The TMB was similar in patients achieving CR/PR and those with SD/PD (Extended Data Fig. 5a). Similarly, while we observed a numerically higher ORR (35% versus 25%) in patients with baseline α-fetoprotein (AFP) levels of <400 ng ml−1 (n = 26) relative to patients with baseline AFP levels of >400 ng ml−1 (n = 8), there was no significant difference in disease control rate (58% versus 63%), mPFS (4.1 versus 6.2 months) or mOS (24.4 versus 15.6 months) (Table 1 and Supplementary Fig. 1). We next evaluated the relationship between baseline CD8 infiltration (assessed by the mRNA expression level of CD8A), pretreatment tumor CD274 (PD-1 ligand 1) and KDR mRNA expression levels, and the T cell-inflamed gene expression profile of 15 biomarkers and the response to treatment (Extended Data Fig. 5b). These biomarkers broadly characterize an immune-inflamed phenotype and have previously shown potential utility in distinguishing responders from nonresponders to anti-PD-1-based therapies in HCC30,31. However, we did not observe any relationship between these pretreatment markers and the response to the PTCV plus anti-PD-1 therapy.
In contrast to the lack of differentiation based on pretreatment tumor biomarkers, an exploratory post hoc analysis demonstrated a positive correlation (P = 0.025) between the number of neoantigens included in the PTCV and the clinical response achieved. Sequencing of the patients’ tumors identified a median of 30 vaccine-targetable neoantigens (range 4–67). Among patients receiving a vaccine encoding ≥30 neoantigens, 7 of 17 (41.2%) had an objective response. Conversely, among those receiving a vaccine encoding <30 neoantigens, only 4 of 17 patients (23.5%) had an objective response. There was a significant difference in the number of targeted neoantigens between the CR/PR group and the SD/PD group (Extended Data Fig. 5c).
We next evaluated pretreatment versus on-treatment (week 9) tumor biopsy samples to determine whether changes in the expression levels of biomarkers associated with T cell activation and infiltration, consistent with a PTCV-mediated effect, could explain the observed clinical response in responders (CR/PR, n = 9) and nonresponders (SD/PD, n = 13). The expression of the T cell biomarkers CD8A, CD8B, CCL5, CXCR6, LCK and TIGIT was significantly increased in responders but not in nonresponders (Extended Data Fig. 6a,b). Although these analyses were exploratory, the results are broadly consistent with the proposed mechanism of action for PTCV-induced clinical responses. In contrast to anti-PD-1 monotherapy, which reinvigorates preexisting antitumor immunity, therapeutic cancer vaccines can prime new antitumor immune responses, providing a potential rationale for why inflamed and noninflamed tumors responded similarly to the therapy. Furthermore, the relationship between the number of neoantigens included in the PTCV and the clinical response achieved suggests that features of tumor antigenicity or of the vaccine itself drive clinical benefit with PTCV plus anti-PD-1 therapy.
Vaccination elicits neoantigen-specific responses
In our study, predicted neoepitopes were selected for inclusion in the PTCVs using a pipeline for called variants based on an in silico analysis of the exome and transcriptome sequencing data from each patient (Methods). While the tumor RNA-sequencing (RNAseq) data confirmed that the nonsynonymous somatic variants included in the PTCVs were expressed, we did not experimentally confirm the processing and presentation of the predicted epitopes in the tumor.
Twenty-two patients with available peripheral blood mononuclear cell (PBMC) samples were evaluated for the presence of vaccine-induced neoantigen-specific responses before and after treatment using the ex vivo interferon-γ (IFNγ) enzyme-linked immunosorbent spot (ELISpot) assay. The criteria for ELISpot positivity are described in Methods. In almost all patients, treatment was associated with an increase in the magnitude of cumulative PTCV neoantigen-specific T cell responses (P < 0.0001) (Fig. 3a).
a, Vaccine-induced responses assessed by IFNγ ELISpot assays without cytokine stimulation (n = 22). Cumulative magnitudes were collected from positive epitopes before and after treatment. The postvaccination response is the ‘best’ (highest magnitude) response for each patient across time points. SFU, spot-forming units. b, Total neoantigens (gray bars) and positive neoantigens before (black bars) and after (red bars) vaccination in each patient’s PTCV assessed by IFNγ ELISpot. c, Percentage of positive responding epitopes by groups. The definition of a neoantigen-specific ELISpot response can be found in Methods. d, Representative density plots (patient 22) of the T cell markers CD69, Ki67, CD107a, IFNγ and TNF upon stimulation with patient-specific PTCV epitope pools. e,f, Polyfunctionality assessed by Boolean gating of CD4+ or CD8+ cytokine-producing populations. T cell activation (CD69 and CD107a; e) and proliferation (Ki67; f) were assessed together with the double-positive expression of GZMA and perforin 1 (PRF1) to evaluate the cytolytic potential of neoantigen-reactive T cells. Four patients (patients 7, 11, 18 and 22) were analyzed in d–f. g, T cell clones expanded in the periphery and the new or expanded clones enriched in the matched tumor sample for each patient (n = 14). Total PBMC and tumor-associated T cell expansion were calculated by comparing posttreatment to pretreatment PBMC or tumor samples (differential abundance statistical analysis). h, Cumulative frequencies of peripherally expanded TCR rearrangements in tumor biopsy samples. i, Expanded clone numbers in tumor biopsy samples. j,k, TCR clonality (j) and repertoire richness (k) in tumor biopsy samples (n = 14). PD (red), SD (gray), and CR/PR (blue). Error bars correspond to the upper s.e.m. of each group. Simpson clonality reports the distribution of TCR rearrangements in a sample, in which 0 indicates an even distribution of frequencies and 1 indicates an asymmetric distribution. TCR repertoire richness reports the mean number of unique rearrangements. Lower numbers indicate focused TCR diversity. Filled symbols in c, e and f and open circles in h and i represent individual patients; the box extends from the 25th to the 75th percentile; the line inside the box is the median; and the whiskers extend from the minimum to the maximum value. Significance between groups was evaluated by a two-tailed Mann–Whitney test (c); significance within groups was evaluated by a two-tailed Wilcoxon rank test (a, h–k).
PTCV treatment was also associated with an increase in the number of encoded neoantigens eliciting an immune response. In 19 of 22 patients (86.4%), the number of vaccine-encoded neoantigens with T cell reactivity was higher after than before treatment (Fig. 3b). Two patients with PD, treated with PTCVs encoding 4 and 11 neoepitopes, did not yield any detectable ELISpot responses either before or after treatment; one patient with SD (20 neoantigens) had a reduced number of reactive epitopes detected after treatment relative to their pretreatment baseline. Individual epitope analyses across the cohort revealed PTCV encoded neoantigen-specific T cell responses to a median of 64.0% (range 0–100.0%) epitopes after treatment compared to 11.8% (range 0–85.3%) epitopes before treatment. PTCV immunization resulted in a significant increase in positive epitopes in both clinically responding and nonresponding patients (Fig. 3c).
A positive correlation was observed between the total number of neoantigens included in the PTCV and the number of positive neoantigen responses detected by ELISpot assays (P = 0.0007, Spearman correlation coefficient) (Extended Data Fig. 7a). We evaluated the magnitude of IFNγ response by quartiles and observed that patients in the top quartile had a trend toward longer OS compared to patients in the bottom quartile (mOS 30.2 versus 15.7 months) (Extended Data Fig. 7b). Patients with CR or PR showed a trend toward a greater magnitude of IFNγ response (Extended Data Fig. 7c). Immune responses were observed against neoepitopes with predicted high binding affinity (kd <500 nM), as well as against those with predicted medium or low binding affinity (kd 500–2,000 nM), to human leukocyte antigen (HLA) class I molecules (Extended Data Fig. 7d).
Neoantigen-specific responses were confirmed in a subset of four responding patients (one with CR, three with PR) through intracellular staining of PBMCs stimulated with patient-specific neoepitope pools in vitro. Upon neoantigen stimulation, both CD4+ and CD8+ populations presented an increased activation profile as determined by the individual expression of the CD69, Ki67, CD107a, IFNγ and tumor necrosis factor (TNF) markers (Fig. 3d). Boolean gating confirmed an increasing trend of active (CD69+CD107a+) (Fig. 3e) and proliferative (Ki67+) (Fig. 3f) polyfunctional CD4+ and CD8+ T cells with cytolytic capabilities (GZMA+PRF1+) after stimulation. Taken together, these data indicate that vaccination is capable of eliciting polyfunctional neoantigen-specific CD4+ and CD8+ responses with cytolytic potential. We next characterized T cell clonal expansion, trafficking, neoantigen specificity, and clonal and subclonal genetic profiles as exploratory endpoints.
Vaccination enriches T cell clone expansion and infiltration
Complementarity-determining region 3 (CDR3) regions of the T cell receptor β-chain (TCRβ) were sequenced from paired pretreatment and posttreatment (weeks 9–12) PBMC and tumor biopsy samples in 14 patients with available paired tumor biopsy samples. Although anti-PD-1 therapy is not known to modulate the diversity of tumor-reactive T cell clones32,33, we hypothesized that the addition of the PTCV to anti-PD-1 therapy would lead to both an increase in abundance and a broadening of the circulating HCC-reactive T cell clonal repertoire, which would subsequently traffic to the tumor microenvironment. Consistent with this hypothesis, we observed significant T cell clonal expansion in 14 of 14 (100%) patients in both the peripheral blood and tumor using a differential abundance statistical framework (Fig. 3g). The median number of new or expanded T cell clones in the periphery was 47 (range 24–132), of which a median of 21 (range 6–71) T cell clones were also new or expanded in the posttreatment tumor. The median increase in the cumulative frequencies of the significantly expanded clones was 1.94% (range 0.35–8.70%) (Supplementary Fig. 2a,b). We identified an increase in both the abundance and number of expanded T cell clones within the tumor after treatment, which was also identified in the peripheral blood after treatment (Fig. 3h,i). Importantly, we observed higher frequencies and numbers of T cell clones newly present in the tumor after vaccination (Supplementary Fig. 3a,b). Additionally, we found significantly increased TCR clonality (P = 0.035) (Fig. 3j) but no significant change in TCR repertoire richness in the tumor (P = 0.216) (Fig. 3k). These data suggest that therapy with PTCV results in the expansion of T cells in the periphery, with T cells trafficking to the tumor.
Vaccination drives effector T cell memory clonal expansion
To characterize the vaccine-induced T cell response further, we performed single-cell RNAseq/TCR sequencing (TCRseq) of peripheral blood at the 12-week postvaccination time point. In four samples obtained from three patients (patients 6 (SD), 7 (CR) and 8 (PR)), we assigned PBMCs to 14 clusters based on transfer learning from reference PBMC data and analysis of canonical marker genes (Extended Data Fig. 8a–d)34,35,36. Clusters demonstrating the highest expression of genes associated with cytotoxicity, including chemokine C–C motif ligand 5 (CCL5) and granzyme K (GZMK), GZMB or granulysin (GNLY), were CD4+ effector memory T (TEM), CD4+ cytotoxic T lymphocyte (CTL), CD8+ TEM and CD8+ proliferating cells, as well as γδ T and natural killer (NK) cells. In 31,843 of 39,439 cells, a TCR sequence was identified by paired single-cell immune repertoire sequencing. Across all single-cell clusters with an associated TCR, clonally expanded T cell populations, which we defined as more than five cells that shared the same TCR, were most strongly associated with the CD8+ proliferating (odds ratio, 5.84; 95% CI, 4.07 to 8.43; P < 0.001), CD8+ TEM (odds ratio, 5.58; 95% CI, 5.15 to 6.06; P < 0.001) and CD4+ CTL (odds ratio, 4.14; 95% CI, 3.13 to 5.48; P < 0.001) clusters by Fisher’s exact test (Extended Data Fig. 8e).
For each patient, we identified all T cells in the single-cell dataset with a TCRβ that had been identified as clonally expanded in bulk TCRseq of prevaccination and postvaccination peripheral blood (Extended Data Fig. 8f). Of 92 TCRβ sequences found to be clonally expanded by bulk sequencing, 64 sequences were identified in 1,041 cells within the single-cell dataset, of which 84.4% (879 of 1,041) were from patient 8, 10.8% (112 of 1,041) were from patient 7 and 4.8% (50 of 1,041) were from patient 6. The single-cell cluster most strongly associated with vaccine expansion by Fisher’s exact test was CD8+ TEM (odds ratio, 107.40; 95% CI, 88.75 to 130.91; P < 0.001), which comprised 87.3% (909 of 1,041) of vaccine-expanded TCRβ (Extended Data Fig. 8g and Supplementary Fig. 4a). On a per-patient basis, CD8+ TEM was also the most highly represented cluster, being found in 88% (775 of 879), 88% (99 of 112) and 70% (35 of 50) of vaccine-expanded TCRβ for patients 8, 7 and 6, respectively.
We then subdivided the CD8+ TEM cluster into subclusters and identified seven subclusters, among which three (CD8+ TEM_3, CD8+ TEM_5 and CD8+ TEM_6) displayed high expression of multiple genes associated with cytotoxicity, including GZMB and NKG7, and three (CD8+ TEM_1, CD8+ TEM_2 and CD8+ TEM_4) had increased expression of GZMK (a preexhaustion marker)37 (Extended Data Fig. 9). Cytotoxic subclusters accounted for 85% (776 of 901) of vaccine-expanded CD8+ TEM, with CD8+ TEM_5 and CD8+ TEM_3, which comprised 47.2% (429 of 909) and 37.1% (337 of 909) of CD8+ TEM, respectively, being the two largest clusters represented. In contrast, the GZMK-expressing preexhausted subclusters CD8+ TEM_2, CD8+ TEM_1 and CD8+ TEM_4 were less numerous (Supplementary Fig. 4b).
Expanded TCR clones are reactive to PTCV-encoded antigens
Lastly, we sought to validate the neoantigen-specific activity of tumor-infiltrating T cells in two representative patients. The first patient had 42 significantly expanded clones in the periphery, of which 27 were found in the tumor sample after treatment (Figs. 3g and 4a). Three of the most frequent TCR sequences (Fig. 4b) from T cell clones newly present in the tumor after vaccination were found primarily in the CD8+ TEM cluster in postvaccination peripheral blood single-cell sequencing (Fig. 4c). These three TCR sequences were selected and cloned into the pMXs-IRES (internal ribosome entry site)-GFP (green fluorescent protein) retroviral plasmid vector for further studies (Fig. 4d). To characterize the neoantigen-specific cellular response driven by treatment with GNOS-PV02, we stimulated TCR-engineered T cells from patient-derived PBMCs with the patient’s PTCV-specific neoantigen pools. We found T cell activation (CD69+) associated with pool 1 (consisting of peptides corresponding to neoantigens 1–20), which included the most reactive epitopes measured by ELISpot; pool 2 (consisting of peptides corresponding to neoantigens 21–40) served as an internal control for specificity and showed similar levels as the nonspecific epitope (CTA1) control (Fig. 4e). In the second representative patient, we were able to map the new T cells/TCRs to a specific vaccine-encoded epitope. From IFNγ ELISpot analysis, we first identified a strongly immunogenic epitope (ATP1A1-ALB) encoded in the patient’s personalized vaccine (Supplementary Fig. 5a). Patient-derived PBMCs were subjected to in vitro stimulation for T cell enrichment and expansion and then stimulated with ATP1A1-ALB peptides. We found both CD4+ and CD8+ T cells with specific polyfunctional responses (CD69+, Ki67+, CD137+, IFNγ+, IL-2+) against ATP1A1-ALB (Supplementary Fig. 5b). High-frequency TCRs were identified by TCRseq/RNAseq (33 clones expanded in the periphery, of which 15 were found in the tumor) and engineered (Supplementary Fig. 5c,d). Engineered TCRs were stimulated with a pool of epitopes containing all the neoantigens in the patient’s PTCV. Similar to the first patient, we observed CD4+ and CD8+ T cell specificity against the patient-specific neoantigens (Supplementary Fig. 5e) relative to the unstimulated or nonspecific peptide (CTA1)-stimulated controls. These data validate the postvaccination infiltration and increase in the frequency of T cells in the tumor with specificity to vaccine-encoded neoantigens.
a, Most frequent TCRs identified by TCRseq and RNAseq in a patient (before vaccination versus week 9 after vaccination, pairwise scatterplots). Different superscript letters show selected high-frequency new T cell clones detected in PBMCs after vaccination and their abundance in the tumor. Orange, green, and gray circles represent expanded, contracted and not significantly changed T cell clones, respectively. b, CDR3 sequences of the three TCRs (from patient 8; TCR 1, TCR 2 and TCR 3) selected for cloning and their frequency (freq.) in the tumor before (pre-Vax) and after (post-Vax) vaccination. Selected cloned TCRs were present in high frequency only in the peripheral blood and tracked into the tumor after treatment. c, UMAP (Uniform Manifold Approximation and Projection) and stacked barplot indicating the single-cell cluster identities and number of cells for each of the three TCRs selected for cloning. d, Patient-specific clonal TCR sequences were gene optimized and inserted into the pMXs-IRES-GFP retroviral plasmid vector containing the viral packaging signal, transcriptional and processing elements, and the GFP reporter gene. MuLV, murine leukemia virus; Mo-MuLV, Moloney MuLV; LTR, long terminal repeat; AmpR, ampicillin resistance. e, TCR-engineered T cells (GFP+) from unvaccinated patient-derived PBMCs were stimulated for 6 h with epitope pools or the nonspecific epitope CTA1 (10 µg ml−1), and CD69 expression was evaluated by flow cytometry. Peptide pool 1 included the most reactive epitopes measured by ELISpot, whereas pool 2 (consisting of peptides corresponding to epitopes 21–40) served as an internal negative control.
Vaccine-induced immune editing leads to tumor escape
Tumor immune editing and escape are key mechanisms of cancer progression and metastatic dissemination. We used paired tumor biopsy samples to investigate the mechanisms of tumor escape in a patient with nondurable PR (patient 11) treated with a PTCV encoding 29 neoantigens plus pembrolizumab. At the first restaging interval at week 9, the patient had a target liver lesion reduction of −36%. By week 18, the response in the target liver lesion intensified, eventually reaching a −59% reduction, but a new lesion was observed in the adrenal gland (Extended Data Fig. 10a).
TCR/tumor-infiltrating lymphocyte (TIL) analysis of week 9 versus screening biopsy samples revealed the clonal expansion and infiltration frequencies of 25 T cell clones after vaccination (Extended Data Fig. 10b). IFNγ ELISpot analysis detected T cell responses to 9 of 29 epitopes and strong responses to 4 of 29 vaccine epitopes (Extended Data Fig. 10c), none of which were recognizable as driver mutations. Flow cytometry analysis showed a high frequency of antigen-specific, activated (CD69+, Ki67+) CD8+ and CD4+ T cells (Extended Data Fig. 10d). Sequencing of the new adrenal lesion identified 25 neoantigens, including 16 shared with the primary liver lesion (Extended Data Fig. 10e). However, all four vaccine neoantigens with the strongest ELISpot responses were absent in the adrenal lesion, consistent with neoantigen loss resulting from immune editing and clonal escape. The ctDNA analysis results were consistent with CR of the liver-specific tumor clones by week 21, persisting through week 57, but showed an increasing level of the adrenal-specific ctDNA over time (Extended Data Fig. 10f). Tumor tissue biomarker analysis of week 0 versus week 9 liver tumor samples showed robust CD8+ T cell infiltration, whereas the adrenal lesion had low CD8+ T cell density resembling that of the pretreatment liver tumor (Extended Data Fig. 10g). These data are consistent with tumor immune escape in the setting of tumor heterogeneity and the loss of passenger neoantigens targeted by the PTCV, which supports the proposed mechanism of action of PTCV-mediated antitumor immunity.