Research ArticleCancer

Personalized cancer vaccine effectively mobilizes antitumor T cell immunity in ovarian cancer

See allHide authors and affiliations

Science Translational Medicine  11 Apr 2018:
Vol. 10, Issue 436, eaao5931
DOI: 10.1126/scitranslmed.aao5931

The personalized touch in cancer vaccination

Transfer of autologous dendritic cells (DCs) has been investigated as a method of boosting T cell responses in therapeutic vaccines for several diseases. Tanyi et al. report the findings of a clinical study involving recurrent ovarian cancer patients. Patient DCs were pulsed with oxidized tumor lysate before transfer and given alone or in combination with immunomodulatory drugs. The DC vaccine was well tolerated and induced potent antitumor T cell responses, including to new epitopes, that correlated with better prognosis. These results suggest further testing of this vaccination regimen for inducing protective T cell immunity in cancer.


We conducted a pilot clinical trial testing a personalized vaccine generated by autologous dendritic cells (DCs) pulsed with oxidized autologous whole-tumor cell lysate (OCDC), which was injected intranodally in platinum-treated, immunotherapy-naïve, recurrent ovarian cancer patients. OCDC was administered alone (cohort 1, n = 5), in combination with bevacizumab (cohort 2, n = 10), or bevacizumab plus low-dose intravenous cyclophosphamide (cohort 3, n = 10) until disease progression or vaccine exhaustion. A total of 392 vaccine doses were administered without serious adverse events. Vaccination induced T cell responses to autologous tumor antigen, which were associated with significantly prolonged survival. Vaccination also amplified T cell responses against mutated neoepitopes derived from nonsynonymous somatic tumor mutations, and this included priming of T cells against previously unrecognized neoepitopes, as well as novel T cell clones of markedly higher avidity against previously recognized neoepitopes. We conclude that the use of oxidized whole-tumor lysate DC vaccine is safe and effective in eliciting a broad antitumor immunity, including private neoantigens, and warrants further clinical testing.


The current standard treatment of advanced epithelial ovarian cancer (EOC) is debulking surgery combined with paclitaxel and carboplatin chemotherapy. Despite a good initial response, typically, most of the patients relapse and ultimately develop resistance to platinum-based chemotherapy, with no curative therapeutic options available (13). Substantial evidence indicates that ovarian cancers express a multitude of known tumor-associated antigens (47), and a proportion of tumors are infiltrated by intraepithelial tumor–infiltrating lymphocytes, which correlate with improved survival (816). Immunoreactive gene signatures have also been described in ovarian cancer, and they too correlate with longer survival (17). These studies have suggested that ovarian cancer patients could benefit from immunotherapy. However, early results with antibodies inhibiting the immune checkpoint programmed cell death 1 (PD-1) or its ligand (PD-L1) were quite modest, with an average response rate of 10 to 15% and a disease control observed in less than half of the patients (18). In addition to immune checkpoints, important barriers to immunotherapy in ovarian cancer may include the inability of tumor-specific T cells to home to tumors due to tumor-soluble factors (19). For example, vascular endothelial growth factor A (VEGF-A) is implicated in the establishment of a vascular endothelial barrier preventing T cell homing in ovarian and other tumor types, and its blockade induces de novo T cell infiltration in mouse and human tumors (2023). Other important barriers include peripheral tolerance mediated by T regulatory (Treg) cells (24) and a variety of soluble or cell surface immunosuppressive ligands that attenuate the function of effector T cells in tumors (25). Combinations of various immunomodulatory treatments will be required to properly mobilize immunity and fully capitalize on the natural immunogenicity of ovarian cancer.

One other form of immunotherapy that can induce a T cell response is vaccination. Vaccines should expand the pool of available tumor-specific T cells, and they could thus provide important partners for combination immunotherapy. However, the expected potential of cancer vaccines has not been realized in the clinical setting. In part, this could be related to the choice of antigens: Most molecularly defined tumor vaccines, to date, have used a single “self” antigen. The use of multiple tumor-restricted antigens, such as neoepitopes resulting from tumor mutations, represents a promising approach to tumor vaccination (2628). A potential alternative is the use of whole-tumor extracts (29). For example, in an ovarian cancer mouse model, we showed that the combination of a whole-tumor antigen vaccine markedly increased the efficacy of checkpoint blockade immunotherapy (30). Autologous whole-tumor vaccines could, in theory, elicit responses to numerous relevant tumor antigens, including shared antigens and private mutated neoantigens. Multiple approaches have been undertaken to enhance the immunogenicity of whole-tumor antigen vaccines through different lysate preparations (3136). We previously reported on a vaccine developed using autologous dendritic cells (DCs) pulsed with autologous tumor lysate produced from tumor dissociated into single cells, which were then killed through oxidation with hypochlorous acid (HOCl) and lysed using freeze-thaw cycles. This approach to lysate preparation proved superior to ultraviolet B irradiation or freeze-thaw lysis in terms of priming T cell responses against tumor antigens in vitro and showed promise in a preclinical model of ovarian cancer (37). We also showed that production of oxidized lysate-pulsed DCs and intranodal injection of this vaccine was feasible in five patients with ovarian cancer (37).

Here, we advance the clinical development of this intranodal DC-based vaccine approach, combining it with the intravenous VEGF-A blocking antibody bevacizumab and intravenous low-dose cyclophosphamide, two multifaceted and potentially synergistic immunomodulatory interventions, which are commonly used in advanced recurrent ovarian cancer patients (3840). Bevacizumab can induce de novo T cell infiltration in human tumors and can be combined safely with PD-L1 blockade (41), and low-dose cyclophosphamide has been shown to be able to attenuate systemic and tumor-infiltrating Treg cells (4245). We met all the endpoints of the study and demonstrated that this combination was feasible, safe, and well tolerated, and it elicited antitumor immunity, which was associated with improved survival. We also report that this immunization approach based on whole-tumor antigen mobilized a newly detected T cell response against private tumor neoepitopes, which included novel T cell clones with distinct T cell receptors (TCRs) and high functional avidity.


OCDC vaccine and combination treatment is feasible and safe

We conducted a pilot study of tumor vaccination using autologous oxidized whole-tumor cell lysate pulsed on autologous 5-day-old DCs (henceforth referred to as OCDC). Subjects were 25 platinum-pretreated, immunotherapy-naïve patients with recurrent advanced EOC with the specific goal of testing the safety, feasibility, tolerability, and immune effects of the ultimate combination of OCDC with bevacizumab and low-dose cyclophosphamide. The study followed a modified safety phase 1 design, with stepwise escalation of the combination, and included a vaccination-only run-in cohort with just OCDC administered every 2 weeks (cohort 1, n = 5), followed by a cohort where the combination of OCDC plus intravenous bevacizumab was administered (cohort 2, n = 10) and a final cohort with OCDC plus bevacizumab and low-dose intravenous cyclophosphamide (cohort 3, n = 10; fig. S1). Cyclophosphamide (200 mg/m2) was administered about 24 hours before each vaccine dose, and bevacizumab (10 mg/kg) was administered on the day of vaccination. Most patients had received several prior lines of chemotherapy. The clinicopathologic characteristics of patients and treatment details are provided in Table 1. The overall scheme of vaccine development is summarized in Fig. 1A. Peripheral blood monocytes were collected by elutriation (fraction 5) from a 10- to 15-liter leukapheresis and were cultured in cell factories in the presence of human granulocyte-macrophage colony-stimulating factor (GM-CSF) and human interleukin-4 (IL-4) for 4 days to induce differentiation into DCs. DCs were then pulsed overnight with autologous oxidized tumor lysate, followed by treatment with lipopolysaccharide (LPS) plus interferon-γ (IFN-γ) for 6 to 8 hours. OCDCs were harvested and cryopreserved on day 5 until clinical use. All patients completed induction vaccination with five OCDC doses, followed by computerized axial tomography scan computed tomography (CT) evaluation. Patients who did not progress by Response Evaluation Criteria in Solid Tumors (RECIST) and had OCDC doses remaining continued with maintenance vaccination (plus bevacizumab with or without cyclophosphamide, depending on the cohort) until disease progression or exhaustion of vaccine, whichever occurred first. All patients were followed for up for 2 years.

Table 1 Clinical data.

Patient identification number, age, tumor histologic type (ovarian, epithelial ovarian cancer; peritoneal, primary peritoneal serous carcinoma), FIGO stage, number of prior chemotherapy lines (chemos), number of prior recurrences, protocol cohort, combination therapy with vaccine (bev, bevacizumab; cy/bev, cyclophosphamide and bevacizumab), total number of vaccine doses received, clinical response, remission inversion status after vaccination, vaccine, tumor, and new antigen immune response are reported. Vx, vaccine; PD, progressive disease; SD, stable disease; PR, partial response; Y, yes; N, no; NR, nonresponder; R, responder.

View this table:
Fig. 1 Production, feasibility, and safety of vaccine.

(A) Scheme of dendritic cell (DC) vaccine production and administration. (B) Top: DC yield on day 6 is presented as percentage of seeded monocytes. Each dot represents a different patient. Bottom: DC yield correlates with the monocyte counts in the apheresis product. Correlation analysis was performed using the Spearman’s nonparametric correlation test. (C) Percentage of output DCs expressing CD86 or HLA-DR in the vaccine product. Each dot represents a different patient. The line shows the median. DCs from all patients met the release criteria of >60% positivity for both markers. (D) Number of vaccine doses produced per patient. The line shows the median. Each dot represents a different patient. (E) Example of vaccine injection under ultrasound guidance. (F) Number of adverse events in the different treatment cohorts.

Manufacturing of the vaccine product was feasible in all patients. We performed a total of 26 events of leukapheresis for 25 patients, with 3.6 × 109 ± 2.7 × 109 (means ± SD) monocytes obtained per patient. We seeded a mean of 6.9 × 108 monocytes and obtained a mean of 1.8 × 108 DCs at the end of the procedure, representing a yield of 26.9% of cultured monocytes; the DC yield correlated with the monocyte counts in the apheresis product (Fig. 1B). Output cells were >70% CD86+ and human leukocyte antigen (HLA)–DR+ (Fig. 1C), and additional phenotype analysis confirmed their mature DC phenotype (fig. S2). Release criteria were met in all patients, and OCDC vaccine products were negative for mycoplasma, bacteria, or fungi; contained endotoxin (<5 EU/ml); and were free of residual HOCl. A total of 456 doses of OCDC vaccine were generated, with an average of ~16 doses per patient (range, 5 to 45; Fig. 1D). OCDC vaccines were administered through a direct intranodal injection under ultrasound guidance into one to two inguinal lymph nodes bilaterally (Fig. 1E). A total of 174 intranodal injections were performed without complications. All vaccine injections were tolerated well, and most vaccine-related toxicities were transient and grade 1 (Fig. 1F and table S1). There was no toxicity grade >2 attributed to OCDC. The combination was also tolerated well; patients in cohort 2 received an average of six doses of vaccine plus bevacizumab (range, 3 to 16), and patients in cohort 3 received an average of seven doses of vaccine plus bevacizumab and cyclophosphamide (range, 5 to 15). There was no grade toxicity (grade >2) attributed to the treatment as a whole.

OCDC effectively mobilizes a vaccine-reactive immune response

We tested peripheral blood mononuclear cells (PBMCs) collected prevaccination and after five doses of induction vaccination (on-vaccination) for the presence of T cells recognizing tumor antigen, that is, reactive to DCs pulsed with autologous tumor lysate or to primary autologous tumor cells, when available. Vaccine-reactive T cells were readily detectable in on-vaccination PBMCs but were at a significantly lower frequency or undetectable in prevaccination PBMCs (P = 0.001; Fig. 2A). As determined in five representative patients, T cells detected on-vaccination comprised both vaccine-reactive CD8+ and CD4+ T cells, which were detected at a significantly lower frequency or were undetectable prevaccination (Fig. 2B). Most of the CD4+ and CD8+ vaccine-reactive T cells collected on-vaccination were polyfunctional, that is, they secreted more than one cytokine in response to DCs pulsed with tumor antigen, although CD4+ cells were more skewed toward tumor necrosis factor–α (TNFα) and IL-2, and CD8+ toward IFN-γ and TNFα (Fig. 2C). We documented in patient CTE-0001 the assessment of T cell responses after stimulation with tumor cells or DC-presented tumor antigen (Fig. 2D, top). Progressive increase in the frequency of peripheral blood tumor–specific T cells was also observed with continued vaccination over 24 months (Fig. 2D, bottom), which was associated with remission inversion, that is, longer disease-free interval (DFI) on vaccine maintenance (96 months) relative to the DFI after each of the prior chemotherapies (14 and 7 months, respectively) (Table 1). We also documented direct and specific killing of autologous tumor cells in vitro and suppression of autologous tumor growth in vivo in a patient-derived xenograft (PDX) mouse model by on-vaccination T cells (Fig. 2, E and F), which demonstrated a significant increase in tumor-specific CD3+ cells on-vaccination. Clinical benefit was observed in this patient who had recurred twice after chemotherapy before enrolling in cohort 1 (vaccine-only) of this study. After discontinuation of vaccination, she remained in remission for a total of five additional years on no further therapy. A significant increase in tumor-specific CD8+ and CD4+ cells on-vaccination was also observed in three patients who experienced disease stabilization or partial responses in tumor lesions at 12 weeks (for example, CTE-0017 in Fig. 2G and Table 1). In this patient, we also observed IFN-γ–producing T cells responding directly to DC-presented tumor antigen after vaccination composed of CD8+ and CD4+ T cells (fig. S3).

Fig. 2 Response to immunotherapy.

(A) Increased recognition of tumor antigens on-vaccination (Vx) treatment in all responder patients (n = 11). The frequency of interferon-γ (IFN-γ)–secreting T cells in response to autologous DCs pulsed with autologous oxidized tumor lysate measured by enzyme-linked immunospot (ELISpot) is shown. These analyses were performed with cryopreserved cells without any prior in vitro expansion. (B) Blood CD8+ and CD4+ T cell response to autologous DCs pulsed with autologous tumor lysate analyzed pre- and on-treatment shown in five patients with available cryopreserved peripheral blood mononuclear cell (PBMCs). The gating strategy is shown in fig. S7. (C) Intracellular cytokine staining showing the functional profiles of CD8+ and CD4+ T cells producing intracellular IL-2, TNFα, and/or IFN-γ in response to autologous DCs pulsed with autologous oxidized tumor lysate. These analyses (B and C) were performed with cryopreserved cells without any prior in vitro expansion. (D) Top: Representative examples of T cell responses against vaccine and autologous tumor cells detected pre- and on-vaccination. The number of spots per 105 PBMCs, as measured by IFN-γ ELISpot assay, is shown. Bottom: Fold increase in frequency of peripheral blood tumor-specific T cells over 24 months. (E) 51Cr release assay of posttherapy CD3/CD28-expanded PBMCs with autologous tumor cells. [R, responder patient (CTE-001) whose T cells produced IFN-γ in response to autologous tumor cells; NR, nonresponder (CTE-0012; *P < 0.003)]. (F) On-treatment CD3/CD28-expanded T cells suppress the growth of autologous tumor in NSG mice more potently than pretreatment T cells (*P = 0.03). (G) Example of two implants; right pelvic and presacral (arrows) regressing in on-treatment CT (bottom) relative to pretreatment CT (top). (H and I) PFS is significantly longer in patients whose on-treatment PBMCs recognized autologous tumor cells (H) or ex vivo autologous tumor lysate–pulsed DCs (I) relative to patients showing no such responses. Differences between averages of variables were compared using two-tailed t test. Percentages between groups were compared using χ2 test.

Together, we detected increases in IFN-γ–producing T cells responding to DC-presented tumor antigen on week 12 of vaccination in 11 of 22 evaluable patients from whom we could use autologous tumor lysate–pulsed DCs. In addition, we confirmed an increased number of IFN-γ–producing T cells responding directly to tumor cells in 9 of 13 evaluable patients from whom we were able to generate short-term autologous tumor cell cultures (Table 1). We found no significant difference in the number of baseline peripheral blood monocyte or lymphocyte counts between responder and nonresponder patients, and there was no difference either in the number of DC doses produced or in the output DC phenotype between immune responder and nonresponder patients (table S2). Tumor assessment by RECIST revealed that of the 25 patients treated, 2 patients experienced partial response, whereas 13 patients had documented stable disease, which persisted for a median of 14 months from enrolment (range, 4 to 96 months). Patients, whose on-vaccination T cells recognized autologous tumor cells or DC-presented tumor antigen, experienced significantly longer progression-free survival (PFS) on immunotherapy compared to patients whose T cells did not respond to either (Fig. 2, H and I); 8 of 11 vaccine responders and 8 of 9 tumor responders achieved remission inversion, whereas none of the tumor and vaccine nonresponders experienced remission inversion (Table 1). The 2-year overall survival (OS) rates of the responder patients was 100%, whereas the 2-year OS of nonresponders was 25%.

We further investigated whether the better PFS observed in vaccine responders and tumor responders, as opposed to nonresponders, is related to the vaccine effect or simply reflect general immunocompetence of these patients. To address this issue, we stimulated T cells from each patient collected before or after immunotherapy with polyclonal stimulation using staphylococcal enterotoxin B (SEB). We observed a very similar T cell response to SEB between vaccine responders and nonresponders both at baseline and after immunotherapy end of study (EOS) (fig. S4). To further address this point, we looked at gene expression data from pretreatment tumor in 16 of 25 patients and asked whether patients whose tumors exhibited gene signatures indicative of T cell infiltration before therapy had longer PFS or OS 24 months after enrolment. We found that positive immune signature before vaccination did not correlate with the clinical response nor with immune response observed after vaccination nor with clinical outcomes such as PFS or OS. Although these are correlative data, they suggest that perhaps therapy changed the natural course of the disease and that vaccine-induced immune response was more impactful than preexisting immunity (fig. S5 and table S3).

Cyclophosphamide augments OCDC vaccination effect

The main goal of our study was to develop a safe and feasible combination that would produce an effective immunization. We included low-dose intravenous cyclophosphamide before each vaccine dose with the expectation that this would result in enhanced vaccination effects. When comparing responses in patients who received bevacizumab without or with cyclophosphamide, we detected a significantly higher number of patients responding to tumor antigen (DCs pulsed with autologous tumor lysate) in the cyclophosphamide cohort (8 of 10 responders) compared to patients who did not receive cyclophosphamide (3 of 12; P = 0.002; Fig. 3A). Furthermore, although we detected no significant differences in eight other serum cytokines or analytes between the two cohorts, we observed a significant transient increase in serum IFN-γ levels about 24 hours after vaccination in the cyclophosphamide cohort, which was not detected in cohort 2 (Fig. 3B). This was preceded specifically in the cyclophosphamide cohort by a significant transient decline in serum levels of transforming growth factor–β, an immunosuppressive cytokine highly expressed by Treg cells in ovarian cancer (46) observed about 6 hours after vaccination and 30 hours after cyclophosphamide administration (Fig. 3C).

Fig. 3 Increased immune response and survival in patients treated with cyclophosphamide.

(A) Frequency of T cells responding to autologous tumor lysate–pulsed DCs by IFN-γ ELISpot on-vaccination relative to prevaccination analyzed in patients treated with (n = 12) or with no (n = 10) cyclophosphamide. These analyses were performed with cryopreserved cells without any prior in vitro expansion. (B) Serum IFN-γ concentration is significantly increased relative to baseline 24 hours after the first vaccination in cyclophosphamide-treated patients (Cy; n = 6) but not in patients who did not receive cyclophosphamide (No Cy; n = 6). (C) Serum transforming growth factor–β (TGF-β) concentration is significantly decreased 6 hours after the first vaccination (and about 30 hours after the first dose of intravenous cyclophosphamide) relative to pretreatment baseline in cyclophosphamide-treated patients but not in patients who did not receive cyclophosphamide (n = 8 in each group). (D) OS of patients treated with OCDC vaccine plus bevacizumab (Bev) without (cohort 2) or with cyclophosphamide (cohort 3). (E) OS of patients treated with OCDC vaccine plus bevacizumab and cyclophosphamide (cohort 3) relative to matched historic patients (n = 56) from the same institution treated with bevacizumab and cyclophosphamide. Median survival times were computed using Kaplan-Meier methods. Differences in OS or PFS were assessed using the log-rank test.

Consistent with more effective immunization, patients in the cyclophosphamide cohort exhibited significantly higher OS rate at 12 and 24 months compared to patients who received no cyclophosphamide before each vaccination (Fig. 3D). The survival observed in the cyclophosphamide cohort was considered higher than previously reported in this population with bevacizumab-based biological combinations (3840). We compared these results to a control historic group of 56 patients from the same institution matched carefully for disease characteristics and who had undergone secondary debulking and bevacizumab/cyclophosphamide during their course of therapy, but no vaccine. We found that the OS at 2 years was significantly higher (78%) in study cohort 3 patients who received vaccine plus bevacizumab/cyclophosphamide than in the historic group of matched patients who received bevacizumab/cyclophosphamide but no vaccine (44%) (log-rank P = 0.046; Fig. 3E), the latter being similar to the reported survival for this population (3840).

OCDC mobilizes neoepitope-specific CD8+ T cell responses

To date, it remains unknown whether lysate vaccines can elicit a novel T cell response against tumor neoepitopes resulting from nonsynonymous somatic tumor mutations. One of our main hypotheses was that effective whole-tumor lysate vaccination would induce a broad antitumor immune response, including responses against private tumor antigens. To address this question, we analyzed six patients, in whom after five vaccinations, we detected T cells reactive to autologous tumor cells or DCs pulsed with autologous tumor lysate or for whom a residual sample was available for sequencing from the same tumor used for lysate preparation (Fig. 4A). We identified by exome sequencing a median of 90 mutations per tumor (range, 56 to 117; Table 2). We used our fetchGWI (47) method to call nonsynonymous somatic mutations under conditions that favored sensitivity to avoid false negatives. Using the NetMHC algorithm (48), a total of 566 9- or 10-mer candidate neoepitopes were predicted in silico to bind with high affinity (<500 nM) to patients’ cognate class I HLA alleles, with a range of 50 to 133 neoepitopes predicted per patient (Table 2). We interrogated peripheral blood CD8+ T cells collected before and after completion of induction vaccination (five doses) for recognition of all the above predicted neoepitopes identified in each patient, for which we synthesized and tested both the mutated and the wild-type peptides. We stimulated peripheral blood lymphocytes with pools of predicted peptides for 12 days in vitro, followed by rechallenge with the same peptides and analysis by IFN-γ ELISpot (Fig. 4, B and C, and fig. S6). T cell responses against neoepitopes were further validated by multimer staining (Fig. 4B) and/or polychromatic intracellular cytokine staining (Fig. 4, D and E, and fig. S7). Whereas we could detect neoepitope-specific CD8+ T cells prevaccination only in four patients (nine neoepitopes total, with response to one to four neoepitopes per patient; see examples in Fig. 4, B and C), we detected a T cell response against at least one neoepitope in all six patients on-vaccination (16 total neoepitopes, with response to one to six neoepitopes per patient; table S4). These responses were largely specific to the cognate mutated epitopes, and there was no or little recognition of the wild-type counterpart epitopes (Fig. 4, B and C, and fig. S6). Neoepitope-specific CD8+ T cells isolated after five vaccinations were found to be polyfunctional, that is, co-producing IFN-γ, TNFα, and IL-2, in response to the cognate neoepitopes (Fig. 4, D and E). Together, OCDC significantly expanded neoepitope-specific CD8+ T cell responses, including enhancing preexisting responses against some neoepitopes (Fig. 4F, left) and identification of newly detected responses against additional neoepitopes (Fig. 4F, right).

Fig. 4 Effect of vaccination on neoepitope-specific T cell response.

(A) Schematic representation of neoantigen prediction and validation algorithm. (B and C) Representative examples of neoepitope-specific T cell responses detected pre- and on-vaccination. (B) CD8+ T cell response detected in patient CTE-0011 pre- and on-vaccination (left) against the mutant (mut) neoepitope from the SEPT9 gene, but not the native [wild type (wt)] peptide (right). These analyses were performed with cryopreserved cells after one round of in vitro stimulation. Validation of SEPT9-specific CD8+ T cells is shown by flow cytometry using multimer (No Ag, no peptide; PHA, phytohemagglutinin). The analysis of multimer staining was performed with cryopreserved cells after multiple rounds of in vitro stimulation, as described in Materials and Methods. (C) CD8+ T cell responses in patient CTE-0019 against two neoepitopes. PLK3-specific T cells are detected only on-vaccination; ODZ3-specific T cells are detected also prevaccination, but their frequency markedly increased after immunization (left). These analyses were performed with cryopreserved cells after one round of in vitro stimulation. The right panels show the lack of cross-reactivity against native (wt) peptides for both PLK3 and ODZ3. These analyses required a second round of in vitro stimulation. For (B) and (C), significant differences in magnitude compared to either prevaccination or to wt peptides are identified with “*” and refer to criteria described in Materials and Methods. Out-of-scale T cell responses (for positive controls) are identified with “>”. (D) Representative intracellular detection of IL-2, TNFα, and/or IFN-γ in viable CD8+ T cells on-vaccination after exposure to neoepitopes. (E) SPICE analysis shows polyfunctional response by neoepitope-specific CD8+ T cells harvested on-vaccination. These analyses (D and E) were performed with cryopreserved cells after one round of in vitro stimulation, as described in Materials and Methods. The gating strategy is shown in fig. S7. (F) Increased recognition of tumor neoepitopes on-vaccination in all analyzed patients for whom material was available. Change in preexisting responses (left) and newly detected responses to neoepitopes (right) are shown separately. These data show analyses performed with cryopreserved cells after one round of in vitro stimulation, as described in Materials and Methods. Differences between averages of variables were compared using two-tailed t test. Percentages between groups were compared using χ2 test.

Table 2 Neoepitope analysis of clinical study patients.

Patient identification number, total number of nonsynonymous somatic mutations, HLA class I haplotypes, and number of predicted neoepitopes.

View this table:

Vaccine amplification of preexisting neoepitope-specific responses involves identification of newly detected high-avidity T cell clones

To further understand how OCDC expanded preexisting T cell responses against neoepitopes, we analyzed paired pre- and on-vaccination T cells recognizing the same neoepitope in two patients. In both patients, we documented a marked amplification of the preexisting neoepitope-specific response after five vaccinations. This was accompanied by a >2-log increase in the functional avidity of neoepitope-specific T cells, as revealed by 100-fold lower concentration of peptide required to activate the release of IFN-γ (Fig. 5, A and B). We sorted paired pre- and on-vaccination neoepitope-specific T cells recognizing the HAAT neoepitope using HAAT neoepitope peptide multimer and sequenced their TCRα and β chains to characterize their clonal repertoire. We found few dominant clones in the paired populations, which were nonoverlapping in pre- and on-vaccination blood (Fig. 5D and fig. S8), supporting the notion that vaccination modulated the repertoire of neoepitope-specific T cells.

Fig. 5 Neoantigen analysis on patient CTE-0013 who achieved clinical benefit.

(A) Expansion of preexisting CD8+ T cell response to neoepitope (from ODZ3; patient CTE-0019) on treatment (left) and increase in the functional avidity of neoepitope ODZ3-specific T cell clones pretreatment and on-treatment (right). (B) Expansion of preexisting CD8+ T cell response to neoepitope (from HHAT; patient CTE-0013) on treatment (left) and increase in the functional avidity of neoepitope HHAT-specific T cell clones pretreatment and on-reatment (right). Data from (A) and (B) show analyses performed with cryopreserved cells after one round of in vitro stimulation. (C) Cumulative data showing the marked increase in the functional avidity of neoepitope-specific (HHAT) T cell clones pretreatment (pre) and on-treatment (on). (D) T cell receptor–β (TCRβ) V-J segments recombination landscape of sorted HHAT neoepitope-specific CD8+ T cells from patient CTE-0013 [neoepitope-specific T cell responses pre- and on-treatment were fluorescence-activated cell sorting–sorted using peptide–major histocompatibility complex (pMHC) multimer complexes]. V and J segments are represented according to their chromosomal location on the x and y axes; the frequency of each recombination is shown on the z axis and highlighted by colors. Analysis of TCR repertoires pre- and on-treatment showed a remarkable diversity of the two populations, with few distinct dominant clones, supporting the hypothesis of new priming of neoepitope-specific T cells by the treatment. TCRα V-J segments recombination are shown in fig. S8. (E) Calculated binding mode of peptide KQWLVWLFL on HLA-A*0206. The peptide is shown in ball and stick, colored according to the atom types; with carbon in dark and light gray for residues buried or exposed to the TCR, respectively. The TCR (D83-102501-311993) was removed for clarity. Trp3 and Trp6 are predicted to be buried in the same pocket [see (F)]. Anchor residues Gln2 and Leu9 are deeply buried in the MHC surface. (F) Calculated binding mode of peptide KQWLVWLFL on HLA-A*0206, centered on the peptide Trp3 and Trp6 residues. MHC residues are shown in thick lines, colored according to the atom types, with carbon colored in brown. The side chains of peptide residues Leu4 and Val5 were removed for clarity. (G) Calculated TCR/pMHCs for D1 (pre-Vx)–TCRα (111105)–TCRβ (139954), on the left, and D83 (on-Vx)–TCRα (79822)–TCRβ (139954) on the right. TCRα is colored in light blue. Hydrogen bonds and ionic interactions are shown as thin blue lines. Residues are labeled in brown, black, and blue for MHC, peptide, and TCRα, respectively. More numerous favorable interactions can be found between pMHC and TCRα (79822) (right) than with TCRα (111105) (left) (see table S5). EC50, median effective concentration.

To understand the basis of the different avidities between pre- and on-vaccination clones, and having the cognate TCR, peptide, and HLA sequences available, we next used homology modeling (table S5) to compute the predicted molecular interactions of each of two dominant pre-vaccination and two dominant on-vaccination TCRs with the peptide and HLA-A*0206. The peptide/HLA interactions were found to be conserved across all four cases (table S5) and included the canonical hydrogen bonds between the N-terminal ammonium function and the side chains of major histocompatibility complexes (MHCs) Tyr7, Tyr159, and Tyr171 on one side and between the C-terminal carboxylate function and the side chains of MHC Tyr143, Lys146, and hb-Tyr84 on the other side. The side chain of the N-terminal anchor residue Gln2 was buried and made hydrogen bonds with the side chains of MHC Tyr9 and Glu63 and with the backbone carbonyl of Glu63. The side chain of the C-terminal anchor residue Leu9 was also buried in MHC and occupied a nonpolar pocket constituted by the side chains of MHC Leu81, Tyr116, Thr143, and Trp147. The side chains of both Trp3 and Trp6 were predicted to be buried in the same nonpolar and essentially aromatic pocket, constituted of MHC residues His70, Tyr99, His114, Leu156, and Tyr159, with which they could make several π-π interactions. Trp6 also made a cation-π interaction with MHC Arg97. In addition, Trp3 and Trp6 made π-π interactions together. Peptide Leu7 was predicted to be partially buried in a nonpolar pocket formed by MHC Trp147 and Val152. Finally, peptide residues Lys1, Leu4, Val5, and Phe8 were predicted to be exposed to the TCR surface (Fig. 5, E and F). Table S7 lists the predicted interactions between each of the four TCRs and the peptide/HLA-A*0206 complex. The hTRAV38-2 V-segment present in TCRα D1(79822) and D83(102501), which provides the sequences for CDR1α and CDR2α, led to a strong ionic interaction between CDR1α Glu30 and the side chain of peptide Lys1. It also led to numerically more possible interactions with the MHC in that region: 6 for hTRAV05 versus 10 for hTRAV38-2 (Fig. 5G). In addition, according to these predictions, the CDR3α and CDR3β loops of the on-treatment TCRs led to more numerous favorable interactions with the peptide residues than the pretreatment TCRs. As a consequence, the pretreatment TCRs were predicted to make 5 and 8 favorable interactions, respectively, with the cognate peptide compared to 10 interactions for each of the two posttreatment TCRs. In addition, the pretreatment TCRs were predicted to make a total of 15 and 18 favorable interactions with the MHC, compared to 17 and 20 interactions for the posttreatment TCRs. In summary, the posttreatment TCRs were predicted to make qualitatively more favorable interactions with the peptide-MHC (pMHC) compared to the pretreatment ones, supporting the notion that enhanced avidity of on-treatment T cells was, in part, driven by the generation of novel dominant T cell clones endowed with higher-affinity TCRs after five doses of immunization.


We tested a DC-based vaccine pulsed with autologous oxidized whole-tumor lysate based on earlier preliminary data that it can induce a broad immune response to multiple tumor antigens (37), a condition that could be advantageous in the context of combinatorial immunotherapy. Here, we show that such a personalized approach to cancer immunization is feasible, well tolerated, and safe either when given alone or when combined with intravenous bevacizumab and low-dose intravenous cyclophosphamide.

As previously hypothesized, we demonstrated that our vaccination approach can effectively mobilize antitumor immunity in ovarian cancer patients and that postimmunization peripheral blood T cells could recognize autologous tumor cells in addition to DCs loaded with autologous tumor lysate. In some patients that we could analyze who received vaccination for several months, we observed a progressive increase in the frequency of tumor-reactive peripheral blood. Postimmunization peripheral blood T cells were able to kill autologous tumor more effectively than preimmunization peripheral blood T cells, both in vitro and in vivo in PDX mice as demonstrated in one patient. The detection of postimmunization T cells that were reactive to either autologous tumor cells or to DCs loaded with autologous tumor lysate was associated with significantly better clinical outcome. Progression was seen among patients in whom we could not detect any such T cells postimmunization, whereas the median time to progression of those in whom we documented reactive T cells was >15 months among which one-third had not progressed at 24 months. In addition, we observed remission inversion in more than 60% of the patients who responded to vaccine.

Recently, some papers reported the successful induction of responses to tumor neoepitopes after vaccination using synthetic peptides vaccine (27, 28, 49). Here, we report that whole-tumor antigen vaccination elicits a specific response against tumor neoepitopes. We analyzed six evaluable patients in whom we could detect on-immunization T cells reactive to autologous tumor cells or to autologous DCs pulsed with autologous tumor lysate. In all these patients, we documented CD8+ responses against one or multiple neoepitopes on-immunization. Although we found that immunization induced both CD4+ and CD8+ T cells, a feature that is considered important for effective antitumor immune responses (50), our neoepitope discovery algorithm selectively assessed CD8+ responses as it was restricted to class I–restricted peptides. It is thus possible that our study underestimated the magnitude of neoepitope-specific T cells that were elicited by the immunization.

We found that many responses detected after five doses of vaccination were new responses undetectable prevaccination. However, we also saw that vaccination amplified preexisting responses to some neoepitopes, which we could already detect prevaccination. In two cases analyzed in detail, we found a marked increase in the avidity of on-immunization T cells, which recognized cognate neoepitope peptide at more than 100-fold lower concentrations. To understand more how OCDC expanded preexisting T cell response to neoepitopes, we sorted cells specific to one neoepitope, which were then subjected to TCR sequencing to learn more on their clonal repertoire. We identified few immunodominant clones in preimmunization and postimmunization samples, but there was no overlap between the two conditions, indicating that, in fact, the observed amplified response was due to the priming of novel high-avidity T cell clones. We thus applied molecular modeling to compute the strength of interactions between each TCR and its cognate pMHC formed by the neoepitope and its assigned HLA-A*0206. Postimmunization TCRs were predicted capable of higher strength interactions with the cognate pMHC as we have previously demonstrated (51), further substantiating that increased avidity of on-immunization T cells was due to the newly detected clones with higher-affinity TCR. Collectively, these data provide evidence that DCs pulsed with oxidized tumor lysate can effectively induce the detection of novel T cell clones against multiple tumor neoepitopes, which can exhibit high-affinity TCRs and lead to high-avidity responses.

The clinical efficacy of cancer vaccines has been long debated because vaccines alone have not been efficacious in producing objective responses in the clinic (52). The advent of effective immunomodulatory molecules has, however, raised hopes that cancer vaccines could become an important partner in combinatorial approaches, and preclinical data amply support this notion (30, 53). Vaccines targeting tumor neoepitopes could be particularly promising because they could expand neoepitope-specific T cells (2729). Expectedly, neoepitope-specific T cells could exhibit high avidity because they presumably escape thymic selection, which depletes self-recognizing clones bearing high-affinity TCRs. However, concrete evidence supporting this hypothesis is missing to date. Here, we provide evidence that DCs pulsed with whole-tumor lysate can expand novel high-avidity T cell clones with markedly high-affinity TCRs against tumor neoepitopes. These observations suggest that perhaps peripheral tolerance mechanisms suppress at the steady state the emergence of such high-affinity TCR clones, whereas exogenous DCs loaded with immunogenic lysate can drive their expansion.

The study was designed to address safety and feasibility of the combination of DCs with bevacizumab and cyclophosphamide. We pursued the above combination based on the hypothesis that cyclophosphamide would enhance vaccination, and bevacizumab could be a suitable partner for immunotherapy (41, 54) and the notion that the combination of bevacizumab and low-dose cyclophosphamide has proven safe in ovarian cancer patients (3840, 55). In addition to providing encouraging safety data, we show that the cohort who received the full combination exhibited longer OS relative to patients who received only vaccine plus bevacizumab. The addition of low-dose cyclophosphamide resulted in more effective immunization, which was also associated with concordant serum biomarker changes. The 2-year survival of the cohort who received the full combination was considered promising, albeit preliminary, compared to literature and historic institutional controls.

Albeit the limitations that one can encounter with autologous whole-tumor antigen DC vaccines such as the lack of immunogenicity of lysates, the paucity of tumor material sufficient to produce vaccine, and the difficulty of the production itself, these pilot clinical observations along with our biological observations provide encouraging data for the pursuit of DC vaccines pulsed with autologous whole-tumor antigen in the context of personalized combinatorial immunotherapy approaches, where the mobilization of broad antitumor immunity and neoepitope-specific T cells appears as an important strategy.


Study design

Patients with recurrent ovarian, fallopian tube, or primary peritoneal cancer were enrolled in a phase 1 study (NCT01132014). This was a single-center study of three sequential cohorts, with the primary objective to establish the safety and biological activity of OCDC, an autologous DC vaccine pulsed with autologous oxidized whole-tumor lysate, alone or in combination with intravenous bevacizumab or intravenous bevacizumab and intravenous cyclophosphamide (fig. S1). The first five patients received only OCDC vaccine, whereas in the second cohort, 10 patients received OCDC vaccine plus intravenous bevacizumab (15 mg/kg) on the same day every 3 weeks, and in the third cohort, 10 patients received intravenous cyclophosphamide (200 mg/m2) followed by OCDC vaccine plus intravenous bevacizumab (15 mg/kg) on the next day every 3 weeks. The rationale for this study has been reported previously (56). After 12 to 15 weeks of vaccination, patients who had no evidence of disease or were stable were offered maintenance vaccination, whereas those who progressed went off the study.

Inclusion criteria required patients with recurrent ovarian, fallopian tube, or primary peritoneal cancer who were ≥18 years old; had sufficient tumor previously harvested at secondary cytoreductive surgery and stored for lysate preparation; had completed physician’s choice chemotherapy after secondary debulking as indicated; and had a baseline Eastern Cooperative Oncology Group performance status ≤1. The study was approved by the U.S. Food and Drug Administration under BB-IND-14269 and by the University of Pennsylvania’s Institutional Review Board. All patients gave written informed consent before initiation of any study procedures. After enrolment, patients underwent 10- to 15-liter leukapheresis to harvest PBMCs for OCDC manufacturing.

Vaccine manufacturing

As previously described (37) and as depicted in Fig. 1A, vaccine was manufactured at the Clinical Cell and Vaccine Production Facility at the University of Pennsylvania. DC generation was conducted as previously described (37). Briefly, elutriated monocytes of patients were cultured in CellGenix DC media (CellGenix), 2% human AB serum (Valley Biomedical Inc.), l-glutamine (2 mM), penicillin (100 U/ml), streptomycin (100 μg/ml; Cellgro), clinical grade human GM-CSF (500 IU/ml; Leukine, Bayer Healthcare Pharmaceuticals), and animal-free research grade IL-4 (250 IU/ml; R&D Systems Inc.). After 4 days, CD11c, CD14, and HLA-DR were measured on output cells to determine the percentage of cells with a DC phenotype, and these were routinely more than 70%. After overnight autologous lysate loading, DCs were matured with LPS (60 EU/ml; Escherichia coli O:113; gift from A. Suffredini at the National Institutes of Health (NIH)] and recombinant human IFN-γ (2000 IU/ml; InterMune). Vaccine met release criteria in all 20 patients.

Vaccine aliquots (~5 to 10 × 106 DCs per dose) were cryopreserved at −140°C, thawed, and washed before each administration. Patients received five doses of vaccines (~5 to 10 × 106 DCs per dose) intranodally every 3 weeks under ultrasound guidance and continued on a monthly maintenance regimen until disease progression or exhaustion of vaccine supply, whichever came first. Safety was determined using the National Cancer Institute Common Terminology Criteria for Adverse Events version 4.0. Patients underwent CT scan at enrolment and on day 86. Clinical response was based on the RECIST 1.1 and immune-related response criteria. Patients were then either treated with maintenance vaccine plus bevacizumab with or without low-dose cyclophosphamide or went off the study.

All patients underwent blood draws for immune assessment baseline (pretreatment), throughout the study, and at EOS (posttreatment). Samples were transferred and analyzed at the Center of Experimental Therapeutics at the Lausanne Branch of the Ludwig Institute for Cancer Research.

Identification of nonsynonymous tumor mutations

Genomic DNA from available cryopreserved tumor tissue and matched PBMC was isolated using DNeasy kit (Qiagen) according to the manufacturer’s instructions and subjected to whole-exome capture and paired-end sequencing using the HiSeq 2500 Illumina platform. Sequencing was performed at the High-Throughput Sequencing Research Facility of the Penn Genome Frontiers Institute and at BGI@CHOP next-generation core. Data analysis was performed at the Vital-IT Systems Biology Division, Swiss Institute of Bioinformatics (SIB), Lausanne. Somatic variants were called from the exome reads and the reference human genome hg19 by using a software pipeline composed of a genome mapping tool, fetchGWI (47), followed by a detailed sequence alignment tool, align0 (57). We used parameters (a minimum of four reads or 15% of the total reads supporting the alternative allele) that favored high sensitivity to avoid false negatives. Comparison between our method and GATK ( revealed up to 95% concordance on variant calls. Variations present in the tumor samples and absent from the corresponding blood samples were assumed to be somatic.

Neoepitope prediction

High-resolution HLA typing at the A, B, and C locus was performed at the Immunogenetics Laboratory, Department of Pathology and Laboratory Medicine, the Children’s Hospital of Philadelphia, using a next-generation sequencing protocol, as previously described (58). Binding predictions to class I HLA alleles for all candidate peptides incorporating somatic nonsynonymous mutations were performed using the NetMHC algorithm v3.4 (59, 60). Mutant 9- and 10-mer peptide sequences containing the somatically altered residue at each possible position and the corresponding wild-type peptide sequences were selected for further testing.

Statistical analyses

Differences between averages of variables were compared using two-tailed t test for variables with normal distribution or by using Mann-Whitney nonparametric test for non-normal variables. Percentages between groups were compared using χ2 test. Median survival times were computed using Kaplan-Meier methods. Differences in OS or PFS were assessed using the log-rank test. The testing level (α) considered was 0.05. Whenever it was relevant, Bonferroni corrections were applied. Analyses were performed using GraphPad Prism v7.03. The number of patients was restricted on the basis of the ability to perform the procedure and follow the patients successfully for at least 2 years. No power calculation was performed for defining the size of this convenience sample.


Materials and Methods

Fig. S1. Schema of the clinical study.

Fig. S2. Analysis of DC vaccine product.

Fig. S3. Immune response in patient CTE-0017.

Fig. S4. Immune competency of patients.

Fig. S5. Clinical response does not depend on prevaccination T cell gene expression.

Fig. S6. Validation of CD8+ T cell responses against neoepitopes and wild-type peptides.

Fig. S7. Representative example of the gating strategy applied for intracellular cytokine staining analyses.

Fig. S8. TCRα V-J segments recombination landscape of sorted HHAT neoepitope-specific CD8+ T lymphocytes.

Table S1. Adverse events detected in cohorts 1 to 3.

Table S2. Comparative parameters in immune responder and nonresponder patients.

Table S3. Gene set enrichment analysis between clinical responders and nonresponders does not find immune-related pathways differentiating both groups of patients before the vaccination.

Table S4. Description of validated HLA class I neoepitopes identified.

Table S5. Protein Data Bank entries used as templates to model the TCRα, TCRβ, MHC, β-microglobulin, and the peptide epitope.

Table S6. Molecular interactions between the peptide epitope (KQWLVWLFL) and MHC HLA-A*0206, as predicted by homology modeling.

Table S7. Molecular interactions between the TCR and pMHC, as predicted by homology modeling.

References (6166)


Acknowledgments: We thank all the patients who participated in this study. We thank the doctors, nurses, and staff at the Hospital of the University of Pennsylvania for their work enabling the conduct of the clinical study, as well as the staff at the Penn Ovarian Cancer Research Center and the Penn Tumor and Tissue Acquisition Bank at the Department of Pathology for handling tumor and blood samples. We thank D. Siegel and the staff of the Penn Apheresis Unit for conducting all aphereses; A. Brennan and the staff of the Penn Cell and Vaccine Production Facility for manufacturing the vaccines and providing samples for analysis. We thank D. Vanhecke and all the members of the Humanised Mouse Facility of the Department of Oncology at the University of Lausanne for helping design and supporting the experiments using NSG mice included in Fig. 2. Funding: The conduct of the clinical trial was supported by NIH grants P50 CA083638 SPORE in Ovarian Cancer (to G.C.), R21 CA156224 (to G.C.), and 5P30 CA016520-36 Abramson Cancer Center of the University of Pennsylvania Core Support Grant (to B.L.L.), and grants from the Marcus Foundation (to G.C. and L.E.K.) and the Ovarian Cancer Immunotherapy Initiative (to G.C. and L.E.K.). Tumor sequencing was supported by a grant by the Pennsylvania Department of Health (to G.C.; the Department specifically disclaims responsibility for any analyses, interpretations, or conclusions). All immune analyses were supported by the Ludwig Institute for Cancer Research and a grant by the Ovacure Foundation (to A.H., G.C., and L.E.K.). The Vital-IT Center for high-performance computing of the SIB is supported by University of Lausanne/École Polytechnique Fédérale de Lausanne/University of Geneva/University of Bern/Université de Fribourg and The Swiss Federal Government through the State Secretariat for Education, Research and Innovation. Author contributions: L.E.K., G.C., and A.H. jointly conceived the study and its analyses. A.H. developed and supervised the immune analyses. E.O. and S.B. performed the experiments and the data analysis and interpretation. A.R., J.R., R.G., B.J.S., C.I., D.G., and U.D. performed sequencing and data analysis and interpretation. A.M., P.G., B.J.S., S.T., O.M., V.Z., D.D., J.L.T., I.X., P.R., C.F., L.D., P.B., B.C., A.S., C.C., D.A.T., H.L.N., C.H.J., B.L.L., D.J.P., R.M., and D.S.M. provided additional support to the experiments and data analysis. L.E.K. was responsible for provision of study resources, materials, and patient access. S.B., L.E.K., R.G., V.Z., D.G., G.C., and A.H. wrote the manuscript. All authors gave their final approval to the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: Sequence data have been deposited at the European Genome-phenome Archive, which is hosted by the EBI and the CRG, under accession number EGAS00001002803.

Stay Connected to Science Translational Medicine

Navigate This Article