Purpose: Despite the vast number of clinical trials conducted so far, dendritic cell (DC)-based cancer vaccines have mostly shown unsatisfactory results. Factors and manufacturing procedures essential for these therapeutics to induce effective antitumor immune responses have yet to be fully characterized. We here aimed to identify DC markers correlating with clinical and immunologic response in a prostate carcinoma vaccination regimen.

Experimental Design: We performed an extensive characterization of DCs used to vaccinate 18 patients with prostate carcinoma enrolled in a pilot trial of T-cell receptor gamma alternate reading frame protein (TARP) peptide vaccination (NCT00908258). Peptide-pulsed DC preparations (114) manufactured were analyzed by gene expression profiling, cell surface marker expression and cytokine release secretion, and correlated with clinical and immunologic responses.

Results: DCs showing lower expression of tolerogenic gene signature induced strong antigen-specific immune response and slowing in PSA velocity, a surrogate for clinical response. These DCs were also characterized by lower surface expression of CD14, secretion of IL10 and MCP-1, and greater secretion of MDC. When combined, these four factors were able to remarkably discriminate DCs that were sufficiently potent to induce strong immunologic response.

Conclusions: DC factors essential for the activation of immune responses associated with TARP vaccination in prostate cancer patients were identified. This study highlights the importance of in-depth characterization of DC vaccines and other cellular therapies, to understand the critical factors that hinder potency and potential efficacy in patients. Clin Cancer Res; 23(13); 3352–64. ©2017 AACR.

Translational Relevance

Dendritic cell–based vaccines have been widely tested in preclinical and phase I/II clinical studies, most of which led to unsatisfactory results. Here, we analyzed in-depth dendritic cells (DCs) used in a clinical trial vaccinating prostate carcinoma patients and correlated DC phenotype with clinical and immunologic response. Our analysis revealed a gene signature of tolerogenic DCs that was minimally expressed in patients showing a strong immunologic and clinical response. In parallel, it was observed that a similar predictive value could be obtained using a four-protein index based on CD14 expression on DC membrane, and secretion levels of IL10, MCP-1, and MDC. These results, if further validated in larger studies, could be used to tailor DC vaccination only to those patients showing low tolerogenic signature in their DCs. Furthermore, our study provides a framework that can be used for other cell therapies to identify markers correlating with clinical response.

Dendritic cells (DCs) are antigen-presenting cells that are able to activate both innate and adaptive arms of the immune system (1). For this reason, DC-based vaccines represent a promising immunotherapeutic approach in several clinical settings. In fact, since the discovery of monocyte differentiation into DCs (2), over 300 clinical trials have been conducted, which have proven the feasibility and safety of DC vaccines (3). However, despite extensive preclinical and clinical studies, very few clinical trials have demonstrated the desired clinical efficacy. For the majority of trials, the overall response rates have been well below 20% (4). Many reasons have been hypothesized for such low response rates, among which the generation of DCs with suboptimal potency is considered the most relevant (4). It is not known yet how to generate the most potent DCs; furthermore, differences in clinical setting, study design, sources of antigens, and route of administration make it almost impossible to compare results from previously conducted trials to clearly delineate the shared determinants of in vivo efficacy of DC-based vaccines.

Compared with drug therapy, cell therapies are more challenging because of their considerable lot-to-lot and patient-specific variability that in most cases has yet to be sufficiently quantified and characterized (5). In fact, while each cell therapy lot has to be tested for identity, consistency, and potency among other tests, feasibility issues dictate these tests to be focused on a handful of factors and, therefore, they cannot ensure an exhaustive characterization of each lot. An accurate characterization of DCs should ideally assay all the factors affecting their in vivo biological functions: antigen processing and presentation, expression of costimulatory signals, absence or reduced expression of coinhibitory signals, lymph node migration, and secretion of activating cytokines and chemokines. These are all essential features of potent DCs and should be thoroughly tested. As it is impossible to routinely evaluate each product for every cell function using cellular assays, the identification of reliable biomarkers of identity, consistency, and potency of cell therapies is highly encouraged by regulatory agencies beginning in the earliest phases of clinical development of the cellular product (6).

Many factors are known to play a key role in DC-induced activation of the immune system. Secretion of interleukin-12 (IL12) is considered the most important driving factor for Th1 inflammatory T-cell activation (7). Surface expression of CCR7 is necessary for DC migration into lymph nodes and expression of costimulatory molecules (i.e., CD80 and CD86) is essential for the activation of naïve T cells (8). On the other hand, several detrimental factors have also been identified. Secretion of IL10 is considered a hallmark of tolerogenic activities exerted by DCs (8, 9). Similarly, the maintenance of immature/monocytic factors (e.g., CD14) is also known to be characteristic of tolerogenic DCs (10, 11). Even though these and many other molecular factors have been characterized thoroughly for their role in DC function and many of them have been used to discriminate among DCs produced using different differentiation/maturation procedures, it has rarely been determined how these factors are differentially expressed among DCs manufactured using the same differentiation/maturation procedure and whether such difference has a functional relevance.

Recently, our group has shown that even when highly standardized procedures are used to generate monocyte-derived DCs, manufacturing, intradonor- and interdonor-related factors may affect DC phenotype (12). In particular, we observed that while most of the well-known and usually tested DC markers (e.g., CD80, CD86, CD83, HLA-DR) did not show any differences in the expression among DCs generated at different times from different donors, the expression of several genes and the levels of several key secreted cytokines and chemokines showed significant variability among DC products. However, whether such lot-to-lot variability affects the identity, potency, and/or efficacy of DC-based vaccines used in human clinical trials has yet to be determined.

NCI-09-C-0139 (NCT00908258) is a randomized, prospective, pilot study of vaccination with a mixture of wild-type (TARP27-35) and epitope-enhanced (TARP29-37-9V) T-cell receptor gamma chain alternative reading frame protein (TARP) peptides in HLA-A*0201 patients with stage D0 prostate cancer (status postdefinitive treatment of primary tumor, with prostatectomy or radiotherapy, no evidence of visceral or bone metastasis with persistently elevated or rising PSA levels, i.e., biochemical progression) and at increased risk for disease progression based on PSA doubling time (PSADT; ref. 13). In stage D0 prostate cancer, the rate of change of PSA as slope or doubling time has been documented to be a valid predictor of clinical outcome (14). TARP is a tumor-associated antigen expressed in over 90% of prostate and 50% of breast carcinomas (15). The study compared two vaccination regimens: in one, TARP peptides were admixed with Montanide ISA 51 VG plus Sargramostim to generate a peptide emulsion administered by deep subcutaneous injection; in the other, TARP peptide-pulsed autologous dendritic cells (DC) were administered intradermally. TARP DC vaccines were manufactured and administered every 3 weeks at weeks 3, 6, 9, 12, and 15 as part of the primary vaccination series, with a conditional sixth booster at week 36 dependent on documented immunologic and/or clinical responses at week 24. The original 48-week study design was amended and extended to subsequently allow seventh and/or eighth booster doses of vaccine at weeks 48 and 96 after initial immunologic and clinical activity of TARP vaccination was documented.

In the current study, we analyzed 114 peptide-pulsed DC preparations manufactured to vaccinate 18 patients randomized to the autologous TARP peptide-pulsed DC arm to understand which factors are affected by lot-to-lot variability in clinical GMP manufactured DCs and whether such variability has an impact on DC identity, potency, and efficacy. By analyzing DC surface marker expression, gene expression profiles, protein secretion profiles, and culture data, we observed the existence of a tolerogenic DC signature that was negatively correlated with the development of clinical and immunologic response.

Mature DC manufacturing process

DCs were manufactured according to a standard operating procedure established in the Cell Processing Section (CPS), Department of Transfusion Medicine (DTM), Clinical Center (CC), NIH, Bethesda, MD. Briefly, peripheral blood mononuclear cell (PBMC) concentrates were collected by apheresis using a Cobe Spectra Apheresis System (Terumo BCT) from 18 patients enrolled in clinical trial NCI-09-C-0139 (NCT00908258). All patients signed an informed consent approved by the NCI Institutional Review Board. Monocytes were enriched directly from the leukapheresis products by elutriation using the Elutra Cell Separation System (Terumo BCT) according to the manufacturer's recommendations. Monocytes were cryopreserved in aliquots of 3 × 108 cells. DCs were manufactured from single monocyte aliquots after assessing post-thaw viability and purity; in all cases, both were greater than 80%. At the time of culture initiation, thawed monocytes were resuspended in RPMI1640 media, containing 10% autologous heat-inactivated plasma, 10 mcg/mL gentamicin, GM-CSF (Leukine Sargramostin, 2,000 IU/mL, Genzyme), and IL4 (USP grade recombinant human IL4, 2,000 IU/mL, CellGenix) at a final concentration of 1.5 × 106/mL in tissue culture flasks (Corning Incorporated Life Sciences) and incubated at 37°C in 5% CO2. On day 2, fresh cytokines were added to the culture at the same concentrations together with keyhole limpet hemocyanin (KLH, 10 mcg/mL, Stellar Biotechnology Port Hueneme). On day 3, a maturation cocktail containing lipopolysaccharide (LPS; 30 ng/mL, CTEP, NIH, Frederick, MD) and IFNγ (Actimmune Interferon gamma-1b, 1,000 IU/mL, Intermune) was added. Twenty-four hours later, DCs were harvested. After two washes, DCs were resuspended in infusion media made of Plasma-Lyte A and 10% autologous heat-inactivated plasma. DCs were then pulsed in separate aliquots with wild-type (WT) 27–35 and epitope-enhanced (EE) 29-37-9V TARP peptides (NeoMPS, Inc.) at 37°C in 5% CO2. After pulsing, DCs pulsed with the two peptides were combined and tested for recovery, viability, purity, sterility, mycoplasma absence, endotoxin concentration, and expression of surface markers by flow cytometry (see below). Release criteria for each lot were defined on the basis of CD83 expression by flow cytometry and Trypan blue viability set as ≥ 70% and 60% respectively. If the cells met all the release criteria, then 2 × 107 viable DCs were used for vaccination and were administered intradermally to patients. The remaining cells were centrifuged, and the supernatant was used for cytokine profile analysis (see below) and excess DCs were used for RNA extraction.

Flow cytometric analysis

Analysis of expression of surface markers was performed using fluorescently labeled antibodies (Abs) and flow cytometry. The purity of the elutriated monocytes was assessed by flow cytometry using CD33-PE, CD15-FITC, CD3/CD19/CD56-APC, and CD45-APC-Cy7 (Becton Dickinson) and isotype controls (Becton Dickinson). DCs were analyzed after pulsing on day 4. The analysis included the standard “DC panel” adopted in our institution as lot release for mature DC products and other investigational markers. The panel consisted of CD86-FITC, CD83-PE, CD14-APC, HLA-DR-FITC, CD123-PE, CD11c-APC, CD80-FITC, CD54-APC, CCR7-APC, and CD38-FITC (Becton Dickinson). Flow cytometry acquisition and analysis were performed with a FACScanto flow cytometer (Becton, Dickinson and Company) according to CPS standard operating procedures. Spectral overlaps were electronically compensated using single color controls. Quality controls were run before each session according to internal quality control policy.

Gene expression profiling

Total RNA was extracted from the unused fraction of DCs using an miRNeasy kit (Qiagen). Universal Human Reference RNA (Stratagene) was used as reference. Test samples and reference RNA were amplified and labeled using an Agilent kit according to the manufacturer's instructions and hybridized on Agilent Chip (Whole Human genome, 4 × 44 k, Agilent Technologies). The arrays were scanned using an Agilent Microarray Scanner and images analyzed using Agilent Feature Extraction Software 9.5.1.1. The resulting data were uploaded onto mAdb Gateway (http://madb.nci.nih.gov), the Agilent-normalized processed signals retrieved, and analyzed with BRB Array Tools (http://linus.nci.nih.gov/BRB-ArrayTools.html). The processed dataset was subjected to filtration based on signal intensity, quality, and presence across the dataset. A total of 35,753 genes passing these criteria were selected and log 2 base ratios were used for further analysis.

Protein analysis platform

Supernatants of DC-conditioned infusion media were collected, stored at −80°C, and analyzed all together at the end of the collection. The levels of 11 soluble factors were further assessed on a customized antibody-based platform (Aushon) consisting of a multiplex array with 11 different mAbs spotted per well in standard 96-well plates. A sandwich ELISA technique was used to generate signals via chemiluminescent substrate. Light corresponding to each spot in the array was captured by imaging entire plates with a commercially available cooled charge-coupled device (CCD) camera. Data were reduced using image analysis software (Aushon Proteome Arrays) that calculates exact values (pg/mL) based on standard curves. Prior to further analysis, protein concentrations were normalized according to the number of DCs.

Evaluation of immunologic response

Patient evaluation of immunologic response was performed as described by Wood and colleagues (13). For the in vitro stimulation, frozen PBMCs were thawed, resuspended, and plated into wells of 24-well tissue culture plates (Corning, 3524) in media containing a 1:1 ratio of RPMI (Gibco, 21870-084) and Click medium (Sigma, C5572), 10% human AB serum (Gemini, 100-512), 1% penicillin/streptomycin-l-glutamine (Gibco, 10378-016), 1% sodium pyruvate (Gibco, 11360-070), 35 mmol/L HEPES (Gibco, 15630-080) and 50 μmol/L 2-mercaptoethanol (Sigma, M7522). On day 0, PBMCs were stimulated with a mixture of TARP WT 27-35, TARP WT 29-37, and TARP EE 29-37 peptides at 10 μg/mL in the presence of 1,000 IU/mL recombinant human IL7 (Peprotech, 200-07). On day 3, recombinant human IL2 (Teceleukin; Roche, 23-6019) was added (20 IU/mL). On day 5 and 6, half of the culture supernatant was replaced with fresh CTL media and the cells were harvested on day 7 and 8.

All ELISPOT assays were performed in the Laboratory of Cell-Mediated Immunity at Leidos Biomedical Research, Inc. (formerly SAIC-Frederick, Inc.) which is certified by Clinical Laboratory Improvements Amendments. Day 7/8 IVS PBMCs as the effectors and peptide-pulsed autologous day 0 PBMCs as the antigen-presenting cells (APC) were cocultured at a 1:1 ratio (2.5 × 104 or 105 cells). The APC were pulsed with 10 μg/mL TARP or control peptides 2 hours at 37°C before being plated with the effectors.

The day before assay setup, 96-well polyvinylidene fluoride membrane, HTS opaque plates (Millipore, MSIPS4W10) were coated overnight with a 1:100 dilution of anti-human IFNγ capture antibody (1 mg/mL, Mabtech 3420-3-1000) in DPBS. Antibody-coated plates were washed in DPBS and blocked with 5% human AB ELISPOT medium. The effectors and APCs were plated and incubated for 18–20 hours. Then, the plates were washed six times with 0.05% Tween 20 in DPBS. Plates were incubated for 2 hours with a 1:2,000 dilution of the biotinylated secondary antibody, anti-human IFNγ (1 mg/mL Mabtech, 3420-6-1000) in DPBS/1% BSA/0.05% Tween and then washed four times in DPBS. A 1:3,000 dilution of streptavidin alkaline phosphatase (Mabtech, 3310-10) in DPBS/1% BSA was added to each well for 1 hour followed by four washes in DPBS. Finally, the BCIP/NPT substrate, 100 μL/well, (KPL, 50-81-07) was added for 7 to 10 minutes, resulting in the development of spots. Three washes in distilled water were performed to stop reaction. Spots were visualized and counted on overnight-dried plates using the ImmunoSpot Imaging Analyzer system (Cellular Technology Ltd.) and ImmunoSpot software v5.1. ELISPOT results were expressed as the “number of spots per 106 responder cells” after subtracting background spots obtained in wells of effectors with nonpulsed PBMC.

Statistical analysis

ROC curves were generated using the R package “pROC” (16). ROC curve represents an easy visualization tool because it illustrates the performance of the variable under study by plotting specificity versus sensitivity of the test for each possible cutoff; and the AUC summarizes the overall ROC curve and can be considered as a summary statistic of its ability to classify cases correctly. A perfect test would have an AUC of 100%; a worthless test would have an AUC of 50%. According to an arbitrary guideline, AUC values may be classified as follows: 90%–100%, excellent; 80%–90%, good; 70%–80%, fair; 60%–70%, poor; 50%–60%, fail (17). The area under the curve (AUC) was used as a measure of the performance of a classifier and confidence intervals were computed with Delong method. Clinical responses as assessed by changes in slope log PSA (mathematically equivalent to an inverse calculated PSADT) at weeks 24 and 48 or strong immunologic responses (defined as a TARP-specific interferon-gamma ELISPOT count > 500) were used to classify DCs. Clinical response at week 24 was used for patient #203 that went off study during the trial.

Unsupervised hierarchical clustering and principal component analysis (PCA) of the whole dataset were run with Partek Genomic Suite (Partek). Davies–Bouldin Index was calculated with Partek to identify the number of clusters between 2 and 20 that better separates samples in subgroups. Class comparisons to identify genes differentially expressed among DCs were performed with BRB ArrayTools with a P value threshold of 0.001. To control false discoveries, the false discovery rate (FDR) was calculated for each analysis as the ratio of the expected number of false discoveries divided by the number of discoveries as described by Sorić (18).

Weighted Gene Co-expression Network Analysis (WGCNA) was performed using the R package “WGCNA” (19). The analysis was applied only on the most variable quartile (9,112 genes) as suggested by package instructions. To apply more stringent criteria in module definition, we applied a modification to standard protocol. The dataset was split in two and WGCNA was then performed in both datasets. Only genes assigned to the same network in both analyses were considered as forming a module and used in subsequent analysis. Gene Ontology was performed using the Database for Annotation, Visualization and Integrated Discovery (DAVID, http://david.abcc.ncifcrf.gov/; ref. 20) and Network analysis was performed using Qiagen's Ingenuity Pathway Analysis (IPA, Qiagen, www.qiagen.com/ingenuity). The clustering of genes belonging to Module 2 was performed with Cluster (21) and results were visualized with Java Treeview (22). For the meta-analysis, all publicly available tolerogenic DC datasets with clear sample information were selected from GEO and EMBL-EBI database. P values were calculated with Fisher exact test.

The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus (23) and are accessible through GEO Series accession number GSE85698 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE85698)."

CD14/IL10/MCP1/MDC index was calculated as follows: each DC preparation was ranked according to the percent of CD14+ cells and concentrations of IL10, MCP-1, and MDC measured in supernatants in decreasing order (i.e., rank 1 to the highest expressing DC preparations). Then, taking into account that MDC levels negatively correlated with Module 2, whereas CD14, IL10, and MCP-1 levels positively correlated with Module 2, the index was calculated as: MDC rank − (CD14 rank + IL10 rank + MCP-1 rank). In this way, high index scoring DC preparations showed low expression of MDC and high expression of CD14, IL10, and MCP-1.

Clinical study and DC phenotype

Each DC vaccine was manufactured starting from one aliquot of autologous cryopreserved monocytes cultured for 3 days with GM-CSF and IL4. On day 2, KLH was added to the culture as a control antigen. On day 3, cells were matured for additional 24 hours with LPS and IFNγ and then pulsed for 2 hours with WT and EE TARP peptides.

Thirteen of 16 evaluable patients were considered to have achieved a clinical response (decrease in slope log PSA) at week 48 (two additional patients completed the treatments but their week 48 clinical responses were not included as a result of the data analysis cut-off date for the dataset; ref. 13). The development of TARP-specific immune response (assessed by IFNγ ELISPOT) was observed in 10 of 18 evaluable patients (IFNγ ELISPOT was performed on patient samples taken at baseline and weeks 12, 18, and 24 after vaccination). Immune activation against control antigen KLH was observed in the majority of patients (15 of 16 subjects in whom KLH reactivity was assessed). Clinical (change in slope log PSA) and immunologic responses were assessed both qualitatively (i.e., in terms of responder or nonresponder) and quantitatively (Table 1). Interestingly, slope log PSA responses were observed almost independently of TARP-specific T-cell responses; however, a strong immunologic response (defined as a median ELISPOT reading greater than 500) was observed only in patients with a notable decrease in slope log PSA (<0.045; i.e., equivalent to a PSADT that at least doubles upon vaccination and therefore considered to indicate a stronger clinical response).

Table 1.

Clinical and immunologic patient responses

IDGleason scoreBaseline PSA concentrationBaseline PSA doubling time (mo)PSA doubling time at week 48 (mo)Baseline log PSA slopeLog PSA slope at week 48Clinical response at week 24aIntensity of clinical response at week 24aClinical response at week 48Intensity of clinical response at week 48bImmunological responseIntensity of immunological responsec
201 0.91 6.4 43.3 0.109 0.016 Yes −0.172 Yes  −0.093 No 
202 4.99 12.8 16.1 0.054 0.043 Yes −0.017 Yes  −0.011 Yes 210 
203 1.94 6.0 2.7 0.116 0.260 No 0.144 Off Study  Yes 30 
204 0.96 5.9 3.8 0.118 0.184 No 0.039 No  0.066 No 50 
205 4.78 2.9 11.4 0.239 0.061 Yes −0.148 Yes  −0.178 Yes 1180 
206 0.52 5.1 4.2 0.137 0.167 No 0.036 No  0.03 Yes 100 
207 3.53 10.5 12.0 0.066 0.058 Yes −0.002 Yes  −0.008 No 
208 12.00 6.3 8.3 0.110 0.084 Yes −0.054 Yes  −0.026 No 
209 1.33 7.0 −34.7 0.099 −0.020 Yes −0.018 Yes  −0.119 No 
210 13.00 12.2 36.5 0.057 0.019 Yes −0.052 Yes  −0.038 No 
211 1.41 4.1 12.4 0.169 0.056 Yes −0.087 Yes  −0.113 Yes 710 
212 5.88 3.7 4.7 0.187 0.146 Yes −0.002 Yes  −0.041 No 
214 17.30 8.8 23.8 0.079 0.030 Yes −0.028 Yes  −0.049 Yes 4480 
215 9.39 5.5 11.5 0.126 0.060 Yes −0.103 Yes  −0.066 Yes 4370 
216 1.80 7.5 68.7 0.093 0.010 Yes −0.013 Yes  −0.083 Yes 230 
217 0.95 7.9 8.2 0.088 0.080 Yes −0.028 Yes  −0.008 No 
218 0.48 12.6 /d 0.055  Yes −0.025  Yes 50 
220 1.00 4.4 /d 0.160  Yes −0.07  Yes 90 
IDGleason scoreBaseline PSA concentrationBaseline PSA doubling time (mo)PSA doubling time at week 48 (mo)Baseline log PSA slopeLog PSA slope at week 48Clinical response at week 24aIntensity of clinical response at week 24aClinical response at week 48Intensity of clinical response at week 48bImmunological responseIntensity of immunological responsec
201 0.91 6.4 43.3 0.109 0.016 Yes −0.172 Yes  −0.093 No 
202 4.99 12.8 16.1 0.054 0.043 Yes −0.017 Yes  −0.011 Yes 210 
203 1.94 6.0 2.7 0.116 0.260 No 0.144 Off Study  Yes 30 
204 0.96 5.9 3.8 0.118 0.184 No 0.039 No  0.066 No 50 
205 4.78 2.9 11.4 0.239 0.061 Yes −0.148 Yes  −0.178 Yes 1180 
206 0.52 5.1 4.2 0.137 0.167 No 0.036 No  0.03 Yes 100 
207 3.53 10.5 12.0 0.066 0.058 Yes −0.002 Yes  −0.008 No 
208 12.00 6.3 8.3 0.110 0.084 Yes −0.054 Yes  −0.026 No 
209 1.33 7.0 −34.7 0.099 −0.020 Yes −0.018 Yes  −0.119 No 
210 13.00 12.2 36.5 0.057 0.019 Yes −0.052 Yes  −0.038 No 
211 1.41 4.1 12.4 0.169 0.056 Yes −0.087 Yes  −0.113 Yes 710 
212 5.88 3.7 4.7 0.187 0.146 Yes −0.002 Yes  −0.041 No 
214 17.30 8.8 23.8 0.079 0.030 Yes −0.028 Yes  −0.049 Yes 4480 
215 9.39 5.5 11.5 0.126 0.060 Yes −0.103 Yes  −0.066 Yes 4370 
216 1.80 7.5 68.7 0.093 0.010 Yes −0.013 Yes  −0.083 Yes 230 
217 0.95 7.9 8.2 0.088 0.080 Yes −0.028 Yes  −0.008 No 
218 0.48 12.6 /d 0.055  Yes −0.025  Yes 50 
220 1.00 4.4 /d 0.160  Yes −0.07  Yes 90 

Abbreviation: ND, not done.

aAbsence or presence of a clinical response was defined as having negative difference in the slope log PSA at either 24 or 48 weeks minus the pretreatment slope log PSA.

bIntensity of clinical response was calculated as the difference in slope of PSA trend over time observed at time of analysis compared with pretreatment value (e.g., log PSA slope at week 24 –log PSA slope before treatment).

cIntensity of immunologic response was calculated as the median # of spots observed through 7 days in vitro stimulation ELISPOT against wild-type 27-35, epitope enhanced 29-37-9V and wild-type 29-37 TARP peptides tested at week 12, 18, and 24.

dNot available.

Notably, while patient baseline parameters were correlated (e.g., Gleason score and prevaccination PSA doubling time were correlated), clinical response was observed independently of prevaccination Gleason score (r2 = 0.0438), baseline PSA doubling time (r2 = 0.0435), or baseline PSA levels (r2 = 0.0121; Supplementary Fig. S1). As expected, clinical response assessed by the decrease in slope log PSA correlated with changes in PSA doubling time (r2 = 0.6827, P = 0.0012), and PSA decline (r2 = 0.4188, P = 0.0067).

Phenotypically all lots of DC products were positive for CD80, CD83, CD86, CD123, CD11c, CD38, CD54, HLA-DR (all > 95%) by flow cytometry (Fig. 1A). The markers showing significant degrees of variability among DC products were CD14 (ranging from 14% to 90% CD14+) and CCR7 (ranging from 5% to 90%). This variability was dependent on both manufacturing and interpatient factors, but only for CD14 the interpatient variability was substantially greater than manufacturing variability (lot-to-lot for the same patient; Fig. 1B). Interestingly, when we analyzed DC preparations for differential expression among those from patients that achieved a decreasing slope log PSA clinical response (RespDC) versus those from patients that did not (NonRespDC), we observed a trend with RespDC expressing higher levels of CCR7 and lower levels of CD14 compared with NonRespDC (not statistically significant). To analyze how CCR7 or CD14 levels were able to discriminate RespDC versus NonRespDC, we used ROC analysis. The underlying assumption of ROC analysis is that a variable under study (e.g., % of CCR7+ DCs) is used to discriminate between two mutually exclusive states (i.e., RespDC vs. NonRespDC). When qualitative clinical responses were evaluated by ROC curves, both factors led to an AUC of 76.3% based on percent of CD14+ cells and of 69.6% based on percent of CCR7+ cells (Fig. 1C).

Figure 1.

Flow cytometry and culture data analysis. A, Flow cytometry analysis of DCs. Histograms of the expression of surface markers CD86, CD83, HLA-DR, CD14, CD80, CD123, CD11c, CD54, CCR7, and CD38 of a representative DC product; B, Coefficients of variation (CV) of % CD14+, %CCR7+, % viable cells, and final DC yields (as a percentage of final number of viable DC compared with total starting number of monocytes) were calculated for manufacturing (light-gray bars) and interpatient variability (black bars) among all manufactured DCs. Manufacturing-related CV was calculated as the average CV registered among all the DCs generated from each patient, whereas interpatient CV was calculated on patients averaged values; C, ROC curves showing the power of %CD14+, %CCR7+, %viable cells, and final DC yields to discriminate among RespDC and NonRespDC. In a ROC curve plot, the “true positive” diagnosis rate (sensitivity) is plotted against the “false positive” diagnosis rate (1−specificity) for a test with a binary outcome. The AUC summarizes the discrimination of the test, that is, its ability to classify cases correctly. A perfect test would have an AUC of 100%; a worthless test would have an AUC of 50%. AUC values may be classified as follows: 90%–100%, excellent; 80%–90%, good; 70%–80%, fair; 60%–70%, poor; 50%–60%, fail (23).

Figure 1.

Flow cytometry and culture data analysis. A, Flow cytometry analysis of DCs. Histograms of the expression of surface markers CD86, CD83, HLA-DR, CD14, CD80, CD123, CD11c, CD54, CCR7, and CD38 of a representative DC product; B, Coefficients of variation (CV) of % CD14+, %CCR7+, % viable cells, and final DC yields (as a percentage of final number of viable DC compared with total starting number of monocytes) were calculated for manufacturing (light-gray bars) and interpatient variability (black bars) among all manufactured DCs. Manufacturing-related CV was calculated as the average CV registered among all the DCs generated from each patient, whereas interpatient CV was calculated on patients averaged values; C, ROC curves showing the power of %CD14+, %CCR7+, %viable cells, and final DC yields to discriminate among RespDC and NonRespDC. In a ROC curve plot, the “true positive” diagnosis rate (sensitivity) is plotted against the “false positive” diagnosis rate (1−specificity) for a test with a binary outcome. The AUC summarizes the discrimination of the test, that is, its ability to classify cases correctly. A perfect test would have an AUC of 100%; a worthless test would have an AUC of 50%. AUC values may be classified as follows: 90%–100%, excellent; 80%–90%, good; 70%–80%, fair; 60%–70%, poor; 50%–60%, fail (23).

Close modal

In addition to phenotypic expression of surface markers, we also analyzed cell culture data and noticed a great variability in final product viability and DC yield (i.e., the percentage of initial cells that were recovered at the end of DC manufacture), respectively, ranging between 37% and 91% and between 6% and 48%. For these factors, the sources of variability were also traced back to both manufacturing and interpatient differences (Fig. 1B). A nonrandom distribution was also observed for these factors between RespDC and NonRespDC, but with a much lower relevance (AUC based on DC viability was 60.8% and the AUC based on DC yield was 61%). Altogether, these data indicate that lot-to-lot variability can be observed in clinical DC products and that interpatient variability might be responsible for phenotypic differences among RespDC and NonRespDC.

DC transcriptomes clustered according to patient

Next, we analyzed gene expression profiles of 99 DC vaccine products derived from the 18 patients who received at least five vaccines using microarray technology. Unsupervised hierarchical clustering analysis grouped the DC products according to patient (Fig. 2A), confirming the prominent role of interpatient variability in affecting DC lot-to-lot variability shown in our previous report (12). In addition, the node analysis of the unsupervised hierarchical clustering did not show the existence of separated subclusters, but rather indicated that the DC products were spread on continuum levels of variability as indicated by the fact that except for a few outlier samples, the vast majority of DC preparations showed similar interpatient distances. Similar observations were obtained using PCA of the whole dataset and through Davies–Bouldin Index testing on partitioning the dataset into defined numbers of groups (not shown). Altogether, these analyses suggested that clinical DC products show interpatient differences that cannot be easily grouped into well-defined phenotypes. In particular, in both clustering analysis and PCA, RespDC were not separated from NonRespDC, pointing to the fact that interpatient variability exceeds any difference between RespDC and NonRespDC. Similar conclusions could be traced by standard class comparison analysis (see Supplementary Results) strongly suggesting that to delineate differences between RespDC and NonRespDC, more complex models must be implemented to analyze interpatient variability in an unbiased manner.

Figure 2.

Gene expression analysis of DCs. DC products (n = 99) were analyzed by gene expression profiling using Agilent microarrays. A, Unsupervised hierarchical clustering of the DCs based on the whole dataset (35,753 genes). Branches are colored according to patient. B, Similarity matrix analysis of the 1,864 genes belonging to the eight modules identified by WGCNA. On the right, magnification of the top right corner of the matrix to show less abundant modules. Similarity matrix is on a white-to-red gradient, where white represents a correlation equal to 0, whereas red is 1. C, Manufacturing and interpatient variability of the expression levels of the eight modules in clinical DC. For each module, the SD of module expression is shown for both interpatient (black bars) and manufacturing variability (gray bars). Manufacturing variability was calculated as the average SD registered among all the DCs generated from each patient, whereas interpatient variability was calculated on patients' averaged values. D, Average expression levels of the eight identified modules in the DC of each patient. The heatmap is shown on a Blue–White–Red Gradient, where blue represents an expression level below the average, white is an average expression level and red represents an expression above the average.

Figure 2.

Gene expression analysis of DCs. DC products (n = 99) were analyzed by gene expression profiling using Agilent microarrays. A, Unsupervised hierarchical clustering of the DCs based on the whole dataset (35,753 genes). Branches are colored according to patient. B, Similarity matrix analysis of the 1,864 genes belonging to the eight modules identified by WGCNA. On the right, magnification of the top right corner of the matrix to show less abundant modules. Similarity matrix is on a white-to-red gradient, where white represents a correlation equal to 0, whereas red is 1. C, Manufacturing and interpatient variability of the expression levels of the eight modules in clinical DC. For each module, the SD of module expression is shown for both interpatient (black bars) and manufacturing variability (gray bars). Manufacturing variability was calculated as the average SD registered among all the DCs generated from each patient, whereas interpatient variability was calculated on patients' averaged values. D, Average expression levels of the eight identified modules in the DC of each patient. The heatmap is shown on a Blue–White–Red Gradient, where blue represents an expression level below the average, white is an average expression level and red represents an expression above the average.

Close modal

Weighted gene coexpression analysis revealed the presence of eight modules in DCs

To characterize the interpatient variability without any a priori assumption, we applied to our dataset the weighted gene coexpression analysis (WGCNA) to identify modules of genes that are coexpressed (i.e., whose expression changes similarly among different samples) and therefore should be strongly representative of interpatient variability (19). WGCNA revealed the existence of eight modules that were differentially expressed among the DC preparations in our dataset (Fig. 2B). Modules were labeled numerically in decreasing order (i.e., Module 1 being the one made of the highest number of genes). Details about modules and the genes of which they are made are in Supplementary File S1. To dissect the characteristics of the eight modules and define whether the modules reflect manufacture-related variability or interpatient variability, we calculated the manufacturing and interpatient SDs for each module. As shown in Fig. 2C, while Modules 1 and 8 clearly showed a low level of interpatient variability, the other modules showed a much higher degree of interpatient variability indicating that differences in the expression levels of these modules exist among patient DCs (Fig. 2C and D). Interestingly, Modules 2, 3, and 7 showed low manufacturing-related variability, suggesting that levels of expression of these modules are mainly affected by manufacturing-unrelated factors (Fig. 2C). On the other hand, Modules 4 and 6 were characterized by manufacturing-related variability levels comparable with interpatient variability, indicating that genes belonging to these modules were more susceptible to manufacturing-related variability.

Low expression of Module 2 genes correlated with clinical and strong immunologic responses

We then analyzed modules for their differential expression among RespDC and NonRespDC. Notably, only Module 2 showed a statistically significant correlation with clinical response (R = 0.5278, P = 0.035, Fig. 3A), therefore suggesting that expression level of genes belonging to Module 2 may play a role in clinical response. In particular, when we analyzed the expression of Module 2 among different patient DC products, we observed that while upregulation of Module 2 led to mixed clinical responses, downregulation of Module 2 was strongly associated with clinical and immunologic responses (χ2P = 0.008829, Fig. 3B). Next, we evaluated Module 2 expression for its predictive value for clinical response and through a ROC curve we obtained an AUC of 85.5% (Fig. 3C). However, when tested as a predictor of strong immunologic response, Module 2 led to an almost perfect prediction with an AUC of 97.9%. All together these data indicate that lower expression of Module 2 was correlated with more potent DC vaccines that resulted in strong immunologic and clinical responses, even though clinical responses were observed even in patients that received DCs expressing high levels of Module 2.

Figure 3.

Module 2 characterization: correlation with response, gene ontology, and network analysis. A, Plot showing the correlation of Module 2 expression in DC products and the quantitative measure of clinical response at week 48 [measured as the decrease in slope (log PSA) over time]; B, Heatmap of the patient-averaged expression level of the genes belonging to Module 2. Each column represents the average value observed among the DCs manufactured from the same patient. Columns are ordered according to the quantitative measure of clinical response at week 48 with nonresponders on left shown by the color bar on the top of the heatmap (green, no response; yellow, mild response; red, strong response). *, DCs of patients showing strong immunologic response (median ELISPOT count >500). Heatmap is colored on a blue–black–yellow gradient. The top gene cluster of the heatmap shows genes more expressed in Resp-DC, whereas the lower cluster shows genes more expressed in NonResp-DC. C, ROC curves showing the ability of Module 2 expression on DCs to discriminate among clinical responders and clinical nonresponders in black and strong immunologic responders versus nonstrong immunologic responders in red. D, GO analysis of genes belonging to Module 2. P value of the enrichment of each Biological Process family is shown. E, The most relevant network of the genes in Module 2 indicated by Ingenuity Pathway Analysis.

Figure 3.

Module 2 characterization: correlation with response, gene ontology, and network analysis. A, Plot showing the correlation of Module 2 expression in DC products and the quantitative measure of clinical response at week 48 [measured as the decrease in slope (log PSA) over time]; B, Heatmap of the patient-averaged expression level of the genes belonging to Module 2. Each column represents the average value observed among the DCs manufactured from the same patient. Columns are ordered according to the quantitative measure of clinical response at week 48 with nonresponders on left shown by the color bar on the top of the heatmap (green, no response; yellow, mild response; red, strong response). *, DCs of patients showing strong immunologic response (median ELISPOT count >500). Heatmap is colored on a blue–black–yellow gradient. The top gene cluster of the heatmap shows genes more expressed in Resp-DC, whereas the lower cluster shows genes more expressed in NonResp-DC. C, ROC curves showing the ability of Module 2 expression on DCs to discriminate among clinical responders and clinical nonresponders in black and strong immunologic responders versus nonstrong immunologic responders in red. D, GO analysis of genes belonging to Module 2. P value of the enrichment of each Biological Process family is shown. E, The most relevant network of the genes in Module 2 indicated by Ingenuity Pathway Analysis.

Close modal

Module 2 was a tolerogenic DC module

To characterize Module 2 genes, we performed a gene ontology (GO) analysis and among the most over-represented “biological process” GO families we observed: inflammatory response, immune response, chemotaxis, and endocytosis (Fig. 3D). In particular, as shown by a network analysis (Fig. 3E), the dominant Module 2 factors were CD14, IL10, thrombospondin, estrogen receptor 1, insulin-like growth factor-binding protein 4, and hepatocyte growth factor (HGF). Most of these genes are known factors driving immune tolerance and specifically the first two are widely described as the major markers of tolerogenic DCs (10, 24–27). To better understand whether Module 2 represents a module of tolerogenic DCs, we performed a meta-analysis of all the publicly available tolerogenic DC gene expression studies and looked at how genes belonging to Module 2 behaved in these other datasets of tolerogenic DC. In total, we found eight gene expression datasets describing in vitro–generated tolerogenic DCs that could be used for the analysis. In these studies, tolerogenic DC were generated according to different protocols using IL10 alone or in combination with other cytokines, mesenchymal stromal cells, T regulatory cells, or adhesion protein disruption. In five of the eight analyzed datasets, we observed a statistically significance overlap of Module 2 overexpression in tolerogenic DCs (Table 2), strengthening the link between the expression of Module 2 genes and tolerogenic DCs.

Table 2.

Meta-analysis of tolerogenic DC dataset

Dataset # (source)DC ComparisonReferencesP value of overlap with Module 2Top 20 genes in common with module 2 (only for significantly overlapping)
GSM468775 (NCBI GEO) IL10/IL6 mDCs vs. mDCs 10 3.56E−06 TGFBI, TIMP1, GLUL, MMP14, ALDH1A1, ABCA1, MMP2, EPAS1, MSR1, CTSC, DOCK4, IPR2, AFB, CD14, SLCO2B1, CCL26, MS4A4A, CYP3A5, PMP22, AGPAT4
GSE23371 (NCBI GEO) IL10/dexamethasone DC vs. LPS DC 52 7.34E−06 CD163, CD14, C3AR1, MS4A6A, RNASE1, S100A8, LAIR1, PDK4, FCGR2A, SLCO2B1, C1QA, TMEM37, MLFML2B, GLUL, STAB1, S100A9, CD300LF, GIMAP1, NPL, C5AR1 
GSE18921 (NCBI GEO) IL10/IL6 DCs vs. standard DCs 10 7.52E−06 TGFBI, TIMP1, GLUL, MMP14, ALDH1A1, ABCA1, MMP2, EPAS1, MSR1, CTSC, ITPR2, MAFB, CD14, SLCO2B1, CCL26, MS4A4A, CYP3A5, PMP22, FST, P2RY1 
MTAB-286 (EMBL-EBI) DCs grown in presence of MSCs vs. normal DCs 53 7.34E−05 CD163, P2RY6, FOLR2, CCR1, SLCO2B1, S100A9, HBB, FCAR, CTSL3, CTSC, ALDH1A1, IL10, GATM, TNFRSF6B, S100A12, S100A8, TAGAP, C1QA, NLN, ITPR2 
GSE18921 (NCBI GEO) IL10 DC vs. standard DC 10 0.005962 TIMP1, TGFBI, GLUL, MAFB, MMP14, EPAS1, ABCA1, FST, AGPAT4, HIP1, CD14, CTSC, PMP22, SLC39A8, CMKLR1, ITPR2, ALDH1A1, ADAMTS2, FOLR2, P2RY1 
GSE7387 (NCBI GEO) Comparison of gene expression data from induced-regulatory T-cell–treated and -untreated DCs from patients with ITP 54 >0.05  
GSE9241 (NCBI GEO) E-cadherin–stimulated DCs vs. bacteria-activated DCs 55 >0.05  
GSE18921 (NCBI GEO) IL10/TGFb1 DC vs. standard DC 10 >0.05  
Dataset # (source)DC ComparisonReferencesP value of overlap with Module 2Top 20 genes in common with module 2 (only for significantly overlapping)
GSM468775 (NCBI GEO) IL10/IL6 mDCs vs. mDCs 10 3.56E−06 TGFBI, TIMP1, GLUL, MMP14, ALDH1A1, ABCA1, MMP2, EPAS1, MSR1, CTSC, DOCK4, IPR2, AFB, CD14, SLCO2B1, CCL26, MS4A4A, CYP3A5, PMP22, AGPAT4
GSE23371 (NCBI GEO) IL10/dexamethasone DC vs. LPS DC 52 7.34E−06 CD163, CD14, C3AR1, MS4A6A, RNASE1, S100A8, LAIR1, PDK4, FCGR2A, SLCO2B1, C1QA, TMEM37, MLFML2B, GLUL, STAB1, S100A9, CD300LF, GIMAP1, NPL, C5AR1 
GSE18921 (NCBI GEO) IL10/IL6 DCs vs. standard DCs 10 7.52E−06 TGFBI, TIMP1, GLUL, MMP14, ALDH1A1, ABCA1, MMP2, EPAS1, MSR1, CTSC, ITPR2, MAFB, CD14, SLCO2B1, CCL26, MS4A4A, CYP3A5, PMP22, FST, P2RY1 
MTAB-286 (EMBL-EBI) DCs grown in presence of MSCs vs. normal DCs 53 7.34E−05 CD163, P2RY6, FOLR2, CCR1, SLCO2B1, S100A9, HBB, FCAR, CTSL3, CTSC, ALDH1A1, IL10, GATM, TNFRSF6B, S100A12, S100A8, TAGAP, C1QA, NLN, ITPR2 
GSE18921 (NCBI GEO) IL10 DC vs. standard DC 10 0.005962 TIMP1, TGFBI, GLUL, MAFB, MMP14, EPAS1, ABCA1, FST, AGPAT4, HIP1, CD14, CTSC, PMP22, SLC39A8, CMKLR1, ITPR2, ALDH1A1, ADAMTS2, FOLR2, P2RY1 
GSE7387 (NCBI GEO) Comparison of gene expression data from induced-regulatory T-cell–treated and -untreated DCs from patients with ITP 54 >0.05  
GSE9241 (NCBI GEO) E-cadherin–stimulated DCs vs. bacteria-activated DCs 55 >0.05  
GSE18921 (NCBI GEO) IL10/TGFb1 DC vs. standard DC 10 >0.05  

Low IL10 concentrations correlated with strong immunologic response

Next, we analyzed media supernatants obtained from the last 6 hours of DC culture (see Materials and Methods) to characterize the cytokine/chemokine secretion profiles of DC immediately before they were injected. We tested 93 supernatants (90 of which corresponding to the same DC we tested by gene expression) by ELISA for the presence of 11 proteins: IFNγ, IL10, IL12p70, IL6, IP10 (CXCL10), MCP1 (CCL2), MIG (CXCL9), TNFα, I-TAC (CXCL11), MDC (CCL22), and TGFβ1. Interestingly, we observed high levels of both manufacturing- and interpatient-related variability for most of the tested proteins, with coefficients of variation ranging between 0.27 and 0.67 for manufacturing-related variability and between 0.33 and 1.34 for interpatient-related variability (Fig. 4A). When tested for their predictive value of clinical response, none of the proteins showed an AUC greater than 80% indicating that single cytokine concentrations in supernatants may not be good predictors of DC efficacy. However, when we tested protein concentrations for their predictive value of strong immunologic response (similarly to what was observed with Module 2 genes) we observed an impressive predictive value for IL10 with an AUC of 95.8% (Fig. 4B). In particular and as expected, minimal levels of IL10 were detected in the supernatants of those DC that led to a strong immunologic response compared with the ones that did not. A similar result was obtained when we tested the IL12/IL10 ratio. In this case, we observed an AUC of 94.3%, with the highest IL12/IL10 ratios leading to strong immunologic responses (Supplementary Fig. S2).

Figure 4.

Cytokine/chemokine secretion profile analysis of DCs. A, Coefficients of variation (CV) of supernatant concentrations of indicated cytokines/chemokines calculated for manufacturing (gray bars) and interpatient variability (black bars) among 93 manufactured DC products. Manufacturing-related CV was calculated as the average CV registered among all the DCs generated from each patient, whereas interpatient CV was calculated on patient-averaged values. B, ROC curves showing the ability of IL10 concentrations measured on DC supernatants to discriminate among clinical and clinical nonresponders in black and strong immunologic responders versus not strong immunologic responders in red. C, Correlation of Module 2 expression with CD14, IL10, MDC, and MCP-1. Each dot represents one DC sample. X, y, and z coordinates represent concentration levels of IL10, MDC, and MCP-1, respectively, size of dots is proportional with percent of CD14+ cells, color represents Module 2 expression level (red–yellow–white gradient, with red being the lowest expression level and white the highest). An interactive 3D plot is available in Supplementary Fig. S3. D, Plot showing the correlation of Module 2 expression with the CD14+/IL10/MCP-1/MDC index. E, ROC curves showing the ability of CD14+/IL10/MCP-1/MDC index to discriminate among clinical responders and nonresponders in black and strong immunologic responders versus not strong immunologic responders in red.

Figure 4.

Cytokine/chemokine secretion profile analysis of DCs. A, Coefficients of variation (CV) of supernatant concentrations of indicated cytokines/chemokines calculated for manufacturing (gray bars) and interpatient variability (black bars) among 93 manufactured DC products. Manufacturing-related CV was calculated as the average CV registered among all the DCs generated from each patient, whereas interpatient CV was calculated on patient-averaged values. B, ROC curves showing the ability of IL10 concentrations measured on DC supernatants to discriminate among clinical and clinical nonresponders in black and strong immunologic responders versus not strong immunologic responders in red. C, Correlation of Module 2 expression with CD14, IL10, MDC, and MCP-1. Each dot represents one DC sample. X, y, and z coordinates represent concentration levels of IL10, MDC, and MCP-1, respectively, size of dots is proportional with percent of CD14+ cells, color represents Module 2 expression level (red–yellow–white gradient, with red being the lowest expression level and white the highest). An interactive 3D plot is available in Supplementary Fig. S3. D, Plot showing the correlation of Module 2 expression with the CD14+/IL10/MCP-1/MDC index. E, ROC curves showing the ability of CD14+/IL10/MCP-1/MDC index to discriminate among clinical responders and nonresponders in black and strong immunologic responders versus not strong immunologic responders in red.

Close modal

Module 2 expression correlated with CD14, IL10, MDC, and MCP-1 secretion

Given that it is not possible to routinely test DC by gene expression profiling, we analyzed how the expression of Module 2 genes correlated with the other analyzed factors that can be tested more easily. As expected, we observed a statistically significant correlation between Module 2 expression levels and percentages of CD14+ DCs assessed by flow cytometry (r = 0.71; P < 0.0001) and IL10 secretion levels (r = 0.604, P < 0.001), confirming at a proteomic level, the observations made on gene expression profiles. Also, Module 2 expression correlated positively with secreted concentrations of MCP-1 (CCL2; r = 0.537, P < 0.0001) and negatively with level of MDC (CCL22; r = −0.534, P < 0.0001). Given that none of these proteins was able to replace Module 2 for its predictive value as single factor, we evaluated whether by combining all four proteins we were able to obtain a better correlation with Module 2. As shown in Fig. 4C, and better in Supplementary Fig. S3, the combination of all four proteins correlated better with Module 2 expression. We, therefore, calculated for each DC product that we analyzed by gene expression, CD14 expression by flow and supernatant analysis by ELISA (n = 89) a CD14/IL10/MCP1/MDC index (made by adding up DC ranks of the expression level of the four proteins, see Materials and Methods for details) and observed that it strongly correlated with Module 2 expression (R = 0.867, P < 0.0001, Fig. 4D). Next we tested its predictive value for both clinical and strong immunologic responses and we obtained AUC of 88.7% and 97.2%, respectively (Fig. 4E). Altogether, these data suggest that the analysis of CD14 expression by flow cytometry combined with IL10, MCP-1 and MDC cytokine concentrations was able to discriminate between RespDC and NonRespDC.

DC-based cell therapies represent a promising approach to activate immune responses against tumors even though the vast majority of clinical trials have failed to show efficacy for such approach. Several reasons for such disappointing results have been identified: suboptimal generation and delivery of DCs, inappropriate selection of immunogenic tumor–associated antigens, systemic inactivation of the immune system in advanced tumors, and the ability of the established tumor microenvironment to inhibit T-cell function. These factors have all been widely described and analyzed as possible justification of poor clinical trial result outcomes (3, 28–31). It has also been recently recognized that response evaluation of immunotherapies, especially cell-based therapies, should be based on different criteria compared with standard chemotherapy drugs and treatments, implying that the previous reports should be careful reevaluated (6).

In this study, we focused our analysis on the DC products administered to stage D0 prostate cancer patients, which allowed eliminating issues related to systemic tumor burden and a local immune-tolerizing microenvironment, and observed a strong correlation between DC phenotype and slope log PSA responses (a well-established surrogate for clinical outcomes in the stage D0 population) and immunologic responses. In particular, we identified a gene signature made up of several well-known tolerogenic DC factors such as CD14 and IL10 that was able to discriminate RespDc from NonRespDC. The differential expression of CD14 and IL10 was confirmed at the proteomic level and observed that MCP-1 and MDC protein levels correlated (directly or inversely, respectively) with the expression level of the tolerogenic gene expression signature. Even though IL10 secretion levels were able to predict strong immunologic responses, it was only by combining CD14, IL10, MCP-1, and MDC measures that it was possible to obtain an index able to replace the tolerogenic gene expression signature in its ability to discriminate both clinical and strong immunologic responses.

Lot-to-lot variability is a critical issue for DC-based immunotherapies. Our data revealed a correlation between the phenotype of DCs used as vaccines and the induction of clinical and immunologic responses. Even though functional analyses are needed to support a causative role of the identified phenotype, our observations further strengthen the need for extensive characterization of cellular products used in preclinical and early-phase clinical trials to identify manufacturing- and interpatient-related issues that may hamper identity, consistency, and potency of final DC products. In a previous report, we described a framework for preclinical analysis of cell therapies for the identification of factors affecting the consistency of cell products (12), but only by using accepted surrogates for clinical outcomes such as slope log PSA used in this study was it possible to correctly determine which factors play a role in product efficacy. The consistency of DC-based products is critical considering that many reports have highlighted how DCs generated from patient monocytes show phenotypic differences compared with those manufactured from healthy donor monocytes (32–35). Therefore, lot-to-lot variability should be carefully characterized for each cell product in the early phases of product development to determine which factors should be analyzed routinely to control and manage product consistency.

The generation of potent DCs capable of inducing a strong antitumor immune response is highly sought after, but a consensus concerning optimal DC generation protocols is still lacking. The changes in slope log PSA and TARP-specific immunogenicity following therapeutic vaccination observed in the current study were encouraging and highly statistically significant (13). Clinical responses were observed in 15 of 18 evaluable patients at 24 weeks and 13 of 16 evaluable patients at 48 weeks, whereas immunologic responses were detected in 10 of 16 evaluable patients. However, and more interestingly, our results suggest that even among DC products manufactured following identical standard operating procedures, it is possible to identify more potent DCs. The detrimental role of tolerogenic signals on DC function, such as expression of CD14 and secretion of IL10, has been widely discussed in literature (8, 10, 27, 36), but the direct involvement of these signals in clinical DC products has not yet been described. DCs generated with our protocol and expressing low levels of tolerogenic genes, as determined by scoring low on our CD14/IL10/MCP-1/MDC index, strongly correlated with the induction of strong immunologic and clinical responses. How to more consistently manufacture DCs with such a potent phenotype is under investigation in our laboratory, but the possibility of predicting which patients are more likely to benefit from vaccination is already an appealing scenario that will be tested further in forthcoming clinical trials at our institution.

DC culture conditions strongly affect DC phenotype: differentiation and maturation signals, their concentrations, and the duration of stimulation strongly shape DC activity (3, 8, 36). In this study, we focused in interpatient and lot-to-lot DC phenotypic differences and identified gene and protein markers correlating with patient response. In particular, a lower expression of tolerogenic features was observed in patient that showed clinical and immunologic response. Many factors may be the cause of the expression of tolerogenic genes and proteins in some DC vaccines: it has shown that individual genetic polymorphisms alter DC response to LPS (37, 38), also patient serum levels of IL10 or other cytokines can alter DC phenotype and activity (39, 40). Therefore, it might be that for those patients whose DC show higher tolerogenic phenotype changes in manufacturing process could result in more potent DC vaccines, such as the use of different maturation signals (e.g., other TLR agonists) or replacement of autologous serum-conditioned media with serum-free ones (41).

Response to DC-based vaccine depends on several factors. In the current study, by analyzing DC administered to patients with relatively low tumor burdens, that is, micrometastatic disease, as the only evidence of disease is PSA biochemical progression, we were able to more directly link DC phenotype with clinical and immune responses. However, it is reasonable to expect that in more complex clinical settings, additional factors related to DC phenotype as well as unrelated factors (e.g., overall patient immune system status following multiple chemotherapies, tumor phenotype, and tumor burden) should also be considered. Therefore, complex data modeling should be developed to be able to extract precious information on DC phenotype and identify factors that correlate with clinical and immunologic outcomes. In our study, despite the clear induction of both clinical (81%) and immunologic response (62.5%), there was no association between the two outcomes. This may depend on multiple factors. First, as pointed out by previous studies (42, 43), clinical response may be the result of a minimal number of TARP-specific T cells (and therefore undetectable by ELISPOT) that with their cytotoxic activity led to the activation of non-TARP–specific T cells that are responsible of the clinical response. Moreover, DCs may have exerted an antitumor effect by activating NK cells (44), B cells (45), or by a bystander activation of T cells (46). On the other side, it has to be considered that factors in the tumor microenvironment, such as programmed death-ligand 1 (PD-L1), CTL-associated protein 4 (CTLA-4), myeloid-derived suppressor cells (MDSC), tumor-associated macrophages, and T regulatory cells, are known to affect T-cell response against tumor (47–49).

In this study, when we used standard statistical tools for the analysis of gene expression data (i.e., t test–based class comparison), we were not able to observe statistical differences among RespDC and NonRespDC. It was only when using a novel unsupervised method for the selection of gene modules which were coexpressed across the dataset (i.e., WGCNA) that we were able to uncover and identify the tolerogenic gene signature. In fact, some of RespDC did express similar levels of the tolerogenic signature of NonRespDC. What is the mechanism behind the ability of these DCs to induce clinical responses in the absence of a conventional immunologic response will be tested in future clinical trials, but considering the multiplicity of effects DCs can exert (50), it is possible that these DCs worked by activating immune cells other than T cells (51) or that T-cell responses in PBMCs did not reflect localized tissue T-cell responses that were more potent but could not be assessed. However, further investigations are needed to explore such hypotheses, including studies that involve expanded patient immunomonitoring and/or systems immunology.

To our knowledge, this is the first study to perform an in-depth characterization of therapeutic DC vaccine products by combining flow cytometry phenotyping with gene expression profiling and cytokine/chemokine secretion profiling. With this approach, we were able to observe a specific DC phenotype that correlated with the postvaccination clinical and immunologic responses observed in this study. Although these data must be validated in a larger cohort of patients, they highlight the importance of extensive characterization of clinical and preclinical DC, as well as other cellular therapies, for the understanding of critical factors that may hinder their identity, consistency, potency, and most important functional efficacy in patients. Once validated, these discoveries might allow the pretreatment identification of patients that will more likely benefit from DC vaccine therapy and/or drive the development of manufacturing protocols that consistently lead to more potent DCs.

M. Terabe is a consultant/advisory board member for Intensity Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L. Castiello, M. Sabatino, J.A. Berzofsky, D.F. Stroncek

Development of methodology: L. Castiello, D.F. Stroncek

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Castiello, M. Terabe, L.V. Wood

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Castiello, J. Ren, J.A. Berzofsky

Writing, review, and/or revision of the manuscript: L. Castiello, M. Sabatino, J. Ren, M. Terabe, L.V. Wood, J.A. Berzofsky, D.F. Stroncek

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Castiello, H. Khuu

Study supervision: M. Sabatino, L.V. Wood, D.F. Stroncek

This research was supported in part by the Intramural Research Program of the NIH, Department of Transfusion Medicine, Clinical Center and Vaccine Branch, Center for Cancer Research, National Cancer Institute.

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.

1.
Ueno
H
,
Schmitt
N
,
Klechevsky
E
,
Pedroza-Gonzalez
A
,
Matsui
T
,
Zurawski
G
, et al
Harnessing human dendritic cell subsets for medicine
.
Immunol Rev
2010
;
234
:
199
212
.
2.
Sallusto
F
,
Lanzavecchia
A
. 
Efficient presentation of soluble antigen by cultured human dendritic cells is maintained by granulocyte/macrophage colony-stimulating factor plus interleukin 4 and downregulated by tumor necrosis factor alpha
.
J Exp Med
1994
;
179
:
1109
18
.
3.
Castiello
L
,
Sabatino
M
,
Jin
P
,
Clayberger
C
,
Marincola
FM
,
Krensky
AM
, et al
Monocyte-derived DC maturation strategies and related pathways: a transcriptional view
.
Cancer Immunol Immunother
2011
;
60
:
457
66
.
4.
Anguille
S
,
Smits
EL
,
Lion
E
,
van Tendeloo
VF
,
Berneman
ZN
. 
Clinical use of dendritic cells for cancer therapy
.
Lancet Oncol
2014
;
15
:
e257
67
.
5.
Stroncek
DF
,
Jin
P
,
Ren
J
,
Feng
J
,
Castiello
L
,
Civini
S
, et al
Quality assessment of cellular therapies: the emerging role of molecular assays
.
Korean J Hematol
2010
;
45
:
14
22
.
6.
Vatsan
RS
,
Bross
PF
,
Liu
K
,
Theoret
M
,
De Claro
AR
,
Lu
J
, et al
Regulation of immunotherapeutic products for cancer and FDA's role in product development and clinical evaluation
.
J Immunother Cancer
2013
;
1
:
5
.
7.
Carreno
BM
,
Becker-Hapak
M
,
Huang
A
,
Chan
M
,
Alyasiry
A
,
Lie
W-R
, et al
IL-12p70–producing patient DC vaccine elicits Tc1-polarized immunity
.
J Clin Invest
2013
;
123
:
3383
94
.
8.
Kalinski
P
,
Urban
J
,
Narang
R
,
Berk
E
,
Wieckowski
E
,
Muthuswamy
R
. 
Dendritic cell-based therapeutic cancer vaccines: what we have and what we need
.
Future Oncol
2009
;
5
:
379
90
.
9.
Vieira
PL
,
de Jong
EC
,
Wierenga
EA
,
Kapsenberg
ML
,
Kaliński
P
. 
Development of Th1-inducing capacity in myeloid dendritic cells requires environmental instruction
.
J Immunol
2000
;
164
:
4507
12
.
10.
Torres-Aguilar
H
,
Aguilar-Ruiz
SR
,
González-Pérez
G
,
Munguía
R
,
Bajaña
S
,
Meraz-Ríos
MA
, et al
Tolerogenic dendritic cells generated with different immunosuppressive cytokines induce antigen-specific anergy and regulatory properties in memory CD4+ T cells
.
J Immunol
2010
;
184
:
1765
75
.
11.
Chitta
S
,
Santambrogio
L
,
Stern
LJ
. 
GMCSF in the absence of other cytokines sustains human dendritic cell precursors with T cell regulatory activity and capacity to differentiate into functional dendritic cells
.
Immunol Lett
2008
;
116
:
41
54
.
12.
Castiello
L
,
Sabatino
M
,
Zhao
Y
,
Tumaini
B
,
Ren
J
,
Ping
J
, et al
Quality controls in cellular immunotherapies: rapid assessment of clinical grade dendritic cells by gene expression profiling
.
Mol Ther
2013
;
21
:
476
84
.
13.
Wood
LV
,
Fojo
A
,
Roberson
BD
,
Hughes
MSB
,
Dahut
W
,
Gulley
JL
, et al
TARP vaccination is associated with slowing in PSA velocity and decreasing tumor growth rates in patients with Stage D0 prostate cancer
.
Oncoimmunology
2016
;
5
:
e1197459
.
14.
Vickers
AJ
,
Brewster
SF
. 
PSA velocity and doubling time in diagnosis and prognosis of prostate cancer
.
Br J Med Surg Urol
2012
;
5
:
162
8
.
15.
Epel
M
,
Carmi
I
,
Soueid-Baumgarten
S
,
Oh
SK
,
Bera
T
,
Pastan
I
, et al
Targeting TARP, a novel breast and prostate tumor-associated antigen, with T cell receptor-like human recombinant antibodies
.
Eur J Immunol
2008
;
38
:
1706
20
.
16.
Robin
X
,
Turck
N
,
Hainard
A
,
Tiberti
N
,
Lisacek
F
,
Sanchez
J-C
, et al
pROC: an open-source package for R and S+ to analyze and compare ROC curves
.
BMC Bioinformatics
2011
;
12
:
77
.
17.
Hanley
JA
,
McNeil
BJ
. 
The meaning and use of the area under a receiver operating characteristic (ROC) curve
.
Radiology
1982
;
143
:
29
36
.
18.
Sorić
B
. 
Statistical “Discoveries” and Effect-Size Estimation
.
J Am Stat Assoc
1989
;
84
:
608
10
.
19.
Langfelder
P
,
Horvath
S
. 
WGCNA: an R package for weighted correlation network analysis
.
BMC Bioinformatics
2008
;
9
:
559
.
20.
Huang
DW
,
Sherman
BT
,
Lempicki
RA
. 
Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources
.
Nat Protoc
2009
;
4
:
44
57
.
21.
Eisen
MB
,
Spellman
PT
,
Brown
PO
,
Botstein
D
. 
Cluster analysis and display of genome-wide expression patterns
.
Proc Natl Acad Sci U S A
1998
;
95
:
14863
8
.
22.
Saldanha
AJ
. 
Java Treeview–extensible visualization of microarray data
.
Bioinformatics
2004
;
20
:
3246
8
.
23.
Edgar
R
,
Domrachev
M
,
Lash
AE
. 
Gene Expression Omnibus: NCBI gene expression and hybridization array data repository
.
Nucleic Acids Res
2002
;
30
:
207
10
.
24.
Gregori
S
,
Tomasoni
D
,
Pacciani
V
,
Scirpoli
M
,
Battaglia
M
,
Magnani
CF
, et al
Differentiation of type 1 T regulatory cells (Tr1) by tolerogenic DC-10 requires the IL-10-dependent ILT4/HLA-G pathway
.
Blood
2010
;
116
:
935
44
.
25.
Rutella
S
,
Bonanno
G
,
Procoli
A
,
Mariotti
A
,
de Ritis
DG
,
Curti
A
, et al
Hepatocyte growth factor favors monocyte differentiation into regulatory interleukin (IL)-10++IL-12low/neg accessory cells with dendritic-cell features
.
Blood
2006
;
108
:
218
27
.
26.
Krispin
A
,
Bledi
Y
,
Atallah
M
,
Trahtemberg
U
,
Verbovetski
I
,
Nahari
E
, et al
Apoptotic cell thrombospondin-1 and heparin-binding domain lead to dendritic-cell phagocytic and tolerizing states
.
Blood
2006
;
108
:
3580
9
.
27.
Pulendran
B
,
Tang
H
,
Manicassamy
S
. 
Programming dendritic cells to induce T(H)2 and tolerogenic responses
.
Nat Immunol
2010
;
11
:
647
55
.
28.
Whiteside
TL
. 
Immune modulation of T-cell and NK (natural killer) cell activities by TEXs (tumour-derived exosomes)
.
Biochem Soc Trans
2013
;
41
:
245
51
.
29.
Whiteside
TL
. 
Immune suppression in cancer: effects on immune cells, mechanisms and future therapeutic intervention
.
Semin Cancer Biol
2006
;
16
:
3
15
.
30.
Vasaturo
A
,
Di Blasio
S
,
Peeters
DG
,
de Koning
CCH
,
de Vries
JM
,
Figdor
CG
, et al
Clinical implications of co-inhibitory molecule expression in the tumor microenvironment for DC vaccination: a game of stop and go
.
Front Immunol
2013
;
4
:
417
.
31.
Hargadon
KM
. 
Tumor-altered dendritic cell function: implications for anti-tumor immunity
.
Front Immunol
2013
;
4
:
192
.
32.
Decker
P
,
Kötter
I
,
Klein
R
,
Berner
B
,
Rammensee
H-G
. 
Monocyte-derived dendritic cells over-express CD86 in patients with systemic lupus erythematosus
.
Rheumatology
2006
;
45
:
1087
95
.
33.
van den Heuvel
MM
,
Vanhee
DD
,
Postmus
PE
,
Hoefsmit
EC
,
Beelen
RH
. 
Functional and phenotypic differences of monocyte-derived dendritic cells from allergic and nonallergic patients
.
J Allergy Clin Immunol
1998
;
101
:
90
5
.
34.
Cuellar
A
,
Santander
SP
,
Thomas
MDC
,
Guzmán
F
,
Gómez
A
,
López
MC
, et al
Monocyte-derived dendritic cells from chagasic patients vs. healthy donors secrete differential levels of IL-10 and IL-12 when stimulated with a protein fragment of Trypanosoma cruzi heat-shock protein-70
.
Immunol Cell Biol
2008
;
86
:
255
60
.
35.
Kvistborg
P
,
Bechmann
CM
,
Pedersen
AW
,
Toh
HC
,
Claesson
MH
,
Zocca
MB
. 
Comparison of monocyte-derived dendritic cells from colorectal cancer patients, non-small-cell-lung-cancer patients and healthy donors
.
Vaccine
2009
;
28
:
542
7
.
36.
Butterfield
LH
. 
Dendritic cells in cancer immunotherapy clinical trials: are we making progress?
Front Immunol
2013
;
4
:
454
.
37.
Wurfel
MM
,
Park
WY
,
Radella
F
,
Ruzinski
J
,
Sandstrom
A
,
Strout
J
, et al
Identification of high and low responders to lipopolysaccharide in normal subjects: an unbiased approach to identify modulators of innate immunity
.
J Immunol
2005
;
175
:
2570
8
.
38.
Peng
JC
,
Abu Bakar
S
,
Richardson
MM
,
Jonsson
JJ
,
Frazer
IH
,
Nielsen
LK
, et al
IL10 and IL12B polymorphisms each influence IL-12p70 secretion by dendritic cells in response to LPS
.
Immunol Cell Biol
2006
;
84
:
227
32
.
39.
Beckebaum
S
,
Zhang
X
,
Chen
X
,
Yu
Z
,
Frilling
A
,
Dworacki
G
, et al
Increased levels of interleukin-10 in serum from patients with hepatocellular carcinoma correlate with profound numerical deficiencies and immature phenotype of circulating dendritic cell subsets
.
Clin Cancer Res
2004
;
10
:
7260
9
.
40.
Bellone
G
,
Carbone
A
,
Smirne
C
,
Scirelli
T
,
Buffolino
A
,
Novarino
A
, et al
Cooperative induction of a tolerogenic dendritic cell phenotype by cytokines secreted by pancreatic carcinoma cells
.
J Immunol
2006
;
177
:
3448
60
.
41.
Fučíková
J
,
Rožková
D
,
Ulčová
H
,
Budinský
V
,
Sochorová
K
,
Pokorná
K
, et al
Poly I: C-activated dendritic cells that were generated in CellGro for use in cancer immunotherapy trials
.
J Transl Med
2011
;
9
:
223
.
42.
Lurquin
C
,
Lethé
B
,
De Plaen
E
,
Corbière
V
,
Théate
I
,
van Baren
N
, et al
Contrasting frequencies of antitumor and anti-vaccine T cells in metastases of a melanoma patient vaccinated with a MAGE tumor antigen
.
J Exp Med
2005
;
201
:
249
57
.
43.
Lonchay
C
,
van der Bruggen
P
,
Connerotte
T
,
Hanagiri
T
,
Coulie
P
,
Colau
D
, et al
Correlation between tumor regression and T cell responses in melanoma patients vaccinated with a MAGE antigen
.
Proc Natl Acad Sci U S A
2004
;
14631
8
.
44.
Van Elssen
CHMJ
,
Oth
T
,
Germeraad
WTV
,
Bos
GMJ
,
Vanderlocht
J
. 
Natural killer cells: the secret weapon in dendritic cell vaccination strategies
.
Clin Cancer Res
2014
;
20
;
1095
103
.
45.
Palucka
K
,
Banchereau
J
. 
Cancer immunotherapy via dendritic cells
.
Nat Rev Cancer
2012
;
12
:
265
77
.
46.
Kamath
AT
,
Sheasby
CE
,
Tough
DF
. 
Dendritic cells and NK cells stimulate bystander T cell activation in response to TLR agonists through secretion of IFN-alpha beta and IFN-gamma
.
J Immunol
2005
;
174
:
767
76
.
47.
Gabrilovich
DI
,
Nagaraj
S
. 
Myeloid-derived suppressor cells as regulators of the immune system
.
Nat Rev Immunol
2009
;
9
:
162
74
.
48.
Chang
C-H
,
Qiu
J
,
O'Sullivan
D
,
Buck
MD
,
Noguchi
T
,
Curtis
JD
, et al
Metabolic competition in the tumor microenvironment is a driver of cancer progression
.
Cell
2015
;
162
:
1229
41
.
49.
Zarour
HM
. 
Reversing T-cell dysfunction and exhaustion in cancer
.
Clin Cancer Res
2016
;
22
:
1856
64
.
50.
Steinman
RM
,
Banchereau
J
. 
Taking dendritic cells into medicine
.
Nature
2007
;
449
:
419
26
.
51.
Bray
SM
,
Vujanovic
L
,
Butterfield
LH
. 
Dendritic cell-based vaccines positively impact natural killer and regulatory T cells in hepatocellular carcinoma patients
.
Clin Dev Immunol
2011
;
2011
:
249281
.
52.
Jansen
BJH
,
Sama
IE
,
Eleveld-Trancikova
D
,
van Hout-Kuijer
MA
,
Jansen
JH
,
Huynen
MA
, et al
MicroRNA genes preferentially expressed in dendritic cells contain sites for conserved transcription factor binding motifs in their promoters
.
BMC Genomics
2011
;
12
:
330
.
53.
Aldinucci
A
,
Rizzetto
L
,
Pieri
L
,
Nosi
D
,
Romagnoli
P
,
Biagioli
T
, et al
Inhibition of immune synapse by altered dendritic cell actin distribution: a new pathway of mesenchymal stem cell immune regulation
.
J Immunol
2010
;
185
:
5102
10
.
54.
Zhang
X-L
,
Peng
J
,
Sun
J-Z
,
Liu
J-J
,
Guo
C-S
,
Wang
Z-G
, et al
De novo induction of platelet-specific CD4(+)CD25(+) regulatory T cells from CD4(+)CD25(-) cells in patients with idiopathic thrombocytopenic purpura
.
Blood
2009
;
113
:
2568
77
.
55.
Jiang
A
,
Bloom
O
,
Ono
S
,
Cui
W
,
Unternaehrer
J
,
Jiang
S
, et al
Disruption of E-cadherin-mediated adhesion induces a functionally distinct pathway of dendritic cell maturation
.
Immunity
2007
;
27
:
610
24
.