Melanoma-specific antigen-associated antitumor antibody reactivity as an immune-related biomarker for targeted immunotherapies.

Background: Immunotherapies, including cancer vaccines and immune checkpoint inhibitors have transformed the management of many cancers. However, a large number of patients show resistance to these immunotherapies and current research has provided limited findings for predicting response to precision immunotherapy treatments.
Methods: Here, we applied the next generation phage display mimotope variation analysis (MVA) to profile antibody response and dissect the role of humoral immunity in targeted cancer therapies, namely anti-tumor dendritic cell vaccine (MelCancerVac®) and immunotherapy with anti-PD-1 monoclonal antibodies (pembrolizumab).
Results: Analysis of the antibody immune response led to the characterization of epitopes that were linked to melanoma-associated and cancer-testis antigens (CTA) whose antibody response was induced upon MelCancerVac® treatments of lung cancer. Several of these epitopes aligned to antigens with strong immune response in patients with unresectable metastatic melanoma receiving anti-PD-1 therapy. Conclusions: This study provides insights into the differences and similarities in tumor-specific immunogenicity related to targeted immune treatments. The antibody epitopes as biomarkers reflect melanoma-associated features of immune response, and also provide insights into the molecular pathways contributing to the pathogenesis of cancer. Concluding, antibody epitope response can be useful in predicting anti-cancer immunity elicited by immunotherapy.

Introduction

Knowledge of the immunosuppressive tumor microenvironment has markedly improved within the last decade (reviewed in ref. [1]). To achieve immunogenicity, tumor cells must express antigens capable of eliciting immune activation. The identification of applicable tumor antigens is indispensable for the development of effective cancer immunotherapy. Most known tumor antigens are considered canonical if derived from protein-coding regions in contrast to noncanonical antigens that include sequences outside protein-coding regions or that are generated by antigen-processing[2]. Melanoma cells are considered highly immunogenic with well-described tumor-associated antigens (TAAs)[3], including cancer-testis antigens (CTAs)[4] and neo-antigens carrying novel epitopes of self-antigens[5]. Some well-known examples include carcinoembryonic antigen (CEA), B melanoma antigen 1 (BAGE), G antigens (GAGEs), cancer/testis antigen 1 (CTAG1; also known as NY-ESO1), and melanoma-associated antigens (MAGEs) (Rev in ref. [6]). The antigenic repertoire is a critical factor for immunosurveillance and cancer progression[7]. However, most studies have focused on the role of T cells in these battles[8], while considerably less is known about B cell response[9]. Humoral response against cross-reactive autoantigens has been detected in different cancers[10]. A burst of recent publications is pointing to the role of antibodies contributing to tumor control[11] as cancer-associated autoimmunity targeting non-malignant tissues may reflect favorable disease outcome[12]. On the other hand, the reasons underlying the immunogenicity of the tumor, or the lack of it, are not well understood[13]. The antitumor immunity can result from many factors including MHC genetic variation, tumor mutational load, tissue microenvironment[13], but also by cell stress, reactivation of embryonic or gonadal transcription, epigenetic instability, aberrant RNA splicing, and others[14,15]. For example, it is argued that the capture of either apoptotic or necrotic cancer cells by macrophages and dendritic cells in the tumor microenvironment may lead to immune suppression or stimulate inflammatory pathways contributing to antitumor cytotoxicity[16].

Discoveries in cancer biology have led to new strategies in awakening tumor immunogenicity, including checkpoint blockade, adoptive cellular therapy, and cancer vaccines, underscoring the role of the immune system in waging the war on cancer tissue. Among these are monoclonal antibodies that target cancer immune checkpoint inhibitors (ICIs) including anti-CTLA-4, anti-PD-1, and anti-PD-L1/2 antibodies that are able to restore anticancer immunity and are widely used for the management of various cancers, including melanoma[17]. Immunogenicity of CTAs has led to the use of melanoma-associated antigens as promising candidates for novel cancer treatments[18,19]. In addition to monoclonal antibodies, cancer vaccines, in particular those based on dendritic cells (DCs) as vectors for antigen delivery, are a major focus of current developments[20]. To date, personalized neoantigen-based DC vaccines are evolving and have shown clinical success in melanoma and other solid tumors[21].

Biomarkers associated with clinical prognosis of the cancer and/or severe immune-related adverse effects (irAEs) of the drugs are areas of active investigation. Different biomarkers have been tackled with variable success, such as levels of PD-L1[22], genetic mutations[23], inflammatory cytokines[24], and the presence of tumor-infiltrating lymphocytes (reviewed in ref. [25]). Tumor infiltrating B lymphocytes contribute to anti-tumor immunity by promoting antibody response to tumor antigens[26,27]. High titer antibodies against melanoma differentiation antigens (TRP1/TYRP1, TRP2/TYRP2, gp100, MelanA/MART1) were observed in responder group of melanoma patients treated with ICI mAbs (monotherapies with Nivolumab, Pembrolizumab or Ipilimumab, or the combination of Nivolumab and Ipilimumab)[28,29]. However, pre-treatment autoantibody profiles in melanoma patients were reported to predict ICI treatment-associated toxicity[30]. Connectedly, DC vaccines also stimulate robust antibody response[31-33] and in some cases, this is associated with prolonged recurrence-free survival[32]. Despite big hopes, clinical benefit of immunotherapies has remained limited only to a subset of patients[34,35] and it is currently undetermined whether increase or decrease in immune response to specific tumor antigens is beneficial to the patient[36,37].

Here, we explore the use of a high precision approach called mimotope variation analysis (MVA), a next generation random peptide phage display method to delineate cancer therapy-associated antibody immune response at epitope resolution. We hypothesize that the pre-existing and treatment-induced antibodies against specific antigen targets could reflect the response elicited by anti-tumor drug and that this response could be predictive of cancer immunogenicity and thus, sensitivity to immune therapy. We generate data to test this hypothesis by immunoprofiling analysis of the anti-melanoma antibody response in the sera samples from the phase II clinical trial of patients with non-small cell lung cancer (NSCLC) receiving autologous DC therapy based on allogenic melanoma cell lysate (MelCancerVac®)[38,39]. We correlate the findings on melanoma-specific antigen profiles with those from a group of patients with unresectable metastatic melanoma receiving anti-PD1 (pembrolizumab) treatment as a part of their standard-of-care. We verify the melanoma-antigen specificity using MVA-based competition, and further determine a three-epitope biomarker signature of melanoma-specific antibody response elicited by both immunotherapies. Our results demonstrate the relevance of antibody epitope profiling to better understand the fine line separating beneficial immunosurveillance from harmful autoimmunity in the anticancer immune response elicited by different types of therapy.

Methods

Study population

The present study analyzed samples from a total of 119 individuals from 2 different clinical cohorts of NSCLC and melanoma patients and their appropriate controls, whose clinical characteristics are shown in Table 1 and Supplementary Table 1. The study was conducted in accordance with the guiding principles of the Declaration of Helsinki and the study participants gave informed consent before enrollment.

Table 1

Description of clinical cohorts.

Cohort	Cohort (n = 119 individuals)
Sub-cohort	NSCLC sub-cohort			Melanoma sub-cohort
	(n = 34)			(n = 85)
Group	CTRL-NSCLC	NSCLC	MelVac-CTRL /MelVac	CTRL-Mel	PEM-Mel
Individuals	Controls without cancer (n = 10)a	NSCLC without MelCancerVac® therapy (n = 18)b	NSCLC with MelCancerVac® therapy (n = 6)c	Healthy controls (n = 80)a	Melanoma patients with pembrolizumab therapy (n = 5)a
Age (mean ± SD)	65.3 ± 8.4	59.2 ± 7.9	55.7 ± 8.4	38.5 ± 10.7	67.6 ± 9.2
Gender (M/F/NA)	5/5/0	7/8/3	4/2/0	42/38/0	2/3/0
Samples	∑ samples = 130

NSCLC – non-small cell lung cancer patients; CTRL-NSCLC – non-cancer controls for NSCLC group; MelVac – NSCLC patients who received MelCancerVac® vaccine; MelVac-CTRL – paired samples of MelVac group taken before vaccination; CTRL-Mel – healthy controls for melanoma group; PEM-Mel – melanoma patients receiving pembrolizumab treatment; a – 1 sample per person available to researchers; b – 1 sample per person available to researchers, except for 3 patients (NSCLC1, NSCLC2, and NSCLC7) who had 2 samples available; c – 1 pre- and 1 post-vaccination sample of the patient available to researchers, except for one patient with 1 pre- and 3 post-vaccination samples. F – female, M – male, n – number of individuals; NA – not available.

The NSCLC patient cohort (n = 24) included longitudinal study of patients diagnosed with advanced NSCLC, who participated in the phase II clinical trial evaluating the effectiveness of MelCancerVac® vaccine[38,39] (Supplementary Table 2). The clinical trial, completed at the time of this study, was designed and carried out by Dandrit Biotech A/S and approved by European Medicines Agency (https://www.clinicaltrialsregister.eu/ctr-search/trial/2006-002202-54/DK). Out of the 24 study participants, 6 NSCLC patients donated blood samples before vaccination (group: MelVac-CTRL) and after receiving MelCancerVac® (group: MelVac), while 18 NSCLC patients had not received any doses of the vaccine at the time of sample donation (group: NSCLC).

The melanoma group comprised of patients with unresectable and metastatic melanoma (n = 5, ICD-10: C43; group: PEM-Mel), who received KEYTRUDA® (anti-PD-1 monoclonal antibody pembrolizumab, Schering-Plough Labo NV) immunotherapy as a part of standard-of-care. Serum samples of melanoma patients were collected 3 weeks after the first immunotherapy treatment, when patients came to receive the second dose (European Medicines Agency guidelines for KEYTRUDA therapy) and were provided by EGeen International (Mountain View CA, USA; ethical permit: 236/T-5).

Control groups included subjects with no history of cancer (n = 10, group: CTRL-NSCLC), with approvals for recruitment to the study from the Ethics Committee of the University Hospital of Liège (permit: 2018/77), and healthy blood donors (n = 80, ICD-10: Z52.0; group: CTRL-Mel) from the Blood Center of North Estonia Medical Center with the approval of the Ethics Review Committee on Human Research of the National Institute for Health Development, Estonia (permit: 1045).

Mimotope variation analysis (MVA)

MVA, the next generation phage display method was used to determine individual immunoprofiles reflecting antibody repertoires for the study cohort[40,41]. Two µl of serum or plasma, previously precleared to plastic and E. coli/wt M13 phage particles, was incubated with 5 µl of phage library (~5 × 1011 phage particles, derivative of Ph.D.-12, NEB, UK) overnight at +4 °C. The human immunoglobulin G (IgG)-captured phages were pulled down by protein G-coated magnetic beads (NEB, S1506S). IgG-bound phage DNA was extracted and samples were barcoded and sequences amplified by PCR. Pooled samples were analyzed by Illumina sequencing (50 bp single end read, Brigham Young University DNA Sequencing Center, Utah, USA).

MVA with DDM-1.7 cell line lysate competition

MelCancerVac® (DanDrit Biotech, Denmark/Enochian Biosciences, USA) is a therapeutic cellular vaccine based on autologous dendritic cells pulsed with the lysate of allogeneic melanoma cells (DDM-1.7) expressing several tumor antigens, including melanoma-associated antigens[42]. In MVA competition assay, freshly produced lysate from DDM-1.7 melanoma cells (Cellin Technologies, Estonia) was used to pre-block the study samples before MVA assay. Briefly, 30 µl of cell lysate (3 mg/ml) was incubated with 2 µl of serum or plasma before overnight incubation with the phage library and MVA was conducted as described.

Data analysis and peptide antigen clustering

Data were processed with peptide data sets cleaned of sequencing errors and known artefacts, and counts normalized to 3 million reads[40,41]. Final dataset of 12-mer peptides consisted on average of 3.26 × 106 peptide sequences (5.8 × 105 unique) per sample, with a combined total of ~4.2 × 108 peptide sequences. SPEXS2 exhaustive pattern search algorithm[40,41] was used to group similar peptides and reveal enriched recognition patterns (epitopes) in the studied peptide sets (Supplementary Fig. 1a). Each sample was analyzed separately for identification of sample-specific epitopes that had ≥4 fixed amino acid positions. For data analysis of MelVac samples, the identification of epitopes was performed in a discriminative manner, where peptide sets from MelVac-CTRL and MelVac samples of the same patient were compared to each other. Epitopes that represented peptides that were at least 2-fold more enriched in the query sample (MelVac) as compared to paired sample peptide set (MelVac-CTRL) and with a hypergeometric p-value < 1 × 10−8 were selected for further analysis. For melanoma cohort (n = 5, PEM-Mel) the identification of epitopes was performed as non-discriminatory, where patient-specific epitopes were identified in comparison to a random-generated peptide set. Epitopes that represented peptides that were 10-fold more enriched in the query (PEM-Mel) than randomly generated reference peptide set and had a hypergeometric p-value < 1 × 10−8, were selected for further analysis. Altogether 54,055 core epitopes for melanoma and 18,021 epitopes for MelVac groups were selected, representing a dataset of melanoma-specific antibody immune response. In addition, pairwise comparison of MelVac-CTRL and MelVac sample datasets generated 17,690 pre-treatment-specific core epitopes.

Sequence alignment

The set of melanoma-associated antigens used in sequence alignment were chosen from Weinert et al., 2009 data describing genes expressed in the DDM-1.7 melanoma cells[42] (Supplementary Fig. 1b). Sequences of the epitopes of the antigens were downloaded from Immune Epitope Database (IEDB[43], date accessed: 24.09.2020, www.iedb.org). Altogether, the IEDB database contained 2234 epitopes of 102 proteins expressed in the melanoma cell lysate DDM-1.7[42]. All antigen alignments were conducted using custom Excel VBA scripts.

For sequence similarity analysis, 2234 linear IEDB epitopes were exactly aligned with 54,055 melanoma and 18,021 vaccination-specific epitopes generated with SPEXS2. Thirty-five database entries (altogether 34 unique proteins) with sequence identity to at least 1 epitope from both melanoma and vaccination-specific epitope sets were recruited for further antigen-specific analysis. Primary protein sequences were downloaded from UniProtKB database[44] using accession codes matching IEDB epitope entry names (date accessed: 09.10.2020, www.uniprot.org). These 35 protein sequences were aligned with 54,055 melanoma, 18,021 vaccination-specific, and 17,690 pre-vaccination-specific epitopes, with the criteria that every fixed amino acid from SPEXS2-determined epitopes was to match with the protein sequence. Out of these, altogether 8562 epitopes aligned to sequences of 35 melanoma-associated antigens.

ELISA

Human cytomegalovirus (CMV) and Epstein-Barr virus (EBV) serostatuses were measured from blood samples with ISO-17025 accredited methods. In brief, serological analyses were performed with anti-CMV ELISA (IgG) method (EUROIMMUN EI 2570–9601G) and with anti-EBV-CA ELISA (IgG) method (EUROIMMUN EI 2791–9601G) according to the manufacturer’s specifications. Absorbance was measured at 450 nm with SpectraMax Paradigm (Molecular Devices). For CMV serology, 41 samples tested positive, 13 negative and 2 samples were borderline and therefore excluded from further correlation analyses. For EBV serology, all measured samples were conclusive: 35 tested positive, 3 samples were negative.

Statistics and reproducibility

The study included 119 independent study subjects. Samples donated at different time points were considered as paired samples of the individual (n = 130). Technical replicates are defined as the same sample profiled in independent MVA experiments. No randomization or blinding to sample characteristics was conducted, samples were divided into groups based on clinically relevant diagnoses. Group-wise comparisons of median values were visualized using violin- or boxplots with individual data points, and statistical significance is shown where applicable. To evaluate the reproducibility of MVA data, the values of peptide abundance in two technical replicates were compared using Pearson’s correlation coefficient analysis (R package “ggpubr”) and the correlation value between replicates was established as R = 0.95 (P < 0.0001). Other samples were not measured repeatedly.

Statistical analysis

Statistical analyses were conducted with R statistical programming language v.4.0.4 and RStudio environment v.1.4.1106[45,46]. Data were analyzed, graphs were produced and visualized using R packages “reshape2”, “tidyverse”, “precrec”, “ggpubr”, “ggsci”, “scales”, “patchwork”, “egg”, “ggalt” 2021 versions[45-58].

Cosine similarity indices (CSIs) for sample comparisons based on top 2500 peptide abundance values and composition were calculated with the cosine function in R package “lsa”[59].

Top 50 immunodominant characteristics were defined from group-specific epitopes generated in SPEXS2 analysis. For post- (Vac, n = 6) or pre- (Pre, n = 6) vaccination samples the abundance of group-specific epitopes (18,021 for Vac and 17,690 for Pre, respectively) were calculated as the number of IgG-bound peptides containing the epitope sequence in the sample. The 50 epitopes with the highest abundance values were selected for analysis. Z-scores for the comparison of antibody response to top 50 immunodominant characteristics were calculated individually for each patient. First, the mean of top epitope abundance values across both Pre and Vac samples was calculated, then the mean was subtracted from the value of each epitope (mean centered) and the result divided by the standard deviation (autoscaled). For graphical presentation the values are capped off at the 97.5th percentile value of each patient.

Boxplots were generated using the style of Tukey with R packages “ggpubr” or “ggplot2”[47,48]. In figures the upper, middle and lower boxplot lines represent the 75th, 50th, and 25th percentiles, while whiskers represent the largest or smallest value within 1.5 times interquartile range above the 75th percentile or below the 25th percentile, respectively. The p-values of two-sided Wilcoxon Rank Sum test were visualized with “ggpubr” or “ggplot2” packages[47,48].

Wilcoxon Rank Sum test (with continuity correction, base R “stats” package[46]) was used to assess the group-differentiating features of 8562 unique epitopes aligning to melanoma-associated antigens, while custom Excel VBA script was used to determine the sensitivity and specificity while maximizing Youden’s index for each biomarker. MedCalc® Statistical Software (v.19.7.2, www.medcalc.org; 2021) was used to conduct logistic regression and ROC analysis of 15 epitopes as a combinational test.