Non-small cell lung cancer (NSCLC) is the most common malignancy and the leading cause of cancer death. Chemotherapy with platinum doublets±bevacizumab, a monoclonal antibody (mAb) to the vascular endothelial growth factor (reserved for patients with non-squamous histology and low risk of bleeding only), has been for many years the standard treatment for patients in metastatic (m) and advanced stage of disease (stage IIIB-IV), a good performance status (Eastern Cooperative Oncology Group (ECOG) 0–1) and no druggable driver mutations/rearrangements of epidermal growth factor receptor (EGFR), anaplastic lymphoma kinase/echinoderm microtubule-associated protein-like 4 (ALK/EML4) and ROS proto-oncogene 1 (ROS1).1 2

In the last few years, effector immune checkpoint blockade (ICB) with mAbs to programmed cell death receptor-1 (PD-1; nivolumab and pembrolizumab) and PD-1 ligand-1 (PD-L1; atezolizumab, avelumab and durvalumab) alone or in combination with chemotherapy, radiotherapy or bevacizumab has grown as a solid treatment for these patients with remarkable results in terms of clinical benefit, progression-free survival (PFS), and overall survival (OS) which has been observed in more than 30% of the cases.3–5 On the other hand, these immune-oncological treatments may be associated to more or less severe immune-related adverse events (irAEs) and to high costs. It is, therefore, conceivable that reliable biomarkers predictive of response and irAEs are eagerly awaited.6–8

On this basis, we led a real-world multicenter retrospective investigation aimed to evaluate the predictive role of different human leukocyte antigen (HLA) alleles in patients undergone PD-1 blockade with nivolumab by taking in consideration that: (1) these mAbs promote the rescue of cytotoxic T lymphocytes (CTLs) function silenced by the PD-1/PD-L1 or PD-L2 immune checkpoint at tumor sites,8 (2) HLA molecules play a critical role in triggering CTL-mediated tumor cell killing, T cell priming and clonal expansion,9–12 and finally, (3) there is a strong correlation between expression of specific HLA alleles and development of autoimmune disease.13

Here, we correlated the germline expression of the most frequent HLA alleles in the loci A, B, C and DRB1 (reflecting a β subunit of class II HLA-DR) with outcome and risk of iraEs in patients with mNSCLC undergone to salvage therapy with nivolumab treatment.

This retrospective study was carried out in patients with mNSCLC who received a real-world salvage therapy with nivolumab. The results of the study fulfilled the primary endpoint to investigate the possible correlations among patients’ outcome and irAEs frequency with the germinal expression of multiple class I HLA and DRB1 loci. Variability of class I HLA genes has been related to ICB response in a number of malignancies including melanoma and colorectal cancer.16 17 The association between autoimmunity and antitumoral effect induced by immunotherapy is revealed by IRAEs that more strongly emerge after the treatment with anti-PD-1 and anti-PD-L1 than in response to anti-cytotoxic T-lymphocyte protein 4(CTLA-4) antibody.18 Then, it could be useful to study HLA genetic variation in order to define biomarker of response to ICIs in terms of benefit and adverse events. The rationale of the study is based on the evidence that HLA molecules play a critical role in engaging an antigen-specific immune response acting both on primary cross-priming and peripheral T cell effector response. The results of the present study indeed showed that the PFS of these patients was strongly correlated to their germinal expression of HLA-A*01 and/or A*02 alleles, the most frequent class I HLA-A allele in the Caucasian population and Mediterranean geographical area.19 On the contrary, patients expressing less frequent class I A allele and not A*01 and A*02 showed a very poor prognosis. We also showed that the expression of homozygosis in the locus A (mainly represented by A*01 and A*02 alleles) correlated with their OS confirming the critical role for these molecules in promoting an efficient and prolonged anticancer immune response in patients subjected to PD-1 blockade. In our patient population, however, the best survival was observed in those patients presenting germinal heterozygosis in the locus DRB1 and, therefore, class II HLA-DRB1 molecules, which, differently from class I HLA, are not directly involved in the CTL-mediated killing of tumor cells enforced by PD-1 blockade, but rather to cross priming, T and B cell modulation20 and autoimmunity.21 A recent study showed that no significant correlations were observed for HLA class I zygosity and PFS or OS in three independent cohorts, and tumor genomic characteristics, such as EGFR and ALK alterations, should take in consideration to determine benefit from ICB in NSCLC. Moreover, the authors suggest that decreased expression of MHC class I and antigen presentation through MHC class II should be evaluated in NSCLC to predict response to ICB.22 In this context, our analysis confirmed the previous finding that occurrence of irAEs, AAbs rise, a low baseline inflammatory status and the kind of chemotherapy regimen received prior PD-1 blockade were all correlated to patients’ survival.23 Class II HLA molecules alike HLA-DRB1 bind peptides recognized by T cell receptors (TCRs) on CD4⁺ T cells with helper and/or T regulatory function; thus, they can modulate the differentiation and the proliferation of other lymphocyte subsets including CTLs and B cells.20 We believe that the occurrence of autoimmunity in these patients, in line with the results of other immunological studies, is consequential to the cross priming of antigens released from the tumor site in an immunological context of CTL mediated immune rejection6 24 25 whose cross priming may be facilitated and redirected to the B cell compartment by the germline expression of specific and multiple class II HLA alleles.

All together the results of our study are in line with the knowledge that mAbs to PD-1, such as nivolumab, exert their antitumor effects by reactivating tumor-specific CTLs26 whose existence largely precedes the diagnosis and the immunological treatment and that are consequence of a long-lasting immune response to multiple tumor-associated (TAA) and tumor-specific antigens (TSA).12 27

The ultimate effectors of PD-1 blockade are primed CTLs able to recognize target cells with a specific TCR which, in turn, binds TAA/TSA derived 9–10 mere-peptides (epitopes) bound to homologous class I HLA molecules on their surface. On the other hand, T cells priming and expansion occur within specialized lymphoid organs where T cell precursors use their TCRs to recognize peptide epitopes/class I–II HLA complexes on antigen-presenting cells able to uptake and process antigenic material released in the blood and lymphatic stream by the tumor tissues.9–12 Epitope peptide binding to HLA molecules occurs in the cytoplasm operated by the proteasome system, and it is constrained by definite HLA allele/locus specific amino acid consensus motifs. The latter have been largely described in the literature and specific network algorithms have already been defined to detect epitopic sequences and their potential binding affinity for specific HLA alleles by scanning the amino acid sequence of viral and tumor-derived protein antigens.28–30 A large number of dominant immunogenic epitope peptides from various TAA/TSA and viruses specific for the most frequent class I HLA alleles have already been identified using these specific algorhythms. Many of these peptides have been synthesized and tested in several studies worldwide.31 32 Some of these peptides derived from prostate-specific antigen, carcinoembryonic antigen, human telomerases and thymidilate synthase have also been evaluated in clinical trials of active specific immunotherapy in patients with cancer obtaining contrasting results.31–38 Moreover, the sequence divergence between alleles of HLA-I genotype is related to the increased probability of neoantigen recognition by tumor-infiltrating T cells and subsequently ability of different individual immune system response to checkpoint blockade immunotherapy.39 The main lesson from these studies was that protein antigens present one or more epitope sequence with class I/II HLA allele-specific binding motifs and that both their number and binding affinity for different HLA alleles and TCRs are not uniform. It is consequential that patients with different class I and II HLA germinal haplotypes present divergent immune reactivity to the tumor due to a completely different pool of epitopes available for immune priming, CTL recognition and tumor cell killing even in the presence of the same target antigens. Several studies have already shown that gene and/or upstream downregulation and/or one or more HLA allele and antigen-peptide processing in tumor cells are effective mechanisms of tumor immune escape and resistance to immune checkpoint in patients with cancer.40 41 Another study supports the hypothesis that an allelic variant of killer cell immunoglobulin-like receptor (KIR), KIR3DS1, mainly expressed on natural killer (NK) cells, is associated to resistance to PD-1 blockade in lung cancer.42 The recent results of a large survey by Chowell et al, including a pool of 1535 patients with advanced cancer receiving CTLA-4 and/or PD-1/PD-L1 blockade, showed that germinal homozygosis in at least one of the three class I HLA locus (A, B and C) were predictive of a poor survival.16 These results were in line with our findings concerning heterozygosis in the B, C and DRB1 loci, whereas apparently in contrast with the finding that class I HLA-A locus heterozygosis was correlated to a worse outcome in our series. This difference could be simply explained considering that they do not focus on class I HLA-A and on the other loci separately; additionally, their analysis includes patients with various malignancies with different prognosis and treatment responsiveness and finally, they pool together, patients who received anti-CTLA-4 and anti-PD-1/PD-L1 immune checkpoints inhibitors that act on different phases of the immune response and present a non-homogenous range of activity. Therefore, we performed an additional analysis on the 375 patients with mNSCLC who received PD-1 blockade and included in Chowell’s survey (supplemental NIHMS980063_table_S1)16 finding no outcome differences correlated with class I A, B and C allele heterozygosis. In the same patients’ series, we were unable to extrapolate significant data on the outcome of patients with heterozygosis in the locus A, considering the following: (1) patients with HLA-A locus homozygosis (53 out of 375) showed a twice shorter follow-up; (2) they have not achieved the median survival when the data were collected (before January 2018); (3) the two groups had not received a uniform immunological treatment.

All together these findings suggest that the different response to nivolumab observed in patients with less frequent class I HLA-A alleles might be a consequence of a divergent CTL response, whose entity is dependent on the different amount of high affinity epitope peptides (epitope repertoire) bound to their specific class I HLA profile on the locus A. It is, therefore, possible to speculate that patients with a germline expression of the most frequent class I HLA-A alleles in a specific geographical area (A*02 and/or A*01) also present a more appropriate CTL epitope peptide repertoire aimed to protect them in a specific environment.

If these data should be confirmed on a large series, it is reasonable to propose the combined use of ICB and allele-specific anticancer vaccines in patients who failed to rise a spontaneous tumor antigen-specific immune response. Recent findings concerning the application of PD-1/PD-L1 blockers in combination with front-line chemotherapy in patients with advanced NSCLC compared with the same chemotherapy doublet alone were partially sorrowed by the occurrence of a high risk of early deaths (within the first 60 days of treatment) due to severe immunological disorders. Even though, it is possible to speculate that immune system over-stimulation combining danger signals and massive antigen release consequent to secondary cytotoxic agents (anticancer drugs/radiotherapy/antiangiogenic agents) might be detrimental in patients who already show an inappropriate immune response.43 44 On these bases, we believe that an accurate multimodal definition of tumor immunogenicity at baseline as well as a germline class I and II HLA allele characterization deserves to be studied on a large scale in these patients in a contest of RWE in order to define not only responsive patients but also patients with high risk of lethal irAEs and to propose new treatment modalities combining immune checkpoint inhibitors and class I HLA allele-specific anticancer vaccines.