Targeting the microbiota in melanoma anti-PD1 therapy
The composition of the gut microbiome determines the efficacy of cancer therapy by modulating the anti-tumor immune response through the training of infiltrating myeloid and antigen-presenting cells in the tumor. This has been shown in murine models, in which mice with distinct gut microbiota profiles exhibited differential tumor growth and differences in response to PD-1 blockade. These differences could be eliminated by cohousing of animals, indicating it could be transmitted. Fecal microbiota transplant (FMT) together with anti-PD-1 therapy resulted in nearly full tumor rejection in mice with melanoma.
The influence of the microbiome on the effect of immune checkpoint blockade has also been shown in several clinical studies. In one study of melanoma patients treated with PD-1 inhibitors, significant differences were observed in the gut microbiome between responders and non-responders, with responders having higher diversity and more Ruminococcaceae bacteria [1]. These patients also had enhanced systemic and antitumor immunity. In another study, a significant association was observed between microbiome composition and clinical response to PD-1 blockade in melanoma patients, with Bifidobacterium longum, Collinsella aerofaciens, and Enterococcus faecium all more abundant in responders [2].
One issue is that different studies have identified a wide variety of different bacterial species that are associated with response. In a meta-analysis of several studies, metagenomics identified Ruminococcaceae, Lachnospiraceae, and Bifidobacteriaceae as being associated with a response to anti-PD-1 treatment, while Bacteroidaceae were generally associated with a lack of response.
The potential role of FMT is being assessed in a phase II trial at the University of Pittsburgh, in which fecal samples obtained from long-term PD-1 responders is combined with additional anti-PD-1 treatment in melanoma patients who previously failed to respond to PD-1 blockade [3]. To date, 16 anti-PD-1 refractory patients have received FMT from PD-1 responders, with one complete response, two partial responses and three with stable disease. Consistent with previous observations, responders tended to have higher frequency of Ruminococcaceae, Lachnospiraceae, and Bifidobacteriaceae, while Bacteroidaceae were more frequent in patients with disease progression. FMT induced a rapid and persistent alteration and instability in the microbiome composition, although each patient generally maintained a distinct microbiome based on their existing taxa before receiving FMT. The majority of taxa that were present in the donor but not the recipient colonized the donor gut and were persistent unless the patient was treated with antibiotics. Overall, the microbiota composition after FMT reflected colonization with the donor-specific taxa but perturbation of the microbiome resulted in altered abundance of different taxa of both donor and recipient origin.
FMT in anti-PD-1 refractory melanoma is most likely to induce a response in patients with the immunological potential to respond but with an unfavorable microbiota that can be corrected. However, anti-PD-1 refractory patients may fail FMT for various reasons. These may include the absence of an adequate immunological response regardless of microbiota composition, the FMT lacking the taxa needed to improve anti-PD-1 response, or the FMT failing to induce a perturbation of the microbiome that favors a response, due either to technical reasons or possibly because of incompatibility between donor and recipient microbiome.
Prediction of response to checkpoint inhibition: is there a simple but not simplistic way?
A subset of patients with metastatic melanoma have durable responses to immunotherapy, while others develop potentially serious immune-related adverse events. Reliable biomarkers that can predict response to immune checkpoint inhibition are needed but remain elusive. PD-L1 expression, and TMB are used in the clinic but have limitations, as do other baseline characteristics that have been proposed, e.g. lactate dehydrogenase (LDH) and ECOG performance status. Active areas of research to predict immune checkpoint inhibitor response include biomarkers in the blood and microbiome, genomic profiling of the T cell regulome, auto-antibody signatures for immune-related toxicity and microRNA (miRNA) profiling.
Another possible approach is to integrate machine learning technology on histology specimens with clinical data to predict immune checkpoint inhibitor response. Previously, our group developed a deep convolutional neural network pipeline that could discriminate between malignant and normal lung tissue. In addition, the network was trained to accurately predict the most frequently mutated genes in lung tumors including STK11, EGFR, FAT1, SETBP1, KRAS and TP53 [4]. This machine learning framework was then adapted to whole slide image analysis of tissue from patients with metastatic melanoma who had lymph node and/or subcutaneous tissue resected before first-line anti-CTLA-4 and/or anti-PD-1 therapy [5].
Google Inception v3 was used as a foundation architecture for computational image analysis. Using hematoxylin and eosin-stained slides, a neural network segmentation classifier was trained to identify the tumor from the surrounding microenvironment. Because the dataset included lymph node and subcutaneous tissue sections, the classifier was trained to identify lymphocyte clusters and connective tissue in addition to tumor compartments. This segmentation classifier distinguished regions of interest with high accuracy. A response classifier was then trained to identify whether patients responded to checkpoint blockade or were resistant and progressed. A logistic regression classifier that combined neural network output with clinical and demographic variables (ECOG performance status) augmented the prediction accuracy. Class activation mapping revealed that cell nuclei played an important role in the decision to classify samples as progressive disease or response, with more and larger nuclei associated with a prediction of progression. Based on these findings, machine learning on metastatic melanoma tissue histology shows potential for predicting immunotherapy response, especially if integrated with clinical data.
Another approach is to assess the relationship between body mass index (BMI) and outcomes. Studies have suggested a link between BMI and response to checkpoint blockade, including the counterintuitive phenomenon in which a high BMI seems to confer a survival benefit [6]. However, there are substantial discordances in these data. For example, we found that patients who were overweight or obese has similar progression-free survival (PFS) as patients with normal BMI [7]. However, there was a moderate but non-significant association between higher BMI and better PFS in patients receiving first-line treatment. In addition, a significant survival benefit was observed in overweight and obese patients receiving combination immunotherapy, whereas this was not seen in patients who received anti-PD-1 or anti-CTLA-4 monotherapy.
One possibility is that static measurements do not consider changes in body weight and nutritional intake over time. Moreover, in some patients a lower BMI might be associated with disease progression. To investigate this, we tested the association between BMI changes before the start of immunotherapy and treatment outcomes in patients with melanoma, lung cancer or other cancers [8]. A pretreatment decrease in BMI and low baseline prognostic nutritional index were associated with worse outcomes, including PFS and OS. However, baseline BMI category was not significantly associated with any treatment outcomes, indicating that dynamic BMI changes rather than static assessments correlate with treatment response.
Immunotherapy-induced anti-cancer responses
The anti-cancer immune response involves B and T cell responses against the cancer surfaceome, which includes non-mutated shared antigens, mutated epitopes, and the extracellular domains of transmembrane proteins, as well as intracellular proteins released from tumor cells. There is considerable evidence that shared antigens are important, and genes that are upregulated or amplified in cancer and are associated with worse survival outcomes are potential therapeutic targets. An example of an overexpressed normal gene is vestigial-like (VGLL)-1 that is a cancer-placenta antigen. VGLL-1-specific cytotoxic T lymphocytes (CTLs) can recognize and kill human leukocyte antigen (HLA)-matched allogeneic tumor cell lines derived from different cancers in an antigen-specific manner, indicating that VGLL-1 may constitute an immunotherapeutic target in multiple cancer types [9].
Peptides that are presented on tumor human leukocyte antigen (HLA) molecules are mostly derived from short-lived proteins (SLiPs) and defective ribosomal products (DRiPs) bound to HLA and transported to the cell surface. However, these rapidly degraded products are less available for cross-presentation by antigen-presenting cells and as such are not typical targets of the immune system.
Blocking proteasomal degradation leads to stabilization of DRiPs and SLiPs and the formation of autophagosomes that contain not only DRiPs/SLiPs, but also damage-associated molecular patterns (DAMPs) and chaperone molecules that facilitate cross-presentation. These autophagosomes can be harvested by membrane disruption and fractionation to create the DRibbles vaccine product. The first allogeneic human DRibbles vaccine, DPV-001, was derived from autophagosome products of two non-small-cell lung cancer (NSCLC) cell lines, one of mixed histology and one from an adenocarcinoma [10]. DPV-001 contains multiple toll-like receptor agonists and > 130 potential NSCLC antigens. In a phase II trial, patients with stage III NSCLC received cyclophosphamide induction therapy, before being randomized to DPV-001 alone, or in combination with granulocyte–macrophage colony-stimulating factor or imiquimod. Patients receiving DPV-001 had a significant increase in total (CD4 and CD8) T cells versus controls and the increase in CD4 T cells was similar to that seen in patients receiving ipilimumab.
Antibody responses to over-expressed antigens were detected as ‘waves’ with possible co-coordination of B cell and T cell responses after vaccination. T cell contraction, a natural component of the T cell response to antigens, may be responsible for this anomaly. Co-stimulation with T cell agonists, such as OX40, GITR or 4-1BB, may augment vaccine-induced T cell expansion, maintenance, and function. In preclinical models, DRibbles vaccine and anti-OX40 co-stimulation led to tumor regression and improved survival in a breast cancer murine model [11]. Similarly, anti-GITR and anti-PD-1 antibodies in combination with DRibbles vaccine resulted in better survival in a pancreatic cancer model [12]. Three triplet immunotherapy trials are now ongoing; a phase 1b trial of multivalent autophagosome vaccine with or without OX40 antagonist with nivolumab in patients with triple-negative breast cancer (NCT02737475), a phase 1b trial of multivalent autophagosome vaccine with or without GITR antagonist with anti-PD-1 in patients with head or neck squamous cell carcinoma (NCT04470024), and a trial of neoadjuvant/adjuvant GVAX pancreas vaccine with or without nivolumab and urelumab in patients with surgically resectable pancreatic cancer (NCT02451982).
The role of CD39 in melanoma
The ATP-adenosine pathway is a key modulator of innate and adaptive immunity within the tumor microenvironment (TME). CD39 is the rate-limiting enzyme in the conversion of ATP to immunomodulatory adenosine. Extracellular adenosine in the TME favors escape from antitumor immunity and tumor progression. CD39 and adenosine receptors are upregulated in response to stimuli such as hypoxia, tissue damage and remodeling, and chronic inflammation. High levels of CD39 have been reported in various tumors and CD39 as a therapeutic target has been an active area of investigation.
CD39 blockade with the broad ectonucleotidase inhibitor sodium polyoxotungstate (POM-1) has shown improved antitumor immunity and decreased metastatic burden in preclinical models. However, concerns over its lack of specificity, potential toxicity and limited therapeutic half-life have hindered its clinical development. More recently, anti-CD39 antibodies have been generated. A novel anti-mouse CD39 antibody, which specifically binds to CD39-expressing cells and potently inhibits CD39 ATPase activity in vitro has demonstrated potent activity against MC38 colon adenocarcinoma tumors [13]. Anti-CD39 monotherapy was as potent as anti-PD-1 antibody and more effective than anti-CD73 antibodies and adenosine A2 receptor antagonists in this model. Anti-CD39 was also shown to sensitize anti-PD-1 resistant tumors by increasing CD8+ T cell infiltration. Inhibition of CD39 enzymatic function led to an accumulation of extracellular ATP, which in turn led to an activation of myeloid cells via ATP receptor P2X7. The antitumor activity of anti-CD39 required CD39 and P2X7 co-expression on intratumor myeloid subsets and active interleukin (IL)-18 release to facilitate expansion of intratumor effector T cells.
Targeting CD39 also suppresses experimental lung carcinoma metastases. The antimetastatic activity of anti-CD39 was NK cell and interferon (IFN)-γ dependent, and anti-CD39 enhanced IFN-γ production and CD107a expression in lung-infiltrating NK cells following tumor challenge [14]. Efficacy of anti-CD39 required enzymatic blockade but not FcR engagement. Anti-metastatic activity was dependent on the P2X7-NLRP3 inflammasome pathway. Anti-CD39 also combined with other NK cell activating agents to suppress experimental lung metastases.
An anti-human CD39 antibody enhanced CD4+ and CD8+ T cell proliferation and Th1 cytokine secretion [IFN-γ, tumor necrosis factor (TNF)-α and interleukin (IL)-2] in vitro. Anti-human CD39 antibody also enriched intratumoral human CD8+ T cells and suppressed human B-cell lymphoma following autologous Epstein-Barr virus-specific T cell transfer. First-in-human trials of the anti-CD39 antibody in patients with advanced cancer have recently been initiated.
Adrenergic receptors: a non-canonical immune checkpoint?
The sympathetic nervous system has a role in regulating immune responses. The tumor is innervated by the sympathetic nervous system and, in response to stress, these nerves secrete norepinephrine. Chronic stress may be detrimental because it suppresses effector immune cells while activating immunosuppressive cells. β-adrenergic receptors are expressed by many cell types in the TME and β-adrenergic receptor signaling acts through multiple mechanisms to promote tumor survival, growth, and metastasis [15].
In murine models, β-adrenergic receptor antagonists (i.e., β-blockers) have been shown to improve the antitumor immune response. Chronic adrenergic signaling in mice exposed to stress promotes tumor growth. Reducing β-adrenergic receptor signaling through the use of the β-blocker propranolol facilitated conversion of tumors to an immunologically active TME and was associated with a significantly increased efficacy of anti-PD-1 checkpoint blockade [16]. Retrospective studies have also suggested that incidental use of β-blockers is associated with better survival outcomes in cancer patients. In patients with metastatic melanoma who received immunotherapy, OS was improved by use of pan-β-blockers [17]. In a small prospective study of patients with stage IB-IIIA cutaneous melanoma, those willing to take propranolol as an adjuvant treatment had an approximately 80% reduction in risk of recurrence versus those who did not take propranolol [18].
The potential of β-blockers to improve response to checkpoint blockade has been further explored in a phase I trial of propranolol and pembrolizumab in combination in patients with locally advanced and metastatic melanoma [19]. Nine patients received increasing doses of propranolol in combination with pembrolizumab 200 mg every 3 weeks. No dose-limiting toxicities were observed, and the most frequent treatment-related adverse events were rash, fatigue, and vitiligo. Objective response rate (ORR) was 78%. Perceived Stress Score decreased over time, but baseline score did not predict response. Responders tended to have a higher ratio of inflammatory (T helper cells and CTLs) to regulatory cells [myeloid-derived suppressor cells (MDSCs) and regulatory T cells (Tregs)] in the TME at baseline, suggesting this may predict for outcome. Responses seen in this small study support the potential for pan-β-adrenergic blockade to synergize with anti-PD-1 inhibition. A phase II multicenter study is currently underway.
Escape mechanisms in melanoma
HLA class 1 antigen-processing machinery (APM) defects are frequently present in malignant tumors. Mutations in the HLA class I genes themselves, abnormalities in their regulation and/or defects in HLA class I-dependent antigen processing can underlie HLA class I downregulation. Beta 2 Microglobulin (β2m) mutations inhibit HLA class I heavy chain-β2m-peptide trimolecular complex formation on melanoma cells. However, structural defects in HLA class I heavy chain, β2m, and HLA class I APM components are caused by mutations in only a low percentage of malignancies, at most 25% of the total defects. Multiple regulatory mechanisms can be involved but the most frequent cause of HLA class I APM defects in malignancies is represented by abnormalities in epigenetic pathways. This can involve reduced HLA class I APM components chromatin accessibility, methylation of the promoter regions of HLA class I APM components by DNA methyltransferases, lack of histone acetylation due to histone deacetylase (HDAC) overexpression, and histone H3 lysine 27 trimethylation (H3K27me3) by polycomb repressive complex 2 (PRC2). It may also be due to inhibition of transcription factors for HLA class I APM components, by downregulation of NLRC5 that acts as a transcriptional activator of major histocompatibility complex (MHC) class I gene and interferon regulatory factor 1 (IRF1). In addition, HLA I expression is regulated by MAPK pathway including activation via EGFR, HER2, RAS, RAF, ALK or RET mutations/overexpression resulting in STAT1 inactivation. Lysosomal degradation of HLA class I heavy chain-β2m-peptide trimolecular complexes may also result in HLA class I downregulation on malignant cells, with autophagy resulting in expression of cargo receptor NBR1 involved in trafficking of trimolecular complexes to the lysosome and overexpression of PCSK9, a secreted protein binding to an extracellular region of HLA class I heavy chain that mediates endosomal-lysosomal degradation of trimolecular complexes.
Most HLA class I APM defects can be corrected by counteracting the abnormalities in epigenetic and/or regulatory mechanisms. Abnormalities in the expression, regulation and/or function of components of this machinery have been associated with the development of resistances to T cell-based immunotherapies. Restoring sensitivity to checkpoint inhibition may be achieved by using targeted strategies to enhance HLA class I expression. Restoration of HLA class I APM component expression in Merkel cell carcinoma cells by treatment with HDAC inhibitors in vitro provided the rationale for combining this approach with immune checkpoint inhibition. In a patient with Merkel cell carcinoma with complete loss of HLA class I expression and resistance to checkpoint inhibition, treatment with the HDAC inhibitor panobinostat restored HLA class I expression, increased CD8+ T cell infiltration, and resulted in disease stabilization following anti-PD-L1 treatment with avelumab.
Tumor cell-derived exosomes can also contribute to immune-cell dysfunction in cancer. Both normal cells and tumor cells in the TME produce exosomes, which are mixed populations of normal cell-derived and tumor cell-derived vesicles. The tumor antigen chondroitin sulfate proteoglycan 4 (CSPG4) can be used as a marker to separate exosomes released by cancer cells from exosomes released by non-malignant cells. Melanoma cell-derived exosomes inhibited C-type lectin CD69 expression, induced apoptosis, suppressed proliferation in CD8+ T cells and downregulated activating receptor NKG2D expression in NK cells while non-melanoma cell derived exosomes were enriched in immunostimulatory proteins [20]. Melanoma cell-derived exosomes may be a major mechanism of tumor-induced immune suppression and as a barrier to immunotherapy.
Electrochemotherapy in metastatic melanoma
Electrochemotherapy (ECT) involves the application of high intensity electric pulses which increase the permeability of cell membranes, allowing the direct diffusion of cytotoxic drugs into cells. In addition to its use in skin cancers, ECT can be employed for the treatment of deep-seated lesions, including in the intestinal tract, pancreas, liver, and bone.
ECT can provide a durable benefit in melanoma as shown in the study of 60 patients that reported 48% had a complete response, that was long-lasting after one ECT session, and 13 patients (45% of complete responders) disease-free after a mean duration of follow-up of 27.5 months [21]. Similarly, a larger multicenter study, reported 67% 1-year OS and 74% melanoma-specific survival, indicating that ECT is a highly effective local treatment for melanoma metastases in the skin [22]. Coverage of deep margins, previous irradiation of the treated area and tumor size (< 3 cm) were all significantly associated with complete response to ECT. More recently, 11-year data from 28 centers across Europe that included 987 patients with 2482 tumor lesions were analyzed [23]. ORR was 85%, with 70% complete responses and 15% partial responses. Response rates were high across different histotypes, including metastases of malignant melanoma (82%), basal cell carcinoma (96%), breast cancer metastases (77%), squamous cell carcinoma (80%) and Kaposi's sarcoma (98%). Higher response rates were achieved with lesions < 3 cm. Hexagonal electrodes were generally used for larger tumors, but linear array electrodes provided better tumor control for tumors < 3 cm. Intravenous administration of bleomycin was more effective than intratumoral injection in tumors > 2 cm in size.
As with radiotherapy, ECT may induce an abscopal effect which suggests the potential for combining with immunotherapy. This may be due to the release of tumor antigens stimulating inflammatory response, cytokine production, complement activation, increased MHC class I expression and T cell activation as a result of checkpoint inhibition. In 15 patients treated with ipilimumab who underwent ECT for local disease control and/or palliation of cutaneous lesions, a local objective response was observed in 67% of patients (27% complete response) [24]. According to immune-related response criteria, a systemic response was observed in nine patients, resulting in a disease control rate of 60%. Response was associated with a significant decrease in levels of circulating Tregs. In another study of 127 melanoma patients treated with ipilimumab, the addition of local peripheral treatment (local irradiation, skin directed ECT or selective internal radiotherapy of liver metastases) significantly prolonged OS [25].
A randomized study to compare ECT with wide excision in patients with local cutaneous melanoma undergoing sentinel lymph node biopsy is planned.