Immune correlates of clinical outcome in melanoma

Summary

Melanoma has long been recognized as a potentially immunogenic tumour, but only recently has it become clear that the reason for this resides in its many ultraviolet (UV)‐induced mutations and expression of multiple autoantigens which can be targeted by the immune system. The first successful applications of immune‐based treatments included passive immunotherapy using high‐dose interleukin (IL)‐2 and/or adoptive transfer of natural killer (NK)‐cells, as well as active immunotherapy using whole cell‐derived or peptide vaccines. In the intervening decades, it has become clear that these approaches can lead to durable responses in stage III/IV melanoma, and even to functional cures – but only in a vanishingly small fraction of patients. With the advent of immune checkpoint blockade first with anti‐cytotoxic T‐lymphocyte 4 (CTLA‐4), then with anti‐programmed cell death 1 (PD‐1) antibodies, and combinations thereof, the small percentage of responding patients may be increased to half, a major accomplishment in this refractory disease. Improved techniques for identifying mutation‐derived neoantigens and thus more sophisticated active immunotherapies, probably combined with checkpoint blockade, currently hold great promise for further increasing the fraction of responding patients. As additional immunomodulatory antibodies and therapies become available, it will be increasingly important to develop diagnostic tools to determine which particular therapy is likely to elicit the best response for the individual patient. Practically speaking, therapy selection and efficacy monitoring on the basis of the results of a blood test would be most desirable. The purpose of this review is to consider the feasibility of identifying ‘immune signatures’ for predicting responses and determining mechanisms responsible for success or failure of these immunotherapies.

Introduction

Since Sir Frank Macfarlane Burnet proposed the concept of cancer immunosurveillance in the 1950s,1 and others demonstrated the existence of tumour immunogenicity in animal models, the old idea that the human immune system can recognize and control neoplasia has attracted a great deal of attention. However, for many decades, conflicting data in animal models and in human patients led to scepticism regarding the whole concept. Despite enormous efforts in the field, the inability to reliably translate such concepts into meaningful clinical therapies fed such scepticism among oncologists and even among immunologists, so that even only a decade ago many considered that immune‐based treatments would never become effective first‐line therapies. Nonetheless, from the early days of experimental immunotherapy, rare patients responding to one or another form of immune‐based treatment continued to encourage the belief that continuation of attempts to develop both passive and active approaches would eventually prove worthwhile. Passive immunotherapy included the use of antibodies with specificity for antigens expressed or over‐expressed by tumours, and here some notable successes were achieved for certain indications, such as herceptin for breast cancers over‐expressing human epidermal growth factor receptor 2 (HER2)/neu.2 The general applicability of antibody‐based approaches for treating most tumours has remained limited, but technical modifications to enhance antibody efficacy by engineering the antibody itself,3 or even transplanting the antibody specificity into T‐cell receptors4 may yet increase the relevance of antibody‐based therapies. Since the 1980s much effort has also been expended on other forms of passive immunotherapy, most notably using cytokines, especially interleukin (IL)‐2 with or without infusions of natural killer (NK) cells [lymphokine‐activated killer (LAK) cell therapy].5

In all these early trials, late‐stage melanoma was commonly targeted because of a lack of effective alternative therapies and because of the perception that this tumour might be more immunogenic than most.6 This idea may have arisen because of the reported occurrence of rare spontaneous melanoma regressions, accompanied by vitiligo,7 leading to the belief that this represented a special case of the anti‐cancer effects of autoimmunity, which would not be applicable in most other tumours. Thus, the majority of earlier immunotherapy trials focused upon melanoma, with even the earliest studies seeking immunological correlates of response to treatment8 and exploring the concept of combining immunotherapy with chemotherapy,9 an idea in which there has been a recent resurgence of interest.10 At the same time that passive transfers of cells, cytokines or antibodies were being developed, the emerging technical ability to identify the antigens recognized by T‐cells resulted in the isolation of human leucocyte antigen (HLA) class I‐restricted short peptides preferentially presented by tumour cells.11 This knowledge was rapidly translated into experimental active immunotherapies for melanoma using such short peptides as vaccines and initiating a wide range of trials attempting to stimulate immune responses by active immunization. In most cases this resulted in disappointment, because even when vaccine‐specific T‐cells could be found in the patient's blood there were rarely any correlations with clinical efficacy, which remained achievable in only a tiny minority of patients. In many cases, this is believed to be because the dynamic and evolving tumour‐immune cell interactions lead to eventual tumour ‘escape’ from the immune response, via a plethora of different mechanisms to avoid both innate12 and adaptive immunity13 or both.14 Many of these escape mechanisms have been identified in the meantime, and their existence in patients validated, allowing therapeutic approaches to be developed that target these mechanisms. Nonetheless, even the first experimental trials, in certain instances exceptional individual patients have survived for far longer than would have been expected and may even have been cured. These early efforts treating patients in the 1990s who are still alive today unequivocally demonstrated the feasibility of developing effective immunotherapies.15

However, it was the advent of so‐called checkpoint blockade that has truly revolutionized cancer therapy during the last few years, as data from trials (again, mostly on melanoma as the ‘pioneer‐type’ cancer) have emerged. These document not only initial clinical responses, but show beyond doubt that responses can be durable for many years of recurrence‐free survival. For the first time the ‘C’ word may be used not for ‘cancer’, but for ‘cure’.16 The first such checkpoint blockade immunotherapy to be approved was a CD152 antibody, which blocks the binding of the negative receptor cytotoxic T‐lymphocyte antigen 4 (CTLA‐4) on T‐cells to its ligands CD80 and CD86 on antigen‐presenting cells, among other possible mechanisms of action.17 This immunomodulatory pathway is a physiological regulatory mechanism essential for controlling adaptive immunity, but exploited by cancer to avoid immune destruction. Approval by the Food and Drug Administration (FDA) of the monoclonal antibody ipilimumab (Yervoy™) for treating melanoma in 2011 was to be the first of many approvals for this target in several different solid tumours. The CD152–CD80/86 pathway is, of course, only one of many physiological regulatory mechanisms required to keep the immune system in check (Fig. 1). Of these, antibodies against the programmed cell death 1 (PD‐1)‐negative receptor on T‐cells and its ligands programmed cell death ligand (PD‐L)1 and PD‐L2 on tumour (and other) cells have proved even more effective than anti‐CTLA‐4 antibodies, and with fewer side effects. These agents are now licensed for a wide range of solid tumours and may even become recommended first‐line therapies for non‐small‐cell lung cancer in the very near future. Combinations of anti‐CTLA‐4 and PD‐1 antibodies are even more effective, yielding clinical responses in 50% of melanoma patients.18 These striking results hold out hope that combinations of these and other antibodies, together with other approaches to boost immunity (e.g. active vaccination; adoptive immunotherapy) and, most importantly, to prevent tumour escape from the newly released or de‐novo‐generated immune responses, will allow effective cures in all patients.

Figure 1.

Potential targets for immunomodulatory antibody therapy in cancer.

Thus, prerequisites for patient‐individualized effective cancer therapy could ideally be as follows:

1. Assessment of the patient's tumour status

2. Assessment of the patient's peripheral immune status

3. Select combination therapies based on (1) and (2)

4. Monitor the patient's immune status during therapy and modify when necessary.

The purpose of this paper is to consider these options and their feasibility for treating metastatic melanoma at the present time.

Assessment of the patient's tumour

Surgery remains and will continue to remain the mainstay of melanoma therapy at the early stage of disease. Unfortunately, many patients present with metastases requiring systemic therapy. In most cases, tumour biopsies or resected lesions will be available for pathology. Clearly, in rare cases of occult primary or visceral metastases only, this will not be possible and so these patients would need treatment without knowledge of the characteristics of their tumour. Even when tumour is available, such is the heterogeneity and instability of mutation and gene expression in melanoma (and many other tumours) that the characteristics of one resected deposit may not be representative of other deposits, including lesions newly arising during therapy. Nonetheless, it is clear that much important information can be acquired from studying the nature and distribution of immune cells infiltrating the tumour, and from assessments of the characteristics of the cancer cells and the stromal cells constituting the tumour itself. This has been most comprehensively reported for resected specimens of primary colon cancer, but has also been extended to other solid tumours. This type of analysis is less well‐developed for melanoma, or for solid tumour metastases in general, and may well reflect outcome of the dynamic interactions between cancer and immunity at the time of resection. Thus, it is conceivable that immune signatures (i.e. constellations of immune parameters) within the tumour at baseline may be predictive of outcome in cancers other than colon carcinoma only, but follow‐up serial biopsies during treatment would be challenging and potentially dangerous. However, one report of the results of serial biopsies in a small number of melanoma patients has appeared, confirming the principle that such monitoring could be useful at the individual patient level.19 It is unlikely that such an approach could be routinely applied, even in melanoma. Hence, equally informative blood tests would be superior in that they would be less invasive for the patient, less dangerous in terms of possibly facilitating metastasis, and repeatable essentially at will. There is reason to believe that immune cells in the periphery do indeed reflect what is in the tumour, even down to the clonally enriched T‐cell receptor (TCR) specificities thought to be responsible for mediating specific anti‐tumour immunity. This was illustrated in a study of lymphocytes recognizing tumour‐specific mutant neoantigens and asking the question of whether such T‐cells were present in the circulation or limited to the tumour‐infiltrating lymphocytes (TILs). In the latter, it is most often the PD‐1⁺ CD8⁺ T‐cells that recognize these products – and hence benefit from anti‐PD‐1 therapy – and this was found also to be the case in the peripheral blood of three of four melanoma patients examined.20 Importantly, the tumour antigen specificities and TCR repertoires of the CD8⁺ PD‐1⁺ T‐cells both in the TILs and the circulation appeared similar, allowing isolation or monitoring of neoantigen‐specific T‐ cells in patients’ peripheral blood. These data are also consistent with earlier pilot studies correlating higher peripheral TCR diversity with an increased likelihood of clinical benefit from treatment with ipilimumab.21

In addition to analysing the immune cell infiltrate in solid tumours, recent efforts have been directed at determining the expression of checkpoint receptor ligands such as PD‐L1 and PD‐L2 in the tumour, either by the cancer cells themselves or the stromal cells. This represents the classical pathology approach, seeking to categorize tumour samples by immunohistochemistry, requiring large‐scale studies for standardization. A recent multi‐centre study reported that of 451 evaluable melanoma patients, 76% had PD‐L1‐positive tumours, on the basis of > 1% of tumour cell staining, often taken as the cut‐off in such cases. This study22 concluded that tumour PD‐L1 expression correlated with the clinical response rate to the anti‐PD‐1 antibody pembrolizumab, as well as with overall survival and progression‐free survival, but importantly also concluded that some patients with apparently PD‐L1‐negative tumours also responded.

More excitingly, technical developments are now beginning to allow genetic mutation analyses of tumour samples to determine which of the (in melanoma, very many) mutations actually give rise to antigens recognizable by T‐cells. Such identified neoantigens could then be applied as active cancer vaccines or, if the TCR specificities required to recognize them are found to be absent from the patient's repertoire, TCR could be identified in other subjects and engineered into the patient's own T‐cells for adoptive immunotherapy. An early study along these lines investigated this in the context of the clinical response to ipilimumab by analysing pretreatment melanoma samples versus germline samples from the same 110 patients.23 It was reported that the overall mutational load, neoantigen load and presence of cytotoxic T‐cells in the tumour were all significantly associated with clinical benefit, but no particular neoantigen was associated in general with patient response. Hence, it will be necessary to examine each individual patient to determine which antigens are relevant for that patient. One method for achieving this has been illustrated in two melanoma patients with extended (> 10 years) survival after treatment by passive immunotherapy using their own cultured TILs. It could be shown that the TILs recognized certain peptides predicted from an analysis of the multiple non‐synonymous somatic mutations from the tumours.24 Fascinatingly, in sequential biopsies taken at different recurrences, the concomitant loss of neoantigen expression by the tumour and disappearance of the neoantigen‐specific T‐cells was seen, at the same time that novel neoantigen‐specific T‐cells emerged. This latter event presumably accounted for the long‐term survival of these patients. Had such novel neoantigen‐specific T‐cells been absent from these patients’ repertoires, then they may have succumbed to disease, as is unfortunately the case for most patients. This report illustrates the importance of tracking individual patients and determining the dynamics of the ongoing immune‐tumour interactions to guide therapeutic decisions. For example, identifying the emergence of neoantigens against which the patient was unable to respond would indicate the necessity of selecting alternative means of targeting these mutations. This could include the identification of the appropriate neoantigen‐specific TCRs in unrelated healthy subjects and their adoptive transfer back to the patient, as alluded to above.25

Pioneering work from a small number of centres has now documented the feasibility of identifying cancer‐specific neoantigens in a more general manner and generating vaccines for active immunotherapy, either in the form of peptide or RNA vaccines. These have been used in a small number of melanoma patients, in some cases in combination with PD‐1 checkpoint blockade, with extremely encouraging results. Thus, in two recently published papers whole exome and RNA sequencing compared tumour and normal tissue to identify potential neoantigens in melanoma. In one study, personalized RNA vaccines were prepared encoding 10 separate neoantigens per patient26 and in the other synthetic peptides representing up to 20 neoantigens per patient were used as immunogens.27 In these two tours de force, vaccinated patients were subjected to detailed immune monitoring to correlate their T‐cell responses with the very encouraging clinical responses observed. In the RNA vaccine study, all patients had a T‐cell response to multiple neoantigens, and in one case tumour escape was documented by loss of β2‐microglobulin, and thus HLA class I expression, from the tumour. In another patient, additional PD‐1 checkpoint blocked was required to achieve a clinical response.26 In the peptide vaccine study, complete responses were also observed when PD‐1 blockade was used in combination with the personalized vaccine.27 Together these studies illustrate the technical feasibility of the personalized approach and the requirement for individual patient monitoring and treatment adjustment.

Assessment of the patient's peripheral immune status

For establishing baseline peripheral immune status, and importantly for follow‐up, it was argued above that a blood test would be optimal. Even without knowledge of the tumour‐infiltrating immune cells or the cancer antigen or checkpoint ligand expression at baseline, there may be certain characteristics or combinations of immune parameters that correlate with clinical outcome. As with any biomarker, these should be viewed in the first instance only as markers of outcome and not necessarily as providing information on the mechanisms involved in tumour control or tumour escape, and are only relevant if associated with a robust clinical outcome. Thus, differences between patients and controls may or may not point to clinically relevant characteristics, which cannot be predicted a priori. For example, in our own studies we found that on average the percentages of CD4⁺ regulatory T‐cells (T_regs) in the blood tended to be higher in stages III and IV melanoma patients than in controls, but that survival of patients with above or below median levels of these cells was no different.28 In contrast, while average percentages of certain phenotypically defined so‐called ‘myeloid‐derived suppressor cells’ (MDSC) were not markedly different in patients and controls, those patients with higher than median levels of these cells nonetheless experienced significantly worse survival.28 We found that a signature of higher levels of MDSC was always associated with poorer survival, whether melanoma patients had been pretreated conventionally or with various different immunotherapies, including checkpoint blockade with ipilimumab. There is a large literature consistent with this finding, also in other cancers, as we ourselves have shown for breast cancer.29 However, associations with survival were not sufficiently strong for determination of MDSC levels to be useful for predicting survival or selecting treatment at baseline, or to monitor ongoing treatment, at the individual patient level. Because MDSC are hard to define phenotypically and functional assays are also unwieldy and poorly specific, we expended some effort in attempting to define more detailed MDSC phenotypes correlating more precisely with patient survival. We used an extended panel of markers for this, employing mass‐spectrometry flow cytometry (CyTOF). Thus far, we have been unable to demonstrate a sufficiently close correlation between a patient's level of extended‐phenotype MDSC and survival, although combinations of markers for T‐, B‐ and NK‐cells in addition to MDSCs are likely to be more informative.30 One advantage of using CyTOF is the ability to test far more surface markers than possible with optical flow cytometry, using smaller amounts of patients’ blood, an important consideration for clinical monitoring. We are currently in the process of validating the results from this pilot study using more patients and a wider range of markers. It seems likely that combinations of all available techniques, including functional tests as well as surface marker phenotyping, will be necessary if we are ever to achieve the goal of predicting individual patient outcome sufficiently accurately to be in a position to select the most appropriate therapy for that individual patient from a range of potential treatment options. In a first attempt to introduce some level of functional testing into such monitoring schemes, we established in‐vitro culture systems to assess the ability of each patient's peripheral blood immune cells to respond to synthetic peptides representing selected tumour‐associated antigens known to be relevant in melanoma (e.g. NY‐ESO‐1, Melan‐A).31 A response in these assays requires that the patient's antigen‐presenting cells are functional, that their T‐cell receptor repertoire includes TCRs able to recognize the antigen, that their T‐cells are functionally intact as assessed by the production of pro‐ and anti‐inflammatory cytokines and that these responses are not prevented by T_regs or MDSCs. It is therefore a very demanding signature, assessing several different parameters simultaneously. Using this approach, which could potentially be applied also to neoantigen‐specific responses, we determined that an unopposed proinflammatory in‐vitro T‐cell response to certain (but not all) tumour‐associated shared antigens correlated with survival of late‐stage melanoma patients.32 When combined with phenotypical assessments of the level of MDSCs, the correlation became significantly stronger. This was shown to be the case in breast cancer (and so not limited to one potentially unusual tumour type) and also in older patients (and so not negatively affected by immune ageing). These results document that such combined ‘immune signatures’ do not necessarily apply solely to one type of cancer that might not have been representative, and also that the immune system of older patients (> 80 years of age) was functionally intact, at least in this respect.29, 33 This is important, given the common perception that ageing of the immune system (‘immunosenescence’) contributes not only to decreased protection against infectious disease, but also cancer, as cited in many reviews (e.g. ref. 34).

Selection of combination therapies at the individual patient level

Given the current situation of an increasing number of potential immune‐based therapies that could be offered to a particular patient, and the fact that even with the best approach thus far feasible it is likely that not all patients will respond, it would clearly be valuable to predict outcomes and synergies. This is also increasingly important because it is becoming clear that some combinations of agents active by themselves may be mutually inhibitory,35 and even more so considering that in some cases it may be that immunomodulation can cause faster tumour growth,36, 37 perhaps reflecting the old concept of immune stimulation of cancer.38 In one of these studies this may be particularly worrisome in people over the age of 65 years, implying that there may indeed be some subtle effects of ageing on immunity in this respect.36 The discussion above has provided some notion as to how this selection of optimal therapies and avoidance of undesirable outcomes might be accomplished in an ideal world. In practice, applying the most sophisticated approaches thus far possible is unlikely to be feasible for every patient. Thus, while on one hand it seems clear that personalized medicine is needed to reliably treat cancer, on the other hand, a one‐size‐fits‐all approach, as at present, is the simplest. For routine clinical use, ideally a set of parameters already tested in hospital laboratories would be the simplest to fit into the work flow. To this end, and with the specific question of predicting clinical responses to ipilimumab in metastatic melanoma patients, we sought laboratory parameters that most hospitals would be able to establish as routine procedures without special training or accessing overly sophisticated equipment. Our first results suggest that a baseline signature of low levels of the commonly used marker lactate dehydrogenase, together with low absolute monocyte counts, low MDSCs but high absolute lymphocyte counts and, intriguingly, high not low CD4⁺ T_reg levels were associated with a favourable outcome following single‐agent ipilimumab treatment.39 Additionally, increases in absolute lymphocytes counts and circulating CD4⁺ and CD8⁺ T‐cells as well as γ/δ T‐cells were associated with better responses to this agent during therapy.40, 41 We are currently extending these analyses to melanoma patients treated with anti‐PD‐1 antibodies and combinations of anti‐CTLA‐4 with anti‐PD‐1 antibodies. Even if these correlations cannot realistically be expected to reach 100%, it should become possible to select much more accurately those patients with the best chance of responding to a particular therapy and thus increase clinical success coupled with decreased costs and side effects.

Patient monitoring

As discussed above, current techniques for monitoring melanoma immunity biosignatures have become highly sophisticated in recent years, requiring a large investment of multidisciplinary resources to be available in loco for treating and monitoring each patient separately. Although such centres currently exist in only a handful mainly of academic hospitals and are expensive, it is not unfeasible to propose more widespread future use of diagnostic techniques as a base for treatment selection and monitoring efficacy. Such an ‘immunoholistic’ approach would take into account the characteristics of the tumour and the changing dynamic of the cancer‐immune system interaction. Combination therapy to focus the adaptive immune system on the relevant cancer antigens (such as active vaccination together with checkpoint blockade, as referenced above) together with measures to prevent tumour escape (such as MDSC down‐regulation, neutralization of the inhibitory activity of the tumour microenvironment, etc.) (for reviews, see Refs 42, 43), amplification of anti‐tumour innate immunity and judicious use of immunoenhancing radio‐ and chemotherapy,44 as well as possible ‘epigenetic rejuvenation’ of ‘exhausted’ CD8⁺ T‐cells, offer hope that efficient anti‐cancer immunity could be monitored, modulated and maintained by modifying therapies appropriately over the course of treatment, and thus elicit cures in all patients. Given the remarkable success of even some single‐agent immunotherapies in certain patients which, considering the redundancy of immune control mechanisms and the great propensity of cancer to evolve resistant variants, is even more remarkable, it is all the more likely that future protocols based on individual patient factors will increase treatment success rates. There are numerous novel approaches and combination therapies currently being trialled in melanoma, including combining checkpoint blockade with oncolytic viruses,45 with variants of adoptive immunotherapies using novel ex‐vivo primed cells,46 or by using different immunomodulatory antibodies recognizing thus far little‐tested receptors such as CD27,47 as well as new candidates for checkpoint blockade targets such as T‐cell immunoreceptor with Ig and ITIM domains (TIGIT) binding its CD155 ligand on melanoma cells.48 We should also not forget that there may still be a place for one of the earliest forms of successful immunotherapy, perhaps using high‐dose IL‐249 in modified regimens. The ability to monitor ongoing immune correlates of anti‐tumour efficacy will facilitate eradication of the cancer by enabling rapid adjustment of therapeutic protocols during the whole course of treatment.

Conclusions

Selecting and adjusting personalized immune therapies in melanoma will require examination of the patient's tumour to determine which antigens to target and which tumour escape mechanisms to aim to counteract, by means of monitoring the dynamic interactions between the immune system and the tumour. The latter will need to be based preferentially on assays that can be performed using peripheral blood [although advances in imaging, e.g. positron‐emission tomography (PET) scans, may facilitate non‐invasive monitoring of immune cells at tumour sites, thus far only possible in small animals50]. Available data so far suggest that phenotypical determination of the nature of the immune cells in blood, and assessing their antigenic specificities and functions, is indeed informative for the clinical course of the patient. A major proviso in all these considerations is that the situation in the periphery sufficiently mirrors that in the tumour, especially concerning T‐cell clonal specificities, their relative state of ‘exhaustion’ and how far this is still discernible in locations distant to the local activity of suppressive mechanisms in the tumour itself. Only comparative analyses of these biomarkers in correlation with clinical outcome will resolve this issue but, surprisingly, there are very few such comparisons to be found in the literature. While even the most sophisticated biomarkers are unlikely to achieve 100% specificity and sensitivity, our ability to correlate constellations of such markers with a particular clinical outcome is improving rapidly. As more data are gathered on novel immunotherapeutic strategies, it is anticipated that accuracy of predicting the outcome of a particular treatment will also rapidly improve.

Disclosures

The author declares no competing interests.