Applications of systems biology in cancer immunotherapy: from target discovery to biomarkers of clinical outcome

Abstract

Immunotherapies are coming to the forefront as a treatment paradigm in cancer with multiple US FDA approvals in recent years and a better understanding of their therapeutic mode of action. The control of tumor growth by the immune system is orchestrated by a complex array of cellular interactions and molecular pathways, both in the immune cells as well as the tumor. Although research over the past three decades has elucidated many aspects of tumor immunosurveillance, given the inherent complexity of the immune cell phenotypes and function, high-throughput molecular profiling (‘omics’) approaches have now become essential to support the discovery and development of new therapies. Technologies, such as DNA and protein microarrays, deep sequencing, mass spectrometry, as well as the computational methods for their analyses, are advancing the contributions of systems biology towards the development and mechanistic understanding of cancer immunotherapies. In this review, the authors illustrate this through some recently reported studies.

Keywords::

• bioinformatics
• cancer immunotherapy
• immune response
• molecular profiling
• omics
• pharmacodynamic biomarker
• predictive biomarker
• systems biology

Figure 1. Biospecimens, molecules and omics platforms that can be used in the context of cancer immunotherapy studies.

Each platform can have a multitude of applications, some of which are described in the text.

Adapted with modifications from [39].

Figure 2. A conceptual integrated framework for the discovery of antigens from omics studies in cancer patients, database mining and algorithmic predictions of immunogenicity.

A number of platforms can be applied to study both tumor and peripheral samples. Since the omics platforms can have a high false discovery rate, integration of data across orthogonal platforms can enable rapid discovery of antigens that have a higher probability of validation in subsequent preclinical testing.

Figure 3. Examples of omics platforms being used to assess post-treatment therapeutic responses against the tumor, such as the extent of lymphocyte infiltration, tumor destruction and epitope spread against ‘secondary’ antigens (or epitopes).

mRNA profiling may be used to assess the extent of tumor infiltration of lymphocytes. Protein or peptide microarrays can be used to assess the diversity of antibody (B cell) responses and next-generation sequencing (T-cell receptor sequencing) can be used to assess the diversity of T cell clones. Specific examples are provided in the text.

Systems biology is viewed as an approach that integrates and analyzes data from high-content omics platforms, using computational and statistical methods, to better characterize biological phenomena [1–4]. Rather than probe a single element in depth, sometimes referred to as a ‘reductionist approach’, technologies such as microarrays, deep sequencing, high-throughput proteomics or metabolomics, aim to broadly study the state of the biological systems, from which hypotheses can be generated for further exploration. In cancer immunotherapy, the applications of systems biology have ranged from new target discovery, to the mechanistic understanding of therapies and the identification of biomarkers for patient stratification according to therapeutic response. Through known examples we will broadly cover these applications, after providing a brief overview of systems biology in drug development and recent advances in cancer immunotherapy. There are numerous examples of systems biology applications in cancer biology, basic immunology and infectious disease research. Therefore, where suitable, we will draw examples from these related areas as well.

Systems biology applications in drug discovery & development

Since DNA microarrays were developed about 20 years ago, omics profiling has been extensively used for addressing essential issues in drug development, such as, the elucidation of the biological pathways associated with disease development, demonstration of therapeutic mechanism of action (MoA), development of molecular classifiers for identifying subpopulations in disease and drug response and prediction of drug toxicity [1–3,5–8]. Many other platforms and methods have been developed over the years that can be included in the broad spectrum of approaches for systems biology, such as mass spectrometry for proteomics and metabolomics, protein and antibody microarrays for monitoring antibody or protein levels, whole-genome single nucleotide polymorphisms (SNP) genotyping, and next-generation sequencing (NGS) for DNA and RNA analytics.

One potential benefit of the systems biology approaches is the ability to generate a more comprehensive understanding of biological pathways and their relationships with disease or drug response. This can enable data-driven decision making in many phases of drug discovery and development [1,2,5–8]. Results from these approaches can facilitate the de-risking or re-prioritizing of therapeutic candidates in development in order to improve their probability of success in clinical testing. Regulatory agencies, such as the US FDA, also encourage the application of such approaches, as these may improve and streamline the assessments of safety and effectiveness of the therapies [9].

To date, the applications of omics and systems biology have been most widespread in oncology, where characterizing the molecular complexity and heterogeneity of the tumor is critical not only for the development of targeted therapies, but also for the identification of responder patient populations for such treatments. Work in oncology provided the first successful examples of the utility of gene-expression profiling for disease subtyping [10–12], disease prognosis [13,14] and multi-gene diagnostic tests [14–16]. Significant progress is also being made in the areas of autoimmune and infectious diseases, where systems biology is being used in characterizing disease pathogenesis, early disease detection, and understanding the MoA of vaccines and adjuvants [17–21]. The use of systems biology approaches in the above areas has improved the trajectory of therapeutic development, and it has the potential to be useful in similar ways in the domain of cancer immunotherapy.

Below we discuss known applications of systems biology in cancer immunotherapy. For the purpose of this review, we segment the discussions into sections such as new antigen discovery, understanding therapeutic mechanisms of action, and clinical biomarker discovery. In reality however, these topics are intimately related. For example, better characterization of a therapeutic target and its mechanism can significantly inform the development of biomarkers of in vivo efficacy.

Recent advances in cancer immunotherapy

The engagement of the immune system to control tumor growth and the evasion of tumor destruction from the immune system has come to be recognized as a hallmark of cancer pathogenesis [22–24]. Over the past 30 years, research and clinical development of cancer immunotherapies have resulted in a deeper understanding of the interplay between the immune system and tumor development [23–28]. The importance of the immune system in controlling tumor growth is supported by several key clinical observations: (i) cancer incidence is increased in immunocompromised individuals [29–31], (ii) increased infiltration of T lymphocytes into the tumor is predictive of survival [32,33], and (iii) therapies that engage or activate the immune system, including cytokines, therapeutic vaccines, immune checkpoint inhibitors and monoclonal antibodies, have proven to be effective in treating cancers [25,34,35].

Recent advances in the field have been rapid, and have led to the regulatory approval of two therapies by the FDA: (i) sipuleucel-T or Provenge^® (Dendreon Corporation, WA, USA), an autologous cellular therapy, activated using an antigen cassette consisting of prostatic acid phosphatase (PAP, a prostate antigen) and granulocyte–macrophage colony-stimulating factor (GM-CSF), used for the treatment of asymptomatic or minimally symptomatic, hormone-refractory, metastatic prostate cancer [35,36]; and (ii) ipilimumab or Yervoy^® (Bristol–Myers Squibb, NY, USA), an immune modulating antibody directed against CTLA-4, a negative regulator of T-cell activation, for the treatment of metastatic melanoma [34,37]. Immunotherapies currently undergoing clinical testing represent a complex array of approaches, but may be broadly classified into: (i) cellular therapies (such as autologous dendritic cells [DCs], antigen-presenting cells [APCs], adoptive T-cell therapies, or, genetically engineered T cells containing chimeric antigen receptors or CARs), (ii) cytokine and adjuvant therapies, (iii) immune-modulating antibodies such as those targeting immune checkpoints CTLA-4 or PD-1, and (iv) cancer vaccines containing tumor antigens (e.g., MAGE-A3, NY-ESO-1 and MUC-1) [23,28,38]. Although tumor targeting monoclonal antibodies (e.g., rituximab or trastuzumab) constitute an important mode of cancer immunotherapy [28], for the purpose of this review, we only cover those modalities (mentioned above) that directly target the immune system for its activity against the tumor, or involve tumor targeting by engineered immune cells. Given the promising results from ongoing clinical trials, it appears likely that several immunotherapies will gain regulatory approval for treating various cancers in the coming years.

Applications of systems biology in cancer immunotherapy

Various omics platforms can be used to interrogate biospecimens from the subjects under study (see Figure 1 and [39]). Of these, mRNA expression profiling (transcriptomics) has been the most widely used and the resources for the interpretation of such data (statistical and bioinformatics methods, as well as databases) are well developed. However, other high-content platforms, including proteomics (mass spectrometry, protein and antibody microarrays [40–42]) and NGS (nucleic acid sequencing and quantification) [43,44] are also routinely used to investigate distinct aspects of discovery and development in cancer immunotherapy.

Discovery of targets for cancer immunotherapy

Therapeutic targets for cancer immunotherapy can be broadly classified into two categories. One class of targets is typically expressed on the cells of the immune system, and depending on the therapeutic objective, can be activated (as by cytokine or agonistic antibody) or repressed (as by a checkpoint inhibitor) to modulate the immune function for activity against the tumor. Antigen targets, by contrast, are expressed by the tumor cells, and can be used as immunogens to specifically direct the immune effector activity against a malignancy. While the requirements related to immunogenicity confer certain restrictions on antigen selection, the range of the tumor antigens for an immunotherapy may also be wider than those that can be targeted by small molecule or monoclonal antibody therapies. For example, targets such as transcription factors and intracellular proteins, may be targeted in an immunotherapy (e.g., [45]), since the antigens targeted by the immune system do not need to be ‘druggable’ (as is usually required by the small molecule inhibitors) [46] or expressed on the cell surface (as required by monoclonal antibody therapies). With the emphasis on the discovery of tumor antigens, we discuss below some of the primary considerations for selecting targets for immunotherapies, and how systems biology approaches are being used to discover such targets.

The selection of the suitable tumor antigen(s) is critical for antigen-targeting cancer immunotherapies [47,48]. Ideally, these targets should possess a number of attributes, including tumor or tissue specific alterations (aberrant expression, mutation, splicing or post-translational modifications) and immunogenicity [47]. Since off-target, immune responses may be potentially severe, long-term and difficult to control [49], tumor or tissue specificity is a particularly important consideration in the selection of an antigen. Some antigens used in immunotherapies are mutated or over-expressed specifically in the tumor (such as immunotherapies targeting KRas [50], p53 [50] and Her2/neu [51]). There are also immunotherapies that have targeted antigens which show restricted normal expression in a tissue that is considered non-essential (such as the testicular germ cells), but aberrant expression in certain tumors (viz. the cancer testis antigens) [52]. Antigens that are tissue specific (such as, gp100 for melanocytes [53], PAP [35,36] or PSA [54] for the prostate and CD19 for B cells [55]) have also been used effectively as targets for treating malignancies originating from those tissues. As antigen targets should ideally have tumor or tissue specific alterations, the mining of the gene expression or mutation databases [56–58] have contributed to the discovery of many tumor antigens over the years [59–62]. The application of deep sequencing now has the potential to unravel tumor gene alterations at a resolution previously not possible [43,62].

The efficacy of an immunotherapy relies upon enhancing the immune system’s ability to suppress tumor growth by the tumor-targeting lymphocytes. Therefore, immunogenicity is a critical consideration for selecting a tumor antigen. Immunity against a tumor antigen depends on the recognition of the peptides from the antigen presented to the lymphocytes by the MHC molecules on the target (tumor) cells, as well as the professional APCs. The repertoire and activity of lymphocytes targeting auto-antigens is generally limited by the immune system’s natural tendency to delete or suppress auto-reactive effector lymphocytes. Tumor-specific alterations may increase the immunogenicity of the antigens by expressing new epitopes that the immune system may not be tolerized to.

Immunogenicity can be assessed by the frequency with which the antigen peptides are presented by the MHCs and the frequency with which naturally occurring cellular or humoral responses to the antigen are observed. There are experimental as well as computational methods for assessing immunogenicity. These include high-throughput humoral profiling techniques for detecting antibodies, and the identification of MHC-restricted epitopes using computational predictions or mass spectrometry. Serological analysis of recombinant cDNA expression libraries (SEREX) [63], has been used for over a decade to identify a host of well-known cancer testis antigens [52], such as NY-ESO-1, and the MAGE, LAGE, GAGE families [64,65]. SEREX involves the construction of cDNA expression libraries and protein expression followed by autologous serological screening for detecting reactivity of those proteins with antibodies present in the sera of the cancer subjects. While SEREX has been a productive approach, it is a relatively laborious technique and may not readily yield quantitative data. High-content protein microarrays provide an alternative approach for the identification of antibodies against tumor antigens [61,66]. Antibodies against hundreds or thousands of self-antigens can be quantified rapidly using these microarrays. For identifying MHC-restricted epitopes, both computational [67,68] and experimental [69–72] methods are available. Computational prediction of MHC-restricted epitopes can be challenging because of their high degree of heterogeneity in length and sites of natural truncations. Therefore, mass spectrometry has often been used as an experimental alternative to identify and quantitate the peptide epitopes from MHC-peptide complexes [69–73].

The application of high-throughput approaches, and in particular, the integrative analysis of data across platforms, can significantly facilitate the discovery and characterization of tumor antigens (Figure 2). For example, it is now possible to rapidly screen the sera from cancer patients using protein microarrays to identify tumor antigens with detectable serum antibodies [61,66]. This information can be integrated with cancer gene expression or alteration data [56–58,74] to identify tumor antigens with data supporting both their tumor alterations as well as immunogenicity. Since the omics platforms may produce hits with high false discovery rates, integration of information from orthogonal platforms can enable the identification of antigen candidates with a higher probability of success in follow-on validation studies. Walter et al. have pursued such an integrative approach for the discovery of tumor antigens in renal cell carcinoma [75]. By combining tumor gene expression profiling with mass spectrometric identification of MHC-restricted peptides from tumor specimens, they identified a set of immunogenic antigen peptides. These peptides were then used to develop a vaccine, which is under clinical investigation. In a Phase II study of this vaccine, immune responses against these peptide epitopes were associated with improved overall survival [75].

Tumor progression in patients can be frequently driven by multiple genetic events, evolving over the course of the disease. It may therefore be ideal to use multiple tumor antigens in an immunotherapy, and if possible, tailor the therapy for a patient with the specific tumor antigens that are altered in that patient [50]. Recently, Castle et al. have explored how deep sequencing technologies can be used to survey the mutational profile of tumor antigens in individuals and how that profile may be used, in principle, to enable the formulation of individualized immunotherapies [62,76].

In this section, we have discussed several important considerations for immunotherapy target selection and the omics methods for the identification of tumor antigens. The application of various omics platforms, followed by judicious integration of data, has the potential to significantly enhance the rate of discovery and prioritization of tumor antigens for future therapeutic design.

Understanding therapeutic MoA

The molecular pathways that are involved in the immune surveillance of a tumor and the tumor’s escape of the immune system are complex. Hypothesis driven investigation is often insufficient to obtain the necessary mechanistic understanding of the responses associated with an immunotherapy. Omics platforms can be particularly useful in such situations. In this section, we discuss applications of transcriptomics, proteomics and high-throughput sequencing for the characterization of MoA. We describe the applications of transcriptomics in detail and provide examples of a few unique applications of proteomics and high-throughput sequencing in the context of cancer immunotherapies.

Transcriptional profiling for understanding MoA

Since the DNA microarray has been an established platform for interrogating cellular pathways, it has been used in numerous studies for investigating MoA. Here we discuss some methods for interpreting transcriptomic data for elucidating MoA and provide examples of their applications in cancer immunotherapy. We also discuss related research in basic immunology for characterizing gene-sets(also referred to as signatures or modules) associated with immune cell lineages, signaling pathways, activation and differentiation (e.g., [77–83]). These gene-sets are key to the interpretation of new transcriptomic data, especially from complex biological samples such as peripheral blood mononuclear cells (PBMCs) or whole blood.

One way to interpret a novel whole-genome gene-expression data-set is to analyze that data using information from known biological pathways or gene-sets derived from previous studies. This approach is often referred to as gene-set enrichment analysis [84–86]. There are well-curated databases for gene-sets or signatures (e.g., MsigDB [84], PAGED [87] and GeneSigDB [88]) and biological pathways (e.g., Gene Ontology [89], KEGG [90] and Reactome [91]). In addition, there are repositories of raw transcriptomic data from which gene-sets can be derived (gene expression omnibus or GEO) [92]. There are several statistical methods for determining enrichment [85,86], which assess whether genes from a previously known gene-set or pathway are significantly over-represented amongst a list of genes identified from a new experiment that is being analyzed (e.g., genes that are perturbed in response to a given therapy). Despite its usefulness and widespread use, there are some limitations of gene-set based analyses for understanding MoA. For example, gene-sets are frequently incomplete and context (e.g., tissue or cell-type) dependent. Since the quality of the gene sets determine, in large part, the success of these approaches, incomplete or context-specific gene-sets can result in misleading inference about the mechanism. Also, the computational methods for performing enrichment analysis are still under active investigation [85,86,93]. As progress is made towards better definition of the biological pathways and gene-sets, and improved computational methods for enrichment analyses, such methods can only become more useful for the interpretation of transcriptomic data from new studies.

We highlight a few examples of the application of gene expression profiling and data interpretation for the mechanistic characterization of cancer immunotherapies. The applications can be in multiple areas, including the characterization of the cell therapy products, understanding in vivo response, and identification of immune characteristics of subjects associated with clinical response to immunotherapy.

DC or APC therapies constitute an important modality for cellular immunotherapy [94–96]. Detailed characterization of these cellular products presents a particular challenge in their development. Cellular products have an inherent complexity, and achieving consistency and uniformity in the preparation of such products require thorough quality control and understanding of a range of product parameters. Gene-expression profiling can be helpful in characterizing the activity and potency of such products. DCs for therapeutic applications are generated through culture with a wide range of activating factors (such as cytokines, adjuvants and so on). Depending on the precise nature of the cytokines or adjuvants used in their production, DCs may have different functional properties. Moschella et al. cultured DCs with IFN-α-2b and GM-CSF (to produce mature dendritic cells called ‘DCAs’), and GM-CSF and IL-4 (to produce ‘DC4’ cells), and compared the changes in the post-culture DCs using DNA microarrays [97]. Comparison of mRNA profiles between DCA and DC4 revealed that the expression of genes involved in antigen presentation was increased in DCA relative to DC4 (such as TAP-2, HLA-DQ and HLA-DM). DCA also showed upregulation of TLRs, several TNF family ligands (e.g., TRAIL, TNFα and Fas), and inflammatory chemokine receptors (e.g., CCR2) for the migration of DCs to inflamed tissues. Consistent with expression profiling, functional assays showed that DCAs were more potent mediators of naive T-cell responses and tumor-cell cytotoxicity than DC4 [97]. Similar work was done by Han and colleagues, where they compared the global gene expression profiles of immature DCs matured with lipopolysaccharide (LPS; a TLR4 agonist), IFNγ, IL-1β and TNF-α, to evaluate different clinical DC maturation protocols [98]. Understanding the association of particular genes and pathways in the product with its clinical efficacy will help facilitate the development of reliable and effective production methods for cellular therapeutics as well as establish robust methods of quality control for such products.

Transcriptional profiling from complex and heterogeneous biological specimens such as whole blood and PBMCs, along with the related bioinformatics methods for data interpretation, can significantly enable the elucidation of the MoA associated with immunotherapeutic responses. Often in clinical trials, the availability of biological materials is limited. Therefore, a detailed phenotypic or functional characterization of the immune cells by conventional assays (such as multi-parameter flow-cytometry, cell proliferation or ELISpot) may not be feasible. In such cases, whole blood or PBMCs can be analyzed using transcriptional profiling (which requires a relatively small amount of the specimen) and the data interpreted with gene-sets [19,20,99,100]. Significant progress has been made in recent years in characterizing the gene-sets that are associated with different immune-cell lineages [83,101–103], cell signaling pathways [82], cellular differentiation and activation (e.g., naive, memory and effector T cells [80,83,104] or monocyte differentiation to macrophages and DCs [105]). The computational methods for the analysis of complex samples, such as whole blood, are also under development [99,106]. Taken together, these methods can significantly benefit the studies investigating in vivo MoA and clinical biomarkers from easily accessible peripheral samples such as whole blood and PBMCs.

Transcriptomic profiling of tumors is also providing insights into the in vivo MoA of cancer immunotherapies. Multiple cancer immunotherapy clinical studies are now showing evidence that increased lymphocyte infiltration within the tumor, reflected by a higher expression of lymphocyte markers, cytokines and chemokines, might be associated with better clinical outcome from the therapies [107–109]. Pre-treatment gene-expression signatures (consisting of T-cell markers and cytokines) from melanoma and non-small-cell lung tumor were found to be associated with overall survival upon treatment with a MAGE-A3 vaccine in patients with tumors expressing the MAGE-A3 antigen [107,110]. More recently, Ji et al. have arrived at similar a conclusion for ipilimumab (anti-CTLA-4 antibody) using gene-expression profiling of tumor biopsies collected from melanoma patients before and three weeks after the start of treatment in a Phase II clinical trial [108]. Transcriptional profiling of pre-treatment tumors indicated that patients with high expression levels of immune related genes (T-cell activation, antigen presentation) were more likely to show clinical activity upon treatment with ipilimumab. In patients with clinical activity, ipilimumab treatment increased the expression of genes involved in immune responses (IFN-γ-inducible genes and Th1-associated markers) and decreased the expression of melanoma antigens and cell-proliferation genes. These post-treatment gene expression changes were associated with the lymphocyte infiltrate into the tumors. Thus, gene expression profiling is providing valuable mechanistic insights into the in vivo activity of the immunotherapies against the tumor.

Applications of proteomics in assessment of antigen spread or epitope spread

It is proposed that an effective cancer immunotherapy, which leads to the destruction of tumor tissue, may subsequently lead to priming of T or B lymphocytes against tumor antigens that are not contained in the therapy (secondary antigens). This phenomenon is often referred to as antigen spread, epitope spread or determinant spread (Figure 3) [111–113]. A broadened immune response against tumor antigens may promote more efficient killing of the tumor cells [111]. Epitope spread has been observed in response to PSA-targeting immunotherapy [114], CTLA-4 blockade in melanoma and prostate cancer [113,115], Her2/neu vaccination for breast cancer [116,117], and MUC1 DC vaccination in renal cancer patients [118]. Data from early studies suggest that a broadened immune response may occur with a higher frequency in clinical responders compared with non-responders [111,113,119].

Evaluation of humoral response against a diverse set of self-antigens can be explored rapidly using the high-content protein or peptide microarrays [61,66,115,120]. A few studies have now reported the detection and validation of post-therapy antibody responses to tumor antigens using this technology [115,120]. For example, Kwek et al. reported the discovery of PAK6 as a tumor antigen against which both humoral as well as CD4⁺ T-cell responses were observed post-CTLA-4 blockade in prostate cancer patients [115]. The authors also found that immunization with PAK6 could be protective in mouse tumor models. It is expected that future clinical studies will apply such high-throughput technologies to study the breadth of the immune response to assess the therapeutic activity against the tumor in vivo.

Application of NGS in lymphocyte receptor sequencing & monitoring of T-cell clones

High-throughput NGS technologies [43] have numerous applications in the domain of DNA and RNA analyses [44,121–123]. Among these, one that is uniquely applicable to the study of immunotherapy is the sequencing and quantification of antigen receptors on T and B lymphocytes [124–128]. Since the antigen receptor molecules from certain B cells are present in secreted form as antibodies, and there are high-throughput techniques available for identifying the specificities of antibodies (e.g., protein or peptide microarrays), the application of NGS has been particularly useful in the high-throughput assessment of the diversity of receptors on the T cells (which are present only on the cell surface). The T-cell receptor (TCR) genomic loci undergo somatic recombination, as well as addition or subtraction of nucleotides at the recombination junctions, to generate a wide array of T cells with specificities against a wide array of antigens [125,126,128,129]. The potential diversity of T-cell repertoire has been estimated to be greater than 10¹⁴ [129]. High-throughput DNA and RNA sequencing technologies now allow for a much deeper assessment of this diversity than previously possible [124–128]. Although the techniques and analytical approaches are still emerging and being refined [130,131], the applications are already addressing important issues in cancer diagnosis and immunotherapy development [132–135]. We provide some examples of such applications in both preclinical as well as clinical studies.

The population of individual clonotypes is dynamic and capable of rapid expansion or contraction in response to immune stimulation or treatment with an immunotherapy. In cancer immunotherapies, sequencing of TCR (TCR-sequencing) can help to describe the dynamics of the T-cell repertoire before and after therapy and may aid in characterizing the efficacy of specific T-cell clones in vivo. Using such an approach, Chervin et al. identified the T-cell clones present in the tumors following adoptive T-cell therapy in mice [135]. A library of TCRs with a range of binding properties to a model antigen, SIY, was introduced into T cells and adoptively transferred into a B16 melanoma mouse model containing SIY-positive tumors. High-throughput sequencing of the TCRs from tumor-infiltrating lymphocytes and lymphoid organs allowed them to determine the TCRs that were selected for in vivo. They observed that CD8⁺ T cells expressing the highest-affinity TCR variants were deleted in both the tumor-infiltrating lymphocyte population and in the peripheral lymphoid tissues. The same high-affinity TCR variants were however preferentially expressed by the intra-tumoral CD4⁺ T cells, suggesting that the CD4⁺ T cells played a role in antigen-specific tumor response. The results enabled the authors to characterize the clonotypes of the transduced TCRs that controlled the survival and tumor infiltration of the transferred T cells. Such applications can aid in the in vivo characterization of T cells and the development of adoptive T-cell therapies. Other applications have included the detection of minimal residual disease in leukemia/lymphoma patients [132,133], and assessment of the T-cell repertoire after adoptive transfer of activated autologous T cells to enhance immunological function in neuroblastoma patients treated with high-dose cytotoxic therapy [134].

Biomarkers of clinical outcome

Numerous profiling studies over the years have shown that cancer is highly complex at a molecular level, with significant variability within a given indication. Tumor gene-expression profiling has led to the identification of diagnostic markers, some of which are widely used in clinical practice [14,15]. Increasingly, diagnostic markers are allowing more efficient and successful development of new therapeutics by enabling the identification of patients most likely to respond to a given therapy.

It has been noted that ideal biomarkers of clinical outcome should have some of the following properties: i) readily detectable in bodily fluids, ii) suitable for quantitative analysis, iii) associated with relevant clinical outcomes (e.g., overall survival), and iv) indicative of the in vivo mode of action [136]. The development of biomarkers predictive of clinical outcome to cancer immunotherapies is still relatively new. However, omics technologies are contributing to the discovery and testing of such markers from preclinical and clinical studies. Recent studies across various immunotherapeutic modalities are starting to reveal some consistent themes. There is promise that such markers can be identified and translated into clinical practice. Below we discuss some early examples, with the focus on predictive (pre-treatment) and pharmacodynamic (post-treatment) markers of clinical outcome.

Predictive (pre-treatment) markers of clinical outcome

As mentioned earlier, pre-treatment tumor gene-expression profiles, indicative of host immune response to the tumor, were observed to be associated with clinical outcomes upon MAGE-A3 vaccine [107,110] and ipilimumab [108] treatments. Therapeutic outcome in subjects with specific gene-expression signatures are being evaluated as secondary endpoints in the current Phase III MAGE-A3 vaccine trials [110]. Beyond tumor gene expression profiles, peripheral whole blood or PBMC profiles hold promise as potential sources of genomic signatures predictive of response to immunotherapies. Early efforts to develop gene expression signatures from PBMCs that are predictive of clinical outcome to a peptide vaccine for the treatment of advanced prostate cancer have been reported [137]. Recently, multiple groups have also reported the discovery and validation of gene expression signatures from whole blood that are prognostic of overall survival in castrate-resistant prostate cancer [138,139]. T-helper cell activation and signaling pathways (iCOS-iCOSL and CD28 signaling) were found to be associated with prognosis [139]. Given such results, and their suggestion that immune status can predict disease course, it is conceivable that in the future studies, predictive expression signatures associated with clinical outcome to immunotherapy may be detected from whole blood or PBMCs. This would provide the tools for identifying interpretable RNA-based predictive biomarkers from easily accessible biological materials (peripheral rather than tumor samples).

SNPs and other DNA variations can be very useful as predictive markers. Relative to RNA, DNA variations are easier to assess since DNA is more stable as biological material and the number of variable states can be limited (as in the case of SNPs, somatic mutations or certain rearrangements). There are early examples where SNPs in candidate genes have been shown to be associated with response to immunotherapies [140,141]. Genome-wide studies, associating SNPs to clinical efficacy of immunotherapies, are currently lacking. SNP platforms that are focused on genes with known immune functions may provide a cost-effective way to perform large-scale genotyping studies in immunotherapy trials [142,143]. In addition, where available, integration of SNP data with other molecular data, such as gene expression, can significantly enhance the identification and annotation of functional SNPs associated with clinical outcomes [144–146]. As larger clinical studies are completed and evaluated, we may see additional results of SNP associations with clinical efficacy as well as validation of the early results.

Pharmacodynamic (post-treatment) markers of clinical outcome

Robust and reliable pharmacodynamic markers of clinical efficacy is a critical need in cancer immunotherapy development [136,147,148]. Clinical outcome to immunotherapy treatment is often not reflected accurately by the conventional imaging measures for solid tumors, such as the RECIST (Response Evaluation Criteria in Solid Tumors) or the WHO criteria [149,150]. Tumor infiltration by active immune cells following therapy may be misleadingly diagnosed as tumor growth and disease progression by these above measures. In cases where immunotherapeutic intervention leads to prolonged survival without showing reduced tumor progression by the above measures, reliable early read-outs would provide useful guidance, both to the developers of novel therapies, as well as to the clinicans managing their use in patients.

Immune response monitoring in clinical studies has traditionally been performed using techniques such as ELISA for measuring antibody levels, multi-parameter flow cytometry for characterization of immune cell phenotypes, lymphocyte proliferation assays, ELISpot for quantitating antigen-specific cells and multiplexed cytokine/chemokine measurements [148,151]. Where the therapy contains tumor antigen(s), a significant effort is usually focused on characterizing the B- or T-cell responses against the immunizing antigen(s) [75,152,153]. In some cases, immune responses against the immunizing antigen(s) have been shown to be associated with clinical outcome [75,152]. In addition, in cell therapies, specific molecular markers of cellular activation in the product may be associated with clinical efficacy [152].

Systems biology is gradually becoming an important component in immune response monitoring and pharmacodynamic biomarker discovery. Applications include the monitoring of: i) immune cell characteristics, ii) the diversity of antibody responses (potentially related to epitope spread), iii) dynamics of T-cell clones and iv) immune characteristics of the tumor microenvironment. We discuss two examples below.

Earlier, we noted that Ji et al. have found gene expression patterns associated with lymphocyte infiltration and tumor destruction following treatment with ipilimumab (CTLA-4 blockade) [108]. Further clinical studies are needed to establish whether these post-treatment expression signatures can be used as biomarkers of overall survival benefit upon ipilimumab therapy.

Early studies have demonstrated the utility of protein microarrays for the high-throughput measurement of antibody responses post-therapy [115,120]. In clinical studies of GVAX^® (Cell Genesys, CA, USA; GM-CSF-secreting whole cell immunotherapy) in prostate cancer, Nguyen et al. have reported the use of protein microarrays to identify specific antibody responses induced post-therapy [120]. An antibody response against TARP (T cell alternative reading frame protein, a protein linked to prostate carcinogenesis) was significantly associated with the survival duration [120]. Thus, protein microarrays appear to hold promise for the discovery of post-treatment serum antibody biomarkers of clinical outcome that can be translated for routine immune monitoring in patients. Such antibody responses have many of the characteristics of an ideal biomarker: readily accessible for quantitative analysis, and intimately tied to the in vivo therapeutic mechanism.

Expert commentary & five-year view

Omics technologies and systems biology are having an impact in many aspects of cancer immunotherapy discovery and development. There are elegant and practical examples of target discovery, mechanistic characterization and predictive or pharmacodynamic biomarker identification. We have been able to cover some of these in the current review. We believe we will see more successful applications in the near future, as new therapies are tested in clinical studies and new technologies for addressing the unique needs in immunotherapy become more mature and widely used. Application of the omics platforms has served as a complementary approach to the conventional immunological assays (such as multi-parameter flow-cytometry, ELISpot, ELISA and so on) for the generation of hypotheses and candidate biomarkers. Data from larger immunotherapy clinical trials will help to confirm the mechanistic themes and biomarkers that are emerging from the preclinical and early clinical studies. There are lessons to be learned from the omics applications in oncology, basic immunology and infectious diseases; advances in those areas should be leveraged for interpretation of the data that are being generated from the cancer immunotherapy studies.

New technological developments in omics platforms (e.g., NGS) are likely to affect the discovery efforts. In certain areas, multiple platforms have now become available, so researchers have the flexibility to use the one best suited to address the purpose of the study. For example, there are now a multitude of platforms for measurement of gene expression, including microarrays, NGS (RNA-sequencing) and multiplexed-qPCR. Therefore, a judicious selection of the appropriate, fit-for-purpose, platform is important. For addressing new target discovery or understanding MoA, RNA-sequencing or DNA microarrays may provide more value since they survey expression at the genome-wide level. For candidate biomarker discovery and translation to a limited quantitative panel, or for studying formalin-fixed, paraffin-embedded tumor tissues (where RNA is degraded), multiplexed qPCR or digital mRNA counting with color-coded probes can be a better choice [154].

Genomic analyses of whole blood and PBMCs have a tremendous potential for enabling the discovery of mechanisms and biomarkers associated with clinical outcome to immunotherapeutics. Methods are being developed for such analyses [19–21,39,99,106]. Data interpretation in this domain will benefit from the work being done to characterize the gene expression signatures associated with purified immune cell-types, or their differentiation, activation and signaling states. However, such approaches can have some issues. Methods for cell purification can alter their activation or signaling states, and certain expression patterns associated with purified cell-types or cellular activation may not be discernible from whole blood or PBMCs. Continued development and improvement of the methods for the analysis of complex immune cells will be important.

Discoveries using systems biology rely heavily upon the integration of heterogeneous data. Integration of omics data with clinical parameters or immune cell phenotypes (assessed from multi-parameter flow-cytometry) is often critical for understanding and interpretation of MoA and biomarkers [155,156]. Large amounts of data can be generated even from small-scale omics experiments, which require appropriate database management systems and software for storage, integration and mining. Data from the very same biological specimen or patient are frequently stored in heterogeneous representations, making their integration difficult. Biologists often lack basic understanding of data management issues and the appreciation of how their data can be integrated across platforms; and professionals in database management or software development may lack a basic understanding of the nature of the biological questions being addressed. This can create a technical or communication gap between the professionals who are involved in the integration and mining of the data.

It will therefore be most fruitful to approach the systems biology challenges in a multidisciplinary manner that brings together expertise from immunology, cancer biology, computational biology and statistics, along with the appropriate infrastructure and tools needed to integrate and analyze the experimental data generated from a multitude of platforms [156–158]. Appropriate statistical and network reconstruction methodologies are also needed for integrative mining of the data and generating comprehensive biological models from the omics data (e.g., [159,160]). Such environments and methodologies are essential to obtain the best mechanistic insights possible into the therapies and to identify the robust and interpretable biomarkers that can be translated into practical clinical use for the benefit of patients receiving cancer immunotherapies.

Key issues

• • The development of cancer immunotherapies is complex and challenging as it involves many poorly understood mechanisms of tumor targeting by the immune system and tumor’s immune evasion.

• • Unbiased systems biology approaches can be important for many aspects of cancer immunotherapy discovery and development, starting from new tumor antigen discovery to the discovery of biomarkers reflective of the activity and efficacy of immunotherapies.

• • There are many omics approaches and related analytical methods that are mature and well suited for the study of cancer immunotherapies, such as transcriptomic analyses of complex immune cells, mass spectrometry, T-cell receptor sequencing and protein microarrays.

• • Although omics data from large Phase III clinical trials is limited in cancer immunotherapy, the current data from preclinical and early stage clinical studies are showing several consistent mechanistic themes across different modalities.

• • Data from larger clinical studies are required to confirm the emerging mechanistic themes and for the validation of pre-treatment and pharmacodynamic biomarkers of clinical outcome. This should enable the identification of patients most likely to benefit from the immunotherapies, as well as faster development times and improved probability of success in future clinical testing.

• • Experimental and computational methods for omics analysis of complex immune cells (e.g., whole blood, PBMCs) should significantly benefit the discovery of biomarkers of clinical efficacy. More of such methods need to be developed and made widely available.

• • Integrative mining of data emerging from different platforms is often the key for obtaining deeper insights into MoA, new targets and biomarkers. To facilitate this, a diverse set of expertise is needed to support discovery and development efforts. It will be critical to use an interdisciplinary approach and integrative analysis in order to make rapid progress through the discovery and development cycles of cancer immunotherapies.

Acknowledgements

The authors thank Mark Frohlich for a critical review of the manuscript and many helpful comments. The authors also thank Kenneth Brasel and Harini Kandadi for useful discussions.

Financial & competing interests disclosure

The authors are all employees of Dendron Corporation. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

No writing assistance was utilized in the production of this manuscript.