Today, there is a huge effort to develop cancer immunotherapeutics capable of combating cancer cells as well as the biological environment in which they can grow, adapt, and survive. For such treatments to benefit more patients, there is a great need to dissect the complex interplays between tumor cells and the host's immune system. Monitoring mechanisms of resistance to immunotherapeutics can delineate the evolution of key players capable of driving an efficacious antitumor immune response. In doing so, simultaneous and systematic interrogation of multiple biomarkers beyond single biomarker approaches needs to be undertaken. Zooming into cell-to-cell interactions using technological advancements with unprecedented cellular resolution such as single-cell spatial transcriptomics, advanced tissue histology approaches, and new molecular immune profiling tools promises to provide a unique level of molecular granularity of the tumor environment and may support better decision-making during drug development. This review will focus on how such technological tools are applied in clinical settings, to inform the underlying tumor–immune biology of patients and offer a deeper understanding of cancer immune responsiveness to immuno-oncology treatments.
Cancer immunotherapy has significantly revolutionized treatment of patients with cancer by improving overall survival, quality of life, and our understanding of the processes involved in the development of a clinically beneficial antitumor immune response. Despite these advances, only a small percentage of patients receiving immuno-oncology (IO) treatments experience durable clinical responses to-date (1). Exemplified by the high attrition rate of molecules in early clinical development, there is refueled interest to develop cancer immunotherapeutics capable of combating the biological environment in which the tumor grows and survives (2). Thus, the emerging question is: What additional considerations must be taken into account to improve further on patients benefiting from such treatments? One approach could be to undertake early clinical development trials of new immunotherapeutic modalities in a patient-centric fashion. However, this might not fully uncover the underlying biology and the antitumor immune drivers as seen by efforts over the past years, and the challenges remain pertinent (3). On the contrary, if during development our focus shifts toward a balance between patient- and mode-of-action (MOA)–centric approaches, this promises to decode the causal biology, delineate the key players, and may support better decision-making during drug development (4).
To broaden the success of IO therapies especially in the emerging era of IO combinations, it is imperative to dissect the complex interplays between host tumor cells and antitumor immune responses by considering the host's genetic makeup, immunocompetency, functional diversity, and immune-surveillance. Resistance to cancer immunotherapy is multifactorial and is identified in the form of primary, adaptive, or acquired resistance representing the constant battle for survival between tumor cells, immune cells, and the tumor microenvironment (5). Resistance mechanisms are rarely driven by a single mechanism and can range from constitutive programmed death-ligand 1 (PD-L1) expression, loss of human leucocyte antigen (HLA) expression, alterations of antigen processing machinery, alterations in IFN signaling, loss of target expression, T-cell exhaustion, suppressive immune populations, and many others (6). This necessitates investing into multidimensional assay platforms to characterize the resistance profile of each tumor allow concomitant interrogation of multiple biomarkers beyond single biomarker approaches to uncover such dynamics (7). Such multifactorial biomarker approaches may also contribute toward increasing the success rate of IO drug development by incorporating biomarker components such as patient enrichment strategies in early clinical trial designs (8). To date, there are more than 1,000 IO combinations currently undergoing clinical evaluations (9) and some have already demonstrated clinical benefit. These can include combinations between IO modalities (i.e., checkpoint inhibitors: PD-1, PD-L1, LAG3, etc.; immune modulators: IDO-1, CCR4, etc.; vaccines; ACT, etc.) and either radiotherapy, targeted therapies (MEK, VEGF, PI3K, etc.), chemotherapy agents, and others (10–13).
The purpose of this review is not to discuss every available innovative technology contributing to the aforementioned, but rather provide an overview of technological tools available today and how they are applied in new molecule clinical development to inform the underlying tumor–immune biology of patients and support informed decision making. Thus, elucidating the puzzle that drives effective anticancer immunity can inform the clinician under what conditions such treatment could provide maximum benefit to the patients. Zooming into cell-to-cell interactions using innovative single-cell spatial transcriptomics, advanced tissue histology approaches, and molecular immune profiling tools promise to delineate our considerations for improved IO drug design. Bringing these large and multifactorial datasets from the individual patient to larger patient cohorts requires, on the one hand, appropriate sample/matrix considerations and, on the other, concerted bioinformatic/computational approaches that consider assay variability and limited number of patients evaluated. These efforts promise to foster biomarker identification and transform clinical development by offering a deeper understanding of cancer immune responsiveness to IO.
Advanced Histopathology Tools to Dissect the Tumor Microenvironment
To understand the prevalence of IO targets, their cell-to-cell crosstalk, and their pharmacodynamic changes during treatment, it is of paramount importance to simultaneously assess the phenotype and spatial distribution of any cell or structure. Histopathology represents the most powerful way to understand cancer, identify treatment-related immune contexture changes within the tumor microenvironment (TME), and interpret other OMICs data. It is an integral part of assessments that can be used to inform early cancer immunotherapy drug development, contingent on considerations related to tissue location and its size, primary or metastatic, sampling time point, host tissue, and presence of invasive margin (14, 15). Pathology is now evolving into two orthogonal and complementary directions: multiplex stainings to enable cell phenotyping with spatial analysis and artificial intelligence (AI)–based approaches to extract more information from slides.
The constant TME evolution makes the identification and characterization of changes in such a spatially organized ecosystem rather challenging, especially because the various cellular components can assume diverse phenotypes. The presence of tumor-infiltrating lymphocytes (TILs) has shown prognostic significance using the definition of immune phenotypes based on CD8 cell density and distribution and may provide an entry point to immune tumor classification enabling patient stratification across different tumor types (16, 17). However, TME classification on such basis does not seem to be sufficient because prognosis can be influenced by several microscopic features such as tumor architecture, stroma quality, desmoplasia, tumor budding, topological distribution, and ratios of the different immune cells (or their combined location) and invasive growth patterns (18, 19). Hence, a deeper and more comprehensive understanding of the tumor–host interactions that not only incorporates cellular composition, but also the phenotype within a certain tissue compartment by combining immune-, cancer-, and host tissue–based parameters, can enable a more holistic understanding of each tumor setting (20).
The emergence of digital pathology (DP) together with multiplexing technologies on top of the routinely used tools allows for the concurrent quantification of several proteins, including the assessment of marker coexpression and their spatial location. Usually, up to five protein markers can be tested on the same formalin-fixed paraffin-embedded tissue (FFPET) section, because accuracy, reliability, and reproducibility can be challenging (21). The choice of platform is guided by the scientific questions to be addressed, the intended use, access, and cost (Table 1; ref. 22). To characterize further spatial resolution and TME molecular alterations, the following approaches have been considered: (i). libraries of oligonucleotide (barcode) directly conjugated antibodies incorporated on the slides and (ii) MS-based platforms with multiplexed ion beam imaging methods using metal-based labels (23, 24). Despite these advances, the major drawback of high-multiplexed technologies is that they do not allow for complete tumor architecture elucidation of the whole tissue sample because they usually focus on certain regions of interest. Finally, scanning conditions of the DP slide images and imaging algorithms must be validated and bioinformatic/statistical support is required to enable meaningful interpretation of the huge data volume.
|.||Key characteristics .|
|Technology .||Strengths .||Weakness .||Routine clinical application .|
|Advanced Pathology tools|
|Ag-specific T cells||pMHC||pMHC||pMHC|
|TCR VB||TCR VB||TCR VB|
|.||Key characteristics .|
|Technology .||Strengths .||Weakness .||Routine clinical application .|
|Advanced Pathology tools|
|Ag-specific T cells||pMHC||pMHC||pMHC|
|TCR VB||TCR VB||TCR VB|
Note: Several technologies are used to-date and are characterized by strengths and weaknesses. The choice is driven by how these could be used in a clinical setting and whether they can inform drug development.
Digitized slides are annotated to define or discriminate different compartments (normal/tumor tissue, tumor boundaries, cancer nests, stroma, etc.). AI tools based on machine learning/deep learning (ML/DL) algorithms are now widely developed and can support identification of such features (25). Algorithm-generated outputs include object counts and location and allow both cellular and spatial statistical analysis. Consequently, the validation of these algorithms is a necessary critical step prior to supporting clinical development.
Advancements in ML/DL can leverage both the full potential of each single slide and large existing datasets of digital whole-slide images. Image-derived features can be used for advanced triaging, diagnostics, and monitoring treatment-related immune contexture changes (26). Such approaches could inform patient selection/enrichment by developing composite scores using parameters, such as immunophenotyping, molecular signatures, stroma quality, tumor nest characteristics, and others. ML/DL-based algorithms can reveal in an unbiased and highly reproducible automated manner “hidden tissue features” not obvious to the pathologist. Consequently, new guidelines for the use of AI in the field of pathology are currently being rewritten (27, 28). Development of such algorithms however, presents with challenges such as the difficulty in using small datasets for training and validation, the need to overcome multiple tissue artifacts, tissue quality (e.g., fragmented tissue biopsies etc), access to properly curated and annotated datasets, and finally the demanding computer processing requirements (Table 1; ref. 29).
The use of multiplex approaches with ML/DL algorithms can generate an increasing amount of data requiring advanced visualization tools and pathologist-guided biostatistical analysis. The combination of these qualitative and quantitative spatial tumor descriptions with other biomarkers and clinical data is paving new ways in supporting early clinical cancer immunotherapy drug development.
Transcriptomic Approaches to Enable TME Characterization
To further uncover the drivers of effective antitumor immunity beyond the cellular phenotype and spatial distribution characterization of the TME, it is imperative to also undertake an in-depth characterization of the complex TME heterogeneity using genomic information to complement protein-based analysis. This is achieved by applying transcriptome analysis technologies such as RNA sequencing (RNA-seq), single-cell RNA sequencing (scRNAseq), and spatial transcriptomics.
Developed more than 10 years ago, RNA-sequencing enables transcriptome wide analysis (30). By isolating and sequencing the total amount of RNAs from a given sample, one can simultaneously study transcriptional profiles, differential gene expression, SNP characterization, as well as epigenetic RNA modifications (31). Only the total cellular content of mRNAs can be assessed though and the results represent average values of the pooled populations that can be influenced by the abundance of certain cell types that are not necessarily the drivers of an antitumor response. To overcome this, laser microdissected tissue was considered as an improved alternative (32). Consequently, because there is a requirement for significant RNA quality and quantity, these approaches are seldomly used in large multicenter oncology trials. With the exception of gene fusion and somatic mutations, most of the recent RNA-seq results assessing the tumor–immune-(stromal) contexture are mostly descriptive signatures. Although these data allow interrogation of the underlying resistance/escape mechanisms following CPI treatments, they have found limited applicability in early drug development albeit in some instances they offered prognostic significance (Table 1; ref. 33).
More recently, scRNAseq was developed in an effort to provide high resolution phenotypic and functional cellular analysis within the TME (34). It offers dissection of the crosstalk between tumor cells and TILs, identification of new tumor targets, responses to therapies, and can uncover new treatment options (35, 36). Using scRNAseq, the dynamic functional state of TILs and that of immunosuppressive regulatory T-cell subpopulations have been well studied and characterized and consequently these studies support one of the thesis that clinically relevant antitumor immunity is dependent on a pool of stem-cell–like memory effector T cells residing outside the tumor rather than preexisting antigen-experienced and potentially exhausted TILs (37–39). Stemming from their enormous phenotypic and functional plasticity, scRNAseq offers new possibilities to also study myeloid subpopulations and their origin in the TME (40). Kim and colleagues report that resident tumor-associated myeloid cells are gradually replaced by monocyte-derived immune suppressive cells (41). Moreover, using scRNAseq has led to the identification of immune modulatory and inflammatory fibroblasts and myofibrobalsts within the TME, suggesting that these stromal cells may be directly involved in cancer progression and resistance to cancer immunotherapy (42). The aforementioned can collectively decipher complex interactions among multiple cell types in TME and demonstrate how scRNAseq could be used in clinical settings to monitor such interactions.
Most of the emerging clinical scRNAseq data originate from freshly resected tumor material, because viable single-cell sorting prior to RNA analysis is required. However, such material is not readily available from patients participating in early clinical drug development trials. Moreover, on-treatment patient monitoring of new immunotherapeutic modalities requires repeated sampling and while this may be less of a concern for blood-based samples in hematologic diseases, the operational feasibility of fresh or fresh frozen paired (pre- and on-treatment) tumor biopsies for solid tumors needs to be carefully considered (Table 1). To this end, tissue single nucleus instead of scRNAseq might be a viable alternative for clinical routine monitoring, allowing more stable results for conserved tissue samples (43).
With scRNAseq technologies soon to become a potential analytic method for routine clinical assessment, the next evolution is already on the horizon. Spatial transcriptomic approaches are addressing the main drawback of scRNAseq, that being the required dissociation of cells from their physiologic spatial contexture, as well as the loss of significant cellular information (shape, cell–cell, and cell–matrix interactions). They use either directly labeled FISH or position-specific barcoded beads to provide high-resolution spatial mapping of expression profiles directly within the tissue and can have a resolution ranging from a hundred to nearly single-cell level (44). Emerging data suggest that the accumulation of immune infiltrates in certain regions of the tumor may be associated with tumor subclones, tumor-neoantigen burden, and the risk for relapse as described for patients with lung cancer (16). Similarly, compartmentalization of potential immune suppression and specific niches of tumor subpopulation–enriched gene networks were revealed in patients with squamous cell carcinoma (45).
A key challenge in developing and implementing OMICS technologies is the standardization and validation of the various analytic steps that need to be considered on the basis of the intended use to enable patient data comparability (Table 1; ref. 46). These include quality and quantity of total RNA needed, single cell/nuclei isolation technologies, library construction, the actual depth of sequencing, and finally the analysis of data (47). Although the latter remains a key challenge, it also offers new opportunities to link data from different other sources such as technologies described herein (48, 49). Despite the aforementioned limitations of OMICs technologies, emerging data from clinical trials may still be used to increase the confidence in the hypothesized MoA of single agent and combination treatments in early clinical cancer immunotherapy drug development.
Flow Cytometry and Analysis of Immune Cell Population Dynamics
In addition to tissue-based proteomic approaches, flow cytometry (FCM) is also a fundamental tool in IO. It can inform on the underlying tumor–immune biology interactions by enabling concomitant interrogation of multiple biomarkers across several immune cell types. Multiplexing allows highly sensitive analyses to be undertaken at a single-cell level that can uncover quantitative or qualitative changes and the unequivocal assignment of cellular identity and differentiation status. Depending on the intended use, markers of activation, antigen specificity, and exhaustion can simultaneously be quantified. This facilitates the deconvolution of immune cell population dynamics to an unprecedented granularity. Applications are extremely diverse and so is method complexity and instrumentation. This ranges from a highly targeted 3-/4-plex assay with short turnaround time to employing high-performance instruments capable of multiplexing 20 and more parameters requiring highly sophisticated data analysis capabilities (50, 51).
In early clinical cancer immunotherapy drug development trials, FCM allows detailed pharmacodynamic biomarker measurements intended to confirm MoA and demonstrate proof of mechanism (POM) for new immunotherapeutic modalities. An advantage in this context is the possibility to obtain informative results within a comparatively short turnaround time. Moreover, FCM is used to help dose finding by either employing receptor occupancy as a measure for target engagement (52), protein phosphorylation in case of kinase inhibitors (53), target cell depletion (54), or activation marker upregulation (55). For instance, FCM can be used for the detection of extracellular vesicles (EVs) as these have been implicated as prognostic and diagnostic markers in cancer and other diseases (further description is found in Liquid Biopsy section; ref. 56). However, challenges in detection have been associated with lack of sensitivity, sample preparation, and standardization issues, suggesting that at least some of the published studies may lack reproducibility. In addition, adjusting the configuration of instruments optimized solely for the detection of EVs is required, as well as development of advanced sample preparation strategies such as enrichment by an immunocapture method prior to analysis (57). Moreover, monitoring chimeric antigen receptor (CAR) T cells by FCM, can be a valuable alternative to conventional methods, such as PCR or sequencing. The added value over the molecular methods typically assessed on bulk tissue is that coexpression of functional markers and viability can be interrogated. This can for example, be highly relevant in the case of CAR T cells engineered to lack expression of additional markers designed to enhance or prolong successful engraftment (58, 59).
Sample types routinely used for FCM, are either whole blood or enriched cells from whole blood. However, as long as single-cell suspensions can be derived, any type of specimen can be used (bone marrow, solid tumor tissue, or lymph nodes). Regarding hematologic disorders, FCM is routinely used to inform diagnosis and treatment decisions and to determine prognosis or monitor minimal residual disease and treatment efficacy (60, 61). Solid tumor tissue biopsies are also used for FCM analysis but to a lesser extent. Reasons are mostly due to the amount of material required and the operational feasibility of obtaining paired samples from patients participating in early drug development trials (Table 1).
Overall, FCM is evolving rapidly with two major areas of development. The first is developing standardization solutions to increase reproducibility among laboratories and the second is to further enable higher multiplexing capabilities (47, 62). Alternative technologies such as mass cytometry or spectral cytometry can offer more in-depth immunophenotyping and enable the identification of novel functional immune cell subsets due to the more unbiased combination of markers- a consequence of higher multiplexing capabilities. This has, in turn, again fueled a fascinating increase of advances in the field (63). However, these more recent approaches remain highly exploratory with standardization still required prior to seeing them routinely applied in cancer drug development (64). Alternatively, large immune cell landscaping panels may be used for discovery, subsequently triggering the development of more targeted panels to confirm the initial findings in additional patient cohorts.
Among single-cell multiomic approaches, liquid biopsies are a rapidly evolving field that can aid delineation of the key antitumor drivers. Key advantages of this approach is that it is less invasive, enables longitudinal sampling and faster turnaround time than tissue biopsies (Table 1; ref. 65). Hence, it allows for disease monitoring with relative ease and can potentially identify earlier any recurrence. The broad term of liquid biopsies refers to circulating tumor DNA (ctDNA)-based testing, the identification of circulating tumor cells (CTCs), or tumor-derived exosomes, extracellular vesicles, or RNA material in the blood of patients with cancer (66, 67).
ctDNA levels may fluctuate during treatment and can correlate with disease response or relapse across a wide range of cancers (68). Although in early clinical cancer immunotherapy drug development, the utility of ctDNA technologies is mostly exploratory and focused on patient stratification and drug response monitoring, such technologies promise to revolutionize drug development and treatment (69). Using ctDNA to screen healthy individuals, cancer can be detected much earlier (70). Through disease risk determination and surveillance in neoadjuvant or adjuvant settings, patients with residual disease can be identified, enabling additional or prolonged therapy options to be considered (71). Screening for mutations in metastatic disease settings can potentially offer late-stage patients access to new investigational drugs in combination with targeted therapies (72). Moreover, ctDNA allows for drug efficacy monitoring and thus treatment decisions can be taken earlier leading potentially to improved patient outcomes (73). Finally, detection of secondary resistances and monitoring of patients in remission for cancer recurrence can be undertaken (74, 75).
ctDNA is emerging as a potential surrogate endpoint in cancer immunotherapy drug development (76). Analyses could predict the risk of progression in patients with long-term responses to, for example, PD-L1 inhibitors and combined with other tumor-extrinsic features, enable the identification of patients achieving durable clinical benefit after a single infusion of PD-L1 inhibitors, suggesting that ctDNA dynamics are influenced by both tumor and immune factors and can be used to classify patient outcomes (77, 78). Such studies provide the rationale for tailored treatment strategies in long-term responders to immunotherapy, favoring treatment discontinuation in patients with undetectable ctDNA and consolidative treatments in case of ctDNA persistence.
However, these tools also present several technical challenges such as the need to detect very low levels of ctDNA and differentiate tumoral versus nontumoral sources of circulating DNA. To this end, somatic fingerprinting–based technologies can utilize digital PCR or next-generation sequencing (NGS), with or without knowledge of patient's tumor mutational status (79). Clonal hematopoiesis of indeterminate potential (CHIP) can lead to false positives but can be corrected by sequencing peripheral blood mononuclear cells (PBMCs) and/or tumor tissues (80). Technologies based on DNA fragment methylation, utilize NGS to detect epigenetic signatures unique to ctDNA avoiding CHIP issues, requiring, however, signature validation (81).
CTCs are a selective representation of primary and metastatic tumor lesions detected and isolated noninvasively from blood. Enumeration of CTCs is an FDA approved test that can inform on disease progression and response to any intervening therapy (64). In the circulation, CTCs are subject to modulations by immune cells and their analysis can provide key insights related to tumor driver mutations, genetic makeup, and phenotype (82).
The number of CTCs among indications fluctuates and can be a prognostic biomarker associated with either worse progression-free survival (PFS) or a reduction in tumor burden (83). Beyond CTC enumeration, PD-L1 expression by CTCs is the most characterized after PD-1/PD-L1 blockade therapy (84, 85). The loss or reduction of PD-L1 expression on CTCs after treatment could predict clinical benefit (86, 87). These studies provide a strong rationale to use PD-L1 expression levels by CTCs as a response biomarker, although the findings must be confirmed in large randomized clinical studies. Furthermore, advancements in single-cell omics technologies as described in earlier sections, is now endowing the CTC field with the potential of extensively elucidating further tumor cell heterogeneity, clonal distribution, molecular fingerprint, and resistance mechanisms (88, 89). A major challenge for CTC analyses, however, remains their limited number in circulation requiring further development of current enrichment strategies.
Finally, tumor-derived exosomes and extracellular vesicles can also provide genetic information about the tumor and a real-time monitoring of tumor dynamics (90). Ascertaining their content especially with respect to types and levels of particular miRNAs, DNA, and proteins, can inform on tumor survival, invasion, metastasis, immunomodulation, etc., and potentially be used as surrogate biomarkers for diagnosis and treatment efficacy assessment (91, 92). However, challenges associated with their isolation, assay reproducibility, and clinical validation have hindered their broader use and applicability so far in early clinical cancer immunotherapy drug development.
Overall, liquid biopsies offer an attractive opportunity for early clinical drug development to use longitudinally minimally invasive molecular diagnostics, to rapidly inform benefit from new immunotherapeutic modalities. Although tumor biopsies remain the gold standard in characterizing TME contexture as a correlate to disease outcome and benefit as described above, with the ever-increasing sensitivity and specificity of assays measuring ctDNA and CTCs, it will not be long before liquid biopsy assays become routine clinical practice.
Detecting Ag-specific T Cells
For most IOs to be effective, there is a strong dependence on the availability of T-cell clones capable of recognizing tumor antigens (Ag), to provide effective antitumor killing. Such T cells, however, are generally of low frequency and their detection has been rather challenging till the discovery of tetrameric peptide-MHC-complexes (pMHC; ref. 93). pMHC are used in cancer immunotherapy to assess the quantity and specificity of a given T-cell clone.
Because the early days of Ag-specific T-cell detection with tetramers, the valency of pMHC has been greatly optimized by multimerizing monomeric pMHC on other scaffold molecules (94). These higher order pMHC, enable the detection of T cells with low-affinity T-cell receptor (TCR)/pMHC interactions and achieve sensitivity of detection as low as 0.02% of the total CD8 T-cell pool. By employing heavy metal–tagged pMHC for use with mass cytometry (95), high-dimensional data on Ag-specific T cells have been generated and through combinatorial staining with metal-tagged tetramers, it is now possible to quantitate frequencies in the range of 0.001%–0.01% (96).
The elucidation of tumor mutational burden (TMB), representing total number of somatic mutation frequencies in cancer cells (97), as a response biomarker to checkpoint blockade necessitates the enumeration of T cells recognizing tumor-presented neoantigens (peptides derived from mutated genes). pMHC can identify potentially immunogenic neoantigens among hundreds of tumor mutations. Using metal-tagged pMHC, we recently showed that neoantigen-specific CD8 T cells can be consistently detected in patients with NSCLC during treatment with, for example, anti–PD-L1 antibodies such as atezolizumab. Interestingly, neoantigen-specific T cells of responding patients typically exhibit a differentiated effector phenotype compared with a memory-like one of patients with progressive disease (95). Rizvi and colleagues, was able to correlate tumor regression in patients with NSCLC with neoantigen-specific T-cell responses in periphery (98).
However, implementing pMHC more widely in early cancer immunotherapy trials is rather challenging. Tumor-Ag–specific T cells are rare in circulation and TIL analysis is difficult to undertake due to limited access to tumor biopsies. Furthermore, the highly polymorphic nature of HLA limits pMHC use to only higher abundance HLA alleles. Finally, due to the inherent limitations of the TCR/pMHC interaction, fully functional T cells are hard to detect because the TCR affinity threshold required for staining with pMHC is higher than that required for efficient T-cell activation. To this end, using magnetic nanoparticles presenting neoantigen-loaded pMHC multimers at high avidity by barcoded DNA linkers, allows the detection of the desired cell population (99).
Nevertheless, the direct quantification of tumor-specific T cells may guide development of new immunotherapeutic modalities. In patients with melanoma, circulating T cells capable of recognizing common melanoma–associated antigens had a strong positive prognostic impact on patient survival (100). Furthermore, identification and characterization of rare tumor-specific T cells using pMHC can also be a powerful pharmacodynamic biomarker to measure ongoing antitumor responses and may explain why some patients can respond to immunotherapies. Changes in pMHC-positive T cells in periphery of patients with renal cell carcinoma (RCC) correlated with increased TILs posttreatment with atezolizumab and bevacizumab in metastatic RCC (101). Because tumor Ags vary by indication and mostly consist of neoantigens unique to the individual patient, a truly personalized immune monitoring approach is urgently needed to enable pMHC analysis in IO trials.
Among the four different mature TCR chains, the functional TCR-Vb chain is the most diverse and is encoded by numerous combinations that enable sensitive differentiation within and among individual repertoires. Analyzing this TCR-Vb diversity in blood or tissue samples either at baseline or longitudinally can inform on the quantitative representation of certain T-cell clones in terms of other cell types, diversity, and clonality (102). Responders among PD-1/PD-L1 blockade exhibited both higher TCR-Vb clonality prior to treatment and higher clonal expansion posttreatment. These findings suggest that the expansion of posttreatment T-cell clones consisted of novel clonotypes (103–105). More recently, blood TCRseq could predict responders and showed good correlation with tumor TCRseq (34, 106). This exciting new development is in line with other findings demonstrating that circulating tumor-reactive T cells can be isolated from the blood of with cancer. Consequently, it opens the door for further analyses of such cells irrespective of the availability of a tumor biopsy. In addition, single-cell TCRseq is emerging as an enhanced technology, which apart from informing on T-cell clonality and diversity, can potentially be used to identify tumor-reactive TCR heterodimers that can be utilized for adoptive T-cell therapy approaches (107).
With many different novel immunotherapies and combinations thereof, currently tested in the clinic, the tracking of Ag-specific T-cell responses using pMHC and TCR-Vb, could be a useful biomarker (Table 1). It may contribute toward a better appreciation of whether such therapies can trigger an effective antitumor T-cell response as a surrogate of POM demonstration.
Additional Technologies Utilized in IO
As mentioned earlier, TMB is regarded as a key element of T-cell–mediated antitumor immunity and has been shown to predict survival after cancer immunotherapy (108). As a response prediction marker in the context of checkpoint inhibitors, TMB was reported to outperform PD-L1 expression status in certain indications (109). Although FDA granted approval of TMB as a clinical decision-making biomarker for pembrolizumab in June 2020, it is still controversially discussed (110). TMB can also be assessed in circulation from blood ctDNA (bTMB) and several studies are exploring the association between bTMB, prognosis, and IO response prediction (111). Assessing mutational burden with whole-genome sequencing in the clinical setting, however, is rather challenging and thus targeted gene panels are employed to facilitate data analysis and interpretation (Table 1; ref. 112). However, although quite advanced as clinical decision enabling tool, standardization and harmonization of TMB approaches is still to be established to enable comparisons of clinical patient data across different tumor types and treatment interventions (113).
Soluble cytokine monitoring
Cytokine monitoring in early clinical cancer drug development can be used as a functional measure of immune activation, either as a prognostic/pharmacodynamic assessment tool or for ascertaining the safety profile of treatment (114). Nowadays, multiplex protein arrays offer rapid analysis of several analytes quantifying the totality of the cytokine milieu. More recently, transcriptomic approaches are becoming more popular, because they can provide more granularity into the cellular contributors of antitumor immunity (115).
Most technologies offer rapid and robust readouts and are analytically validated. Sampling for peripheral cytokine measurements is noninvasive and can be performed longitudinally. However, careful consideration must be given on the matrix used, collection material, sample preparation, and long-term stability (116). However, establishing upper and lower detection limits of quantification for high and low abundance cytokines is rather challenging and the latter has prohibited such assessments using tumor tissue so far (117).
Elevated levels of several growth-promoting and antiapoptotic inflammatory cytokines at diagnosis have been generally associated with poor prognosis (118). Similarly, proinflammatory and anti-inflammatory cytokines levels after IO treatment correlated with a trend toward better recurrence-free survival or with stable disease (119, 120). After the safety events of the agonistic anti-CD28 TGN1412 phase I trial, cytokine monitoring as safety biomarkers became key in early cancer drug development. Sudden and dramatic posttreatment increases in several cytokines demonstrates how an immune-mediated cytokine storm could culminate in multiorgan failure (121). Consequently, given the spectrum of biological events that can induce clinically manifested CRS, accurate monitoring, and reporting of plasma cytokine alterations following early cancer drug development is strongly recommended. Overall, monitoring cytokine changes remains a formidable challenge that can provide, however, extremely valuable information on the intricate interactions of the antitumor immune response (Table 1).
The use of qualitative and quantitative outputs of medical imaging has received renewed interest over the recent years as a noninvasive and high-throughput approach to support disease characterization. Radiomics applies advanced image-based tumor phenotype feature analyses, such as tumor intensity, shape, texture, pixel interrelationships, etc. from large sets of radiographic medical images to identify patterns irrespective of the radiographic modality (PET, MRI, and CT; refs. 122–124). These imaging outputs can subsequently be incorporated with ML-based analyses that link them with clinical data or biological features (i.e., genomics, proteomics, histopathology, etc.).
The goal with radiomics is to uncover prognostic signatures and disease characteristics that can inform for instance on how patients can respond to treatment. In the context of cancer immunotherapy, several reports describe radiomics-based signatures with immunologic features such as intratumoral immune cell infiltration and response to checkpoint inhibitors (125–128). To this end, radiomics enables the simultaneous analysis of multiple lesions as compared with biopsy-based technologies that focus only on a single tumor lesion. The statistical connection of imaging patterns to clinical and molecular features expands our opportunities to include the TME in its entirety to our analyses. However, on the critical side it needs to be mentioned that (as for all mathematical evidence–driven correlations) that a biological causality is not necessarily given (129, 130).
As a “not so new kid on the block” the gut microbiome is a rapidly emerging area of investigation in cancer immunotherapy and collectively refers to the diverse bacteria, fungi, and viruses that colonize an individual's gastrointestinal tract and are identified using NGS. Stool collections have been and are rapidly being incorporated into clinical protocols as a means to characterize the gut microbiome and its possible correlation with response to IO treatment (131, 132).
Antibiotic usage was demonstrated to negatively correlate with response to cancer immunotherapies but the contributions of timing and class of antibiotic treatment is still under investigation (133). Gut bacterial composition was shown to predict both clinical response to PD-1/PD-L1 blockade and safety (134–137). Although specific bacterial species related to treatment response can vary across studies, patients with greater baseline bacterial diversity consistently responded more efficiently to therapy. Aggregate analyses across healthy and cancer cohorts revealed an overall shift in the gut microbiomes of patients with nongastrointestinal cancers compared with healthy donors. This includes an overall decrease in Shannon diversity, which takes into account the richness and evenness of species, in patients with cancer compared with their age- and sex-matched healthy counterparts (138).
Several ongoing clinical trials are evaluating the safety and efficacy of checkpoint inhibitors in combination with microbial therapeutics, following the preclinical findings that oral administration of 11 gut bacteria elicited both local and systemic CD8+ T-cell responses suggesting their biotherapeutic effectiveness (139). In these trials, multiple types of microbial interventions are being tested including fecal microbiome transplants or transplant of a complete stool sample, single species of bacteria and consortia, or rationally selected groups of bacteria (140–142).
Initial data highlighting the role of gut microbes in immunotherapy are exciting, but the field is young and hampered by small sample numbers that limit robust statistical analyses. In addition, mechanistic insight into how these gut microbes promote systemic antitumor immunity remains elusive (Table 1). Although there is still work to be done and no microbial therapeutics are yet approved for cancer patient treatment, the potential for the gut microbiome to unlock new areas of biology, and subsequently, novel targets and biomarkers for antitumor immunity is promising and worthy of continued investigation.
Concluding Remarks and Outlook
Immune responsiveness to cancer presents with an enhanced structural and functional diversity among individuals. Deciphering the nature of the intricate interplays between what is usually an immune-resistant cancer phenotype and the drivers of an effective antitumor immune response, represents the cornerstone of successful early clinical cancer immunotherapy drug development. Consequently, our increased understanding of the tumor-immune contexture is revolutionizing how biomarker research enables a thorough characterization of the dynamic interactions that drive antitumor immunity and how cancer immunotherapy drug design is approached. The ultimate goal is to identify not only patients that are better suited for a given IO treatment, but also why not all patients respond by achieving safely, a durable clinical response. These efforts promise the development of true precision IO-guided therapies. Given the enormous diversity of antitumor immune response, it is very likely that no single biomarker will become the sole predictor of response to IO for any given indication. Rather, a multifactorial composite biomarker workflow, as depicted in Fig. 1A, that includes several biomarkers analyzed using advanced bioinformatic tools, is expected to better reflect the comprehensive efforts needed to guide early clinical cancer immunotherapy drug development because the combined outcome from different technologies in this holistic approach holds more power.
Preceding any selection of specific technology approaches supporting cancer immunotherapy development, the respective key biological hypotheses and questions need to be formulated according to the IO drug development stage. For instance, if as per the MOA of a new drug it is expected to overcome resistance or escape mechanisms by engaging new T cells in the tumor, to fully understand the drugs' potential, the aim would be to monitor both pre/posttreatment immune contexture changes occurring in the TME (Fig. 1B). To achieve this to a significant level of granularity, characterization not just of T-cell–intrinsic factors but also of T-cell–extrinsic ones, need to be considered. Consequently, the identification of how and which actionable technologies can specifically support the intended outcome is crucial. A key consideration must not only be the potential readouts offered by different tools but also, especially for prospective patient sampling approaches, the balance between scientific rationale, interventional patient burden, and operational feasibility. Special attention at this step must be given as to how the tailored selection and combination of individual technological approaches may be integrated to maximize the outcome from any given scientific query. This type of composite approach when applied early enough in clinical cancer immunotherapy drug development can offer physicians and drug developers the option in identifying patients that might not benefit to switch treatment prior to any tumor progression. Following this forward approach, the integrated analyses and resulting data related to a given question, can leverage efforts toward a reverse approach whereby a specific technology (-ies) can be considered to address emerging questions related to furthering drug development. Finally, as for all technologies discussed herein and particularly if composite scores are developed, the generalization of the findings might not be broader applicable, if the same markers within a signature are not validated across different studies and if outcome data are not appropriately tested in training and validation cohorts.
In conclusion, in the age of refueled interest to develop cancer immunotherapeutics capable of driving beneficial and sustained antitumor immune responses, generating longitudinal multiomics data, promises to lead into new biomarker discoveries. We are entering a new era of true precision IO that necessitates the unearthing of predictive biomarkers, understanding of mechanisms of resistance, and design of combination therapies with higher clinical success. These new ways of thinking can only be instigated by implementing robust, validated, spatial omics modalities coupled to systems biology bioinformatics approaches. Key to this success is the use of comprehensive workflows to better and more critically inform early clinical cancer immunotherapy drug development.
M.A. Cannarile reports other from Roche Diagnostics GmbH during the conduct of the study and outside the submitted work. B. Gomes is an employee of F. Hoffmann-La Roche AG. M. Canamero reports other from Roche Diagnostics GmbH during the conduct of the study and outside the submitted work; in addition, M. Canamero has a patent pending for Metric to quantitatively represent an effect of a treatment. B. Reis reports other from Hoffman-La Roche Ltd. outside the submitted work. A. Byrd is an employee of Genentech Inc. J. Charo reports a patent for WO2018220100A1 issued, a patent for WO/2000/012121 issued, a patent for WO/2001/016174 issued, a patent for WO/2008/006458 issued, a patent for WO 2007/107380 issued, and a patent for EP20100163251, US 2011/0189141 A1 issued. M. Yadav is an employee and stockholder of Genentech/Roche. V. Karanikas reports other from F. Hoffmann-La Roche AG during the conduct of the study and outside the submitted work.
We thank all patients enrolled in early clinical development oncology trials and agreeing to provide specimens for biomarker analyses without which the advancements described herein would have not been made possible. We thank all members of the Early Clinical and Biomarker development groups at Roche-Genentech for ongoing discussions, collaboration, and scientific contributions over the past several years.