Guest Commentary: Opportunities for personalized cancer immunotherapy directed by immune-profiling
The successful use of monoclonal antibodies that target CTLA-4-expessing cells, or that interrupt PD-1 signaling, has heralded a new era of cancer immunotherapy. However, not all types of cancer are sensitive to these approaches and, in those that are, too many patients do not respond. To broaden efficacy, the hunt is on for additional immune pathways that can be manipulated to improve response rates, particularly when used in combination with PD-1-blockade. Earlier this year, the reported failure of the IDO-inhibitor Epacadostat to enhance the response rate to anti-PD-1 treatment brought disappointment, highlighting the continuing inability to predict the clinical fortunes of what seem scientifically sound approaches.
How can this be improved? A widely held view is that we need to better understand a patient’s immune landscape, and particularly the immune landscape within their tumor(s). This should, in the future, allow clinicians to tailor the drugs used in order to trigger the appropriate immune response and/or inhibit the key immunosuppressive pathways that are at play, thereby delivering personalized immunotherapy. There are currently three main approaches to characterizing the immune landscape of solid tumors: transcriptional profiling, single-cell cytometric analyses and histology. Each of these disciplines brings its own benefits and, with rapid technological advances, powerful datasets are now achievable.
Transcriptional profiling allows collection of the broadest datasets. These can come from the use of commercially supplied packages, including “immunology,” “oncology” or “immune-oncology” panels, or from unbiased RNA-sequencing. Single-cell next-generation sequencing (NGS) currently provides the deepest insight into gene expression and is particularly powerful for discovery science.
However, whilst there will undoubtedly be interesting targets that are missed using defined panels, the ease and speed of data analysis (over the relatively cumbersome bioinformatics required for NGS) mean the more targeted analyses are likely to have greatest utility for day-to-day immune profiling. Recent advances have combined RNA and protein analyses from the same sample, such as NanoString’s nCounter system, which comes with the advantage of being able to handle FFPE-preserved material.
Whilst transcriptional analyses can provide breadth of information, in general this does not include information on levels of expression of particular genes in defined cell populations. Taking molecular profiling in the setting of PD-1 blockade as an exemplar, the ligand PDL-1 can be expressed by tumor cells, by infiltrating innate immune cells, or by PD-1+ T cells themselves. FACS sorting [fluorescence-activated cell sorting] to provide individual populations for downstream transcriptional analyses is feasible, but requires some a priori decisions regarding the cells of interest, as the number of cell populations that can be sorted from a single sample is limited. It also only tells us about expression at the population level, rather than at the individual cell level.
Single-cell NGS can bring the necessary resolution but, as discussed above, is unlikely to be widely used as a routine. Recently launched platforms seek to balance breadth of gene expression with single-cell detail and also functionality. For example, BD’s Rhapsody system uses microwells to capture RNA from individual cells on uniquely barcoded beads, prior to pooling, multiplex PCR and sequencing. Available gene panels provide coverage of hundreds of genes from thousands of sampled cells.
Cytometry in multiple dimensions: From light to mass to four-letter coding
Flow cytometry has been the workhorse technique of cellular immunology for decades, and fluorescently labeled antibodies remain the most widely used platform. The range of cell populations identifiable in a sample has been limited by the numbers of colors available. Although the most advanced cytometers can now confidently handle around 20 colors, newer technologies are pushing the boundaries further. Utilizing metal-conjugated antibodies, CyTOF combines flow cytometry with time-of-flight mass spectrometry to deliver powerful data-sets on the expression of dozens of molecules. But even CyTOF has so far remained in the double-digit range and will be restricted by the numbers of metals available for conjugation. It is also relatively slow (around 1,000 cells per second, less than a 1/10th the pace of traditional FACS) and it is considerably more expensive.
The consensus is that DNA barcoding offers a solution to really scale up molecular profiling. Antibodies are tagged with unique DNA sequences, labeled cells are lysed and the barcodes read on a DNA sequencer. This approach has the potential to be nuanced to include transcriptional analysis of the same sample. Although some other current technologies combine protein and mRNA profiling on the same sample (e.g. NanoString), these only provide data at the sample level rather than the single-cell level, so a technology that can offer this would be a step forward.
Whilst utilization of the methods described above can provide information on which cells are expressing what molecule, they cannot tell us where those cells are. Sticking with the example of PD-1 blockade, how valuable is it to know which cells are expressing PDL-1, if we do not know how close they are to T cells (that are or are not) expressing PD-1?
Histological examination has always been central to understanding tumor heterogeneity, and advances in multiplex staining, whole-slide scanning and image analysis permit extraction of qualitative or quantitative information with 2D integrity maintained. Tissue microarrays allow high-throughput screening, and slide scanning allows the morphological, pathological, spatial distribution and temporal changes resulting from treatment to be visualized across large areas of tissue.
Advances in multiplex detection technologies allow simultaneous detection of multiple immune biomarkers in a similar manner to flow cytometry, but with the added value that spatial visualization brings. There are currently two main approaches for high-plex staining. The SIMPLE method uses direct (no signal amplification) or indirect (low/moderate signal amplification) staining of individual targets.1 Once imaged, the sections undergo a combination of chromogen bleaching or fluorescence quenching and antibody stripping.
This process repeats iteratively until the required number of images is acquired, before merging the datasets into a rendered multiplex image. Alternatively, the highly-sensitive Tyramide-based detection methods employ a sequential approach for target detection, signal amplification and antibody stripping.2 Currently there can be six cycles of staining before imaging using a spectral imaging platform, but this should soon increase to nine. Tyramide detections currently deliver 7-plex staining, and can be combined with RNAScope for mRNA detection.
Both of the approaches outlined above have advantages and disadvantages, and both share the complexities of target validation and staining sequence. However, robotic staining systems and whole-slide scanners favor Tyramide-based approaches within a consistent high-throughput environment, as opposed to the SIMPLE method which is more interventionist and laborious. Ultra-high-plex approaches of the future may again call on DNA barcoding to raise the technical bar. The reagents and technologies supporting this field are evolving rapidly, and will require a new generation of skilled histotechnologists to deliver.
Which approach to choose?
Each of the approaches discussed here brings benefits and has limitations: NGS brings the greatest depth, but at a population level; single-cell analyses by can bring detail on relative expression of proteins or mRNA, but with reduced depth and a lack spatial information; histology brings that information, but here “multi-plexing” currently means the number of markers available in a fairly standard flow cytometry run. As ever, choice is dictated by the questions that need to be answered. Personalized medicine is delivered by asking, “What is the best (available) treatment for this patient?” For cancer immunotherapy, it is obvious that an understanding of the extent and quality of the immune infiltrate in a tumor will help to guide the clinician and patient. For simplicity, tumors have been divided into three categories determined by the presence and location of T cells—notably, “immune deserts,” “immune-excluded” or “inflamed.” Logically, and in broad terms, this order reflects sensitivity to current immunotherapies.
Spatial information from histology can have prognostic value. Using numbers and location of CD3+ and CD8+ T cells, HalioDx’s Immunoscore Colon test on resected tumors has shown value in predicting relapse risk for stage I-III colon cancer. This has not been extended to predictions of response to immunotherapy. Two immunohistochemistry tests for PDL-1 expression have been FDA-approved to accompany Keytruda (anti-PD-1) and Tecentriq (anti-PDL-1), and the search for other companion diagnostic tests will remain intense, particularly for the alternative immunotherapies currently in clinical trials.
Depending on the target pathways, these might well be better characterized by transcriptional or cytometric analyses than by histology. There are now some commercial providers offering CyTOF for broader characterization of immune cell populations in tumors, for example.
It is important to remember that we are dealing with an immune response. As such, ex-vivo characterization can only tell us so much and a great deal more can be learned by functionally testing that response; for example, a T cell’s response to T cell receptor activation, or an innate immune cell’s response to toll-like receptor activation (with or without addition of drug). Functional readouts might then be cytometric or biochemical characterization, changes in transcription, multiplexed analysis of secreted products or, of course, cytotoxic activity against tumor targets.
It is likely that the high-end, complex and expensive approaches discussed here will be best deployed to provide preclinical datasets. The true benefit will come from distilling the salient information from these to enable development of directed, validated and robust tests. This is in the distance, but a lack of technology will not be the obstacle.
Stephen M. Anderton is group head of translational biology at Concept Life Sciences and Michael Millar is chief technical officer (histology) at Aquila BioMedical.