This program is supported in part by educational grants from AstraZeneca, Bristol-Myers Squibb, Genentech, Incyte, and Prometheus Laboratories Inc.
Help the New York Academy of Sciences bring late-breaking scientific information about the COVID-19 pandemic to global audiences. Please make a tax-deductible gift today.DONATE
Posted July 21, 2017
In the fight against cancer, nothing since the advent of chemotherapy has shown as much promise as the emerging field of immunotherapy. Currently, billions in funding are being directed towards understanding the complex interplay between the immune system and the tumor microenvironment (TME). But as immuno-oncology (IO) becomes more sophisticated and increasingly prevalent in the clinic, the field requires new tools and techniques to better exploit the relationship between tumors and the immune system.
To date, studies have produced an enormous amount of new and significant data on the subject; in order to make sense of it, researchers have begun adapting quantitative approaches rooted in fields such as mathematics, physics, computational biology, bioinformatics, and bioengineering. This pioneering interdisciplinary work has opened up whole new arenas for study, including an understanding of innate and adaptive immune ecosystems, and the discovery of novel tumor antigens and tumor-intrinsic immune modulators.
This program is supported in part by educational grants from AstraZeneca, Bristol-Myers Squibb, Genentech, Incyte, and Prometheus Laboratories Inc.
On March 1, 2017, the Cancer and Signaling Discussion Group presented Quantitative Approaches in Immuno-Oncology to foster cross-disciplinary discussions of current immunotherapy research and ways to expand its uses going forward, highlighting new means for acquiring and interpreting data, cutting-edge tools to better understand the tumor microenvironment (TME), and innovative methodologies to enable precision immunotherapy to treat patients more effectively.
Keynote speaker Harlan Robins of Fred Hutchinson Cancer Research Center explained that many of the tools traditionally used to measure the immune response are not applicable in the case of cancer, as the tumor immune response is subtler, requiring more specific and sensitive tools. In contrast to other cells in our body, T and B cells of the adaptive immune system possess unique DNA, in order to attack a diverse range of pathogens. When a T or B cell responds and binds to a particular pathogen, receptor DNA clonally expands in a quantitative manner to effectively mount an immune response. Robins developed the immunoSEQ assay to overcome amplification bias that occurs in standard multiplex PCR to more accurately quantify T and B cell clones.
Sequencing the T cell receptor (TCR) allows researchers to quantify T cells within the tumor. This number can predict how well a patient will respond to immuno-oncology drugs, which act by heightening the immune response already at play within the tumor. For example, Robins found that the dendritic cell vaccine worked vastly better in glioblastoma patients with massive T cell infiltration than in those with lower T cell counts. In fact, this therapy increased survival time in these patients from 12 months to 70-80 months. Tumor-infiltrating lymphocyte (TIL) counts also had predictive value in evaluating the response to anti-programmed cell death (PD-1) therapy. Additionally, a more diverse T cell repertoire, with less specificity for the tumor, correlated negatively with treatment response. Altogether, this data indicate that immuno-incompetent patients, who are less likely to respond to immunotherapy, can be identified through quantification of TIL and TCR clonal expansion.
Uncovering aspects of the tumor immune ecosystem and tumor evolution has driven recent work by Dana Pe’er of Memorial Sloan Kettering Cancer Center. Pe’er is studying the effect of anti-PD1 or anti-cytotoxic T-lymphocyte associated protein (CTLA)-4 treatments on several T cell markers in mice to better understand their effects on the tumor immune ecosystem. Organizing the data in a PhenoGraph, which uses social network algorithms to detect single-cell phenotypes and subpopulations of T cells, Pe’er detected 15 distinct TIL clusters following the different treatments. Analysis of the T cell populations using this method indicated that different mechanisms of action exist for anti-PD1 and anti-CTLA4 treatments, with anti-CTLA4 being more robust and outperforming anti-PD1.
While looking at the canonical T cell markers was informative, Pe’er set out to get a more comprehensive, global view of the ecosystem. Pe’er developed BISCUIT— Bayesian Inference for Single Cell ClUstering and ImpuTing—which clusters and normalizes single-cell RNA-seq data based on different immune cell types. Using this method, Pe’er’s team created an atlas of the breast tumor immune system. Comparing blood, healthy breast, and cancerous breast tissues from the same patient, they found high diversity of tumor immune cell types and tissue specific differences in the tumor-immune environment with tumor infiltrating T cells being generally more active and cytotoxic than their healthy tissue counterparts.
Tumors evolve continuously in response to strong selective pressures from therapies, becoming more aggressive and resistant and creating a central obstacle to treating cancer. Dan A. Landau of Weill Cornell Medical College studies tumor evolution in chronic lymphocytic leukemia (CLL), a blood cancer that remains incurable despite various helpful therapies.
Using whole exome sequencing (WES) of a tumor, Landau can detect the frequency of a particular mutation. In doing this, he defines two types of clonal frequencies: the clonal mutations, which affect all cells, and the subclonal mutations, which affect a subpopulation. If a particular mutation occurs only in a subclonal population, this mutation probably occurred much later in the disease progression. Landau compared this to human evolution: most people have four limbs, with little diversity, whereas hair color is highly diverse. Thus, four limbs must have evolved earlier than the traits for hair color. Looking at large subsets of pre-treatment clones, Landau can determine which clones evolve over a certain time point, whereas other do not. Particular subclonal drivers were associated with worse prognosis and aggressive disease, representing a prognostic biomarker.
Similar techniques can be applied to study how cancer therapy accelerates tumor evolution. Using 59 CLL patient samples pre- and post-treatment, Landau found that 97% of CLLs significantly evolved in relapse. Additionally, many of these CLLs had expansion of a particular clone already detected in the pre-treatment sample. Using mathematical models, Landau can extrapolate forward to determine how large a clone size will be in a particular amount of time, or can extrapolate backward to determine the number of resistant clones that were present before the treatment began. In this manner, Landau can better understand the mechanism by which one clone can win over another and cause tumor evolution. Additionally, knowing the growth curves for resistant clones, Landau can predict earlier, possibly after a few months, whether a patient will respond to the therapy, rather than waiting several years.
It remains unclear why some cancers are unresponsive to checkpoint blockade. Christina Leslie of Memorial Sloan Kettering Cancer Center is using ATAC-seq techniques to study chromatin accessibility states in dysfunctional T cells of the tumor to see if epigenetic failure within the tumor acts as a barrier to checkpoint blockade. Data from a liver cancer mouse model indicate that the functional states of tumor-associated T cells correlate with epigenetic states. Eight days following expression of a mutant antigen, or neoantigen, the T cells remained plastic and susceptible to therapeutic reprogramming. However, at day 30, still with no tumors present, T cells were fixed and could no longer be rescued by therapeutics. Increased accessibility to transcription factor binding sites in the chromatin was responsible for driving T cell dysfunction and could be delayed through pharmacological inhibition of the transcription factors. Further, Leslie found that although tumor-specific T cells in the plastic and fixed states both had high levels of PD-1, the populations capable of therapeutic reprogramming had low levels of CD38 and CD101, whereas the fixed populations displayed high expression of CD38 and CD101. Thus, CD38/CD101 states may represent a biomarker to distinguish between plastic and fixed states and predict which patients are more likely to respond to immunotherapy.
The second session of the day consisted of data blitz presentations. Jacqueline Buros of Icahn School of Medicine at Mount Sinai discussed two types of biomarkers: predictive, to guide treatment selection, and prognostic, to predict overall outcome. But as the number of therapies and biomarkers increase, researchers still lack the means for acquiring the necessary data to fully unlock its potential. Buros suggests tailoring prescriptions by modeling treatment response mechanisms using all pertinent studies in order to aid biomarker discovery and direct treatment choices.
Anna S. Tocheva of New York University School of Medicine studies the signaling cascade downstream of PD-1 to better understand affected pathways following PD-1 ligation. Tocheva showed that immediately after ligation, PD-1 initiates a signaling cascade involving dephosphorylation of several key proteins in early TCR signaling events. Additionally, PD-1 ligation results in auto-phosphorylation of proteins associated with transcriptional regulation of cell cycle progression, cell proliferation and cellular metabolism that may represent potentially druggable targets.
The day’s final session revolved around computational modeling of the immune response. DNA of the adaptive immune system undergoes stochastic gene editing, known as VDJ recombination, to produce high diversity. Curtis G. Callan of Princeton University uses next-generation sequencing to assay the large, complex immune sequences from individual animals and calculate the probability that a particular sequence will be selected for TCR production. Callan developed an approach to infer the statistical ensemble upon which a TCR sequence is produced and what factors influence the choosing of this sequence. This approach can be applied to several applications to better analyze and understand immune function. For example, Callan can generate synthetic TCR repertoires to model the actual TCR sequences. A given neoantigen will be recognized by a finite set of complementarity-determining region (CDR)-3 of an antibody. This approach can be applied to immunotherapy to determine if any given human immune system will have a high probability of including a particular CDR3 for a given neoantigen.
To better understand immune repertoire diversity, Aleksandra Walczak of Ecole Normale Superieure is interested in the probability of particular TCR sequences occurring in different individuals. Surprisingly, there is a certain degree of random overlap between the prevalent clones amongst most individuals. The immune repertoire begins to form before birth, when the insertion enzyme—which later contributes largely to TCR diversity—is down-regulated. For this reason, individuals start out with a limited amount of diversity. Analysis of TCR sequence data from humans showed that many of the TCR clones produced pre-birth persist for approximately 36 years, explaining the commonalities between individuals. However, mouse studies showed that gene insertions quickly increase in the TCR after birth to expand diversity and increase immunity. Further, the optimal repertoire for two different people varies vastly even for the same environment, indicating personalized responses to the same stimuli.
Walczak’s group developed a high-throughput affinity and expression assay, Tite-Seq, to quantitatively define the link between an antibody’s sequence and its binding affinity for a particular antigen. Tite-Seq measures affinities for a library of mutant antibodies by analyzing binding titration curves. Using a mutational library of the CDR3H and CDR1H regions of the well-studied antibody scFv as a model, Walczak found that affinity is dependent on the number of contacts a residue makes within the receptor. Additionally, mutations in the CDR3 region had a larger effect on affinity and may explain why this region is more likely to contain mutations in functional receptors.
Many immunotherapies work because the tumor is not recognized by the immune system as ‘self’ due to mutations and the production of neoantigens. James R. Heath of the California Institute of Technology utilizes WES and RNA-seq on pre-therapy tumors to identify expressed mutations, and predicts the ability of these neoantigens to bind to the major histocompatibility complex (MHC). These neoantigen-MHC complexes are attached to metallic nanoparticles to create a highly sensitive “fishhook” to capture tumor-reactive T cells from patient tissues. Strikingly, this technique detected a baseline anti-tumor immune response in a glioblastoma patient with hardly any other traditional treatment options, with hope that they may respond to immunotherapy. Heath is currently delving deeper into how some neoantigens rather than others are selected, despite having stronger binding affinity for the MHC. Molecular dynamic simulations are beginning to give a better understanding of how binding and dissociation rates between the neoantigen and MHC influence neoantigen selection, and further studies are warranted for how the T cell interacts with the neoantigen.
Benjamin Greenbaum of Icahn School of Medicine at Mount Sinai is using observations from viral genome evolution to understand how tumor cells evolve and avoid self-recognition by the innate immune system. Viruses have evolved away from certain motifs to avoid recognition by pattern recognition receptors (PRRs) in the human immune system. Greenbaum created a mathematical model to quantify the entropic and selective forces guiding viral evolution to adapt to a new environment by avoiding certain motifs. Previous studies have identified the over-expression of satellite repeats and non-coding (nc) RNA in several cancers, with some ncRNA’s, like HSATII, being cancer-specific. Applying the viral-evolution algorithm to cancer cells, Greenbaum found that in the latter, the ncRNA expressed preferentially contain many repetitive elements that differ from the normal motif usage. HSATII is one such ncRNA that fell vastly outside the normal motif usage and was in fact immunostimulatory. However, the presence of HSATII correlates with poor prognosis and low TIL, possibly representing an example of ‘bad’ inflammation. A continuing application of the viral approach to cancer uses virus vaccine predictions to predict how well a patient and particular clone will respond to checkpoint blockade immunotherapies.
David R. Kaufman of Merck outlined three key biomarkers to accurately predict how well a patient will respond to the checkpoint blockade drug pembrolizumab, which acts by blocking the inhibitory PD-1 signals to activate the immune response. Previous studies have shown that pembrolizumab has increased activity in extending overall survival in non-small cell lung cancer patients expressing PD-L1, the ligand for PD-1, over patients without PD-L1 expression. Furthermore, patients with high mutational load attributed to deficient DNA mismatch repair (dMMR) and high microsatellite instability (MSI-H) have been shown to respond better to immunotherapy. The presence of many mutations increases the antigenicity of the tumor, making it look more foreign to the immune system upon treatment.
Kaufman introduced a new biomarker, a gene expression profile (GEP) of 18 genetic markers, including PD-L1, across multiple cell types, to give a generalized idea of the TME. A T cell-inflamed GEP correlates with a more favorable tumor immune microenvironment that can benefit from immunotherapy. Data indicate that this is a robust measurement, as good a biomarker as mutational load, and can be used independently or complementarily.