Treating Cancer Beyond Tissue of Origin
Researchers are increasingly concluding that the conventional classification of cancer, based on organ or tissue of origin, limits the effectiveness of treatment provided to patients. Patients with a particular type of cancer, such as lung cancer, have traditionally been treated using a defined set of drugs. This set is determined via clinical trials enrolling patients based exclusively on that specific tissue of origin. But as researchers and regulators are learning, cancer should be classified and treated based on mutational signatures of patient genomic profiles rather than tissue of origin, which delivers personalized medicine to each patient.
Here, we look at research conducted on classifying cancer based on biomarker expression levels that provide better target specification for therapies, irrespective of tissue of origin. Tissue agonistic approaches are central to the Cellworks multi-omic biosimulation.
Biomarker Expression Classification
The Pan-Cancer analysis project was launched by The Cancer Genome Atlas (TCGA) to analyze similarities and differences across tissue types and to identify molecular alterations that define cancer lineages or transcend lineages and re-group tumors from various tissue types. The ultimate aim of this project was to impact clinical decision-making which could give rise to clinical trials designed on the basis of biomarkers that bring together sub set of tumors from different tissues.
TCGA published their analysis of tumors at different molecular levels - epigenomic (chromatin), genomic (DNA), gene expression (RNA), and proteomic (protein) levels. Results showed that most tumor samples bear unique alterations within each specific molecular profile, which are not necessarily present in other samples from the same tissue of origin, hinting at the importance of personalization in treating cancer. Varying molecular alterations in different samples (even samples from different tissues) can lead to malfunctioning of the same signaling pathway and can be aggregated together as a class. This opens up the possibility of using therapies that are effective in treating one tissue of origin class to instead treat different tissue of origin classes with similar molecular profiles.
In one of the studies from the Pan-Cancer project published by F Chen, entitled Multiplatform-based molecular subtypes of non-small-cell lung cancer, non-small-cell lung cancer (NSCLC) cases were characterized with at least three methods to identify molecular profile. As a result of the analysis, nine major subtypes of NSCLC were discovered (Table 1). This classification is clinically significant and is based on the expression level of key biomarkers, which can be targeted with appropriate drugs. For example, some of these subtypes bear high expression of checkpoint genes that should lead clinicians to use checkpoint inhibitor therapy in patients with similar genomic make-up. On the other hand, another subtype showed increased activation of p38/MAP Kinase and PI3K/AKT/mTOR pathways that warrant the use of p38 and mTOR inhibitors to curb the tumor growth.
Table 1. NSCLC subtypes and their therapeutic implications. Adapted from Chen et al.
High SOX2 (but lower compared to SQ. 2a/SQ.2b); CT antigen expression
Higher SOX2 compared to SQ.1; CT antigen expression; better OS association
Higher SOX2 compared to SQ.1; distinct methylation patterns from those of SQ.2a; CT antigen expression; better OS association
LCNEC-associated; poor AD differentiation; CT antigen expression
Platinum-based drugs; Immunotherapy (CT)
CIMP; high immune cell infiltrates; immune checkpoint pathway activation
High immune cell infiltrates; CT antigen expression; immune checkpoint pathway activation
Immunotherapy (CT); Immune checkpoint
High immune cell infiltrates; immune checkpoint pathway activation; lower neoepitope count and mutation rate; better OS association
p38; mTOR; Immune checkpoint
High proportion of never-smoker/long-term non-smoker patients; lower mutation rate; high p38 and mTOR pathway activation; better OS association
CIMP; high proportion of never-smoker/long-term non-smoker patients; lower mutation rate; high p38 and mTOR pathway activation; better OS association
Selecting Drugs to Treat a Mutational Network
The Cellworks Omics Biosimulation platform selects drugs based on the gene mutations and interactions that characterize a specific patient's tumor. If check-point genes were found to be the drivers in that patient's cancer, check-point inhibitors would be among the selected drugs tested on a patient's profile in silico. Similarly, if the MAP Kinase or PI3K/AKT/mTOR cascades were upregulated, the corresponding inhibitors would be part of the selection. However, the problem may not be limited to those signaling cascades. Given that other driver mutations could exist in the patient profile, the final recommendation would be based on the interaction between driver pathways, selecting drugs that would be efficacious on the mutational network as a whole and not limited to suppressing one pathway alone.
A Closer Look at Molecular Characterization
Let's go a bit deeper into this topic starting with an extensive pan-cancer study led by Zhang Y., where researchers focused on PI3K/AKT/mTOR pathway and examined its components in more than 10,000 tumor samples of 32 cancer types. By using multiple molecular characterization platforms, several subsets of genetic variants related to this pathway that could have functional relevance have been identified:
- Cases with non-silent somatic mutation or copy alteration involving selected PI3K/AKT/mTOR pathway members (n = 4,468 cases)
- Additional cases with non-silent mutation involving selected receptor tyrosine kinase (RTK)-associated genes (n = 415 cases)
- Cases with high phospho-AKT but with none of the above somatic alterations (n = 764 cases)
- Cases with low phospho-AMPK but with none of the above somatic alterations (n = 394 cases)
- Cases not aligned with any of the above (unaligned, n = 1,058 cases).
Manual Decisions vs Computational Modeling
Stratifying patients according to mutational status of genes involved in this pathway might increase the response rate to mTOR inhibitors. Some cancers however, showed high mTOR pathway activity, without genomic alterations in the genes related to this pathway, possibly due to epigenetic alterations. This highlights the importance of data types (e.g. protein activation) beyond genomic sequence information and the importance of integrating multiple data types to evaluate a patient's tumor to guide treatment. Too much information makes manual decisions complex and difficult.
The Cellworks computational model automates the processing of patient tumor data. The more omic information provided the better. In the example described above, had the epigenetic data been missed, it would impact the recommended therapy and in turn the patient response. A patient profile could have multiple upregulated pathways, working simultaneously to drive tumor growth. Targeting one driver pathway based on a one-mutation one-drug approach may appear to do the job. However, another pathway could compensate and override the effect of the drug making treatment ineffective.
Eliminating Obstructing Signals
It is important to recognize that signals caused by molecular signatures that define histological background of cancer might obstruct the detection of more relevant signals identifying cancer subsets transcending the tissue of origin. In a third Pan-Cancer study, Creighton et al. described a computational way to eliminate such obstructing signals and classify cancers without this confounding effect. By using gene expression data of 10,224 tumors of 32 major types, they came up with 10 molecular classes (Table 2). These included classes characterized with increased immune checkpoint gene expression, CT antigen expression and the class that comprise 4% all cancers and showed the features of neuroendocrine tumors, all of which may have clinical relevance in diagnosis and therapy.
Table 2. Tissue-independent molecular classes. Adapted from Creighton et al.
High differential expression of oxidative phosphorylation genes, glycolysis genes, and pentose phosphate pathway genes.
Lack of strong associated expression patterns; can serve as a comparison group for the other classes.
Strong association with immune checkpoint pathway; differential expression profile associated with immune cell infiltration; mesenchymal signature; NRF2/KEAP1 pathway signature; Wnt pathway signature.
Differential expression profile associated with neuroendocrine tumors and with normal cells and tissues of the central nervous system; CT antigen expression.
Represents basal‐like breast cancer; TP53‐related alterations, MYC amplification and expression; YAP1 target expression; high expression of pentose phosphate and TCA cycle genes; immune checkpoint pathway; CT antigen expression.
Epithelial signature; normoxia signature; YAP1 target expression.
Mesenchymal signature; hypoxia signature; Wnt pathway signature; Notch pathway signature; NRF2/KEAP1 pathway signature; low differential expression of miR‐200.
High differential expression of fatty acid metabolism genes; mesenchymal signature; hypoxia signature; Wnt pathway signature; Notch pathway signature; NRF2/KEAP1 pathway signature; high differential DNA methylation and low differential expression of miR‐200; differential expression profile associated with normal cells and tissues of the central nervous system; immune checkpoint pathway (observed in TCGA cohort only).
Wnt pathway signature; Notch pathway signature; NRF2/KEAP1 pathway signature.
Immune checkpoint pathway; differential expression profile associated with immune cell infiltration; YAP1 target expression.
This classification serves patients well by looking at more than one pathway or aberration type. However, clustering patients into a fixed set of classes, by definition means a) focusing on certain tumor signatures that are representative of the class, and b) ignoring mutations that are not signatures of the class. We could personalize each patient into their own class - in keeping with the fact that each patient's cancer is unique. This concept is core to the Cellworks Omics Biosimulation approach.
These studies highlight the importance of molecular characterization of a tumor at different levels and integration and interpretation of the characterization data in guiding clinical decisions. We are presented with novel classifications that point to clinically relevant differences among tumors of the same tissue, or tumors bearing the same aberrant signaling pathway among tumors of different tissue. However, this is still an attempt at grouping patients into classes.
Prescribing Treatments Based on a Patient's Mutational Profile
Taking a patient's mutanome in its entirety is likely to place each patient in a class of their own. Cellworks interprets a patient's whole exome sequence, and epigenomic profile to determine appropriate targeted therapies for the patient. Cellworks' genomic disease physiology modeling and biosimulation techniques allow us to identify treatments personalized for each patient and provide therapy options where no standard exists. Cellworks' biosimulation selects drugs that will be effective for a patient based on their mutational profile, and not the tissue of origin. This approach improves treatment efficacy and avoids the unwanted side effects caused by ineffective treatment regimes, for patients predicted to be non-responsive to standard therapies. Prescribing treatment based on a patient's mutational profile, is the most effective approach to treating this heterogenous disease.