The initial foray into genomic sequencing two decades ago has led to the discovery of key driver mutations and spurred drug development, thereby transforming the management of imminently lethal diseases.
For non-small cell lung cancer, specific, oncogene-directed therapies—for EGFR, BRAF, ERBB2, and MET mutations, ALK, ROS1, NTRK, and RET fusions, tumor mutation burden, microsatellite instability (MSI), and PD-L1 expression—have significantly improved patient outcomes.
Similarly, for patients with cancers possessing homologous recombination deficiency (HRD), molecular diagnosis has dramatically enhanced survival benefits for patients with breast, pancreatic, ovarian, and prostate cancers. Yet, these new paradigms only scratch the surface.
Precision medicine benefits less than 10% of cancer patients
As pointed out by critics, today’s precision medicine helps fewer than 10% of cancer patients. In part, this is because we don’t have drugs for every oncogenic driver and virtually none for transcription factors.
On the other hand, NGS annotation and diminutive panels limit molecular diagnosis to genes used for disease classification, drugs that have regulatory approval, and some drugs that are in clinical trials. But the cancer research enterprise has been able to credential 800 oncogenes, 1200 tumor suppressor genes, 450 genes involved in DNA repair and more than 2500 microRNA that regulate protein translation.
Gene expression analysis reveals the consequences of transcriptional regulation and epigenetic control, often presenting a different perspective than one gets from genome-only analysis. Finally, post-translational protein modifications in the proteome may be genomically and transcriptomically normal, but are prime mediators of the malignant cascade.
3D molecular diagnosis gives better picture of cancer biology
Multiomic analysis of the genome, transcriptome, and proteome reveal that cellular function is regulated in three separate but integrated dimensions, perhaps best exemplified by the Star Trek vision of three-dimensional chess. The disruption of these processes is mechanistically linked to the hallmark behaviors of cancer and determines whether particular treatment strategies succeed or fail.
A three-dimensional view of molecular abnormalities provides a vision of tumor biology equivalent to going from a 2D chest x-ray to a 3D CT scan or from an M-MODE echocardiogram to a 3D rendering of the heart beating.
Similarly, we can obtain profoundly illuminating insights from 3D molecular analysis. The number of pixels making up an image is also analogous to increasing the number of genes. The vast difference between a 0.3T and 3.0T MRI image represents a similar experience in going from 150-gene panels to whole exome (23,500-gene) panels.
Getting the fullest possible picture of both genomic and transcriptomic factors is the molecular equivalent of looking in both directions before crossing the street—at rush hour.
If you recall the difference of watching a sporting event on a 1960’s TV and the experience of viewing a 1024 x 768 high-definition screen, there is a similar analogy to create with the evolution of today’s genomic profiling.
In the former, one may barely see the ball; in the latter, not only can the viewer see the ball, but one also can appreciate the player’s facial expressions and see the effort, anxiety, and exuberance.
In fact, a vast amount of genomic information never makes the NGS report—especially chromosomal copy number aberrations—and physicians receive a highly filtered report. Overly stringent regulatory standards for validation may prevent official reporting of genomic abnormalities, even though the gene residing immediately adjacent to the one in question may be validated.
Predetermined thresholds filter out copy number gain less than 5x. Limitations of sequencing depth and library preparation prevent confident identification of shallow or even deep deletions throughout the chromosomal complement.
Nevertheless, hemizygous deletions have the equivalent impact as loss of function mutations, and modest CNV gains or amplifications imitate gain of function mutations in perturbing the signaling pathways they govern.
Though less than 5x amplification may be useful for excluding overt oncogene addiction, lesser forms of oncogene engagement may still be playing a role in disease pathogenesis or drug responsiveness.
By analogy, an addicted individual need not be totally addicted to narcotics or alcohol to have a substance abuse problem that impairs driving ability or psychosocial functioning.
Beyond knowing detailed CNV information including the base pair coordinates of deleted and amplified segments of chromosomes, functional copy number analysis based on transcript expression level simplifies the interpretive task, allowing us to discard CNV abnormalities that are compensated for by the transcriptional and epigenetic regulation.
Gene expression also identifies disease drivers emanating from dysregulated gene expression of otherwise normal genes. Thus, getting the fullest possible picture of both genomic and transcriptomic factors is the molecular equivalent of looking in both directions before crossing the street—at rush hour.
The cancer network is characterized by disease-specific nodes
The information that is missing depicts a far more complex disease network than is represented by showing a few mutations in minimally annotated NGS reports.
In fact, this complexity represents tumor biology. It provides a fascinating portrait of many dozens of dysregulated signaling pathways which display redundancy, crossover, convergence and positive feedback loops that conspire to activate a smaller number of key transcription factors and kinases that drive the malignant phenotype and hallmark behaviors of cancer.
The nodes of convergence in the network represent the master regulators. These govern the transcriptional regulation of the cancer and constitute the cells’ regulatory logic—making it malignant rather than normal and profoundly equipped to maintain homeostasis under the duress of treatment.
Besides homologous recombination deficiency, there are potentially dozens of synthetic lethal interactions—i.e., Achilles heel vulnerabilities—arising in a cancer network which also form key nodes in the cancer network.
These remarkable insights include the realization that these key disease-specific nodes in the network are most often arising from normal genes.
NTCT for complex cancers: Meeting the therapeutic imperative
Attacking multiple key disease-specific nodes in a complex network causes the network to collapse. This gives rise to a new approach to combinatorial therapy design, specifically employing network-targeting combination therapy (NTCT).
Conversely, perturbation of complex networks with single agent treatment strategies may only minimally perturb a network which is prone to adapt. Hence, the challenge for oncologists is to design a combinatorial therapy that overcomes this adaptation to fundamentally re-program cell fate or induce an irreparable insult that drives a definitive efficacy and survival benefit.
The commercial practice of defining the minimal number of genes needed to address regulatory drug approvals in panels fails to capture the complex biology of most solid tumor malignancies.
Fortunately, the raw data in commercially available WES panels contains information about all forms of DNA repair, oxidative stress resistance, evasion, uncontrolled proliferation, blockade of apoptosis, angiogenesis, invasion and metastases, chromosomal instability, and other hallmark behaviors of cancer.
Most of the time, a look at the raw data reveals an immense chromosomal instability often affecting hundreds of genes. A dive into gene expression profiling reveals the secrets of epigenetic and transcriptional dysregulation. Quantitative proteomics reveals protein expression abnormalities that were unsuspected from genome-only profiling.
Key disease-specific nodes are defined by master regulators and synthetic lethal vulnerabilities
Usually, NGS results reveal a striking and potentially infinite heterogeneity among diseases considered to be the same on clinical grounds and by light microscopy, thus creating an N-of-1 conundrum for studying cancer and designing individual treatments.
Conversely, biosimulation uncovers the cancer network and shows considerable overlap in disease-specific nodes, thus simplifying heterogeneity, providing a roadmap for combinatorial drug design, and making the problem of studying diverse N-of-1 diseases tractable.
By leveraging the totality of molecular results, computational biosimulation can dissect and elucidate the mechanisms of hallmark behaviors of cancer for each patient’s cancer, providing a functional anatomy of the disease drivers.
Ultimately, disease classification based on nodal patterns that drive specific disease subsets will lead to NTCT that promises vastly superior disease control and cancer survival.
Artificial intelligence for the practicing oncologist
Not surprisingly, the vast amount of new cancer biology is dauntingly complex. Additionally, the exponential growth in the knowledge base surpasses the ability of anyone, clinician or PhD scientist, to keep up. Thus, a chasm has emerged between the forefront of knowledge that can help patients and clinical practice.
Biosimulation is not just a mere luxury, but a necessity that transcends the paradigm of treating in the dark. It removes the uncertainty of not quite understanding what will happen when we give a drug or why one person’s disease is aggressive and another’s not so. It illuminates why one cancer responds to a particular strategy and another does not.
By accruing molecular information into a computational working model of normal and malignant cellular functions together with the latest insights from cancer biology that are relevant for an individual patient’s disease, biosimulation addresses the otherwise unmet need of understanding complex malignancies at a mechanistic level, stratifying the effectiveness specific treatment strategies (including chemotherapy, immunotherapy, radiation, and non-oncology drugs), and defines what must be done to take down a specific disease.
Without requiring that every practicing oncologist become an expert cancer researcher, biosimulation can bridge the gap between the patient in the clinic and the latest insights of cancer biology.
This comprehensive multidimensional picture of cancer provides an engineering-level insight into what makes an individual cancer unique in terms of its behavior and response to specific treatment strategies. Network modeling of cancer discloses the definitive vulnerabilities in the disease network.
Based on comprehensive genomic information, a biosimulated “digital twin” disease model can herald new insights that reveal what makes one person’s cancer similar to other cancers, but also identifiy the uniqueness in the signaling network that produces outlier or exceptional responses, or treatment opportunities that would be otherwise unsuspected.
Removing the molecular blindfold
In medicine, generally speaking, enhanced understanding leads to enhanced actionability. More than simply generating molecular data about the disease, computational biosimulation can be used to translate comprehensive molecular diagnosis into treatment insights pertaining to a wide range of treatment choices.
The molecular blindfold is removed by not simply ordering comprehensive NGS, but by developing a signaling pathway understanding of the biological consequences in a manner that provides actionable mechanistic insight into the disease.
In this sense, biosimulation is not just a mere luxury, but a necessity that transcends the paradigm of treating in the dark. It removes the uncertainty of not quite understanding what will happen when we give a drug or why one person’s disease is aggressive and another’s not so. It illuminates why one cancer responds to a particular strategy and another does not.
In most cases, individual drugs have a combination of sensitizing and resistance mechanisms. How are conflicting mechanisms resolved? With computational biosimulation, the relative contribution of various genetic aberrations, dysregulated signaling pathways and master regulators are integrated by the model in a manner that stratifies which approach is most likely to succeed.
Although the randomized trials tend to conclude that a higher number of drugs is more efficacious, biosimulation presents the magnitude of benefit for incremental strategies and reveals widely varying assessments of individual drugs among patients with different genomic aberrations.
As such, biosimulation permits tailoring treatment to exploit what is unique in the individual’s cancer as well as what is shared with other patients to administer the most appropriate therapy.
Designing personalized treatments
For more than 25 years, it has been inconceivable to manage a disease like breast cancer without knowing its HER2 status. In time, it will be just as inconceivable to design any patient’s treatment regimen without first understanding the complex adaptive biological network that defines the disease behavior and determines treatment response.
Just as it would be unthinkable to construct an aircraft without an engineer’s help and insight, clinicians will soon find it inconceivable to treat a patient in the molecular dark, or as aptly described by one of my patients: “Throwing it at the wall and hoping that it sticks.”
Instead we can capitalize upon multi-omic information to take precision medicine forward with a giant leap.
