Today we’re joined by Ivan Kairatov, a biopharma expert whose work at the intersection of technology and oncology is reshaping our understanding of cancer. We’ll be exploring a breakthrough in liver cancer research, delving into the specific molecular triggers that drive tumor growth, the innovative use of spatial transcriptomics to map the precancerous environment, and how machine learning is being harnessed to create a powerful predictive score that could identify high-risk patients long before tumors become a life-threatening reality.
The protein MYCN has long been linked to liver cancer. How did your mouse model, which combined MYCN overexpression with active AKT, clarify its direct role in tumorigenesis, and what does this specific pathway suggest for developing future therapeutic interventions?
For years, MYCN was a known suspect in liver cancer, but its precise role was murky. We needed to move beyond correlation to causation. Our approach was to create a mouse model that didn’t just have MYCN present but forced its overexpression alongside another key player, an always-active form of AKT. The results were dramatic and unequivocal. A staggering 72% of these mice developed aggressive liver tumors within just 50 days. This wasn’t a subtle change; it was a powerful demonstration that this specific combination is a potent driver of tumorigenesis. It’s crucial to note that overexpressing either gene alone failed to produce tumors, which tells us that targeting the synergy between these pathways, not just a single protein, could be the key to developing highly effective future therapies.
Spatial transcriptomics revealed a specific “MYCN niche” of 167 genes. Could you elaborate on how this technique works to map a precancerous microenvironment, and why is understanding the location of gene activity, not just its presence, so crucial for predicting tumor development?
Think of traditional gene analysis as creating a smoothie from a fruit basket—you know what fruits are in it, but you have no idea how they were arranged. Spatial transcriptomics, on the other hand, is like taking a detailed photograph of that basket. It allows us to see not only which genes are active but precisely where in the tissue they are being expressed. In our study, this meant we could watch the liver tissue over time and pinpoint the exact neighborhoods where MYCN levels were rising before any tumor was visible. We discovered a distinct cluster of 167 genes that consistently switched on or off in these specific spots, forming what we call the “MYCN niche.” Understanding this location is everything. It shows us that cancer doesn’t just arise from a single bad cell, but from a permissive microenvironment, a “fertile soil” that enables a tumor to take root and grow.
Your team developed a machine-learning model that identifies the MYCN niche with 93% accuracy. What key characteristics of the gene-expression pattern does the algorithm look for, and what were the main challenges in translating this model from mouse data to reliably predict risk in human tissues?
The beauty of the machine-learning model is its ability to see complex patterns that would be invisible to the human eye. It isn’t just looking for the presence of those 167 genes; it’s analyzing the subtle interplay, the precise signature of their collective expression level and spatial arrangement that defines the MYCN niche. With this sophisticated pattern recognition, we achieved an impressive 93% accuracy in identifying these high-risk zones. The primary challenge in moving from mouse to human data is always complexity and variability. Mouse models are controlled environments, whereas human tissues are influenced by a lifetime of unique genetics, diet, and environmental exposures. We had to ensure our algorithm was robust enough to cut through that noise and find the core biological signal of the MYCN niche, which, thankfully, proved to be a conserved and powerful indicator of risk across species.
The MYCN niche score proved more predictive when calculated from non-tumor liver tissue. Why does this precancerous microenvironment offer a stronger predictive signal than the tumor itself, and how could this finding reshape how clinicians approach biopsies and risk assessment for patients?
This was perhaps the most significant finding of our work. A tumor is chaotic—a whirlwind of mutations and genetic instability. Trying to predict future behavior from that chaos is incredibly difficult. The surrounding, seemingly healthy tissue, however, tells a different story. If that tissue contains the MYCN niche, it signals that the fundamental conditions for cancer growth are present. It’s the “why” behind the cancer, not just the “what.” This completely flips the script on conventional practice. Instead of focusing a biopsy solely on the known tumor, clinicians could begin sampling the adjacent non-tumor tissue to calculate this risk score. It could allow us to identify patients with a high risk of recurrence or developing new tumors, enabling proactive surveillance or preventative treatments for the first time.
What is your forecast for integrating spatial transcriptomics and machine learning into routine cancer diagnostics and preventative care over the next decade?
I am incredibly optimistic. Over the next decade, I foresee these technologies moving from the research lab into the core of clinical pathology. We will shift from a reactive to a predictive and preventative model of cancer care. Imagine a patient with liver disease undergoing a routine biopsy; instead of just looking for existing cancer, the sample will be analyzed using spatial transcriptomics. An AI model will then generate a personalized risk score, like our MYCN niche score, telling the clinician the likelihood of that patient developing cancer in the future. This will enable tailored surveillance schedules and early interventions, fundamentally changing patient outcomes. It’s about detecting the storm on the horizon, not just waiting for the rain to fall. This fusion of spatial biology and artificial intelligence represents the future of truly personalized medicine.
