#485 — The New Science of Cancer

In this extensive conversation, Siddhartha Mukherjee explores how cancer science has evolved over the past 15 years. He establishes that while all cancers share physiological commonalities—uncontrolled cell division and hijacking of normal cellular pathways—each individual cancer specimen is genetically distinct, requiring personalized treatment approaches.

On prevention, Mukherjee reveals a surprising fact: since the 1960s, no significant preventable chemical carcinogen has been identified that could substantially reduce cancer mortality, despite billions in research funding. However, a new class of carcinogens called "inflammogens" has recently been discovered. These substances don't cause mutations directly but rather change the cellular environment ("soil") to enable dormant cancer cells to grow by inducing chronic inflammation. Particulate air pollution and asbestos may work through this mechanism. Mukherjee discusses successful chemoprevention for estrogen receptor-positive breast cancer using drugs like tamoxifen, and emphasizes the critical importance of HPV vaccines, which have demonstrated near-zero cervical cancer risk in randomized trials.

Regarding detection, Mukherjee addresses the controversy around liquid biopsies and cell-free DNA tests using Bayesian reasoning. He explains that the fundamental problem with screening healthy populations is low base rates of cancer—even highly accurate tests produce predominantly false positives when applied to low-risk populations. He illustrates this with the "needle in a haystack" analogy: a 90% sensitive and specific detector will still find mostly hay. His solution is not universally screening everyone but rather identifying higher-risk populations (those with genetic mutations, family history, or prior cancer) where the prior probability of cancer is higher. He warns against false positives driving unnecessary biopsies and invasive procedures.

On whole-body MRI screening, Mukherjee acknowledges higher type-1 error rates but suggests temporal scanning (baseline plus follow-up) could provide valuable information about newly emerged lesions versus long-standing incidental findings, though this approach remains understudied.

For minimal residual disease monitoring, Mukherjee identifies this as the optimal use case for sensitive tests like cell-free DNA. Detecting early relapse signals in previously treated cancer patients shifts the Bayesian prior probability favorably, enabling early intervention before resistance develops.

On treatment breakthroughs, Mukherjee documents significant progress: cancer mortality has decreased from 200 to 140 deaths per 100,000 over 20 years. Specific advances include immunotherapy for lung cancer and bladder cancer, extended survival in breast cancer and myeloma, and CAR-T cell therapies achieving 50-60% cure rates in relapsed childhood acute lymphoid leukemia. However, CAR-T cells have struggled with solid tumors due to the hostile microenvironment surrounding these tumors.

Mukherjee explains that drug development increasingly uses incremental advances—the "crampon" metaphor—where each therapeutic breakthrough reveals resistance mechanisms that inform the next treatment generation, gradually transforming fatal diseases into chronic and eventually curable conditions.

Regarding AI, Mukherjee distinguishes between multiple task-specific AI applications rather than general artificial intelligence. He emphasizes that effective drug discovery AI requires teaching the rules of medicinal chemistry rather than pure pattern recognition. His company, Manus AI (co-founded with Gujwal Singh and Reid Hoffman), focuses on this rules-based approach. AI shows promise in target discovery (identifying proteins to inhibit), molecular discovery (designing molecules), radiological diagnosis (as a companion diagnostic), and adaptive clinical trials. Prevention represents a largely untapped area for AI given the high-dimensional, multi-vector problem of predicting cancer risk across genetics, exposome, microbiome, and behavior.

On generic drug availability, Mukherjee notes that major immunotherapy drugs are coming off patent soon, which should dramatically reduce costs. He acknowledges past quality control problems documented in the Vanity Fair article by Katherine Eban regarding generic drug manufacturing, but argues these are solvable through continuous independent audits and quality checks rather than only FDA inspections.

Mukherjee expresses optimism about cancer's future, noting that many previously incurable cancers have become chronic diseases and some curable. He cites recent RAS inhibitors for pancreatic cancer that doubled survival from 6 to 13 months—a crucial first foothold for future progress. However, he acknowledges that cancer as a disease of aging cannot be completely eliminated, though incidence can be reduced through lifestyle changes and obesity treatment.

On the political situation, Mukherjee expresses concern about the Trump administration's anti-science stance, including funding cuts to the CDC and USAID, increased bureaucratic scrutiny of research, and lack of scrutiny where needed. He notes that rebuilding damaged institutions takes years after swift destruction. Most troublingly, he documents that U.S. drug imports from China have increased from $5 billion in 2020 to a projected $60-70 billion in 2025, representing a dangerous offshoring of pharmaceutical innovation and manufacturing when supply chain resilience should be prioritized.