The oncology landscape is currently dominated by the rapid ascent of antibody-drug conjugates (ADCs). Often described as the "guided missiles" of cancer therapy, these complex molecules combine the precision of monoclonal antibodies with the lethal power of cytotoxic chemotherapy. With approximately 15 FDA-approved products now on the market and a segment valuation projected to hit $57 billion by 2032, the momentum behind ADCs is undeniable.
Yet, for all their promise, ADCs remain plagued by a persistent clinical mystery: why do tumors that test positive for a drug’s target—such as HER2—frequently fail to respond to treatment?
To answer this, Gilead Sciences has entered a strategic collaboration with Nucleai, a leader in AI-driven spatial biology. By analyzing clinical trial datasets through the lens of computational pathology, the two companies are working to bridge the disconnect between what a pathologist sees on a slide and what an ADC actually encounters within the complex architecture of a tumor.
The "Target-Positive" Paradox: Why Conventional Assays Fail
The core premise of an ADC is elegant: the antibody acts as a GPS, tethering to a specific protein on the surface of a cancer cell, which then internalizes the drug to release its toxic payload. Theoretically, this maximizes the therapeutic window, delivering chemotherapy directly to the tumor while sparing healthy tissue.
In practice, however, the biology is far more intricate. Dr. Ken Bloom, head of pathology at Nucleai, explains that the industry has historically relied on immunohistochemistry (IHC) to determine if a patient is a candidate for ADC therapy. "Initially, what we were told was that ADCs are basically targeted chemotherapy," Bloom notes. "But in practice, it’s far trickier than that."
The Inside-Outside Epitope Problem
A major point of failure lies in the diagnostic tools themselves. Conventional IHC staining is excellent at identifying the presence of a protein, but it often fails to distinguish between the intracellular and extracellular domains of that protein.
For an ADC to work, it must bind to the extracellular epitope—the part of the protein sticking out of the cell. However, most standard pathology antibodies used in clinical diagnostics are engineered to detect the intracellular domain, as these regions are generally more stable during tissue processing.
This creates a "hidden secret" in pathology: we are testing for the presence of a protein by looking at its internal side, even though the drug is fighting the battle on the outside. In the case of HER2, for example, truncated fragments (like p95HER2) can retain the internal domain while lacking the external target for drugs like trastuzumab. Consequently, a patient may be labeled "HER2-positive" by a standard lab test, but the drug has no binding site to dock onto.
Chronology: From Multiplex Analysis to Clinical Retrospection
The relationship between Gilead and Nucleai began several years ago, rooted in the foundational science of multiplex immunofluorescence. Initially, the collaboration focused on preclinical research, utilizing spatial biology to map how immune cells and tumor cells interact within the tumor microenvironment (TME).
As the utility of these insights became clear, the partnership evolved. On August 11, the companies officially disclosed a broader initiative: Nucleai is now performing a deep-dive retrospective analysis of whole-slide images (H&E and IHC) from Gilead’s clinical studies across multiple oncology indications.
The goal is to link specific spatial tissue features—such as cell density, neighborhood organization, and protein expression patterns—with actual patient clinical outcomes. By moving beyond simple "positive/negative" scoring, the teams are attempting to build a more granular, predictive model of drug efficacy.
Beyond Staining: The Rise of Computational Pathology
The collaboration hinges on the distinction between traditional digital pathology and true computational pathology. For decades, digital pathology served primarily to digitize slides, allowing pathologists to view images on screens rather than through glass. Computational pathology, by contrast, treats the slide as a data-rich environment.
Quantifying the "Neighborhood"
"Now we care about the neighborhood that a cell is in," says Dr. Bloom. "What’s the relationship between it and other cells in the vicinity? What’s the density of things?"
Computational tools allow researchers to:
- Analyze Spatial Distribution: Assessing whether the target protein is expressed in areas of the tumor that are accessible to the drug’s delivery mechanism.
- Identify Drug-Efflux Activity: Detecting cellular features that suggest a tumor is actively pumping out the drug, rendering the payload ineffective regardless of target density.
- Account for Heterogeneity: Mapping how biomarker expression varies across different regions of the same tumor—a factor that standard IHC scores often average out or miss entirely.
Official Perspectives: The Synergy of Human and Machine
Gilead views this collaboration as a critical expansion of its internal translational medicine and biomarker capabilities. According to Meghna Das Thakur, senior director of oncology biomarkers at Gilead, the partnership combines Nucleai’s specialized AI-driven spatial analytics with Gilead’s deep clinical and scientific expertise. The objective is to generate insights into patient stratification with greater speed and accuracy than conventional methods allow.
Dr. Bloom emphasizes that this does not replace the pathologist; rather, it empowers them. He remains a staunch advocate for "human-in-the-loop" AI, famously noting that "a fool with a tool is still a fool." The vision is to provide pathologists with an "AI overreader" that reduces the variance caused by human fatigue, training differences, and the subjective nature of manual slide scoring.
"When you bring an AI overreader into the process, everything gets better," Bloom argues. "It catches mistakes. It gets all pathologists more uniform, which was always a big problem."
Implications: A New Standard for Clinical Trials?
The findings from this partnership could have profound implications for how future oncology trials are designed and how drugs are prescribed.
1. Refined Patient Selection
If the team can identify biomarkers that are more predictive than standard IHC, clinical trials could be enriched with patients most likely to respond to a specific ADC. This not only increases the probability of success for pharmaceutical companies but, more importantly, spares patients from the toxicity of ineffective treatments.
2. Standardizing the "Pathological Truth"
A significant hurdle in current research is the lack of reproducibility across laboratories. By developing computational assays that are stable and reproducible, the industry could move toward a global standard for biomarker testing, ensuring that a patient in a community clinic receives the same diagnostic precision as one in a top-tier research center.
3. Improving Communication
The impact also extends to the clinical workflow. Currently, pathology reports are often dense, narrative-heavy documents that can be difficult for oncologists and surgeons to interpret quickly. Computational tools can transform this data into intuitive, color-coded spatial maps, providing a visual "map" of the tumor that clarifies the therapeutic strategy.
The Path Forward
While the timeline for the publication of these findings remains fluid, Dr. Bloom suggests that a peer-reviewed presentation of the candidate biomarkers could arrive as early as late 2024 or early 2025.
The industry is watching closely. As Bloom acknowledges, there is currently no well-trodden path for integrating these complex AI-driven diagnostics into standard pharmaceutical development. However, the leaders in the space—like Gilead and Nucleai—are currently in the process of forging that path.
"I think there still isn’t a pathway that’s been forged clearly yet for somebody else to follow," Bloom said. "But the good news is that there are several leaders out there attempting to forge that first path. What you’re going to see, as soon as the first one hits, is a wave that follows."
As the $57 billion ADC market continues to mature, the ability to look past the surface—to decode the micro-geography of the tumor and understand the true availability of its targets—will likely be the defining factor in determining which therapies reach patients and which are left behind in the lab. Through the integration of AI and spatial biology, the "guided missile" of modern oncology may finally be getting the precision navigation it requires.
About the Author
