Antibody-drug conjugates (ADCs) have emerged as a busy niche in oncology. The class now spans 15 FDA-approved products and is projected to reach $57 billion by 2032. ADCs aim to widen chemotherapy’s therapeutic window by using a tumor-targeting antibody to concentrate a potent cytotoxic payload in cancer cells while limiting exposure to healthy tissue.
Ken Bloom, MD, head of pathology at Nucleai, explained the concept: “I’m going to put a GPS signal on my chemotherapy target, and it’s going to zone in just to the tumor cells and release its payload.”
However, Bloom noted that “in practice it’s far trickier than that.” A tumor may appear target-positive on a pathology slide, yet the drug’s cell-surface target could be altered or inaccessible. Extracellular proteases in the tumor microenvironment can cleave some ADC linkers, releasing their payload prematurely. Cancer cells can also expel certain payloads using drug-efflux pumps.
Closing the gap between pathology and drug delivery
Nucleai is collaborating with Gilead Sciences to address the gap between pathology findings and ADC performance. On August 11, they disclosed that Nucleai analyzed H&E and IHC whole-slide images from multiple Gilead clinical studies across various oncology indications, linking tissue features with clinical outcomes.
The partnership began with preclinical multiplex immunofluorescence work and has expanded into retrospective analysis of clinical trial data. Meghna Das Thakur, senior director of oncology biomarkers at Gilead, highlighted Nucleai’s “specialized expertise in AI-driven spatial biology and tissue analytics,” particularly in analyzing complex tumor microenvironments at scale.
Combining Nucleai’s capabilities with Gilead’s scientific and clinical expertise aims to generate insights more efficiently, Thakur said.
Candidate biomarkers emerge from retrospective analysis
Bloom anticipates a presentation and peer-reviewed publication later this year or early next, though the timeline remains flexible. The retrospective analysis has identified several candidate biomarkers that outperform standard immunohistochemistry.
“These biomarkers offer further insight beyond just quantifying expression levels,” Bloom said. They are undergoing deeper analysis.
Read Also: Scientists uncover enzyme activity with advanced cell imaging
Conventional IHC expression scores can miss therapeutically relevant information because staining confirms a target’s presence without showing whether an ADC can engage the relevant extracellular epitope.
The inside-outside epitope issue is well-documented, with HER2 as a key example. The Ventana 4B5 IHC antibody detects the intracellular domain of HER2, identifying both full-length and truncated forms. In contrast, trastuzumab binds the extracellular domain.
A 2011 Cancer Research review explains that “p95HER2 fragments arise through at least 2 different mechanisms: proteolytic shedding of the extracellular domain and translation of HER2 mRNA from internal initiation codons.” These fragments lack the extracellular region targeted by trastuzumab.
“Pathologists focus on the intracellular side because tissue processing preserves it better,” Bloom said. “But it’s the extracellular region that matters for drug binding.”
He added, “Most pathology antibodies target the intracellular membrane, yet the extracellular region is critical for drug engagement.”
Computational pathology adds new dimensions to tumor analysis
The discrepancy between IHC detection and drug binding has clinical implications. A 2015 study of trastuzumab-treated breast cancer patients found discordant results in 15% of cases when measuring HER2’s intracellular and extracellular domains. Higher extracellular-domain expression correlated with longer disease-free survival.
A 2022 study co-authored by Bloom highlighted another issue: conventional HER2 assays, designed to distinguish amplified from unamplified tumors, offer limited resolution within the lower range of HER2 expression relevant for drugs like trastuzumab deruxtecan.
The challenge, Bloom said, is developing pathology assays that capture therapeutically relevant features while remaining stable and reproducible across laboratories.
Throughout his 40-year career, Bloom noted that new pathology tools primarily improved consistency in existing tasks. Computational pathology, however, can extract complex relationships from slides, such as cell density, spatial distribution, and cellular neighborhoods.
Read Also: Novo Nordisk’s ziltivekimab fails Phase 3 trial
“We now analyze the neighborhood around a cell,” Bloom said. “Quantifying these relationships adds depth to conventional pathology and may explain why tumors with the same biomarker behave differently.”
These advancements extend beyond drug development. Bloom cited studies showing that AI assistance improves pathologist accuracy and consistency, partly by reducing fatigue and training disparities.
“AI overreaders enhance the process,” he said. “They catch mistakes and improve uniformity, a long-standing issue in pathology.”
Computational pathology relies on human expertise. Bloom views these tools as a way to achieve clinical-grade accuracy while keeping physicians involved. “A fool with a tool is still a fool,” he said, a line he’s used since the early days of digital pathology.
He also highlighted communication benefits. Computational tools can visualize tumor and non-tumor tissue in distinct colors, offering non-pathologists an intuitive spatial understanding that narrative reports cannot provide.
“AI alone isn’t enough, but a well-trained physician with AI will outperform one without,” Bloom said.
Establishing computational pathology in pharma
The adoption of this model across pharma remains uncertain. Bloom sees several organizations leading the way, but a clear, repeatable path has yet to emerge.
“A defined pathway isn’t established yet,” he said. “However, several leaders are forging the first path. Once one succeeds, a wave of adoption will follow.”