The opening plenary session of the American Association for Cancer Research (AACR) meeting took place in an enormous room, packed to the gills with attendees, at the San Diego Convention Center on Sunday morning.

First, the meeting’s program committee co-chairs, Christina Curtis, PhD, professor of medicine, genetics, and biomedical data science at Stanford University and Keith Flaherty, MD, physician investigator at the Cancer Center, Mass General Research Institute, took the stage to share their joint vision that led the direction of the meeting, which Flaherty said, sets the agenda for the research community.

“As we developed the program,” noted Curtis, “we tried to account for a technological tipping point in a few ways.” One way was to institute technology focused sessions. For example, the first plenary session’s focus was on artificial intelligence. Another was to weave together different aspects of research by pairing basic science, technology, and clinical results into the same sessions.

They made a concerted effort, they said, to foster a sense of exploration by creating sessions with talks that inform one another, capture interest, and expand research horizons.

That sentiment was felt in the opening plenary with a quartet of researchers offering a wide view of cancer research. First was Aviv Regev, PhD, executive vice president at Genentech Research and Early Development in a talk titled, “From cell atlases to medicines.”

Regev noted that it is important to focus on malignant cells and microenvironment cells as well as the circuits that exist within each of these cells. This complexity means that the cancer community has an enormous task and a difficult shared mission. One way to tackle this is to hone on the fundamental units—the cells.

A pioneer in single-cell genomics, Regev is a leader in the construction of the human cell atlas, which includes millions of tumor cells. Another rich source of information are H&E stained slides which exist for many cancer patients. Hoping to combine those two tools, she asked, “What if genomics could be performed on H&E stained slides?” To tackle that question, Regev and her team developed and trained the algorithm SCHAF, which learns to predict a tissue’s single-cell omics dataset from a histology image.

Large data can, with the help of machine learning algorithms, help identify biological events. When the quantity grows large enough, it becomes quality, she asserted.

Regev continued to explain how this work, and other technologies including perturb-seq, perturb-view, and a cancer perturbation atlas can translate these data into treatment options for patients by aiding in target selection or iterating into new medicines.

The next speaker in the session, Jakob Nikolas Kather, MD, professor of medicine and computer science at Technical University Dresden, Dresden, Germany, continued with the AI theme with a talk titled, “Artificial intelligence-based biomarkers in cancer histopathology.”

Kather noted that cancer patients have unstructured data (i.e., pathology slides and radiology images). His goal is to make those data “AI-able.” Kather’s interdisciplinary team of both medical and computational experts has developed machine-learning models that can use an H&E slide to predict microsatellite instability in colorectal patients. It’s not perfect, he said, but it’s “good enough to be useful.” Other biomarkers, like homologous recombination deficiency (HRD), can also be predicted from H&E slides. This same principle can be used to predict clinical outcomes and response to immunotherapy in hepatocellular carcinoma based on gene expression signatures.

Jakob Nikolas Kather
Jakob Nikolas Kather, MD at the AACR 2024 plenary session [LeMieux]
New approaches include multimodal models (a deep learning model that can incorporate multiple types of data, for example, genomics and histopathology) and the emerging foundation models which take a two-step approach. First, a deep learning model is trained on images—for example, a million slides anonymized with no other information (which, he emphasized, would have been useless previously). The patterns create a foundation model that can be fine-tuned. With foundation models, every researcher does not need to make their own model. They can use one that is already available and fine tune it to their specific tasks. The advantage is better models that need less data, can be generalized, and can work with less amount of tissue (for example, a biopsy).

Kather cautioned that they “still have to take many steps to take this to the clinic.” For example, pathology labs have to be digitalized and regulators have to be in the conversation. Kather co-authored a recent paper in Nature Medicine titled, “Large language models should be used as scientific reasoning engines, not knowledge databases.” But, he stressed in his talk that, “the technology is here now.”

Kather ended his talk by adding (while chuckling) that, “The field is moving so fast, this will all be outdated next year.”

The session moved from AI to medicinal chemistry, with a talk by Benjamin F. Cravatt, professor in the department of chemistry at Scripps Research about “Activity-based proteomics: Cancer target and ligand discovery on a global scale.”

Presenting his work on activity-based protein profiling (ABPP)—chemical probe discovery within a druggable space that can extend to the undruggable space. More specifically, he spoke about ABPP guided covalent ligand discovery that targets enzyme active sites. This method has resulted in enzyme inhibitors that are now in clinical development. For example, they identified covalent inhibitors to pharmacologically control N-Ras.

To explore this more broadly, Cravatt sought to do a proteome-wide covalent ligand discovery by cysteine-directed ABPP. There are numerous cysteines in the human proteome that can be covalently liganded.

One way to translate this work to drug development is by industrializing the platform, which was done by Vividion Therapeutics. But an academic lab cannot scale up to that level. To develop an alternative approach, Cravatt turned to the design of their chemical libraries to develop next-generation electrophilic compound libraries for expanded druggable space. This was done in collaboration with Stuart Schreiber, PhD, now at Arena BioWorks and Bruno Melillo, PhD, institute investigator, department of chemistry at Scripps Research.

Cravatt finished his talk by telling the story of the identification of a steroselective ligand targeting FOXA1—a pioneer transcription factor (one that can bind the site in open DNA and in closed heterochromatic DNA) required for breast and prostate cancers. FOXA1 is both a driver and also mutated in cancer. They found stereoselective, site specific, covalent ligands for FOXA1 that bind in a DNA-dependent manner, enhance FOXA1 binding to suboptimal DNA sequences, and remodel FOXA1 binding to chromatin, leading to global changes in accessibility.

Last on the lineup was Nobelist Carolyn Bertozzi, PhD, the Baker Family Director of Sarafan ChEM-H, and professor of chemistry, at Stanford University. Bertozzi, a pioneer in bioorthogonal reactions, presented a talk entitled, “Next-generation cancer therapies enabled by biorthogonal chemistry.”

Bertozzi—whose talk was delivered so well, it made organic chemistry easy to understand—jumped into glycan biology and how glycoproteins, glycolipids, and the newly described glycoRNA are major determinants of cell surface interactions.

All of these glycans create a “molecular landscape that allows cells to interact that are important in the progression of cancer,” she said.

Bertozzi first became interested in how the patterns of glycans change when cells undergo physiological changes (during disease for example) in the 1980s. Tumors overproduce sialoglycans, or glycans with sialic caps. When this happens, she said, “a well manicured garden turns into a tropical forest.”

When immune cells and cancer cells interact, the immune cells need to recognize the cancer cells to destroy them. But immune cells also have inhibitory receptors. Cancers can develop a phenotype where they can block the inhibitory signal, allowing them to survive. One immunomodulatory receptor on immune cells are siglecs (siaclic scid Ig-like lectins).

Bertozzi hypothesized that tumors are rich in sialoglycans because they can engage siglecs on an immune cell. This would open up a major axis of immune modulations that could explain why patients do not respond to checkpoint inhibitors

In an effort to disrupt siglec and sialogycan binding, Bertozzi took a page from the playbook of the T-cell checkpoint inhibitors. In order to drug this biology, they focused on a method to strip the sialic acids off of the cell—like a lawnmower that would park on the surface of the targeted cancer cell—so that they cannot engage siglec cells. They used biorthogonal chemistry to create Trastuzumab-sialidase conjugate (T-sia) as a prototype glycan editor and showed that, when tested in a Her2+ murine breast cancer model, the tumors turn from cold to hot.

This work, which is now being expanded at Palleon Pharmaceuticals in Waltham, MA, is expected to enter the clinic in 2025.

The last part of Bertozzi’s story was an investigation into how cancer cells overproduce sialogycans and how that is connected to other aspects of cancer biology. Collaborating with Dean Felsher, MD, PhD, professor of medicine and pathology at Stanford University, they are uncovering that MYC regulates genes that regulate siglec biosynthesis.

Together, the four speakers told inspiring stories while presenting novel technology. They set the tone for the rest of the meeting and lived up to the meeting’s description of “inspiring science, fueling progress, and revolutionizing care.”

Previous articleAACR News: Pancreatic Cancer Outcomes May Improve with Presurgical Nivolumab/Chemo Combo
Next articleAdvances in Induced Pluripotent Stem Cell Research