Molecular profiling revolution helps overcome unclear pediatric cancer diagnoses

In 2005, astronomer Mike Brown, enabled by new technologies, triggered a scientific crisis by discovering the dwarf planet Eris beyond Pluto’s orbit. Eris was more massive than Pluto, but it did not fit the definition of a planet, causing a crisis in classification. Either the definition of “planet” would expand to become meaningless to astronomers, or they would have to accept that Pluto no longer qualified to keep the category useful, leading to its status as a dwarf planet.

Pediatric cancer diagnosis has undergone a similar reckoning, empowered by new technologies and computational approaches, subcategorizing tumors into more useful groupings. Traditionally, physicians relied largely on what tumor cells looked like under the microscope to classify disease and guide treatment. However, some tumors that appeared to have nearly identical form and structure (morphology) behaved very differently. Others did not fit in any established categories. At St. Jude, researchers have helped drive a molecular profiling revolution that is making those unclear diagnoses far less common, allowing physicians to group tumors more accurately and match children to more appropriate therapies.

“Particularly for pediatric tumors, which are rare, it was historically challenging to fit them into categories,” said Brent Orr, MD, PhD, Department of Pathology. “A lot of my research has been focused on taking things that have been defined by morphology and then looking at them in a molecular way, such as methylation profiling, and applying machine learning to that data to categorize them.” 

Methylation profiling examines chemical marks on DNA that change gene expression. By looking at methylation patterns, scientists can understand which genes are turned off and on, seeing what is similar and different between tumors at the genomic level. This analysis often far exceeds the accuracy and precision of looking under the microscope, making it easier to diagnose even rare forms of disease accurately.

Dissolving a confusing diagnosis

One of the clearest demonstrations of that shift came from tumors once broadly classified as primitive neuroectodermal tumor, or PNET. The name described how the tumors looked, but it did not explain why patients with the same diagnosis often had very different outcomes.

“One of the first groups we looked at with this approach was PNETs,” Orr said. “When we look at this confusing diagnosis by methylation profiling, most could be better classified as other things.”

“We started with PNET as a group because they behaved aggressively and we could not determine how best to treat them,” added Jason Chiang, MD, PhD, Department of Pathology. “We saw that PNETs don’t belong together; they are more appropriately grouped with different tumor types, explaining their differing responses to therapy.”

Published in Cell, the implications of these results for patients were substantial. Historically, tumors diagnosed as central nervous system (CNS)-PNETs were treated aggressively with craniospinal irradiation, a treatment with long-lasting side effects, due to their high risk of spread. The study revealed that many of these tumors were actually high-grade gliomas, which typically only require focal radiation. By correctly reclassifying these tumors, clinicians could spare patients from the unnecessary craniospinal irradiation. Conversely, accurately identifying the tumors that require intensive regimens ensures they still receive the most appropriate therapy. By effectively dismantling the CNS-PNET category, researchers established a more precise diagnostic framework, paving the way for properly tailored therapeutic strategies.

Subtyping existing tumors to remove confounding categorizations

More often than eliminating a category, molecular profiling has revealed that many tumor types can be broken down into more specific and useful groups. Paul Northcott, PhD, Center of Excellence in Neuro-Oncology Sciences director and Department of Developmental Neurobiology member, launched his scientific career by subcategorizing medulloblastoma. Despite being one of the most common pediatric brain tumors, medulloblastoma had historically been considered a single category with inconsistent treatment responses.

“We used the best generation platform available at the time to profile a large set of medulloblastoma tumors,” Northcott remembered. “That ultimately led to the four groups, WNT, Sonic Hedgehog, Group 3 and Group 4, which we still use today.” St. Jude was the first institution to launch a clinical trial based on the molecular groups of medulloblastoma, SJMB12, that used these definitions to help guide therapy.

Chiang’s group applied the same logic to high-grade gliomas, brain tumors that had a history of highly varied treatment responses. He and his colleagues published the results in Neuro-Oncology. “We realized that although these tumors look similar under the microscope, they could actually be entirely different diseases, such as infant-type hemispheric glioma,” he said. “That matters because we have good inhibitors for the recurrently mutated genes that drive the infant-type tumors, but we would be using incorrect and more intensive therapies than needed if we labeled them as high-grade gliomas.”

Discriminating between diagnoses with circulating tumor DNA

Even for cancers that are easier to diagnose, such as Wilms tumor, the most common pediatric kidney cancer, uncertainty can remain. “Most cases of unilateral (single kidney) Wilms tumor have really good outcomes,” said Andrew Murphy, MD, Department of Surgery. “But there are subsets of patients with adverse biology that continue to do poorly. They have biomarkers that predict needing more intense therapy, such as changes in chromosomes one and sixteen, but we do not have a good method to diagnose them without an invasive biopsy.”

The issue has been finding a way to discriminate between the two groups before taking a biopsy. Biopsies can disrupt the tumor in a way that may lead to upstaging, automatically causing the cancer to be treated with a higher intensity treatment. To address this gap, Murphy is working with other St. Jude researchers to determine if sequencing circulating tumor DNA in a person’s blood, ctDNA, can be used to diagnose Wilms tumor. 

“We are going to look at biospecimens from our St. Jude Wilm’s tumor 21 clinical trial,” Murphy said. “We’ll look at the feasibility of taking a blood sample and using ctDNA instead of performing a surgical biopsy.”

If successful, this approach could help physicians better distinguish which tumors need intensified therapy and which do not, reducing unnecessary long-term toxicities while preserving aggressive treatment for the children most likely to benefit.

Northcott’s group has taken a similar approach for brain tumors, where biopsies can be especially difficult and distinguishing relapsed disease from a second cancer is critical for treatment planning. “We created a computational framework that uses ctDNA from cerebrospinal fluid samples to classify pediatric brain tumors,” he said.

Published in Nature Cancer, Northcott’s lab created Methylation-based Predictive Algorithm for CNS Tumors (M-PACT), which uses artificial intelligence to classify tumors. It successfully distinguished between relapsed and second cancers, showing the potential promise of ctDNA-based diagnosis to overcome current diagnostic challenges in multiple settings.

Molecular profiling reveals deviations from traditional classifications

Molecular profiling has also shown that pediatric cancer classification is more complex than once thought. Some tumors arising in different parts of the body can share important molecular features, while others sharing a hallmark mutation may still belong to different biological groups.

Pineoblastoma, a rare tumor of the pineal gland, illustrates the first case. Northcott’s group published their work profiling pineoblastoma in Cancer Cell, revealing a dependency involving light-sensing genes. That signature was shared across anatomically distinct central nervous system tumors, including tumors of the pineal gland, retina and cerebellum, pointing to a possible dependency that spans traditional tumor boundaries.

Chiang’s group found the reverse pattern with the gene FOXR2. At one point, any central nervous system (CNS) tumor with a mutation in FOXR2 was categorized as a CNS neuroblastoma. However, when the scientists analyzed DNA and RNA data from multiple brain and CNS cancers in the St. Jude Cloud, they found that multiple other tumor types also had altered FOXR2 — and that many did not respond to the treatment for CNS neuroblastoma, explaining previously confounding results. This work was published in Neuro-Oncology.

Together, these studies show how molecular profiling is refining diagnosis in both directions: splitting apart tumors that were grouped too broadly and revealing connections where traditional classifications drew lines too narrowly.

Learning from unclear diagnoses

Despite these advances, some cancers defy categorization. “We still run into outlier tumors that don’t classify with any existing groupings,” Chiang said. “In these cases, we do four things: First, we do our best to describe it; second, we acknowledge what we don’t know; third, we have a consensus meeting to determine the best therapeutic strategy; and fourth, we document the treatment response.” 

Like the discovery of Eris forcing astronomers to reconsider Pluto, molecular profiling has compelled cancer researchers to rethink long-standing definitions. At St. Jude, those efforts are steadily replacing ambiguous diagnoses with more accurate and useful ones, bringing physicians closer to matching every child’s tumor with the treatment most likely to help.

“In the past, we often had unclear or unhelpful diagnoses because their subtype was ‘hidden’ from our tools at the time,” Orr said. “Now, through methylation profiling, sequencing and computational approaches, we are one step closer to getting the right diagnosis and treatment for every child with cancer we see.”