Breast cancers can be classified into subgroups that hint at the aggressiveness of the cancer and the likelihood that the patient will experience a recurrence years after their initial diagnosis.

Now researchers at Stanford Medicine have shown that these subgroups can be bundled into three main groups based on structural variations in their DNA – including repeats, or amplifications, of cancer-associated genes called oncogenes on chromosomes and the presence of small DNA circles untethered to the rest of the genome. These variations are established early during cancer development and are maintained as the disease advances and metastasizes.

Understanding the role these variations play in tumor evolution, and how to stop them, will aid physicians in making decisions and point to new targeted therapeutic interventions, the researchers believe. This robust classification system could also differentiate breast cancer patients who are most likely to benefit from aggressive early intervention from those who may be able to safely defer such aspects of treatment.

“My lab has had a long-standing interest in understanding how aggressive breast tumors arise, why they are resistant to therapy, and why they are prone to recur in distant organs,” said Christina Curtis, PhD, the RZ Cao Professor and a professor of oncology, of genetics, and of biomedical data science. “This research shows that breast tumors develop key structural variants that set the tumor on its course very early in its development. In short, some are born to be bad. It emphasizes the importance of robust biomarkers and of intervening early in the course of the disease.”

Curtis is the senior author of the research, which was published Jan. 8 in Nature. Formal postdoctoral scholar Kathleen Houlahan, PhD, postdoctoral scholar Lise Mangiante, PhD, former research assistant Cristina Sotomayor-Vivas, and graduate student Alvina Adimoelja are the lead authors.

Broad classification

Curtis, the director of artificial intelligence and cancer genomics at the Stanford Cancer Institute, has been researching the evolution of breast cancer tumors for over a decade. For many years, breast cancers have been classified in broad strokes according to the types of protein receptors they make. Tumors with elevated levels of receptors that bind to estrogen or progesterone are called hormone-receptor positive tumors, others with elevated levels of a receptor called HER-2 are called HER-2 positive tumors, and a third type called triple-negative does not express either of the hormone receptors or HER-2.

The majority of hormone positive breast cancers are estrogen positive; therapies for these cancers lower estrogen production, block estrogen binding to its receptor, or degrade estrogen receptors on the surface of the cancer cells.

Of the three, triple-negative breast cancers – constituting about 10% of newly diagnosed cases – are often considered the most difficult to successfully treat and tend to recur early. Hormone-receptor positive cancers, which are the most common, are often successfully treated with a combination of hormone therapy, chemotherapy, surgery, and radiation. HER-2 positive breast cancers – about 15% to 20% of all cases – are aggressive but can be successfully treated with drugs that block HER-2 activity.

In 2012, Curtis and her colleagues used machine-learning techniques to compare the DNA sequences from a patient’s healthy cells with the DNA and RNA sequences from their breast tumors. This gave a molecular snapshot of any genetic alterations the tumor may have developed as well as the effect of those alterations on when and how the cells’ genes were expressed. The study identified 11 clinically significant subgroups – far more than had been previously identified on the basis of receptor expression. These subgroups had varied prognosis, but it wasn’t clear at the time how to use this information to guide patient care.

A subsequent study of 75,000 people with estrogen-receptor positive breast cancer showed that even after five years of hormone therapy, and even in the clinical group with the lowest risk, breast cancer recurrences continued. Curtis and her colleagues wanted to know why and whether the subgroups they had defined could better delineate this risk.

In 2019, they showed that overlaying the receptor status of breast tumors with their subgroup classification could predict which hormone-receptor positive tumors were likely to recur long after initial diagnosis and treatment. Specifically, four of the eight estrogen-receptor positive subgroups were much more likely than others to return even 10 or 20 years after diagnosis. Combining these four high-risk groups, they found that one-quarter of women with hormone-receptor positive, HER-2 negative breast tumors faced a nearly 50% chance of their breast tumors recurring even decades after their initial diagnosis. This elevated recurrence risk surpassed even that of people with triple-negative breast cancer and mirrored that of HER2-positive breast cancers before the approval of trastuzumab, also known as Herceptin, which transformed patient outcomes.

The approach could also identify patients with triple-negative tumors who were unlikely to experience a recurrence more than five years after diagnosis and treatment and those who were more likely to experience a recurrence. This type of patient stratification is helpful in pinpointing who might need an aggressive treatment early in their disease course or more intensive monitoring in subsequent years as well as others who may be able to safely bypass harsher treatment approaches.

But it still wasn’t entirely clear what was driving the differences among the subgroups.

“We wanted to take a step back,” Curtis said. “Each of the four higher risk subgroups has copy number events – duplications or amplifications of specific oncogenes involving different regions of the genome. These patterns of genomic copy number change were similar to that seen in HER2-positive disease. If we look at these tumors in an unbiased way and deconstruct these different types of mutations, what could we learn about their processes that give rise to these characteristic events? Would we discover something different?”

A look at the genomic architecture

When Curtis and her team assessed the genomic architecture – the mutations and structural variations in a cancer cell’s DNA – of nearly 2,000 breast cancers of varying stages, from ductal carcinoma in situ (stage 0) to advanced metastatic disease (stage 4), they found they were able to categorize the tumors into three groups based on oddities in their genomes.

They found that the high-risk hormone-receptor positive subgroups strongly overlapped with the HER-2 positive subgroup: Each had complex but localized amplifications of cancer-associated genes as well as small DNA circles called extrachromosomal DNA, or ecDNA, chock-full of oncogenes. Other recent studies have implicated ecDNAs, which often ignore normal cellular regulatory mechanisms, as key drivers of cancer growth and evolution.

“Here we have two different molecular subtypes, which we treat differently in the clinic but that strongly overlap in their patterns of chromosomal instability,” Curtis said.

Triple-negative tumors had genomes that were globally unstable, accumulating alterations across the genome; a subset of these also showed signs of being deficient in their ability to repair DNA damage. “The whole genome shows scars,” Curtis said. “It’s not limited to particular oncogenes.”

In contrast, garden-variety hormone-receptor positive, HER-2 negative breast cancers with typical risks of recurrence have relatively stable genomes.

The structural variations that define each group were present in the earliest stages of the disease and were maintained as the tumors grew and spread throughout the body. They also correlated with whether and how immune cells infiltrate and respond to the tumor.

Understanding the foundational importance of structural variations and genomic architecture on cancer development can hint at new therapeutic options. For example, the researchers speculate that existing drugs designed to target impaired DNA repair pathways in patients with BRCA1 and BRCA2 mutations (leading to inherited forms of breast cancer) might also benefit the approximately 13% of people with DNA repair-deficient, estrogen-receptor positive breast cancers. Other tumors that rely on focal amplifications and ecDNA might be vulnerable to compounds that target their respective drivers or ensuing replication stress. Still other approaches may directly target the mutational processes that propagate these events.

“These early, sometimes catastrophic mutational events happen decades prior to the diagnosis of the tumor, emphasizing opportunities for earlier interventions,” Curtis said. “Despite the complexity of their genomes, there are constraints and only so many evolutionary paths for a tumor to follow. We now have an understanding of how and when these complex alterations arise and their accompanying vulnerabilities.”

Christina Curtis is a member of Bio-X and of the Stanford Cancer Institute and is a Chan Zuckerberg Biohub investigator.