Breast cancer can be classified into different molecular subtypes determined by unique genetic fingerprints. Details about a tumor’s genetic makeup help oncologists identify the best approach to the treatment of advanced breast cancer. In an exciting advancement, team of scientists from the UK has just developed a novel tool to accurately classify breast cancer subtypes. The research, published in the journal Genome Biology, could improve treatments and targeting of treatments for the disease.
Cancer arises due to genetic changes that cause normal cells to develop into tumors. As we learn more about breast cancer, we are seeing that it is not one single disease — the mutations in the genes that cause different cancers are not alike, and this is why tumors respond differently to treatment and grow at different rates.
Clinicians already divide tumors into a few different types, and targeted treatments are available for some types of the disease. For instance, women with tumors that test positive for a cancer-promoting protein called HER2 often respond well to the drug Herceptin, which isn’t effective against other types of tumors.
But in a frustratingly high number of cases, scientists can’t explain why one woman will respond to a given treatment and another woman won’t — even though they both might have tumors that are estrogen-receptor-positive, for example.
Because patient responses can’t be precisely predicted, doctors tend to err on the side of caution, often administering treatments in cases in which they may provide little benefit (despite significant side-effects). Having a more detailed system of tumor categories can not only help avoid that problem, but also tailor treatment to individual patients and predict women’s survival and prognosis more accurately.
Ten unique subtypes of breast cancer
With that in mind, researchers at Cancer Research UK and the University of Cambridge have been working on a new tool called the IntClust system, which uses genomic technology to create a classification system with enough detail to accurately pinpoint which type of breast cancer a patient has, and therefore what treatment would be most appropriate.
The IntClust system was first introduced in 2012, when scientists used it to identify ten unique subtypes of breast cancer, each associated with distinct clinical outcomes and providing new insights into the underlying biology and potential molecular drivers.
Before the initial research was published the team conducted two large studies — a primary analysis on 1,000 tumor samples and then a validation study on an additional 1,000 samples — to confirm the results. The analysis turned up remarkably precise results, finding that IntClust consistently identified the same 10 subgroups, even for different groups of tumor samples. Moreover, the 10 subtypes were found to be strongly predictive of response to treatment and prognosis.
But as with many of these vast genetic explorations, the study revealed as much unexplored terrain as it mapped – exposing the complexity faced when diagnosing and treating breast cancer.
Translational research
Translating vast quantities of genetic data into something that can be routinely used by doctors is a huge challenge. In the lab, research teams take years to piece together the details of what makes tumors tick, but for doctors – and most importantly patients – this information is needed in a matter of days rather than years.
In their 2012 study, the team employed a whole range of analytical techniques to identify the 10 subtypes. In their new paper, they’ve checked to see if the subtypes could still be spotted accurately by using just one of these techniques – measuring the activity levels of genes within the samples (known as expression levels). This is important – doctors need a test with enough detail to accurately spot which ‘type’ of breast cancer a patient has, but one that’s simple – and cheap enough – to be reproduced around the world. In other words, they need a map showing the clear boundaries between countries, cities and towns, but not necessarily the color of each individual front door in that area
To test the system, the scientists looked picked a list of 612 genes from the original study and used the initial 997 tumor samples to ‘train’ a computer program to spot the 10 subtypes based on how active the 612 genes were. Next, they used the IntClust system to analyze another collection of around 1000 tumors from the original study. Crucially, the gene activity data from the second set of samples was accurately grouped into the 10 distinct subtypes.
But what about breast cancer data beyond the samples used to develop IntClust?
“We wanted to really test the accuracy of the system. So we tried it out on as many collections of breast cancer samples – or ‘datasets’ as possible,” says Dr Raza Ali, lead scientist on the new study. “Only by challenging our system in this way can we confirm the accuracy of the 10 IntClust subtypes.”
Findings confirmed in re-validation study
On a study-by-study basis, the team turned to the gigabytes of data available from studies around the world – encompassing over 7,500 breast tumors from more than 40 studies – and set about grouping these samples. The same 10 subtypes emerged once again from each study, confirming their 2012 findings – the IntClust system is a ‘real’ phenomenon. But they didn’t stop there. Next, the team looked at how well their IntClust system performed against the two main systems in use today. The first – called PAM50 – splits breast tumor samples into five groups, and the second – known as SCMGENE – identifies four groups.
Crucially, the researchers found that IntClust was at least as good at predicting patients’ prognosis and response to treatment as the existing systems. However, the new system also identified a previously unnoticed subgroup of tumors. The new subgroup was present in just 3.1 percent of women, all of whom had very poor survival rates, leading the researchers to conclude that the subtype may be resistant to treatment. Determining the genomic signatures for this group could help flag these high risk cancers early, and having the genomic data for these could aid in the investigation of new avenues for treatments for this type of cancer, the researchers say.
“Our findings highlight the potential of this approach in the era of targeted therapies, and lay the foundation for the generation of a clinical test to assign tumors to IntClust subtypes,” says Dr. Raza Ali, lead author from Cancer Research UK Cambridge Institute. “By looking at the genetic data we can gather important information about what’s driving these deadly tumors, which could be used to develop new targeted treatments in the future.”
While IntClust still requires too much training to be used by most clinicians, the detail and accuracy of the system could be of great use to breast cancer researchers, who will be able to investigate the reasons that certain groups of cancer respond better to certain treatments, in order to find clinical markers, or to identify new targets for breast cancer treatments.
Meanwhile, in a similar study published earlier this week, scientists identified a new biomarker that can accurately predict tumor metastasis in women with breast cancer. And in another related study, published in July, researchers from Rockefeller University identified a protein that makes breast cancer cells more likely to migrate to other parts of the body. What’s more, the protein, called TARBP2, points to the possibility of new cancer therapies that target this “master regulator” that helps set metastasis in motion.