NEW YORK (GenomeWeb) – A team of researchers have been looking at various types of genetic biomarkers to provide oncology researchers and clinicians with crucial information to better predict patient outcomes.
In a study published last week, for example, they found that copy number alterations (CNAs) in cancer driver genes can offer more valuable information on cancer aggressiveness and the likelihood of patient death than mutations in these genes, a finding they are planning to further explore next year in partnership with Northwell Health.
By analyzing genomic changes in public datasets, the authors identified factors driving aggressive disease while sidestepping some of the confounding issues that plague targeted biomarker studies, including patient-selection bias and post-hoc hypothesis testing. Their findings also improve on typical prognostic markers like tumor de-differentiation and lymph node infiltration, which can provide incomplete information on a patient's most likely clinical course. Their insights could be a boon for researchers and clinicians seeking improved strategies for patient risk assessment and treatment stratification.
"There is a lot that is known about the genetic differences between normal cells and cancer cells, however, there is much less known about what differentiates benign cancer cells from aggressive metastatic cells," said Jason Sheltzer, a fellow at Cold Spring Harbor Laboratory and one of the authors on the study. Having access to clinical and genomic information from large-scale studies such as The Cancer Genome Atlas (TCGA) "gave us an avenue for investigating this question," he added.
According to the paper, published last week in the journal eLife, the researchers analyzed genomic profiles from nearly 18,000 tumors with known patient outcomes and found that mutations in cancer driver genes contained little information on patient prognosis. However, CNAs in these same driver genes were found to have "significant prognostic power."
Previous studies have linked CNAs and cancer patient outcomes. A 2016 study done by researchers at the Seoul National University College of Medicine, for example, evaluated the prognostic importance of a specific copy number change in breast cancers in over 900 invasive breast cancer cases. Another 2016 study, by researchers at the University of Manchester, used copy number changes to classify small-cell lung cancer patients as either chemosensitive or chemorefractory. A more recent study identified copy number profiles that correspond to poor outcomes in patients with a subtype of liver cancer.
In contrast, the eLife study found a much larger number of CNAs associated with health outcomes in a variety of tumor types. Specifically, they found 108 significant associations between copy number changes and patient outcomes compared to just 23 associations between driver gene mutations and patient outcomes. Furthermore, for 28 of the 30 driver genes that were analyzed in the study, copy number information proved more prognostic than mutational status in many more cancer types. The researchers also showed that focal CNAs were linked to worse outcomes compared to broad CNAs. These CNAs remained prognostic even when controlling for cancer stage, grade, TP53 status, and aneuploidy. Finally, they identified a subset of CNAs that were predictive of patients' response to certain treatments.
As a first step, the researchers devised a data analysis pipeline and validated it on data from just under 10,000 patients with 16 kinds of cancer from the TCGA dataset. "What we did was develop some systematic tools that allowed us to ask similar questions again and again of the different pieces of the data," Joan Smith, a software engineer at Google and the first author on the paper, explained. The study, however, was not part of any Google-sponsored research initiative and the company has no involvement in the study. Smith was involved in this research on her own time.
Specifically, for each tumor type and dataset, the research generated statistical models that linked the presence of genetic features with clinical outcomes. These models generate scores that represent the significance and direction of the associations.
The researchers used these tools to analyze the clinical impact of CNAs in 30 cancer driver genes and found that the copy number of both oncogenes and tumor suppressors was linked to patient outcomes. Specifically, they found that amplification of EGFR, PIK3CA, and BRAF, and deletion of CDKN2A, RB1, and EP300 were strongly associated with shorter patient survival times in four or more cancer types, respectively. Furthermore, CNAs proved prognostic even in cases where mutations in the genes had not previously been linked with patient outcomes. For example, although mutations in PIK3CA are not informative for patient outcome, PIK3CA copy number alteration was associated with outcomes in glioma as well as in breast, colorectal, , lung-squamous, pancreatic, and prostate cancers.
Sheltzer and Smith also found that the copy number changes remained prognostic even when cohorts were grouped according to cancer subtype. For example, they found that amplifications in MYC and PIK3CA were prognostic across multiple tumor subtypes. However, the prognostic value of CNAs was more difficult to ascertain for rarer tumor subtypes due to small sample sizes.
As part of their analysis, the researchers also looked for links between mutations in coding regions and patient outcomes. Of their analysis of the 30 most frequently mutated cancer driver genes, only two – EGFR and TP53 – were associated with prognosis in more than two tumor types. In contrast, mutations in other common driver genes — including KRAS, PIK3CA, and BRAF — were not significantly linked with patient outcomes. One exception was driver genes in glioma cases, where their analysis revealed alterations associated with both poor and favorable patient prognosis.
They checked to see if the patients had been treated with targeted therapies that could have alleviated the harmful effects of the driver mutations. However, few patients received treatment prior to sample collection, according to their findings, and removing those patients from the population did not have a significant effect on the association scores.
A separate analysis within the study focused on 1,000 patient-derived xenografts and revealed several biomarkers that indicated susceptibility to cancer therapies. In total, 49 percent of the prognostic biomarkers identified in this study indicated that the patients were susceptible to cancer therapies. For example, the researchers found that CDKN2A deletions were correlated with sensitivity to a combination of a CDK4/6 inhibitor and an mTOR inhibitor. They also found that gliomas with mutations in STAG2 were sensitive to the PARP inhibitor olaparib.
Sheltzer and Smith have shared their findings, including the biomarkers they identified and their associated scores, in a public database. In addition to CNA information, the resource also includes data from other unpublished studies they conducted looking at the prognostic value of other kinds of information such as methylation and microRNA data. Researchers could, for example, "look up their favorite gene in our databases and see [that] there is this interesting association in kidney cancer or this association in brain cancer," Sheltzer said. "That might help spur future research to answer [some of] those questions that we cannot answer just yet."
They also plan to further validate their findings on addtitional datasets. Ultimately, the hope is to find a way to help clinicians incorporate tumor CNA information into their treatment plans. As part of their next steps, the researchers are partnering with Northwell Health, a non-profit healthcare organization, in New York to analyze data from patients that are treated within the system. "With this paper published, we are ready to start setting up the structure for that data collection and hopefully that will be something we work on for 2019," Sheltzer said. It is not clear at this stage how much data the researchers will have access to but because Northwell treats thousands of patients annually, Sheltzer noted, so the numbers could be significant.
Through these follow-up studies, they hope to find answers to questions such as what effect CNAs have on tumors. For example, "[we] have a good understanding of what a cell with wild type PI3 kinase looks like and what a cell with mutant PI3 kinase looks like," Sheltzer explained. "What we don’t understand is how having five copies of PI3 kinase or ten copies of PI3 kinase or 20 copies of PI3 kinase affects cells. That's a level that we really don't have a good answer for yet."
The researchers also plan to publish at least one additional paper that will share the results of their explorations of the prognostic value of other kinds of genetic information collected from patients. This includes the effects of microRNAs, methylation, and transcription on cancer prognosis. "We looked at all sorts of features of primary tumors … [and] we are hoping to include that in our next paper," Sheltzer said. "For this first one, we just focused on … mutations and copy number changes."