Pan-cancer validation of a lung-adenocarcinoma-derived gene-expression-based prognostic predictor

Abstract

Gene expression profiling can be used to classify lung adenocarcinoma tumors into molecular subtypes that also correlate with patient prognosis. Knowing these subtypes can help clinicians save lives by directing surgery and adjuvant therapy efforts, provided adequate clinical validation. It is less well-known how generalizable prognostic gene signatures derived in one malignancy are in a pan-cancer context. In this study, the single sample predictor CLAMS (Classifier for Lung Adenocarcinoma Molecular Subtypes) was tested in over 14,000 samples from 32 cancer types to classify samples into better (named TRU) or worse (named nonTRU) prognosis. Of the malignancies that presented both CLAMS classes, survival outcomes were significantly different for cancer in the breast, brain, kidney, and liver. In addition, samples classified as better prognosis by CLAMS in these organs were generally of lower grade and had better/intermediate prognosis according to other type-specific classifications. As example, most breast tumor TRU samples were from the luminal A subtype and had a low risk of recurrence prediction. Furthermore, samples classified by CLAMS as better prognosis were always less proliferative than their worse prognosis counterparts in a pan-cancer context. We also identified other malignancies that have a potential prognostic component in cell proliferation, such as mesothelioma. This takes us one step closer to understanding how gene-expression-based single sample predictors act, and how to derive tools useful for prognostication that are efficient across organs.

Popular Abstract

Deborah Figueiredo Nacer de Oliveira

Can one program predict patient prognosis of different types of cancer?

When it comes to treating cancer, what they say is true: the more you know your enemy, the better your chances are of winning the war. This knowledge is beneficial both for institutions and patients. The better you know how to treat cancer, the more institutions can save resources when tending to patients’ needs, allowing them to help more people for longer. More information is also better for the patients: if you receive the right type of therapy in the right amount, you will likely have the best results with the least recovery time and side effects. While we are still not able to truly personalize cancer treatment, we are able to find increasingly smaller groups of patients that seem to respond more uniformly when treated equally. This is the goal in the healthcare system: to have the best possible treatment for each patient or patient subgroup.

Cancer is not just one disease; it is many different malignancies that can be further divided into subtypes. Subtypes can be linked to better or worse prognosis based on the likelihood of disease progression and remission. One way of identifying such subtypes is by measuring how much specific genes are being used by tumor cells in a patient. If gene A is more used than gene B, then that sample is more likely to be of a specific class. When we measure many genes, sample classification can become more complex and even problematic. This is where computer-aided classification becomes important, such as by using machine-learning algorithms. Comparison of gene use is then done by algorithms (programs) to determine whether a sample should be considered as belonging to a group having better or worse prognosis. In our group studying lung cancer, we developed a program named CLAMS (short for Classifier of Lung Adenocarcinoma Molecular Subtypes) which does exactly that. CLAMS originally classifies samples of this specific lung cancer (adenocarcinoma) into prognosis subtypes based on molecular characteristics (the gene use comparison mentioned above).

To help even more patients, we would ideally have programs that can do the same thing for every type of cancer. To see how much our program could help when it came to cancer outside of the lungs, we used CLAMS to classify over 14,000 tumor samples from 32 different cancer types. We found that CLAMS could be used for other malignancies as well, mainly in the breast, brain, liver, and kidney. This was shown by differences in patient survival between the groups classified as better versus worse prognosis by the program. For these types where CLAMS showed good results, patients classified as better prognosis had less chances of dying in the same time span than patients classified as worse prognosis. We also found that CLAMS classification generally agreed with other malignancy-specific subtyping tools when it came to prognosis.

One interesting fact about CLAMS is that its prognosis classification seems to be linked to cell proliferation: tumor samples that more frequently used genes connected to cell division and growth were more often linked to worse development of the disease. Looking into cell proliferation with more detail, we found that it can be used to divide patients into groups with better and worse prognosis reflected in survival over time in other cancer types as well, not just the ones CLAMS could be used for. This study allowed us to understand more about how programs that compare genes for subtyping tumor samples behave when used on different cancer types with unique characteristics. This will hopefully lead us to developing tools that are more efficient for several kinds of cancer, saving time, resources, and above all, lives.

Master’s Degree Project in Bioinformatics 45 credits 2020
Department of Biology, Lund University

Advisor: Johan Staaf
Division of Oncology, Lund University, Medicon Village

• author: Oliveira, Deborah
• supervisor: Johan Staaf
• organization: Degree Projects in Bioinformatics
• course: BINP51 20192
• year: 2020
• type: H2 - Master's Degree (Two Years)
• subject: Biology and Life Sciences
• language: English
• id: 9012098
• date added to LUP: 2020-06-02 11:22:22
• date last changed: 2020-06-02 11:22:22