Lung cancer is the leading cause of cancer-related deaths in Quebec and Canada, killing more individuals than prostate, colon and breast cancers combined. In a recent collaborative study, GCI researchers profiled the cellular composition and spatial organization of the tumour immune microenvironment in over 475 lung cancer patient tumours using highly multiplex imaging technology.听
.
Meet the co-first authors
Mark Sorin, GCI trainee, is a MD-PhD student from the Department of Human Genetics in the labs of Prof. Logan Walsh and Dr. Jonathan Spicer. His training in this MD-PhD program was instrumental to his success. Follow him on Twitter @sorin_mark
Elham Karimi, Research Associate in the lab of Prof. Logan Walsh at the GCI.
Morteza Rezanejad, Ph.D., is a senior Machine Learning Scientist working in the biotech industry. He contributed to this article in his personal capacity as a postdoctoral fellow at the University of Toronto. Follow him on Twitter @mo_rezanejad
听
听
听
听
听
听
What is new in this paper?
Mark: Lung cancer is the largest cause of cancer-related death in both men and women. A big challenge in how we treat lung cancer is that we do not know which early-stage patients will be cured and which patients will have a cancer recurrence after surgery. We used imaging mass cytometry, a novel technology that allows researchers to comprehensively characterize the immune cells located within and adjacent to the tumour, to describe the tumor microenvironment of lung cancer. Imaging mass cytometry identifies cells by visualizing them with up to 50 markers, significantly more than was previously possible. We combined artificial intelligence with imaging mass cytometry to predict post-surgical progression for lung cancer patients by using a tiny 1mm2 section of their tumor. Our findings can help guide clinical management after surgery by predicting which early-stage lung cancer patients would benefit, and which patients would be harmed by chemotherapy.
Elham & Morteza: This paper is about using imaging mass cytometry (IMC) to characterize the tumor and immunological landscape of 416 samples from patients with lung adenocarcinoma across five histological patterns. Using an in-house built-in system, we were able to segment more than 1.6 million cells and identify their phenotypes. We show that by using spatial information that is obtained from images we can improve our understanding of microenvironmental features.
How does it add to what was already known?
Mark: The tumor microenvironment is an ecosystem of interactions between tumor cells, immune cells, and various structural cells. While the tumor immune microenvironment of lung cancer has been profiled before, the development of a new technology called imaging mass cytometry has enabled to resolve this ecosystem at a depth that was not previously possible. Our study represents the most comprehensive spatial resolution of the tumor microenvironment in lung cancer and our cohort of 416 patients gave us the power to resolve new important relationships between immune cells and survival. We show that many different cell types can interact in cellular communities that we called neighborhoods and that B cell neighborhoods had a strong association with improved survival outcomes. Overall, our study highlights the importance of spatial information in studies of the lung cancer tumor microenvironment.
Elham & Morteza: We think one of the key findings of this study is that this paper pinpoints the importance of image-based profiling methods in tumor immune microenvironment analyses. Here, by using the imaging mass cytometry technology, we show that it is possible to look for specific relationships that are significant in cellular communications and also potentially relate them to the prediction of clinical variables.
How did you discover it? What led you to ask the question and explore it this way?
Mark: One of the big challenges for the use of artificial intelligence in biomedical research is the availability of large datasets. Our cohort, which was composed of 416 patients and over 1.6 million cells allowed us to have enough power to perform these machine learning predictions.
Elham & Morteza: We think it is fair to say that we were inspired by other works in the domain that show that the cellular spatial information that is available in our images is important and can be predictive of a substantial amount of information. We were lucky to have access to a cohort of a relatively large number of patients and we tried our best to extract the most accurate cellular information from our images (including correctly segmenting the shapes of nuclei as well as assigning those cells detected with a correct phenotype). Going through these processes, we came up with many hypotheses and tried every single one of them to see which one was supported by evidence from our processed data. This was a completely incremental process, and it happened over a long span of time.
Why is this finding important?
Mark: A substantial proportion of early-stage lung cancer patients will have a recurrence of their cancer after surgery and could benefit from chemotherapy. Unfortunately, many of these patients will decline chemotherapy after surgery to avoid treatment toxicity. If we could know which patients will have a recurrence after surgery, we could tailor our use of chemotherapy to these patients and avoid chemotherapy toxicity in patients who will be cured by surgery. Our findings are important because these predictions were made with the use of a 5-micron section of a single 1 mm2 tumor core, material which is readily available from surgical resections or biopsies.
Elham & Morteza: In our opinion, this study is important from another important perspective too, which is, there is important spatial information encoded in the images that could be exploited at many levels, given that the imaging technology can be applied at the multi-plex level. The use of images in this domain is still relatively new and through this study, we are showing that this community can benefit from this modality of information alongside other modalities of information for future research. Besides, this research paves the way to find answers to more important questions, i. e., given enough cores for a cohort of data, is it possible to predict recurrence with high accuracy? Our findings demonstrate that we have some initial evidence that we can use to build a foundation for future exploration.
What are the practical implications?
Mark: One of the challenges of using our artificial intelligence predictions is that imaging mass cytometry is not available in the clinic. Clinical pathology services use lower-plex technologies such as immunofluorescence which are often limited to three markers. As a result, we set out to determine the minimum number of markers required to make meaningful predictions about the progression of lung cancer patients after surgery. We found that by using six markers, we were able to come very close to our findings with our full imaging mass cytometry panel. These findings are very encouraging for the possibility of using artificial intelligence to inform the post-surgical clinical management of lung cancer patients by increasing the cures rates for patients likely to have a cancer recurrence and limiting toxicity for those likely to be cured by surgery.
Elham & Morteza: We are hopeful to expand our study extensively to a much larger dataset so that we can build a system that uses a minimal number of markers and is able to make meaningful predictions that could help clinicians with patients suffering from lung cancer.