Investigating The Locations of Lymphocytes in Healthy Lungs Using Cyclic Fluorescent Antibody-Stained Imaging Data

Abstract

The aim of this research project was to create a bioinformatic image analysis pipeline for the investigation of lymphocytes in healthy human lungs. The lung is an important organ to study as it provides a fast way for pathogens to enter the body. Therefore, the pulmonary immune system must be further investigated in both healthy and diseased tissue. In order to do comparative studies, we need to carefully characterise healthy tissue, which was done in this study using multiplexed image analysis.The data consisted of three formalin-fixed paraffin-embedded (FFPE) tissue samples, divided into six regions of interest (ROIs) which were imaged using a multiplexed cyclic fluorescence-based antibody staining approach with 26 antibodies. State-of-the-art downstream analysis methods and tools were applied to stitch the images, segment the cells, perform dimensionality reduction, and investigate the geographical properties of the data. A manual gating approach was used to infer cell types based on marker intensity expression levels and binary thresholds. The cell type inference was tested using multiple statistical plotting methods. A cell visualisation app was created to provide an interface for further analysing dimensionality reduction plots, with which cell type misclassifications could be examined and explained. The locations of T cells in the tissues were investigated using distance analysis methods and no particular trends were found regarding which structural regions of the tissue the cells were located in. Regardless, this methodology was a successful approach to analysing multiplexed imaging data. In the future, it can be applied to further classifying lymphocytes in healthy lungs, for example to distinguish circulating cells from tissue-resident memory cells. This workflow can also be used to do future comparative studies of patient samples.

Popular Abstract

An Image Analysis Pipeline For Investigating Immune Cells in Healthy Lungs

Although a lot of research nowadays focuses on diseased tissue, it is just as important to investigate healthy tissue as well. Immune cells are not only present in infected lungs but also in the lungs of a healthy person. If immune cells are found in lungs, it can be due to different reasons: They could be memory cells circulating the body which just happen to be in the lung tissue at the time of data collection, or they could be a type of memory cell that stays in the tissue after infection. This second type of immune cells is called “tissue-resident memory cells” and while they help the body respond faster to incoming threats if they return a second time, they can also turn against the body and cause autoimmune diseases. Researching these cells further is just one reason why the study of immune cells in healthy lungs is very relevant. It is also important to know the behaviour and locations of immune cells in healthy tissue so that one can know the difference between healthy and diseased patients.



In this study, we had access to lung image data, and the aim was to create a functioning analysis pipeline to find out more about immune cells. The data consisted of six different image sample sections that were stained with different antibodies, which are proteins that bind to very specific other proteins in the tissue. Before staining, the antibodies were also attached to fluorescent chemical compounds that can be picked up by a camera. The fluorescent dye then shows up only on the cells that are marked by the specific antibody, and this is how images of different structures in the sample can be captured. For the data used in this study, a multiplexed cyclic imaging technique was used. That means that each sample was stained, a picture was taken, and the staining was removed again, many times over and over in a cyclic procedure. Through this, the same sample can be imaged with many different biomarkers showing many different cells and structures in the tissue, without harming the tissue in the sample.

We then used different bioinformatic tools and methods to analyse the data. From a computational point of view, images are nothing more than matrices of pixel intensity values. So how do you quantify an image with so many cells to analyse it computationally? The first step is to perform cell segmentation, which means that an artificial intelligence algorithm is used to identify every single cell on an image and save its coordinates. Having detected every cell with as high accuracy as possible, the next step is to be able to distinguish between them. In order to find out more about immune cells, one has to know which of the ~10,000 cells in one image are immune cells and which are not. Often, in research, an expert is called in to do this task and to manually identify each and every cell. Instead, here, we used a computational approach. We verified the cell type identification methods with various statistical plots.

Next, we used a method called dimensionality reduction where the goal is to find emerging patterns or clusters of the data. Based on the cells’ biomarker intensity values, the algorithm creates patterns that show which cells are closer together and therefore likely somehow related, and which ones are further apart and probably different. By using this method, we could also confirm that our computed cell types were quite accurate, because the dimensionality reduction algorithm formed the same clusters as our cell type groupings.

There were of course also some ambiguities when it came to the inferred cell types. The cells that were ambiguous were classified as having a “mixed” cell type and then further investigated. When checking only the “mixed” cells with the dimensionality reduction methods, we found some very clear and distinct clusters. To analyse those further, we created an app which the user can use to find out more about the cells in the different clusters. By hovering over cells or selecting groups, more information about those cells is displayed that would otherwise not be available. Also, the cells can be traced back to their original locations in the image which is also automatically displayed in the app. With this new method, we were able to find out that a lot of the ambiguous cells came from transition regions between cell types. For example, in the area around an airway, there are epithelial cells, which line the airway, but they are surrounded by a thin layer of smooth muscle cells. Cells that are on the edge of one region can have fluorescent signal from both different antibodies on them, making them ambiguous and their classification hard.

Finally, we also investigated the geographical distances between cell types and locations of immune cells. We did this by calculating pairwise distances between the immune cells and other structural cell types, and by calculating every immune cell’s closest neighbour. The immune cells seemed to be very evenly distributed across the tissue and not extraordinarily close to any one structural region.

Overall, the pipeline we created was successful. It can be used to further research immune cell types in healthy lungs, or it can be applied to diseased data in the future.



Master’s Degree Project in Bioinformatics 60 credits 2024
Department of Biology, Lund University
Advisor: Dr. Nils Norlin
Department of Experimental Medical Science, Faculty of Medicine

• author: Schulz, Felicia
• supervisor: Nils Norlin
• organization: Degree Projects in Bioinformatics
• course: BINP52 20232
• year: 2024
• type: H2 - Master's Degree (Two Years)
• subject: Biology and Life Sciences
• language: English
• id: 9176800
• date added to LUP: 2024-10-22 16:17:00
• date last changed: 2024-10-22 16:17:00