No Label? No Problem!

Today, the gold standard for clinical cancer diagnosis is the visual inspection of H&E-labeled tissue thin sections by a pathologist. Coming in at a close second is the immunohistochemical stain. But proper evaluation of these labeled samples depends on the pathologist’s expertise and the reliability and reproducibility of the staining – aspects that may vary from one situation to the next.

To overcome that obstacle, we established a technique for classification of tissue thin sections that is both label-free and inter-/intra-observer-independent. Our infrared (IR) imaging uses spatially resolved infrared spectra as fingerprints for biochemical disease status. The spectra are automatically classified through bioinformatics, eliminating the variability introduced by a subjective observer.

In this new approach, we image the unstained tissue with an infrared microscope; the image is then classified bioinformatically. The resulting index color images represent the tissue classification – including cancer type, subtype, tissue type, inflammation status, and even tumor grading. As an example, we established a label-free classification of thoracic tumors and their subtypes with a sensitivity of 91 percent and a specificity of 97 percent compared with histological annotation (1). We even achieved an accuracy of 96 percent in the differential diagnosis of the subtypes of lung adenocarcinoma.

The main hindrance for clinical use has been the slow data acquisition speed of older IR imaging systems. In a pioneering study, we showed for the first time that a new laser-based wide-field IR imaging microscope could be used to accurately classify colorectal cancer tissue 180 times faster than previous IR technologies (2) – speed that makes it suitable for clinical use. Colorectal cancer is one of the most common tumor diseases and has high survival rates if caught at an early stage. We studied 100 samples of stage II and III colorectal cancer tissue and 20 tumor-free tissue samples and developed a workflow that enabled us to classify tissue for diagnosis in about 30 minutes (for large thin sections; smaller regions of interest take only a few minutes). Better yet, our new method carried a sensitivity of 96 percent and a specificity of 100 percent. What does this mean? In a very short time, we can gain an understanding of the tumor and its microenvironment without the risk of operator or equipment bias.

The spectral data obtained from the microscope can easily be combined with omics techniques to provide both spatial and molecular resolution. We recently demonstrated this approach for diffuse malignant pleural mesothelioma, a type of cancer mainly caused by asbestos exposure (3) to identify proteins expressed differently in two different tumor subtypes. The next step is detailed bioinformatic analysis to select biomarker candidates from the proteins identified. In our demonstration, all of the clinical immunohistochemistry biomarkers used today for mesothelioma could be identified on a small number of test samples.

Such automated image analysis will one day support pathologists in their daily routines and provide a second opinion in challenging diagnostic situations. It’s our hope that it will pave the way for precise diagnostics and more specific biomarkers in precision medicine.