Currently, diagnosis and therapy strategies for cancer patients strongly relies on cell- or tissue-based visual assessment of tumor markers.
However, in most clinical, pathology and research centers the criteria used for their visual classification is merely qualitative and thus strongly operator-dependent. An inaccurate biomarker evaluation can be as harmful as an inappropriate drug choice in cancer treatment, leading to erroneous treatments with no benefit or even injury for the patient.
Thus, there is an urgent need to introduce novel and quantitative methods to improve the assessment of these biomarkers.
In this project, our goal is to develop new bioimaging quantitative tools that will combine morphological characteristics, protein expression profiles and detailed spatial analysis at subcellular level of cells and tissues. Namely, we will focus on cell receptors, cell adhesion and cytoskeleton molecules associated to gastric cancer.
To accomplish this goal, we need (1) to develop novel tracers that improve pattern recognition and facilitate quantification of molecular and cellular/tissue features, and (2) to design new methods to automatically or semi-automatically compute and integrate cell phenotypic and molecular characteristics.
Specifically, novel types of fluorescent probes as quantum dots will be used and combined with advanced optical image acquisition to attain accurate biomarker information. Alterations in tissue architecture, cell shape, polarity and size, nucleus-cytoplasm relative size and positioning, will be also numerically evaluated to discriminate cancer.
The team is well prepared to undertake this project since over the last two years, we have already designed algorithms that are able to compute representative profiles of expression of molecular biomarkers from archive in situ immunofluorescence images. The innovative potential of this technology comes from the fact that it can rigorously map and quantify the level of expression of such proteins, as E-cadherin, in heterogeneous cell populations. Additionally, we have already performed other pilot studies in which we applied this analytical approach: 1) to characterize and quantify cancer related biomarkers such as p120 and other proteins from the adhesion complex; 2) to determine the therapeutic impact of biological treatments in breast cancer related signaling; 3) to evaluate aneuploidy and determine cell cycle phases from archive in situ heterogeneous cancer cell populations; and 4) to assess morphometric cellular features to identify cell populations with aberrant adhesion (See preliminary results in attachment).
Based on previous experience and preliminary data obtained, our specific tasks in the project are:
(1) To generate new methods of signal emission to improve image acquisition for quantification purposes. More specifically, we will explore the application of quantum dots to capture molecular and organelle information, performing multiplexed image tracking. Quantum dots are semiconductor nanocrystals with the ability to be conjugated to proteins, yielding clear advantages in comparison to organic dyes: broad excitation spectra, narrow emission spectra, tunable emission peaks, long fluorescence lifetimes, and diminished photobleaching. To improve the spatial resolution of the tracer’ signal, new pre-processing algorithms for signal conditioning and noise removal, signal registration, alignment and reconstruction will be developed, considering the specificities of cell populations and tissues.
(2) To generate and validate novel algorithms that integrate phenotypic and molecular characteristics for stratification purposes. This task aims at computing numerical features related to tissue/cell morphology and protein expression from cell population and tissue images, usually distorted and corrupted by noise. For that we need to design new algorithms in a Bayesian framework, able to deal with this type of highly ill-posed inverse problem and capture proteomic and phenotypic data.
For diagnostic classification purposes we will provide a binary indicator of gastric cancer (GC) and the corresponding likelihood factor. Data mining techniques will be used to guide us in the selection of the most GC-correlated features, which will be combined by the classifier in a non-linear manner and tuned using a machine learning approach.
In this project we will exploit gastric cancer as a model since IPATIMUP is internationally recognized as a reference centre for gastric cancer research. Importantly, the team has already available a set of biological reagents and a tumor bank, awarding us a significant competitiveness for validating our bioimaging approaches and to standardize the appropriate parameters for identifying and characterizing cancer cells.