Summer Undergraduate Research Experience in Bioimaging, Engineering & Technology

Information on 2008 SUREBET Participants

Stamping of ECM proteins onto glass for selective cell adhesion

Complex 3D patterns can be created by combining patterning methods with the ability of bio-molecules to arrange themselves into organized structures. Micro contact printing is one such patterning technique that offers the opportunity to pattern substrates with wide range of feature sizes. The advantages of microcontact printing include low cost, high resolution and its application to non-planar surfaces.

Patterning extracellular matrix proteins such as Fibronectin on activated functionalized substrates result in stable patterns. This is as a result of strong chemical bonds that form between fibronectin and the functionalized surfaces. Fibronectin islands then serve as cell adhesive regions. Micropatterning of bio-molecules on surfaces has a number of applications, which include modulation of cell-substrate interactions in biomaterials and tissue engineering, the fabrication of multi-analyte biosensors, and genomic arrays.

Molecular Biosensor Database (MBD)

Nowadays the automated analysis of microscopy images is an indispensable technique for exploring cell pathways. The design of a new way to label proteins is a critical step to improve the determination of the subcellular location of proteins. This project deals with the creation of a database, named Molecular Biosensor Database (MBD), to assist with the creation of new protein markers. The data stored in the MBD database are annotations of newly developed biosensors which label proteins within cells. This biosensor is composed of a fluoromodule, which is a combination of an apomodule (a genetically expressed gene), a fluorogen (a non-fluorescent dye that fluoresces in the apomodule), and a biological fusion partner (the target protein). Other important aspects of this project are to connect the MBD with the Protein Subcellular Location Imaging Database (PSLID) to store and analyze biosensor images and to provide a public access to the data via an internet interface. When images of cells with biosensors are acquired, they are then analyzed with tools available in PSLID.

CaCo2 Cell Sheet Imaging for Location Proteomics

Automated analysis of the location of proteins is useful in determining statistical differences in protein patterns that may be undetectable to the human eye. The goal of this project is to collect three-dimensional, fluorescence microscope images of the CaCo2 cell line. Each image will contain multiple cells. Each cell has one of two possible tagged proteins. This will be accomplished by infection of CaCo2 cells with GFP, and raising cell lines from 1 cell colonies. Then two of the cell lines will be mixed. In order to evaluate our automated methods, a ground truth must be established. One of the 2 GFP patterns will also be tagged with a nuclear DS-Red. All the cells will also be tagged with Phalloidin and DRAQ5. These will aid in segmentation. There will be 4 channels acquired. Thus far, cells have been infected with GFP and transfected with DS-Red. They each have been sorted using a flow cytometer (1 cell per well in a 96-well plate). The DS-Red positive cells will be infected with GFP and sorted for GFP and DS-Red positives. The cells have been maintained and will be used to create multicell images for analysis.

Extracting frequency features from 3D fluorescence microscopy images

Fluorescence microscopy images of microtubules differ in their patterns from the centrosome. In order to perform 3D microtubule image matching, features that capture spatial variations were implemented. Discrete cosine frequencies from sub-blocks of 3D images were extracted and used for image matching.

Quaternion Feature Sets

We explore the value of color features in classification of biomedical images. For example, pathologists use color when examining histological images. In the otitis media dataset, acute middle ear infection cases exhibit a distinctive marked redness; non-acute cases have instead more of amber and gray, while the normal cases are grayish-pinkish in color. The question now is: what is a color feature? What is most commonly done is to merely extract the grayscale-based feature from each of the (usually three) color channels individually, resulting in a feature vector that can be considered a color feature. The issue is the potential loss of discriminating information by treating each of the color features separately. No matter the representation, a single number cannot represent the apparent color of a single image pixel although the “true” color might be, that is, the hue in HSV. Recognizing this property of color we must find features that take advantage of the inherent vector nature of color. The mathematical framework we will use for this task is referred to as quaternions, which can be thought of as ``hypercomplex'' numbers. While one could still use vectors and matrices, using quaternions allows for the representation of a single color pixel as a single quaternion, allowing for the use of most usual algebraic operations in an intuitive way, such as a Fourier transform. Using quaternions to unify our representation of color images, we can develop color-based feature sets and algorithms for classification. One immediate direction is to update existing compatible (not all algorithms lend themselves to, some of which exist already).

Gene Knockdown using RNA Interference in NIH3T3 Fibroblasts

RNA Interference is a recently discovered gene knockdown mechanism by which short (20-23 nt) RNA molecule associates with mRNA by complementary base-pairing and mediates degradation or translational inhibition of the RNA. In our post-genomic era, we need tools for probing the functions of genes. RNA Interference (RNAi) enables screens of gene function, whereby many different genes (or even all the genes in the genome) are downregulated in parallel and the results are computed by automated methods. While RNAi represents an exciting concept, there are some potential pitfalls. Certain cell types do not readily internalize exogenous RNA molecules, limiting efficiency. In order to facilitate uptake by cells, transfection reagents are required, though these agents can have toxic effects to some cell types. Finally, the dynamics of gene knockdown can vary between cells types. We sought to characterize gene knockdown by RNAi in NIH3T3 cells in this work. Our results include the type and amount of transfection reagent required for optimal transfection and knockdown efficiency while limiting toxicity. We found that in this cell type, the knockdown effect is persistent for at least 7 days. Continuing work is expanding on this work in other cell types, including neural stem cells.

Applying A Graphical Model Segmentation Method to CaCo-2 Images

Cell segmentation is a necessary preprocessing step in subcellular pattern recognition and has been an active area of research over the last decade. Different segmentation methods have been developed, yet little has been done to standardize comparison between them. In this work, we evaluated methods to segment images into single cells by defining criteria for comparing them. We tested Graphical Model, Watershed and Voronoi segmentation approaches. The images we segmented are fluorescence micrographs of CaCo-2 cells, grown in a monolayer. The cells were stained with DRAQ5 for the nuclei and phalloidin for the cell membrane. In this work, we developed a preprocessing method that allows for the Graphical Model segmentation to be applied to the CaCo-2 data; to do this we defined an approximate method for extracting cell edges from the cell membrane image. We first manually segmented the multiple-cell images to obtain a ground truth against which the different segmentation methods can be compared. We chose one image to train parameters for the different automatic segmentation approaches, and then applied the automatic methods to the rest of the images. We defined and calculated metrics that determine the region and pixel overlap between the automated methods and the ground truth, and used these to evaluate the different methods. We found that Watershed segmentation performed best in all of the metrics, while the Graphical Model segmentation gave lowest performance on a majority of the metrics. This low performance indicates that improved edge detection is needed for Graphical Model segmentation.

Hyaluronic acid derivatives for cellular encapsulation

Enclosing cells in capsular environments is of interest in areas of research ranging from, for example, the treatment diseases which can be cured through cellular or protein replacement to understanding basic biological questions. The development of materials for encapsulation which have cell-interactive properties may facilitate these lines of research. Hyaluronic acid (HA), for example, is a native extracellular matrix constituent which has been shown to induce cell adhesion and proliferation. It is inherently biocompatible, contributes to matrix structure, and is actively engaged and remodeled by cells. The synthesis derivatives of HA which carry opposing charges will allow the creation of capsules by polyelectrolytic complexation. These capsules may provide cell-interactive three dimensional environments which can be engineered for diverse applications.