|Send to printer »|
Feature Articles : Oct 15, 2008 ( )
Exploiting Advances in Image Analysis
Technology Provides an Unbiased Perspective at the Population and Single-Cell Levels!--h2>
With recent technological advances, we witnessed a profound transformation in our ability to delve into intimate subcellular structures and understand dynamic processes in their true complexity, unveiling details that previously were beyond reach.
In areas from microbiology, drug design, and forensics to astronomy, defense, and art-fraud detection, these advances have become reality thanks to image analysis, which increasingly shapes the bridge across seemingly unrelated disciplines.
Which is the most exciting feature that all image-analysis applications have in common, irrespective of the experimental setting? “Making the invisible visible,” said Gaudenz Danuser, Ph.D., associate professor of cell biology at the Scripps Research Institute.
At the “International Conference on Intelligent Systems for Molecular Biology (ISMB)” held recently in Toronto, he exemplified this in the case of intracellular forces, and he showed that using computer vision it is possible to study small deformation of cellular structures such as the actin cytoskeleton, and then deduce the force necessary to generate those deformations.
While certain imaging techniques explore populations as a whole, it has become increasingly clear that cell-to-cell fluctuations within apparently homogeneous populations are more significant than previously envisioned. Image analysis at the single-cell level needs, therefore, to gain momentum in order for us to comprehend a variety of cellular processes in their true complexity.
“Image analysis at the level of single cells is still an orphan field compared to, for example, MRI, CT, or radiology,”, said Dr. Danuser, who leads one of the few groups (perhaps three or four worldwide) that systematically use computer vision to understand the mechanistic basis of certain cellular functions.
Dr. Daniel’s lab is working on new mathematical methods to measure in images, sequentially, the interactions between multiple molecular components and their activities in different combinations and in various experiments. The lab subsequently wants to build a model of the probed pathways, which may consist of tens to hundreds of components, an approach Dr. Danuser calls computational multiplexing.
His lab applies this framework to studies of cell division, cell migration, and endocytosis, basic cellular functions known for their pronounced heterogeneity. In this way, automation addresses a problem that arises with manual measurements—which are more prone to bias.
“I think that is really the number-one limiting factor in cell biology research,” emphasized Dr. Danuser, “that consciously and subconsciously investigators select the data they need to prove their hypothesis and ignore 99% of all the other data within an image that potentially would say the opposite.”
Image analysis plays increasingly important roles in understanding cell-to-cell variations under a multitude of circumstances. Jeremy Gunawardena, Ph.D., senior lecturer in systems biology at Harvard Medical School and director of the Virtual Cell Program, focuses on signal transduction in mammalian cells. Signal-transduction cascades are responsible for processes as diverse as growth and development, immune and inflammatory responses, or malignant transformation.
According to the long-held conventional view, between the arrival of a signal to the plasma membrane and the onset of nuclear transcription, the intervening events were thought to be largely passive and their information processing capabilities received relatively little attention. It is now increasingly clear that these intermediary steps are far from simple, and instead involve “a complex network of proteins, a lot of positive and negative feedback, with many types of switches and cascades, and motifs,” said Dr. Gunawardena.
His group focuses on Erk signaling involved in the epidermal growth factor pathway, in an attempt to gain insight into the information processing that takes place along those intermediary pathways.
In another common assumption, cells within a population were thought to differ only slightly from each other, which is “not so bad, if they are mostly doing the same thing,” explained Dr. Gunawardena, because the mean would always provide a good representation of any cell within the population.
But in a surprising discovery made by his post-doctoral fellow Yangqing Xu, Ph.D., made and presented at the conference on “Systems Biology of Human Disease (SBHD)” held in Boston, it was revealed that Erk activation is not well represented by the mean, but instead follows a bimodal distribution, with high levels of activation in some cells and low levels in others, with the proportions of cells in the two modes varying in a time and EGF dose-dependent manner.
The researchers subsequently used high-throughput imaging to gain insight into cell-to-cell variability and characterize the statistical distribution of Erk activation across the population as a function of dose and time.
While cells within the population turned out to all share the same network, they exhibited differences concerning the network starting places. And since the starting conditions were different, the activation threshold could differ from cell to cell as well. A mathematical model confirmed that this mechanism could account for the observed bimodal variation. Additionally, in subsequent work that is being prepared for publication, experiments revealed that cells do not make activation decisions independently—rather they are “talking” to each other.
Imaging can artificially be divided into in vitro and in vivo approaches, and while both approaches have experienced rapid expansion, certain platforms that bridge the two are particularly appealing. Such is the case with the highly sensitive time-lapse imaging platform based on MATLAB software that Mei Zhang, M.D., Ph.D., senior scientist in the department of biology at Synta Pharmaceuticals, developed with Mats Holmqvist, Ph.D., of the Novartis Institute for Biomedical Research.
This platform comes with several advantages such as the ability to study primary cell lines and to perform quantitative analysis, and it provides an in vitro system to address targeting issues in drug development and predict the in vivo impact of therapeutic agents on cardiomyocyte physiology.
Furthermore, it is anticipated that this platform can investigate the cardiomyocyte toxicity of therapeutic agents under development. “It is an exciting time to look at the imaging field for drug discovery. The progress is amazing,” said Dr. Zhang.
While drug discovery saves lives, we should appreciate that prevention is better than a cure. In addition to drug design, imaging is an increasingly important player in prophylactic applications. Heart disease is responsible for 30% of worldwide mortality. Despite the huge number of lives cardiovascular conditions claim annually, about one-third of the affected individuals are unaware of being at risk.
“One of the things we are not doing a good job at is identifying individuals who have the potential to have a heart attack before they actually have it,” noted Simon Robinson, Ph.D., director of discovery biology at Bristol Myers Squibb Medical Imaging. “We are treating patients after the event rather than preventing them from having the event.”
Since prevention is so much more effective than treatment, both in terms of lives saved and of resources spent, developing new cardiovascular imaging agents becomes an important priority. At the “Drug Discovery and Development Therapeutics” meeting held in Boston, Dr. Robinson presented a recently discovered lead agent that enables noninvasive magnetic resonance imaging of arterial wall lesions.
This development allows visualization of the arterial cell wall thickening that is associated with atherosclerosis, along with the shape and size changes that are caused. In contrast, currently employed agents only look at the vascular lumen, and while they can identify arterial-wall changes that impinge into the lumen, they miss lesions that expand outward and do not change the arterial lumen despite having the potential to grow and rupture.
“Angiography only tells part of the story,” said Dr. Robinson. This is just one of the reasons that visualizing arterial-wall lesions will be essential to change the patient populations from a post- to a pre-event.
Some experimental approaches involve high volumes of data or greatly dynamic cellular events, and image analysis can now accomplish feats that just years ago were undreamt of.
David W. Andrews, Ph.D., professor of biochemistry and biomedical sciences at McMaster University, leads a biophotonics facility that bridges the gap between academic and pharmaceutical research.
“From an academic biologist’s or biochemist’s point of view, you want to generate as much data by yourself as possible,” explained Dr. Andrews, adding that while live cell-based assays are attractive, “we often forget how much data that is when performing high-content genetic screens or compound screens.”
At the “ISMB” meeting, Dr. Andrews talked about a recent genetic screen in which two fluorescent proteins were targeted to yeast endoplasmic reticulum, and the yeast subsequently mated against 5,000 strains, each missing one gene. In less than one week, 75,000 images were generated, each with an average of 300 yeasts per image. One terabyte of storage was required.
“High-content screening for me has become really attractive because you can now use automated image analysis to interpret the large data sets generated,” he said.
His group employed multidimensional clustering to sort the images into groups, and revealed the power of remining the data in different ways, at different times, to find clusters of genes involved in specific functions or associated with certain phenotypes—for example, the identification of genes associated with certain morphological characteristics, or of genes involved in lipid biosynthesis.
At the “SBHD” conference, Dinah Loerke, Ph.D., post-doctoral researcher at the Scripps Research Institute, discussed her work on endocytosis, hypothesizing that this process links mechanical changes with intracellular signaling via actin dynamics. A significant challenge in studying endocytosis is that most imaging studies focus on individually selected clathrin-coated pits, which are not representative of the entire population.
“Our approach is not to handpick individual clathrin-coated pits or individual endocytic events, but to develop an assay that allows us to characterize the process on the basis of the entire population, because endocytosis and actin dynamics are heterogeneous processes,” she reported.
By using a recently developed global optimization tracker to visualize 30,000–50,000 coated-pit trajectories, Dr. Loerke was able to characterize the process at the level of the entire population, using statistical analysis and quantitative methods to perform pit-lifetime analysis, and linking endocytosis to actin dynamics.
Since tumor cells are one example of when mechanical properties become profoundly altered upon malignant transformation, the findings promise broad applicability in cancer research.
These are just a few areas that image analysis has revolutionized. It is essential to recognize imaging as an important catalyst that shapes new concepts and provides an unbiased perspective both at the population and single-cell level. In particular, the powerful insight into single-cell events comes just years after Stephen Jay Gould pointed out the dangers associated with ignoring variation and focusing on measures of central tendency, as important sources of erroneous interpretations.
As Gould emphasized, in order to understand the true complexity of biological phenomena, it is essential to focus on variations within entire systems and take into consideration, just as the title of his book so eloquently suggests, the Full House.
© 2016 Genetic Engineering & Biotechnology News, All Rights Reserved