Scientists from the Coriell Institute for Medical Research, Icoria, and Invitrogen have published research demonstrating how combining high-throughput and high-content imaging of tissue samples with gene-expression data can provide a more comprehensive picture of drug toxicity.
The method is based on the concept of tissomics, which uses automated microscopy and machine vision to quantify phenotypic changes in tissue slices, much like the burgeoning field of high-content cellular imaging. Therefore, the work may open the door for high-content imaging platforms to more clinically relevant applications, including biomarker identification and pharmacogenomics.
Andres Kriete, an associate professor for bioinformation engineering and head of a joint bioinformatics initiative between the Coriell Institute and Drexel University, was first author on the study, which appears in the May 5 online version of Cytometry Part A.
Kriete has had a long-standing collaborative relationship with the company formerly known as TissueInformatics, which was acquired by Paradigm Genetics in 2004. Later that year, Paradigm changed its name to Icoria, and early this year, Icoria was acquired by and became a subsidiary of Clinical Data.
Kriete this week told CBA News that the latest research is an extension of the type of work being done at Icoria. Part of the motivation for this research originally, he explained, was to develop an alternative to human analysis of tissue samples, which can often be subjective.
"What is the value of taking gene-expression data from these different tissues, pooling them, and saying, 'This is where the difference is'? You're just muddying the water — you can't differentiate, and you lose sensitivity."
"We see it as an alternative," Kriete said. "I'm not saying that pathologists are doing a bad job here, but we find different tissue features or detect things that maybe pathologists have a hard time finding ways to quantify. They may see certain structures, modifications, or lesions, but we can do it reproducibly, and that's the key differentiator.
"Whenever we run this on a machine, we always get the same results, whereas with a pathologist, there are a lot of things that go into that type of diagnosis," he added. "That is basically a text-based quantification versus our approach, which is very quantitative, and can generate numbers that you can compare."
Automated microscopy for clinical tissue sample analysis is now old news, as several companies, such as Applied Imaging, ChromaVision, TriPath, Ventana Medical Systems, and HistoRx have developed such platforms.
But what hasn't been done to a large extent is coupling data generated by these imaging systems with gene-expression analysis.
"People use very different techniques to say something about what's going on in a tissue," Kriete said. "Visualization and quantification of structures, which we do, is one way; and of course people are interested in gene-expression analysis, proteomics, and the like."
Gene-expression analysis, in particular, has become a very important tool in early-stage drug toxicity testing, or so-called toxicogenomics.
"The problem with that is if you take just a gene-centric approach, you have a hard time dealing with heterogeneity, because every animal reacts a little bit differently," Kriete said. "The same thing is true with many diseases — they are all in different states. What is the value of taking gene-expression data from these different tissues, pooling them, and saying, 'This is where the difference is'? You're just muddying the water — you can't differentiate, and you lose sensitivity."
In the Cytometry A paper, Kriete and colleagues demonstrated how they could increase sensitivity by analyzing gene-expression data from the liver tissue of rats treated with a positive hepatotoxin, negative hepatotoxin, or control compound, and then correlating it with data obtained from automated imaging of treated liver tissue sections.
Icoria's Hepat software coupled with a homemade automated microscope enabled a total of 30 cellular and tissue features including count, average size, density, and nearest neighbor calculations for hepatocytes, non-hepatocytes, vacuoles, and sinusoids. Tissue slides were also assessed by a trained veterinary pathologist.
For gene-expression analysis, the researchers used a GE Healthcare Codelink Uniset Rat I bioarray containing 10,000 oligo probes. The researchers used VectorXpression software from Invitrogen's Informax division to compare gene-expression data, clinical chemistry measurements, and tissue histomorphometric measurements.
The researchers found that "significant, consistent histological differences were not observed by pathologist reading between the three groups of animals examined," the paper states. In addition, they uncovered several genes that correlated to hepatotoxic phenotypes, but had not heretofore been associated with hepatotoxicity in scientific literature.
Scientists using high-content cellular imaging for functional genomics studies are already aware of how valuable multiplexed phenotypic data can be in determining gene function in living cells, especially when correlated to gene-expression data. Analysis of tissue phenotypes, or tissomics, is just one level higher and closer to the clinic, Kriete said.
"The advantage of doing it on a tissue level is that industry is doing this, to a large extent, in drug-target development, so they have to mix in experimentation or treatments with a new drug on animals," Kriete said. "So it's the next step in the development pipeline of a drug. Also, it's closer to pathology. Pathologists in the clinic don't use gene-expression analysis so much now, but they do look at the tissues. It's the next step in the biological hierarchy, but you could do a very similar thing on the cellular level."
According to the researchers, the method could also have application in other areas such as biomarker identification and, eventually, pharmacogenomics.
"Tissues are central in the biological hierarchy and anchor both molecular processes as well as physiological activity, and correlations can be found in either direction to identify biomarkers as well as improve classifications," the researchers wrote in the Cytometry A paper.
"A tissomics-based approach as demonstrated here, if combined with advanced biostatistical data-mining tools, provides a model for a broader range of biomedical applications, but in particular, for the development of a personalized medicine," the researchers wrote.
In addition, the technique may find application in other basic biological research. Kriete himself is applying the method to take a systems approach to understanding the biology of aging.
"Biological age is very different from chronological age," Kriete said. "We want to use gene-expression analysis to find changes in tissues as they age, but let's say you study all 80-year-olds — they show a lot of variability in their gene-expression profiles, because some people are a little bit younger, biologically; and some are a little bit older.
"So you don't have a high sensitivity if you just pool the data," he added. "We take tissues from the same individuals, use this technique as a net of sorts over the gene-expression data, and then we have a much higher sensitivity, and find genes that we normally wouldn't find."
— Ben Butkus ([email protected])