When it comes to understanding schizophrenia or manic depression, researchers have lots of ideas but little data, Sabine Bahn, a clinical psychiatrist who heads a laboratory at the Babraham Institute in Cambridge, UK, told BioArray News. “There are many, many hypotheses, almost as many hypotheses as there are professors.”
That was one of the main reasons, she said, to embark on a microarray study of 150 human brains — the largest such study published thus far, she said. The results of the study, “Application and optimization of microarray technologies for human postmortem brain studies,” appeared in Biological Psychiatry last month.
Bahn and her team analyzed 150 brain samples — from schizophrenia patients, bipolar disorder patients, and controls — using 220 Affymetrix U133A microarrays.
“We wanted to collate empirical evidence first, and hopefully, in due course, we will find a hypothesis which is not based on intuition but on biological data,” she said. Unlike DNA-level studies, like SNP analyses, such a study would address not only genetic but also epigenetic factors that contribute to the diseases.
What she and her colleagues discovered along the way is a number of pitfalls when using microarrays on human postmortem tissues, prompting them to develop a series of quality control steps and criteria that indicate a good quality sample. “We learned it the hard way; we wasted about 50 chips,” Bahn said.
They kept details of every step of the analysis, including when a sample was run, when the washing buffer was changed, and when a new kit was started.
Many of these steps relate to the RNA quality, since RNA can often be degraded by the time the brain is removed — up to 48 hours after death — Bahn said. She and her team found that a single such control step, which includes, for example, an optimized RNA extraction protocol, is not enough to determine a good quality chip, but that “it’s the summary of those that needs to be looked at.”
Subsequently, they investigated the variables that may determine a good quality chip and wrote a meta-analysis script for variables such as noise, average background, number of genes detected, 3’:5’ ratios, and internal standards. They found that failed chips clustered together. The script is available to researchers on request.
Now, “I think we have a better idea about which samples are worth running and which ones we should discard right from the beginning,” she said. Filtering out bad samples early on helps both save resources and get reliable results. “A few bad chips, we found out, can skew your data dramatically,” Bahn said. Also, she would not have repeated runs with certain samples had she known earlier that they were of low quality.
Other than RNA quality, what hampers microarray brain research is the lack of sensitivity, Bahn said. Her group has now started using laser capture microdissection to study individual cell populations in the cortex and to be able “to compare apples with apples.” Different platforms may vary in their sensitivity, and Bahn said she is about to switch to Amersham’s CodeLink microarray platform since its sensitivity is allegedly “superior to the Affymetrix system,” she said.
The other technical limitation of microarrays in her field, Bahn said, is the ability to study splice variants, and to analyze small expression differences. “Small-fold differences … may be biologically more important than a high-fold difference,” Bahn said.
The complex data analysis, she said, requires neuroscientists to collaborate with bioinformatics experts, and she now has two bioinformaticists in her own group. Microarray studies without appropriate bioinformatics input “create a bad reputation for our field because people can’t reproduce it,” she said.
In the end, she said, other platforms like proteomics and metabolomics — though often not exhaustive — go hand in hand with microarrays to “shed a different light onto the problem.”