NEW YORK (GenomeWeb) – Seeking a stably expressed reference gene to normalize real-time reverse transcriptase PCR experiments, researchers in Austria have determined that the existence of a universal reference gene for all tissues and cell types is very unlikely.
In research published last month in PLoS One, the group used a 3D in vitro culture of human endothelial blood-brain barrier-derived cells as a model system.
"At the beginning of this work we were not looking at reference genes ... We wanted to run quantitative RT-PCR experiments in blood-brain barrier cells that were subjected to inflammatory stimuli," corresponding author Idriss Bennani-Baiti told GenomeWeb in an interview this week, adding, "The experiment sounds simple enough, right?"
However, the researchers were unable to find other published studies describing reference genes for human blood-brain barrier cells undergoing inflammation.
The team — which consisted of scientists at the Medical University of Vienna, The University of Vienna, the Vienna General Hospital, and an association of researchers called the B2 Scientific Group — assayed nine commonly used genes. But they found that no single reference gene was stably expressed enough during their experimental condition: addition of tumor necrosis factor alpha (TNF-α).
The group ranked housekeeping genes for expression stability under TNF-α stimulation using three statistical algorithms: BestKeeper, GeNorm, and NormFinder.
"Some of the most-utilized genes — such as GAPDH and beta-2-microglobulin, which almost everyone uses — were actually some of the least stably expressed," Bennani-Baiti said. Interestingly, these two genes were among the few noted to not have seasonal variation in transcription in a recent study covered by GenomeWeb.
The Austrian group showed that variability in stability of the presumed reference genes in their system was correlated with the number of binding sites for a transcription factor known to be induced by TNF-α called nuclear factor kappa-light-chain-enhancer of activated B cells, best known as NF-κB.
To Bennani-Baiti's knowledge, there hasn't been much published research examining regulatory sites in housekeeping genes to understand when and why they might not be stable.
Meanwhile, an author of the Minimum Information for Publication of Quantitative Real-Time PCR Experiments, or MIQE, guidelines told GenomeWeb last year that choice of reference genes may be partly responsible for the the high variability of reverse transcription efficiency. And a survey in 2013 found that many researchers relied on a single reference gene that had not been validated using the specific experimental conditions in their lab.
The issue is so problematic, one study recently reported by GenomeWeb instead developed a way of normalizing efficiency-corrected cycle threshold data to input sample quantity after encountering difficulties finding proper reference genes for clinical cases of stroke.
Bennani-Baiti said he found two metanalyses from disparate fields which suggested only a small percentage of published literature used a validated reference gene or geneset in their expression quantification studies.
"This is really scary, and gives you an idea of the number of publications out there where you really don't know if what they've found is actually valid or not," he said.
The use of non-validated housekeeping genes as reference genes might also be leading to the trend of irreproducible studies.
In the PLoS One study, Bennani-Baiti also used a multi-parameter analysis of published literature looking at gene expression via RT-qPCR and microarray, in about 5,300 different tissue samples, virtually probing more than 300 cell and disease types for universal reference genes.
"What we found out is that there is no single gene, in fact no single gene set, that can be used consistently as a universal reference," he said.
He noted that a recently published study claimed there were 13 such genes, but the Austrian group tested them and they were not stably expressed in their system. He also said analysis of other published literature shows them to be significantly changed in other experiments, including being modulated in response to "at least 100 different drugs," he said.
He noted that the most stable gene we the team found was variable with a p value of 10-8.
The researchers ultimately wished to use their inflammation model to examine expression of a drug transporter-encoding gene called RLIP76, which has been noted to be an important element of the stress defense mechanism adopted by cells. This transporter is important to the blood brain barrier and implicated in diseases such as epilepsy, Alzheimer's, and Parkinson's.
RLIP76 appears to be only moderately induced by inflammation, so "depending on which housekeeping gene we use as a reference, we find it either to be induced some five-fold or actually repressed more than three-fold," Bennani-Baiti said, adding, "You can bet your favorite Pipetman that actually makes a difference to a cell."
For many genes, such as those involved in basal transcriptional machinery or chromosome structure and dynamics, differences in expression levels as small as 20 percent can have "profound effects" on a cell, Bennani-Baiti said.
"Documenting such minimal changes requires really rock-solid reference genes," he said, adding, "The odds of hitting one by chance just because one happens to have primers in the freezer are probably in the range of 1:100,000, accounting for all the genes and alternatively spliced isoforms."
"I think most scientists are rational enough not to play the lottery, so why should they play it with their RT-qPCR experiments?"
The information in this paper could now be used as a starting point for clinical studies of inflammation, but Bennani-Baiti said researchers would have to revalidate the results in their particular study conditions.
Bioinformatic analysis using Transfac to find transcription factor binding sites might help in a simple model, but in real life it might be more complicated to pick candidate reference genes empirically.
Bennani-Baiti said the group will now use the gene set to run future RT-qPCR assays in its model system, and can be certain that the findings are not an artifact of the data normalization method.
In the case of RLIP76, which is a substrate for epilepsy drugs, data now suggests that inflammation can reduce anti-epileptic drug efficacy, and the investigators are now using their model system of human tissue to examine whether non-steroidal anti-inflammatory drugs could improve epilepsy treatment.
The group is also developing a searchable web-based database to be called The Reference Gene Database to collect and document potential reference genes in multiple cells, tissues, disease, and pathophysiological conditions, Bennani-Baiti said.