Scientists can easily track mRNAs from cradle to grave in cultured cells, but monitoring them in tissues presents a challenge. Now, a group led by Hamed Najafabadi at McGill University in Montreal describe a new method claimed to yield more accurate results. As reported October 13 in Nature Communications, the authors fix a bias in a previous algorithm to estimate RNA decay rates in situ. Applying it to brain tissue, the researchers calculated different stabilities of RNAs. They also identified two RNA-binding protein families, RBFOX and ZFP36, and four microRNAs, miR-124, miR-29, miR-9, and miR-128, that dominate regulation of RNA stability. RBFOX1, which binds mRNAs encoding synaptic proteins, drops in Alzheimer’s brains as synaptic RNA decay accelerates.
- New method estimates mRNA decay more accurately.
- Two RNA-binding proteins, four microRNAs control brain mRNA stability.
- In AD, RBFOX1 expression drops and synaptic mRNA decays faster.
Some researchers cautioned that AD sample tissue was limited and that long postmortem intervals compromise some of the RNA data sets, but others welcomed the potential of this approach. “The study adds to the realization that post-transcriptional regulation is of high importance in the brain,” wrote Evgenia Salta at VIB/KU Leuven, Belgium (see full comment below). “This novel approach increases the depth of information that can be extracted from [omics] data sets.”
For years, researchers have used microarrays or RNA sequencing (RNA-Seq) to quantify gene expression, but these offer only a snapshot of steady-state mRNA levels. Reasoning that RNAs still containing intron sequences reflect levels of pre-mRNAs, which are quickly spliced into intron-less mRNAs, scientists realized they could calculate relative mRNA decay rates by subtracting changes in the synthesis rate (introns) from changes in steady-state levels (exons) (Gaidatzis et al., 2015). “It’s a clever method,” said Najafabadi. However, when his group used it to compare mRNA decay rates across different tissues, they ran into problems. “We were seeing results we couldn’t explain,” he told Alzforum. “We saw mRNAs for brain-specific pathways that were less stable in the brain than in non-brain tissues. It didn’t make sense.” The researchers figured out that introns can linger in cells when transcription rates are high and the splicing machinery cannot keep up. Using intron levels as the sole indicator of synthesis leads to an overestimation of mRNA decay for RNAs that are quickly transcribed, while underestimating slowly transcribed RNAs.
RNA Fireworks. More (red) and less (blue) stable mRNAs in the brain have binding sites for two proteins and four microRNAs that stabilize (square) or destabilize (diamond) the nucleic acids. [Courtesy of Alkallas et al., Nature Communications, 2017.]
To correct for this, Najafabadi and first author Rached Alkallas created an algorithm that takes into account the kinetics of RNA processing. To test the new technique, they used RNA-Seq data sets from two breast carcinoma cell lines (Furlow et al., 2015; Minn et al., 2005). The researchers chose these data sets because previous studies using a direct measure of RNA stability that only worked in cultured cells had revealed dozens of mRNAs whose stability differed between the two lines. Najafabadi’s algorithm yielded results that were consistent with the direct measure. Importantly, it roundly outperformed the simple estimates based on intron versus exon levels, which, in some cases, suggested decay rates were faster in one cell line than in the other, when it was the other way around.
The researchers then applied their new tool to obtain one of the first broad surveys of mRNA stability across human tissue. They analyzed RNA-Seq profiles from samples of 20 different tissues (Duff et al., 2015). They found stark, tissue-specific differences in mRNA stability, and that these differences were conserved between humans and mice.
Najafabadi and colleagues then turned their attention to the brain because analysis of its RNA decay profile suggested many processes are regulated differently in this tissue than in others. To get a handle on which factors might be shaping brain RNA stability profiles, the authors scanned the 3' UTRs of brain mRNAs for RNA-binding protein (RBP) and miRNA binding sites. They then correlated those sequences with the stability estimates for brain mRNAs. Strikingly, sites for just the two RBP families—RBFOX and ZFP36—and the four miRNAs—miR-124, miR-29, miR-9, and miR-128—surfaced as the main regulators. One of the RNA-binding proteins, RBFOX, is highly expressed in brain and stabilizes mRNAs. Counterintuitively, the other mRNA stabilizing protein, ZFP36, is essentially absent in brain. It turns out ZFP36 destabilizes neural-specific mRNAs in nonneural tissues.
All of the miRNAs destabilize their target transcripts and, interestingly, all have been implicated in neurodegenerative disorders. For example, researchers have reported miR-128 dysregulation in AD (Lukiw, 2007), miR-29a levels correlated with AD progression (Apr 2017 conference news), and reductions in miR-9 and miR-124 have been found in the brains of people with frontotemporal dementia (Nov 2014 conference news).
Integrating data from more than 2,000 transcripts, the authors then built a network that linked these six key factors to their mRNA targets (see image above). Is this network meaningful? The authors turned to previous experimental data, reviewing crosslinking studies that mapped protein-binding sites in the mRNAs of living cells, and studies on the effects of miRNAs on specific mRNAs. They found that their network better predicted protein-RNA and miRNA binding than did simple examination of sequence motifs. For example, their networks had a 3.6-fold better chance of identifying mRNA targets that were degraded by miR-124 than simply searching for mRNAs with miR-124 binding sites.
Noticing that the RBFOX component of the network was most highly enriched for genes involved in synaptic transmission, Alkallas and Najafabadi wondered if it might be disrupted in AD. The researchers analyzed RNA-Seq data from postmortem prefrontal cortices of six patients with advanced AD, as judged by a Clinical Dementia Rating score of between four and five, with that from five age- and gender-matched controls (Scheckel et al., 2016). Surveying nearly 8,500 mRNAs, they found a generalized destabilization of brain-specific transcripts in the AD brains. The predicted targets of RBFOX took the biggest hit—a drop of about 25 percent on average. With 37 mRNAs, synaptic transmission emerged as the pathway most highly represented among the top 500 destabilized transcripts.
The researchers also found that expression of the RBFOX family member RBFOX1 was lower in brain tissue of AD data sets than in controls. This supports previous transcriptome microarray data from 310 AD patients with moderate to severe AD, which showed a halving of RBFOX1 mRNA (Narayanan et al., 2014). Alkallas found the lower RBFOX1 expression levels in AD samples correlated with a dearth of synaptic mRNAs predicted to be RBFOX1 targets. When the researchers knocked down RBFOX1 in primary human neural progenitor cells, they observed an mRNA stability signature similar to that of AD brains.
Benjamin Wolozin at Boston University, who found that RNA decays faster in a mouse model of tauopathy, welcomed the findings, but worried about the limitations of computational studies. Although he complimented the authors’ correction of a bias in the original exon-intron algorithm, he feared there might be other important biases for which they have still not accounted. “There are lots of potential biases that are simply unknown, particularly when looking at disease states,” he said. Salta noted that further experimental validation is needed. “Widespread alterations in the machinery involved in RNA metabolism that perturb, for instance, RNA splicing or intron stability can occur in neurodegenerative disorders, and this would affect the reliability of the algorithm,” she wrote. Najafabadi acknowledged the limitations, but emphasized that this is a start and the algorithm can be refined as new information emerges.
Walter Lukiw, Louisiana State University, New Orleans, pointed out that the survey of human tissues included total brain RNA from just one person and cerebellar RNA from 10 pooled samples. The cerebellum is relatively spared in AD. He also noted that the RNA-Seq data sets of AD brains included samples with up to 18-hour postmortem intervals. “After about three hours, accurate sampling of the total mRNA of brain tissue is not easily obtainable,” he wrote. Najafabadi acknowledged the problem. “That’s absolutely a concern,” he said. “But given that we are comparing postmortem AD samples with postmortem control samples, we predict that biases introduced by postmortem decay cancel each other out.”
Najafabadi also noted that it remains challenging to separate the contributions of different cell types, and to estimate how neuronal loss affects RNA decay. He would like to collaborate with labs doing single-cell RNA-Seq and test computational methods to tease apart neuron-specific signatures. Studies like these might clarify open questions such as what causes the generalized RNA decay seen in AD brains, and what roles the four miRNAs play.
The when, where, and how of RBFOX dysregulation in Alzheimer’s intrigued researchers. Salta wants to know when and where in the brain RBFOX levels drop. Najafabadi wants to test the effect of overexpressing RBFOX in mouse models of AD. He also wants to examine other neurologic disorders, noting a recent link between RBFOX and autism (Lee et al., 2016). RBFOX1 alternatively splices genes involved in neural development (Zhou et al., 2007) and may play a role in the alternative splicing of Aβ precursor protein (APP), which could affect Aβ aggregation (Alam et al., 2014).—Marina Chicurel
- Look in the MiR: MicroRNA Fans Neurogenesis in Old Alzheimer’s Mice
- Do MicroRNAs Cause Mayhem Across Frontotemporal Dementia Spectrum?
- Gaidatzis D, Burger L, Florescu M, Stadler MB. Analysis of intronic and exonic reads in RNA-seq data characterizes transcriptional and post-transcriptional regulation. Nat Biotechnol. 2015 Jul;33(7):722-9. Epub 2015 Jun 22 PubMed.
- Furlow PW, Zhang S, Soong TD, Halberg N, Goodarzi H, Mangrum C, Wu YG, Elemento O, Tavazoie SF. Mechanosensitive pannexin-1 channels mediate microvascular metastatic cell survival. Nat Cell Biol. 2015 Jul;17(7):943-52. Epub 2015 Jun 22 PubMed.
- Minn AJ, Gupta GP, Siegel PM, Bos PD, Shu W, Giri DD, Viale A, Olshen AB, Gerald WL, Massagué J. Genes that mediate breast cancer metastasis to lung. Nature. 2005 Jul 28;436(7050):518-24. PubMed.
- Duff MO, Olson S, Wei X, Garrett SC, Osman A, Bolisetty M, Plocik A, Celniker SE, Graveley BR. Genome-wide identification of zero nucleotide recursive splicing in Drosophila. Nature. 2015 May 21;521(7552):376-9. Epub 2015 May 13 PubMed.
- Lukiw WJ. Micro-RNA speciation in fetal, adult and Alzheimer's disease hippocampus. Neuroreport. 2007 Feb 12;18(3):297-300. PubMed.
- Scheckel C, Drapeau E, Frias MA, Park CY, Fak J, Zucker-Scharff I, Kou Y, Haroutunian V, Ma'ayan A, Buxbaum JD, Darnell RB. Regulatory consequences of neuronal ELAV-like protein binding to coding and non-coding RNAs in human brain. Elife. 2016 Feb 19;5 PubMed.
- Narayanan M, Huynh JL, Wang K, Yang X, Yoo S, McElwee J, Zhang B, Zhang C, Lamb JR, Xie T, Suver C, Molony C, Melquist S, Johnson AD, Fan G, Stone DJ, Schadt EE, Casaccia P, Emilsson V, Zhu J. Common dysregulation network in the human prefrontal cortex underlies two neurodegenerative diseases. Mol Syst Biol. 2014 Jul 30;10:743. PubMed.
- Lee JA, Damianov A, Lin CH, Fontes M, Parikshak NN, Anderson ES, Geschwind DH, Black DL, Martin KC. Cytoplasmic Rbfox1 Regulates the Expression of Synaptic and Autism-Related Genes. Neuron. 2016 Jan 6;89(1):113-28. Epub 2015 Dec 10 PubMed.
- Alam S, Suzuki H, Tsukahara T. Alternative splicing regulation of APP exon 7 by RBFox proteins. Neurochem Int. 2014 Dec;78:7-17. Epub 2014 Aug 11 PubMed.
No Available Further Reading
- Alkallas R, Fish L, Goodarzi H, Najafabadi HS. Inference of RNA decay rate from transcriptional profiling highlights the regulatory programs of Alzheimer's disease. Nat Commun. 2017 Oct 13;8(1):909. PubMed.