CMS121 and J147 – Two drug that improve memory and slow neurodegeneration associated with Alzheimer’s disease


In mouse models of Alzheimer’s disease, the investigational drug candidates known as CMS121 and J147 improve memory and slow the degeneration of brain cells.

Now, Salk researchers have shown how these compounds can also slow aging in healthy older mice, blocking the damage to brain cells that normally occurs during aging and restoring the levels of specific molecules to those seen in younger brains.

The research, published last month in the journal eLife, suggests that the drug candidates may be useful for treating a broader array of conditions and points out a new pathway that links normal aging to Alzheimer’s disease.

“This study further validated these two compounds not only as Alzheimer’s drug candidates but also as potentially more widely useful for their anti-aging effects,” says Pamela Maher, a senior staff scientist at Salk and a co-corresponding author of the new paper.

Old age is the biggest risk factor for Alzheimer’s disease–above the age of 65, a person’s risk of developing the disease doubles about every five years. However, at a molecular level, scientists aren’t sure what occurs in the brain with aging that contributes to Alzheimer’s.

“The contribution of old age-associated detrimental processes to the disease has been largely neglected in Alzheimer’s disease drug discovery,” says Antonio Currais, a Salk staff scientist and first author of the new paper.

Maher and David Schubert, the head of Salk’s Cellular Neurobiology Lab, previously developed CMS121 and J147, variants of plant compounds with medicinal properties.

Both compounds tested positive for their ability to keep neurons alive when exposed to cellular forms of stress related to aging and Alzheimer’s disease.

Since then, the researchers have used the drug candidates to treat Alzheimer’s in animal models of the disease.

But experiments revealing exactly how the compounds work suggested that they were targeting molecular pathways also known to be important in longevity and aging.

In the new research, Maher, Currais and their colleagues turned to a strain of mice that ages unusually fast.

A subset of these mice was given CMS121 or J147 beginning at nine months old–the equivalent of late middle age in humans. After four months, the team tested the memory and behavior of the animals and analyzed genetic and molecular markers in their brains.

Not only did the animals given either of the drug candidates perform better on memory tests than mice that hadn’t received any treatment, but their brains showed differences at the cellular and molecular levels. In particular, expression of genes associated with the cell’s energy-generating structures called mitochondria was preserved by CMS121 and J147 with aging.

Old age is the biggest risk factor for Alzheimer’s disease–above the age of 65, a person’s risk of developing the disease doubles about every five years.

“The bottom line was that these two compounds prevent molecular changes that are associated with aging,” says Maher.

More detailed experiments showed that both drugs affected mitochondria by increasing levels of the chemical acetyl-coenzyme A (acetyl-coA).

In isolated brain cells, when the researchers blocked an enzyme that normally breaks down acetyl-CoA, or when they added extra amounts of an acetyl-coA precursor, they saw the same beneficial effect on mitochondria and energy generation.

The brain cells became protected against the normal molecular changes associated with aging.

“There was already some data from human studies that the function of mitochondria is negatively impacted in aging and that it’s worse in the context of Alzheimer’s,” says Maher. “This helps solidify that link.”

Maher and Currais are planning future experiments to test the effects of CMS121 and J147 on how other organs age.

They also hope to use the new results to inform the development of new Alzheimer’s drugs; targeting other molecules in the acetyl-coA pathway may help treat the disease, they hypothesize.

“We are now using a variety of animal models to investigate how this neuroprotective pathway regulates specific molecular aspects of mitochondrial biology, and their effects on aging and Alzheimer’s,” says Currais.

Other researchers on the study were Ling Huang, Joshua Goldberg, Gamze Ates, António Pinto-Duarte, Maxim Shokhirev and David Schubert of the Salk Institute, and Michael Petrascheck of The Scripps Research Institute.

Funding: The work was supported by grants from the National Institutes of Health, the Glenn Foundation for Medical Research, the Shiley Foundation and the Edward N. and Della L. Thome Memorial Foundation.

David Schubert is an unpaid advisor for Abrexa Pharmaceuticals, a company working on the development of J147 for Alzheimer’s therapy. The Salk Institute holds the patents for CMS121 and J147.

Age is the greatest risk factor for Alzheimer’s disease (AD) and related dementias. While genetic risk factors have been the major focus of AD drug discovery, it is important to consider the progressively detrimental metabolic processes that take place with normal aging as possible therapeutic targets.

Clinical evidence shows that a decline in cerebral glucose metabolism (hypometabolism) precedes the pathology and symptoms of AD and is more severe than that observed in normal aging (Costantini et al., 2008Cunnane et al., 2011Yin et al., 2014).

Given that energy production from glucose supports the majority of brain activity as well as the maintenance of cellular homeostasis, it is likely that a failure to supply cells with adequate energy contributes to the neuropathological cascade in both aging and age-associated dementias such as AD (Caldwell et al., 2015).

Most of the energy derived from glucose oxidation is produced in mitochondria. Not surprisingly, a number of mitochondrial-dependent functions are found to be impaired during aging and AD (Caldwell et al., 2015Chan, 2006Currais, 2015aLin and Beal, 2006Onyango et al., 2016Swerdlow and Khan, 2004Yin et al., 2016). In fact, brain mitochondrial dysfunction has been hypothesized to be responsible for the pathological hallmarks of AD (Swerdlow and Khan, 2004Yin et al., 2016).

Because the current approach to AD drug discovery has largely failed, we devised a novel drug discovery paradigm based on phenotypic screening assays that mimic numerous aspects of old age-associated neurodegeneration and brain pathology, including energy failure and mitochondrial dysfunction (Prior et al., 2014). We have identified two compounds that are very neuroprotective – CMS121 and J147. J147 is active in transgenic AD animal models (Chen et al., 2011Daugherty et al., 2017Prior et al., 2013Prior et al., 2016). It enhances memory and prevents some aspects of aging in rapidly aging (15 month median lifespan) senescence-accelerated prone 8 (SAMP8) mice when administered early in life (Currais et al., 2015b). CMS121 is a more potent derivative of the flavonol fisetin that maintains most of its biological properties (Chiruta et al., 2012). CMS121 is not as well studied as J147, but we have recently shown that fisetin is able to ameliorate some aspects of aging in SAMP8 mice (Currais et al., 2018). The target of J147 is the alpha subunit of ATP synthase, which engages a neuroprotective response involving the activation of AMP-activated protein kinase (AMPK) (Goldberg et al., 2018). ATP synthase is an anti-aging target in C. elegans (Chin et al., 2014). We have ruled out that CMS121 also targets ATP synthase (unpublished data), and its molecular targets are currently under investigation. However, because the two compounds were developed based upon brain toxicities associated with aging and therefore share similar biological activities in vitro, we hypothesized that they may mitigate some aspects of aging brain metabolism and pathology via a common pathway despite differences in molecular structure and direct targets.

To test this idea, we fed CMS121 and J147 to aged SAMP8 mice and used a multiomics approach to identify modes of action. We first show that both compounds reduce metabolic and gene transcription markers of aging in the SAMP8 model of aging and dementia when administered at a late stage of the aging process. We further demonstrate that both compounds share a mechanism of action that maintains high levels of acetyl-coenzyme A (acetyl-CoA), at least in part, by the inhibition of acetyl-CoA carboxylase 1 (ACC1). Importantly, the compounds increase histone acetylation in cultured neurons and SAMP8 mice at a site on histone H3 that is required for memory formation (Mews et al., 2017). Together, these data show that aging and dementia share a common metabolic pathway related to brain mitochondrial function that can be therapeutically targeted.


Despite the observation that old age is the greatest risk factor for AD, few clinical approaches have focused on treating AD and other dementias by targeting detrimental age-related processes that are relevant to disease onset and development. In part, this is because we still lack a deep understanding of the aging brain. As such, our laboratory has developed a novel drug discovery program based upon cell-based screening assays that mimic multiple aspects of old age-associated neurodegeneration and AD pathology (Prior et al., 2014) and has used a rapidly aging mouse model to study drug candidate effects in the context of aging (Currais et al., 2015b). This approach has led to the identification of CMS121 and J147, two potent neuroprotective small molecules. Because these compounds have different structures, but are effective in the same in vitro models of toxicities associated with the aging brain, we asked if they could be used to identify metabolic changes associated with age-related cognitive dysfunction, leading to the identification of novel therapeutic targets.

CMS121 and J147 were able to reduce age-related cognitive dysfunction, even when administered to SAMP8 mice at a late stage of their lives. In addition, CMS121 and J147 share a common mechanism of action that is associated with the maintenance of mitochondrial health in terms of transcriptional stability and metabolism. Mitochondrial dysfunction is one of the hallmarks of aging (López-Otín et al., 2013), and an acceleration in this dysfunction may be responsible for the onset of AD pathology (Swerdlow and Khan, 2004Yin et al., 2016). Therefore, mitochondria represent an important therapeutic target for the treatment of AD (Caldwell et al., 2015Onyango et al., 2016).

Consistent with these observations, CMS121 and J147 reduced several molecular parameters of brain aging. PCA and drift analysis of the transcriptome showed that the two compounds significantly prevented changes in gene expression associated with aging. The level of suppression by CMS121 was particularly remarkable as evidenced in the correlation plots. Although not as strong, J147 also significantly prevented transcriptomic aging, which is in accordance with what we found in the SAMP8 mice treated from a young age (3 months old) with this drug candidate (Currais et al., 2015b).

The drift suppression was not as strong at the level of the brain metabolites as it was with gene expression. Instead, numerous metabolites that did not change with age were directly altered by the compounds. The effects of CMS121 and J147 on certain metabolic pathways and individual metabolites were very specific, namely the increase in acyl carnitines. Since acyl carnitines are direct substrates for fatty acid oxidation in mitochondria and a source of acetyl-CoA, they further support the conclusion that the compounds have a direct effect on mitochondrial metabolism in the brain. CMS121 and J147 also suppressed the metabolic drift in the plasma, indicating that the compounds may prevent the aging process at the systemic level as well.

Pathway analysis of the transcriptome identified genes that code for proteins that are part of the ETC and ATP synthase as being preferentially maintained at the younger level by the compounds. Mitochondria-related genes were also the top pathways affected by aging, indicating that mitochondrial dysfunction is part of the SAMP8 aging process. Because a decrease in mitochondrial DNA copy number was found with aging, the increased expression of gene transcripts associated with the ETC with aging is likely a compensatory response to a lack of adequate mitochondrial metabolism. The conclusion is supported by the observation that deficits in the respiratory chain and mitochondrial bioenergetics increase as a function of age (Caldwell et al., 2015López-Otín et al., 2013), and by our data showing significant alterations in a number of TCA cycle metabolites in 13 months old SAMP8 mice.

Acetyl-CoA is a key metabolite that bridges glycolysis, fatty acid β-oxidation, fatty acid synthesis and the TCA cycle in the mitochondria. Its metabolism in the TCA cycle leads to the production of metabolites that are used in other metabolic pathways essential for the proper functioning of the cell, including the neurotransmitter acetylcholine which is reduced in forebrain neurons at early stages of the disease (Mufson et al., 2008). Acetyl-CoA is also used by histone acetyltransferases to acetylate histones and thus regulate memory formation (Mews et al., 2017). In addition, the TCA cycle generates reducing potential that is used by the ETC. Therefore, our data have substantial implications for the therapeutic use of CMS121 and J147 in AD, given that deficits in mitochondrial metabolism are thought to play a key role in the cognitive dysfunction of AD patients.

Although acetyl-CoA can be derived from multiple sources, our data argue that the elevation of acetyl-CoA by the compounds is due to the inhibition of ACC1 via its phosphorylation by AMPK. This conclusion is based on the following observations: 1) both compounds cause ACC1 phosphorylation in vivo, primary neurons, HT22 cells and fibroblasts, and increase acetyl-CoA levels in all four models; 2) knockdown of ACC1 increases acetyl-CoA levels and provides neuroprotection; 3) chemical inhibition of ACC1 also increases acetyl-CoA levels and provides neuroprotection; and 4) absence of AMPK prevents the inhibition of ACC1, increases in acetyl-CoA levels and reduces the protection against oxytosis.

One of the few clinical approaches targeting mitochondrial metabolism that has been tested in the context of AD is the use of ketogenic diets. In animals, ketone bodies are mainly produced in the liver from fatty acid β-oxidation through acetyl-CoA and are then transported via blood to other tissues, such as the brain, where they are once again metabolized to generate acetyl-CoA. Administration of a variety of ketogenic diets has been shown to have some positive effects in both AD transgenic mice as well as human AD patients (reviewed in Lange et al., 2017). Because CMS121 and J147 increase acetyl-CoA levels in the brains of animals and in vitro, our data suggest a therapeutic alternative to the ketogenic diet that leads to an improvement in mitochondrial health as well as an improvement in other parameters of aging.

Acetyl-CoA, along with several mitochondrial metabolites, has been directly implicated in the regulation of gene expression via the modification of chromatin (reviewed in Kinnaird et al., 2016). Histone acetylation requires acetyl-CoA that is synthesized in the cytoplasm and nucleus from acetate, citrate or pyruvate. These metabolites are intermediates of the TCA cycle and were all found to be altered with aging in SAMP8 mice. Other TCA metabolites, such as fumarate and succinate have been identified as inhibitors of Jumonji-C (JMJC) domain-containing histone demethylases (JHDMs) and TET demethylases, which rely on the TCA metabolite α-ketoglutarate as a co-substrate.

The levels of all these metabolites were also altered in the brains of 13 months old SAMP8 mice. CMS121 and J147 decreased the levels of succinate and α-ketoglutarate. Therefore, it is not unexpected that alterations in mitochondrial metabolism in the 13 months old SAMP8 mice would be followed by changes at the level of gene expression.

It should be noted that, due to limitations in the amount of material, the transcriptomic analysis was carried out with brain hippocampus and that the metabolomic study was performed with brain cortex. Despite being from two different brain regions, the interpretation of the global effects of the compounds on these parameters with aging still hold valid. This is because network analysis using GAM identified acetyl-CoA as a central metabolite regulated by CMS121 and J147 in both cortex and hippocampus, and this finding was validated in primary neurons and HT22 cells. In addition, both transcriptomics and metabolomics also identified mitochondrial dysfunction as a feature of aging in SAMP8 mice in both tissues.

The link between brain acetyl-CoA metabolism and behavioral improvement caused by CMS121 and J147 may be the increased acetylation of H3K9, known to regulate the expression of genes involved in memory formation (Mews et al., 2017). H3K9 acetylation was reduced in the brains of old SAMP8 mice as well as in the brains of APPswe/PSEN1dE9 transgenic AD mice. Both CMS121 and J147 increased acetylation at this site in the mouse brains as well as in primary cortical neurons. The ability of these compounds to prevent a large number of transcriptomic changes with aging might thus be a consequence of their direct effects on acetyl-CoA-mediated histone modifications.

The reason that the compounds share a common effect on metabolism is likely related to the fact that the same cell-based screening platform was used to identify both. In fact, some of these assays are characterized by mitochondrial dysfunction, such as the glutamate-directed oxytosis assay in HT22 cells (Prior et al., 2014). We show that, despite engaging distinct initial molecular targets in the cell, the compounds activate mechanisms that overlap at the level of mitochondrial metabolism, leading to the same phenotypic outcome.

In conclusion, when administered at advanced stages of aging and cognitive dysfunction, two structurally distinct AD drug candidates preserve brain mitochondrial homeostasis during this period of life, making mice appear younger at the level of cognition, transcription and metabolism. The analysis of molecular pathways based upon data overlapping the action of CMS121 and J147 and respective mechanistic validation identified a unique neuroprotective mechanism that maintains acetyl-CoA levels in cell culture and animal models of aging and AD (Figure 7). Because mitochondrial dysfunction is a hallmark of both old age and AD, it follows that other therapies that target acetyl-CoA in the brain may preserve mitochondrial homeostasis and prevent the metabolic deficits in AD patients that occur with aging.

An external file that holds a picture, illustration, etc.
Object name is elife-47866-fig7.jpg
Figure 7.
Summary diagram.CMS121 and J147 preserve brain mitochondrial homeostasis with aging via a neuroprotective mechanism that regulates acetyl-CoA metabolism.

Study design

Twenty-three 9 months old female SAMP8 mice were fed with vehicle diet (LabDiet 5015, TestDiet, Richmond, IN), and twenty-two and twenty-two 9 months old female SAMP8 mice were fed with CMS121 and J147 diets, respectively (LabDiet 5015 + 400ppm CMS121, LabDiet 5015 + 200ppm J147, TestDiet). Diet treatment lasted for four months until mice reached 13 months of age. At 9 months of age, SAMP8 mice already present a strong phenotype (Currais et al., 2015bCurrais et al., 2012Takeda, 2009).

The dose of J147 used was 200 ppm (~10 mg/kg/day), which previously proved effective in mouse models (Currais et al., 2015bPrior et al., 2013Prior et al., 2016). For CMS121 we used 400 ppm (~20 mg/kg/day), which was chosen based on its greater efficacy than its parent, fisetin, in in vitro assays and positive results with 500 ppm fisetin in the SAMP8 mice (Currais et al., 2018). Eleven 9 months old female SAMP8 mice were used as the baseline control group.

The effect of the compounds was assessed in older SAMP8 mice after the four months of treatment and any age-related changes are defined by comparison to 9 months old SAMP8 mice. All mice were randomly assigned to experimental groups. The number of mice per group was determined based on previous experiments (Currais et al., 2015bCurrais et al., 2012) and was sufficient to attain statistical power.

Six SAMP8 mice fed with control diet, four SAMP8 mice fed with CMS121 diet and four SAMP8 mice fed with J147 diet died throughout the course of this study. Behavioral testing was carried out one month prior to sacrifice and collection of biological material. Data were analyzed by blinded researchers when appropriate.

SAMP8 mice

The SAMP8 line, a naturally occurring mouse line that was developed based on its phenotype of accelerated aging, was acquired from Harlan Laboratories (U.K.). Mouse body weights were measured regularly and no significant differences were found between the groups (Figure 2—figure supplement 2). All experiments were performed in accordance with the US Public Health Service Guide for Care and Use of Laboratory Animals and protocols approved by the IACUC at the Salk Institute.

Tissue preparation

Mice were anesthetized and their blood collected by cardiac puncture. After perfusing with PBS, their brains were removed and dissected to collect cortex and hippocampus. Tissue was prepared for Western blotting, RNA extraction and metabolomic analysis.

Large-scale metabolome analysis

Metabolite measurement and analysis were conducted at Metabolon as previously described (Currais et al., 2015b).

Measurement of acetyl-CoA

Acetyl-CoA levels were determined in protein-free lysates of primary neurons or HT22 cells treated as indicated in the figure legends using a kit from Sigma (MAK039) according to the manufacturer’s instructions. The levels were normalized to the protein in the solubilized pellet using the BCA assay.

Supplemental materials and methods

Cell lines

Mouse hippocampal HT22 cells, generated in our laboratory, were propagated as previously described (Davis and Maher, 1994). AMPK knockout (K.O.) mouse embryonic fibroblasts were a generous gift from Ruben Shaw (Salk Institute). The lack of AMPK expression can be verified by Western blotting (Figure 6—figure supplement 2). To prevent cell misidentification, large batches of each cell line were frozen that are regularly thawed to avoid using the wrong cell line. Cell lines are routinely tested for mycoplasma and no contamination has been found.

Primary neurons

Primary cortical neurons were prepared from day 17 rat embryos and used at 7 days in vitro (7 DIV) (Soucek et al., 2003).

APPswe/PSEN1dE9 transgenic AD mice

Samples from line 85 APPswe/PSEN1dE9 transgenic AD mouse mice and respective wild type (Wt) mice treated with J147 in a reversal experimental paradigm were obtained from our previous study (Prior et al., 2013). In another independent study, 10 months old male Wt and APPswe/PSEN1dE9 mice were fed with vehicle diet (LabDiet 5015, TestDiet, Richmond, IN) or CMS121 diet (LabDiet 5015 + 400ppm CMS121) for three months and brain cortical tissue collected.

Behavioral assays

Elevated plus maze – The maze consisted of four arms (two open without walls and two enclosed by 15.25 cm high walls) 30 cm long and 5 cm wide in the shape of a plus. A video-tracking system (Noldus EthoVision) was used to automatically collect behavioral data. The software was installed on a PC computer with a digital video camera mounted overhead on the ceiling, which automatically detected and recorded when mice entered the open or closed arms of the maze and the time spent in each.

Mice were habituated to the room 24 hr before testing and habituated to the maze for 1 min before testing by placing them in the center of the maze and blocking entry to the arms. Mice were then tested for a 5 min period and their behavior recorded. Disinhibition was measured by comparing the time spent on the open arms to time spent on the closed arms.

Barnes maze – The maze consisted of a flat circular surface (36’ diameter) with 20 equally spaced holes (2’ diameter) along the outer edge. One of the holes led to a dark hide box while the other 19 led to false boxes that were too small to be entered. The latency to enter the hide box was recorded. The test was conducted in three phases. Phase 1 (Training): A hide box was placed under one of the holes. Animals were placed into an opaque cylinder in the center of the maze for 30 s to promote spatial disorientation at the start of the test. After 30 s, the cylinder was removed and the animal explored the maze until it found and entered the hide box.

The number of incorrect entries was scored. If the mouse failed to enter the box within 3 min, it was gently led into the box. The animal remained in the box for an additional 20 s before it was removed from the boxed and gently placed into the home cage. Training is repeated three times a day for four days.

The location of the hide box remained the same during every trial but it was shifted between subjects to reduce the potential for unintended intra-maze cues. Phase 2 (Retention): This phase measures retention of spatial memory following a delay. After a two day break from training, each animal was re-tested for a one day, three-trial session using the same hide box location as before. Phase 3 (Reversal): This phase examines memory reversal. On the day following the retention phase, a new hide box location was established 180 degrees to the original location. The same method as before was used and trials were repeated three times a day over two consecutive days.

Western blotting

Western blots were carried out as described previously (Currais et al., 2015b). The primary antibodies used were: HRP-conjugated rabbit anti-actin (#5125, 1/20,000), acetyl-histone H3 (Lys9) (#9649, 1/100000), phospho-ACC1 (#3661, 1/2000), total ACC1 (#4190, 1/1000), phospho-AMPK (#2535, 1/1000) and total AMPK (#2793, 1/1000), from Cell Signaling Technology; histone H3 (#ab24834, 1/100000) from Abcam. Horseradish peroxidase-conjugated secondary antibodies (goat anti-rabbit, goat anti-mouse or rabbit anti-goat (BioRad) diluted 1/5000) were used.

Whole transcriptome analysis

RNA was isolated from the hippocampus of SAMP8 mice or primary neurons using the RNeasy Plus Universal mini kit (Qiagen). RNA-Seq libraries were prepared using the Illumina TruSeq Stranded mRNA Sample Prep Kit according to the manufacturers instructions. Briefly, poly-A RNA was selected using poly dT-beads. mRNA was then fragmented and reverse transcribed. cDNA was end-repaired, adenylated and ligated with Illumina adapters with indexes. Adapter-ligated cDNA was then amplified.

Libraries were pooled and sequenced single-end 50 base-pair (bp) on the Illumina HiSeq 2500 platform. Sequencing reads were mapped to the mm10 mouse genome or rn6 rat genome using the spliced aligner STAR (2.5.1b) with default parameters (Dobin et al., 2013).

Reference mm10 mouse and rn6 rat genome was downloaded from UCSC. Only uniquely aligned reads were considered for downstream analysis. Expression values were quantified using Homer (4.9.1, Heinz et al., 2010) by counting reads mapped across all gene exons of RefSeq genes and mitochondrial encoded genes.

The differential expression (DE) analysis was performed by edgeR (v3.16.1, McCarthy et al., 2012). Briefly, genes with counts per million greater than one for at least half of the samples were normalized by the default ‘TMM’ method.

The dispersion was estimated by the ‘estimateGLMTagwiseDisp’ function. Genes with a false discovery rate (FDR) < 0.05 and an absolute log2 fold-change >0.3 (about 1.23 fold-change) were identified as significantly differentially expressed.

The data discussed in this publication have been deposited in NCBI’s Gene Expression Omnibus (Edgar et al., 2002) and are accessible through GEO Series accession number GSE101112 (

Transcriptome/metabolome drift analysis

Transcriptional drift analysis was performed as previously described (Rangaraju et al., 2015) with the exception that we removed expressed genes below the 20th percentile. We further normalized all samples setting the overall mean transcriptional drift to 0 to avoid differences in sample counts that affect drift variance across samples. Metabolomic drift variance was calculated by first determining the metabolomics drift (md) of each individual metabolite (md = log(old/young). The metabolomics drift variance is then determined by calculating the variance of md across the entire metabolome or subgroups of the metabolome.

Mitochondrial DNA copy number

Mitochondrial DNA copy number was determined in the hippocampus of SAMP8 mice using the kit from Detroit R and D, Inc (MCN 3) according to the manufacturer’s instructions.

Bioinformatics and statistics

R package ‘gplots’ (Warnes et al., 2016) was used to generate the heatmaps and MSEAs. For the RNA and metabolic heatmaps, all values were mean-centered and divided by the SD of each variable (scaled Z-score). The heatmap of RNA-Seq was z-scaled log2(FPKM+5) for Figure 3E and Figure 3—figure supplement 1C and log2 fold-change for Figure 3C and D and Figure 3—figure supplement 1A. Hierarchical clustering of RNA expression was performed using Euclidean distances and the Ward2 algorithm. K-means clustering of RNA expression was performed using Euclidean distances on z-scaled log2(FPKM+5) of 20077 expressed genes (at least one sample had greater than 0 FPKM). The total number of 20077 expressed genes were used as background for the DAVID pathway analysis for k1 cluster. MSEAs were generated according to HMDB and the ‘Pathway-associated metabolic sets’. Only top pathways are indicated.

PCA of the transcriptome data was performed on the log2(FPKM+5) values of the top 10% most expressed genes (cutoff: RowSums(log2(FPKM+5))>=110.41; 2458 genes selected) by the R base function ‘prcomp’ with data mean centered. Metabolome data was imputed to replace the missing value by the minimal value observed across all samples and transformed by the function ‘glog’ from the R package ‘FitAR’ (W. Huber, A. von Heydebreck, H. Sultmann, A. Poustka, and M. Vingron. Variance stablization applied to microarray data calibration and to quantification of differential expression. Bioinformatics, 18: S96-S10 2002) before PCA analysis. The ellipses showed the 70% confidence interval of a multivariate normal-distribution of sample groups estimated by PC1 and PC2 data.

Pathway enrichment test of the transcriptome data was analyzed by DAVID Bioinformatics Resources v6.8 (Huang et al., 2009a and Huang et al., 2009b). The total number of expressed genes of each individual pair-wise comparisons was used as background for the enrichment test.

Pathway enrichment analysis on metabolic pathways was conducted using the MetaboLync platform (Metabolon). Enrichment scores = (number of significant metabolites in pathway/total number of detected metabolites in pathway)/(total number of significant metabolites/total number of detected metabolites).

The metabolic network was reconstructed by GAM (Sergushichev et al., 2016) with input of the metabolome and/or the transcriptome dataset(s) and visualized by Cytoscape (Shannon et al., 2003).

Random Sampling: The dataset that contained the test statistics (p-value and fold-change) between 13 months+CMS121 and 13 months was used for the random sampling experiment. The Gene IDs and the Compound IDs were randomly shuffled 10 times (sampling without replacement) followed by the network analysis using the same parameters and cutoffs. The probability of observing acetyl-CoA in the resulted network was calculated as pnull

. Random sampling p=pnull^na*(1-pnull)^nb (na = number of experiments that identified acetyl-CoA in the network; nb = number of experiments that did not identify acetyl-CoA in the network). In this experiment, pnull = 0.1 and random sampling p=0.01. For the rat primary neuron transcriptome data, random sampling was performed using the same method with data generated from the comparison between CMS121 and control.

GraphPad Prism six was used for statistical analysis and exact P values are indicated (for p<0.050).

Salk Institute
Media Contacts:
Salk Communications – Salk Institute
Image Source:
The image is in the public domain.

Original Research: Open access
“Elevating acetyl-CoA levels reduces aspects of brain aging”. Antonio Currais, Ling Huang, Joshua Goldberg, Michael Petrascheck, Gamze Ates, António Pinto-Duarte, Maxim N Shokhirev, David Schubert, Pamela Maher.
eLife doi:10.7554/eLife.47866.


Please enter your comment!
Please enter your name here

Questo sito usa Akismet per ridurre lo spam. Scopri come i tuoi dati vengono elaborati.