A Big Data Look at Alzheimer's Disease, Part I

 

Here, we’ll take on Alzheimer’s disease. The subject is huge, of course, so we’ll devote future posts to Alzheimer’s as well. For the near term, we’d like to make the following points:

1. Mouse Alzheimer’s models and EAE (neuroinflammation) models overlap quite significantly, at least when examining upregulated (vs. downregulated) transcripts.
2. Hooray for mouse models, right? Actually, maybe not. That’s because the mouse Alzheimer’s and EAE models fail to intersect with transcripts altered in actual human Alzheimer’s brain tissues.
3. Somatostatin stands out in our analysis as the most consistently altered transcript in the human Alzheimer’s brain.
4. Hoxa5 may be relevant to Alzheimer’s.
5. A study with antiretrovirals may hint at a treatment option.

To tackle the topic, we constructed a number of “canonical” regulation lists. First, there are lists of transcripts that are up/down-regulated in real human Alzheimer’s cases, when compared to healthy control brains. We used a total of 35 different transcriptomic studies to construct the upregulation and downregulation lists. The database IDs for these lists are 123049121 and 123050121. Similarly, we constructed canonical lists for transcripts altered in mouse Alzheimer’s model brains (12 studies: 123633121 and 123634121) and in mouse EAE model brains (8 studies: 123635121 and 123636121). We also built up/down-regulation lists for transcripts altered in the blood of Alzheimer’s patients (123069121 and 123070121); these blood-focused lists are not particularly relevant to this post, but you may wish to tinker with them anyway using the apps at WhatIsMyGene.

Let’s look at upregulation in the three relevant canonical lists:

Had this post been an academic paper, that Venn (via interactivenn.net) would have consumed half a day. Instead; about 3 minutes. It’s easy to see that, from the big-data transcriptomic point of view, the mouse Alzheimer’s models, most (but not all) of which utilize overexpression of amyloid beta precursor, have no relevance to human Alzheimer’s. The same goes for the EAE (experimental autoimmune encephalitis) mouse studies. Of note, the single transcript found at the intersection of human Alzheimer’s and the EAE studies is SERPINA3. The P-value for the mouse Alzheimer’s model/EAE model intersection works out to about 10^-20.

For the sake of argument, let’s pretentiously assume that our own big-data approach invalidates the notion that mouse AD models and EAE studies have anything to do with Alzheimer’s in humans. Why, then, has so much been invested in the mouse studies? Without immersing myself in the literature, I’d offer up several points:

•     Amyloid plaque is the most obvious hallmark of Alzheimer’s. It makes sense to induce these plaques in mice (via knock-ins). These plaques, in turn, induce hallmarks of EAE.
•     Some evidence suggests that neuroinflammation may positively correlate with Alzheimer’s in humans. For example, numerous studies show an inverse correlation between rheumatoid arthritis and Alzheimer’s, with researchers speculating that the use of various anti-inflammatories may provide the crucial link.
•     Some human neurological diseases indeed overlap with both Alzheimer’s and the mouse models and EAE studies. For example, multiple sclerosis offers parallels with both human Alzheimer’s and EAE, and Creutzfeldt-Jakob brains offer parallels with both human Alzheimer’s and the mouse Alzheimer’s models.

Thus, it’s easy to assume that the mouse experiments are moving the field in the correct direction. However, to refute the above points:

•     It’s apparent that the plaques cause neuroinflammation, not the inverse (i.e. knock in APP and generate a mouse with neuroinflammation).
•     Unfortunately, recent studies have blurred the correlations between arthritis, anti-inflammatories, and Alzheimer’s. Even assuming that the negative correlation between RA and Alzheimer’s holds, some have questioned whether anti-inflammatories are the molecules that counter neuro-degeneration.
•     Connecting Alzheimer’s with EAE, using multiple sclerosis or Creutzfeldt-Jakob studies as an “intermediary”, is like children playing the game of “telephone.” A lot can get lost in translation. As an example, look below. Transcripts upregulated in Creutzfeldt-Jakob brains (PMID 30446946) overlap strongly with those upregulated in Alzheimer’s, and the Creutzfeldt-Jakob data overlaps nicely with mouse Alzheimer models (-log(P) of approximately 20 and 12, respectively). This doesn’t alter the nonexistent connection between mouse models and human Alzheimer’s, however.

On the positive side, it’s possible that therapies that remediate plaques in mice might be beneficial for Creutzfeldt-Jakob patients. The mouse Alzheimer’s models are relevant to human disease, but perhaps not human Alzheimer’s.

One simple point: Note that we do find human/mouse overlap between two neurodegenerative conditions. One needn't wonder if somehow our algorithms are failing to convert mouse identifiers to their orthologous human counterparts. 

Below is a Venn diagram examining downregulated transcripts in the three “canonical” lists:

Here, the three sets show even less relevance to each other. If you’re interested, the three transcripts found in the human Alzheimer/mouse model intersection are NPTX2, DUSP6, and PCSK1.

Note Jan 2021: How did I miss this?...Mouse models rarely mimic the transcriptome of human neurodegenerative diseases: A systematic bioinformatics-based critique of preclinical models. A 2015 paper that has apparently been ignored for the most part.

Looking at the subject from a purely transcriptomic/big-data view, it would seem that amyloid plaques are a by-product of Alzheimer’s pathology.

We’ve covered our initial points 1) and 2). We’ll address the next three points, and more, in upcoming posts.

whatismygene.com 

Become a Patron!

A New Tool: Intersect

We've added a new tool that we simply named "intersect." This tool allows you to find the genes that are common to two sets.

If you enter two gene sets of the same type (and click the "same type" box), you'll receive the intersecting genes. This could easily be performed in an Excel spreadsheet using, for example, the "vlookup" function. In this case, you'll receive output fairly quickly because there's no need to dip into any database of gene IDs.

However, this tool can be used for two situations that would be problematic for Excel. First, you can find the intersection of two gene sets that have different ID types (e.g. ENSG0000012345 and STAT6). The algorithm is simple, but requires a fair amount of processing: convert both the gene lists to the format used in our database, then find all common IDs, and then re-convert the IDs to uniprot (e.g. STAT6) format. If a uniprot ID is not available for a particular gene, we may output a different format. Second, you can use our own database IDs as input. This way you can find genes in common with your own set and the sets found in our database. Given the processor-intensive nature of the tool, we limit the size of a list of genes to 400 IDs, unless the "same type" box is clicked. 

Ideally, we'd output the intersecting genes any time you use, say, our "Fisher" tool. Again, though, this is problematic.

Have fun with the tool. What happens if you enter a list of human genes and a list of c elegans genes? You'll get a list of orthologs that are found in both sets, output as human uniprot IDs. We can't claim the process is infallible as, for example, our database may not have a absolutely complete list of c elegans genes that are orthologous to human genes. But it should work quite well.

whatismygene.com 

Become a Patron!

 

The Big-Data View of Resveratrol

I’m fascinated with the transcriptomic/proteomic effects of “health food” supplements. The fascination doesn’t derive from being a “believer”, but from being a skeptic. At least at the level of cell-culture and the force-feeding of mice, however, some of these supplements really do alter gene expression. Here, we’ll do a “shallow dive” on the topic of resveratrol, a compound prominently found in wine and peanuts, and likely available in a local health food shop (possibly in the form of “grape skin extract”).

There are 12 studies relating to the transcriptomic effects of resveratrol in our database. Just type the keyword “resveratrol” into the “relevant genes” app to see these studies. Two of the studies use CRISPR to find genes that enhance sensitivity to resveratrol, while a third compares resveratrol and pinosylvin. We tossed these 3 studies and created lists of “canonically” upregulated and downregulated transcripts following resveratrol treatment in the remaining 9 studies. You can play with these canonical datasets yourself using the dbase IDs 122772121 and 122773121. On the upregulation side, the following genes were found in 3 of the 9 studies:

RETSAT, PIK3IP1, PLA2G4C, ACTA2, ALDOC, GADD45A, CDKN1A, DDIT3, JUN, SERTAD1

On the downregulation side, 2 genes were actually found in 4 of the 9 studies, with a large number (which we won’t list) found 3 times:

KIF20A and RRM2

If you plug KIF20A or RRM2 into the coregulation tool, you'll see that these two genes tend to be up/down-regulated hand in hand (log(P)=-200).

Let’s jump to the juiciest (but not most significant) results. One of resveratrol’s myriad supposed positive effects would be life extension. And, in fact, Fisher analysis of the upregulated genes (122772121) actually suggests relevance. For example, these genes overlap well with those in a study involving metformin treatment of MCF7 cells; for a taste of metformin’s effects, see the review, Metformin as Anti-Aging Therapy: Is It for Everyone? Metformin is an anti-diabetic drug; another anti-diabetic, troglitazone, also overlaps with the effects of the canonically upregulated set. Another overlapping study (SIRT1 activator SRT1720 extends lifespan…) is obviously relevant. A final study leads us deeper into the realm of health food supplements: Identification of Targets of a New Nutritional Mixture for Osteoarthritis Management Composed by Curcuminoids Extract, Hydrolyzed Collagen and Green Tea Extract.

Downregulated transcripts, as suggested above, are more significantly relevant to resveratrol’s effects. These transcripts overlap quite strongly with studies involving the cell cycle (e.g. studies involving CDK4/6 inhibitors, segregation of G1 and S phases, and knockdown of cell-cycle regulators such as DDX6, DHX9, and TRIM33).

Anti-cancer relevance is seen in the results. Most powerfully, genes upregulated in cervical cancer are downregulated (log(P) = -79 via our “Fisher” app) in the canonical downregulation list. Other extremely significant overlaps relate to yet another cervical carcinoma study, laryngeal carcinoma, and high-grade astrocytoma. Inserting both canonical datasets into the “Match Studies” app suggests particular relevance to luminal B (vs A) breast cancer, and myc (vs ras) –driven cancers.

Numerous other drugs in the database seem to offer similar profiles to resveratrol: butylidenephthalide, ly101-4b (an e2f inhibitor), roscovotine, and BET inhibitors, for example. These similarities may be important as resveratrol’s optimal dosage and bioavailability to various organs may be questioned. We do note, however, numerous studies suggesting an effect at relatively low dosages (e.g. Low dose resveratrol improves cerebrovascular function in type 2 diabetes mellitus, Low dose resveratrol ameliorates mitochondrial respiratory dysfunction and enhances cellular reprogramming, and more).

whatismygene.com 

Hello!

This will be the blog that accompanies our  bioinformatic analysis site, whatismygene.com .

On the site, you'll find some background on the various sorts of analyses that can be performed on the site...gene searches, co-expression analysis, bio-enrichment, and a few esoteric tools as well, all backed by a monster database. The explanations are admittedly terse; we can delve deeper into the functions of the site here. We won't shy away from discussions of the weaknesses of our methods as well.

Not mentioned in the site's primer is the fact that whenever we enter new sets of studies into our database, we examine them individually for overlap (via Fisher's exact test) with previously entered studies. This sometimes points out possible errors we've made in data entry, but there's a more basic reason we take this step: curiosity. The result is a large array of observations and hypotheses, none of which have crystallized into something resembling an academic paper. Is there a link between production of dsRNA and Alzheimer's? What really is a "cytokine storm"? Do micro-RNA target lists and micro-RNA knockdown/overexpression studies align at all? Could standard chemotherapy/immunotherapy agents shrink primary cancers while promoting metastasis? Which health foods and traditional medicines actually have interesting transcriptomic effects? We plan on devoting a big chunk of this blog to these sorts of questions.

Become a Patron!