Tinkering with Alzheimer’s
data, I noticed that Somatostatin, the most commonly downregulated protein in
Alzheimer’s studies, was also downregulated in the brains of alcoholics and
bipolar disorder patients. This was somewhat bothersome; could many or most of
the genes I had identified as Alzheimer’s markers simply be broad markers for
neural conditions in general? I thus constructed new “canonical” lists for
brain conditions with at least a handful of post-mortem studies, and compared
these to the canonical Alzheimer’s list. Specifically, we have
up/downregulation lists for Parkinson’s (database IDs 124415121 and 124416121),
alcoholism (124416121
and 124416122), schizophrenia (124417121
and 124418121), depression/bipolar disorder (124419121 and 124419122), and
autism (124420121 and 124421121).
Intersecting studies with studies, one gets P-values via Fisher’s exact test. Below, red signifies strong intersections (beginning at –log(P)>4 and as high as 91), white gives intersections with weak statistical significance, and blue signifies weak anti-correlation (in no cases was anti-correlation particularly significant*). To answer the initial question: no, most genes in our canonical lists are found in only one list; there’s no single set of genes that typifies neural disorders.
A |
upregulated in frontal cortex of
Creutzfeldt-Jakob patients (GSE124571) |
B |
downregulated in frontal cortex of
Creutzfeldt-Jakob patients (GSE124571) |
C |
upregulated in midbrain of
cocaine-addicted subjects (GSE54839) |
D |
downregulated in midbrain of
cocaine-addicted subjects (GSE54839) |
E |
upregulated in postmortem brains of
hiv patients w/neurocognitive disorders who took ARVs (GSE28160) |
F |
downregulated in postmortem brains
of hiv patients w/neurocognitive disorders who took ARVs (GSE28160) |
G |
canonically upregulated in blood of
alzheimer's patients (n=2) |
H |
canonically downregulated in blood
of alzheimer's patients (n=2) |
I |
canonical up in mouse brain
alzheimer's model (12 studies; n>=3) |
J |
canonical down in mouse brain
alzheimer's model (12 studies; n>=2) |
K |
canonical up in mouse eae brain (8
studies; n>=4) |
L |
canonical down in mouse eae brain
(8 studies; n>=2) |
M |
canonical up in Parkinson's brain
(5 studies; n=2) |
N |
canonical down in Parkinson's brain
(5 studies; n>=2) |
O |
canonical up in alcoholic brain (4
studies; n>=3) |
P |
canonical down in alcoholic brain
(4 studies; n>=2) |
Q |
canonical up in schizophrenia brain
(17 studies; n>=3) |
R |
canonical down in schizophrenia brain
(17 studies; n>=3) |
S |
canonical up in depression/bipolar
brain (8 studies; n>=2) |
T |
canonical down in
depression/bipolar brain (8 studies; n>=2) |
U |
canonical up in human Alzheimer's
brain (found in at least 5 of 35 studies) |
V |
canonical down in human Alzheimer's
brain (found in at least 5 of 35 studies) |
W |
canonical up in autism brain (5
studies; n=2) |
X |
canonical down in autism brain (5
studies; n=2) |
What are the genes that are found in two or more of the “canonical” lists?
|
We’ve highlighted genes that are upregulated in one list, but downregulated in others. Of 70 genes found in canonical list intersections, 10 fall into this category. One might give special importance to these genes, speculating that various disorders “hinge” on their expression. At the same time, folks who are interested in defining the causes of particular neural maladies may wish to de-emphasize the remaining 60 genes on the assumption that they are “responders” to messed-up chemistry, not causative. One relevant caveat would be the fact that our canonical lists are created by combining studies that probed a number of brain regions. Thus we treat the brain as a single entity, not an organ with a myriad of distinct cell types, and it’s still possible, for example, that ITPKB would be a very specific marker for Alzheimer’s if you focused exclusively on the hippocampus.
We’ll dish out a few more posts on the subject of Alzheimer’s, and then move on to new subjects (the effects of prolonged vibration of mice? The blood transcriptome of meditating monks? Or just jump into another heavy topic, like cancer or cytokine storms?)
*We should point out that it's difficult to identify strong anti-correlation. If you have 20,000 genes, and two randomly-derived subsets of 100 genes, it's not at all interesting if there's no intersection between the two subsets. If you want to see anti-correlation, you need long lists. Our canonical lists are quite short. Thus, strong anti-correlations may exist, with the blue color hinting at them.
**You can't make a single Venn diagram with 14 lists, but there is a nice Venn-diagram tool that will nevertheless detail all of the different intersecting groups.