Most of the risk factors associated with Alzheimer’s disease from observational studies are likely wrong

Interesting article from Korologou-Linden et al 2022, “The causes and consequences of Alzheimer’s disease: phenome-wide evidence from Mendelian randomization”.

In this study, the authors used a Mendelian randomization approach to examine the causal relationships between various risk factors and Alzheimer’s disease. They used the UK Biobank, which has a massive sample size of >300,000 participants.

They found that genetic variation at one gene — APOE — is far and away the main mediator of Alzheimer’s disease genetic risk. This replicates why Alzheimer’s has been called a quasi-monogenic disease — APOE has that large of an effect.

They don’t hold anything back in the discussion, basically arguing that their study disagrees with observational studies because their methodology is better and observational studies are wrong, because observational studies can’t identify causality.

Instead, they suggest that associations with Alzheimer’s disease in observational studies are due to reverse causation (i.e. they are symptoms of early/prodromal Alzheimer’s disease, rather than causes) or simply due to selection bias.

This means that the things often associated with Alzheimer’s disease in observational studies — body mass index, blood pressure, and physical activity — might not actually increase the risk of disease. Interestingly, Alzheimer’s genetic risk in this study is actually associated with a lower body mass index and body fat in people 53-72 years old.

They also found that Alzheimer’s disease risk is associated with a lower fluid intelligence score, with no causal effect on educational attainment.

One caveat I have with the study is that I’d like to learn about the associations of these risk factors with other forms of cognitive impairment. Laypeople often use the term “Alzheimer’s” to refer to dementia or age-related cognitive impairment in general.

For example, does a higher genetic risk for elevated blood pressure or body mass index causally affect the risk for vascular-associated cognitive impairment? My guess is that they might, which might cause this study to be a bit misleading if the results were taken in the wrong way.

That said, there seems to be a real effect of APOE on the risk of cognitive impairment that is independent of classical risk factors body mass index, blood pressure, and physical activity. And this study helps to parse out how that might be occurring, which will hopefully help to develop better preventive approaches and treatments.


Single cell histone modifications seem to accumulate randomly during aging

One of the most remarkable findings in aging over the past decade is that it’s possible to track the rate of aging based on stereotyped DNA methylation changes across a diverse set of tissues. These are known as epigenetic clocks.

But as anyone in the gene expression field knows, changes in the levels of epigenetic markers between groups (like young vs older) is confounded by cell type proportion differences between those groups.

This cell type proportion confound makes it harder to tell whether the changes in DNA methylation are truly a marker of aging or whether they are due to cell type proportion variations that may be already known to occur during aging, like naive T cell depletion due to thymus atrophy.

Single cell epigenetics has the potential to address this problem. By measuring DNA methylation patterns within individual cells, you can compare the epigenetic patterns within the same cell type between groups, and don’t have to worry (as much) about overall changes in cell type proportion [1].

I was interested to see whether anyone has used single cell epigenetic profiling, which was just come out within the past couple of years, to measure whether changes in epigenetic marks can be seen within single cells during aging.

First, let’s back up a second and talk about epigenetics. Two of the major factors that defines a cell’s epigenome are its DNA methylation patterns and its histone post-translational modifications.

DNA methylation has been studied a bit in single cells. One study looked at DNA methylation in hepatocytes and didn’t find many differences between old and young cells.

However, as a recent review points out, single cell DNA methylation data are currently limited because of sample quantity within each cell, and can’t easily compare methylation patterns between different cells in the same region of the genome.

On the histone modification front, I found a nice article by Cheung et al 2018, who measured histone post-translational modifications (PTMs) in single cells derived from blood samples. They found that in aging, there was increased variability histone PTMs both between individuals and between cells.

So, in summary, here are some future directions for this research field that it would be prudent to keep an eye one:

  1. How much of the changes in DNA methylation seen in aging are due to changes in relative cell type proportions as opposed to changes within single cells? If we assume that age-related changes in DNA methylation will be similar to age-related changes in histone PTMs, then Cheung et al.’s results suggest that the changes in DNA methylation are probably due to true changes within single cells during aging.
  2. Is there a way to slow or reverse age-related changes in DNA methylation or histone PTMs, perhaps targeted to stem cell populations? It’s not clear that this can be done in a practical way, especially if age-related changes are driven primarily by an increase in variability/entropy.
  3. If it is possible to slow or reverse DNA methylation or histone PTMs, would that help to slow aging and thus “square the curve” of age-related disease? Aging might be too multifactorial for a single intervention like this to make a major difference, though.

[1]: I add “as much” here because differential expression analysis in single cell data is far from straightforward, and e.g. has the potential to be biased by subtle differences in the distribution of sub-cell type spectrum between groups.

Shared genetic correlations of psychiatric and neurologic disorders

A few weeks an interesting preprint from Antilla et al. was published. They set out to measure the genetic correlation between a variety of brain disorders — both “psychiatric” and “neurologic” — by comparing risk markers from a set of 23 different GWAS’s. They called themselves the “Brainstorm consortium” (for which they win creativity points). A major finding in their paper is that there is a substantial correlation between psychiatric disorders (e.g., OCD, schizophrenia, MDD, bipolar disorder), while there is less or no correlation among neurologic disorders (e.g., Alzheimer’s, Parkinson’s, MS). This data set is based on comparing polygenic risk variants from individual studies, and it’s certainly possible to draw too strong of conclusions from this type of data, as it is confounded by the societal structure of the people who participated in the studies, among other factors. That said, this should stimulate a number of interesting follow-up studies. One of their most interesting sections is on the genetic correlations between these disorders and other traits:

Screen Shot 2016-05-07 at 5.13.18 PM

Two correlations especially jump out to me here:

  1.  The positive correlation between autism spectrum disorder risk and variants associated with measures of cognitive performance. This fits with at least one finding that there is a positive association between ASD prevalence and socioeconomic status, which is sometimes attributed to increased paternal age, but as this study shows, that is potentially not the whole story. I’m certainly not an expert in ASD epidemiology and this is just my initial impression, and I could totally be off.
  2. The inverse correlation between variants associated with measures of cognitive performance and risk of stroke and intracerebral hemorrage. This fits with my priors that good blood flow is critical for proper brain function. In my experience is not as widely known by people without a medical background (such as myself prior to my preclinical med school training).
Antilla et al. 2016 Analysis of shared heritability in common disorders of the brain. doi:

The two SNPs with the strongest associations to schizophrenia

According to a GWAS published yesterday (using a sample of individuals with European ancestry only), they are in these genes:

1) HLA-DQA1: this codes for a receptor involved in antigen presenting cells, and might be related to autoimmune disorders. This speculative paper bites the bullet on the connection and suggests that better treatment for pathogens could reduce the prevalence of the disorder. The hypothesis seems testable via epidemiological data–does more pathogen treatment (or treatment of a specific type) lead to a decreased incidence of schizophrenia? I doubt there’s a big effect here; otherwise, we’d probably already know.

2) MADL1L: this codes for a protein which helps regulate cell division (mitosis), almost all of which occurs in the brain during development. Although schizophrenia is often considered a developmental disorder, it doesn’t typically present until young adulthood, with some studies reporting a mean age of onset of 30. This suggests that the spatial pattern and distribution of cells set down early in development can probabilistically impact outcomes much later in life. (This is as opposed to a gene being involved in synapse remodeling, which is common throughout adulthood.)

Multiple SNPs within both of these genes have significant p-values, which makes the explanation of linkage disequilibrium seem less likely. (Plus, the authors checked the haplotype maps for this.)


Jia P, Wang L, Fanous AH, Pato CN, Edwards TL, et al. (2012) Network-Assisted Investigation of Combined Causal Signals from Genome-Wide Association Studies in Schizophrenia. PLoS Comput Biol 8(7): e1002587. doi:10.1371/journal.pcbi.1002587

Why do olfactory neurons only express one type of receptor?

In their paper released today, Jafari et al make major strides in answering this question. First, they systematically knocked down 611 transcription factors in Drosophila (~80% of the total) in four representative classes of olfactory sensory neurons. They identified seven whose loss led to a strong decrease in odorant receptor expression.

Next, they showed that knocking down at least one of these seven transcription factors in almost all of the known olfactory sensory neuron classes (32/34) caused that class to stop expressing its olfactory receptor.

rows = transcription factors, columns = olfactory sensory neuron classes; grey = wildtype-like expression, black = no expression, odorant receptor expression detected by ISH; orange = trichoid, one of the three major odorant receptor expression domains; note that the raw data in table s2 is a bit more noisy than the simplified version above, as expected; doi:10.1371/journal.pbio.1001280

In their discussion, the authors mention that the seven transcription factors they found is likely to be an underestimate. This makes sense because the library wasn’t available to screen every transcription factor, and RNAi is stochastic. Regardless, their data set and paradigm should open up many avenues for studying combinatorial transcriptional coding.


Jafari S, Alkhori L, Schleiffer A, Brochtrup A, Hummel T, et al. (2012) Combinatorial Activation and Repression by Seven Transcription Factors Specify Drosophila Odorant Receptor Expression. PLoS Biol 10(3): e1001280. doi:10.1371/journal.pbio.1001280

Odor recognition coding by mice olfactory neurons

In mice, each olfactory neuron expresses exactly one of ~ 1000 types of olfactory receptors. Through combinatorial coding, the system is able to recognize a wide range of odors and their combinations. But how exactly is this diversity of responses achieved?

Nara et al. recently set out to answer this question. They put dissociated mice olfactory epithelium cells on glass coverslips, loaded them with a calcium indicator, and monitored them for changes in calcium signaling following the application of 13 different odorant mixtures.

OSN = olfactory sensory neuron, the actual number of neurons in each group is shown above each bar (out of a total of 217 tested); doi: 10.1523/ JNEUROSCI.1282-11.2011

As you can see above, most neurons responded (i.e., demonstrated an increased calcium concentration) to just one mixture, but some neurons did express receptors which allowed them to respond to many mixtures.

If a neuron responded to a given mixture, it was then tested for a response to each of the individual odorants in that mixture. Here is the response curve for one of their neurons, which responded to many different mixtures:

a sharp line indicates a change in fluorescence intensity, which is a proxy for neural activity; the bars indicate when the odors were added; the final response to KCl indicates neuron viability; Fi = change in fluorescence; doi: 10.1523/ JNEUROSCI.1282-11.2011

This neuron was considered an example of a “broad tuning,” since it responded to different odors with high structural variability.

Although all of the responses marked in red were classified as a “recognition,” some responses (such as 4-2) seem to be stronger than others (such as 6-10).

It might be interesting to analyze multiple replicates of responses from the same odor on the same neuron, to allow the authors to parse signal from noise and see whether these gradations in response are significant.

Cracking the full odorant code will require an even more high-throughput set of experiments, and probably would have to have at least one data point from each type of odorant receptor. This study is a clear proof of principle that such an extension would be possible and valuable.


Nara K, et al. 2011 A Large-Scale Analysis of Odor Coding in the Olfactory Epithelium. doi: 10.1523/​JNEUROSCI.1282-11.2011

Greater variability in neuronal than glial DNA methylation patterns

Iwamoto et al. found this in their recent study, by looking at the post-mortem prefrontal cortex and cerebellum of humans and chimpanzees. They separated nuclei from these regions into neuron and non-neuron (i.e., glia) groups with purities of 95 and 99.9%, respectively, using NeuN as a marker for neurons.

They then used magnetic beads to extract methylated DNA molecules from each of these groups and examined the DNA with a tiling array. By looking at the correlation of tiling array probe replicates within the same group (i.e., either neurons or non-neurons), they were able to tell which group had more variability.

They found that this correlation was lower in neuron samples (average R = 0.850) than in non-neuron samples (average R = 0.875). The effect size is not huge, but it is extremely unlikely to be due to chance (see this for yourself in the histogram in fig 8 if you have access).

This is intriguing, if indirect, evidence that epigenetic patterns have an especially large effect on the function of neurons.

It’d be interesting to see a study take a similar approach but select for specific types of brain cells (with different antibodies) to see if there is still high variation within that one class of cells. This would allow us to distinguish between epigenetic changes due to cell type differentiation and those due to neural activity and experience. Of course, purifying a discrete class of brain cells on the basis of one antibody would likely not be so easy.


Iwamoto K, et al. 2011 Neurons show distinctive DNA methylation profile and higher interindividual variations compared with non-neurons. Genome Research doi:10.1101/gr.112755.110.

Prediciting brain connectivity in the rat via regional gene expression in the mouse

In a study conceptually similar to another one I mentioned a few months ago, Wolf et al discuss their results from comparing and contrasting USC’s rat brain connectivity atlas with the Allen mouse brain gene expression atlas.

To combine these, they first had to map the regions from the rat brain to the mouse brain.

The authors then trained a linear classifier to try to predict whether two given brain regions have a connection or not, using vectors of gene expression of the 500 most predictive genes in both incoming and outgoing connections as features. They use 80% training and 20% testing cross-validation. Finally, they randomly shuffle the brain regions and re-do the analyses to compute empirical p-values.

The Allen brain atlas is hierarchical, which is useful for some analyses but could lead to double counting of brain regions here. So the authors only analyze the brain regions at the lowest level of the hierarchy–the outermost nodes in the circular diagram below.

Only brain regions with more than 5 outgoing or incoming connections were analyzed. They found that connectivity was able to be predicted by the gene expression in many regions. See below for details:

second to outermost ring = outgoing connections (c), outermost ring = incoming connections (d); green with dot = predictions have <5% prob of being as accurate due to chance, yellow = prediction has >5% prob of being as accurate due to chance; doi:10.1371/journal.pcbi.1002040

They also analyzed genes that are thought to be involved in brain disorders like schizophrenia. Schizophrenia-related genes are much more likely to be involved in connectivity patterns that would be expected due to chance, bolstering the hypothesis that this disorder is related to neural connectivity. Defining the null hypothesis here seems a bit tricky though, so we’ll need a wide breadth of studies to help confirm this finding.

It would be interesting to see if the predictive ability of gene expression is higher using the developing mouse brain atlas as opposed to the adult. This is expected because that’s when most long-range axons form, but there is also lots of dendrite rearrangment and plasticity in adults, too.


Wolf L, Goldberg C, Manor N, Sharan R, Ruppin E (2011) Gene Expression in the Rodent Brain is Associated with Its Regional Connectivity. PLoS Comput Biol 7(5): e1002040. doi:10.1371/journal.pcbi.1002040

How neurons remain neurons

A short Feb ’11 review by Oliver Hobert (HT: J Snyder) explains the process. A particular protein called a terminal selector coordinates it, and acts by binding to DNA sequences. One might describe the process as involving three main steps:

1) Initiation. Initiation occurs when neuroblasts terminally divide. An initiator protein binds to the DNA upstream of the gene encoding the terminal selector (in particular, to the “cis-regulatory element” of the DNA). This activates transcription of the terminal selector, and thus its translation as well. Crucially, the initiator protein is itself only expressed for short window of time.

2) Propagation. The terminal selector binds to the cis-regulatory elements upstream of “terminal differentiation genes,” activating their expression. These genes are involved in neural function, such as neurotransmitter metabolism and ion channels. Some also presumably act to arrest the cell’s growth phase in G0.

3) Maintenance. Through a common mechanism known as transcriptional autoregulation, the terminal selector gene maintains its levels by binding to a cis-regulatory element upstream of its own gene, thus activating its own expression. So, long after the initiator protein is no longer present (and indeed for the lifespan of the animal), the expression of the terminator selector gene will remain high, and it will, in turn, continue to activate the expression of the terminal differentiation genes.

This is also an interesting case study in the interplay between chromatin states and the action of transcription factors. New (“de novo”) events of transcription factor binding require the chromatin to “open up” to allow them to bind the DNA. An individual transcription factor protein molecule probably only binds to the DNA for short periods of time (low dissociation constants suggest it’s often on the scale of milliseconds). This also leads to remodeling of the chromatin state via histone modifications, which over the long run might make binding of the transcription factors easier.

But how important are the relative contributions of de novo transcription factor binding, histone modifications, and the initial chromatin state of DNA upstream the terminator selector and terminal differentiation genes? As far as I can tell, these remain somewhat pressing and open questions.


Holbert O, 2011, Maintaining a memory by transcriptional autoregulation, Current Biology. doi:10.1016/j.cub.2011.01.005

Kiełbasa SM, Vingron M (2008) Transcriptional Autoregulatory Loops Are Highly Conserved in Vertebrate Evolution. PLoS ONE 3(9): e3210. doi:10.1371/journal.pone.0003210

Wang Y, et al. 2009 Quantitative Transcription Factor Binding Kinetics at the Single-Molecule Level. Biophys Journal 10.1016/j.bpj.2008.09.040.

Neural macro connectivity and gene expression

French et al explore this link, looking at the correlation between gene expression in the mouse in and connectivity in homologous brain regions of the rat. This is their conceptual scheme:

Of their 142 common regions, 112 have efferent (outgoing) connections, and 141 have afferent (incoming) connections. There are 5216 outgoing connections and 6110 incoming connections.

To see whether the similarity of the connectivity between two regions is related to the similarity of gene expression between those two regions, they looked at the correlation between the connectivity and gene expression matrices. Specifically, they used a Mantel test, which measures the Pearson correlation between every entry in the matrices, and gets significance by permuting the order of the entries and recalculating the correlation after each permutation.

They find that brain regions with similar connectivity tend to have similar gene expression. The Mantel correlation between expression and incoming connectivity patterns (141 regions) is 0.248, with a permutation (exact) p-value < 0.0001, and for outgoing the correlation is 0.226, p also very low.

This is a really good proof of principle that expression and brain connectivity feed back upon one another. As the gene expression and connectivity measures become more precise, these types of analyses will yield valuable insights related to neural development and disease.


French L, Pavlidis P (2011) Relationships between Gene Expression and Brain Wiring in the Adult Rodent Brain. PLoS Comput Biol 7(1): e1001049. doi:10.1371/journal.pcbi.1001049