Arterial aging and its relation to Alzheimer’s dementia

I’m a big proponent of the role of arterial aging in explaining dementia risk variance, in large part because it explains the large role that vascular-related risk factors have in promoting the likelihood of Alzheimer’s disease (AD). However, some data suggests that the burden of ischemic events and stroke cannot explain all of the vascular-related AD risk. Recently, Gutierrez et al. published a nice paper which suggests that non-atherosclerotic artery changes with age may explain some of this residual vascular-related risk of AD. In particular, they used 194 autopsied brains and found five arterial features which strongly correlated with aging, including decreased elastin and concentric intimal thickening. Importantly, these features also correlated with AD risk independently of age.

The authors propose that the arterial aging features are a consequence of mechanical blood flow damage that accumulates over the years. If it is true that the damage is mechanical, it suggests that it may be difficult to reverse with existing cellular and molecular anti-aging therapies. For those people who are interested in slowing down aging, the brain must be a top priority because it cannot be replaced even by highly advanced tissue engineering approaches to replace the other organs. Thus, this sort arterial damage needs to be addressed, but to the best of my knowledge it has not been, which is one of the many reasons that I expect that serious anti-aging therapies are much further out than are commonly speculated in the popular press.

Telomeres: something to fear, or merely a seer?

Attention Conservation Notice: 1200+ words, with an estimated read time of seven minutes, of exuberant extrapolation from a new study which I was not associated with in any way. I am not an expert in telemore biology, nor am I a doctor.

In case you’ve been busy obsessing over whether or not to post things to social media over the past decade (oh wait, that’s me), you may not know that much about telomeres.

Telomeres are repetitive sequences of DNA at the end of chromosomes that get shorter with each cell division. If they get too short, they can promote cellular senescence.

They also are responsible for some of the prettiest pictures of DNA, in which people stain for the telomere “cap” at the end of the chromosomes, such as this classic image:

from the US DoE HGP via Wikipedia User:Gustavocarra

from the US DoE HGP via Wikipedia User:Gustavocarra

Much of the extrapolation of telomeres to aging is built around “Hayflick limit” models of aging.

However, these models do not explain all aspects of aging, and in fact, in most organ systems, cells do not die in large numbers during normal aging (a key exceptions being the thymus) [1]. Instead, cell numbers usually remain constant, because most cells in the body are post-mitotic.

Nevertheless, average telomere length has emerged as a very promising biomarker for aging, and a study on 60,000+ Danish people came out a few weeks ago that gives us a lot of new data on the subject. As far as I can tell, it is the largest such study to date.

Associating telomere length with mortality 

This study measured average telomere lengths in white blood cells (leukocytes) from peripheral blood samples.

They then measured death rates in a 0-22-year follow-up period (with a median follow-up length of 7 years).

Adjusting for age only, the individuals in the shortest decile of telomere length had an increased risk of death of 1.54 (95% CI 1.38 to 1.73).

And even after adjusting for age, sex, BMI, systolic BP, smoking status, tobacco consumption, alcohol consumption, physical activity, and cholesterol level, individuals in the shortest decile of telomere length still had an increased risk of death of 1.40 (95% CI = 1.25 to 1.57).

This shows that telomeres are, at the very least, a biomarker for aspects of aging other than those seen by these traditional clinical parameters.

It’s also possible to go further say “we corrected for nearly everything and still found an effect of telomeres on mortality, so there must be something causal” but that’s fraught with problems — the biomedical literature is truly littered with effects seen in correlational studies that didn’t replicate in causal ones — ibuprofen for Alzheimer’s disease, estrogen replacement therapy, etc.

This doesn’t mean it’s wrong, of course, but that the reference class probability is not as high as you might think.

Luckily, the genetic data in their study offers a way to address this question directly.

Genetic variants of telomere length as a window into causality

The authors genotyped participants for three SNPs that affect telomere length, one associated with each of the following genes:

  • Telomerase reverse transcriptase (TERT)
  • Telomerase RNA Component (TERC)
  • A gene that helps replicate and cap telomeres (OBFC1)

Each of these SNPs has two major alleles, which means that there are three possible states for each individual at each of the SNPs: a) no telomere-shortening alleles, b) one telomere-shortening allele, or c) two telomere-shortening alleles.

Since there are three total SNPs and two alleles at each, there are six total telomere-shortening alleles that each individual could have.

When the authors built a linear “score” from those SNPs, they found that it had a very strong effect on telomere length. I made a visualization of their raw data to show this:

Telomere length versus the sum of telomere-shortening alleles in each individual (allele score), +/-  the standard error; data from doi: 10.1093/jnci/djv074

Telomere length versus the sum of telomere-shortening alleles in each individual (allele score), +/- the standard error; data from doi: 10.1093/jnci/djv074; code

Under the assumption that these alleles don’t influence the aging process in any other way, this allele score offers a tremendous “natural experiment” into the causal role of average telomere length in blood leukocytes in humans.

What the authors found using this natural experiment is that telomere-shortening alleles led to a decrease in the risk of cancer (OR of 0.95 +/- 0.04 per telomere-shortening allele), but had no effect on all-cause mortality (OR = 0.99 +/- 0.02 per telomere-shortening allele).

One suggested mechanism for the cancer effect is that shorter telomeres give potentially cancerous cells a shorter “fuse” before they become senescent. So, people with shorter telomeres may be less likely to develop an actual malignancy.

So, my conclusion is that, despite the correlation seen in the section above, average telomere length likely does not play a causal role in age-related mortality in humans [2]. A few possible critiques of this conclusion:

1) The study not be well-powered enough to detect a difference in all-cause mortality. Some possible solutions here would be to a) increase sample size or b) do studies to find more variants affecting telomere length and then re-do the analysis with increased association power. But, needing a sample size > 60,000 to find an effect would mean that any effect that you did find would be very small.

2) Maybe it is not the average telomere length but rather some other property of telomere length distribution (such as the proportion of cells below or above some “critical threshold”) that is the relevant parameter. This is possible, but biology in generally doesn’t operate on such threshold mechanisms.

3) Josh Mitteldorf, if I understand him correctly, suggests that these three SNPs may affect telomere length only later in life, but not earlier in life. Although I don’t know the actual argument, I suppose it’s possible that telomere lengths early in life play the major causal role in later aging rates, and that these SNPs don’t affect telomere length until later in life. One possible solution here could be a longitudinal study of telomere length, or a cross-sectional study in infancy/childhood. Still, I don’t find either aspect of the argument here very likely.


Genetic evidence suggests that average telomere length (in blood leukocytes, in a Danish population) likely does not play a causal role, or at least does not play a strong causal role, in promoting aging.

That said, average telomere length in blood leukocytes does appear to be an excellent biomarker for aging, capturing large effects not accounted for by measuring traditional clinical factors that affect aging, such as smoking status or cholesterol levels.

This bolsters the evidence for Bojeson’s argument that telomere length should be used as a biomarker for aging and for assessing risk of age-related diseases in routine clinical practice.


Rode L, Nordestgaard BG, Bojesen SE. Peripheral blood leukocyte telomere length and mortality among 64 637 individuals from the general population. J Natl Cancer Inst. 2015;107(6):djv074.

Mitteldorf J, “Large New Survey Tracks Telomere Length and Mortality”. 2015.

Bojesen SE. Telomeres and human health. J Intern Med. 2013;274(5):399-413.

[1]: What about the brain, do I hear you say? A key tenet of the Alzheimer’s disease model is that neuronal loss in aging is not normal and only occurs in Alzheimer’s disease.

[2]: Let me be quite clear and point out the obvious that I agree this is a bad thing. That is, it would be great if telomere length were causal for aging, because it would suggest an obvious intervention to decrease the risk of age-related diseases: therapies to extend telomere length. Unfortunately, this study suggests that such an intervention will probably not have a beneficial effect on age-related mortality.

Parkinson’s GWAS improves when subclassifying patients by pathology, instead of clinical traits

Simón-Sánchez and Gasser discuss this in a stimulating article in the most recent issue of Neurology:

Beecham et al [2015] argue that one of the reasons for this “missing heritability” may be the uncontrolled degree of heterogeneity of [Parkinson’s Disease (PD)] in studies relying on clinical diagnosis alone. To reduce heterogeneity, they performed a GWAS in which only PD cases with autopsy-confirmed Lewy body (LB) pathology, and controls with exclusion of PD neuropathology, were included. Despite the relatively small number of samples used, they found evidence suggesting that common variation in a small region of chromosome 1p32 is associated with LB PD with a p-value just below genome-wide significance level and an odds ratio of 0.64 for the protective allele, which is comparable with that found for SNCA and MAPT in other studies. This association peak lies within PARK10, which was originally identified in a set of Icelandic families.

This “splitting” approach is likely to be valuable in other neurodegenerative diseases, as well.

Age-related changes in white matter are likely not due to demyelination?

That’s the conclusion of an important new paper from Billiet et al that integrated a bunch of different imaging modalities.

First, they used DTI to measure fractional anisotropy, which is a non-specific measure of white matter, measuring axon density, axon diameter, and myelin content. They also used myelin water fraction (MWF) to measure myelin content and the orientation dispersion index (ODI) to measure dendrite/axon dispersion. These were their findings:

  • Total white matter volume does not change in aging
  • Myelin content (via MWF) is not significantly altered in aging in most regions, and in the regions where there is a change, it increases
  • There is a widespread decrease in fractional anisotropy in aging that is especially strong in frontal regions
  • Changes in myelin content (MWF) do not correlate well with this decrease in fractional anisotropy
  • Changes in neurite orientation (ODI) dispersion do correlate with this decrease in fractional anisotropy

Thus, their conclusion is that age-related decreases in the “diffusibility” of white matter (or whatever fractional anisotropy is measuring) are due to changes in axons rather than to changes in myelin.

It’s all cross-sectional, n = 59, and they call this conclusion “highly speculative,” but if true it suggests that axons changes are more early/causal/fundamental to aging than changes in myelin. We need more data on this, especially from older individuals (their study went up to only 70) and from people with dementias, such as Alzheimer’s disease.

Billiet T, Vandenbulcke M, Mädler B, et al. Age-related microstructural differences quantified using myelin water imaging and advanced diffusion MRI. Neurobiol Aging. 2015

Journal club: How well do stem cell division rates predict cancer rates?

Summary: ~ 600 words on a moderately controversial recent paper that has been discussed endlessly elsewhere. Reading this is unlikely to be a good use of your time.

What Tomasetti et al did was to:

  • Choose a set of ~ 30 cancers
  • Find the lifetime incidence for each of those cancers
  • Independently, find estimates for the number of stems cells and the rate of stem cell division in each of the tissues from which those cancers arise, and then use these to calculate the lifetime number of stem cell divisions in that tissue
  • Find the correlation of the lifetime incidence of the cancer with the lifetime number of stem cell divisions in that tissue

In order to make sense of their results, I decided to reproduce their analysis. First, I manually copied and cleaned up the formatting of the data from their pdf (!) [1]. One of the first things I did after loading it into R was to plot it:

This doesn’t look like the plot in their paper, but I realized that was because I hadn’t taken the logs. Once I do, it looks way more similar (although I still flipped the axes):

log log

The Pearson correlation coefficient is much higher in the log-log data (0.80) than in the non-transformed data (0.53). The Spearman correlation coefficient is the same in both (0.81), since the ranks don’t change. This is a good example of how the Spearman correlation is more robust.

So what does figure one mean? I interpret it as showing that there is a positive relationship between the rate of stem cell division in a tissue and the rate of cancer in that tissue (variance explained of ~ 0.66). This then suggests that cancers arising from tissues with a higher rate of stem cell division are more due to the “luck” of whether or not one of those divisions happened to include a mutation, as opposed to a genetic predisposition.

Of course, the rate of stem cell divisions and/or mutations could still be influenced by an environmental factor, but it’d be less likely that any hypothetical environmental factor would affect the risk of mutation in all stem cells in all tissues at the same rates.

However, this does not suggest that “66% of cancers are caused by bad luck”, for many reasons, including the fact that the residuals are not weighted by the proportion that each individual cancer makes up of total cancer rates.

The next section of their paper ranks the cancers based on a score meant to quantify the amount of cancer that occurs above and beyond what you’d expect from the stem cell division rate in that tissue, and then clusters the cancers into two groups based on this ranking.

It’s slightly troubling that they call this one-dimensional k-means clustering “machine learning”. K-means clustering is in fact commonly considered a machine learning method, but at one dimension it reduces to a single breakpoint estimation, which is not quite in the spirit of high-dimensional ML connotation space. [Editor’s note: this was slightly edited for clarity/correctness].

All in all, while the data being in pdf format was slightly annoying and I wasn’t 100% sure of the implementation details of their k-means clustering [2], I was able to reproduce pretty much all of their results. This was a useful exercise for me and it is a good sign for the longevity of their insights. Simplicity is the ultimate sophistication.

1: My code for this post is on github.

2: Since I haven’t read through every line of the paper with a hyperdontic comb, let me state that this is probably my fault for missing something.

Is Alzheimers just normal aging? Part 1

A paper by Robinson et al provides a data point against, evaluating the post-mortem brains of around 150 people 90+ years old. They split their cohort three groups: those with dementia, those with cognitive impairment but no dementia, and those without cognitive impairment. Here is their main result table.

robinson et al

Thus, in their study, we can see that

a) there are pathological differences between dementia and other cognitive impairment, and

b) of the pathology types that they measured, there are no clear pathological differences between cognitive impairment and non-cognitive impairment.

To me this seems to be pretty good data emphasizing the standard theory that Alzheimers is distinct from normal aging. And that normal aging can also present with cognitive impairment, although these individuals can still convert to Alzheimers later in life.

How might nucleosome integrity affect aging?

Attention conservation notice: Rampant speculation on a paper outside of my field, which is unlikely to be simultaneously both new and interesting.

Using a zebrafish egg extract system, Ziehurt et al have shown that in the absence of nucleosomes (ie, in the “DNA beads” condition, as opposed to the “nucleosome beads” condition), there is no nuclear pore complex (NPC) formation. For example, these DNA extracts lack lamin B3.

Without nuclear pore complexes, cells can’t shuttle necessary proteins and other biomolecules in and out of the nucleus, which is… not good for the nucleus. This could be adaptive for the organism, though, if there is foreign or damaged DNA in the cell.

How does this relate to aging?

1) In general, nucleosome-free regions of the nucleus become more common in aging.

2) A different lamin protein called lamin A, which also acts at the NPC, is mutated in the most severe aging accelerating disease I’m aware of: progeria.

3) So, it’s reasonable to ask: in normal aging, does the loss of nucleosome integrity (because of, perhaps, the hand-wavy “accumulated damage”?) lead to lack of NPC accessibility?

From here, another reasonable question is whether you could find a biomarker of NPC integrity from some sort of high throughput method, such as a DNA sequencing method, that would allow you to measure nucleosome integrity easily across aging and intervention states.