Since 2003, I’ve been saying that long telomeres are a path to long life. The opposing view says that nature allows our telomeres to shorten to protect us against cancer. Up until this spring, there has been little evidentiary support for the cancer theory. Now, a major new study uses genetics to argue that longer telomeres increase risk of cancer as much as five-fold. The study contains many statistical checks, but I’m going out on a limb to say I think the experts have made a conceptual error.
Up until now, epidemiological studies in humans and lab studies in animals have shown consistently that shorter telomeres increase risks for all the diseases of old age. People’s telomere length tends to decline with age, but among people of the same age, those with shorter telomeres tend to die sooner.
The new study finds a very different conclusion: that shorter telomere length leads to much lower risk of cancer, while longer telomere length leads to slightly lower risk of heart disease. Put these two together, and you predict pretty much the same life expectancy for people with long telomeres and short telomeres.
The new studies are based on genetics and account for telomere length only indirectly. Nevertheless, it is claimed, they are more reliable than the old studies (based on direct observation) because they are able to eliminate a statistical anomaly that (they claim) is super-important.
I believe the new study is actually less reliable, and that we should believe the more direct studies like the ones I have reported here in the past. My reasons are that
- The previous studies are straightforward, direct correlations. Methodology in the new study relies on very small differences in telomere length, tiny differences that are lost in the noise and very difficult to detect.
- The new studies require very large implicit extrapolation that is not necessary in the old studies. The 50 to 1 extrapolation is very speculative, and it magnifies the noise along with the signal.
- It is likely what these new studies are seeing are actually direct effects of genetics on cancer risk. Even very small (direct) effects of genotype on cancer would appear in their methodology as though they were huge (indirect) effects of telomere length. This is what I believe is happening, and why I don’t trust their results.
I may be wrong about this. I’m questioning seasoned experts in the field based on my general knowledge of statistics.
Two years ago, I reported on a Danish study linking short telomeres to higher mortality, especially heart disease (CV). I took this as clear proof that telomere length was not just a marker of aging but a cause. The implication is that you can live longer by adopting lifestyles and taking supplements that extend your telomeres.
The core of my argument, based on the the Danish study, was this:
- Impact of telomere length on mortality, raw data: 3.38 (meaning that the 10% of people with the shortest telomeres were dying at a rate 3.38 as high as the 10% with the longest telomeres)
- Same calculation, corrected for age: 1.54
- Same calculation, corrected for age and all other hazard variables: 1.40
Conclusion: This demonstrates that age is the biggest factor in mortality, and telomere length is second, with a strong effect, independent of age. All the health variables together have only minor effect compared to age and telomere length.
The Danish study did a multivariate analysis, also called ANOVA. This is a statistical technique designed to separate out the factors that contribute to an outcome (in this case, mortality) and assign percentages of causality. What their analysis revealed was that the strongest cause of increased mortality is age itself, and that telomere length comes second. Everything else, from smoking to depression to a history of infections, is much less important than age and telomere length. I interpret this to say that short telomeres are probably a direct cause of increased disease risk.
A popular theory is that the association of short telomeres with higher mortality is only incidental. Stresses, infections, smoking, etc. cause both shorter telomeres and higher mortality. But these are separate pathways. It is not the shorter telomeres that are causing higher mortality, but short telomeres happen to be associated with higher mortality because both are caused by various stressors in a person’s past. If this is true, you can’t improve your odds of living longer just by extending your telomeres.
But I believe that the Danish study disproves this theory. If the stressor theory were correct, then the Danish analysis would have found that the relationship between stressors and mortality was stronger than the relationship between telomere length and mortality. In fact, they found the opposite.
The New Genetic Study
The result reported by the new study is that longer telomeres creates a very much higher risk of several common cancers. On the other hand, longer LTL (=leucocyte telomere length) protects against heart disease. The protective effect for heart disease is much smaller, but many more people die of heart disease than of these particular cancers. The result is a wash. Longer LTL is neither a net benefit to health nor is it a net risk. People with longer and shorter LTLs have similar overall mortality risk, about the same life expectancy.
Disease | Odds Ratio |
Glioma | 5.27 |
Ovarian cancer | 4.35 |
Lung cancer | 3.19 |
Neuroblastoma | 2.98 |
Bladder | 2.19 |
melanoma | 1.87 |
Testicular | 1.76 |
Kidney | 1.55 |
Endometrial | 1.31 |
Basal cell skin | 1.22 |
Breast Cancer | 1.06 |
Heart disease | 0.78 |
“Odds ratio” refers to a person’s probability of contracting the corresponding disease. For example, the first line means that people whose telomere length is one standard deviation (1 sigma) longer than average have a risk of glioma 5 times greater than people who have average telomere length.
This result gains credibility because it is exactly what the theory would predict. Nature has optimized LTL by compromising between two risks. If the average LTL for our species were longer, then we’d get more cancer. If it were shorter, we’d get more heart disease. The reason there is so much variation among the population, people with much longer and much shorter telomere length, is that it doesn’t matter very much.
So here is agreement between experiment and theory, a tidy situation that scientists like to see. What is more, there is a widely-held belief that the methodology of the new study is more reliable than studies in the past that are more direct and simpler. Nevertheless, I’m about to offer my opinion that the previous studies were right, the theory is wrong, and, in fact the design of the new study is seriously flawed.
This was the latest and far the largest in a series of GWAS studies going back four years [ref, ref, ref]. GWAS stands for Genome-Wide Association Study. The idea is to work around life experience variables that might create a correlation without a causal connection. In the present case, the target is to detect any causal relationship between leucocyte telomere length (LTL) and various diseases, while filtering out associations between LTL and disease risk that might be incidental, as described above. The researchers looked for small genetic differences (called SNPs) that are linked to telomere length. These vary from one individual to the next, and they persist through a lifetime. The next step is to compare numbers of people with a particular SNP variant among those who have the disease and those who don’t have the disease. Are people who have the SNP associated with longer telomeres more or less likely to develop the disease? From the answer to this question, they infer a causal relationship, not between SNP and the disease but between telomere length and the disease.
Observational studies look for a direct relationship between LTL and disease. GWAS studies look for an indirect relationship between SNP and LTL, SNP and disease. The indirect study is widely considered to be a more reliable indicator of causal connection than the direct study. Why?
“Mendelian randomization studies are less susceptible to confounding in comparison to observational studies…Given the random distribution of genotypes in the general population with respect to lifestyle and other environmental factors, as well as the fixed nature of germline genotypes, these results should be less susceptible to confounding and reverse causation than those generated by observational studies.”
The reasoning is that people have their genomes for their entire lives, independent of how they live, what they do, what they are exposed to. By working with the genome, the statisticians can be sure to eliminate the standard objection that (for example):
- Stress directly decreases LTL
- Stress directly increases risk of disease
- Therefore, short LTL will appear to be linked with disease, even though short LTL doesn’t cause disease.
Problems with GWAS studies
But the GWAS methodology also introduces new problems of its own. The main problem is that the statistical sensitivity of the study is seriously reduced. This is because the relationship between SNP and LTL is very weak. All sixteen SNPs together constitute a very small factor among many larger ones that create difference in LTL between one person and the next.
“The selected SNPs correspond to 10 independent genomic regions that collectively account for 2% to 3% of the variance in leukocyte telomere length”
And of course, very few people have all 16 SNPs going in the same direction. The study is forced to work with people who have, for example 10 positive SNPs out of 16 compared to others who may have 5 positive SNPs out of 16.
Their LTL is really quite close together. To compensate for this, the statisticians divide by a small number to extrapolate outwards. For example, the difference between typical people in the study is about 1/20 sigma*. And the difference between risk of glioma (brain cancer) for these people is only about 0.08** . But the difference is reported as “what would have been the risk of brain cancer if the difference had been not 1/20th but one full sigma. They extrapolate exponentially, so the conclusion comes out quite startling: They claim that people with 1 sigma of extra LTL have 5 times greater chance of getting brain cancer.
What they find: people with 0.05 sigma extra LTL have 1.08 times the risk of getting brain cancer.
What they report: people with 1 sigma extra LTL would have (by extrapolation) 5 times the risk of getting brain cancer. |
They conclude that there is a large effect of telomere length on cancer, but they do this by the following reasoning:
- There is a small effect of these genetic variations on telomere length.
- There is a small effect of these genetic variations on cancer risk.
- Dividing the small by the small, they conclude: if the mechanism for these genetic variations affecting cancer is mediated by their effect on telomere length, then the effect of telomere length on cancer must be quite large.
I’m sorry to belabor this, but it’s important, and it’s hidden in the methodology. People who do these studies know that an odds ratio (OR) of 1.08 means nothing. The state of the art in epidemiology is rarely able to attach meaning to odds ratio that is close to 1. It is lost in the nosie. But an OR of 5 is something easy to see. It stands out from the noise and is easy to detect.
The description of the methodology in this study hides the fact that they are working with ORs less than 1.08 and extrapolating exponentially outward to make the ORs look very large and significant.
What I think is really going on
The study finds a large and consistent result that demands some explanation. I’m claiming that the explanation they offer (in terms of telomere length) is wrong. So why do I think they get the results that they did?
A few of the sixteen SNPs that are considered in the study correspond to slight variations on the form of the telomerase molecule. I’m guessing that these mutated forms of telomerase cause an increased risk of cancer. The increased risk doesn’t have to be large. As in my example above, the increased risk for brain cancer would have to be just 8%, and the increased risk for lung cancer (more important because it is more common) only 6%. Because of the extrapolation by an exponent of 20 that is implicit in their methodology, these small effects would be reported as though they were odds ratios of 5 (for brain cancer) and 3 (for lung cancer).
Another possibility is that one or more of the SNPs happen to be more common in a segment of the population that is prone to cancer, for whatever reason. It may be that a particular SNP is more common in an ethnic group that has high smoking rates, or that is prone to melanoma because of lighter skin, or has a diet and lifestyle that leads to a slightly greater risk of cancer. For example, it is known that people of African extraction have SNPs associated with longer telomere length, and they also have higher risks for many cancers, including lung and [ref]. (Africans have lower risk of glioma, so the correlation goes in the wrong direction for this particular example.) At the risk of beating a dead horse, I emphasize again that even a small increased risk would be magnified by the extrapolation that is implicit in the methodology of the GWAS, and appear very large and scary when misinterpreted as an effect of telomere length.
GWAS is also referred to as “Mendelian randomization studies” because they depend very much on the assumption that different SNPs are randomly distributed in the population. Of course, this assumption is not literally satisfied. How significant is the deviation from random distribution? I will be investigating this question, and I’ll let you know what I find.
The Bottom Line
There is a sharp conflict between the new GWAS results [Haycock, 2017] and the observational results [Rode, 2015] reported two years ago. They can’t both be right. If the GWAS results are as Haycock claims, there would have been glaring increases in cancer risk that Rode could not have missed. If Rode is correct, then the methodology of Haycock must be flawed.
The reasoning in GWAS studies depends on a huge extrapolation. I am saying it is more likely that the effect of genetic variations on cancer risk is direct, not (as per Haycock’s assumption) mediated by telomere length. It could be that a very small direct effect of one of these SNPs is reported as though it were a large indirect effect, working via telomere length.
For now, I’m sticking with my previous counsel: Lengthening telomeres is a viable strategy for improving health and longevity. If you take supplements that promote telomerase, you are not adding to your cancer risk. Because of the large net benefit, lengthening of telomeres should be a major target for medical research.
But as I said at the outset, I am criticizing the new study from the outside, and it is quite possible that I have misunderstood the methodology. I have sided with the direct observational studies and I have been skeptical of the GWAS studies, but it may be that the consensus in the field is correct, and that GWAS studies really are more reliable indicators of causality.
I intend to get to the bottom of this, and will report my findings in future columns.
__________
* Sigma is a standard deviation of telomere length in the population at large. If you know what that means, that’s great; if you don’t it doesn’t matter to the logic of what I’m saying.
** Disease risk is typically reported as an odds ratio. In this case, 0.08 would mean that, in their raw data, people in the study with the longer LTLs had a risk of 1.08 times as great as people with shorter LTLs. You get to 1.08 not by adding 1 but by raising e to the power 0.08.