Biogerontologists Score - Thin is Healthy!

Friends, it's coming home. We all knew this will be the century of biogerontology, because of the unceasing demographic change, and because of the force of Taeuber's paradox. These factors mean we are the only hope for genuine medical progress in developed nations (1). We all knew our time would come, but no one knew when.

Given the just recently concluded Tokyo olympic games, let's run with a sports analogy. The trophy is coming home. If research were some sort of game, then the results we will discuss, would be akin to winning the world cup. One high impact win does not make you the best athlete, but it is a beginning that suggests a certain talent.

I've maintained for years that "that low body-weight in healthy people is suggestive of a [caloric restriction] CR-like state (i.e. mild CR)" (In defense of being underweight, August 23, 2017). I also agreed with the reasoning that, at worst, we are looking at two orthogonal phenomena. Genetically-programmed BMI (which could associate with lifespan in either direction) and a reduction of bodyweight (and hence BMI) below the so called ad libitum steady-state, which should be beneficial to longevity (3). This duality complicates the interpretation of epidemiologic data.

The truth is, no one knew at any point in time whether low or high BMI is beneficial to longevity. It was an ongoing research question and everyone who claimed to know, was being hyperbolic and arrogant (I am sure I also claimed to "know" the answer, partly to shift the overton window). Thus, it made me furious whenever I heard well-intentioned people parrot the slogans "thin is bad", also promulgated by well-intentioned, but ultimately clueless, doctors and health-professionals. As if this were settled science! Doctors are NOT experts in diet, longevity or BMI research. They know a lot about different things, but rarely specialize in current research topics.

It made me angry because it was part of overt discrimination against the skinny and underweight, and because it was actively undermining biogerontologic research and outreach. Because how could CR work if being thin is unhealthy! Or so the reasoning went...

On to the evidence
Now a series of papers utilizing a novel epidemiologic technique (2c) called mendelian randomization provided evidence in favour of low BMIs (2). Let's unroll them in chronological order, since at first it did not look so good for us.

Figure: Both actual (A) and genetically-predicted BMI (B) is associated with all-cause mortality in the UK Biobank study (2). Wade et al. 2018. 

This figure is pretty unnerving, but, damn, look at those confidence intervals. It appears that being thin is still bad for you, albeit the "new" optimum is around BMI 23. Did the epidemiologists win? Whereas the aging researchers, and those who holistically evaluated the whole body of literature, made the wrong call?

Maybe the confidence intervals do suggest that something is going on here. One of my first thoughts was, did they really adjust for confounding? I do not think you can handwave away all confounding in Mendelian randomization, even though it helps a lot. Let's say, you are at a low BMI due to another disease that healthy people would rarely have (e.g. frailty due to premature cancer), it may well be that a drop in BMI will be unfavourable; but this may not apply to the 10-20% of the population that has a reasonable diet and doesn't smoke. Important to respect that distinction.

So, as per usual, I followed the google scholar citations to look for updated evidence and comments. Then I noticed a re-analysis of the data including not only the initial Biobank sample (n > 360k) but also the HUNT study (n > 50k), which is smaller while having longer follow-up.

Figure: Ooops, we've got a little bit of confounding going on here Sun et al. 2019 (2). It is particularly noteworthy that the confidence intervals for low BMIs are really tight, but not for high BMIs.

This is glorious: "Subgroup analyses by smoking status, however, suggested an always-increasing relation of BMI with mortality in never smokers and a J shaped relation in ever smokers. [I do not like using the word J, because it is confusing. What is really the difference between an L, J or U when looking at spline curves?]"

The key ways to salvage classic BMI epidemiology have always been 1/ long follow-up (to reduce reverse causation), 2/ exclusion of the elderly from analysis (to eliminate risk factor reversal [6]) and 3/ exclusion of smokers (because these people are on a suicide mission that we biologically don't quite grok, so better to let them go their own way.) Therefore I am not totally surprised that applying one of these - exclusion of smokers - also strengthens the BMI hypothesis in the context of Mendelian randomization.

What could be the reason for the smoker bias? It seems likely that respiratory disease and cancer is the explanation, maybe because low bodyweight could increase mortality from these diseases while decreasing their incidence. Since smokers have a much higher incidence to begin with, this might shift the risk-benefit ratio towards the impact on mortality. More on the matter in our discussion of Yu et al. 2021 (9) below.

What is mendelian randomization and is it any good?
Some genes may affect certrain traits of interest. Imagine a gene that controls the generation of vitamin D precursors in the skin. If we can determine the expected expression of the gene e.g. by knowing that certain polymorphic variants of the gene show higher actvitiy or expression then we can calculate a score for the trait as well. Whereas the production of vitamin D will be biased by behaviour, for example when sick people avoid going out, the same cannot be true for the segregation of genetic alleles (variants). You inherit them more or less randomly and if it is not random the bias is unlikely to be in the same direction for each allele, and unlikely to be in the same direction as in classic epidemiology. Using the mendelian randomization approach we could now study people with genetically higher vitamin D levels and ask if they are healthier or not, we can do the same with variants that affect BMI. While this approach avoids many pitfalls of classic nutritional epidemiology it engenders others. Not only is it difficult to measure variants, but their effects may also be insufficiently characterized (genetic pleitropy) or the genes themselves may be involved in the disease being studied (2d).

We could think of a little toy model. Variants that promote hunger or tissue accretion are clearly anabolic, the biggest paradigm in current aging research is that anti-anabolism is healthy and extends lifespan, thus any variant that increases BMI could be harmful independently of BMI via some other anabolic signaling. (Not sure this is a relevant distinction, but it's just to give you an idea.)

It would be great if we had enough data to know if mendelian randomization is clearly superior. One would need to compare the gold standard, controlled trials, with the two type of epidemiology. So far, we know that mendelian randomization for LDL produces data that agrees with controlled trials (but in this case so does classic epidemiology). Depending on how we interpret the VITAL trial (8), here the mendelian approach is overall in much better agreement with the data than is standard epidemiology (i.e. no benefit on cancer, 2c).

There is no obvious explanation for increased mortality at low BMIs
As far as I can tell, no one has ever provided a satisfying mechanistic explanation why low BMIs should kill you. Yes, lower bodyweight could increase the risk of surgical complications and traumatic injury in the wake of accidents, but these should not be drivers of mortality in the middle aged (or across your whole life). Additionally, immune function could be compromised, although this is again not a major cause of death at this age and neither is frailty or osteoporosis. However, these issues could explain why lower BMIs are harmful to the elderly in some studies. So called risk factor reversal has been observed before. Some risk factors in the young, are protective in the elderly. Thus it makes sense to stay very lean, as long as you are healthy and young, but to bulk up before surgery and when you are older (4).
Even if they exist, small, speculative side-effects of being too lean and underweight should be offset by the improvements in validated risk factors that have been documented after weightloss (e.g. LDL, blood pressure, CRP). Absence of evidence is weak evidence of absence, so obviously this is more of a side-note than an argument.

CRON - caloric restriction with optimal nutrition
Let's repeat this, CR is NOT the same as having a low BMI although the two are clearly related. It is speculated that people having a low BMI are more sensitive to some unhealthy lifestyle choices. For example, if you eat little, you may fail to reach the recommended dietary allowance for minerals, vitamins and phytonutrients (which may not scale linearly with body weight), hence CRON was invented. If you are skinny you have to pay attention to a healthy diet. Furthermore, due to the somewhat catabolic state / low total bodymass, both dieters and skinny people may be susceptible to lack of exercise, as this would super-charge muscle loss.  Put another way:

A recent observational study reported a J shaped association between BMI and all cause mortality and a more profound U shaped association between lean body mass and mortality,(27) suggesting that the higher risk of all cause mortality in the lower range of BMI might be explained by low lean mass rather than low fat mass. Low fat-free mass has also been reported to associate more strongly with the risk of all cause mortality than low fat mass.(28)

Other recent studies
Back in 2017 I made the point that low adiposity shows a more favourable association with mortality than does low BMI. On the whole, it still stands.

Let us look at a meta-analysis of 72 observational studies across 2.5 million participants that was published in 2020 (7). First, I will focus on waist-to-hip ratio (WHR) because it is a well-established marker with the most available data. The association between mortality and WHR looks pretty linear or "J-shaped", but there is no increase at low WHR ratios (so it is a J with a totally flat bottom part, please let's just forget this terminology).


Figure: where is your J and U now, did you lose it somewhere? The infamous U-shaped association betwen mortality and BMI is less pronounced with adiposity, especialy so with waist-to-hip ratio (WHR).

To be fair, some other measures of adiposity like waist cirumference are less consistent with lower-is-better (7), although, even here in 4 out of 5 subgroup analyses there is no increase mortality with low waist circumference. 

The last paper I want to discuss (Yu et al. 2021), before this post gets entirely out of hand, is this observational study from Taiwan that caught my attention (9). On the one hand, it is an interesting study because - let's face it, we Caucasians are fat - it allows us to study people who are really thin. On the other hand it is problematic due to Taiwan's meteoric rise in GDP and living standards. Why is that an issue you say? In poor countries being thin can be a sign of poverty and malnourishment. It is not entirely clear how many people in this study may have acquired their "thin phenotype" in this way.

Overall the paper finds that being very thin, as defined by e.g. low waist circumference, is healthy but not being extremely thin. The twist is that the optimal waist circumference for women was between 65-75cm, which is equivalent to the lowest observed circumferences in the above meta-analysis. Women with below average waist circumference, at what we consider XS (around 65cm), were still close to the mortality optimum, whereas only women below ~62cm had an increase in mortality with the range easily extending to 40cm.

We can understand just how thin people are in this study if we compare to the official statistics: "In the United States, the average adult man has a BMI of 26.6 and the average adult woman has a BMI of 26.5." In contrast, in the Taiwanese study the female BMI was 21.6.

A problem specific to this study is that the observed population was young with a mean age of ~35 whereas the follow-up was only 15 years, so deaths in this study were likely not due tage-related diseases. A related problem would exist if we measured adiposity in 55 year-old people, because then, while we would be able to measure incidence of age-related diseases more reliably, we would not know if adiposity at age 55 is reflective of adiposity at a younger age. We can build a neat post-hoc story explaining how this biases the Taiwanese data. Overall, tuberculosis and respiratory death used to be more common in Taiwan/China than in Europe. This is a relatively common disease that can kill young and middle-aged people. Low adiposity somehow predisposes to respiratory death, as was shown in earlier studies and confirmed here. For example, cancer/CVD/respiratory mortality hazard ratios at the lowest female waist circumference were 1.05/0.95/2.78.

Based on the above shortcomings we can also come up a with a better design, which may help us develop an intuition for this problem. More on this topic can be found in footnote (10). A near perfect observational study would measure time-averaged adiposity between ages 20-40 and age-related morbidity and mortality from age 60+ onwards. That way 1/ the population is old enough to show diseases of age, 2/ the measure of adiposity is accurate since it is averaged over time and fits nicely into the critical window during which CR is most robust (organism is neither immature nor senescent), 3/ we do not measure mortality in the 40-60 window so that we exclude reverse causality (those who died in their 50s might have been sick and thus thin in their 30s)

Summary and future directions
Truly, this is not last word on the matter, but it remains a clear win for the biogerontologist's heuristic. Summing up, it seems that if you are young (<60yo), healthy and do not smoke then being very thin is no problem at all and having a low BMI could be even beneficial. As was predicted by biogerontology and the CR literature.
Sadly, the notion of low-BMI-is-harmful is so ingrained that no amount of research will change the minds of laypeople and average doctors, as even the very editorial written about the Sun paper immediately mischaracterizes their work. This is some of the dumbest stuff I have ever read in the prestigious BMJ. "Overall, public health recommendations that people should aim for a BMI within the normal weight range should remain unchanged.", is still by far the nicest thing from the editorial. Please see (5) for some entertainment.

Why do we care about weight-loss at the extreme tail end of the distribution if most people are fat? I think it bears repeating, if I have not made this clear yet. We can easily fix the obesity pandemic if we ever find the political will to mitigate the obesogenic environment. Aging, in contrast, is neither politically nor biologically easy. We are interested in low BMIs and low adiposity because these phenotypes can inform our opinion about the efficacy of CR and the safety of deploying CR-mimetics, even if this analysis is an imperfect tool for that task.

In the future I hope to see follow-up studies using exactly the same methodology as in the Sun paper, but with exclusion of smokers as a pre-defined primary mode of analysis (see also [5, 6]). Finally, I would like to note that methylation aging clocks and twin studies could be another valuable tool, as shown by the below preprint. For now, I would say the GrimAge methylation clock is too novel a tool to be a serious contender in the BMI/weight/adiposity debate, but I want to mention it because this is where the field is heading. I think the picture tells the story by itself.


"we found a linear relationship between BMI and age acceleration, with underweight individuals displaying the lowest amount of age acceleration, and an incremental increase in the amount of age acceleration through each subsequent BMI category. This is noteworthy given that most studies of BMI suggest that underweight individuals are at higher risk of disease..." (Lundgren et al. 2021)


References and notes
Epistemic status. I'm  stealing the idea of "epistemic status" from the lesswrong forums, because it allows me to actually write instead of trying to produce the perfect article. The notes also serve to provide additional discussion, explanations and critical remarks so I do not feel like I am being unfairly one-sided. I am an aging researcher but not an expert in Mendelian randomization, so I may be misinterpreting the data, although this seems unlikely to me.

Acknowledgments
Just recently I read a post (was it on stackexchange?) suggesting that researchers should write more narrative prose explaining the process of research. This would be something best put into footnotes and I kind of feel like sharing my thoughts today. Inspiration to write this post came from browsing the less wrong forums. No, they are not a reliable source of medical information, but they do attract lots of interesting (sometimse overly libertarian, contrarian) debate. I've recently noticed the rise of Mendelian randomization epidemiology, but I did not have the time nor inclination to update my personal notes on BMI with this new tool until I saw a somewhat unrelated less wrong post linking to an older study: "Variation in smoking and systolic blood pressure had the strongest causal life-shortening effects (5.3 and 5.2 years, respectively) [remember that BMI could also affect SBP and insulin, at least in theory], followed by fasting insulin, body mass index and CAD, while years of education showed by far the most beneficial effect (4.7 years)..."

1. Simply put, aging is the driver of most important diseases (cancer, cardiovascarul disease, diabetes, stroke, frailty, etc.) and the most effective way of slowing them all, is to slow the biological process of aging! Please read The Symmetry Error - are the benefits of healthspan extension offset by extended disease-span? (July 11, 2021) if you are still confused.

2a. Wade, Kaitlin H., et al. "BMI and mortality in UK Biobank: revised estimates using Mendelian randomization." Obesity 26.11 (2018): 1796-1806.

2b. Sun, Yi-Qian, et al. "Body mass index and all cause mortality in HUNT and UK Biobank studies: linear and non-linear mendelian randomisation analyses." bmj 364 (2019).

2c. Larsson, Susanna C. "Mendelian randomization as a tool for causal inference in human nutrition and metabolism." Current Opinion in Lipidology 32.1 (2021): 1-8.

2d. Davies, Neil M., Michael V. Holmes, and George Davey Smith. "Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians." Bmj 362 (2018).
NB: a fantastic primer, some interesting notes on the shortcomings of mendelian randomization: " In the case of C-reactive protein and coronary heart disease, the Mendelian randomisation estimates suggest that targeting CRP is unlikely to be a viable therapeutic target for the prevention of coronary heart disease". Oops.

3a. We biogerontologists thought that weightloss should be beneficial because caloric restriction (CR) extends lifespan across a wide-range of species. The attentive researcher also knew that all surrogate studies in humans favoured weight-loss and low BMIs (e.g. blood pressure). To be fair, the evidence in favour of CR is much more complicated and nuanced that I am letting on, but still favours the idea that CR is beneficial. See the below paper for some up to date discussions

Mulvey, Lorna, et al. "Strain-specific metabolic responses to long-term caloric restriction in female ILSXISS recombinant inbred mice." Molecular and Cellular Endocrinology (2021): 111376.

3b. What is ad libitum? Regarding ad libitum states. Maybe think of it this way, in a very goofy but illustrative way: an elephant or a turtoise has a higher programmed BMI (or body weight) than we do, but it that doesn't mean we are on a diet, or in a CR-state, relative to an elephant. A CR state can only be measured relative to the hypothalamic / neuronal set-point of the individual.

3c. if BMI is orthogonal to the CR phenomenon then the BMI literature could never refute the CR hypothesis, but it would be "safer" if the BMI literature aligned with the CR hypothesis.

4.  It makes me feel a bit "American", as if I am afraid of getting sued or something, when I say this, but I will say it nonetheless, at least this once. I am not your doctor, nor am I a doctor to begin with. Don't do any dumb shit.

5. "These analyses support a causal association between lower BMI and higher mortality below a BMI of about 20-22 [what the flying fucking?]....finding little or no J shape in the association in never smokers but a more pronounced J shape in ever smokers [this is why I dislike the word J-shaped, it is only J-shaped in non-smokers in the sense that the bottom part of the J is entirely straight]...a best guess from the analysis by Sun and colleagues is that the BMI with the lowest mortality might be around 24." No, no, no! You cannot say it and if you can say it, then it is entirely meaningless. The paper disaggregated smokers and non-smokers for a damn good reason. This BMI applies only to an average person that is both a smoker and non-smoker at the same time -- Schrödinger's smoker?

Now let us deconstruct the most important lie:


The point is, they could have criticized the conclusions without lies and distortions (see, I did it in [6]), but they choose not to.

Bradbury, Kathryn E., and Benjamin J. Cairns. "Understanding the relation between BMI and mortality." (2019).

6. Normally exclusion of the elderly strengthens the BMI hypohesis but here the adjustment for age produces very odd results (see the supplementary), perhaps due to a relationship with smoking. This certainly suggests that more research is needed. One could argue that the authors played pick-and-choose with the secondary analysis which they put into the main body of the paper.

"Similarly, the BMI-mortality relation was J shaped or decreasing in younger participants (<65 years), but it was generally increasing in older participants (>65 years). This is not consistent with results of several observational studies in older people, in which overweight categories were associated with a lower risk of all cause mortality.2223 There is an intrinsic limitation in separating age and smoking status, as deaths before age 65 were more common in ever smokers."

7. Jayedi, Ahmad, et al. "Central fatness and risk of all cause mortality: systematic review and dose-response meta-analysis of 72 prospective cohort studies." bmj 370 (2020).

Waist circumference in men is shown as an example that is a bit less consistent with lower-is-better. How do we interpret those graphs given how many different corrections the authors performed? Ultimately, we are asking two things. 1/ Does CR work and 2/ is healthy weight-loss good? Healthy weight-loss would maintain muscle mass while cutting adiposity. The best approximation that we can get of this from "steady-state" studies - i.e. studies measuring the varaible of interest, weight/adiposity, at a single time point - may be provided by BMI-adjusted adiposity because otherwise the co-linearity of BMI and adiposity would unduly influence the results. Conceivably, at low waist circumferences there may be a sizeable number of people who reached the lower levels through muscle loss which might be reflected in a lower BMI (as muscle is heavier than fat).


8. VITAL suggested reduced cancer mortality but not incidence in post-hoc analysis. Make of it what you will.
Also regarding mendelian randomization not being infallible see:
Hegele, Robert A. "Editorial comment: when Mendelian randomization goes astray." Current Opinion in Lipidology 32.2 (2021): 79-80.

9. Yu, Tsung, et al. "Adiposity and risk of death: A prospective cohort study of 463,002 adults." Clinical Nutrition 40.4 (2021): 1932-1941.
NB: long follow-up tends to attenuate the increased mortality risk in this study for both BMI and waist circumference, but smoking does not, which seemed an issue in (2, 5, 6). Whether this is a biologic effect remains to be determined by future studies but interesingly at least one factor of the trifecta usually works to attenuate the excess mortality in most observational studies (i.e. smoking, follow-up, age).

10. In some cases CR can lead to an early life increase in mortality. If we saw the same in humans, e.g. because CR lowers age-related conditions but promotes some other diseases independent of age, it would make the epidemiology very hard to read. Overall we may see a net extension of healthspan even if we see an acute increase in mortality as in the Taiwanese study.





Kommentare

  1. Cell Cynergy Nutrition was created with one purpose in mind, to help you to maximize your Health span. Health span is the length of time a person lives in optimal health and free from serious disease. Lifespan, in contrast, simply refers to the number of years a person lives.

    AntwortenLöschen

Kommentar veröffentlichen