The Death Of Good Cholesterol

Briefly

There were always two types of cholesterol, the good and the bad. Until now. A large new study tells us that good cholesterol might have been an impostor. That's food for the media types. For those who think before they type, the real news is that we are finally getting closer to uncovering the impostors. Thanks to the genetics revolution which seems to be paying off in an unexpected area.  

HDL - The Knight in Shining Armor

In the cholesterol universe there are two camps: good cholesterol, also known as HDL, and bad cholesterol, often referred to as non-HDL cholesterol. The latter comes in a variety of flavors, of which LDL is the most prominent and best known. From many large observational studies we know that high levels of LDL and low levels of HDL associate with an elevated risk for heart disease and stroke. Certain limits have been derived from these studies, above which your LDL shouldn't rise and below which your HDL shouldn't fall. The magic level for HDL is 60 mg/dL blood. Above that limit, we are assured, HDL will even offset some other risk factor, such as age or being of the male persuasion. Given that a large percentage of people fail to achieve these desirable levels, researchers have been eagerly sourcing for pharmaceutical means to increase HDL. Now a new study tells us, that HDL might have to be stripped off its White-Knight title, much for the same reason as "Dr." Karl-Theodor zu Guttenberg, the former German defense minister, had to be stripped off his doctorate last year: for being an impostor.

Epidemiology 101

If you have been following biomedical research for a couple of years, you will have noticed that results are often conflicting. So, you might discount the findings of one study if hundreds of others come to a different conclusion. Only in this case you should pay closer attention, because what Voight and colleagues have produced strikes at the foundation of how we do research in epidemiology, the science which studies the health of populations [1].  To appreciate the gravity of the situation, I need to familiarize you with a basic concept of epidemiological studies: Confounding. I'll use a very simple and hypothetical example.  Let's say we are interested to know the causes of health and disease in children in the hypothetical and impoverished state of Maladipore. The figure to the left represents our astonishing finding that children growing up in a household which owns a TV are significantly less likely to die during childhood than children growing up without the boob tube. The correlation between TV ownership status and survival are very strong and compelling.  On the face of it we could now recommend the prime minister to improve the health of the nation by simply installing a TV in every household in which there are children. If we know that this is nonsense, we take our epidemiology tools and look for another factor which has an influence on TV ownership AND on survival rate.  And so we discover that wealth is this third factor. We call it a confounder. Wealth has confounded our original finding because the wealthy can afford a TV and they can also afford medical care and immunization for their children. Whereas the inability to buy a TV certainly reflects the inability to buy medical care, too. When we repeat our analysis of the data, which we gathered during our observational study, we find that the link between TV ownership and survival disappears once we bring the third variable, wealth, into the equation. Clearly, providing every household with a TV wouldn't have reduced the rate of child deaths. Greater wealth however will. In the case of Maladipore, common sense is all it takes to suspect and find the confounder. In real life it is almost never as simple. When we find an association between cholesterol and heart disease, then we typically have some idea about the way cholesterol might contribute to heart disease. At that stage our ideas are merely hypothetical. The classic way of investigating them is through clinical trials in which we randomize participants into 2 groups, one in which we lower (bad) cholesterol and another in which we don't, the control group. Then we observe them for a period and note the rate at which people in both groups develop heart disease or die from it. If we find that the control group, the one which didn't receive the benefit of having its cholesterol lowered, has a significantly higher rate of falling ill, we conclude that lowering cholesterol is the way to go. Sounds easy, but it isn't. For several reasons. In the case of cholesterol, the time between developing high bad, or low good, cholesterol and suffering a heart attack or stroke is measured in decades rather than in years. We also cannot just experiment with people as we would like to in the name of science. Ethics boards look very closely at the potential risks and benefits associated with what we do in trials. We cannot simply withhold treatment from a control group, with scientific curiosity as the motivation. With these obstacles, we had  to draw our conclusions from observational studies, which tell us a lot about associations but nothing about cause and effect. Until now, we simply had no other choice. But not any more:

It's Mendel All Over Again

With larger and larger databases being developed from genetic research we can now do something else: Mendelian randomization studies. Which is what Voight and colleagues did. The concept behind it is amazingly simple and elegant, though not as brand new as you might think. It has been named after Gregor Mendel, the father of modern genetics, who first observed and described how traits are inherited. As always, a concept is best understood using an example. In the 1980s some researchers thought that very low cholesterol levels might increase the risk of cancer. There was definitely an association being observed between cancer and low cholesterol, but nobody knew which was the cause and which the effect. Or whether there was a third confounding variable, as yet unknown. Now, you can't make a study in which you lower the cholesterol in some people, just to see whether they will develop cancer.  Go and find volunteers for that one. So, Martjin Katan had another idea [2]. In 1986 he pointed out that there existed a certain variation in one gene (the gene which encodes the apolipoprotein E), which, if you had that variation, would give you extremely low cholesterol levels. He also knew, of course, that we inherit our genes from our mother and our father in a random way. That means, your hodgepodge of genes and my hodgepodge of genes are not systematically different from each other. Both are just random assemblies of genes from among all possible variations. In case you inherited that low-cholesterol gene, and I didn't, then it was just the luck of the draw. The important point is, that there is no room for confounding the random selection of genes.  So, if the "unconfoundable" low-cholesterol gene directly affects cholesterol levels and nothing else, then people who carry this gene should be found more often among patients with cancer than among people who are free from cancer. That was Katan's suggestion for a study design to test this cholesterol-cancer hypothesis.
Unfortunately, in 1986 it was impossible to realize this study design. The required genetic data were not yet available. That is changing. While, to the best of my knowledge, Katan's proposal has not been carried out yet, Voight and colleagues used his proposed design to investigate the HDL-heart disease theory.

The Death of Cholesterol?

They looked at a rare gene variant, which, as far as we know today, correlates strongly with HDL concentration, but not with any other cholesterol type. That's important, because we need to disentangle the effects of HDL from those of LDL. In their analysis, using data from 21,000 heart disease patients and 95,000 controls (people free of heart disease), the researchers could not find any association between HDL level and risk of heart disease. But Voight and colleagues didn't leave it at that. They also formulated a genetic risk score using 14 common gene variants with known effects on HDL (but not on LDL) and examined the score's association with heart disease in over 12,000 patients and over 41,000 controls. Again, nothing. Elevated HDL did not show up as the cherished knight in shining armor. What do we make of this? First, that raising HDL cholesterol may not be a way to reduce the risk of heart disease. Therefore, secondly, let's not think that treating a so-called risk factor will reduce risk (more on that in my post "when risk factors for heart disease really suck"). Third, let's hope Pfizer & Co. get this message, too. Because drugs, which treat risk factors but not risk, are like impostors: they never deliver.
1.   Voight, B.F., et al., Plasma HDL cholesterol and risk of myocardial infarction: a mendelian randomisation study. The Lancet, 2012. 2.   Katan, M.B., Apolipoprotein E isoforms, serum cholesterol, and cancer. Lancet, 1986. 1(8479): p. 507-8.
Voight, B., Peloso, G., Orho-Melander, M., Frikke-Schmidt, R., Barbalic, M., Jensen, M., Hindy, G., Hólm, H., Ding, E., Johnson, T., Schunkert, H., Samani, N., Clarke, R., Hopewell, J., Thompson, J., Li, M., Thorleifsson, G., Newton-Cheh, C., Musunuru, K., Pirruccello, J., Saleheen, D., Chen, L., Stewart, A., Schillert, A., Thorsteinsdottir, U., Thorgeirsson, G., Anand, S., Engert, J., Morgan, T., Spertus, J., Stoll, M., Berger, K., Martinelli, N., Girelli, D., McKeown, P., Patterson, C., Epstein, S., Devaney, J., Burnett, M., Mooser, V., Ripatti, S., Surakka, I., Nieminen, M., Sinisalo, J., Lokki, M., Perola, M., Havulinna, A., de Faire, U., Gigante, B., Ingelsson, E., Zeller, T., Wild, P., de Bakker, P., Klungel, O., Maitland-van der Zee, A., Peters, B., de Boer, A., Grobbee, D., Kamphuisen, P., Deneer, V., Elbers, C., Onland-Moret, N., Hofker, M., Wijmenga, C., Verschuren, W., Boer, J., van der Schouw, Y., Rasheed, A., Frossard, P., Demissie, S., Willer, C., Do, R., Ordovas, J., Abecasis, G., Boehnke, M., Mohlke, K., Daly, M., Guiducci, C., Burtt, N., Surti, A., Gonzalez, E., Purcell, S., Gabriel, S., Marrugat, J., Peden, J., Erdmann, J., Diemert, P., Willenborg, C., König, I., Fischer, M., Hengstenberg, C., Ziegler, A., Buysschaert, I., Lambrechts, D., Van de Werf, F., Fox, K., El Mokhtari, N., Rubin, D., Schrezenmeir, J., Schreiber, S., Schäfer, A., Danesh, J., Blankenberg, S., Roberts, R., McPherson, R., Watkins, H., Hall, A., Overvad, K., Rimm, E., Boerwinkle, E., Tybjaerg-Hansen, A., Cupples, L., Reilly, M., Melander, O., Mannucci, P., Ardissino, D., Siscovick, D., Elosua, R., Stefansson, K., O'Donnell, C., Salomaa, V., Rader, D., Peltonen, L., Schwartz, S., Altshuler, D., & Kathiresan, S. (2012). Plasma HDL cholesterol and risk of myocardial infarction: a mendelian randomisation study The Lancet DOI: 10.1016/S0140-6736(12)60312-2