Download Commentary: Mendelian randomization, 18 years on

IJE vol.33 no.1 © International Epidemiological Association 2004; all rights reserved.
International Journal of Epidemiology 2004;33:10–11
DOI: 10.1093/ije/dyh023
Commentary: Mendelian randomization,
18 years on
Martijn B Katan
apoE4 phenotype. I was a molecular biologist by training, and
although I was new to the lipid field I felt at ease with genetics.
So it was obvious to me that there were plenty of people who
carried one or two copies of the apoE2 gene, and that the large
majority of them had relatively low cholesterol levels but were
in other respects quite comparable with people who only
carried copies of the apoE3 and apoE4 variants of the gene.
Importantly, none of these people knew about their genetic
status or their cholesterol levels, because this was before the era
of large-scale cholesterol testing. So the E2 subjects lived their
lives in exactly the same way as the E3s and E4s; a perfect
experiment of nature. Each had had these genes and these low
cholesterol levels from birth, so there was no need for a
prospective study; measuring the apo E phenotype in cancer
patients and matched controls was just as good as measuring it
in a large number of newborns and following them to see who
developed the disease and who did not. It had recently become
feasible to determine apoE phenotypes in thousands of patients
by means of protein electrophoresis (this was before the days of
DNA sequencing), so I thought that a case-control study of apoE
phenotype in cancer patients versus controls could settle the
question of low cholesterol and cancer. As I was not an
epidemiologist and did not have access to patients myself I put
the idea in a letter to the Lancet, hoping that someone else
would take it up. Although no one did6 I still felt pleased with
this clean and logical approach to an important problem.
However, I never felt that I had invented something new,
because genetic experiments of nature had been invoked many
times before to explain the aetiology of diseases; thus in the
days when the role of low density lipoprotein (LDL) cholesterol
in heart disease was controversial, the high rates of coronary
heart disease (CHD) in familial hypercholesterolaemia had been
a strong argument in favour of causality.
I remained interested in applying molecular genetics to the
study of diet and heart disease, but most of our studies turned
out negative, especially those of the genetic basis of hypo- and
hyperresponsiveness to dietary cholesterol.7,8 My most
successful involvement in a Mendelian randomization study was
in a meta-analysis of the methylene tetrahydofolate reductase
(MTHFR) C677T mutation, a study in which I played a minor
but, I like to think, significant role. The MTHFR enzyme converts
folate into its biologically active form, and the mutation in base
pair 677 causes a reduction in the activity of the enzyme which
results in reduced plasma folate and increased homocysteine
levels. The meta-analysis was set up to decide whether folate,
and by implication homocysteine, was causally involved in CHD.
As the data collection progressed it became clear that the relative
risk associated with an increase in homocysteine was lower than
originally thought, and therefore the number of patients
I have no notes left regarding writing ‘Apolipoprotein E
isoforms, serum cholesterol, and cancer’,1 but I think I thought
it up in Hawaii. I passed through Hawaii on my way to the US
from Melbourne, where I had given a talk on diet, low
cholesterol, and cancer at the 7th International Atherosclerosis
Symposium in October 1985. At that time the dangers of a low
serum cholesterol level was a hot topic;2 several scientists
thought that a low cholesterol increased your risk of violent
death or cancer. This rested both on observational associations
and on the outcomes of early cholesterol-lowering trials. The
idea became less plausible after the big statin trials showed no
relation between lowering of cholesterol and the rates of violent
death or cancer, but those studies came later.
I myself was convinced that the association was spurious or
due to reverse causality, i.e. to occult tumours causing a
lowering of cholesterol in future cancer patients. I must confess
that that conviction rested partly on data and partly on
emotion. Cholesterol had been a subject of fierce controversy
for decades, and the scientific debate was frequently distorted
by commercial interests, with dairy, meat, and egg producers on
one side and margarine and oil producers on the other.
Scientists were also divided, and each side distrusted whatever
data the other side came up with.3,4 I had attempted to stay in
the middle but I had always felt that the data in favour of
cholesterol lowering were strong, and I was wary of theories
that a high cholesterol level might be better than a low level.
But what type of data could resolve this issue?
This is where genetics, with its clear causal pathways, should
be able to help out. In my talk in Melbourne5 I had already
argued that patients with the rare genetic disease, abetalipoproteinaemia, do not get premature cancer even though
they have almost zero plasma cholesterol levels. However,
when I thought the matter over after the Melbourne meeting I
realized that there were too few patients with this disease
worldwide to settle the issue. When I tried to think of other
subjects with low cholesterol levels, the apolipoprotein E
polymorphism came to mind. I had first heard about this
polymorphism at the European Lipoprotein Club meetings in
Tutzing in Bavaria; these were informal gatherings where I had
learned a lot about lipids. The apoE-2/3/4 polymorphism had
been frequently discussed there both by its discoverer Gerd
Utermann and by Gerd Assmann, who from early on had an
interest in applying the new genetics to population studies.
After several years of confusion it had become clear that
cholesterol levels increased from the apoE2 to the apoE3 to the
Wageningen Centre for Food Sciences and Division of Human Nutrition and
Epidemiology, Wageningen University, Bomenweg 2, 6703 HD Wageningen,
The Netherlands. E-mail: [email protected]
10
MENDELIAN RANDOMIZATION
required to settle the question grew and grew. My contribution
consisted of encouraging my scientific staff members to keep
going until they had enough patients to settle the matter—in this
case 23 000 patients. The study showed definitively that the 677
mutation is associated with increased risk of CHD.9
So what have the past 18 years taught me about Mendelian
Randomization studies? Anything I could say about the
topic has already been said better in the excellent review by
Davey Smith and Ebrahim,6 but I would still like to stress a few
points.
The first thing I learned is that you need a simple well-defined
phenotype. In the case of apoE the three genetic variants
produced clearly different cholesterol levels, via a logical mechanism (although logic can be deceptive; the apoE2 variant was
originally discovered as the cause of a rare type of
hyperlipidaemia, and it took a while to establish that most of
the homozygous E2/E2 subjects have low, rather than high,
plasma cholesterol). I have to admit that the phenotype which
I chased for over 20 years, namely an exaggerated versus a
reduced response of blood cholesterol to diet, is too fuzzy and
hard too measure to allow proper genetic analyses.
The second thing is that you need large numbers to get a
result that will stick. That implies that only the urgent issues are
worth pursuing; if the question that you are trying to solve is
not sufficiently inspiring and challenging you will be tempted to
give up when the first 100 or even 1000 subjects fail to give you
a clear answer, which is often the case.
Finally, the apoE episode and all that followed taught me
that my early switch from chemistry10 and biochemistry11
into nutrition science did not mean that I had wasted my
time. The matrix algebra that I struggled with in trying to
understand quantum chemistry later eased my path into
biostatistics, and my PhD studies in molecular biology
taught me the genetics that I needed to understand the apoE
11
polymorphism. A training in the basic sciences can come in
handy in unexpected ways.
References
1 Katan MB. Apolipoprotein E isoforms, serum cholesterol, and cancer.
Lancet 1986;i:507–08.
2 Feinleib M. Review of the epidemiological evidence for a possible
relationship between hypercholesterolemia and cancer. Cancer Res
1983;43:25033–75.
3 Keys A, Grande F, Anderson JT. Bias and misrepresentation revisited:
‘perspective’ on saturated fat. Am J Clin Nutr 1974;27:188–212.
4 Mann GV. Diet–Heart: end of an era. N Engl J Med 1977;297:644–50.
5 Katan MB. Effects of cholesterol-lowering diets on the risk of cancer
and other non-cardiovascular diseases. In: Fidge NH, Nestel PJ (eds).
Proceedings VIIth Atherosclerosis Symposium. Amsterdam: Elsevier, 1986,
pp. 657–61.
6 Davey Smith G, Ebrahim S. ‘Mendelian randomization’: can genetic
epidemiology contribute to understanding environmental
determinants of disease? Int J Epidemiol 2003;32:1–22.
7 Glatz JFC, Demacker PNM, Turner PR, Katan MB. Response of serum
cholesterol to dietary cholesterol in relation to apolipoprotein E
phenotype. Nutr Metab Cardiovasc Dis 1991;1:13–17.
8 Weggemans RM, Zock PL, Meyboom S, Funke H, Katan MB. The
Apoprotein A4-1/2 Polymorphism and Response of Serum Lipids to Dietary
Cholesterol in Humans. Utrecht, the Netherlands: 2001, p.55.
9 Klerk M, Verhoef P, Clarke R, Blom HJ, Kok FJ, Schouten EG. MTHFR
677C→T polymorphism and risk of coronary heart disease: a metaanalysis. JAMA 2002;288:2023–31.
10 Katan MB, Giling LJ, Van Voorst JDW. pH dependence of the transient
absorptions in the flash photolysis of 3-methyllumiflavin. Biochim
Biophys Acta 1971;234:242–48.
11 Katan MB, Van Harten-Loosbroek N, Groot GSP. The cytochrome bc1
complex of yeast mitochondria: site of translation of the polypeptides
in vivo. Eur J Biochem 1976;70:409–17.
IJE vol.33/1 © International Epidemiological Association 2004; all rights reserved.
International Journal of Epidemiology 2004;33:11–14
DOI: 10.1093/ije/dyh056
Commentary: Katan’ remarkable foresight:
genes and causality 18 years on
Bernard Keavney
Over the last scientific generation, observational epidemiology
and clinical trials have revolutionized our understanding of
causal risk factors predisposing to a variety of common diseases,
perhaps most strikingly cardiovascular disease. Pretty much every
member of the public now knows that smoking, high blood
University of Newcastle, Institute of Human Genetics, Central Parkway,
Newcastle upon Tyne, UK. E-mail: [email protected]
pressure, high levels of blood cholesterol, and diabetes predispose
to the development of coronary heart disease (CHD), and yet one
does not have to venture too far back into last century to
find a time when all of this was completely unknown. The extraordinary power of large blood-based observational epidemiological studies to identify associations between risk factors and
complex diseases has been one of medical science’s great recent
success stories. No less important have been the data from