These are my notes from week 12 of Harvard’s Genetics 228: Genetics in Medicine from Bench to Bedside course, held on April 24, 2015. Featuring a lecture by Sekar Kathiresan.

Journal club: Exome sequencing, ANGPTL3 mutations, and familial combined hypolipidemia [Musunuru 2010]

Familial hypobetalipoproteinemia is associated with very low (<5th percentile) levels of ApoB and LDL cholesterol. The most common cause is loss-of-function mutations in APOB. In this paper, they had a pedigree with hypobetalipoproteinemia that was negative for APOB mutations, so they undertook exome sequencing to find the cause, and found compound heterozygous nonsense mutations (S17X and E129X) in ANGPTL1. This was back in 2010 when there had only been a single digit number of instances where exome sequencing had identified a new disease gene, so just the fact that they successfully used this approach was still pretty novel.

Reading the paper is a real throwback - those were the days when you could still just throw out any variant that was present in dbSNP! That approach began to be phased out in favor of allele frequency filters the following year [Bamshad 2011]. Also, back then exome sequencing was super expensive, so instead of sequencing all four affecteds right off the bat, they sequenced two of them and then went back and Sangered the candidate causal variants in the other affecteds. They also Sanger sequenced the unaffected carriers, which confirmed the two variants were in trans to each other.

This family had not only low LDL but also low HDL. Interestingly, the LDL phenotype was codominant (hets had intermediate LDL levels) while the HDL phenotype was purely recessive (hets had wild-type HDL levels).

For validation they also went back and Sanger sequenced the gene in people from the Dallas Heart Study who had been phenotyped for LDL levels. They found 12 people heterozygous for 6 different frameshift variants in ANGPTL3, and these 12 people tended to have lower LDL than wild-type individuals (median 77.5 mg/dl vs. 104, p = 0.03).

Sekar Kathiresan: Using human genetics to validate drug targets in heart disease

Myocardial infarction (MI; heart attack) is the leading cause of death both in the U.S. and worldwide. It occurs when athersclerotic plaques build up on artery walls until a thrombus (blockage of red blood cells) occurs. If the heart tissue downstream of that blockage then goes without blood supply for 20 minutes, it will die. The death of heart tissue is called a heart attack.

Until statins, there was no medication for high LDL. If you had high LDL (the population average in the U.S. is 130 mg/dL and the 95th percentile is 190 mg/dL), your doctor would simply tell you to eat better and exercise more. In the late 1980s, the first statin was approved. Statins underwent clinical trials for progressively lower risk groups - for instance, they were investigated for preventing a second heart attack in people who had already had one, and only later was there a trial to prevent a first heart attack. It was eventually shown that statins both lower LDL and lower heart attack risk, but it still took years before it was firmly established that all means of lowering LDL also lower heart attack risk.

While there have been great advances in treating LDL and preventing heart attack, it is believed there are still undisovered genetic factors and pathways influencing risk. If your mother or father had early onset cardiovascular disease, you have about a 3x higher risk of heart attack [Lloyd-Jones 2004]. This odds ratio is fairly robust to controlling for environmental factors, suggesting much of the risk is genetic.

The first study establishing a correlation between serum cholesterol levels and subsequent risk of heart attack was the Framingham Heart Study [Kannel 1961]. The title of the study was Factors of risk in the development of coronary heart disease–six year follow-up experience and it is credited with establishing the term “risk factor” in the English vocabulary. We later learned that the serum cholesterol association in that study needs to be separated into correlations with three separate forms of cholesterol that circulate in the blood. LDL cholesterol is positively correlated with heart attack risk, HDL is negatively correlated, and triglyceride is positively correlated. The triglyceride correlation disappears if you control for certain other factors, whereas the other two correlations are robust to covariates [Emerging Risk Factors Collaboration 2009]. All three are tested on clinical blood lipid panels, and doctors tend to tell patients that LDL is “bad” cholesterol, HDL is “good” cholesterol, and that it is not clear whether triglyceride is important.

The Framingham Heart Study, however, was purely observational, and for decades thereafter, it was difficult to determine whether high LDL caused high heart attack risk, or whether high HDL caused low heart attack risk. For risk prediction - determining which people are at high risk and could benefit from an intervention - it actually doesn’t matter if your classifier is causal or merely correlated. But it matters a lot for therapy - drugging a target will only alter the outcome if the target is causal.

You can try to distinguish causality from correlation is a randomized controlled trial. Randomize people into two groups, treat them with an intervention, measure changes in the biomarker, and measure changes in the clinical outcome. The problem is that you can’t afford to do a randomized trial for every possible variable that might affect the clinical outcome. This is where genetics can come in. In Mendelian randomization [see Hingorani & Humphries 2005] you ask whether genetic variants that affect the biomarker also affect the clinical outcome. Human genetics offers two great advantages:

  1. The genome is unmodified by the disease process, which minimizes reverse causality
  2. Alleles are randomized at meiosis, which minimizes confounders

The biggest limitation of Mendelian randomization is that it relies on an assumption of no pleiotropy [Pickrell 2015]. Pleiotropy is when a genetic variant affects more than one trait. It is basically impossible to prove that there is no pleiotropy. For instance, in the studies of PCKS9 they showed that the two groups (R46L hets and wild-type people) were not significantly different in terms of smoking, diabetes, etc. [Cohen 2006, Kathiresan 2008] but you never know whether there is some other variable you haven’t thought of.

One way to try to get around this limitation is to try to show that the effect size is what would be expected based on the biomarker. For PCSK9 R46L hets, LDL is reduced by 21 mg/dL, a reduction which according to the observational correlation between LDL and heart attack would predict an odds ratio of 0.80 for heart attack. And indeed, the genetics says that R46L hets have about an odds ratio of 0.72 for heart attack.

Although the possibility of pleiotropy confounds positive results from Mendelian randomization, it does not confound the interpretation of negative results. The LIPG N396S variant raises HDL by 6 mg/dL, which - again, based on observational correlation between HDL and heart attack - would be predicted to lower heart attack risk. Yet people with this variant had unaltered risk of heart attack [Voight & Peloso 2012]. And sure enough, a clinical trial of an HDL-raising drug candidate, dalcetrapib, raised HDL from about 42 mg/dL to about 55 mg/dL in treated individuals yet had no effect whatsoever on heart attack risk [Schwartz 2012]. It now appears that HDL is simply a non-causal marker of heart disease risk - which means it can still be useful for doctors to measure, it’s just not a good drug target. It turns out that HDL is correlated with “everything good” - exercise, good diet, etc.

It has proven very difficult to figure out whether triglycerides are causally associated with heart attack risk using Mendelian randomization, because every genetic variant so far discovered that affects TG levels also affects either LDL, HDL, or both. One approach that Sek and collaborators took was to find a very large number of common variants associated with TG, and then ask whether the effect size of each SNP on TG was correlated with its effect size on heart attack [Do & Willer 2013]. A SNP’s odds ratio for LDL is correlated with its odds ratio for heart attack, but its odds ratio for HDL is not correlated with its odds ratio for heart attack. It turns out that the odds ratio for TG is correlated with odds ratio for heart attack, indicating that triglycerides probably are indeed causal for heart attack risk.

Dr. Kathiresan is of the view that we will have greater success in developing drugs if we focus our efforts on targets for which human genetics provides prospective validation [Plenge 2013]. His group is therefore focused on using human genetics to provide validation for additional targets for treating LDL and heart disease risk. Today there are two classes of drug demonstrated to reduce risk of first heart attack: aspirin and statins. They reduce risk by ~15% and ~25% respectively. The evidence for aspirin is mainly in older people; there is little evidence that it reduces first heart attack risk in young people. Monoclonal antibodies against PCSK9 are likely to be approved this summer.

Dr. Kathiresan has recently led two efforts which have validated drug targets using human genetics.

He ahs demonstrated that people heterozygous for LoF mutations in NPC1L1 (which encodes a transporter responsible for uptake of dietary cholesterol) have reduced LDL and reduced heart disease risk, providing validation for the drug ezetimibe [Myocardial Infarction Genetics Consortium 2014]. The effect size of a single LoF allele of NPC1L1 is to reduce LDL by 12 mg/dL and heart attack risk by 53%. One week after the genetic results came out, the results of a clinical trial of ezetimibe were published [Blazing 2014]. At 10 mg/day, ezetimibe reduced LDL by 15 mg/dL, and after 6 years, it had reduced heart attack risk by 6%. The fact that the heart attack risk benefit of ezetimibe was far, far smaller than the benefit from an LoF allele of NPC1L1 is hypothesized to be due to the cumulative protective effects of a whole lifetime of reduced LDL as opposed to only 6 years of reduced LDL.

You can never be too rich or have too low an LDL.

— Eugene Braunwald, quoted here

More recently, it has been discovered that four different APOC3 LoF mutations, with a cumulative allele frequency of 1 in 150 in the United States, are associated with 39% reduction of triglycerides and 40% reduction in heart attack risk [TG and HDL Working Group 2014]. In a fat challenge test (having volunteers suddenly eat 100 grams of fat), people with an APOC3 LoF mutation experience less of an increase in blood triglycerides [Pollin 2008]. There is already one drug in the pipeline - Isis Pharmaceuticals has an antisense oligonucleotide against APOC3 that has already been tested in human volunteers [Graham 2013].