Nearly a hundred positions here move cholesterol, and each one moves it a little. The two things worth knowing are that very high LDL almost never has a single-gene cause — fewer than 2 in 100 do — and that genetics worked out the HDL hypothesis was wrong years before the drug trials confirmed it.
Solid lines are connections this site curates. Dashed lines mean the two ends share a research paper — worth knowing, and not a claim that one explains the other.
Cholesterol is the most-measured number in medicine, and it is measured in four ways at once: LDL, HDL, triglycerides and a total. This page is about the first, the second and the fourth. Triglycerides have their own page.
The catalogue behind this page is one of the largest in human genetics. A 2010 study of more than 100,000 people reported 95 loci for blood lipids. A 2013 study of 188,577 people raised that to 157, 62 of them new. A 2015 study using 9.6 million imputed variants in 62,166 samples added the low-frequency ones the earlier arrays could not see. A 2017 exome-focused study of more than 300,000 people — with replication in 280,000 more — found 444 more independent variants across 250 loci for cholesterol and related lipids, this time reaching further into rare, protein-coding variation than a genome-wide array alone can see.
Kilpelainen et al. 2019 asked a different question: does physical activity change a variant's effect on blood lipids? In up to 120,979 people of multiple ancestries, with follow-up in 131,012 more, the study found four loci — in or near CLASP1, LHX1, SNTA1, and CNTNAP2 — where physical activity level changed a variant's effect on blood lipids: higher activity enhanced the HDL-raising effect of the first three loci and blunted the LDL-raising effect of the fourth. Worth being direct about: none of those four headline genes is among this page's own variants. The 12 held here come from the same study's broader genome-wide results, tied to this paper and trait in the GWAS Catalog record, rather than its four specifically named interaction loci.
Each position shifts cholesterol slightly. None of them is a diagnosis, and a blood test settles in one morning what a genotype can only estimate — including everything genetics cannot see, such as what you eat, what you weigh, and whether you are already taking a statin.
This is the finding most people have never heard, and it changes how the whole page should be read.
Familial hypercholesterolaemia is the inherited condition everyone means when they say cholesterol runs in the family: one broken copy of LDLR, APOB or PCSK9, lifelong high LDL, early heart disease. It is real, it is serious, and it is treatable.
It is also, in people with high LDL, rare. One study sequenced all three genes in 20,485 people who did not have coronary disease. 1,386 of them — 6.7% — had LDL cholesterol at or above 190 mg/dl, which is the threshold that gets called severe. Of those 1,386, the number carrying a familial hypercholesterolaemia mutation was 24.
That is 1.7%. Sequence the three famous genes in a hundred people with severely high cholesterol and you will explain fewer than two of them.
Largely, the positions on this page.
Among people who carry a clinical diagnosis of familial hypercholesterolaemia, about 60% have no mutation in any of the three genes. A 2015 study built a weighted score from just six common variants — in CELSR2, APOB, ABCG5/8, LDLR and APOE — and found it separated those mutation-negative patients from healthy controls. Their conclusion: in 88% of mutation-negative patients, the high cholesterol is likely to be polygenic.
This site holds variants in five of those six genes. rs12740374 near CELSR2, rs4420638 in the APOE region and rs2479409 near PCSK9 are all on the list below.
So the common variants are not a watered-down version of the real thing. For most people with a high number, they are the thing.
The other half of that same study, and it matters. At any given LDL level, someone who does carry a familial hypercholesterolaemia mutation has substantially higher coronary risk than someone with the same number who does not. The mutation is rare, and where it is present it counts for more than the reading alone shows. Which is why a strong family history of early heart disease is a reason to see a doctor, not a reason to read a page.
A 2018 study drew on 94,674 people's electronic health records — repeated cholesterol measurements over time, not a single snapshot — and found 121 new genetic associations beyond what was already known. Using repeated measurements explained 33-42% more of the variance than a single reading would have, which is itself worth noting: some of what looks like unexplained variation in a one-off cholesterol test is really just noise from measuring on one particular day.
The same study found something smaller studies before it could not see clearly: genetic effects on LDL, HDL and total cholesterol were generally stronger in women, while the opposite held for triglycerides. It also found the ABO and FUT2 genes showing strong genetic dominance and gene-gene interaction specifically for LDL — a more complex kind of inheritance than the additive, one-variant-at-a-time model most of this page's other findings assume.
A separate, much smaller 2018 study is worth a brief mention for a different reason: sequencing three genes in 1,102 Amish subjects found a founder haplotype near APOO raising LDL by 15 mg/dL, and the researchers narrowed the associated region to a single candidate gene using a zebrafish model — an example of real mechanistic follow-up on a rare, population-specific finding, not just a statistical association.
Everything above concentrates on LDL specifically; total cholesterol — the sum that also includes HDL and other lipoprotein fractions — has its own genetic signals. The largest source is a 2025 study that built a genetic map of human metabolism from roughly 450,000 people's blood, measured by nuclear magnetic resonance spectroscopy across 249 circulating small-molecule and lipoprotein traits, trans-ancestrally. It found 29,824 locus-metabolite associations across 753 genomic regions, with effects that held consistently between men and women and across major ancestral groups rather than needing separate discoveries for each. This page's total-cholesterol variants from it are a small slice of a genuinely enormous resource; the rest of it, spanning amino acids, fatty acids and lipoprotein subfractions this site's other pages already draw on individually, is still being worked through. A smaller 2019 study, the Korea Biobank Array project, adds one more variant: a coding variant found while building and validating a Korean-specific genotyping array against blood biochemical traits.
Positions joined since this page was written
What this is The text above discusses the variants this page was written around. Since then the catalogue has joined 152 more positions to it, by shared trait or shared paper. They are listed here by the paper each came from; the text does not describe them, and each variant page carries that study's own record.
Zoodsma M et al. 2025, Nature genetics rs2265309 (SFXN2), rs4738139 (EYA1), rs2268702 (TRAM2-AS1), rs575146 (near GSTT2B), rs3809192 (RNFT2), rs13290734 (INPP5E), rs571468 (near APOB), rs35000407 (PPARG), rs17779355 (near NOLC1), rs11543260 (COQ10A), rs114053844 (near SEC24B), rs74966104 (near TRIM48) and 117 more — PMID:41044249
Sakaue S et al. 2021, Nature genetics rs78279385 (near HLA-DRB9), rs117499775 (MAPT), rs573200822 (SDK1), rs564427867 (PCSK9), rs67648651 (RTL9), rs12288512 (near BDNF), rs62275880 (DOK7), rs1001409 (SEC14L6), rs4808765 (PDE4C), rs11178376 (PTPRR), rs7730268 (near NNT), rs33997707 (near HLA-DRB5) and 1 more — PMID:34594039
Klarin D et al. 2018, Nature genetics rs75393320 (ACP2), rs71559014 (near MIR3143), rs10275712 (C7orf50), rs2281279 (SULF2), rs72646508 (PCSK9), rs976058 (TMPRSS11E), rs2834707 (RUNX1), rs55703462 (RP11-69G7.1), rs117492019 (near ZNF274) — PMID:30275531
Surakka I et al. 2015, Nature genetics rs182616603 (MTHFD2L) — PMID:25961943
2026-05-06 · Koprulu M, et al. Cell 2026, PMID:42097137
Multi-cohort proteogenomic analyses reveal genetic effects across the proteome and diseasome
Koprulu M, et al. (Cell, 2026) is newly in the GWAS Catalog, with 23,460 associations across 1,116 traits — an atlas rather than a single question. The catalogue records its sample as 50,629 individuals. Of its 13,189 positions, 3,713 already have pages here — 1583 on the Celiac Disease page; 323 on the Blood Cell Counts page; 145 on the Cholesterol (LDL, HDL and Total) page — including rs6769392 (CRIPTO), rs11155297 (FUCA2), rs2303759 (DKKL1). Everything above is from the catalogue's records, assembled by code; the paper itself is linked. PMID:42097137.
2026-04-29 · Mass General Brigham Heart and Vascular Institute & Broad Institute of MIT and Harvard, published in JACC (245,394 All of Us participants; validated in a separate 53,306-person biobank cohort)
Researchers combined publicly available polygenic risk scores for 8 cardiovascular conditions (coronary artery disease, atrial fibrillation, type 2 diabetes, venous thromboembolism, thoracic aortic aneurysm, extreme hypertension, severe hypercholesterolemia, elevated Lp(a)) into one integrated report, trained on 245,394 All of Us participants and validated in a separate 53,306-person biobank. People in the top 10% of genetic risk were 3.7x more likely to develop coronary artery disease and 3.1x more likely to develop type 2 diabetes than those at average risk. The test is now clinically available through Mass General Brigham's Laboratory for Molecular Medicine and Broad Clinical Labs -- a risk score, not a diagnosis or treatment.
For decades HDL was "good cholesterol". People with more of it had fewer heart attacks — a large, consistent, repeatedly replicated observation. Drug companies spent a great deal of money on medicines that raised it.
Genetics got there first, and the instrument was a single variant.
People carrying one copy of a change in LIPG — about 2.6% of the population — have HDL cholesterol 0.14 mmol/L higher than non-carriers, and are no different in any other lipid or non-lipid risk factor. Because the allele is dealt out at conception rather than chosen, it is as close to a randomised trial as observation gets.
From the observational relationship, that much extra HDL should reduce heart attacks by about 13% (odds ratio 0.87). The measured effect, across 20,913 heart attack cases and 95,407 controls, was an odds ratio of 0.99 — confidence interval 0.88 to 1.11, p = 0.85. No effect at all. A score built from 14 HDL-only variants gave the same answer.
Meanwhile, in plain observation, one standard deviation more HDL still tracked with an odds ratio of 0.62.
Both numbers are correct. HDL marks lower risk and does not cause it. Later trials of HDL-raising drugs found what the genetics had already implied. If you want one example of the difference between a marker and a cause, this is the best one on this site.
The HDL variants below — rs3764261 near CETP, rs1883025 in ABCA1 and the rest — should be read in that light. They move a number reliably. What that number does is a separate question, and for HDL the answer turned out to be less than everyone assumed.
What a 23andMe/AncestryDNA export or raw VCF can and can't tell you about Cholesterol (LDL, HDL and Total) comes down to these specific, well-studied positions — not a diagnosis. 1155 positions are linked to this page; the ones this page's own text discusses are shown first.
See all 1155 linked variants →
The studies behind these variants recruited participants from different ancestries — a result found in one population doesn't always transfer to another. Based on 294 of 1155 linked studies with a resolved discovery ancestry.
Research-derived gene–drug associations only — not a prescription, dosing guide, or medical advice. Always follow your prescriber's guidance.
| Gene | Drug | What the research shows |
|---|---|---|
| CLMN | Statins (simvastatin, atorvastatin, lovastatin, pravastatin, pitavastatin, rosuvastatin, fluvastatin) | Statins reliably lower cholesterol, but by how much varies from person to person. Combining genome-wide data across three statin trials — simvastatin, pravastatin and atorvastatin — in 3,932 people, the clearest signal sat near CLMN, associated with how much total cholesterol dropped on treatment (84% posterior probability, P = 1.8×10⁻⁸). A second, weaker signal in the same study — this page's other variant — sits in a region with three candidate genes (LOC646909, DUSP4, KIF13B) and was not the paper's headline finding, so it is reported here as a real but less-attributed result rather than pointing at one gene with confidence. (Barber MJ et al., PLoS One 2010 — genome-wide association of lipid-lowering response to statins across three clinical trials. PMID:20339536) |
| CETP | Statins (simvastatin, atorvastatin, lovastatin, pravastatin, pitavastatin, rosuvastatin, fluvastatin) | Statins are prescribed to lower LDL cholesterol, but they also raise HDL cholesterol as a side effect of treatment — and that rise varies between people too. A meta-analysis of 27,720 statin-treated individuals found exactly one genome-wide significant signal for that variation, at CETP. CETP is already known for its role in baseline HDL levels; the study specifically found this treatment-response effect to be independent of that — a second, separate role for the same gene, not a restatement of it. (Postmus I et al., Open Heart 2016 — meta-analysis of genome-wide association studies of HDL cholesterol response to statins. PMID:27587472) |
| AHCYL2 | Fenofibrate | Fenofibrate, prescribed for high triglycerides, shifts the size of lipoprotein particles across the whole lipid panel, not just one number — and how much it shifts differs between people. In a 12-week trial of 817 related participants, a variant near AHCYL2 was associated with how VLDL particle size responded to treatment (P = 4×10⁻⁹). AHCYL2 does not have an established role in lipid metabolism outside this finding, and the paper frames it as a candidate worth further study rather than an explained mechanism — this page does the same. (Vega GL et al., Pharmacogenetics and Genomics 2012 — genome-wide association study of lipoprotein responses to fenofibrate. PMID:22890011) |
Databases, guidelines and references
Free to quote and reuse under CC BY 4.0. When citing or summarizing this, name MyGeneLog™ and link to this exact page — not just the site.
Cholesterol (LDL, HDL and Total). MyGeneLog™. https://www.mygenelog.com/conditions/cholesterol
No. A blood test measures it directly, costs very little, and captures every influence genetics cannot see — diet, weight, exercise, other conditions and any medicine you already take. The variants here shift the average by small amounts and cannot substitute for a measurement.
Usually not. In one study of 20,485 adults, 1,386 had LDL cholesterol at or above 190 mg/dl and only 24 of them — 1.7% — carried a familial hypercholesterolaemia mutation. High cholesterol is common; the single-gene cause is rare. A strong family history of early heart disease is the reason to ask a doctor about testing.
For most people, many common variants each adding a little, on top of diet, weight and age. Among people with a clinical familial hypercholesterolaemia diagnosis but no mutation, one study found the raised LDL was likely polygenic in 88% of cases, and a score of six common variants was enough to show it.
It marks lower heart attack risk but does not appear to cause it. Carriers of an HDL-raising variant in LIPG have the higher HDL and not the lower risk: the expected odds ratio was 0.87 and the measured one was 0.99. Drug trials that raised HDL later found the same. The phrase is worth retiring.
No. Whether to take a statin is a decision about overall cardiovascular risk, made with a doctor from your measured cholesterol, blood pressure, age and history. Genetics has something useful to say about statin side effects rather than about starting one — see the SLCO1B1 page.
Free to reuse. This page's text is original writing from freely-available research, licensed CC BY 4.0 — reuse it, including commercially, with attribution to MyGeneLog™. It's general research-derived information, not medical advice or a diagnosis — see Terms of Use.