SLCO1B1, ABCG2 and statins: what one gene says, and what the pair adds

Two transporters between the tablet and the liver

Statins lower cholesterol by blocking an enzyme inside liver cells, HMG-CoA reductase, so the liver is where a dose does its work^[2]. What fails to get in stays in the circulation, and the guideline calls that systemic exposure the putative cause of the link between these genes and muscle symptoms^[2]. OATP1B1 sits on the blood-facing side of liver cells and takes statins in. BCRP sits on the gut-facing side of the cells lining the intestine and on the bile-facing side of liver cells, and pushes its substrates out^[1]. A weak OATP1B1 leaves more statin in the blood. A weak BCRP changes how much rosuvastatin is absorbed and how fast it leaves, and the net effect is again more of it in the blood^[2].

SLCO1B1: one variant and the studies behind it

The finding came from SEARCH, a statin trial of 12,000 people. Its investigators scanned about 300,000 markers in 85 participants who developed definite or incipient myopathy while taking 80 mg of simvastatin a day, and in 90 on the same dose who did not. One signal stood out, in SLCO1B1, and it led to rs4149056, a change of one letter (c.521T>C) that swaps one amino acid in the transporter. Each copy of the C allele multiplied the odds of myopathy by 4.5, and two copies by 16.9. The authors attributed more than 60 percent of the myopathy cases to the variant, and the association held in a second trial in which 20,000 people took 40 mg^[3].

Later studies pointed the same way, most clearly for simvastatin and for the more severe cases. In STRENGTH, 509 people were randomised to atorvastatin, simvastatin or pravastatin; a composite of myalgia, raised creatine kinase or stopping for side effects occurred in 19, 27 and 50 percent of those with zero, one and two copies of the allele, and the excess was largest on simvastatin^[4]. A British study recruited through general-practice records found an odds ratio of 2.06 per copy for myopathy and 4.09 for severe myopathy^[5]. Its genome-wide follow-up found the same variant, at 5.15 for severe myopathy on several statins, and no other variant with a comparable effect^[6]. JUPITER, which gave 4,404 people 20 mg of rosuvastatin, found no association between rs4149056 and clinically reported myalgia, with a hazard ratio of 0.95 per copy^[7].

How much the variant changes each statin was measured at the University of Helsinki, in small studies in which healthy volunteers took single doses. Four people with two copies of c.521C had 221 percent more simvastatin acid, the active form, in their blood than 16 people with none^[8]. In a second study, the same comparison raised atorvastatin exposure by 144 percent and rosuvastatin by 65 percent, a larger effect on atorvastatin than on rosuvastatin that its authors called unexpected^[9].

Star-allele names add a layer. rs4149056 sits inside two SLCO1B1 alleles: *5, which carries that change alone, and *15, which also carries c.388A>G, a change that on its own (as *37) leaves the transporter working normally^{[12, 13]}. CPIC classes both *5 and *15 as no-function alleles; one of them makes a person's phenotype decreased function, two make it poor function^[2]. In Europeans *15 is much the commoner, at about 15 percent of alleles against 2 percent for *5^[14]. Other alleles raise transporter function (*14, *20). In September 2025 CPIC's expert panel reassessed alleles that are more frequent in populations under-represented in the early studies, and revised the function of *39, *41 and *45^[15].

What CPIC 2022 says for each statin

CPIC's first two SLCO1B1 guidelines, in 2012 and 2014, covered simvastatin alone^[16]. The 2022 guideline replaced them, added ABCG2 and CYP2C9, and wrote a recommendation for each statin, based on the risk of muscle symptoms rather than on cholesterol lowering^[2]. It is addressed to prescribers choosing a statin and dose, and it states that statin therapy should neither be stopped nor avoided on the basis of these genotypes in people who have a reason to take it^[2].

Simvastatin and lovastatin are the two statins where the guideline's first answer to a reduced SLCO1B1 result is a different statin. For the others it sets a starting-dose ceiling or adds a caution above a dose. The table only applies once someone has a genotype; CPIC states that its document is a guide to using available results, not a judgement on whether to test^[2].

ABCG2: a pump that matters for one statin

The ABCG2 variant CPIC grades is rs2231142, c.421C>A, which turns glutamine 141 into lysine (Q141K). Its A allele goes with 30 to 40 percent less of the protein^[2]. The measurement behind the guideline again comes from Helsinki: among 660 healthy Finnish volunteers, 9.5 percent of alleles carried c.421A, and in 32 of them who took single doses, four people with two copies had 144 percent more rosuvastatin in their blood than 16 with none and 100 percent more than 12 with one copy. Atorvastatin rose by 72 percent^[10]. A companion study found fluvastatin up by 72 percent and simvastatin lactone, the inactive form, by 111 percent, with no significant change in simvastatin acid or pravastatin^[11].

CPIC grades the gene for rosuvastatin only. It found the strongest evidence there, judged the atorvastatin data insufficient for a recommendation, and says the risk of myopathy itself is unknown: the ABCG2 advice rests mainly on the exposure data^[2]. Higher exposure also shows up as a stronger effect. In a genome-wide study of 6,989 JUPITER participants of European ancestry, ABCG2 was one of three loci where variants reached genome-wide significance for how far rosuvastatin lowered LDL cholesterol^[17]; CPIC puts that down, probably, to more of the drug reaching the liver^[2]. Ancestry matters beyond this one variant: rosuvastatin exposure was 2.31 times higher in Chinese than in white volunteers living in Singapore, a gap SLCO1B1 genotypes did not explain^[18], and the guideline asks prescribers to consider Asian ancestry when choosing a rosuvastatin dose^[2].

Rosuvastatin with and without ABCG2

Without ABCG2, rosuvastatin follows the SLCO1B1 row of the table above: the desired starting dose at normal function, the same with a caution above 20 mg at decreased function, and a starting dose of 20 mg or less at poor function. With ABCG2, CPIC writes a grid of four SLCO1B1 rows by three ABCG2 columns, one cell for each pair of phenotypes^[2]. The figure below draws the nine cells most people fall into, read from the same CPIC data file our report quotes.

Read across any row and one copy of the ABCG2 variant changes nothing in the wording. It changes the strength: in the top two rows, CPIC's recommendation drops from strong to moderate. Two copies set a lower starting-dose ceiling in every row: 20 mg when SLCO1B1 is normal, 10 mg when it is decreased or poor. The fourth row, increased SLCO1B1 function, reads like the normal row. When SLCO1B1 cannot be assigned, CPIC writes a line based on ABCG2 status alone^[2].

That sentence is the limit of the grid. Every study of muscle symptoms in CPIC's review looked at one gene at a time, and the guideline rates the recommendations for the highest-risk pairs as optional because clinical and pharmacokinetic data on the pairs are limited^[2]. The 10 mg cells are arithmetic on two single-gene effects, not a measurement in people who carry both.

How common each variant is, in Finland and elsewhere

gnomAD, a public reference of sequenced exomes and genomes, splits its counts by genetic ancestry group, and Finland has a group of its own: about 32,000 people for these two positions. The SLCO1B1 variant is more frequent in Finns than in other Europeans, and the ABCG2 variant less frequent^{[19, 20]}. The 660 Helsinki volunteers gave 9.5 percent for c.421A, with a 95 percent confidence interval of 8.1 to 11.3^[10]. gnomAD's Finns give 7.3 percent, outside that interval: a frequency belongs to the people sampled as much as to the country.

Put those counts into the grid and the Finnish picture is lopsided. About 37 percent of Finns carry at least one c.521C, so SLCO1B1 alone moves more than a third of the population off the top-left cell. Two copies of c.421A, the only ABCG2 genotype that changes CPIC's wording, occur in 0.49 percent, about one Finn in 200; in East Asian ancestry it is 9.1 percent, about one in 11. Both genes reduced at once, in any combination, comes to about 5.3 percent of Finns by our estimate. These shares count only the two tagged variants, so they leave out rarer SLCO1B1 alleles in both directions.

What a raw data file can read

Both genes are graded on single-letter variants, which is the kind of position genotyping chips are built to read. We checked the ten public chip exports on our data page^[21] on 10 October 2026. The 23andMe (three chip versions), AncestryDNA (two) and Genes for Good files carry rows for both rs4149056 and rs2231142. The 2018 MyHeritage, 2019 Living DNA and both FamilyTreeDNA exports have no row for rs4149056, and three of those four lack rs2231142 too. Neither variant is a strand-ambiguous A/T or C/G pair, so a chip's letters can be oriented. They can still confuse a reader: ABCG2 lies on the reverse strand of chromosome 4^[22], so the change papers write as c.421C>A appears in a raw file as G and T.

# rsid      chromosome  position  genotype
rs4149056   12          21331549  TC
rs2231142   4           89052323  GT

One SLCO1B1 position cannot tell *5 from *15, which is harmless for the phenotype because both are no-function alleles, and it cannot see any other allele. CPIC warns that a person carrying a rare reduced-function allele may be assigned normal function when a test defaults to *1^[2]. The list of such alleles includes *48, a deletion of the whole gene^[13], which a chip row cannot show. A plain VCF has a different gap: a missing line at rs4149056 may mean the reference letter or a position that was never read. A gVCF says which, and aligned reads in a BAM or CRAM let a star-allele caller examine the whole gene.

What the Aimosti report shows for these genes

The chip view of our sample report belongs to a fictional person with one copy of c.521C and two of c.421A, which lands in one of the highlighted cells: its ABCG2 card quotes CPIC's rosuvastatin line for SLCO1B1 decreased function with ABCG2 poor function. When a chip has rs2231142 but no rs4149056, as one FamilyTreeDNA export does, the ABCG2 card quotes CPIC's line for an unassigned SLCO1B1 instead of assuming a normal one. Two limits are worth knowing. From a chip or variant file the SLCO1B1 card restates simvastatin only; the other six statins are in the table above and, for three of them, on the SLCO1B1 page. And the gVCF and Deep Read cards do not pair the two genes, although both come from the same file. The ABCG2, rosuvastatin and simvastatin pages restate CPIC's single-gene tables.

Muscle symptoms have more causes than one gene

The guideline's own list of other factors is long: a higher statin dose, interacting drugs, older age, low body mass, female sex, an underactive thyroid, intense exercise and Asian or African ancestry. Dose is the strongest independent predictor, with muscle symptoms about six times as common on high doses as on low ones. Some interactions run through the same two transporters: the guideline names ciclosporin as one that involves both^[2]. A genotype is one input among these.

Blinded trials show how much of the aching is the statin. The Cholesterol Treatment Trialists pooled 19 placebo-controlled trials with 123,940 participants: 27.1 percent of those given a statin reported muscle pain or weakness, against 26.6 percent of those given placebo. In the first year the statin added a 7 percent relative excess, which the authors read as one report in 15 being caused by the drug, and after the first year there was no significant excess^[25]. Their conclusion was that more than 90 percent of the muscle complaints from trial participants on a statin were not caused by it^[25].

Whether genotype-guided prescribing prevents muscle symptoms has not been shown; the guideline says prospective data are lacking^[2]. One randomised trial gave primary-care physicians at eight Veterans Affairs practices in Boston the SLCO1B1 results of 408 patients not yet on a statin, either at once or a year later. Of the patients, 29 percent had a genotype linked to higher simvastatin myopathy risk; among them, one was prescribed simvastatin, in the group whose doctors had not yet seen the result. LDL cholesterol a year on was no worse where the doctors had the result, and the trial recorded five cases of statin muscle symptoms in all^[26]. Its question was whether handing doctors the result harmed cholesterol control, and it did not.