Pharmacogenomic Star Allele

A star allele is a named haplotype of a pharmacogene: a defined set of variants that travel together on a single chromosome copy, written as the gene symbol plus an asterisk and a number, like CYP2C19*17 or NUDT15*3. You have two copies of each pharmacogene, so you have two star alleles. The pair is a diplotype, written CYP2C19*1/*17. Each allele carries a function assignment (normal, decreased, no function, increased), and the pair maps to a metabolizer phenotype.

The four phenotype categories most reports use are poor metabolizer (two no-function alleles), intermediate metabolizer (one no-function or two decreased), normal metabolizer, and ultrarapid metabolizer (extra functional copies or increased-function alleles). The idea that inherited variation in drug-metabolizing enzymes explains large differences in drug exposure between people goes back decades and predates sequencing entirely, built from family studies of isoniazid acetylation and debrisoquine hydroxylation.¹²

How it works

Star alleles exist because single variants are not the useful unit here. Function comes from the combination on one chromosome. CYP2D6*4 is defined by rs3892097 (the 1846G>A splice defect) plus a set of linked upstream and coding variants. If you see rs3892097 as heterozygous in a VCF, you know one chromosome is likely *4, but you do not know what the other chromosome carries until you resolve the full haplotype.

Definitions live at PharmVar, which publishes per-gene tables mapping each numbered allele to variants in HGVS notation relative to a reference sequence. HGVS is the grammar: NM_000106.6:c.100C>T means “in this RefSeq transcript, position 100 of the coding sequence, C becomes T,” and p.(Pro34Ser) is the protein-level consequence. Coordinates differ between genomic (g.), coding (c.), and protein (p.) contexts, which is the most common source of confusion when comparing a PharmVar table to a GRCh38 VCF. PharmVar entries also distinguish core alleles from suballeles: *2 versus *2.001, *2.002, which share the function-defining variants but differ elsewhere.³

Function assignment is a separate, curated step. CPIC expert panels review in vitro and clinical evidence and assign each allele a function label, and those labels get revised. NAT2 was recently recurated into a consolidated set of clinical function assignments after decades of inconsistent slow/rapid acetylator definitions.⁴

In your own data

From a 30x WGS BAM or CRAM, the practical path is a dedicated caller rather than reading the VCF by hand. PyPGx (the successor to Stargazer) handles the hard genes, and the hard genes are the point: CYP2D6 sits next to two pseudogenes, CYP2D7 and CYP2D8, with which it recombines, producing hybrid alleles like *36 and *68 and whole-gene deletions (*5) and duplications. Stargazer showed that calling these requires reading depth-of-coverage alongside variant calls, not variants alone.⁵ A caller that only reads a VCF will silently miss *5 and report you as a normal metabolizer.

Concretely, what to check:

• Run the caller on the aligned reads, not on a joint-called VCF: pypgx run-ngs-pipeline CYP2D6 out-dir --variants your.vcf.gz --depth-of-coverage doc.zip --control-statistics control.zip. The control statistics come from a copy-number-stable region and normalize for library depth.
• Record the PharmVar version in your output. Allele definitions change, and a reanalysis of the same samples across definition versions changed diplotype calls and downstream phenotype assignments for a meaningful fraction of individuals, including SLCO1B1 calls relevant to statin guidance.⁶
• Check phasing. If your pipeline used statistical phasing (Beagle, SHAPEIT) rather than read-backed phasing, heterozygous variants more than a read length apart were assigned by population reference, not observed. For a rare compound heterozygote this fails quietly.
• Check coverage over the gene body. CYP2D6 exon 9 and the GC-rich promoter frequently drop below 10x on PCR-based library preps.
• Do not compare a consumer array report to a WGS call and assume agreement. Arrays genotype a fixed probe set and default everything else to *1, so *1 means “no tested variant found,” not “reference haplotype confirmed.”

Limitations

Star alleles are gene-by-gene and enzyme-centric. They describe metabolism, not drug response, and drug response also depends on transporters, targets, liver and kidney function, other drugs, and diet. Ultrarapid and poor metabolizer phenotypes predict exposure; they do not predict outcome on their own.⁷

Coverage of allele diversity is uneven across ancestries because the definition tables were built largely from European and East Asian cohorts. Novel haplotypes in underrepresented populations get called *1 by default.⁸ NUDT15 is the counterexample worth knowing: the clinically important no-function alleles were characterized largely through East Asian and Hispanic cohorts.³

Finally, a diplotype is a measurement, not a prescription. Translating one into any decision about a medication is a clinician’s job, and the call should be confirmed in a clinical laboratory before anyone acts on it.

Woolf Software builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.