The Best Genetic Test for Antidepressants Is a CYP2D6/CYP2C19 Genotype

If you want a genetic test related to antidepressants, testing for which drug will work is not defensible. The defensible thing to test is how fast you clear the drug. That means accurate diplotype calls for two genes, CYP2C19 and CYP2D6, which between them handle most SSRIs and tricyclics. Those two genes have published dosing guidelines from the Clinical Pharmacogenetics Implementation Consortium (CPIC) and matching labels from the FDA and the Dutch Pharmacogenetics Working Group. Everything beyond that, including the color-coded “consider with caution” bins on commercial psychiatric panels, rests on evidence that does not support gene-by-gene drug selection. A diplotype is your pair of star alleles, one per chromosome, such as CYP2C19 *1/*17.

The practical consequence is that you do not need a $400 branded psychiatric panel. You need a correct genotype at two loci and a clinician who will interpret it alongside your history. If you already have whole-genome sequencing, you can produce that genotype yourself with open tools, and the calls will usually be better than what a targeted array returns.

Why response-prediction panels fail

The appeal of a panel like GeneSight is that it promises to answer the question patients ask: which antidepressant will work for me. The genetics does not support that answer. Efficacy of antidepressants is highly polygenic. The candidate variants marketed as predictors are SLC6A4 5-HTTLPR, HTR2A, COMT, BDNF, and MTHFR. They have not replicated at effect sizes that would change a prescription.

The clearest demonstration came from the large pharmacogenetic arm of STAR*D. A genome-wide association study of citalopram response in that cohort produced no variants reaching genome-wide significance, with the top signals falling short and failing replication ¹. Earlier candidate-gene work in the same cohort, including an association study of FKBP5, illustrates the pattern: nominally significant findings in specific subgroups that do not translate into a prescribing rule ². Two decades of this literature have produced consistent pharmacokinetic findings and inconsistent pharmacodynamic ones ³.

A recent systematic review of antidepressant pharmacogenetic studies in major depressive disorder reached the same conclusion. Metabolizer genes show reproducible effects on drug exposure and tolerability, while receptor and transporter candidates remain unreplicated ⁴. Reviews of the field have long flagged the same structural problems: small samples, heterogeneous outcome definitions and weak phenotyping of “response” itself ⁵⁶.

We treat combinatorial panels as opaque for a specific technical reason. They return a proprietary weighted score across many genes rather than the underlying diplotypes. You cannot audit the call, you cannot recompute it as PharmVar allele definitions change and you cannot separate the CYP2C19 result you should trust from the HTR2A result you should ignore.

What does hold up

The pharmacokinetic story is narrow and solid. CYP2C19 is the main route for citalopram, escitalopram and sertraline. CYP2D6 dominates for paroxetine, fluvoxamine partly, venlafaxine and most tricyclics. Loss-of-function or gain-of-function alleles in these genes shift plasma drug concentration severalfold, which is a measurable pharmacokinetic fact rather than an inferred behavioral one ⁷.

The allele frequencies are high enough to matter at an individual level. CYP2C19*2 is a splice-defect null allele found in roughly 15% of chromosomes in European ancestry populations and higher in East Asian populations. CYP2C19*17 is a promoter variant associated with increased expression and is also common in European and African ancestry groups. Combining them gives a wide spread of predicted metabolizer phenotypes across any given population, from poor to ultrarapid.

Knowing your status tells a prescriber whether a standard starting dose is likely to produce unusually high or unusually low exposure for a given drug. It gives no answer on which drug to take. That is a conversation with a physician or a pharmacist, and CPIC’s guidelines are written for them. The argument for genotyping now rather than waiting for better science is that the pharmacokinetic evidence is already actionable while the pharmacodynamic evidence is not ⁸.

Calling CYP2C19 and CYP2D6 from your own WGS

CYP2C19 is easy. It is a well-behaved locus, the important alleles are defined by single-nucleotide variants and a standard 30x short-read genome with a GATK or DeepVariant VCF will resolve it.

CYP2D6 is one of the hardest genes in the pharmacogenome. It sits next to the pseudogene CYP2D7, with which it shares about 94% sequence identity. Real samples carry whole-gene deletions (*5), tandem duplications with copy numbers up to six and CYP2D6/CYP2D7 hybrid genes. Short-read aligners mismap across that region, so a plain variant caller will silently report a wrong diplotype.

The pipeline we would use on a 30x short-read genome aligned to GRCh38:

1. Run Cyrius on the CRAM or BAM directly. It was built for CYP2D6 from short reads, models copy number from read depth and handles the common hybrid structures. Output is a diplotype string such as *1/*4x2.
2. Run PharmCAT. Preprocess first with pharmcat_vcf_preprocessor to normalize and split multiallelics against the reference FASTA, then run the named allele matcher and phenotyper. PharmCAT does not call CYP2D6 from the VCF by default. Feed it the Cyrius result with --matcher-outside-calls, formatted as a two-column TSV of gene and diplotype.
3. Read the JSON, not just the HTML. *.match.json shows which allele definitions were matched and which positions were missing, which is where quiet failures live.
4. Cross-check CYP2D6 with a second caller, Aldy or StellarPGx. Disagreement between callers on CYP2D6 is common and usually means a structural variant. That is the case to escalate to a clinical laboratory rather than resolve yourself.

Two failure modes deserve attention. First, phase. A *2/*17 and a *2+*17/*1 configuration are different, and unphased short-read data cannot always distinguish them. Read-backed phasing with whatshap helps when heterozygous sites are close together. PacBio HiFi or Oxford Nanopore reads spanning the whole locus solve it properly. Second, no-call inflation. If a position defining a rare allele has low depth, the matcher will default toward *1, and you will read a normal-metabolizer report that reflects missing data rather than a normal genotype.

Choosing between WGS, a targeted panel and a clinical test

A clinical CLIA pharmacogenomic test is the right choice if the result will be used in prescribing. It is validated, it usually includes CYP2D6 copy number by a targeted assay and a clinician can put it in your chart.

Whole-genome sequencing is the better choice if you want the data. One sequencing run covers CYP2C19 and CYP2D6. It also covers the rest of the pharmacogenome that matters elsewhere in medicine.

• DPYD
• TPMT
• NUDT15
• SLCO1B1
• CYP2C9 with VKORC1
• HLA-B*57:01 and HLA-B*15:02 by imputation

It is reanalyzable as allele definitions get updated, which targeted arrays are not.

Genotyping arrays are the weakest option and the most common. They interrogate only a preselected list of positions. If your allele is not on the list, you are assigned the reference allele by default. Coverage gaps fall unevenly across ancestries. Arrays designed around European allele sets under-detect CYP2D6*17, *29 and *41, which are common in African ancestry populations. Ancestry-dependent sensitivity is a known limitation of candidate-variant approaches in psychiatric genetics generally ⁹.

Questions people also ask

Does GeneSight tell me which antidepressant will work? No. Its pharmacokinetic components (CYP2C19, CYP2D6) reflect real biology, but the combinatorial score that sorts drugs into categories depends on pharmacodynamic candidate genes without replicated effects on response ⁴³.

Is pharmacogenomic testing recommended before starting an antidepressant? Guidelines provide dosing guidance once a CYP2C19 or CYP2D6 result exists. Routine testing for everyone is not something they recommend. Whether to test before a first prescription is a clinical decision to make with your prescriber.

Can I get my metabolizer status from 23andMe or AncestryDNA raw data? Partially and unreliably. Those arrays include some CYP2C19 positions but do not resolve CYP2D6 copy number or hybrids, so a third-party interpretation of that file can be confidently wrong.

What about measuring drug levels instead? Therapeutic drug monitoring measures plasma concentration directly and reflects genotype plus everything else, including interacting drugs and liver function. It answers the exposure question more directly than genotype does. Ordering and interpreting it requires a clinician.

Does MTHFR genotype matter here? No. MTHFR variants appear on psychiatric panels but do not support antidepressant selection. Treat that section of any report as noise.

Footnotes

1. Holly A. Garriock, Jeffrey B. Kraft, Stanley I. Shyn, et al. A Genomewide Association Study of Citalopram Response in Major Depressive Disorder. Biological Psychiatry, 2010. https://doi.org/10.1016/j.biopsych.2009.08.029 ↩

2. Magnus Lekman, Gonzalo Laje, Dennis Charney, et al. The FKBP5-Gene in Depression and Treatment Response—an Association Study in the Sequenced Treatment Alternatives to Relieve Depression (STAR*D) Cohort. Biological Psychiatry, 2008. https://doi.org/10.1016/j.biopsych.2007.10.026 ↩

3. Chiara Fabbri, Giulia Di Girolamo, Alessandro Serretti. Pharmacogenetics of antidepressant drugs: An update after almost 20 years of research. American Journal of Medical Genetics Part B: Neuropsychiatric Genetics, 2013. https://doi.org/10.1002/ajmg.b.32184 ↩ ↩²

4. Caroline W. Grant, Karina Delaney, Linsey E. Jackson, et al. Comprehensive Characterization of Antidepressant Pharmacogenetics: A Systematic Review of Studies in Major Depressive Disorder. Clinical and Translational Science, 2025. https://doi.org/10.1111/cts.70255 ↩ ↩²

5. Chiara Fabbri, Concetta Crisafulli, Marco Calabrò, et al. Progress and prospects in pharmacogenetics of antidepressant drugs. Expert Opinion on Drug Metabolism & Toxicology, 2016. https://doi.org/10.1080/17425255.2016.1202237 ↩

6. Diane B. Miller, James P. O’Callaghan. Personalized medicine in major depressive disorder — Opportunities and pitfalls. Metabolism, 2013. https://doi.org/10.1016/j.metabol.2012.08.021 ↩

7. Kristina A. Malsagova, Tatyana V. Butkova, Arthur T. Kopylov, et al. Pharmacogenetic Testing: A Tool for Personalized Drug Therapy Optimization. Pharmaceutics, 2020. https://doi.org/10.3390/pharmaceutics12121240 ↩

8. Kristi Krebs, Lili Milani. Translating pharmacogenomics into clinical decisions: do not let the perfect be the enemy of the good. Human Genomics, 2019. https://doi.org/10.1186/s40246-019-0229-z ↩

9. Kathryn L. Evans, David J. Porteous, Walter J. Muir, et al. Nuts and bolts of psychiatric genetics: building on the Human Genome Project. Trends in Genetics, 2001. https://doi.org/10.1016/s0168-9525(00)02170-3 ↩