Pharmacogenetic Testing in Psychiatry: What It Measures and How to Get It From Your Own Genome

Pharmacogenetic testing in psychiatry works for a narrow, well-defined purpose: predicting how fast you clear certain antidepressants and antipsychotics, and flagging two HLA alleles associated with severe skin reactions to carbamazepine and oxcarbazepine. It does not predict which drug will make you feel better. The genes that carry actionable guidance are CYP2C19, CYP2D6, and the HLA alleles HLA-B*15:02 and HLA-A*31:01, with weaker support for a handful of others. The pharmacodynamic markers that commercial panels advertise most heavily, notably the serotonin transporter promoter variant (SLC6A4 5-HTTLPR), HTR2A, and COMT, do not have evidence sufficient to guide prescribing ¹². If you have 30x whole-genome sequencing, you can generate the same clinically actionable calls yourself with open-source tools, at higher resolution than most panels, and hand a printed diplotype report to a prescriber. You cannot act on it alone: dosing and drug selection require a clinician, and the CPIC guidelines that translate these calls into recommendations are written for prescribers.

What the genetics measures, and what it does not

The mechanism here is pharmacokinetic. CYP2C19 and CYP2D6 are hepatic cytochrome P450 enzymes that oxidize most SSRIs, tricyclics, and several antipsychotics. Loss-of-function or gain-of-function haplotypes change plasma exposure at a fixed dose, sometimes by several-fold. A CYP2C19 poor metabolizer taking escitalopram reaches substantially higher steady-state concentrations than a normal metabolizer, which is why regulatory labels and CPIC guidance specify reduced starting doses or an alternative agent for that phenotype. This is a claim about drug concentration, and it is well supported by concentration data across cohorts ³¹.

The step from concentration to clinical outcome is where the evidence thins. Randomized trials of combinatorial panels have generally shown small effects: the largest of them did not meet its primary symptom-improvement endpoint, though secondary response and remission measures favored the guided arm, and later trials in veterans found that genotype-informed prescribing changed which drugs were prescribed more than it changed symptom trajectories. Reviews of clinical usability converge on the same conclusion: the actionable content is CYP2C19 and CYP2D6 metabolizer status plus HLA risk alleles, and proprietary weighted algorithms that combine a dozen genes into a color-coded recommendation add interpretive noise rather than validated signal ¹². Separately, psychiatric polygenic scores for diagnosis or treatment response are research instruments. They do not serve as clinical predictors of who responds to which antidepressant ⁴.

The genes with real evidence

CYP2C19 drives clearance of citalopram, escitalopram, sertraline, and the tertiary-amine tricyclics. Those tricyclics are amitriptyline, imipramine and clomipramine. The enzyme also makes a contribution to fluvoxamine clearance. The important haplotypes are common and easy to genotype: *2 (rs4244285), *3 (rs4986893), and the increased-expression promoter variant *17 (rs12248560).

CYP2D6 drives clearance of the following drugs.

• paroxetine
• fluvoxamine (partly)
• venlafaxine
• vortioxetine
• atomoxetine
• aripiprazole
• risperidone
• the secondary-amine tricyclics (nortriptyline, desipramine)

It is also the hardest pharmacogene in the genome to call. The locus sits in a segmental duplication next to the paralog CYP2D7. It carries whole-gene deletions (*5). It also carries duplications and multiplications (*1x2, *2x2, *4x2), and CYP2D6-CYP2D7 hybrid genes. CPIC converts the diplotype into an activity score, then into a phenotype. A score of 0 is a poor metabolizer, and 0.25 to 1.0 is intermediate. A score of 1.25 to 2.25 is normal, and above 2.25 is ultrarapid.

For the antiepileptic mood stabilizers, HLA-B*15:02 carries a strong association with carbamazepine-induced Stevens-Johnson syndrome and toxic epidermal necrolysis. That association is concentrated in individuals of Southeast Asian ancestry. HLA-A*31:01 is associated with carbamazepine hypersensitivity more broadly. These are the pharmacogenetic results in psychiatry with the largest absolute consequence ³.

Calling these genes from your own whole genome

We would use PharmCAT for everything except CYP2D6, and Cyrius for CYP2D6. Start from an aligned CRAM or BAM on GRCh38 at roughly 30x mean depth.

PharmCAT needs a VCF that distinguishes a reference call from a missing call, so joint-called or gVCF-derived input is much better than a variants-only VCF. Call the PharmCAT positions explicitly:

gatk HaplotypeCaller -R GRCh38.fa -I sample.cram \
  -L pharmcat_positions.vcf -ERC BP_RESOLUTION -O sample.g.vcf.gz

Then run the preprocessor, which normalizes representation, splits multi-sample files, and maps your contig names onto PharmCAT’s expectations:

pharmcat_vcf_preprocessor -vcf sample.g.vcf.gz -refFna GRCh38.fa
pharmcat -vcf sample.preprocessed.vcf -reporterJson -reporterHtml

The output includes a named-allele match report, a phenotype table with CPIC activity scores, and an HTML report keyed to CPIC and DPWG guidance. Read-backed phasing improves diplotype resolution where two heterozygous sites could belong to either haplotype, so run WhatsHap first and keep the PS tags:

whatshap phase --reference GRCh38.fa -o phased.vcf.gz sample.preprocessed.vcf sample.cram

PharmCAT deliberately declines to call CYP2D6 from a VCF, because copy number and hybrid structures are invisible to a position-based caller. Use Cyrius, which models depth and paralog-discriminating positions across the CYP2D6/CYP2D7 pair:

star_caller.py --manifest manifest.txt --genome 38 --threads 8 --prefix sample --outDir cyrius_out

Cyrius reports a diplotype plus a raw call string that tells you when the structure is ambiguous. StellarPGx (Nextflow) and Aldy 4 are reasonable cross-checks, and disagreement among them is a signal to treat the result as provisional rather than to pick a favorite. Before trusting any of them, check coverage over CYP2D6 (GRCh38 chr22:42,126,499-42,130,865) with samtools depth. GC-rich stretches and mismapping to CYP2D7 produce local dropouts, and a region averaging under about 15x should not be used for a copy-number call.

For HLA, extract reads from chr6:29.5-33.1 Mb and type with HLA*LA or OptiType at four-digit resolution, which is what *15:02 and *31:01 require. If you have RNA sequencing from blood, arcasHLA on the transcriptome is faster and typically cleaner, since HLA genes are highly expressed in leukocytes.

Why consumer array files are not enough

A 23andMe or AncestryDNA raw file will give you CYP2C19 *2, *3, and *17 reliably, because those are single common SNVs on the array. It will not give you CYP2D6. Imputation fails in segmental duplications, and arrays do not measure copy number at this locus. The rarer loss-of-function alleles that matter in non-European ancestries (*17, *29, *41 patterns) are inconsistently tagged. Third-party interpretation services that report a CYP2D6 phenotype from array data are inferring a structure they did not measure. Sequencing-based calling is the difference between a diplotype and a guess ⁵.

Interpretation caveats a prescriber will ask about

Genotype gives you the enzyme you were born with. Phenoconversion is the reason that is not always the enzyme you have today. Strong CYP2D6 inhibitors, including fluoxetine, paroxetine, and bupropion, can convert a genotypic normal metabolizer into a functional poor metabolizer, so a patient already on one of those drugs may behave as though they carry two null alleles. Inflammation and liver disease shift activity as well. A genotype report is one input to a clinical picture that includes current medications, and this is precisely the kind of integration a psychiatrist or clinical pharmacist does and you should not attempt alone ³⁶.

The second caveat is scope. A negative result across CYP2C19 and CYP2D6 tells you nothing about whether a given antidepressant will work. Treatment response is polygenic and heavily environmental, and no current panel predicts it ⁴⁷.

Cost and regulation

Commercial combinatorial panels (GeneSight, Genomind, and others) are laboratory developed tests run under CLIA certification rather than FDA-cleared devices, and their list prices have historically been in the high hundreds to low thousands of dollars, with patient out-of-pocket caps often set far below list. Single-gene CYP2C19 or CYP2D6 genotyping through a hospital or reference laboratory generally costs a few hundred dollars and returns the same actionable content. In November 2018 the FDA issued a safety communication warning against genetic tests marketed with unapproved claims about response to specific medications, and followed with enforcement letters to individual laboratories, which is the origin of the “FDA warning” people search for. Formal cost-effectiveness evidence for psychiatric pharmacogenetic testing remains mixed and is sensitive to assumptions about how often results change prescribing ⁸⁹.

Questions people also ask

Does genetic testing for psychiatric medications work? It works for predicting drug exposure through CYP2C19 and CYP2D6 metabolizer status and for flagging HLA hypersensitivity risk. Evidence that a combinatorial panel improves symptom outcomes is modest and inconsistent across randomized trials ¹².

Has GeneSight been approved by the FDA? No. It is offered as a laboratory developed test under CLIA, and its combinatorial interpretive algorithm has not been FDA cleared or approved. Some individual gene-drug pairs it reports do appear in FDA drug labeling.

Can any doctor order pharmacogenetic testing? Generally yes. Any licensed prescriber can order these tests, and psychiatrists commonly do, though interpretation with a clinical pharmacist or pharmacogenomics service is more useful than the vendor’s color-coded report.

How much does it cost in Canada? Provincial plans typically do not fund psychiatric pharmacogenetic panels, so patients usually pay out of pocket, in the range of a few hundred Canadian dollars for targeted genotyping. CADTH has reviewed the technology and the evidence base for exactly this reason.

Which genes should a minimal panel include? CYP2C19, CYP2D6, HLA-B*15:02, and HLA-A*31:01 cover nearly all of the actionable content in psychiatry. Everything beyond that is currently research. It is not guidance ⁶.

Footnotes

1. Roos van Westrhenen, Katherine J. Aitchison, Magnus Ingelman-Sundberg, et al. Pharmacogenomics of Antidepressant and Antipsychotic Treatment: How Far Have We Got and Where Are We Going?. Frontiers in Psychiatry, 2020. https://doi.org/10.3389/fpsyt.2020.00094 ↩ ↩² ↩³ ↩⁴

2. Ron H. N. van Schaik, Daniel J. Müller, Alessandro Serretti, et al. Pharmacogenetics in Psychiatry: An Update on Clinical Usability. Frontiers in Pharmacology, 2020. https://doi.org/10.3389/fphar.2020.575540 ↩ ↩² ↩³

3. Aminah Alchakee, Munazza Ahmed, Leen Eldohaji, et al. Pharmacogenomics in Psychiatry Practice: The Value and the Challenges. International Journal of Molecular Sciences, 2022. https://doi.org/10.3390/ijms232113485 ↩ ↩² ↩³

4. Patrick F. Sullivan, Arpana Agrawal, Cynthia M. Bulik, et al. Psychiatric Genomics: An Update and an Agenda. American Journal of Psychiatry, 2018. https://doi.org/10.1176/appi.ajp.2017.17030283 ↩ ↩²

5. Katherine A. Johansen Taber, Barry D. Dickinson, Modena Wilson. The Promise and Challenges of Next-Generation Genome Sequencing for Clinical Care. JAMA Internal Medicine, 2014. https://doi.org/10.1001/jamainternmed.2013.12048 ↩

6. Filippo Corponi, Chiara Fabbri, Alessandro Serretti. Pharmacogenetics in Psychiatry. Advances in Pharmacology, 2018. https://doi.org/10.1016/bs.apha.2018.03.003 ↩ ↩²

7. Margret R. Hoehe, Deborah J. Morris-Rosendahl. The role of genetics and genomics in clinical psychiatry. Dialogues in Clinical Neuroscience, 2018. https://doi.org/10.31887/dcns.2018.20.3/mhoehe ↩

8. Katherine Payne, Sean P. Gavan, Stuart J. Wright, et al. Cost-effectiveness analyses of genetic and genomic diagnostic tests. Nature Reviews Genetics, 2018. https://doi.org/10.1038/nrg.2017.108 ↩

9. Kurt Christensen, Dmitry Dukhovny, Uwe Siebert, et al. Assessing the Costs and Cost-Effectiveness of Genomic Sequencing. Journal of Personalized Medicine, 2015. https://doi.org/10.3390/jpm5040470 ↩