Why these three genes make a coherent doctoral project
Pharmacogenomics offers doctoral candidates an unusual advantage: a small number of gene and drug pairs already carry prescribing guidance strong enough to act on. That turns a descriptive genetics project into a clinically interpretable one. CYP2C19, CYP2C9 and SLCO1B1 are a defensible trio because each has a published implementation guideline, each affects a widely prescribed cardiovascular or metabolic medicine, and all three can be captured by a single genotyping strategy.
For CYP2C19, the clinical anchor is clopidogrel. Clopidogrel is a prodrug requiring CYP2C19-mediated bioactivation, so intermediate and poor metabolisers achieve reduced active metabolite exposure. Current guidance advises avoiding standard-dose clopidogrel in these phenotypes after acute coronary syndrome or percutaneous coronary intervention, in favour of prasugrel or ticagrelor where not contraindicated (Lee et al., 2022). For CYP2C9, the best-established pair is warfarin, where dosing algorithms combine CYP2C9 reduced-function alleles with the VKORC1 promoter variant rs9923231 and clinical covariates (Johnson et al., 2017). CYP2C9 also moderates fluvastatin exposure (Cooper-DeHoff et al., 2022). For SLCO1B1, the variant of interest is c.521T>C (rs4149056), which reduces hepatic uptake of several statins and increases the risk of statin-associated musculoskeletal symptoms, most markedly with simvastatin (Cooper-DeHoff et al., 2022).
The Emirati case for studying them is empirical rather than rhetorical, and the relevant evidence comes from research groups based in the country. Targeted resequencing of 100 pharmacogenes in a healthy Emirati sample, by a team led from United Arab Emirates University, found rs4149056 at a minor allele frequency of 0.187, higher than in several reference populations, and identified reduced-function CYP2C9 alleles in roughly a third of participants (Al-Mahayri et al., 2020). An exome-based study of clopidogrel-related genes by the same university's ASPIRE Precision Medicine Research Institute sequenced 298 healthy Emiratis, reported CYP2C19*2 at a minor allele frequency of 0.193 and *3 at 0.004, and classified about 37 per cent of participants as intermediate or poor CYP2C19 metabolisers (Khasawneh et al., 2024). Work in the neighbouring Qatari population, using whole-genome sequencing of several thousand individuals, reported actionable SLCO1B1 and CYP2C9 diplotype frequencies above global averages (Jithesh et al., 2022). The national Emirati Genome Programme is building population-scale sequence data that should eventually support finer estimates, although its pharmacogenomic analyses were still appearing as preprints when this article was reviewed. The gap your thesis addresses is therefore not whether variation exists, but how it is distributed, how it maps onto prescribing, and what follows for practice.
Two design lessons follow from how those teams worked. First, each stated what its data could not show. The Emirati clopidogrel study was conducted in healthy volunteers, so it could describe phenotype distributions but not clinical consequences, and the authors also reported that exome coverage was inadequate for three promoter variants they had intended to examine (Khasawneh et al., 2024). Second, the choice of assay sets the ceiling on what can be concluded: a panel or exome that misses an allele cannot find it, which is why the limitation belongs in the methods chapter rather than the discussion.
When you state the gap, be specific. Avoid the claim that Gulf populations are unstudied, which the work above contradicts. Say instead which estimates are missing: diplotype-level phenotype distributions, frequencies of alleles beyond the common two or three, or the link between genotype and the dose actually prescribed. A precise gap is easier to defend at the viva than a general one.
Turning the topic into an answerable question
Most weak pharmacogenomics proposals fail at the same point: they state a topic rather than a question. A doctoral question must specify a population, an exposure, a comparison where relevant, and an outcome that your data can actually measure. The practical constraint is access. What you can access determines which of three designs is honest.
| Design | What it requires | What it can conclude |
|---|---|---|
| Population genotype survey | A defined Emirati sample with consent for pharmacogene genotyping | Allele and diplotype frequencies, predicted phenotype distribution, comparison with other populations |
| Genotype and prescribing linkage | Genotype data plus validated prescribing or dispensing records | Proportion of patients receiving a drug for which their genotype implies a guideline change |
| Genotype and outcome study | Genotype plus adjudicated clinical outcomes and adequate numbers | Association between phenotype group and an outcome such as bleeding, recurrent ischaemia or muscle symptoms |
The third design is the most clinically compelling and the most likely to be underpowered. Outcome studies for rare but important endpoints need participant numbers that a single candidate rarely assembles within a doctoral timeline. A defensible compromise is to make the survey or linkage study the primary objective, and treat an outcome analysis as an exploratory secondary objective with the power limitation declared in advance. An Emirati warfarin study illustrates what a modest, well-specified sample can deliver: with 90 treated patients and 117 controls, a regression model including rs9923231, age and a second VKORC1 variant accounted for about half the variance in average warfarin dose (Al-Mahayri et al., 2019).
Getting allele calling and phenotype assignment right
This is the section examiners probe hardest, because it is where undergraduate-level genotyping is most easily mistaken for pharmacogenomics. Two distinct steps are involved, and each needs its own methodological justification.
Step one: define alleles as haplotypes
Star alleles are haplotypes defined by specific variant combinations, curated and versioned by the Pharmacogene Variation Consortium (Gaedigk et al., 2018). CYP2C19*2 is defined by rs4244285, a splice-defect variant, and CYP2C19*3 by rs4986893, while CYP2C19*17 is an increased-function allele tagged by rs12248560. For CYP2C9, the two classic reduced-function alleles are *2 (rs1799853) and *3 (rs1057910). The SLCO1B1 decreased-function haplotypes, including *5 and *15, carry c.521T>C. Your methods chapter should state the PharmVar version used, list every variant your assay interrogates, and state explicitly which alleles your panel cannot distinguish. A genotype that excludes less common alleles will default them to the reference allele, biasing the normal metaboliser group upward.
Step two: translate diplotypes into standard phenotypes
Phenotype labels must follow the consensus terms agreed for clinical pharmacogenetic reporting: ultrarapid, rapid, normal, intermediate and poor metaboliser for drug-metabolising enzymes, and normal, decreased or poor function for transporters such as the product of SLCO1B1 (Caudle et al., 2017). Record the translation rule and its source rather than describing the mapping in prose, and preregister how ambiguous diplotypes and indeterminate calls will be handled. Reviewers can then reproduce every phenotype count from your genotype table.
- Test your own draft for reproducibility: give the methods chapter and your variant call file to a colleague outside your group and ask them to regenerate your phenotype counts. If they cannot, the chapter is incomplete, however carefully the analysis was actually done.
- State which alleles your assay cannot detect, as the Emirati clopidogrel study did for the promoter variants its exome data could not cover (Khasawneh et al., 2024).
- Record the PharmVar version and the translation table version, since both are revised.
Designing around consanguinity and population structure
Emirati and wider Gulf populations have historically high rates of consanguineous marriage, which raises homozygosity and can produce genotype distributions that depart from Hardy-Weinberg expectation for reasons unrelated to assay performance. Treat equilibrium testing as a descriptive and diagnostic step rather than a quality gate. Use an exact test rather than the chi-square approximation, particularly for alleles with low counts, where the asymptotic test is unreliable (Wigginton et al., 2005).
- Report observed and expected genotype counts for every variant, with the exact test p value and the test used.
- Interpret departures jointly with call rate, read depth and duplicate concordance, so assay artefact and genuine population effects are distinguishable.
- Estimate relatedness or inbreeding coefficients where genome-wide data allow, and state how related participants were handled in frequency estimates.
- Give allele frequencies with confidence intervals, since a sample of one or two hundred yields wide intervals for alleles below about five per cent.
- Prespecify the comparison populations and the data release used, rather than selecting comparators after seeing your results.
Recruitment framing matters as much as analysis. Define Emirati ancestry by a documented, stated criterion, such as self-reported national citizenship with ancestry reported for a specified number of generations, and record it as a variable rather than treating it as a filter applied silently. Evidence suggests that genetically inferred ancestry and self-reported national identity do not align perfectly in the region, so a thesis that conflates the two invites a straightforward challenge.
Justifying the sample size
Sample size justification differs by objective, and a common error is to use a power calculation designed for association testing when the primary objective is estimation. For a frequency survey the appropriate calculation is precision based: choose the width of confidence interval you are prepared to defend for the least common allele of interest, then derive the required number of chromosomes. For a phenotype distribution, the quantity of interest is the proportion of participants in actionable phenotype groups, and published Emirati and Gulf estimates give a reasonable planning value (Khasawneh et al., 2024; Jithesh et al., 2022).
For an outcome analysis, power depends on the frequency of the actionable phenotype, the event rate in the comparison group and the effect size you consider clinically meaningful. State all three. If the resulting requirement exceeds your realistic recruitment, say so in the proposal and reposition the analysis as exploratory. Examiners accept an honest power limitation; they rarely accept a post hoc explanation for a null result that was never powered to be informative.
Ethics, governance and data stewardship
Genomic research in the United Arab Emirates operates under national regulation as well as institutional review, and the approvals pathway can take longer than candidates expect. Build the approval timeline into the research plan rather than treating it as administrative overhead, and confirm current requirements with your institutional research office and the relevant national authority at the time you apply, since the regulatory framework in the region has been revised in recent years.
- Specify what consent covers: the named genes, any wider sequencing, future reuse, and whether data may be shared outside the country.
- State in advance your policy on actionable incidental findings, and who would return them.
- Describe the data management plan concretely: where identifiable data are held, who can access them, and how long they are retained.
- Address community engagement where participants are recruited through families or tribal networks, since consanguinity means individual results can carry information about relatives.
- Record the approval references and protocol version in the thesis, so the audit trail is visible to examiners.
Be careful with clinical implication in your writing. A frequency study establishes that a genotype-based recommendation would apply to a given share of a population. It does not establish that implementing testing improves outcomes in that setting, which is a separate question requiring a different design. Keeping that distinction explicit protects the thesis from overclaiming and often supplies the strongest future-work section.
Writing the design so it survives examination
Reviewers and examiners look for the same chain of reasoning: a specific gap, a question that closes part of it, a design capable of answering that question, an analysis plan fixed before the data arrive, and conclusions no stronger than the design supports. Write the methods chapter early, in enough detail that it doubles as your working protocol, and revise it as decisions change rather than reconstructing it afterwards.
- State the gene and drug pairs and the guideline that makes each clinically interpretable (Lee et al., 2022; Cooper-DeHoff et al., 2022; Johnson et al., 2017).
- Name the population, the ancestry criterion and the recruitment route.
- Specify the genotyping platform, the variants interrogated and the PharmVar version used for allele definitions (Gaedigk et al., 2018).
- Give the diplotype to phenotype translation rule and its source (Caudle et al., 2017).
- Prespecify the statistical plan, including equilibrium testing, interval estimation and the comparison populations (Wigginton et al., 2005).
- Declare the limitations of the design before presenting results, not after.
A thesis built this way makes a modest but durable contribution: locally valid frequency and phenotype estimates, an explicit account of how they were produced, and a clear statement of what would need to be done next to turn them into prescribing practice. That is a stronger doctoral claim than an ambitious outcome study the sample could never support.
Key takeaways
- Anchor the project in gene and drug pairs that already carry implementation guidance, so the genetics has a clinical interpretation (Lee et al., 2022; Cooper-DeHoff et al., 2022).
- Treat star-allele calling as haplotype assignment against a versioned reference, and state which alleles your panel cannot detect (Gaedigk et al., 2018).
- Use the consensus phenotype terminology so your counts are comparable with other studies (Caudle et al., 2017).
- Interpret Hardy-Weinberg departures in light of consanguinity and structure as well as assay quality, using an exact test (Wigginton et al., 2005).
- Match the sample size calculation to the objective: precision for frequency estimation, power for association testing.
- Existing Emirati and Gulf data justify the hypothesis but do not replace local sampling (Al-Mahayri et al., 2020; Khasawneh et al., 2024; Jithesh et al., 2022).
