26-SNP Panel Aids Guiding Androgenetic Alopecia Therapy and Provides Insight into Mechanisms of Action

Abstract

Inter-individual variability in response to androgenetic alopecia (AGA) therapies remains a therapeutic challenge. This study evaluated the clinical and mechanistic utility of a 26-SNP pharmacogenetic panel in guiding treatment decisions. By using a database containing data from 252 individuals stratified by genotype, overall response rates were high (85.6–91.0%), exceeding published benchmarks for minoxidil, finasteride, and dutasteride. SNP association analysis identified rs1042028 in SULT1A1 as a robust predictor of poor response across all three drugs (minoxidil: p = 2.4 × 10⁻⁸, OR = 0.09; dutasteride: p = 0.023, OR = 0.21; finasteride: p = 0.025, OR = 0.11). For dutasteride, the TT genotype of rs39848 in SRD5A1 was also associated with reduced efficacy (p = 0.018, OR = 0.02). SNP–SNP interaction analysis revealed significant epistatic effects between genes involved in prostaglandin signalling and oxidative stress response, including PTGFR × MUC1 (p = 5.38 × 10⁻⁶) and GPR44 × FUT2 (p = 9.4 × 10⁻⁵). Network enrichment analyses further supported drug-specific mechanistic clusters. Importantly, no statistically significant differences in response were observed between pharmacogenetically guided treatment groups (p > 0.1), suggesting successful genotype-based alignment. Together, these findings demonstrate that SNP-informed therapy can enhance efficacy, clarify drug mechanisms, and provide a foundation for precision treatment in AGA.

1. Introduction

Inter-individual variability in drug response represents a major challenge in clinical pharmacology, with implications for both therapeutic efficacy and the risk of adverse drug reactions [1,2,3,4]. Such variability is driven by a combination of environmental, physiological, and genetic factors, the latter increasingly recognised as critical determinants of pharmacokinetics and pharmacodynamics.

Pharmacogenetics examines how inherited genetic variants affect drug metabolism and response, notably in enzymes (e.g., CYP2D6, CYP2C19, CYP2C9), transporters (e.g., SLCO1B1), and targets (e.g., VKORC1) [2,3,4].

These variants influence drug levels and outcomes, such as bleeding risk with warfarin (VKORC1, CYP2C9) and reduced clopidogrel efficacy (CYP2C19). Their relevance has led to over 200 drugs incorporating pharmacogenetic data in regulatory labels (FDA, EMA) and to the publication of clinical guidelines by CPIC and DPWG [2,5,6,7,8,9,10,11,12].

Despite advances, pharmacogenetic testing adoption remains limited. Pre-emptive genotyping has proven beneficial, as shown in real-world and trial settings [13,14]. The PREPARE trial reported a 30% reduction in adverse drug reactions using a 12-gene panel covering key metabolising enzymes and HLA alleles [15,16,17,18,19]. Institutional programs like PREDICT (Vanderbilt) and RIGHT 10K (Mayo Clinic) have embedded decision support into electronic health records for guiding statin, antidepressant, anticoagulant, and cancer therapy [20,21,22,23,24]. Economic analyses suggest cost savings by avoiding adverse events. However, barriers persist, including clinician training gaps, reimbursement variability, a need for population-specific allele data, and workflow integration. Overcoming these is critical for broader, effective pharmacogenetic implementation [2,25,26,27].

Pharmacogenetic-guided prescribing tailors drug choice and dosage to a patient’s genetic profile to enhance efficacy and reduce adverse effects. Variants in genes like CYP450, SLCO1B1, and HLA can predict reduced response or increased toxicity. Incorporating this data supports more precise, safer, and more effective therapy [25,28].

Minoxidil is a first-line treatment for androgenetic alopecia, with proven efficacy in both topical and oral forms. In a 12-month study of 904 men, 62% had reduced bald area and 84% showed some regrowth with 5% topical minoxidil [29]. Trials comparing 2% and 5% solutions found the latter increased mean hair density by ~70% over placebo, with 45% showing visible regrowth. In alopecia areata, 5% formulations led to ≥70% regrowth in 33% of patients, rising to 52% with consistent use. A meta-analysis of 2933 patients on oral minoxidil showed 35% had significant, 47% moderate, and 26% stable outcomes. These results confirm minoxidil’s effectiveness, while highlighting response variability [30,31,32,33,34,35,36].

Finasteride and dutasteride are oral 5α-reductase inhibitors that reduce dihydrotestosterone (DHT) levels in scalp tissue, addressing a primary driver of follicular miniaturisation in androgenetic alopecia. Finasteride selectively inhibits type II 5α-reductase, achieving approximately 60–70% reduction in scalp DHT concentrations, with long-term observational studies in over 3000 men indicating that 11.1% achieved significant regrowth, 36.5% moderate improvement, and 39.5% slight increases in hair density over three years of continuous therapy [37,38]. Dutasteride offers broader inhibition of both type I and II isoenzymes, achieving systemic DHT suppression of over 90%, and has demonstrated superior efficacy in meta-analyses and randomised trials, consistently producing greater hair count gains compared to finasteride. In women under 50, dutasteride has shown greater improvement in hair thickness than finasteride over three years of therapy, while mesotherapy approaches in female patients have yielded 62.8% improvement from baseline compared to 17.5% in controls [39,40]. Notably, combination therapy with oral minoxidil and finasteride has demonstrated additive benefits in men, with a retrospective evaluation showing that 92.4% of patients maintained or improved hair density over 12 months, and 57.4% experienced marked improvements, supporting the potential of combined pharmacologic strategies in achieving superior outcomes [32,33,41,42,43].

Hormonal therapies such as estradiol and spironolactone represent important systemic treatment options, particularly in female-pattern hair loss where androgen-mediated mechanisms play a variable role. Estradiol-based topical formulations, often combined with minoxidil, have demonstrated clinical improvement in hair density in postmenopausal women, although their efficacy tends to be lower than that of 5α-reductase inhibitors [43]. Spironolactone, an aldosterone antagonist with anti-androgen receptor blocking activity, is widely employed in women with hyperandrogenic features or in those demonstrating poor response to monotherapy with minoxidil. It offers systemic anti-androgenic effects while being generally well tolerated at typical oral doses, and is often integrated into multi-modal regimens aiming to reduce progression of follicular miniaturization and support maintenance of hair density in female patients [44,45,46].

Integrating pharmacogenetic testing into the management of androgenetic alopecia offers a compelling opportunity to personalize therapy by identifying genetic variants that influence drug metabolism, target sensitivity, or activation pathways. Variants in genes encoding 5α-reductase isoenzymes or androgen receptors may modulate individual responses to finasteride and dutasteride, while polymorphisms in SULT1A1 influence the activation of minoxidil in hair follicles. By stratifying patients according to these genetic profiles, clinicians could better predict therapeutic efficacy and optimize drug selection and dosing, thereby improving response rates and minimizing adverse effects within a precision-medicine framework for hair loss treatment [5,6,30,47,48].

Personalised medicine is gaining relevance in dermatology, addressing variability in treatment response and tolerability. Pharmacogenomic testing helps guide systemic therapies, while compounding pharmacies tailor topical formulations by adjusting concentrations, vehicles, or actives, to improve adherence [49,50,51]. These strategies are used in inflammatory dermatoses and skin cancers. Given the heterogeneity in response to androgenetic alopecia treatments—especially 5α-reductase inhibitors and minoxidil—applying personalised approaches here is a logical next step. Combining pharmacogenomics with customised formulations may improve efficacy, safety, and adherence [13,49,50,51,52].

Minoxidil is a prodrug activated by follicular sulfotransferases, particularly SULT1A1, to form minoxidil sulfate—the active compound responsible for hair growth. It likely acts by opening ATP-sensitive potassium (K-ATP) channels in dermal papilla cells, inducing membrane hyperpolarisation, triggering anagen entry, and prolonging its duration [53]. In vitro, minoxidil promotes dermal papilla cell proliferation, increases VEGF and prostaglandin E2 levels, and may enhance perifollicular vascularisation and follicle survival. It also appears to activate β-catenin signalling in keratinocytes, though this is based on cell models [34,53]. Additionally, minoxidil stimulates adipose-derived stem cell migration, angiogenesis, and secretion of growth factors (e.g., CXCL1, PD-ECGF, PDGF-C), which promote dermal papilla proliferation via ERK/MAPK pathways. These mechanisms, together with variability in SULT1A1 activity, may explain differences in clinical response [34,53,54].

Finasteride and dutasteride are competitive inhibitors of 5α-reductase, the enzyme responsible for the conversion of testosterone to dihydrotestosterone (DHT), the principal androgen driving follicular miniaturisation in androgenetic alopecia. Finasteride selectively inhibits the type II isoenzyme, predominant in the inner root sheath of hair follicles, reducing scalp DHT concentrations by approximately 60–70%, thereby lowering androgen receptor activation in dermal papilla cells and slowing miniaturisation [39]. In contrast, dutasteride inhibits both type I and type II isoenzymes, the former being abundant in sebaceous glands and also present in follicular structures, achieving more complete systemic DHT suppression of approximately 90–95% and demonstrating greater efficacy in hair count and thickness gains in clinical trials [55]. Pharmacodynamic studies have confirmed that dutasteride is approximately three times more potent than finasteride at inhibiting type I 5α-reductase and around 100 times more potent against type II, contributing to its superior clinical performance in reversing miniaturisation [39]. The long-lasting binding of both drugs to their enzyme targets ensures sustained suppression of DHT production even after plasma clearance. This dual isoenzyme inhibition strategy underscores the rationale for dutasteride’s use, particularly in patients showing suboptimal response to finasteride, despite both agents sharing the fundamental mechanism of reducing local androgenic signalling in scalp hair follicles [39,56].

Understanding minoxidil’s mechanism supports integrating pharmacogenomics into AGA treatment. Variability in response is partly linked to follicular sulfotransferase activity, suggesting SULT1A1 genotyping could guide patient selection and dosing. Likewise, polymorphisms in 5α-reductase or androgen receptor genes may influence response to finasteride and dutasteride. Our planned bioinformatics analysis will assess these variants and their links to treatment outcomes, focusing on androgen metabolism, growth factor signalling, and inflammation. Integrating genetic and clinical data aims to shift AGA therapy toward precision medicine—optimising drug choice, dosing, and combinations, to enhance efficacy and reduce side effects [34,53].

This paper will investigate these pharmacogenetic and mechanistic aspects of androgenetic alopecia treatment in detail, using bioinformatics analyses to explore genetic variants and their potential impact on drug response. By examining single-nucleotide polymorphisms, gene–gene interactions, and relevant metabolic pathways, we aim to clarify sources of variability in efficacy for minoxidil, finasteride, and dutasteride, and to assess how these mechanisms may be influenced by individual genetic profiles. This approach seeks to inform a more precise, personalised strategy for managing androgenetic alopecia, supporting the rational selection and optimisation of therapies to improve clinical outcomes.

2. Materials and Methods

2.1. Database and Ethics Compliance

This study used only fully anonymised clinical and genetic data obtained from an internal database generated as part of routine industry activity, with all direct and indirect identifiers removed to prevent any possibility of re-identification. No patient contact was involved, and no new data collection was performed for the purpose of this study. The use of anonymised data for research is consistent with the principles of the Declaration of Helsinki (Paragraphs 23–25 and 32), which permit retrospective research without additional ethical review when individuals cannot be identified, and aligns with the General Data Protection Regulation (GDPR, Recital 26), under which fully anonymised data fall outside the scope of personal data regulation. This approach complies with EU and international standards for the ethical use of health data, and ensures full respect for regulatory and ethical principles governing retrospective research using anonymised datasets.

As this was a retrospective analysis of anonymised data, no direct inclusion or exclusion of patients occurred. Instead, data points were filtered from the total dataset for points that had undergone genotyping, carried a prior clinical diagnosis of AGA with documented staging, and for whom follow-up data was available. Data entries with incomplete clinical, genetic, or treatment follow-up information were excluded from the analysis. Details regarding treatment categorisation, outcome definitions, and response grading are provided in Section 2.2.

2.2. Data Structure and SNPs

The dataset consisted solely of fully anonymised clinical and genetic information from 252 individuals diagnosed with androgenetic alopecia, originally collected as part of routine medical care, and stored in a secure internal database. No personal identifiers were included, ensuring compliance with data protection and ethical standards governing the use of anonymised health information for research purposes. Available clinical variables included diagnosis, sex, age, body mass index (BMI), prescribed treatment, and therapeutic outcome. Treatments covered topical or oral formulations of minoxidil, spironolactone, estradiol, finasteride, and dutasteride, administered according to standard clinical practice, with outcomes assessed at baseline and after at least six months of therapy. Efficacy was recorded as the percentage reduction in affected scalp area, determined by professional evaluation supported by photographic and trichoscopic documentation, following established clinical protocols [29,57,58]; however, only the final clinical decision and evaluation were considered, due to the anonymity of the data. Genetic data comprised 26 preselected single-nucleotide polymorphisms (SNPs) linked to pharmacological and hormonal pathways relevant to hair growth and drug metabolism. These SNPs originated from previous routine genotyping procedures, and were not specifically generated for this research (

Table 1. List of the 26 SNPs included in the dataset, along with their corresponding gene annotations. These variants were selected, based on their known or hypothesised involvement in drug metabolism, hormone signalling, inflammatory response, or other pathways relevant to the pharmacogenetics of androgenetic alopecia. Major and minor allele frequencies in our population matched the dbSNP database on NCBI, as expected.

Genes	rsIDs	Major Allele Frequency	Minor Allele Frequency
GPR44	rs545659	T	C
GPR44	rs533116	C	T
PTGFR	rs6686438	G	T
PTGFR	rs1328441	C	T
PTGFR	rs10782665	G	T
PTGES2	rs13283456	C	T
SULT1A1	rs1042028	C	T
GR-alpha	rs6198	T	C
CYP19A1	rs2470152	G	A
SRD5A1	rs39848	T	C
SRD5A2	rs523349	C	G
ACE	rs4343	A	G
COL1A1	rs1800012	C	A
IGF1R	rs2229765	G	A
NQO1	rs1800566	G	A
CRABP2	rs12724719	G	A
BTD	rs13078881	G	C
SLC23A1	rs33972313	C	T
MTHFR	rs1801133	G	A
GC	rs2282679	T	G
FUT2	rs602662	G	A
ZPR1	rs964184	C	G
MUC1	rs4072037	T	C
SLC30A3	rs11126936	G	T
TMPRSS6	rs855791	G	A
DMGDH	rs921943	C	T

). Due to the retrospective nature of the dataset, potential confounding factors such as treatment adherence, formulation variability, and dosing differences could not be fully controlled. Response classification was based on clinical follow-up and physician-reported improvement, which may introduce heterogeneity.

2.3. Bioinformatics and Statistical Treatment

All analyses were performed in the R statistical computing environment, using a combination of base statistical packages and the SNPassoc package for genetic association studies. Hardy–Weinberg equilibrium (HWE) calculations were applied to all genotyped variants as a quality control step, to ensure the absence of major genotyping errors and to verify consistency of genotype frequencies with population expectations. This was done using the tableHWE() function in SNPassoc, first on the present dataset of 252 patients and then cross-validated with an independent cohort of 26,000 patients, whose data were previously published by our group [59]. No significant deviations from equilibrium (p > 0.05) were observed after correction for multiple testing, supporting the reliability of the genotyping data and frequency distribution. SNP association and interaction analyses were subsequently conducted using model-based approaches appropriate for the study design and phenotype distribution. For functional enrichment and pathway-level interpretation of results, Cytoscape (version 3.10.3) was used to visualise and explore gene interaction networks derived from the significant variants identified.

2.3.1. SNP Association Analysis

To assess the relationship between individual genetic variants and treatment outcomes, we performed SNP association analysis using the WGassociation() function from the SNPassoc package in R. This function fits generalized linear models for each SNP independently, and supports evaluation under five genetic inheritance models: codominant, dominant, recessive, overdominant, and log-additive. Genotypic data were preprocessed using the setupSNP() function to ensure proper formatting and handling of SNP-level variables. Associations were tested using treatment response as a binary outcome (responder vs. non-responder), and models were run both in the overall population and stratified by treatment subgroup (i.e., minoxidil, finasteride, dutasteride, estradiol/spironolactone). Covariate-adjusted analyses were also performed using age, gender, and BMI as additional terms in the models when relevant. The p-values were generated for each SNP analysed, in view of the clinical outcome. This framework enabled the identification of SNPs associated with therapeutic efficacy, both overall and in drug-specific contexts [60].

2.3.2. SNP Interaction Analysis

To explore potential interactions between SNPs that may jointly influence treatment response, we used the interactionPval() function from the SNPassoc package. This function performs a two-dimensional analysis, evaluating the additive and interaction effects between all possible SNP pairs, using likelihood ratio tests. The method calculates and compares models for each pair: individual SNP effects, additive joint effects, and full interaction models. The resulting p-values provide insight into both marginal and epistatic effects, supporting the identification of SNP combinations that may have a synergistic or modifying influence on therapeutic outcomes.

2.3.3. Odds Ratio and Chi-Squared

To further characterise the relationship between specific genotypes and treatment response, we calculated odds ratios and performed chi-squared tests using genotypes expected to influence the pharmacological activity of each drug. This analysis focused on selected SNPs with known or hypothesised relevance to the mechanism of action of minoxidil, finasteride, or dutasteride. For each genotype–drug combination, we compared the proportion of responders and non-responders, to estimate the magnitude of effect using odds ratios with 95% confidence intervals. Statistical significance was assessed using the chi-squared test, allowing us to identify associations consistent with pharmacogenetic mechanisms influencing drug efficacy.

2.3.4. Enrichment Analysis

To explore potential biological interactions and support functional discovery among the genes associated with significant SNPs, enrichment analysis was conducted using Cytoscape. Gene symbols corresponding to the relevant SNPs were imported into Cytoscape and analysed using the built-in STRING app, which integrates known and predicted protein–protein interactions from curated databases and experimental sources. The resulting interaction networks were visualised and analysed to identify clusters of related genes and potential enrichment in biological processes, molecular functions, and signalling pathways. This systems-level approach enabled functional discovery by revealing biologically meaningful connections underlying the pharmacogenetic variability in response to hair loss treatments [61].

3. Results

3.1. Overall Descriptive Statistics

The study included a total of 252 patients diagnosed with androgenetic alopecia, of whom 218 (86.5%) demonstrated a clinically meaningful therapeutic response, as defined by percentage reduction in affected scalp area. The cohort was composed of 107 male and 145 female patients, with observed response rates of 81.3% and 90.3%, respectively.

Minoxidil, either as monotherapy or in combination, was the most frequently prescribed treatment, administered to 222 patients. Among these, 190 patients were classified as responders, corresponding to a response rate of 85.6%. When used as monotherapy, minoxidil was associated with a lower response rate of 71.4% (15 out of 21 patients), suggesting improved efficacy when combined with other pharmacological agents.

Dutasteride was prescribed to 109 patients, and was associated with a response rate of 89.0% (97 responders). Finasteride-treated patients (n = 41) showed a response rate of 87.8% (n = 36). A total of 100 patients were treated with estradiol, of whom 91 responded to therapy, yielding a response rate of 91.0%. Spironolactone was administered to eight patients, and seven of these were classified as responders, resulting in a response rate of 87.5%.

Combination therapies including minoxidil and a second agent demonstrated consistently high response rates. Among 93 patients treated with minoxidil and dutasteride, 81 responded to treatment (87.1%). Similarly, minoxidil combined with finasteride was prescribed to 38 patients, yielding 33 responders and a response rate of 86.8%. The combination of minoxidil and estradiol was used in 65 patients, with 59 responders (90.8%).

3.2. Single SNP Association Analysis

An SNP-by-SNP association analysis was conducted to identify genetic variants associated with treatment response in androgenetic alopecia, using five inheritance models (codominant, dominant, recessive, overdominant, log-additive) across the full patient population and pharmacologically defined subgroups. Results are summarized by drug exposure group, to support downstream interpretation of SNP–treatment interactions.

In the overall cohort (n = 252), the strongest association was detected for rs1042028 (SULT1A1), which reached statistical significance under the dominant model (p = 3.25 × 10⁻⁷). Notably, rs39848 (SRD5A1) also showed consistent associations across three models—dominant (p = 0.023), recessive (p = 0.046), and log-additive (p = 0.010)—suggesting a potential relationship between SRD5A1 variation and treatment outcomes in the broader patient population (Figure 1).

In stratified analyses, clearer drug-specific patterns emerged. Among minoxidil-treated patients (n = 222), rs1042028 (SULT1A1) remained highly significant (dominant model, p = 2.42 × 10⁻⁸), consistent with its role in minoxidil metabolism. rs2282679 (GC) also showed associations across the recessive (p = 0.015) and log-additive (p = 0.030) models. rs921943 (DMGDH) was again associated under multiple models, most notably dominant (p = 0.0033) and log-additive (p = 0.012). rs523349 (SRD5A2) appeared with nominal significance under the codominant (p = 0.030) and recessive (p = 0.027) models. Several other variants including rs1800012 (COL1A1) and rs13078881 (BTD) also reached nominal significance. Figure 2 shows the summary of the results of minoxidil-treated patients.

In the finasteride subgroup (n = 41), the most prominent associations involved rs545659 (GPR44), with significant values under the codominant (p = 0.00068), recessive (p = 0.018), and overdominant (p = 0.0086) models. In this group, rs1800566 (NQO1) also showed associations under codominant (p = 0.00030), recessive (p = 0.018), and overdominant (p = 0.0028). Other notable SNPs included rs1800012 (COL1A1), rs855791 (TMPRSS6), and rs1801133 (MTHFR), which demonstrated significance in at least one inheritance model (Figure 3).

For dutasteride-treated patients (n = 109), associations were again observed for rs1042028 (SULT1A1) (codominant, p = 0.044), as well as for rs39848 (SRD5A1) (codominant, p = 0.021; log-additive, p = 0.021). rs2282679 (GC) showed strong associations across three models—codominant (p = 0.00027), dominant (p = 0.0066), and recessive (p = 0.00025). rs964184 (ZPR1) also reached significance under codominant (p = 0.056) and recessive (p = 0.023), with rs4072037 (MUC1) showing low p-values under codominant (p = 0.0096) and recessive (p = 0.035). These findings suggest a distinct profile of SNPs more closely associated with hormonal treatment response (Figure 4).

Analysis of patients treated with both finasteride and dutasteride reinforced these patterns. rs1042028 (SULT1A1), rs2282679 (GC), and rs1800012 (COL1A1) were among the most consistently associated SNPs. The SNP rs855791 (TMPRSS6) was also statistically significant under codominant (p = 0.0039), dominant (p = 0.031), and overdominant (p = 0.00088) models.

For patients treated with estradiol and spironolactone, the most notable association was observed for rs921943 (DMGDH), which remained significant across multiple models—codominant (p = 3.18 × 10⁻⁵), dominant (p = 1.41 × 10⁻⁵), and log-additive (p = 7.08 × 10⁻⁵). A weaker association was also seen for rs6198 (GR-alpha), with p-values < 0.05 across dominant, overdominant, and log-additive models.

Adjustment for clinical covariates including BMI, gender, and their combination, did not reveal additional SNPs with improved significance or consistency. While a few SNPs retained nominal p-values < 0.05 after adjustment in specific subgroups, these effects were not more robust than the unadjusted findings, and are not presented in detail.

Table 2. Summary of SNPs associated with treatment response across patient subgroups. The table presents single-nucleotide polymorphisms (SNPs) that demonstrated statistical significance (p < 0.05) in at least one inheritance model within at least one treatment group. For each SNP, the associated gene is listed, and the minimum p-value observed across all tested models (codominant, dominant, recessive, overdominant, log-additive) is reported, per group. Subgroups include all patients, and those treated with minoxidil, finasteride, dutasteride, both finasteride and dutasteride, or estradiol plus spironolactone. Empty cells indicate no significant association was observed in that subgroup.

rsIDs	Gene	All Patients	Dutasteride	Estradiol + Spironolactone	Finasteride	Finasteride + Dutasteride	Minoxidil
rs1042028	SULT1A1	0.00566		0.0466	0.0226	0.00290	0.0128
rs11126936	SLC30A3				0.000304	0.00457
rs13078881	BTD				0.0104
rs1328441	PTGFR	0.01299		0.0276		0.0306
rs1800012	COL1A1	0.0328	0.000125		0.000398		0.0146
rs1800566	NQO1	0.0101	0.0210
rs1801133	MTHFR		0.00257
rs2229765	IGF1R				0.0104
rs2282679	GC			0.0294		0.0273
rs39848	SRD5A1		0.0325		0.000681	0.00196
rs4072037	MUC1			0.0134
rs4343	ACE			0.0179		0.0254
rs523349	SRD5A2	0.0172	0.0219		0.00867	0.000883	0.0419
rs545659	GPR44	0.00310		0.000576	0.0311		0.00329
rs602662	FUT2		0.0227			0.0104
rs6198	GR-alpha	0.00566		0.0466	0.0226	0.00290	0.0128
rs855791	TMPRSS6				0.000304	0.00457
rs921943	DMGDH				0.0104
rs964184	ZPR1	0.01299		0.0276		0.0306

summarises the relevant results for all groups.

3.3. SNP–SNP Interaction Analysis

To explore potential epistatic relationships influencing treatment outcomes, we performed SNP–SNP interaction analyses using likelihood ratio models to evaluate all possible pairwise combinations. Among the statistically significant results (p < 0.05), several interactions emerged involving genes previously implicated in drug response or androgen metabolism.

In the minoxidil-treated subgroup, the most significant interaction was found between rs1328441 (PTGFR) and rs4072037 (MUC1) (p = 5.38 × 10⁻⁶). Additional interactions involving PTGFR variants included rs6686438 with rs11126936 (SLC30A3) (p = 1.86 × 10⁻⁵) and rs10782665 with rs1800566 (NQO1) (p = 4.58 × 10⁻⁵). Notable combinations also included rs533116 (GPR44) with rs602662 (FUT2) (p = 9.41 × 10⁻⁵), rs6198 (GR-alpha) with rs602662 (FUT2) (p = 0.00045), and rs6198 with rs855791 (TMPRSS6) (p = 0.00523). Figure 5 summarises the SNP interaction found for the minoxidil-treated subgroup.

In patients treated with finasteride, significant interactions included rs6198 (GR-alpha) with rs964184 (ZPR1) (p = 8.89 × 10⁻⁵), rs523349 (SRD5A2) with rs964184 (p = 0.00038), and rs523349 with rs1800566 (NQO1) (p = 0.00617). An additional interaction was observed between rs39848 (SRD5A1) and rs1801133 (MTHFR) (p = 0.00533).

In the dutasteride subgroup, the variant rs533116 (GPR44) showed a significant interaction with rs2470152 (CYP19A1) (p = 0.00051). Further relevant combinations included rs2470152 (CYP19A1) with rs602662 (FUT2) (p = 0.00171) and rs921943 (DMGDH) (p = 0.00340), rs39848 (SRD5A1) with rs4343 (ACE) (p = 0.00283), and rs523349 (SRD5A2) with rs1800566 (NQO1) (p = 0.00333).

Among patients receiving estradiol and spironolactone, significant SNP–SNP interactions included rs6686438 (PTGFR) with rs6198 (GR-alpha) (p = 2.12 × 10⁻⁵), rs523349 (SRD5A2) with rs1800566 (NQO1) (p = 2.43 × 10⁻⁵), and rs6198 (GR-alpha) with rs964184 (ZPR1) (p = 0.00104).

These findings reveal distinct sets of genetic interactions associated with treatment response in each pharmacological subgroup, highlighting variant combinations of potential relevance for pharmacogenetic profiling.

3.4. Pharmacogenetic Association by Genotype

To evaluate whether therapeutic response differed across the main treatment combinations observed in the cohort, we performed a logistic regression analysis using treatment group as a categorical predictor of binary response status (responder vs. non-responder). These groups, namely, minoxidil alone or in combination with dutasteride, finasteride, estradiol, or spironolactone, represented the most common pharmacogenetically guided therapy regimens in the study population. The model, which included no covariates, showed no statistically significant differences in response probabilities between any group and the reference (all p > 0.05), nor between groups in pairwise comparisons adjusted for multiple testing (all Tukey-adjusted p > 0.1). Although combination therapies showed slightly higher predicted response probabilities (e.g., 0.88–0.91) than minoxidil monotherapy (≈0.68), the overlapping confidence intervals and non-significant contrasts indicate that these differences are not statistically meaningful. This means that when treatment choice was guided by pharmacogenetic profiling, all therapy regimens achieved comparable overall outcomes within this cohort (Figure 6).

To further evaluate whether treatment allocation was associated with differential clinical outcomes, we performed a chi-squared test of independence comparing the distribution of responders and non-responders across all treatment categories, including both monotherapies and drug combinations commonly observed in the cohort. The test yielded a non-significant result (p = 0.377), indicating no statistical evidence of an association between treatment type and therapeutic response. These findings corroborate the logistic regression analysis, reinforcing that within this genetically stratified population, the proportion of responders did not differ significantly between the main treatment strategies assessed.

Multivariate logistic regression models were used to assess the relationship between selected SNP genotypes and therapeutic response to minoxidil, dutasteride, and finasteride. These models focused on pharmacologically relevant variants previously implicated in drug metabolism and androgenic signalling. Across all three treatments, the TC genotype of the SNP rs1042028 (SULT1A1) consistently demonstrated a significant association with reduced odds of response: OR = 0.09 (95% CI: 0.03–0.24, p < 0.001) for minoxidil; OR = 0.21 (95% CI: 0.04–0.82, p = 0.023) for dutasteride; and OR = 0.11 (95% CI: 0.01–0.76, p = 0.025) for finasteride. Additionally, for dutasteride, the TT genotype of the SNP rs39848 (SRD5A1) was also associated with reduced response (OR = 0.02, p = 0.018), indicating a potential impact on drug activity. These findings suggest that these specific genotypic combinations may impair therapeutic efficacy.

Notably, the directionality of these results implies that patients lacking these genotypes, i.e., those carrying the alternative homozygous forms, may have comparatively higher odds of responding to therapy. In particular, absence of the TC genotype of rs1042028 or the TT genotype of rs39848 could be indicative of better pharmacological outcomes, possibly due to preserved enzyme function or more favourable drug metabolism. All other tested SNPs showed no statistically significant associations, although some displayed broad confidence intervals due to low frequency or limited sample size. Figure 7 summarises the odds ratios and confidence intervals for each treatment-specific model, reinforcing the role of rs1042028 and rs39848 as potential negative predictors of drug response in androgenetic alopecia.

3.5. Functional Enrichment and Network Analysis

The SNP-SNP interaction network was built using only statistically significant pairwise interactions (p < 0.05) and visualised through Cytoscape, providing an integrative representation of genetic relationships potentially involved in treatment response. The overall interaction network (Figure 8) demonstrated a dense and interconnected structure linking key genes such as PTGFR, GPR44, SRD5A1/SRD5A2, ACE, MUC1, SLC30A3, NQO1, FUT2, and NR3C1, which are frequently involved in drug metabolism, prostaglandin signalling, steroid processing, and inflammatory response. Notably, genes like PTGFR and GPR44 appeared repeatedly across multiple interactions, suggesting their central role in modulating drug effects. These hubs connected with a wide range of partners such as SULT1A1, BTD, IGF1R, COL1A1, and MTHFR, further supporting their potential as regulatory bottlenecks within the pharmacogenetic landscape.

When stratifying the analysis by drug, minoxidil-associated interactions showed a distinct and tightly connected subnetwork (Figure 9), prominently involving PTGFR, GPR44, NR3C1, and FUT2. Particularly relevant connections included rs1328441 (PTGFR) × rs4072037 (MUC1), rs6686438 (PTGFR) × rs11126936 (SLC30A3), and rs10782665 (PTGFR) × rs1800566 (NQO1), all with p-values < 5 × 10⁻⁵. These interactions suggest that minoxidil response may be modulated by pathways related to prostaglandin F2-alpha receptor signalling, cellular oxidative stress response, and glucocorticoid receptor-mediated regulation, potentially reinforcing the known vasodilatory and anti-inflammatory mechanisms of minoxidil. Dutasteride and finasteride subnetworks (Figure 10 and Figure 11) displayed fewer, but still meaningful, interactions. For dutasteride, genes like CYP19A1, FUT2, SRD5A1, and ACE emerged, while for finasteride, SRD5A2, ZPR1, and MTHFR were prominent, again suggesting intersections with androgen metabolism and folate-related enzymatic pathways. These stratified subnetworks demonstrate how genotype combinations might contribute to inter-individual variability in drug response, highlighting a complex and drug-specific genomic architecture that warrants further functional validation.

4. Discussion

The present study was designed within the framework of pharmacogenetics, aiming to improve therapeutic outcomes for androgenetic alopecia by aligning drug selection with individual genetic profiles [62,63]. Building on the established notion that inherited genetic variation significantly modulates drug metabolism, efficacy, and toxicity, we focused on assessing both the clinical performance of genotype-guided therapies and the genetic underpinnings influencing treatment response. While pharmacogenetics has been successfully applied in diverse medical areas, including anticoagulation, oncology, and psychiatry, its integration into dermatology—particularly in the management of hair loss—remains limited [2,64,65,66,67,68]. By leveraging SNP-based association models and interaction networks, this work investigates not only the impact of specific variants on drug efficacy, but also how gene–gene interactions may help clarify the biological mechanisms underpinning response to key agents such as minoxidil, finasteride, and dutasteride.

In our genotype-stratified cohort of 252 patients, we observed high response rates across all treatments: dutasteride (89.0%), finasteride (87.8%), and minoxidil (85.6%). Combination therapies such as minoxidil with dutasteride (87.1%), finasteride (86.8%), or estradiol (90.8%) were similarly effective, with minimal subgroup variation. These results exceed reported benchmarks—finasteride monotherapy achieves significant regrowth in ~11.1% of men, with most showing only moderate or slight improvement [37,69,70]; dutasteride shows ~62.8% improvement in selected female populations; and 5% topical minoxidil produces visible regrowth in ~45%, while oral formulations yield significant improvement in ~35% [29,32,57]. Compared to these outcomes, our uniformly high response rates suggest that pharmacogenetic matching can reduce inter-individual variability. These findings support the hypothesis that aligning treatment with genetic profiles enhances efficacy and may help equalise outcomes across different therapeutic options by ensuring optimal drug selection from the start.

SNP association analysis revealed treatment-specific genetic signals consistent with known drug pathways. The strongest association was rs1042028 in SULT1A1 (p = 2.42 × 10⁻⁸), particularly in minoxidil responders, confirming its role in bioactivation [34,53].

Other minoxidil-associated variants included rs2282679 (GC), related to vitamin D binding and follicular cycling, and rs921943 (DMGDH), involved in methylation and oxidative metabolism. The appearance of rs523349 (SRD5A2) suggests a link to androgen-driven hair loss, rather than direct minoxidil action. For both minoxidil and finasteride, rs545659 (GPR44) and rs1800566 (NQO1)—linked to prostaglandin signalling and oxidative stress—were notable. Variants in MTHFR, COL1A1, and TMPRSS6 appeared in finasteride and combination groups, implicating follicular structure, vascularisation, and iron metabolism. In the dutasteride group, rs2282679 (GC) and rs39848 (SRD5A1) supported roles for drug metabolism and DHT synthesis [71,72,73,74]. These SNP-level associations explain inter-individual variability, and align with plausible pharmacological mechanisms.

SNP–SNP interaction analysis revealed epistatic relationships that clarify drug-specific mechanisms and support the biological relevance of key variants. In the minoxidil group, rs1328441 (PTGFR) × rs4072037 (MUC1) (p = 5.38 × 10⁻⁶) suggests interplay between prostaglandin signalling and epithelial remodelling—central to minoxidil’s action. Other interactions, such as PTGFR × SLC30A3 and PTGFR × NQO1, highlight roles for oxidative stress and zinc-dependent pathways [34,53,75]. Recurrent involvement of GPR44 and NR3C1 supports roles in anti-inflammatory signalling and androgen modulation. In finasteride and dutasteride groups, interactions involving SRD5A2, SRD5A1, ZPR1, and MTHFR implicate androgen metabolism and folate pathways. Combinations like GPR44 × CYP19A1 and SRD5A1 × ACE in dutasteride-treated patients suggest broader endocrine modulation, aligning with dutasteride’s dual-enzyme inhibition. These SNP interactions likely represent functional gene networks, enhancing the understanding of treatment response mechanisms.

The integrative gene–gene network analysis supports a pharmacologically coherent model in which the response to each drug class is modulated by overlapping, but distinct, genetic modules. In the minoxidil subnetwork, the centrality of PTGFR, GPR44, NR3C1, FUT2, and SULT1A1 aligns with its multifactorial mechanism involving prostaglandin modulation, sulfotransferase-mediated bioactivation, and glucocorticoid signalling. These connections parallel known in vitro findings of VEGF induction, K-ATP channel activation, and prostaglandin-mediated follicle cycling. For dutasteride and finasteride, the convergence around SRD5A1, SRD5A2, MTHFR, and NR3C1 reinforces the central role of androgen metabolism and receptor regulation. Interestingly, FUT2 and ZPR, associated with epithelial signalling and cell proliferation, also appear across multiple subnetworks, suggesting potential shared modifiers of follicular response. The success of combination therapies such as minoxidil plus dutasteride or finasteride may therefore reflect the simultaneous targeting of both independent and partially overlapping pathways: minoxidil acting via vascular and anagen-promoting mechanisms, and 5α-reductase inhibitors acting via DHT suppression and androgen-receptor downregulation. From a genetic perspective, the overlap of responsive genotypes (e.g., SULT1A1, SRD5A1, GC) across subgroups supports the idea that combination therapy may benefit from additive or complementary genotype effects, thereby enhancing efficacy even in the absence of statistically greater response rates.

The strong association of rs1042028 in SULT1A1 with minoxidil response, alongside its central role in the gene interaction network, reinforces the necessity of follicular sulfotransferase activity for minoxidil bioactivation [47,48]. In parallel, SNP–SNP interactions involving PTGFR and GPR44, genes linked to prostaglandin F2-alpha and D2 signalling, support minoxidil’s proposed modulation of local prostaglandin pathways, known to influence follicular cycling and perifollicular vascularisation. Together, these findings align with known mechanisms of minoxidil, including its role in promoting anagen entry, VEGF expression, and anti-inflammatory activity via glucocorticoid and prostaglandin-mediated routes [47,53,54,75,76].

The identification of specific genotypes associated with reduced therapeutic response provides strong support for the clinical utility of pharmacogenetic testing in guiding treatment decisions. In particular, the TC genotype of rs1042028 in SULT1A1 emerged as a consistent negative predictor of the response to minoxidil (OR = 0.09), highlighting the central role of this sulfotransferase enzyme in drug activation and efficacy. Similarly, the TT genotype of rs39848 in SRD5A1 significantly reduced the odds of response to dutasteride (OR = 0.02), likely reflecting compromised inhibition of type I 5α-reductase. These findings suggest that individuals carrying these genotypes may require alternative agents or combination strategies to achieve optimal outcomes. Notably, despite the use of multiple pharmacologic regimens, the lack of significant differences in predicted response probabilities across treatment groups further indicates that genotype-driven selection was successful in matching each patient to an effective therapy. Thus, pharmacogenetic profiling not only identifies poor responders prospectively, but also enables therapeutic tailoring that can equalise efficacy across diverse treatment pathways.

5. Conclusions

This study provides supportive evidence that pharmacogenetic stratification may enhance therapeutic response in androgenetic alopecia, potentially minimising variability across drugs and combinations. SNP-level associations aligned with known mechanisms of minoxidil, finasteride, and dutasteride, and interaction analysis highlighted functionally relevant gene networks. While the findings are robust and biologically consistent, certain limitations must be acknowledged. The retrospective, anonymised nature of the dataset limited control over adherence, dosing, and formulation, and subgroup sizes were small for some therapies, although combination therapies were markedly relevant, statistically. Additionally, outcome assessment was based on clinical follow-up without formal severity scales, introducing possible heterogeneity. These constraints underline the need for prospective, standardised studies, to further validate genotype-guided therapy and support its broader clinical integration.