Association of sleep among 30 antidepressants: a population-wide adverse drug reaction study, 2004–2019

Drug-gene associations

Human phase I and phase II drug metabolizing enzyme data that includes cytochrome P450 (CYP) inhibitor or inducer information are referenced from the Consensus Guidelines for Therapeutic Drug Monitoring in Neuropsychopharmacology Update 2017 and the FDA Drug Development and Drug Interactions Table of Substrates, Inhibitors and Inducers (CDER, 2017; Hefner, 2018).

Statistical analysis

Disproportionality analyses of psychotropic adverse events resulting in somnolence were calculated using Fisher’s exact test, via a two-by-two contingency table, using the case/non-case method (Egberts, Meyboom & Van Puijenbroek, 2002; Rochoy et al., 2018). A minimum of 1 case is required and the final estimated reporting odds-ratios (ROR) and corresponding 95% confidence interval (95% CI) are reported in the text. All graphical illustrations as well as all computations were performed using R (version 3.5.2, R Foundation for Statistical Computing, Vienna, Austria) (Böhm et al., 2016; R Core Team, 2013). The reporting odds-ratio-95% confidence intervals not crossing a value of one is interpreted as statistically significant.

Somnolence and 5-HT2C, NET, and Alpha-1 adrenergic receptor binding affinities

This study assessed and ranked the reporting odds of somnolence among thirty antidepressant compounds using post-marketing adverse drug reactions reported to the FAERS. As shown in Table 1, seven groups of antidepressants resulted in equivalent reporting odds of somnolence. Consistent with the original hypothesis, the compound with the highest reporting odds of somnolence indeed has the strongest 5-HT2C receptor binding affinity, amoxapine (Ki = 2 nM), among the 30 antidepressant compounds tested (Roth et al., 2000; Tatsumi et al., 1997). Further, amoxepine’s binding affinities for NET and H1 receptor are Ki = 16 nM and 25 nM, respectively. Atomoxetine, a norepinephrine reuptake inhibitor and generally indicated in patients with ADHD, was ranked second in somnolence overall and binds to the NET with the lowest Ki value of 3.5 nM among the thirty antidepressant compounds in this study (Roth et al., 2000; Tatsumi et al., 1997). Mirtazapine, a tetracyclic antidepressant, was ranked 11th in reporting odds of somnolence and had the strongest H1 receptor binding affinity with a Ki of 0.14 nM (Roth et al., 2000; Tatsumi et al., 1997). When considering the glymphatic system, doxepin was the antidepressant compound with the lowest alpha-1 adrenergic receptor binding affinity (Ki = 23.4 nM) and ranked 12th in reporting odds of somnolence (n = 139) ROR = 3.4 (95% CI [2.9–4.1]), but also has a low H1 receptor Ki value of 0.24 nM (Roth et al., 2000; Tatsumi et al., 1997).

Maprotiline, a tetracyclic antidepressant, was ranked third among the thirty antidepressants and binds to the H1 receptor (Ki = 1.7 nM) and NET (Ki = 11.1 nM) with high affinity (Roth et al., 2000; Tatsumi et al., 1997). Similarly, mianserin, another tetracylic antidepressant, binds to the 5-HT2C and H1 receptors with Ki values of 2.59 nM and 1nM, respectively (Roth et al., 2000; Tatsumi et al., 1997). In contrast, vortioxetine, a serotonin modulator and stimulator, was least associated with somnolence and potently inhibits the serotonin transporter (SERT) with a Ki = 1.6 nM, however weakly binds to the 5-HT2C receptor with a Ki = 180 nM and the NET with a Ki = 113 nM (D’Agostino, English & Rey, 2015; Di & Kerns, 2015; Roth et al., 2000).

A recent meta-analysis that analyzed 276 studies and 14 second-generation antidepressants, evaluated somnolence in patients undergoing treatment for major depression, reported that fluvoxamine (NET = 1,892 nM, 5-HT2C Ki = 6,700 nM) and mirtazapine (H1 inverse agonist Ki=0.14 nM, 5-HT2C Ki = 39 nM) resulted in the highest odds of somnolence (Alberti et al., 2015). Moreover, in the same study, buproprion (NET Ki = 5,600 nM, H1 = 6,700 NM) was reported to have the lowest odds of somnolence and highest odds of insomnia (Alberti et al., 2015; Roth et al., 2000). Nevertheless, the odds of causing somnolence are not related to popularity index of antidepressants. A recent scientometric analysis reported that fluoxetine has the highest popularity index (4.0) (Tran et al., 2019). The popularity indexes of aforementioned antidepressants are as follows: fluvoxamine (1.00), bupropion (0.71), mirtazapine (0.61), vortioxetine (0.1), desvenlafaxine (0.08), vilazodone (0.04) and levomilnacipran (0.03). (Tran et al., 2019).

Antidepressant-associated somnolence and CYP2C19 poor metabolizers

Patients who are CYP2C19 Poor Metabolizers—that is have single-nucleotide polymorphisms (SNPs) allele combinations of CYP2C19 *2/*2-3 and have a fraction of activity (FA) of 0.005—when compared to CYP219 Normal Metabolizers (*1/*1 FA = 1 and *2-3/*17 FA = 0.8) will have an increase in drug exposure (Fermier et al., 2018). The allele frequency of the functionally inactive CYP2C19*2 variant is seen at rates of 18.1% in African, 10.1% in Admixed American, 18.3% in European, 31.0% in East Asian, and 34.0% in Southeast Asian populations (Zhou, Ingelman-Sundberg & Lauschke, 2017). Based on the study results, the antidepressant compounds with the highest reporting odds of somnolence and not primarily metabolized by CYP2C19 substrates are: vilazodone, mirtazapine, and duloxetine. Further, antidepressants with the lowest reporting odds of somnolence and not primarily metabolized by CYP2C19 are: levomilnacipran, milnacipran, vortioxetine, and buproprion.

Antidepressant-associated somnolence and CYP2D6 poor metabolizers

Increased drug exposures are also evident in patients who are prescribed CYP2D6 drug substrates and have CYP2D6 loss-of-function DNA sequence variants leading to being phenotypically classified as CYP2D6 Poor Metabolizers (*3-8/*3-8 FA = 0.01 and *10/*10 FA = 0.1) (Fermier et al., 2018). The CYP2D6*4 functionally inactive variant has the highest allele frequency rate and is seen in 11.9% in African, 15.7% in Admixed American, 15.5% in European, 0.4% in East Asian, and 11.6% in Southeast Asian populations (Zhou, Ingelman-Sundberg & Lauschke, 2017). The CYP2D6*5 inactive variant is seen at a 6.5% allele frequency in the Southeast Asian population (Zhou, Ingelman-Sundberg & Lauschke, 2017). Based on the study results, milnacipran and levomilnacipran are antidepressant compounds with low and no association with somnolence, respectively. Further, tablet-based antidepressants that are significantly associated with somnolence and have no to minor CYP2D6 contribution are: desvenlafaxine and mirtazapine. Esketamine is FDA approved in conjunction with an oral antidepressant (DailyMed., 2019b).

Pharmacogenomics, therapeutic drug monitoring, and clinical pharmacologists

In cases where CYP450 metabolizer status are unknown or even unclear from pharmacogenomic testing reports due to concomitant medications resulting in drug-drug-gene interactions, directly measuring drug plasma concentrations via therapeutic drug monitoring (TDM) is often most efficient (Eugene & Eugene, 2018; Qin et al., 2019). Of note, CYP2C19 *1/*2-3 (FA = 0.3) genotypes are phenotypically CYP2C19 Intermediate Metabolizers and known to have a reduced fraction of activity relative to CYP2C19 Normal Metabolizers (FA = 1) (Fermier et al., 2018). In contrast, excessive metabolic rates, resulting in sub-therapeutic antidepressant plasma concentrations, are seen in patients who are CYP2C19 Ultra-rapid Metabolizers and have variant combinations of CYP2C19 *17/*17 (FA = 2.03) and *1/*17 (FA = 1.59) (Fermier et al., 2018). Thus, when considering the results for selecting an antidepressant with an association with somnolence, or even extrapolating from other studies, there may be wide inter-individual variability in antidepressant therapeutic effect and adverse effect. More importantly, the United States Food and Drug Administration expresses concern to physicians and other healthcare professionals that the association of current pharmacogenomics testing reports and antidepressant efficacy are not yet established (FDA, 2018).

In various countries, physicians specializing in clinical pharmacology routinely utilize pharmacogenomics as a means to avoid drug toxicity by mainly using TDM, applying Bayesian dosing methods, and other techniques with the aim of patient drug safety (Borobia et al., 2017; Eichelbaum, Dahl & Sjöqvist, 2018). In the U.S., there needs to be a significant increase in clinical pharmacology training programs that provide direct entry following medical school, as is done internationally, to have more clinical pharmacologists within the hospitals to carry out traditional patient care roles, consult with other physicians, interpret the pharmacogenomics testing reports, and advance the field. One major limitation of using pharmacogenomics testing reports alone to guide drug decision-making is that most patients are prescribed more than one medication resulting in drug-drug interactions as well as drug-drug-gene interactions. Whereas, adding TDM will increase precision in dosing and avoid toxicity based on established safety plasma ranges using pharmacokinetics (Hefner, 2018). Further, drug selection based on pharmacogenomics testing reports alone without also accounting for established efficacy is that often pharmacogenomic testing systematically recommends CYP3A4 substrate alone, as well as compounds not dependent on cytochrome P450 oxidative metabolism, despite established therapeutic efficacy for all compounds that gained regulatory approval. More specifically, a pharmacogenomics study found that among twenty-two antidepressants, desvenlafaxine and levomilnacipran (renally excreted) were recommended greater than 90% of the time, while in less than 10.5% of the time citalopram, duloxetine, escitalopram, fluoxetine, fluvoxamine, mirtazapine, paroxetine, and sertraline were never in the green (i.e., use as directed) category (Macaluso & Preskorn, 2018). However, in a systematic review and network meta-analysis comprised of 522 clinical trials with 116,477 participants found that (in order of decreased efficacy): amitriptyline, mirtazapine, duloxetine, venlafaxine, paroxetine, milnacipran, fluvoxamine, escitalopram, nefazodone, and sertraline were the top ten most efficacious; while levomilnacipran ranked 14th and desvenlafaxine was ranked 20th out of the 21 assessed antidepressants in the study (Cipriani et al., 2018).

Study limitations

The study limitations are that the reporting odds-ratios are generally hypothesis generating due to the entire population of patients experiencing the side-effect are not known, but are dependent on voluntary submission of the adverse event to the FDA. Further, the term somnolence may be reported and recorded as sedation and this study did not factor in the term sedation in the analysis. For example, in the adverse reactions section of atomoxetine’s package insert states that both somnolence and sedation was recorded as somnolence (DailyMed., 2018). For the newest compound in this study, the FAERS results for esketamine showed 61 adverse events for sedation, 288 as other cases, and 349 total reported cases resulting in a ROR = 140 (95% CI [111–176]) for sedation alone. Statistically, the overall interpretation of the results are that the strength of the ADR association with the drugs-under-test and are not indicative of risk (Montastruc et al., 2011). Further, psychotropic medications exhibit target and off-target effects influencing the cholinergic (acetylcholine), gamma-aminobutyric acid (GABA)-ergic, glutaminergic (glutamate), glycinergic, and L-Arginine nitric oxide pathways, among others, so the association of somnolence may not be directly associated with the glymphatic system alone.