This exploratory prospective study was approved by the Tokyo Dental College Ethics Committee (Tokyo, Japan; approval No. 919; February 13, 2019) and Tokyo Metropolitan Institute of Medical Science Ethics Committee (Tokyo, Japan; approval No. 18–46; March 7, 2019). Patients were recruited at Tokyo Dental College Suidobashi Hospital, Tokyo, Japan, between April 2019 and March 2020. Genetic analyses were conducted at the Tokyo Metropolitan Institute of Medical Science, Tokyo, Japan. All of the patients provided written informed consent before participating in this study.

Patients who were scheduled to undergo dental procedures under intravenous sedation were eligible for this study. The inclusion criterion was age between 20 and 60 year. The exclusion criteria were (1) American Society of Anesthesiologists (ASA) physical status III or higher, (2) known allergy to midazolam, (3) psychotropic drug use, including benzodiazepine use, within the past 3 months, and (4) non-Japanese descent.

No premedication was administered. The patients were seated in a dental chair in the semi-supine position with the head up at a 40-degree angle. Vital signs, including noninvasive blood pressure, pulse rate, and oxygen saturation (measured by pulse oximetry [SpO2]), and Bispectral Index (BIS) values were monitored. A BIS electrode was applied on the patient’s forehead. BIS values were continuously recorded with a BIS monitor (Covidien, Tokyo, Japan). The Ramsay sedation scale was used to evaluate the patients’ level of sedation: score of 1, patient anxious and agitated or restless or both; 2, patient cooperative, oriented and tranquil; 3, patient drowsy, but responds to commands; 4, asleep, brisk response; 5, asleep, sluggish response; 6, no response [14]. Patients with no clinical response to midazolam were deemed to have a Ramsay score of 1. To evaluate the amnesic effect of midazolam, the patients were requested to memorize a word and the region where we would collect buccal swab samples during sedation.

After recording baseline vital signs and determining the BIS value, midazolam 0.05 mg kg− 1 (Teva Takeda Pharma, Aichi, Japan) was administered intravenously in approximately 1 min. Five minutes after initial midazolam administration, the Ramsay score was recorded by a dentist anesthesiologist (Y. K.) in all cases. The same variables as baseline were also recorded. If SpO2 was less than 90%, then a jaw thrust maneuver was performed to maintain the patient’s airway, and supplemental oxygen was supplied via a nasal cannula. Any adverse events, including airway obstruction, desaturation, apnea, and unstable hemodynamic, were recorded.

Buccal swab sampling and saying a word occurred immediately after Ramsay score recording. Buccal swab samples were collected for genotyping from either left or right buccal mucosa, and the patients were told a word twice by the dentist anesthesiologist (Y. K.). The word was selected in a randomized manner from the following words (in Japanese): apple, banana, grape, orange, and peach. After data collection, the dental procedure began. The level of sedation was adjusted by the attending dentist anesthesiologist using midazolam with or without propofol. After finishing the procedure, the patients were moved to the recovery room. Amnesic effects, namely the memories of buccal swab sampling and saying a word, were assessed 30 min after confirming that the patients were near their baseline level of consciousness. The patients were considered to have anterograde amnesia if they recalled neither the word nor the left or right side where the buccal swab samples were collected.

Total genomic DNA was extracted from buccal mucosa samples using the QIAamp DNA Micro kit (Qiagen, Hamburg, Germany) according to the manufacturer’s instructions. The extracted DNA was stored at 4 °C until analysis. After all of the clinical data were collected, genotyping was performed on an Infinium Asian Screening Array-24 v1.0 BeadChip (Illumina, San Diego, California, USA) according to the manufacturer’s instructions. Genotypes were called using GenomeStudio v2.0.5 with the Genotyping v2.0.5 module (Illumina). Quality control was performed for the DNA samples and genetic markers. Samples were excluded if the sample call rate was less than 97%. Genetic markers were excluded if each quality metric did not meet the hard cutoff thresholds that are recommended by Illumina, including call frequency < 97%, Cluster Sep ≤ 0.3, AA R Mean ≤ 0.2, AB R Mean ≤ 0.2, and BB R Mean ≤ 0.2. The other multi-variable metrics were higher or lower than the hard cutoff values. In addition to the quality control criteria, variants with a minor allele frequency less than 5% (i.e., rare and low-frequency variants) were filtered out from statistical analyses.

Although no data on genome-wide genetic variants or the sedative effect of midazolam were available, we performed a priori power analysis using Quanto 1.2.4 [15] based on Cohen’s small, medium, and large R2 effect sizes of 0.02, 0.13, and 0.26, respectively [16]. The type I error rate, adjusted for multiple testing, was set at 7.6 × 10− 8 (0.05/660,000) because the Infinium Asian Screening Array-24 v1.0 BeadChip contains approximately 660,000 markers. Based on a two-sided α error rate of 7.6 × 10− 8 and statistical power of 0.8, the sample sizes were calculated as 1,914 for a small effect size, 278 for a medium effect size, and 128 for a large effect size. We sought to recruit at least 128 patients during the 1-yr exploratory study period to examine strong genetic associations with the sedative effect of midazolam.

We used two parallel approaches in this study. First, we tested all genetic variants using the genome-wide approach. Second, we tested variants of genes that are related to midazolam sensitivity using the candidate gene approach. For the candidate gene analyses, we focused on GABAA receptors. Although there are 19 GABAA receptor subunits (α1–6, β1–3, γ1–3, δ, ε, θ, π, and ρ1–3), the α4, α6, π, and ρ subunits are insensitive to classic 1,4-benzodiazepines [17,18,19,20,21]. Thus, we chose genes that encode the α1, 2, 3, 5, β1–3, γ1–3, δ, ε, and θ subunits in humans for candidate gene analyses. We selected 372 SNPs of 13 genes (GABRA1, GABRA2, GABRA3, GABRA5, GABRB1, GABRB2, GABRB3, GABRG1, GABRG2, GABRG3, GABRD, GABRE, and GABRQ) as the candidate SNPs.

Multivariate linear regression analyses were conducted to investigate the association between genotypes and Ramsay sedation scores as an index of the sedative effect of midazolam. Ramsay scores were used as numeric phenotype values for the dependent variable. Genotypes were used for independent variables. Age, sex, and body mass index (BMI) were included as covariates to adjust potential confounding factors in the linear regression models. Additive, dominant, and recessive genetic models were used for the analyses. Genetic variants on the X, Y, and mitochondrial chromosomes (except for the pseudoautosomal region) were excluded from the analyses of the dominant and recessive models. Values of P were adjusted for multiple testing with false discovery rate correction (Q value) [22]. Statistical significance was set at a Q value less than 0.05. Deviations from Hardy-Weinberg equilibrium were tested using Fisher’s exact test at a type I error rate of 0.001.

Additional analyses were performed for genetic variants that had a Q value less than 0.05 in the genome-wide association analyses and candidate gene analyses. To explore the association between the amnesic effect of midazolam and genetic variants, multivariate binominal logistic regression analyses were conducted using the covariates age, sex, and BMI. The presence of anterograde amnesia was used as the phenotype value for the dependent variable. Genotypes were used for independent variables. The genetic models that were used for the analyses were the same as the linear regression analyses. Statistical significance was set at a P value less than 0.05 after correcting for multiple testing using Bonferroni adjustment.

All of the statistical analyses were performed using PLINK 1.90 beta 6.18 [23, 24]. All of the tests were two-tailed. According to the PLINK sample information file format, the presence of anterograde amnesia was coded as the following: 1 (patients with recall [‘control’]) and 2 (patients with amnesia [‘case’]). Male and female were also coded as 1 and 2, respectively. Continuous variables are presented as means (standard deviation) or medians (interquartile range). Categorical variables are described as numbers with percentages. Effect size estimations are reported as regression coefficient (β) or odds ratio (OR) with 95% confidence intervals (CIs).