Adverse event reports corresponding to the 2004–2019 period were obtained from the FDA website (11). Duplicate reports (comprising 13,632,002 cases) were eliminated, as previously described (12), and the remaining 11,438,031 reports were analyzed. Arbitrary drug names, including trade names and abbreviations, were manually annotated to unified generic names using the Medical Subject Headings descriptor ID. Next, reports of BP were defined using the preferred terms “pemphigoid” in MedDRA (version 23.0). FAERS data analysis was performed as previously described (8) and Supplemental Table 1 . Volcano plots were produced, and Z scores were used instead of P values to save space.

The IBM MarketScan Commercial and Medicare Supplemental databases corresponding to the period from January 2017 to December 2019 were purchased from IBM^® Watson Health^®. The dataset comprised data of 43,723,094 employees, dependents, and retirees in the United States, with primary or Medicare supplemental coverage. The outpatients’ data including the year of birth and sex was extracted (36,399,469 cases). Then, those who received formulations belonging to ATC codes A10BH and C09AA03 were defined as patients prescribed DPP4 inhibitors and lisinopril, respectively ( Supplemental Table 2 ). BP cases were defined according to the ICD10 system, based on BP-related symptoms ( Supplemental Table 3 ). To produce a time-series analysis of MarketScan data, the software packages “survival” (13) and “MatchIt” (14) in R software v4.1.2 and RStudio (2022.07.1 Build 554 software, R Foundation for Statistical Computing) were used as previously described (9). The incidence of BP associated with the use of DPP4 inhibitors was first evaluated via Poisson regression analysis, and the obtained results were expressed as incidence rate ratio (IRR), along with the 95% confidence interval (CI) and Z scores. Patients who received any DPP4 inhibitor were categorized into two groups (with and without lisinopril), and 1:1 propensity score matching was performed to eliminate deflections in the number of patients with risk factors. Thereafter, propensity score-matched pairs were created by matching two groups using the nearest-neighbor method based on a 0.2-caliper width. The following categories were used as confounding factors: renal failure (N17–19), liver disorder (K70–77), hypertension (I10–15), diabetes (E10–14), heart disease (I05–09, I20–52), cerebrovascular diseases (I60–69, G45, and G46), cancer (C00–D09), neurologic diseases (G20–22, G30–32, F00–03, G35, G40, and G41), autoimmune diseases (M05–09, L40–41, L43, L93, and M32), loop diuretics (ATC codes C03CA01, C03CA02, C03CC01, and C03CA04), and steroids (ATC code H02A). Further, using matched cohort pairs, the daily and cumulative dosages as well as the periods of administration of DPP4 inhibitors and lisinopril were quantified and compared. The defined daily dosage of each drug was then determined according to the WHO website (15). Next, the cumulative incidences of BP were compared between cohorts with and without lisinopril use via conventional survival analysis, and Kaplan–Meier plots were used to visualize survival curves. Statistical significance was evaluated using log-rank tests, and Cox proportional regression analysis was performed to calculate hazard ratios. The number of at-risk individuals indicated the number of patients who may experience BP onset each month.

Clinical laboratory test results involving hemoglobin A1c (HbA1c) in the IBM MarketScan^® Lab database were defined using the Loinc codes “4548-4” (n = 1,388,265) and “17856-6” (n = 51,213). We only used data with the result unit including “OF TOTAL HGB,” “TOTAL HGB,” “A1C,” “OF HGB,” “PERCENT,” “%,” “HB,” “T.HGB,” “% OF TOTAL HGB”, and “OF T.” Further, we eliminated data of cases that were reported as “< 0” or “> 100” in our analysis, and data corresponding to patients who underwent HbA1c testing more than twice (before and after the first prescription of DPP4 inhibitors) were analyzed (n = 8,921). The patients were categorized into two groups (with and without BP), and for each group, a two-tailed Wilcoxon matched-pairs signed rank test was performed to evaluate changes in HbA1c from before to after prescription of DPP4 inhibitors. The differences in HbA1c changes between the two groups were further compared using unpaired t-tests with Welch’s correction. Additionally, the patients (n = 8,921) were divided into two groups (with and without lisinopril treatment), and their HbA1c values were analyzed as described above.

Vildagliptin was purchased from Cayman Chemical (Ann Arbor, MI, USA), sitagliptin from MedChemExpress (Monmouth Junction, NJ, USA), and lisinopril dihydrate and A779 were purchased fromTokyo Chemical Industry Co., Ltd. (Tokyo, Japan). Vildagliptin, sitagliptin, and A779 were dissolved in dimethyl sulfoxide, and lisinopril was dissolved in ultrapure sterile water. Thereafter, each solution was diluted with the buffer used for each respective experiment.

Recombinant human M-CSF was purchased from PeproTech (Rocky Hill, NJ, USA), and phytohemagglutinin (PHA), phorbol-12-myristate-13-acetate (PMA), and ionomycin were purchased from Sigma-Aldrich (St. Louis, MO, USA). Brefeldin A solution was purchased from Thermo Fisher Scientific (Cleveland, OH, USA).

Deidentified peripheral blood samples were obtained from healthy volunteers at Japan Tobacco, Inc., who had not received any drugs for the last 7 days and provided written informed consent to participate in the study. For the isolation of human peripheral blood mononuclear cells (hPBMCs), whole blood samples were diluted with sterile PBS in a 1:1 ratio, overlaid on Lymphoprep^TM (Axis-Shield PoCAS, Oslo, Norway) in SepMete-50^TM (VERITAS Corporation, Tokyo, Japan), and were centrifuged at 1,200 × g and 25°C for 10 min. Thereafter, the top layer containing hPBMCs was collected and washed twice with PBS. Human monocytes were isolated from the hPBMCs using a Human Pan Monocyte Isolation kit (Miltenyi Biotec Inc., Auburn, CA, USA) according to the manufacturer’s instructions. hPBMCs and monocytes were washed and suspended in RPMI-1640 (Wako, Osaka, Japan) supplemented with 10% heat-inactivated FBS (Sigma-Aldrich) and 1% penicillin/streptomycin (Gibco, Carlsbad, CA, USA); thereafter, the cells were plated as indicated below.

hPBMCs (5 × 10⁵ cells/well) were cultured in the presence or absence of each compound. Following preincubation with the respective compound for 30 min at 37°C under 5% CO₂, the cells were stimulated by adding PHA (final concentration 5 μg/mL) and were incubated for 48 h at 37°C under 5% CO₂. For surface phenotyping, the cells were washed in 2 mM EDTA-PBS, centrifuged at 400 × g for 1 min, and resuspended in 50 μL Human BD Fc Block™ (BD Biosciences, Franklin Lakes, NJ, USA) diluted to 1:200 using 2 mM EDTA-PBS for 15 min at 4°C, and were then washed with 2 mM EDTA-PBS. Staining was performed using Brilliant Stain Buffer (BD Biosciences) with combinations of monoclonal antibodies ( Supplemental Table 4 ) with Fixable Viability Dye eFluor 780 (Thermo Fisher Scientific) for 20 min at 4°C. Thereafter, the cells were washed with Stain Buffer (BD Biosciences) and fixed with fixation buffer (BD Biosciences). Next, the samples were washed and resuspended in staining buffer. Specifically, for intracellular staining, the cells were centrifuged at 400 × g for 1 min and were resuspended in 50 ng/mL PMA, 1 μg/mL ionomycin, and 3 μg/mL brefeldin A solution diluted in RPMI-1640 with 10% FBS. After incubation at 37°C for 3 h under 5% CO₂, the cells were washed and resuspended in 50 μL Human BD Fc Block diluted to 1:200 with FACS buffer (PBS with 2 mM EDTA and 5% FBS) for 15 min at 4°C. Each sample was then fixed and permeabilized using Foxp3/Transcription Factor Fixation/Permeabilization Concentrate and Diluent (Thermo Fisher Scientific). This was followed by staining using V500 anti-human CD3 at 1:50 (clone: UCHT1, 561416; BD Biosciences), BV421 anti-human CD4 at 1:200 (clone: OKT4, 317434; BioLegend, San Diego, CA, USA), APC anti-human IL17A at 1:100 (clone: BL168, 512334, BioLegend), and PE anti-Foxp3 at 1:100 (clone: 236A/E7, 12-4777-42, Thermo Fisher Scientific) for 20 min at 4°C. The samples were then washed and resuspended in a permeabilization buffer. Finally, the cells were subjected to flow cytometry and were analyzed using a BD LSRFortessa™ X-20 system (BD Biosciences). Data were collected using FACSDiva software (BD Biosciences) and were analyzed using FlowJo 10.8.0 (TreeStar, Ashland, OR, USA).

Monocytes (1 × 10⁴ cells/well) isolated from PBMCs using a MACS system were cultured in the presence or absence of each compound. Following preincubation with the compounds for 30 min at 37°C under 5% CO₂, the cells were stimulated with or without 50 ng/mL M-CSF and were cultured for seven days at 37°C under 5% CO₂. Total RNA was isolated from the cells using an RNeasy 96-well Kit (Qiagen, Hilden, Germany) according to the manufacturer’s instructions, and quantitative RT-PCR was performed using a QuantStudio 7 Flex system (Thermo Fisher Scientific) with a TaqMan^® RNA-to-Ct 1-Step Kit (Thermo Fisher Scientific) and a TaqMan probe (Thermo Fisher Scientific; human MMP9, Hs00957562_m1; human CD163; Hs00174705_m1, human TNF Hs00174128_m1; human ACE2, Hs01085333_m1; human ACE, Hs00174179_m1; human DPP4, Hs00897386_m1; human MasR, Hs00267157_s1; and human GAPDH, Hs02786624_g1). PCR was conducted after reverse transcription at 48°C for 15 min and activation of AmpliTaq Gold DNA polymerase at 95°C for 10 min, followed by 45 cycles of 95°C for 15 s and 60°C for 1 min. To determine the Ct values, data were analyzed using the QuantStudio 6 and 7 Flex Real-Time PCR System software version 1.2 (Thermo Fisher Scientific). RNA expression values were then normalized to the expression level of the housekeeping gene (GAPDH), were calculated using the ΔΔCt method. Results are reported as expression levels relative to the control or M-CSF stimulation samples.

Statistical analyses of FAERS and MarketScan data were performed using R software version 4.1.2, RStudio (2022.07.1 Build 554 software, R Foundation for Statistical Computing), and Graphpad Prism for Windows version 6.07 (GraphPad Software Inc., San Diego, CA, USA). The R package “survival” was used to perform time-series analysis and to compare cumulative incidence, the log-rank test was used. The R package “MatchIt” was used to perform 1:1 propensity score matching and to compare population characteristics, and Chi-square tests were performed. Statistical analyses of changes in HbA1cs were performed using two-tailed Wilcoxon matched-pairs signed rank test and unpaired t-test with Welch’s correction. One-way analysis of variance and Tukey’s multiple comparison tests were used to analyze data collected following in vitro experiments, using GraphPad Prism software for Windows version 6.07 (GraphPad Software). Statistical significance is reported at p < 0.05.

First, we investigated the association between the use of a given drug and the incidence of BP based on FAERS data by calculating the reporting odds ratio (ROR) and Z scores during disproportionality analyses ( Supplemental Table 1 BP_DrugA). We observed a class of DPP4 inhibitors that showed a strong association with BP emergence and high RORs and Z scores ( Figure 1A ). We selected all DPP4 inhibitors that occurred in FAERS data, i.e., vildagliptin, linagliptin, sitagliptin, alogliptin, teneligliptin, anagliptin, and saxagliptin, for further analyses, and metformin, frequently used with DPP4 inhibitors, exhibited high ROR and Z score. Additionally, 18 drugs, categorized as ‘likely’ or ‘probably’ associated with the risk of BP incidence in a previous review (16), showed statistically significant RORs ( Table 1 ). After excluding patients who received concurrent metformin or any of the 18 risk drugs, we calculated the RORs of DPP4 inhibitors, which remained significant ( Table 2 ). When the patients were stratified by age and sex, BP incidence increased in all categories. Furthermore, the elderly population (aged ≥ 80 years) showed a higher odds ratio, as also observed in a previous study (17).

We searched for concomitant drugs that reduced the RORs of BP in patients receiving DPP4 inhibitors using FAERS data. Thus, several concomitant drugs, including lisinopril, metoprolol and losartan, that could decrease the ROR of BP in patients treated with DPP4 inhibitors were identified ( Figure 1B and Supplemental Table 1 DPP4iBP_DrugB). Among these drugs, lisinopril, which by itself did not affect BP risk, showed the highest absolute Z score. The other identified angiotensin (Ang)-converting enzyme (ACE) inhibitors did not have a significant effect on DPP4 inhibitor-associated BP.

To examine the causal relationship between the use of DPP4 inhibitors and/or lisinopril and BP onset, we analyzed MarketScan data. The ingredients of the DPP4 inhibitor and lisinopril formulations are shown in Supplemental Table 2 . Further, BP cases were defined based on BP-related symptoms, as listed in Supplemental Table 3 . Therefore, considering the time distribution of the first event after enrolment in the MarketScan database, the number of patients who were initially prescribed DPP4 inhibitors or lisinopril was markedly higher during the first six months and plateaued after seven months ( Supplemental Figure 1 ), indicating that patients prescribed DPP4 inhibitors or lisinopril within six months after enrolment may have used these drugs before enrolment. We thus excluded patients who received a diagnosis of BP or a prescription of DPP4 inhibitors or lisinopril during the 0- to 180-days run-in period from the study cohort. Thereafter, we evaluated the overall association between the use of DPP4 inhibitors and the onset of BP by estimating the IRR of BP. The DPP4 inhibitor group showed high IRR values ( Table 3 ). We also evaluated the effects of changes in blood glucose levels on BP onset, showing that prescription of DPP4 inhibitors reduced HbA1c levels with or without BP onset. However, changes in HbA1c levels did not differ between the groups ( Supplemental Figure 2 ), indicating that prescription of DPP4 inhibitors increased the risk of BP, regardless of their blood glucose level-reducing effect. To estimate the effects of lisinopril on BP onset after prescription of DPP4 inhibitors, we defined the concomitant use of lisinopril as the use of lisinopril after prescription of DPP4 inhibitors, and we divided the DPP4 inhibitor cohort into two populations: those who received lisinopril (n = 34,480) and those who did not (n = 78,069). In the next step, we performed 1:1 propensity score matching ( Table 4 ) to eliminate known confounding factors associated with the onset of BP and the prescription of DPP4 inhibitors and lisinopril (18, 19). In the matched cohorts, combination with lisinopril showed significantly reduced BP incidence, with a hazard ratio of 0.53 (95% confidence interval [CI]: 0.35–0.81, p = 0.003 in the log-rank test; Figure 2 ). Further, in these matched cohorts, daily and cumulative doses as well as the DPP4 inhibitor administration period were equivalent in each pair, with or without lisinopril use ( Table 5 ).

Several studies have suggested that the production of matrix metallopeptidase 9 (MMP9) from immune cells is associated with skin-blister formation during BP (20) and that ACE inhibitors suppress MMP9 expression and inhibit MMP9 activity (21–24). Therefore, we hypothesized that the molecular mechanism underlying the bidirectional effects of DPP4 inhibitors and lisinopril on BP is closely related to the regulation of MMP9 function. It has been reported that monocytes, macrophages, neutrophils, and eosinophils are sources of MMP9 in cases with BP (25–28). However, eosinophil infiltration into the skin is significantly less frequent in BP patients treated with DPP4 inhibitors than that in patients who do not receive DPP4 inhibitors (29). It was also reported that the reduction of skin infiltrating and circulating eosinophils was more pronounced in DPP4 inhibitor-associated BP with a noninflammatory phenotype (30). These data indicated that contribution of eosinophils in the DPP4 inhibitors received BP patients with a noninflammatory phenotype might less than in the idiopathic BP patients. In addition, monocytes enhance neutrophil-induced blister formation in ex vivo model (26). Hence, we focused on MMP9 production by monocytes and macrophages using hPBMCs. In the experiment, monocytes from healthy volunteers were stimulated using M-CSF for seven days, with or without vildagliptin, which showed the highest ROR, in line with the analyses of FAERS data. We used 1 and 10 μmol/L vildagliptin because its plasma concentration is up to 10 μmol/L in clinical use (31, 32); we observed that the vildagliptin-treated cells showed higher MMP9 mRNA expression, compared to their untreated counterparts, in a concentration-dependent manner ( Figure 3A ). Further, to determine the effects of DPP4 inhibitor treatment on T cell differentiation, hPBMCs were stimulated with PHA, with or without vildagliptin and sitagliptin, and the proportions of Th1, Th17, and Treg cells were analyzed using flow cytometry. Thereafter, the classification of each T cell subset was determined using cell surface markers ( Supplemental Figures 3, 4 ), based on a previous review (33) or a combination of IL-17 and Foxp3 ( Supplemental Figure 5 ). However, we did not observe any distinct changes in these cell populations following treatment with vildagliptin or sitagliptin. We subsequently investigated the effect of lisinopril on DPP4 inhibitor-induced MMP9 mRNA changes. We used 10 μM lisinopril, a concentration that is more than 10-fold higher than the plasma concentration recommended for clinical use (34) but does not inhibit MMP9 directly (21), to evaluate the maximum inhibitory effect of lisinopril on ACE. The increase in MMP9 mRNA expression induced by vildagliptin was reduced by the concomitant of lisinopril treatment; however, using lisinopril only did not affect MMP9 mRNA levels ( Figure 3B ). To identify the phenotype of these cells, we measured TNF (M1 type) and CD163 (M2 type) mRNA levels, which showed an increase in mRNA levels of the M2 type marker following vildagliptin treatment. We also observed that this increase was reduced by lisinopril combination treatment ( Figure 3C ). mRNA levels of the M1 type marker tended to decrease under the vildagliptin treatment; however, the effect was not significant ( Figure 3D ).

ACE and ACE2 are the key enzymes responsible for the generation of several components of the renin-angiotensin system (RAS), including Ang II, Ang III, Ang1-7, and Ang1-9, and RAS components affect various cell responses via their receptors. We examined the effects of DPP4 inhibitors on ACE and ACE2 mRNA expression in monocytes/macrophages, which showed that vildagliptin increased ACE2 mRNA expression, but did not affect ACE mRNA levels ( Figures 4A, B ). Further, combination with lisinopril suppressed the upregulation of ACE2 mRNA by vildagliptin ( Figure 4A ). In contrast, DPP4 mRNA expression was not affected by either drug ( Figure 4C ), suggesting that ACE2 mRNA expression plays an important role in enhancing MMP9 mRNA expression in monocytes/macrophages. Recent studies have suggested that Mas receptor (MasR) and Ang1-7 modulate M1/M2 macrophage polarization (35, 36). Further, based on in vivo chronic kidney disease models, DPP4 inhibitors were reported to enhance Ang1-7 production through ACE2 expression (37). We thus evaluated whether DPP4 inhibitors affected the Ang1-7/MasR axis, showing that vildagliptin did not affect MAS1 (MasR) expression ( Figure 4D ). We then used A779, a MasR inhibitor, to examine the effects of vildagliptin on the Ang1-7/MasR axis, showing that the vildagliptin-induced increase in MMP9 mRNA levels was suppressed by combination with A779; however, A779-only treatment did not affect MMP9 mRNA levels ( Figure 5 ).