This was an interventional, multicentre, randomised, open-label, three-group, full replicate crossover study. Study participants were screened and assigned to one of three study intervention groups (groups 1–3) at three investigational sites in the United States (n = 2; Kansas City and Los Angeles) and Canada (n = 1; Montreal). Group 1 participants were recruited at the Kansas City site, group 2 participants were recruited at the Kansas City and Montreal sites and group 3 participants were recruited at the Los Angeles site. The study was designed to confirm bioequivalence between each of the three dose levels of 2G oral semaglutide (1.5, 4, 9 mg) and the three dose levels of 1G oral semaglutide (3, 7, 14 mg) at SS: 9 mg versus 14 mg (group 1), 4 mg versus 7 mg (group 2) and 1.5 mg versus 3 mg (group 3). Doses for the 2G formulation were selected based on a previous comparative bioavailability study (NCT04097600) [27]. The study protocol was approved by appropriate health authorities according to local guidelines and by the Institutional Review Board/Independent Ethics Committee. The study was conducted in accordance with the Declaration of Helsinki [31] and International Council for Harmonization Good Clinical Practice guidelines [32]. Participants provided written informed consent prior to the commencement of any study-related activities. This study is registered with ClinicalTrials.gov (NCT05227196).

Key inclusion criteria included healthy male and female adults aged 18–64 years with a body mass index between 21.0 and 32.0 kg/m2. Participants were required to be generally healthy as judged by the investigator. Full eligibility criteria are in Table S1 in the Supplementary Appendix. In accordance with regulatory guidelines, healthy participants were chosen and test conditions standardised to minimise variability in factors that may affect the assessment of PK parameters [28, 29].

The study comprised a screening period of ≤ 28 days; a dose-escalation period for groups 1 and 2; and four steady-state periods, each of 5 weeks’ duration (Fig. 1). Participants underwent a screening visit 1–28 days before recruitment to group 1, 2 or 3. Within each group, participants were randomised (1:1) to one of two treatment sequences of alternating oral formulations (1G and 2G) of semaglutide as detailed below (Fig. 1). An open-label approach was considered acceptable as the PK endpoints were not judged to be affected by knowledge of treatment allocation. It was not considered feasible to attain blinding by changing the formulation of the visually distinct tablets (e.g. by encapsulation) as this would affect absorption [29, 33]. However, the sponsor staff who were involved in the evaluation of PK data were blinded to treatment allocation until after database lock of the PK data. Randomisation was followed by 2 weeks of dose escalation for group 1 or 1 week of dose escalation for group 2 (Fig. 1). Each week of dose escalation was initiated with a single on-site dose on the first day, followed by self-administered once-daily home dosing for 6 days. Dose escalation ended with an on-site visit before initiation of the first 5-week steady-state period. Participants in group 3 did not undergo dose escalation and proceeded directly to the first steady-state period.

Study design. aNot all visits are shown in the diagram. 1G first generation, 2A dose escalation visit (group 1: day 1), 2B dose escalation visit (group 1: day 8; group 2: day 1), 2G second generation, N planned number of participants (please refer to Results and Fig. 2 for the actual number of participants enrolled); PK pharmacokinetic

The four steady-state periods comprised a total duration of 20 weeks. For all groups, the 5-week steady-state periods started with a single on-site dose followed by self-administered once-daily home dosing. The half-life of semaglutide is approximately 1 week, therefore the participants were dosed daily for 5 weeks (corresponding to five half-lives) in each period to ensure evaluation of bioequivalence under SS conditions. A Test (T) and Reference (R) crossover design of TRTR/RTRT was used, whereby the two treatment arms within each group alternated between 2G oral semaglutide and 1G oral semaglutide (Fig. 1). Each steady-state period ended with a 24-h on-site PK sampling visit (visits 5, 7, 9 and 11) before proceeding to the next steady-state period (or follow-up). Participants arrived the day before the last dose in the 5-week period and stayed on-site overnight. The following morning, body weight was measured, and a pre-dose blood sample was taken before participants received the last dose of the 5-week period, and repeated blood sampling was subsequently performed over the next 24 h for the PK analyses supporting evaluation of the co-primary endpoints. Participants were instructed to fast overnight (≥ 6 h) prior to each dosing and before on-site visits; water intake was not allowed ≤ 2 h before dosing. During the 30 min post-dose, no intake of food, liquid or other oral medication was allowed. Initiation of breakfast meal was carried out 30–35 min post-dosing for both on-site and home dosing.

A follow-up visit was scheduled ≥ 5 weeks after administration of the last dose for safety assessments. The planned total duration of study participation for each participant was 177–217 days (Fig. 1), depending on the group.

It was aimed to have an even distribution of male and female participants with ≥ 40% of each sex recruited into each group.

The primary study objective was to confirm bioequivalence of the two oral formulations of semaglutide (2G vs 1G) administered once daily at three dose levels at SS in healthy participants, namely 2G oral semaglutide 1.5 mg versus 1G oral semaglutide 3 mg, 2G oral semaglutide 4 mg versus 1G oral semaglutide 7 mg, and 2G oral semaglutide 9 mg versus 1G oral semaglutide 14 mg. The co-primary endpoints were area under the semaglutide plasma concentration–time curve during a dosing interval at SS (AUC0–24h,SS) and maximum semaglutide plasma concentration at SS (Cmax, 0–24h,SS). The co-primary endpoints were derived after the last dose administered at the end of each steady-state period, i.e. after dosing on days 49/42/35 (visit 5), 84/77/70 (visit 7), 119/112/105 (visit 9) and 154/147/140 (visit 11) for groups 1/2/3, respectively (Fig. 1). Supportive secondary endpoints were semaglutide plasma concentration 24 h after last dose at SS (Ctau, 24h,SS) and time to maximum semaglutide plasma concentration at SS (tmax, 0–24h,SS). The percentage swing was calculated post hoc as ((Cmax, 0–24h,SS − Cmin, 0–24h,SS)/Cmin, 0–24h,SS)*100, where Cmin, 0–24h,SS is the minimum semaglutide plasma concentration at SS.

In line with EMA, FDA and PMDA guidelines, a crossover design was chosen to leverage the lower within-participant variability in PK measures and thereby limit the number of participants required for the study [28,29,30]. A replicate design was applied so that the within-participant variability for the reference product (1G oral semaglutide) could be determined, to enable the application of widened or reference-scaled bioequivalence acceptance criteria for highly variable drugs [28, 29].

The sample size was calculated to provide at least 90% power for showing bioequivalence between the 1G and 2G formulations for all three dose groups, based on the standard acceptance limits (0.8000–1.2500) to ensure sufficient power to confirm bioequivalence in accordance with the EMA, FDA and PMDA guidelines. Based on data from the previous bioavailability study [27], the expected AUC0–24h,SS treatment ratios (2G versus 1G) were 1.10 for group 1, and 1.08 for group 2 and group 3 and a coefficient of variation (CV) of 45% was considered reasonable. Due to the high correlation in treatment differences between AUC0–24h,SS and Cmax, the sample size only needed to be calculated for one of the endpoints as the other was expected to be approximately similar. The sample size was determined to provide a marginal power of 96.6% for AUC0–24h,SS for each group, resulting in at least 90% combined power to confirm bioequivalence for all three groups. A total of 160 (group 1) and 122 (groups 2 and 3) participants would need to be randomised to confirm bioequivalence for all three groups for AUC0–24h,SS, assuming that 85% of participants would complete the steady-state periods.

All dose comparisons of 2G oral semaglutide versus 1G oral semaglutide (1.5 mg vs 3 mg, 4 mg vs 7 mg and 9 mg vs 14 mg) were conducted separately. The primary endpoints were derived from the plasma concentration–time curve for 0–24 h following 5 weeks’ administration of 2G oral semaglutide and 1G oral semaglutide; derivations were based on actual time since dosing. AUC0–24h,SS was approximated using the linear trapezoidal method. Missing AUC concentrations were handled by the linear trapezoidal method given that the AUC can be reliably calculated. Missing plasma semaglutide concentrations were ignored provided that Cmax could be reliably determined. Missing endpoint data for both AUC and Cmax (e.g. due to missing concentrations that prevent AUC derivation) were assumed to be missing completely at random as they were not expected to be related to treatment. Primary endpoint analysis was performed on the full analysis set (FAS), comprising all randomised participants who were exposed to at least one dose of study product and completed at least one period on 2G oral semaglutide and at least one period on 1G oral semaglutide. All safety evaluations were made on the safety analysis set (SAS), which included all participants who were exposed to at least one dose of study product.

For each comparison of 2G oral semaglutide versus 1G oral semaglutide, two one-sided hypotheses were tested for AUC0–24h,SS and Cmax, 0–24h,SS values at each dose level: 1: H0: μ ≥ 1.25 against Ha: μ < 1.25, and 2: H0: μ ≤ 0.80 against Ha: μ > 0.80 (where μ = the geometric mean under treatment with 2G oral semaglutide, divided by the geometric mean under treatment with 1G oral semaglutide); a (one-sided) alpha of 5% was used for each comparison for each one-sided test, controlling for type 1 error at 5%.

For EMA bioequivalence analysis, co-primary endpoints were log-transformed and analysed separately using an analysis of variance (ANOVA) model with treatment (2G oral semaglutide or 1G oral semaglutide), period (PK-sampling visits), sequence (starting with 2G oral semaglutide or 1G oral semaglutide) and participant nested within sequence as fixed effects. For group 2 (in which participants were recruited from different sites), site was added to the model, so the full model included site, treatment, period nested within site, sequence, sequence nested within site and participant nested within sequence and site. The estimated treatment difference and the corresponding two-sided 90% confidence interval (CI) were back-transformed to the original scale. The 90% CI was then used for testing bioequivalence. The within-participant CV of the reference product (1G oral semaglutide) was > 30% for Cmax, 0–24h,SS, thus qualifying as a highly variable drug, in line with recent reports [22, 28]. Therefore, the acceptance limits were extended for Cmax, 0–24h,SS depending on the observed CV [28]. The standard limits of 0.8000–1.2500 were used for AUC. To establish bioequivalence, the Cmax, 0–24h,SS estimated treatment ratio was required to be fully contained within 0.8000–1.2500 and the 90% CI of the Cmax, 0–24h,SS estimated treatment ratio fully contained within 0.6984–1.4319 for group 1, 0.7273–1.3749 for group 2 and 0.7328–1.3646 for group 3, and the 90% CI of the AUC0–24h,SS treatment ratio was required to be fully contained within 0.8000–1.2500. Additional details are provided in Supplementary Table S2.

For FDA bioequivalence analysis, as the estimated within-participant standard deviation for oral semaglutide, s(WR), was ≥ 0.294 for both co-primary endpoints, the reference-scaled average bioequivalence (RSABE) analysis was the primary analysis for AUC0–24h,SS and Cmax, 0–24h,SS [34]. For the RSABE analysis, only participants with a value for an endpoint at all four steady-state periods contributed to the statistical analysis of the respective endpoint. To establish bioequivalence, the estimated treatment ratio should be fully contained within 0.8000–1.2500 and the 95% upper CI for (μT − μR)2 − θSσ2WR, based on Howe’s approximation, should be ≤ 0. Additional details are provided in Supplementary Table S2.

The PMDA guideline required that the 90% CI of the AUC0–24h,SS and Cmax, 0–24h,SS treatment ratios were between 0.8000 and 1.2500 [30]. The analysis used the same ANOVA model as for the EMA guidelines above, except that the standard acceptance interval of 0.8000–1.2500 for both AUC and Cmax was used.

A treatment-emergent adverse event (TEAE) was defined as an adverse event (AE) that occurs or worsens after receiving a treatment or intervention. TEAEs were coded using the current Medical Dictionary for Regulatory Activities (MedDRA) version and were reported in MedDRA version 26.0 [35]. TEAEs were summarised using descriptive statistics.