This randomized cross-over study was conducted at the Norwegian University of Science and Technology, in Trondheim, Norway. The study was approved by the Regional Committee for Medical and Health Research Ethics (REK-263493) and pre-registered in clinicaltrials.gov (NCT05042414, 13/09/2021). Results concerning the effect of exercise on human milk adiponectin concentrations have been published previously (6). Inclusion criteria were females aged ≥18 years, exclusively breastfeeding a singleton, term infant aged 6–12 weeks, living in the Trondheim area, and being able to walk or run on a treadmill for >50 min. Exclusion criteria were known cardiovascular disease or diabetes mellitus type 1 or 2. Gestational diabetes was not an exclusion criterion. All participants signed an informed, written consent prior to assessments. After baseline assessments, the participants underwent three conditions in random order: REST (sitting), moderate-intensity continuous training (MICT), and high-intensity interval training (HIIT) (Figure 1). There was a washout period between conditions of ≥48 h. A computer random number generator developed at the Faculty of Medicine and Health Science at NTNU was used to randomize the sequence of the conditions for each participant on the first test day. The last author performed the randomization and received the allocation order on screen and by e-mail. The participants did not get information about the sequence of conditions and were only informed about which condition they were undertaking on the test days. Neither participants nor investigators were blinded due to the nature of the intervention (exercise). All methods were performed in accordance with the relevant guidelines and regulations.

On a separate day before the three conditions, we collected data on background characteristics, physical activity levels, body composition, and maximum oxygen uptake (VO₂max). These assessments were undertaken >48 h before the first condition. We used a bioimpedance scale (InBody720 BioSpace Co., Republic of Korea) to estimate body composition. We measured VO₂max during a maximal effort exercise test on a treadmill using direct analysis of expired gases (MetaLyzer II, Metasoft, CORTEX Biophysik, Germany). The participants wore heart rate monitors (Polar, Finland) during the exercise testing, and we used the highest recorded heart rate during the test as estimates of individual heart rate maximum (17). The participants completed questionnaires on background characteristics and the International Physical Activity Questionnaire (18). They were requested to avoid exercise >48 h prior to laboratory assessments and experimental procedures. The participants recorded their dietary intake from the evening before the first test day and on the first test day. This recording included the type of foods consumed, amount, and timing of intake. We asked them to repeat the same dietary intake on the later test days, including the type of foods consumed, the amount, and the time of day (Figure 1).

The participants consumed breakfast at 08:30 h and lunch at 13:00 h on all test days, and all three experimental conditions took place at 10:00 h. Immediately after the human milk collection at 11:00 h, the participants consumed a standardised snack. They chose between a banana or a crispbread with cheese and the snack was the same for all conditions for each participant. During the REST condition, the participants rested for 45 min seated in a comfortable chair at the laboratory. MICT was completed as 48 min of walking or jogging at an intensity corresponding to 70% of heart rate maximum, whereas the HIIT protocol was 4 × 4 min HIIT at 90–95% of heart rate maximum, as previously described (6). We used the recorded heart rate with 5-min intervals during MICT and the average heart rate in the last 2 min of every work-bout during HIIT to estimate the actual exercise intensity as a measure of compliance with the exercise protocols.

The participants were supplied with electronic breast pumps (Medela Swing Flex, Medela AG, Switzerland). Human milk was sampled at four time-points on the experimental test days: at 07:00 (before breakfast), 11:00 (immediately after exercise/rest), 12:00 (1 h after exercise/rest), and 15:00 (4 h after exercise/rest) (Figure 1). Only the 07:00 h sample was collected in the fasted state. We asked the participants to provide us with ≥25 mL at each time-point, from the same breast. The participants stored the first (07:00) and last (15:00) samples in their home freezer and transported these on ice to the laboratory. The samples from the remaining time-points were obtained in the laboratory and immediately stored at −80° until analysis. The participants could breastfeed their infant at any time during the test days and we did not control for this in the analysis.

The milk was thawed at room temperature before centrifugation at 10,000 × g for 60 min. We carefully removed the fat layer on the top using tweezers and used the skimmed milk for analysis. We used enzyme-linked immunosorbent assay (ELISA) for quantitative measurement of insulin (IBL International GmBH, Germany, product number RE53171), using a Dynex DS2 automation system programmed with DS-Matrix software (Montebello Diagnostics AS, Norway). The intra-assay variability for the insulin kit is <3% and inter-assay variability is 6%. All samples from each participant were measured in duplicate wells using the same microtiter plate and kit. We reduced the time for the final incubation step to 12 min (instead of 15 min in the manufacturer’s instructions), based on pilot testing of the kits, otherwise, we followed the manufacturer’s instructions. The range of the ELISA assay was 1.76–100 μIU/mL and all measurements were obtained in the linear range of the assay, with a coefficient of determination for the standard curve of 1.

We did not do a formal sample size calculation due to the exploratory nature of the research question but aimed to include 20 participants. The advantage of cross-over studies is that it allows comparison at the individual level rather than the group level and fewer participants are required in a cross-over design compared with a parallel group design. To determine the effect of HIIT and MICT on human milk adiponectin concentrations, we used a linear mixed model with time and the interaction between time and condition as fixed effects and participant ID as random effects. The first time-point (07:00 h) and REST were used as reference categories for time and condition, respectively. Data were log-transformed to achieve normally distributed residuals. We consider p-values <0.05 as statistically significant. We used IBM SPSS 29.0.1.0 for the statistical analysis. Conditional effect estimates were calculated by exponentiating the coefficients from the linear mixed model analysis, which gives multiplicative factors and represents estimated effects in individuals with equal random intercepts.

Recruitment began in August 2021 and concluded in May 2022. We ended the recruitment when we had reached the pre-specified number of participants that we aimed to include. We included 20 participants, who completed all three conditions (Figure 2). Table 1 shows their baseline characteristics. No adverse events were reported. Insulin was detected in all milk samples, with an average concentration of all 240 samples of 12.3 (SD 8.8) μIU/mL (range 3.2–57.2 μIU/mL). Within the 12 samples obtained from each participant, the average SD was 5.2 μIU/mL (Supplementary Table S1). Table 2 shows the observed data for insulin concentrations at each time-point in the three conditions.

The washout period between conditions was on average 7.3 (SD 2.0) days between the first and second condition and 6.9 (SD 1.4) days between the second and third conditions. For some of the samples, there was a small deviation (average 2.5 min, SD 10.1 min) from the prescribed time-points for sampling. In the MICT condition, the participants exercised at an intensity of 70% (SD 1) of heart rate maximum, whereas the average intensity during the last 2 min of HIIT was 96% (SD 2) of heart rate maximum.

There was no statistically significant effect of either exercise condition on human milk insulin concentration at any of the sampling time-points (Figure 3). However, there was a tendency (p = 0.093) of attenuated increase in insulin concentrations 4 h after exercise after HIIT, compared with the REST condition (Figure 3 and Table 3). Compared with the milk collected at 07:00 h, insulin concentrations were higher in the samples collected immediately after exercise (11:00 h) and in those collected 4 h after exercise (15:00 h, Table 3). The estimated effect of MICT and HIIT at each time-point on can be found by multiplying the exponentiated beta coefficients (estimate) for the main effect of time at each time-point with the exponentiated estimate for the interaction between time and exercise. The estimated relative change from the first to the last time-point was 72% (e^0.54) for REST, 70% (e^0.54* e^−0.01) for MICT, and 41% (e^0.54* e^−0.20) for HIIT.