Participants

The Regional Ethical Review Board of Linköping, Sweden approved the study protocol, which was carried out in accordance with the Declaration of Helsinki (Swedish Medical Product Agency with the Clinical Trial Number EudraCT: 2017-000317-22, registered April 13, 2017, and Regional Ethical Review Board, Dnr 2017/111-31, approved April 10, 2017). All women gave written informed consent.

Healthy postmenopausal women (55–74 years of age) from the mammography screening program at Linköping University Hospital who were categorized as having BI-RADS D (i.e.,,, extremely dense) breast tissue according to the Breast Imaging Reporting and Data System [26] were invited to participate in the study. Inclusion criteria were postmenopausal, defined as more than 12 months of amenorrhea; a baseline breast density score of BI-RADS D for the left breast; no serious co-morbidity; willingness to take ASA for 6 months; and willingness to avoid regular intake of NSAIDs outside of the trial. Exclusion criteria were any sex steroid use within the last 3 months, including systemic hormone replacement therapy; use of contraceptives, including hormonal intrauterine devices; use of anti-estrogen therapy, including selective estrogen modulators or degraders and aromatase inhibitors; previous interventions of the breast; current regular use of NSAIDs; known intolerance to or contraindications for NSAIDs; diabetes; current smoking; uncontrolled hypertension; or any contraindication for MRI, including allergy to contrast agents. Multimodal MRI followed by microdialysis was performed for all women at baseline and after 6 months.

MRI data acquisition and analysis

A 3 T Ingenia MRI scanner (Philips Healthcare, Best, The Netherlands) with a dual-breast 16-element breast coil was used for MRI data acquisition. Co-localization and analysis of lean tissue fraction (LTF) and proton density fat fraction (PDFF), diffusion-weighted images (DWI), and perfusion dynamics were performed using Matlab (Version 2020b, The Mathworks Inc, Natick, Massachusetts, USA). Quantitative relaxometry (qMRI) was also performed.

Magnetic resonance spectroscopy

A magnetic resonance spectroscopy (MRS) voxel (i.e.,,, volume of interest [VOI]) with a minimum size of 10 × 10 × 10 mm3 (but up to 20 mm side length in some participants) was located in the upper lateral quadrant of the left breast, which was the same location as the microdialysis catheter. MRS data were acquired with TR 2000 ms and TE 35, 70, 140, and 180 ms with PRESS as the volume selection method in 16 averages. Post-processing was performed using LCModel (http://s-provencher.com/lcmodel.shtml).

Data from different acquisitions (i.e.,,, perfusion, PDFF, LTF, and diffusion) were extracted from the same VOI as was used for MRS using eTHRIVE pre-contrast agent injection. Volumetric and positional information from image metadata were used to find the corresponding voxel coordinates of the VOI for perfusion, LTF, and DWI volumes, respectively.

Lean tissue fraction

Axial 3D four-point echo turbo field MRI images were acquired with anterior–posterior frequency encoding using an initial TE 1.15 ms and ΔTE 1.15 ms, TR 10 ms, flip angle 10°, field of view (FoV) 300 × 316 mm2, 120 × 127 scan matrix, and 1.8 mm slice thickness. Data acquisition was performed separately before and after intravenous contrast agent injection.

Water- and fat-separated MRI images were constructed using four echoes [27], and LTF maps were constructed as the ratio of lean tissue volume to total volume. Quantitative fat images were computed by calibrating the original fat images using adipose tissue as an internal intensity reference, which also allowed adipose tissue volume to be quantified within the segmentation [28, 29]. LTF data were extracted from LTF images within the MRS VOI.

Diffusion measurements

Diffusion data were obtained using an echo-planar (EPI) single shot DWI with a TR/TE 11,671/91 ms, EPI factor 75, three averages, fat suppression with 'SPectral Attenuated Inversion Recovery' (SPAIR), FoV 300 × 317 mm2, slice thickness 3.0 mm, gap 0.3 mm, resolution 2.2 × 2.2 mm2 (recon 1.0 × 1.0 mm2), and b-values 0, 100, and 850 s/mm2. ADC maps were then computed using manufacturer-provided MRI scanner software. ADC data were extracted from ADC images within the MRS VOI.

Perfusion measurements

Perfusion data were obtained from a 3D spoiled gradient echo with TR/TE 2.5/1.27 ms, flip angle 12°, matrix 136 × 140, FoV 300 × 309 mm2, recon resolution 1.0 × 1.0 mm2, 4.4 mm slice thickness, and -2.2 mm gap. Sixty dynamics were collected during 3:10 min and five dynamics after 7 min, with each dynamic lasting 3 s. Contrast injection started at the same time as perfusion sequences at a rate of 2 mL/s, resulting in six or seven dynamics without contrast media. The contrast medium was 10 mL gadoteric acid (Clariscan®, GE Healthcare Limited, Little Chalfont, England) followed by 30 mL saline using a MEDRAD® MRXperion injector (Bayer HealthCare, Whippany, NJ, USA).

Perfusion characteristics were measured using both a mathematical fitting procedure on the entire MRS VOI and a blinded visual review procedure by radiologists in a selected single slice in the same location as the MRS VOI.

Mathematical modeling. Mathematical fitting was performed by extracting temporal data from the entire MRS VOI. To calculate the area under the perfusion curve (AUC) and time constant (tau), the mean intensity values within each temporal VOI were normalized to baseline values and applied to all acquisitions in the perfusion series. Normalized data points were fitted in Matlab using a non-linear least-squares algorithm as previously described [27].

Radiological review. Representative time points were selected through the region of interest in the upper lateral quadrant of the left breast just before the perfusion curve peaked (usually around 120 s (i.e.,,, “wash-in”)) and after the curve peaked (usually around 300 s (i.e.,,, “wash-out”)). If the curve was constantly rising, times around 120 s and 300 s were chosen. Wash-in and wash-out were expressed as the signal increase (%) relative to baseline intensity. This procedure was performed using IntelliSpace Portal (ISPv11.0, Philips Healthcare, Best, Netherlands).

Relaxometry

Relaxometry (and synthetic MRI) was performed using 3D-QALAS [30], a multi-dynamic 3D gradient multi-echo sequence with TR/TE 5.0/2.3 ms, matrix 179 × 178, FoV 300 × 320 mm2, and slice thickness 1.8 mm. Quantitative R1 and R2 relaxation rates and proton density PD maps were generated by SyMRI (SyntheticMR, Linköping, Sweden). Entire breasts were segmented from the whole acquisition volume, but an additional 5 mm was removed from the edges to suppress the contribution of skin tissue. The images were further segmented based on selected ranges in quantitative R1, R2, and PD into fibroglandular tissue, fat, and remaining tissue including milk ducts and blood vessels. Based on the quantitative R1, R2 and PD maps synthetic T1-, T2- and PD-weighted images were reconstructed.

Conventional imaging sequences

Conventional clinical sequences for diagnostics were also acquired, such as STIR (TR/TE 4370/65 ms, inversion time 240 ms, matrix 300 × 253, FoV 300 × 320 mm2, slice thickness 3.0 mm, gap 0.3 mm, recon resolution 0.7 × 0.7 mm2), T2 3D (TR/TE 2000/308 ms, matrix 378 × 397, FoV 300 × 317 mm2, slice thickness 2.0 mm, gap -1.0 mm, recon resolution 0.7 × 0.7 mm2), and T1 3D fat suppressed (TR/TE 5.9/2.9 ms, matrix 500 × 531, FoV 300 × 317 mm2, slice thickness 1.0 mm, gap -0.5 mm, recon resolution 0.5 × 0.5 mm2).

Microdialysis procedure

Prior to the insertion of microdialysis catheters, 0.5 mL lidocaine (10 mg/mL) was administrated intracutaneously. Microdialysis catheters (M Dialysis AB, Stockholm, Sweden), which consisted of a 20-mm-long tubular dialysis membrane (diameter 0.52 mm, 100,000 atomic mass cut-off) glued to the end of a double-lumen tube, were inserted via a splittable introducer (M Dialysis AB) connected to a microinfusion pump (M Dialysis AB) and perfused with 154 mmol/L NaCl and 60 g/L hydroxyethyl starch (Voluven®; Fresenius Kabi, Uppsala, Sweden) at 0.5 µL/min. One catheter was placed in the upper lateral quadrant of the left breast directed toward the nipple, and one catheter was placed in abdominal subcutaneous (s.c.) fat as previously described [31,32,33,34,35,36,37,38,39]. After a 60-min equilibration period, outgoing perfusate was stored at − 80 °C for subsequent analysis.

Protein quantification

Microdialysis samples were analyzed using multiplex proximity extension assay (Olink Bioscience, Uppsala Sweden) as previously described [40,41,42]. Briefly, 1 μL sample was incubated with proximity antibody pairs tagged with DNA reporter molecules. The DNA tails formed an amplicon by proximity extension, which was quantified by high-throughput real-time PCR (BioMark™ HD System; Fluidigm Corporation, South San Francisco, CA, USA). The generated fluorescent signal correlates with protein abundance by quantitation cycles (Cq) produced by BioMark Real-Time PCR software. To minimize variation within and between runs, data were normalized using both an internal control (i.e.,,, extension control) and an interplate control and transformed using a predetermined correction factor. Pre-processed data were provided in the arbitrary unit normalized protein expression (NPX) on a log2 scale, which was then linearized using the formula 2NPX. A high NPX value corresponds to a high protein concentration. Values represent relative quantification, meaning that absolute levels between different proteins cannot be compared.

Statistical analyzes

This trial was primarily designed to test the hypothesis that in vivo extracellular levels of inflammatory proteins decrease in women treated with ASA compared with untreated women. For sample size calculations, an effect size of 0.8 for IL-6, IL-8, and CCL5 was assumed based on mean and standard deviations from previous reports [9]. Considering a power of 0.8 and a significance level of 0.05, 25 participants per treatment arm were required. Secondary MRI outcomes were analyzed using an exploratory approach. Data were analyzed using paired Student’s t-tests, and correlations were calculated using Spearman’s rho. Proteomic data were analyzed using the two-stage set-up method of Benjamini, Krieger, and Yekutieli with a false discovery rate (FDR) set at 5%. A p < 0.05 was considered statistically significant. Statistical analyzes were performed with Prism 9.0 (GraphPad Software, USA).