The objective of this study was to investigate if women with RA respond differently than a matched group of women without RA to a meal with red meat, reflected in the postprandial serum metabolome.

The study had a parallel single meal design. Samples were collected before the meal and at 30 min, one, two, three and five hours post ingestion of the meal. Women with RA were matched on age and body mass index (BMI) on group basis with women without RA.

Women with diagnosis of RA (International Classification of Diseases-code M05.9 or M05.8) in the age span 20–70 years listed at the Sahlgrenska University Hospital were identified through the Swedish Rheumatology Quality Register. Potential participants (n = 934) were contacted by letter. Women in the same age span without diagnosis for RA or other rheumatic diseases were recruited as controls by advertisement in social media, by word of mouth and by posters at official noticeboards. Inclusion criteria for patients with RA were ≥ 2 years since RA diagnosis and no changes in disease modifying anti-rheumatic drugs (DMARDs) the past 3 months. Exclusion criteria were underweight (BMI < 18.5 kg/m2) or obesity (BMI ≥ 30 kg/m2), diagnosis of cancer, diabetes, inflammatory bowel disease, celiac disease, allergy or intolerance to any of the foods served in the study, pregnancy, lactation, use of any lipid lowering medication, glucocorticoids or interleukin-6 (IL-6) inhibiting therapy during the past 4 weeks, smoking, hemoglobin levels ≤ 100 g/L, glycated hemoglobin (HbA1c) above reference range (18–50 years: > 42 mmol/mol and > 50 years: > 46 mmol/mol).

The PIRA (Postprandial Inflammation in Rheumatoid Arthritis) trial is Registered at Clinicaltrials.gov (NCT04247009) and was approved by the Swedish Ethical Review Authority (Dnr 2019-05242).

C-reactive protein (CRP), erythrocyte sediment rate (ESR), hemoglobin and HbA1c were measured in fresh samples according to the clinical routine at the Sahlgrenska University Hospital laboratory. Tender and swollen joints were examined by trained nurses at the Department of Clinical Rheumatology Research Center at the Sahlgrenska University Hospital. The patients filled out visual analogue scales (VAS) for global health, pain, and fatigue. Disease Activity Score 28-joints (DAS28) was calculated with erythrocyte sedimentation rate [19]. The participants’ medications were controlled through interviews and from patient records, and those fulfilling any of the exclusion criteria were excluded.

Physical activity was assessed based on scales between 1 and 5 on habitual physical activity and intentional physical exercise. Based on this, a physical activity index between 1 and 4 was calculated, resembling that previously validated by Wareham et al. [20]. Dietary quality index was assessed based upon food frequency questionnaires, whereby an index ranging from 1 to 12 was constructed, as previously described by the Swedish Food Agency, to assess habitual quality of diet [21].

The intervention meal consisted of two hamburgers. The burgers contained 130g minced meat (Produced by Scan, by Swedish meat products, 60% beef, 40% pork), 25 g egg, 8 g breadcrumbs and served with 84 g (2 slices) toasted white bread (Jättefranska, Pågen AB), 10 g (2 leaves) romaine lettuce, 10–20 g (4 slices) cucumber, 20–30 g (2 slices) tomato and 20 g vegan hamburger dressing (Hamburger dressing 250 mL, Rydbergs AB). All burgers were cooked to mid-temperature 75̊ °C in non-stick pans with 1 teaspoon of canola oil.

The burgers were cooked in two batches (December 2019, August 2021), due to the Covid-19 pandemic that halted the PIRA trial for 18 months. The differences in macronutrients between these two batches of meat burgers were negligible. Protein and fat content were analyzed by Eurofins Food & Feed Testing, Sweden. Carbohydrate content was calculated based on food labels; fiber content not included. This meal contained 35 g protein, 40 g fat, 47 g carbohydrates and about 700 kcal.

The main outcome in this study report was differences in postprandial metabolite pattern at 3 h after the study meal, between women with and without RA. Secondary outcomes were differences in metabolite patterns in the fasting state and other postprandial timepoints as well as differences in incremental area under the curve (AUCmin) for quantified metabolites.

During the postprandial meal challenges, a catheter was placed, and serum samples were taken in the fasting state and after 30 min, one, two, three and five hours post ingestion of the meal. Serum was collected in letting tubes (BD Vacutainer, 5 mL, reference no 367624), left for 30 min in room temperature, thereby refrigerated for 30 min before centrifugation for 10 min in 2600g. After centrifugation and isolation procedures, all samples were immediately stored in − 20 °C, and at earliest convenience transferred to − 80 °C for storage until analysis.

Metabolites were quantified by Nuclear Magnetic Resonance (NMR)-analysis; serum samples were prepared according to In Vitro Diagnostics Research (IVDr) standard operating procedures (Bruker BioSpin; www.bruker.com/products/mr/nmr/avanceivdr.html). In brief, serum samples were thawed at room temperature for 30 min, then centrifuged at 3500×g for 1 min at 4 °C. Thereafter, 325 μl of serum was transferred with a SamplePro L liquid handler (Bruker BioSpin) to a deepwell plate (Porvair, cat. no 53.219030) containing 325 μl NMR buffer ((75 mM sodium phosphate, pH 7.4, 0.08% 3-(trimethylsilyl) propionic-2,2,3,3-d4), 0.04% sodium azide, 20% v/v D2O) per well. The plate was shaken at 400 rotations per minute, 12 °C for 5 min in a Thermomixer Comfort (Eppendorf). Finally, 600 μl sample was transferred to 5 mm SampleJet NMR tubes with the SamplePro L. The sample tubes, deepwell plate and SampleJet rack were kept at 2 °C during the preparation in the SamplePro L robot.H NMR data was acquired on a Bruker 600 MHz Avance III spectrometer equipped with a room temperature 5 mm BBI probe and a cooled SampleJet sample changer. In brief, 1D NOESY (‘noesygppr1d’ pulse sequence), 1D CPMG (‘cpmgpr1d’) and 2D J-resolved (‘jresgpprqf’) spectra were acquired according to the standard IVDr parameter settings at 37 °C. A pre-acquisition temperature stabilization time of 300 s was used. Before measurement, all samples were kept at 6 °C in the SampleJet. Experimental parameters are available upon request. The 1H-NMR spectra were aligned by setting the TSP-d4 to 0 ppm using icoshift and the spectra were bucketed using the function “opt_bucket.m” [22]. This function used initial size of bucket = 0.04 and slackness = 0.5. The 1D NOESY data were also submitted for B.I.-Lisa lipoprotein profiling and B.I.Quant-PS 2.0.0 automatic quantification of a subset of metabolites through a remote secure Bruker server, generating in total 39 B.I.Lisa and 41 B.I.Quant-PS variables. After quality control using Brucker data on significant correlation for each sample and excluding metabolites where more than 50% of the samples had less than 85% significant correlation, 23 quantified serum metabolites remained for analysis. Included quantified metabolites were: trimethylamine-N-oxide, alanine, creatine, creatinine, glutamine, glycine, histidine, isoleucine, leucine, N,N-dimethylglycine, phenylalanine, tyrosine, valine, acetic acid, citric acid, formic acid, lactic acid, succinic acid, acetoacetic acid (acetoacetate), acetone, pyruvic acid, glucose, and dimethylsulfone.

For quantified metabolites, AUCmin, i.e., the area above the lowest value of all time points, were calculated. Phenylalanine is converted to tyrosine by phenylalanine (4)-hydroxylase (PAH) and to estimate the enzyme conversion we calculated the phenylalanine/tyrosine ratio as a proxy for PAH activity [23].

All multivariable analyses were performed using SIMCA software v.17.0 (Umetrics AB, Umeå, Sweden) and no samples were excluded in any of the analysis. 188 buckets within the chemical shift range of − 0.05 to 8.0 ppm, excluding the water peak at 4.5–5.0 ppm, were included in the multivariable analysis.

Principal component analysis (PCA) model was used to explore clustering patterns of observations and trends in the data in relation to known factors and outliers. Separation of classes and variables related to separation in the data according to classification of diagnosis of RA or not were evaluated using an Orthogonal Projections to Latent Structures with Discriminant Analysis (OPLS-DA) in fasting samples and at different time points (variable concentration at time X minus variable concentration at time 0). Cross-validation groups were set to 7 (default). The validity of OPLS-DA models was assessed using permutation tests (n = 999). Validated prediction models for performance are presented using Receiver Operating Characteristics (ROC) Curve for OPLS-DA models. Also, cross-validated predictive residuals (CV-ANOVA) visual comparison between scores and cross-validated scores, the cumulative amount of explained variation in the data summarized by the model (R2X[cum] and R2Y[cum]), and the predictive ability of the model (Q2[cum]) are presented. Class discriminating variables of interest from the OPLS-DA models were selected if the model had a significant (p < 0.05) CV-ANOVA and the permutation plot showed that the model had a sufficient quality.

Statistical analyses were performed using SPSS version 25 (SPSS Inc., Chicago, IL, USA). Mann–Whitney U-test was used since the subjects only were matched on a group level. Univariate tests were performed to compare baseline characteristics and metabolites between women with RA and without RA. Univariate tests were also performed for AUCmin and metabolites at different time points if they were found either to have a significant AUCmin or driving the separation in OPLS-DA models. In this explorative study, data are presented as median (inter quartile range (IQR)) with significance set at α = 0.05, i.e., not corrected for multi testing. In addition, to be able to explore if fasting metabolites were associated to certain health variables spearman rank correlation was performed for all participants for BMI, age, HbA1c, Hb, physical activity, and diet index and for metabolites and disease related variables (VAS, CRP, ESR, tender and swollen joints, and DAS-28) for women with RA. A sensitivity test adjusting for BMI and age was also performed for this analysis. For these additional correlation tests significance was set at α = 0.01.

The primary objective of the PIRA trial was to measure postprandial levels of IL-6. The sample size was calculated for alpha = 0.05 and 80% power, using data from previous studies [24, 25], which suggested a difference of 1.5 pg/mL in IL-6 could be expected with a standard deviation of 2.0. The study aimed at 30 patients with RA and 30 controls completing the trial. However, a sample size of about 20 participants/group has previously been found sufficient to compare metabolite patterns between patients with RA and controls in fasting samples [13, 17, 26], indicating that this is a sufficient group size.