Cohort description and study design

Eleven female breast cancer patients with invasive ductal carcinoma (gradings 2–3) were initially recruited into this pilot study at the General Hospital Vienna, Austria. All patients gave permission to sampling by signing a written informed consent, and the study was approved by the ethics committee of the Medical University of Vienna (1863/2017). Patient and tumour characteristics were assessed, including age, BMI, histological type and grading (Table 1). The proliferative index according to MIB-1 and the receptor status of all patients was assessed, as well as the RCB class and the occurrence of ductal carcinoma in situ (DCIS). All patients were non-diabetic. Patients P04 and P07 were excluded from the study due to incomplete data points and different type of chemotherapy, respectively.

Table 1 Clinical characteristics of the female breast cancer patients included in this pilot study. IDC invasive ductal carcinoma, MIB-1 Ki-67 labelling index using MIB-1 antibody, RCB residual cancer burden, DCIS ductal carcinoma in situ, ER oestrogen receptor, PR progesterone receptor, HER2 human epidermal growth factor receptor 2

The stratification to specific molecular subtypes determined the cytotoxic chemotherapy treatment schedule. First-line cytotoxic chemotherapy included mainly 5-fluorouracil, epirubicin and cyclophosphamide (FEC). This was individually replaced or complemented by docetaxel, abraxane, carboplatin and/or antibody therapy. The patients also received an anti-emetic drug combination consisting of dexamethasone, ondansetron (zofran) and emend. Except for patient P10, the patients were given G-CSF the day after chemotherapy.

For this study, the patients were monitored during the first cycle of chemotherapy. Samples were obtained at four time points, i.e. (1) before the start of the infusion (baseline, t0); (2) 2 h after start of infusion (t1); (3) after completion of infusion (t2); and (4) 6–12 days after infusion (recovery, t3). At these timepoints, matched blood plasma (EDTA) and eccrine sweat from the fingertips were collected for plasma multi-omic analysis and sweat metabotyping, respectively. Individual blood plasma samples were aliquoted for dedicated proteomic, metabolomic as well as fatty acid/oxylipin workflows. Additionally, clinical routine blood parameters were assessed before the first chemotherapy cycle (baseline, t0) and 7–10 days after infusion (recovery, t3).

Eccrine sweat collection

Sampling units (precision wipes, Kimtech Science, Kimberly-Clark Professional, USA, 0.5 in. diameter) were prewetted with aqueous solution (3 µL, LC–MS grade H2O) and stored in labelled Eppendorf tubes as previously described [35]. Sweat collection involved rinsing hands with warm tap water and subsequent drying with disposable paper towels. After a lag time of 1 min, the sampling unit was placed between thumb and index finger using a clean tweezer and gently held for 1 min. Then, the sampling unit was transferred back to the Eppendorf tube and stored at 4 °C until further processing.

Eccrine sweat metabolomicsSample processing

Aqueous solution (120 µL, VWR Chemicals, LC–MS grade) containing caffeine-(trimethyl-d9) and N-acetyl-tryptophan-d3 (each 1 pg∙µL−1) and formic acid (FA, 0.2%) was added to the sampling unit in the Eppendorf tube. Metabolites were extracted by pipetting up and down 15 times. The sampling unit was pelleted on the bottom of the tube, and the supernatant was transferred into HPLC vials equipped with a 200 µL V-shape glass insert (both Macherey–Nagel GmbH & Co.KG) to be analysed by LC–MS/MS.

Data acquisition

A Q Exactive HF (Thermo Fisher Scientific) mass spectrometer was coupled to a Vanquish UHPLC System (Thermo Fisher Scientific). A Kinetex XB-C18 column (100 Å, 2.6 µm, 100 × 2.1 mm, Phenomenex Inc.) was used for chromatographic separation. Mobile phase A consisted of water (0.2% FA) and mobile phase B of methanol (0.2% FA). Water, FA and methanol were purchased from VWR Chemicals (Vienna, AT in LC–MS grade). The following gradient was employed: 1–5% B in 0.3 min and then 5–40% B from 0.3–4.5 min, followed by a column washing phase of 1.4 min at 80% B and a re-equilibration phase of 1.6 min at 1% B resulting in a total runtime of 7.5 min. Flow rate was set to 500 µL min−1, the column temperature to 40 °C and the injection volume was 10 µL. All samples were analysed in technical duplicates. Electrospray ionization was performed in positive and negative ionization mode. MS scan range was m/z 100–1000, and the resolution was set to 60′000 (at m/z 200). The four most abundant ions of the full scan (Top 4) were selected for HCD fragmentation applying 30 eV collision energy. Fragments were analysed at a resolution of 15,000 (at m/z 200). Dynamic exclusion was set to 6 s. The instrument was controlled using Xcalibur software (Thermo Fisher Scientific).

Data analysis

Intensity extraction was performed as follows: MSConvert and ProteoWizard were used to convert raw files into mzML files. Those were loaded into MZmine (Version 3.4.27) [49] and processed using the modules ADAP chromatogram builder and feature resolver [50]. This was followed by 13C isotope filtering, retention time alignment, blank background subtraction and filtering of features that had invalid isotope patterns. Then, gap filling was performed, duplicates filtered and ion adduct identity determined. Feature information including m/z values, retention time (RT), ion identity and areas under the curve (AUCs) in the specific raw files were exported as a.csv file for further processing.

Feature annotation was performed as follows: An.mgf file including precursor and merged MS/MS information was exported from MZmine and loaded into SIRIUS (Version 5.8.0) [51]. The molecular formula identification was performed with the standard orbitrap instrument settings (mass accuracy 5 ppm). The CSI:FingerID module was used to compute MS/MS data into fragmentation trees which were used to determine candidates for possible molecular structures and database searches. The results for the top hits containing a COSMIC score (Confidence score) were exported into a.tsv file to later merge feature IDs, areas and structural information into one single file.

Using RStudio with R (Version 4.3.1), the data exported from MZMine and SIRIUS were loaded and processed to receive a clean data matrix for statistical evaluation. This process included the import of manually integrated internal standard areas (caffeine-(trimethyl-d9) for positive ionization mode and N-acetyl-tryptophan-d3 for negative ionization mode) using Skyline (Version 22.2.0.351) to receive uniform data which was used for the normalization of areas under the curve (nAUCs) accounting for instrument performance. After calculating the means of technical replicates, the intensities were merged with the feature identification data received from SIRIUS. The finished data frame, containing m/z values, RT, feature ID, annotation and nAUCs, was exported as a.csv file for statistical processing.

Using Perseus (Version 2.0.10.0), the main columns containing the feature nAUC for each sample were first categorically annotated with group information (time points, donors, RCB). All nAUC values were transformed by applying log2 and then adding a fixed value of + 20. Features with less than 50% valid values at every time point were removed. Then, imputation from the lower gaussian distribution was conducted according to standard procedures of down shift 1.8σ and width 0.3σ. Volcano plots and PSA plots were created after filtering for features that had at least 50% COSMIC score according to SIRIUS.

Raw files generated by the Q Exactive HF instrument were manually reviewed using Xcalibur Qual Browser (Version 4.0, Thermo Fisher Scientific).

Blood plasma collection

Venous blood samples were collected in K3 EDTA tubes and kept at room temperature for exactly 30 min before centrifugation at 4 °C and 2000 g for 15 min. Plasma was aliquoted into five Eppendorf tubes and stored at – 80 °C until further processing.

Plasma metabolomics

Metabolomics of patient plasma samples (10 µL) were assessed by a targeted assay. We used the MxP® Quant 500 Kit (Biocrates Life Sciences AG, Innsbruck, Austria, product number 21094.12), which enables the detection and (semi)quantification of up to 631 analytes, including acylcarnitines, an alkaloid, an amine oxide, amino acid related metabolites, bile acids, biogenic amines, carboxylic acids, ceramides, cholesteryl esters, cresol, diacylglycerols, dihydroceramides, fatty acids, glycerophospholipids, glycosylceramides, hormones, indole derivatives, nucleobase-related metabolites, sphingolipids, triacylglycerols, the sum of hexoses and one vitamin/cofactor. Of the initial 631 possible metabolites, we excluded those that were found below LOD in more than 50% of samples or that showed an unacceptable accuracy (< 80% or > 120%) based on internal standards or blank measurements. This reduced the number of metabolites to 487. Measurements were carried out using liquid chromatography tandem mass spectrometry (LC–MS/MS) and flow injection (FIA)-MS analyses on a Sciex 6500 + series mass spectrometer coupled to an ExionLC AD chromatography system (SCIEX, Framingham, MA, USA), utilizing the Analyst 1.7.1 software with hotfix 1 (also SCIEX) according to the kit procedure.

The LC–MS method in positive ion mode was performed as follows: The UHPLC autosampler and column oven were held at 10 °C and 50 °C, respectively. The injection volume was 5 µL. Eluent A was water (0.2% formic acid), and eluent B was acetonitrile (0.2% formic acid). A run time of 5.8 min was applied. Eluent B remained at 0% B for 0.25 min, followed by a linear gradient to 12% B at 1.5 min and 17.5% B at 2.7 min. The percentage of eluent B further linearly increased to 50% B at 4 min and 100% B at 4.5 min, where it remained until 5 min. Eluent B was reduced to 0% at 5.1 min and remained at 0% until 5.8 min. The flow rate was 0.8 mL/min until 4.5 min, then increased to 1 mL∙min−1 until 4.7 min and remained until 5.1 min. Finally, the flow rate was reduced again to 0.8 mL∙min−1 until 5.8 min. A scheduled MRM experiment was run in positive polarity. Detection window was 30 s, target scan time was 0.15 s and 3 ms pause between mass ranges. Q1 and Q3 were held at unit resolution. Source parameters were CUR 45, voltage 5.5 kV, temperature 500 °C, ion source gas at 60 for nebulizer and 70 for Turbo V heater (arbitrary units).

The LC–MS method in negative ion mode was performed as follows: The UHPLC autosampler and column oven were held at 10 °C and 50 °C, respectively. The injection volume was 15 µL. Eluent A was water (0.2% formic acid), and eluent B was acetonitrile (0.2% formic acid). A run time of 5.8 min was applied. Eluent B remained at 0% for 0.25 min, followed by a linear gradient to 25% B at 0.5 min and 50% B at 2.0 min. Eluent B further linearly increased to 75% B at 3 min and 100% B at 3.5 min, where it remained until 5 min. Eluent B was reduced to 0% at 5.1 min and remained at 0% until 5.8 min. The flow rate was 0.8 mL/min until 3.5 min, then increased to 1 mL/min at 4.7 min and remained until 5.1 min. Finally, the flow rate was reduced to 0.8 mL/min until 5.8 min. A scheduled MRM experiment was run in negative polarity. Detection window was 30 s, target scan time was 0.15 s and 3 ms pause between mass ranges. Q1 and Q3 were held at unit resolution. Source parameters were CUR 35, voltage –4.5 kV, temperature 650 °C, ion source gas at 40 for nebulizer and 40 Turbo V heater (arbitrary units).

FIA methods were performed as follows: The UHPLC autosampler was held at 10 °C. eluent was acetonitrile (0.2% formic acid). The injection volume was 20 µL. A run time of 3 min at 100% eluent B was applied. The flow rate was 0.03 mL/min from 0 to 1.6 min and then increased to 0.20 mL/min until 2.4 min and remained until 2.8 min. Then, the flow rate was reduced to 0.03 mL/min until 3 min. The MRM experiments was run in positive polarity with 3 ms pause between mass ranges. Q1 and Q3 were held at unit resolution. Source parameters were CUR 20–30, voltage 5.5 kV, temperature 200 °C, ion source gas at 30–40 for nebulizer, and 50–80 for Turbo V heater (arbitrary units).

The required standards, quality controls (QCs) and eluents were included in the kit, as well as the chromatographic column (product number 21117.1). Phenyl isothiocyanate (Sigma-Aldrich, St. Louis, USA) was purchased separately and was used for derivatization according to the kit manual. Preparation of the measurement worklist as well as data validation and evaluation were performed with the software supplied with the kit (MetIDQ-Oxygen-DB110-3005, Biocrates Life Sciences). Data was normalized according to median QC level 2.

Plasma oxylipin and fatty acid analysisSample preparation

Oxylipins and fatty acids were enriched from blood plasma similarly to a previous report [30]. Briefly, plasma (400 µL) was freshly thawed on ice and proteins were precipitated with cold ethanol (1.6 mL, abs. 99%, –20 °C; AustrAlco). Ethanol contained an internal standard mixture of 12S‐HETE‐d8, 15S‐HETE‐d8, 5‐Oxo‐ETE‐d7, 11,12‐DiHETrE‐d11, PGE2‐d4 and 20‐HETE‐d6 (each at 100 nm, Cayman Europe, Tallinn, Estonia, Supplementary Table S1). After centrifugation (30 min, 4536 g, 4 °C), supernatants were transferred into new Falcon™ tubes (15 mL), and ethanol was evaporated via vacuum centrifugation at 37 °C until the original sample volume (400 µL) was restored. Solid phase extraction was performed by loading samples with Pasteur pipettes onto pre-conditioned StrataX SPE columns (30 mg∙mL−1, Phenomenex, Torrance, CA, USA). After washing with water (5 mL, LC–MS grade, VWR) samples were eluted with ice-cold methanol (500 µL, VWR) containing 2% FA. Methanol was evaporated under a gentle stream of dinitrogen at room temperature. The dried samples were then reconstituted in 150 µL reconstitution buffer (v/v 65: 31.5: 3.5 water: acetonitrile: methanol + 0.2% FA) for data acquisition.

Data acquisition

Chromatographic separation was performed using a Thermo Scientific™ Vanquish™ (UHPLC) system equipped with a Kinetex® C18‐column (2.6 µm XB-C18 100 Å, 150 × 2.1 mm, Phenomenex Inc.). The mobile phase A consisted of water + 0.2% FA and mobile phase B consisted of acetonitrile/methanol (v/v 90:10) + 0.2% FA. A gradient flow profile was applied starting at 35% B and increasing to 90% B (1–10 min), further increasing to 99% B within 0.5 min and held for 5 min. Solvent B was then decreased to the initial level of 35% within 0.5 min, and the column was equilibrated for 4 min. The total run time was 20 min per sample. The flow rate was kept at 200 μL min−1 and the column oven temperature at 40 °C. The injection volume was 20 µL, and all samples were analysed in technical duplicates. Mass spectrometric analysis was accomplished with a Q Exactive™ HF orbitrap (Thermo Fisher Scientific, Austria), equipped with a HESI source in negative, as well as positive ionization mode. A spray voltage of ± 3.5 kV and a capillary temperature of 253 °C were applied. Auxiliary gas was set to 10 a.u. and sheath gas was set to 46 a.u. The MS scan range was set to m/z 250–700 with a resolution of 60′000 (at m/z 200) on the MS1 level. A Top 2 method was selected, as well as an HCD fragmentation with normalized collision energy of 24. An inclusion list covering 33 m/z-values specific for well-known eicosanoids and MS precursor molecules was imported (Supplementary Table S2). The resulting fragments were additionally analysed on the MS2 level at a resolution of 15′000 (at m/z 200).

Data analysis

For data analysis, analytes were compared with an in-house established database on the MS1 level based on exact mass and retention time (degree of identification shown in Supplementary Table S3) using the TraceFinder software (version 4.1). Subsequently, MS/MS fragmentation spectra were manually compared with reference spectra of in-house measured, commercially available standards or to reference spectra from the Lipid Maps depository library (July 2018). Relative quantification of the identified analytes was then performed on the MS1 level using the TraceFinder software (version 4.1). The resulting peak areas were loaded into an R software package environment (Version 4.2.0) and log2-transformed. The mean peak area of the internal standards was subtracted from the analyte peak areas to correct for variances arising from sample extraction and LC–MS/MS analysis. Then, each log2-transformed area was increased by adding (x + 20) enabling missing values imputation. Missing values were imputed using the minProb function of the imputeLCMD package (Version 2.1).

Plasma proteomicsSample processing

Plasma samples (EDTA-anticoagulant) were diluted 1:20 in lysis buffer (8 m urea, 50 mm triethylammonium bicarbonate (TEAB), 5% sodium dodecyl sulphate (SDS)) and heat denatured (95 °C, 5 min). The protein concentration was determined using a bicinchoninic acid (BCA) assay. A total of 20 µg protein per sample was digested using the ProtiFi S-trap approach. Solubilised protein is reduced with dithiothreitol (64 mm) and thiols carbamidomethylated using iodoacetamide (48 mm). After diluting with trapping buffer (90% v/v methanol, 0.1 M TEAB), the samples were loaded onto S-trap mini cartridges. Samples were extensively washed and then digested with trypsin/Lys-C (1:40) at 37 °C (2 h). Peptides were eluted, dried and stored at –20 °C.

Data acquisition

Plasma proteomic analysis was performed as described previously [30]. Dried peptide samples were dissolved with 5 µL FA (30%) containing synthetic peptide standards and additionally with 40 µL loading solvent (97.95% water, 2% acetonitrile, 0.05% trifluoroacetic acid) and transferred to HPLC vials equipped with a 200 µL V-shape glass insert (both Macherey–Nagel GmbH & Co.KG). A Dionex UltiMate3000 nanoLC-system (Thermo Fisher Scientific) was used for chromatographic separation. The injection volume was 1 µL, and the peptides were pre-concentrated on a pre-column (C18 Pepmap100, 2 cm × 75 µm, Thermo Fisher Scientific) at a flow rate of 10 µL∙min−1 using mobile phase A (99.9% H2O, 0.1% FA). Subsequent peptide separation was achieved on an analytical column (1.6 µm C18 Aurora Series emitter column, 25 cm × 75 µm, from IonOpticks) by applying a flow rate of 300 nL min−1 and using a gradient of 7–40% mobile phase B (79.9% acetonitrile, 20% water, 0.1% FA) in 43 min. The total LC run was 85 min including washing and equilibration steps. Mass spectra of peptides were acquired in a data-dependent analysis mode using a timsTOF Pro mass spectrometer (Bruker Daltonics) equipped with a captive spray ion source (1650 V). The instrument was operated in the Parallel Accumulation-Serial Fragmentation (PASEF) mode, and a moderate MS data reduction was applied. The scan range was set to m/z 100–1700 and 0.60–1.60 V s cm−2 with a ramp time of 100 ms. All experiments were performed with 10 PASEF MS/MS scans per cycle leading to a total cycle time of 1.16 s. The collision energy was ramped as a function of increasing ion mobility from 20 to 59 eV and the quadrupole isolation width were set to 2 Th for m/z < 700 and 3 Th for m/z > 700.

Data analysis

Protein abundance profiles were obtained by label-free quantification (LFQ) of the proteomics data. Using the publicly available software package MaxQuant (Version 1.6.17.0) together with the Andromeda search engine, raw data was searched against the SwissProt homo sapiens database (Version 141,219 containing 20′380 entries) including an allowed peptide tolerance of 20 ppm, a maximum of two missed cleavages, carbamidomethylation on cysteines as fixed modification as well as methionine oxidation and N-terminal protein acetylation as variable modifications. A minimum of one unique peptide per protein was set as search criterium for positive identifications. Additionally, the “match between runs” option was applied, using a 0.7 min match time window and a match ion mobility window of 0.05 as well as a 20 min alignment time window and an alignment ion mobility of 1. A false discovery rate (FDR) of ≤ 0.01 was set for all peptide and protein identifications. The identified proteins were filtered for reversed sequences as well as common contaminants and annotated according to the different study groups using Perseus software (Version 1.6.14.0). Then, LFQ intensity values were log2-transformed, and proteins were additionally filtered for their number of independent identifications. Protein had to be conditionally identified in 70% of at least one sample group. Finally, missing values were replaced from a normal distribution with width 0.3σ and down shift 1.8σ.

Statistical analysisEccrine metabolomics

The untargeted metabolomics from eccrine sweat included a COSMIC confidence score of ≥ 50% for feature identification [52]. One-way ANOVA statistics were applied to evaluate significant metabolite time courses with four time points and nine subjects per time point. The volcano plots to compare the eccrine sweat metabolome of specific time points (t0 vs t1) or of RCB comparisons (at t0 or t3) included two-sided t-tests. These P values were multiple testing corrected based on false discovery rate of 0.05, including 250 permutations and were calculated in Perseus software (Version 2.0.10.0).

Plasma metabolomics

One-way ANOVA statistics were applied to evaluate significant metabolite time-courses with four time points and nine subjects per time point. For plasma metabolomics, the volcano plots of RCB comparisons (at t0 or t3) were obtained by plotting Log2-fold changes and permutation-based (n = 250) multiple testing corrected P-values. A Log2-fold change cut-off of ± 1 and a –Log10 (adjusted P value) of 1.3 was chosen.

Plasma fatty acids/oxylipins

One-way ANOVA statistics were applied to evaluate significant metabolite time-courses with four time points and nine subjects per time point. RCB comparisons (at t0 or t3) of single fatty acids/oxylipins were analysed for significant changes between groups using Kolmogorov–Smirnov tests.

Plasma proteomics

A false discovery rate of 0.01 for identification was applied on protein and peptide levels. One-way ANOVA statistics were applied to evaluate significant metabolite time-courses with four time points and nine subjects per time point. Comparisons of time points were calculated by paired t-tests using multiple testing corrected P values, based on false discovery rate of 0.05, including 250 permutations. Those were performed in Perseus software (Version 2.0.10.0).

Clinical parameters

Single parameters were evaluated for significant changes by a paired t-test between baseline (Pre) and recovery (Post).