Chemicals

Analytical grade ethanol, analytical grade methanol and the LC-MS grade formic acid were acquired from VWR International (Fontenay-Sous-Bois, Paris, France). The LC-MS grade acetonitrile was acquired from Fisher Scientific (Loughborough, Leicestershire, United Kingdom). Deuterated standards, DL-phenyl-d₅-alanine and L-tryptophan-(indole-d₅) were acquired from Cambridge Isotope Laboratories Inc. (Andover, Massachusetts, United States of America). Ultra-pure type I water was generated using the Merck Millipore Synergy UV system.

Serum samples

The LNB patients’ serum samples for this study were collected as part of routine diagnostics and were included in our previous study²⁷, which compared the efficacy of intravenous ceftriaxone and oral doxycycline in the treatment of LNB. All individuals participating in the original study provided informed consent and ethical approval was granted by the National Committee on Medical Research Ethics in Finland. The diagnosis of LNB was confirmed according to the European Federation of Neurological Societies (EFNS) guidelines and criteria⁸. The present investigation adhered to the ethical principles outlined in the Declaration of Helsinki for medical research involving human material and data.

To maintain confidentiality, all samples were coded, ensuring that no personal identifiable information was handled, except for age and sex. The sample material comprised serum samples from 81 individual definite LNB patients. Among them, 68 patients had both acute pretreatment and post-treatment samples, with the post-treatment samples serving also as control samples for the pretreatment samples. Additionally, six patients with only pretreatment samples and seven with only post-treatment samples were included in the analysis. Consistent with standard clinical practice, the samples were stored at -20 °C, a common storage temperature for routine diagnostic samples, rather than -80 °C, which is primarily used when samples are collected for research purposes. The freezers and freezer rooms used for storage did not have a frost-free system. The sample collection and antibiotic treatments were conducted in Turku University Hospital, Turku, Finland (49 patients) and in Helsinki University Hospital, Helsinki, Finland (32 patients).

Sample preparation and analytical workflow

The sample preparation method was designed with a systematic approach to ensure the precise analysis of serum metabolites. By implementing controlled procedures, the method optimizes consistency and a reliable foundation for subsequent UHPLC-Orbitrap-MS/MS analysis, enhancing the precision of metabolomic profiling.

Random sampling was conducted, while ensuring that samples matching the same patient were analyzed during the same analysis time. Samples were left to melt slowly, while vortexing them at 750 rpm. After vortexing, 150 µL of serum was transferred to a new Eppendorf tube. Next, 50 µL of deuterated 7.125 µM L-tryptophan-(indole-d₅) aq. internal standard (IS) solution was added to the serum and vortexed for 15 min at 750 rpm. To precipitate macromolecules, mainly proteins, from the serum, 750 µL of cold methanol was added to the mixture and the solution was vortexed for 30 min at 750 rpm. Following the precipitation (Supplementary Information Fig. S1-S3 and Table S1), samples were centrifuged for 20 min at a relative centrifugal force (RCF) of 21,913 g. The supernatant (700 µL) was transferred to a new Eppendorf tube and dried under vacuum.

The dried samples were dissolved with 150 µL of deuterated 10 µM DL-phenyl-d₅-alanine aq. IS solution. The dissolved samples were vortexed for 15 min at 750 rpm before undergoing centrifuge filtration through 0.2 μm polytetrafluorethylene (PTFE) micro centrifugal filters for 20 min at an RCF of 9,056 g.

Following filtration, the samples were pipetted into a 700 µL 96-well plate and analyzed with UHPLC-Orbitrap-MS/MS preceding in silico metabolomics (Supplementary Information Table S2) and statistical analysis. Blank samples, prepared in the same manner as the serum samples, were used as controls to monitor the sample preparation process and assess matrix effects during the analysis.

UHPLC-Orbitrap-MS/MS analysis

The development of our UPHLC-Orbitrap-MS/MS method involved optimization of parameters, with a specific focus on the detection of low molecular weight serum metabolites. Employing positive ion mode, we fine-tuned the MS instrument to ensure accurate detection and quantification of metabolites. A mass range of 70−1,050 Da was found most suitable and was used for the metabolomic and statistical analysis.

Prior to the UHPLC-Orbitrap-MS/MS analysis, the order of serum samples was randomized, and the instrument underwent calibration using Thermo Scientific™ Pierce™ LTQ Velos ESI positive ion calibration solution. The serum samples were analyzed using an ultrahigh-resolution UHPLC-PDA-HESI-QOrbitrap-MS/MS platform, which comprised of an Acquity UPLC^® system equipped with a photodiode-array (PDA) detector (Waters Corporation, Milford, MA, USA). This system was coupled to a Q Exactive™ hybrid quadrupole-Orbitrap™ mass spectrometer (Thermo Fisher Scientific GmbH, Bremen, Germany) via heated electrospray ionization (HESI) source.

The reversed phase (RP) column used in the study was an Acquity UPLC^® BEH Phenyl 1.7 μm 2.1 × 100 mm column (Waters Corporation, Wexford, Ireland). Elution was carried out using acetonitrile (A) and 0.1% formic acid (B) as eluents at a flow rate of 0.5 mL × min^− 1. The elution profile was as follows: 0.0–0.5 min, 0.1% A in B (isocratic gradient); 0.5–7.5 min, 90% A in B (linear gradient); 7.5–8.5 min, 90% A in B (isocratic gradient, column wash); 8.5–8.6 min, 0.1% A in B (linear gradient); 8.6–10.1 min, 0.1% A in B (isocratic gradient, column stabilization). The injection volume was 5 µL with a full loop overfill factor of 3. The UV data, in the range of 190–500 nm, was collected with the PDA detector during the 10.1-minute analysis, while high-resolution MS data was recorded between 0.0 and 7.5 min. The UHPLC flow was diverted to waste during the column wash and stabilization periods.

MS data was recorded in positive ion mode within mass ranges of 70−1,050. Full MS scan resolution was set at 70,000, with Automatic Gain Control™ (AGC) target value at 3 × 10⁶ and a maximum injection time (IT) of 200 ms. Data-dependent MS/MS (dd-MS²) scans were collected at a resolution of 17,500, with AGC target value of 1 × 10⁵ and maximum IT of 50 ms. The dd-MS² scan utilized Top N technique, with maximum precursor multiplexing per scan (MSX) count of 1 and loop count of 5, resulting in a Top N value of 5, where five of the most intensive ions were selected for dd-MS² fragmentation. Additionally, a lock mass of m/z 214.08963 was utilized to enhance mass accuracy.

The settings for the HESI-source included a capillary temperature of 380˚C, a spray voltage of 3,800 V, a sheath gas (N₂) flow rate of 60 (arbitrary units), and an auxiliary gas (N₂) flow rate of 20 (arbitrary units). The S-lens average response factor (RF) level was set to 60 (arbitrary units), and the probe heater temperature was maintained at 300˚C. ISs, comprising both DL-phenyl-d₅-alanine and L-tryptophan-(indole-d₅), were analyzed in duplicate with every 10 injections.

In silico analysis of metabolomic data and molecular feature identification

Generated UHPLC-MS/MS data in RAW format was directly subjected to metabolomic analysis using Compound Discoverer 3 (Thermo Fisher Scientific Inc., Waltham, MA, USA; version 3.1.0.305). Analysis employed the built-in “Untargeted Metabolomics with Statistics Detect Unknowns with ID using Online Databases and mzLogic” workflow. Compound Discoverer 3 is software designed for LC-MS data processing, with automated identification, quantification, annotation, and interpretation of the results. Manual valuations and identifications were done using Xcalibur™ software version 4.1.31.9 (Thermo Fisher Scientific GmbH, Bremen, Germany). The metabolomic data contained metabolite-specific information such as the compound-specific exact mass, retention times, and corresponding integrated peak areas for all the detected MFs. The metabolomic quantification file was then transformed into CSV format for further statistical analysis. The parameters used in the metabolomic analysis and the details regarding the MS databases employed for both manual and in silico identification can be found in Supplementary Information Table S2. The final identified MFs are categorized according to their identification confidence levels (see Supplementary Information Table S3). These classifications align with the 5-level system proposed by Schymanski et al. (2014), which is rooted in the Metabolomics Standards Initiative (MSI) framework by Sumner et al. (2007)^28,29. The level 1 identification corresponds to validated identification (confirmed structure), level 2 to putative identification (MS/MS match to literature), level 3 to tentative structure (database/literature match to molecular formula), level 4 to matching molecular formula (including isotope distribution, charge state and adduct ion formation), and level 5 to a unique feature.

Statistical analysis

Differences between the pretreatment and post-treatment samples of all 26,978 MFs were summarized with descriptive statistics and studied with Wilcoxon signed-rank test, due to the non-normality of the distributions. The compounds which were statistically significant (level set at 0.001) in both datasets (raw and standardized) were examined more closely and visually with a volcano plot. For further analysis, 91 of these MFs were selected based on the largest difference between samples.

Linear mixed models for repeated measures were used for further analysis of all 91 MFs variables to evaluate the effects of patients’ characteristics. The models included one within-factor (timepoint: pretreatment, 12 month post-treatment), and several between factors (sex, age, antibiotic treatment, hospital, acetaminophen use and LNB symptom duration). The compound symmetry covariance structure was used for time. Logarithmic transformations were used to achieve the normal distribution assumption of MF variables.

The normality of variables was evaluated visually and tested with the Shapiro-Wilk test. Tests were performed as two-sided with a significance level set at 0.05. The analyses were carried out using RStudio (version 2023.03.0.386) based on R (version 4.3.0; RStudio, PBC, Boston, MA, USA).