The number and composition of the mass spectrum profile (MSP) library used reference library have a significant impact on the accuracy of the microflex Biotyper taxonomy assignment. To extend the coverage to the identification of members of non-pathogenic Clostridia, we used 142 sequenced strains. These strains are the main representatives of the human gut commensals of the class Clostridia (Table 1; Supplementary Table 1) and have been chosen to include as many geographically diverse strains as possible, to cover each species as broadly as possible. The 142 strains belonging to 21 genera and 47 species were confirmed for their species identification using full-length 16S rRNA gene sequencing. Fifty-eight (58) 16S rRNA gene full-length identified strains were used to validate the newly generated MALDI-TOF MS database. Strain information is summarized in Supplementary Table 1 (strains used for library construction), and Supplementary Table 2 (validation strains).

The culture medium used for each strain is given in Supplementary Table 1 and included the following: Brain Heart Infusion with Inulin (BHII Agar): BHI Agar (BD Difco, cat. number 279830, Franklin Lakes, NJ, USA) supplemented with inulin from chicory (Sigma, cat. number I2255, Darmstadt, Germany), 1 g/l. Schaedler Agar (Oxoid, cat. number CM0497, Ireland), and modified Gifu Anaerobic Broth (mGAM) (HiMedia, cat. number M2079). All media were prepared according to manufacturer instructions and supplemented with 15 g agar/l whenever appropriate. Bacteria were cultured under strictly anaerobic conditions using an anaerobic workstation (Anaerobic chamber; Coy Laboratories, Ann Arbor, MI, USA) containing a gas mixture of 10% CO₂, 5% H_2, and 85% N₂. Bacteria were incubated at 37°C for 3–4 days before MALDI-TOF MS analysis. Culture conditions are summarized in Supplementary Table 1 (strains used for library construction) and Supplementary Table 2 (validation strains).

Genomic DNA was extracted from bacterial cultures using a bead-beating device (SpeedMill BeadBeater for 30 s or with the vortex adapter on maximum speed for 10 min) followed by the Wizard genomic DNA purification kit (Promega, cat. number A2920, Dübendorf, Switzerland) according to manufacturer instructions. Total DNA was quantified by absorbance at 260 nm using the NanoDrop® ND-1000 Spectrophotometer (Witec AG, Littau, Switzerland). DNA samples were stored at-20°C until further analysis.

To confirm the identity of the isolates, the full region of the 16S rRNA gene was amplified by PCR using the FIREPol® DNA polymerase (Solis BioDyne, Tartu, Estonia) using primers 27F (5′-AGA GTT TGA TCC TGG CTC AG-3′) and 1492R (5′-GGT TAC CTT GTT ACG ACT T-3′) with the following conditions: 95°C for 3 min; 35 cycles of 95°C for 30 s, 50°C for 30 s, 72°C for 30 s; then 72°C for 7 min (Frank et al., 2008). PCR products were visualized using standard gel electrophoresis (1% agarose in 1× Tris-acetate-EDTA [TAE] buffer) at 100 V for 30 min (Bio-Rad, Cressier, Switzerland), purified using the Promega Wizard™ SV Gel, and PCR Cleanup System according to the manufacturer’s protocol and sequenced bi-directionally to obtain near full-length 16S rRNA gene sequence using Sanger sequencing at Microsynth (Balgach, Switzerland). To identify the closest homologs, the obtained sequences were aligned using the Basic Local Alignment Search Tool (BLAST) (Altschul et al., 1990). We defined sequence homology ≥99% as a species match, and sequence similarity ≥97% as a genus match (Becker et al., 2004). To build the phylogenetic tree, 16S rRNA gene sequences were aligned and trimmed using MUSCLE and a maximum-likelihood phylogenetic tree was constructed with 1,000 bootstrap replicates using MEGA 11 (Tamura et al., 2021).

As recommended by Bruker-Daltonics for new library creations, ethanol-formic acid extraction was used to prepare the sample for MSP creation. Formic acid is routinely used when using MALDI-TOF in bacteriology since acidic pH of the matrix improves the extraction of the ribosomal proteins (Croxatto et al., 2012). Briefly, two to three colonies were suspended in 300 μl of high-pressure liquid chromatography (HPLC)-grade water (Sigma-Aldrich, Buchs, Switzerland) and mixed with 900 μl of 100% ethanol (Sigma-Aldrich, Buchs, Switzerland). After centrifugation at 15,000× g for 2 min, the supernatant was removed, and the pellet was dried. For sample extraction, 50 μl of formic acid (70% in water) was added to the bacterial pellet, the tube was thoroughly mixed by vortexing and 50 μl acetonitrile (Sigma) were added to the mixture. The mixture was centrifuged at 15,000× g for 2 min. For each strain, 1 μl of the supernatant containing the bacterial extract was spotted on a 96-spot steel plate (Bruker-Daltonics) in 15 replicates and allowed to dry at room temperature. One microliter of bacterial test standard (BTS, Bruker-Daltonics) was pipetted on two MALDI target spots for each plate to allow for calibration during acquisition and processing. Subsequently, the samples were overlaid with 1 μl of α-cyano-4-hydroxycinnamic acid (HCCA) matrix and air dried again prior to measurement. A Bacterial Test Standard (BTS) was used to calibrate the instrument, before each acquisition session.

MALDI-TOF MS was performed with a Bruker-Daltonics microflex LT/LH bench-top mass spectrometer. Protein mass spectra of samples were acquired in a mass range of 2000–20,000 Da using the flexControl 3.4 program (Bruker-Daltonics) and a laser frequency at 20 Hz with a linear positive mode. The default operating conditions were as follows: ion source 1, 20 kV; ion source 2, 18.25 kV; pulse ion extraction, 370 ns. Each spectrum was summed up to 240 laser shots (40 laser shots/position x 6 different positions). For each strain, 30 single spectra were generated from two independent cultures with 15 technical replicates. All raw spectra have been deposited on Zenodo with the access number 10.5281/zenodo.7573939.

The quality of raw spectra was evaluated using flexAnalysis 3.4 software (Bruker-Daltonics). To ease the quality check of the raw spectra, smoothing parameters and baseline subtraction parameters were applied. For baseline subtraction we used the multipolygon, minimum, penalized least squares, and penalized B-Spline baseline subtraction method, and for smoothing, we used the Savitzky-Golay smoothing method. For each analysis day, the eight known masses of the bacterial test standard (BTS) spectrum were controlled to be within the tolerance range of +/− 300 ppm to prove mass accuracy. Spectra showing any flatline, outliers, dramatic mass shifts, and anomalies were deleted. Flatline spectra were defined as spectra with a completely flat line and outliners were defined as spectra with a unique peak that was detected in no other spectra of the same strain. Mass shifts were assessed on a random peak between 6,000 Da to 7,000 Da and we only accepted spectra with peak shifts below 500 ppm. After processing the raw spectra, a minimum of 27 high-quality spectra per strain were selected and merged using the MALDI Biotyper Compass Explorer software (Bruker-Daltonics) to create the mass spectral profiles (MSPs). The merged spectra per strain constituting the MSP were verified to have a log score greater than 2.7 and a peak frequency greater than 75%. The final library was compiled based on the MSP of 142 strains (Figure 1A). An MSP dendrogram was constructed using a Euclidean distance measure and an average linkage algorithm with the MALDI Biotyper Compass Explorer 4.1. The final library has been deposited on Zenodo with the access number 10.5281/zenodo.7573939.

To validate the CLOSTRI-TOF database, we performed a blind test of 58 full-length 16S rRNA gene sequenced isolates in two collaborating laboratories (Figure 1B; Supplementary Table 2). Validation was performed using the extended direct transfer (eDT) technique. Briefly, a few bacterial colonies were directly smeared onto a MALDI-TOF MS target and covered with a 1 μl of saturated HCCA matrix solution as is generally recommended for routine bacterial identification. In case no spectra were detected, we repeated the analysis using the ethanol-formic acid extraction method (as previously described). The raw spectra generated were compared to the two combined databases (Bruker-Daltonics MALDI Biotyper Compass MSP library v. 4.1.100 and CLOSTRI-TOF) (Figure 1B). The identification results were classified using the score values proposed by Bruker: the highest matching scores of the spectra were represented by ranges indicating high confidence identification (2.000–3.000), low confidence identification (1.700–1.999) and no organism identification possible (0.000–1.699).

To test the library on a set of previously unknown isolates, we used the CLOSTRI-TOF database combined with the original Bruker-Daltonics MALDI Biotyper Compass MSP library v. 4.1.100 to identify a total of 326 anaerobic cultures isolated from stool samples of healthy Swiss individuals recruited in the EnrichBut project (Swiss Ethics Number: BASEC ID 2021-00199) (Figure 1C). In brief, stool samples were collected from four self-declared healthy Swiss adults between 30 and 60 years of age who did not consume antibiotics two weeks prior to sample collection. Each stool donor was provided with a custom-made stool sampling kit that allows for the preservation of the oxygen sensitive bacterial groups under anaerobic conditions. Stool samples were stored anaerobically at room temperature prior to processing. Upon receipt (within 48 h of collection), the stool samples were resuspended (1:10 w/v) in anaerobic phosphate-buffered saline (PBS). The suspensions were then serially diluted in PBS and 100 μL was plated onto Columbia Blood, BHII, Schaedler and mGAM Agar. The plates were incubated at 37°C for 5 days. Between 20 and 30 colonies showing different colony morphologies were picked from each medium and re-streaked three times on fresh agar plates to ensure purity of the isolates. The isolates were cultured in liquid broth and mixed with 20% glycerol and stored at – 80°C.

The isolates were prepared for MALDI-TOF MS identification using the extended direct transfer (eDT) technique and were identified based on three technical replicates. The obtained spectra were compared either against the commercial Bruker-Daltonics library or the combined library from Bruker-Daltonics and CLOSTRI-TOF.

For the 142 strains analyzed, we generated 2,130 target positions and a total of 4,260 individual mass spectra. The quality of the spectra was high, with an average of 1–3 of the raw spectra removed per strain prior to MSP generation. Thus, we could generate a robust library for the 47 species.

To test the newly established library, we subjected 58 additional strains (validations strains, Supplementary Table S2), not included in the set of strains used for library construction but previously identified based on their 16S rRNA gene sequences to MALDI-TOF MS. Combining the original Bruker database with the CLOSTRI-TOF database plugin allowed reliable identification of at the species level (score value ≥2.0) of 57/58 strains (98.3%) in laboratory A and 54/58 (93.1%) in laboratory B (Figures 2). Of note, several strains belonging to the genus Blautia were identified with a high score (≥2.0) but were incorrectly assigned at the species level. Further, Coprococcus eutactus was correctly identified in both laboratories, but with a score between 1.7 and 2.0. All the validation strains (58/58) (100%) were reliably identified at the genus level (score ≥ 1.7 but <2.0) in both laboratory A and B. Thus, using the combination of the original Bruker database and the CLOSTRI-TOF database, we were able to reidentify most of the validation strains to species level and all strains to genus level.

To investigate the taxonomic assignment relationship between the microflex Biotyper and 16S rRNA gene-based sequencing, we compared dendrograms generated by either of these techniques. In most cases, the two methods provided identical results where strains belonging to the same species clustered together (Figures 3, 4). However, different strains of Agathobacter rectalis, Blautia spp., Faecalicatena fissicatena, and Lacrimispora celerecrescens formed two or three distinct MALDI-TOF-based clusters (Figure 3). Likewise, the maximum-likelihood phylogenetic tree based on 16S rRNA gene sequences was unable to cluster together strains of the same species of Agathobacter rectalis, Blautia spp., Faecalicatena fissicatena, and Lacrimispora celerecrescens (Figure 4).

We next analyzed a bank of 326 strains isolated from the feces of healthy Swiss volunteers (Supplementary Table 5) using the microflex Biotyper. Of the 326 analyzed strains, 170 (52.1%) were identified with the original Bruker database, while 156 (47.9%) could not be identified/had no match in the library at a score > 1.7. Combining the original library with the CLOSTRI-TOF library plugin, we increased the number of strains identified at a score > 2.0 to 264 (80.9%) (Figure 5), which corresponds to an identification of formerly unknown strains of 60%.