ChipFilter: Microfluidic-Based Comprehensive Sample Preparation Methodology for Microbial Consortia

Abstract

The metaproteomic approach is an attractive way to describe a microbiome at the functional level, allowing the identification and quantification of proteins across a broad dynamic range as well as the detection of post-translational modifications. However, it remains relatively underutilized, mainly due to technical challenges that should be addressed, including the complexity of extracting proteins from heterogeneous microbial communities. Here, we show that a ChipFilter microfluidic device coupled to a liquid chromatography tandem mass spectrometry (LC–MS/MS) setup can be successfully used for the identification of microbial proteins. Using cultures of Escherichia coli, Bacillus subtilis, and Saccharomyces cerevisiae, we have shown that it is possible to directly lyse the cells and digest the proteins in the ChipFilter to allow the identification of a higher number of proteins and peptides than that by standard protocols, even at low cell density. The peptides produced are overall longer after ChipFilter digestion but show no change in their degree of hydrophobicity. Analysis of a more complex mixture of 17 species from the gut microbiome showed that the ChipFilter preparation was able to identify and estimate the amounts of 16 of these species. These results show that ChipFilter can be used for the proteomic study of microbiomes, particularly in the case of a low volume or cell density. The mass spectrometry data have been deposited on the ProteomeXchange Consortium via the PRIDE partner repository with the data set identifier PXD039581.

Introduction

From the introduction of metaproteome and metaproteomics concepts,^1,2 studies have been done on various microbial communities, for example, intestinal microecology, marine biology, soil biology, aerosols composition, and studies of food composition to explain food quality, safety, and allergies.³ One of the major advantages of metaproteomics studies is that they provide functional information and reveal the microorganism functions and interactions at the protein level, complementary to metagenomic and metatranscriptomics data.^4,5 In addition to the data that can be deduced from genomic experiments, such as species identification and quantification, proteomic analyses can answer some other questions: enrichment of a sample in a certain type of post-translational modification, functional analysis of the proteins expressed, etc. Nevertheless, metaproteomics remains relatively underutilized, mainly due to the challenges that remain in extracting proteins from heterogeneous microbial communities.

Extracting proteins from different microbial communities includes various challenges without even mentioning sampling problems. While a universal extraction protocol providing good protein yields from a wide range of samples would be desirable, this objective does not seem achievable given the heterogeneity of matrices and microorganisms’ characteristics.⁶ Commonly, protein extraction includes a direct cellular lysis step, which is attained via chemical lysis using detergents and stabilizing agents; physical lysis (heat, pressure, or snap-freezing); or enzymatic lysis. The choice of the detergent is crucial and greatly impacts the quality of the results.⁷ Proteins can be purified using different methods, such as filter-based methods [filter-assisted sample preparation (FASP)],⁸ precipitation with acids like trichloroacetic acid (TCA),⁹ separation on a polyacrylamide gel using electrophoretic mobility,¹⁰ or solid-phase separation.¹¹ Each method has advantages, such as protein fractionation, recovery, and washing off the detergent. Then, proteins are in most cases identified by mass spectrometry (MS) using a bottom-up approach, in which proteolytic peptides are analyzed to generate protein inferences. The choice of the preanalytical strategy must consider the heterogeneity of microbial cells, having varied cellular membranes that cannot be lysed by a universal method and have species-specific challenges such as high nucleic acid contents for bacteria or a wall structure for fungi. Furthermore, combining several cell lysis procedures increases the risk of losing low-abundance proteins, experimental time, and handling steps. Other challenges include automation, repeatability between biological replicates, and interference of detergents with subsequent purification and analysis techniques. Therefore, developing a new strategy or revising existing methods is necessary for better metaproteomics sample preparation.

Microfluidics offers multiple advantages in the sample preparation of microorganisms for proteomic analysis, including automation, low-volume sample handling, safety, and fast processing. Microfluidic technology has been successfully applied for the separation of bacterial and viral particles from bioaerosols,¹² physical cell lysis,¹³ and chemical sample processing by utilizing immobilized trypsin.¹⁴ In a previous work,¹⁵ we proposed a ChipFilter Proteolysis (CFP) microfluidic device as a reactor for the miniaturization of protein sample processing and digestion steps. The CFP design is closely related to the experimental setup of FASP. The microchip has two reaction chambers of 0.6 μL volume separated by a protein filtration membrane made using regenerated cellulose to concentrate or retain large polypeptides while releasing small molecules of less than 10 kDa. The yeast protein extract and whole human cell proteome were successfully analyzed using CFP.

This study aims to assess a CFP-based workflow for sample preparation for liquid chromatography tandem mass spectrometry (LC–MS/MS) analysis in the context of microbiology and metaproteomics. The workflow described for CFP¹⁵ was modified to introduce microbial cells directly into the device to perform all steps necessary for sample preparation starting from microbial cell lysis to proteolysis. On a mix of three microorganisms, CFP offers performance advantages compared to other methods including modified FASP (mFASP), in-gel, and in-solution proteolysis. More proteins and peptides are identified with CFP than with compared protocols, even at low cell density. The nature of the generated peptides was studied to better understand the influence of the microfluidic system in tryptic digestion. Finally, CFP was utilized to prepare a sample mixture of 17 complex microbial species, leading to the identification of proteins from 16 species.

Materials and Methods

Materials

Octyl-β-d-glucopyranoside (ODG), the protease inhibitor, dithiothreitol (DTT), ammonium bicarbonate (ABC), and iodoacetamide (IAM) were purchased from Sigma-Aldrich. Trifluoroacetic acid (TFA) was purchased from Thermo Fisher Scientific, and acetonitrile (ACN) was purchased from Fisher Scientific.

Trypsin Gold, mass spectrometry grade, was provided by Promega, and lysozyme (50 ng/mL) was provided by Thermo Fisher Scientific.

Luria–Bertani (LB) agar (Merck) plates were independently inoculated with Saccharomyces cerevisiae, Escherichia coli, and Bacillus subtilis and incubated at 37 °C overnight. An inoculum into LB broth was made, and cells were cultivated until the optical density reached 1 at 600 nm. Before harvesting, cells were counted using a glass slide under a microscope. The cells were collected by centrifugation at 2400g for 5 min at room temperature. A single wash with phosphate buffer saline was performed before pelleting and storing at −80 °C until further use. The three cell types were considered in different cell densities for the experiments. Two different dilutions of the cells were taken as a mix of 1:1:1 cell number ratio, 10E6 of each cell type for Mix 1 and 10E2 of each cell type for Mix 2.

A standard whole-cell mixture consisting of 21 representative strains from 17 species of the gut microbiota (Zymo Research, ref D6331) was divided into 10 aliquots and stored in the storage solution provided by the manufacturer at −80 °C. This standard contains 18 bacterial strains including 5 strains of E. coli (JM109, B-3008, B-2207, B-766, and B-1109), 2 fungal strains, and 1 archaeal strain in staggered abundances, theoretically ranging from 20.01 to 0.0009% considering the cell number.

Murine feces were collected from healthy female murine housed in the Brain Plasticity Laboratory, ESPCI-PSL, Paris. After collection, samples were stored at −80 °C until use.

ChipFilter Method

The design and fabrication methodology of the microfluidic device has been explained previously.¹⁵ For the comprehensive sample preparation, cells suspended in ABC were directly loaded into the device in a total volume of 30 μL. Cells were introduced into the ChipFilter using a piston syringe (Agilent) and a syringe pump (Harvard Apparatus) maintaining a flow rate of 0.01 mL/min. The sequential injection of lysis buffer 1 [1% (w/v) of ODG and the protease inhibitor in 150 mM Tris–HCl pH = 8.8], 20 mM DTT in lysis buffer 1, 50 mM IAM in lysis buffer 1, and 50 mM ABC buffer was achieved using a flow-EZ pressure module, a flow controller, M-switch (Fluigent), and the software Microfluidic Automation Tool (Fluigent). The flow rate and volume were maintained in two stages at 2 μL/min for 45 μL and 1 μL/min for 30 μL with the upper pressure limit at 900 mbar. Finally, proteolysis was performed by introducing 20 μL of trypsin (a final concentration of 0.1 μg/μL in 50 mM ABC) at room temperature. A constant flow of 50 mM ABC was maintained for 150 min to ensure the mixing of the proteins with trypsin. The resulting proteolytic peptides were directly transferred in the flowthrough to the sample loop of the LC instrument. The elution volume was regulated to capacitate the sample loop volume. Proteolytic peptides were finally concentrated in a trapping column (C18 Pepmap, 300 μm i.d. × 5 mm length, Thermo Fisher Scientific).

For the standard gut microbiota, four replicates of 75 μL of the mixture corresponding to approximately 3.94 × 10E8 cells were thawed in ice. For cell lysis, lysis buffer 2, which has the same composition as lysis buffer 1, with supplemented lysozyme (final concentration of 0.5 mg/mL) was used. All of the subsequent steps were done as described above for mixed cells.

Microbial cells were extracted from the feces sample using a low-speed centrifugation method.¹⁶ Briefly, 25 mg of the fecal sample was suspended in 500 μL of PBS pH 7.4 with 5% protease inhibitor cocktail. Centrifugation at 250g for 5 min at 4 °C was done, and the supernatant was collected in a new tube. The pellet was resuspended, and the same steps were repeated four times. Finally, the collected supernatant was centrifuged at 21,000g for 30 min at 4 °C. The pellet was resuspended in 200 μL of lysis buffer 2 and subjected to mechanical agitation with glass beads. A final volume of approximately 350 μL was obtained; 50 μL of this lysate was loaded into the ChipFilter for proteolysis as described above.

Modified Filter-Aided Sample Preparation

This protocol was modified from the original FASP protocol⁸ to compare the two approaches (FASP and ChipFilter) that use very similar designs under identical chemical conditions. Accordingly, the samples were resuspended in lysis buffer 1 (sample/lysis buffer 1 ratio of 1:5 v/v) and incubated at 37 °C for 15 min. The contents are then transferred to Microcon-10 kDa centrifugal filter units (Merck) and centrifuged at 15,000g for 30 min to remove the flowthrough. Reduction was done with 20 mM DTT at 37 °C for 2 h, and then, the tubes were centrifuged at 15,000g for 30 min with the flowthrough removed. Alkylation was performed with 50 mM IAM for 2 h in the dark at 37 °C, and the tubes were centrifuged at 15,000g for 30 min to remove the flowthrough. Two washes with 500 μL of 50 mM ABC buffer were performed to remove the reagents. Digestion was performed with 2 μg of trypsin in 50 mM ABC buffer overnight at 37 °C. Elution of peptides was done by centrifugation twice at 15,000g for 20 min with 50 mM ABC to ensure maximum peptide recovery. Finally, the peptides were acidified with 0.1% (v/v) TFA. Desalting was performed with C18 Zip Tips (Merck) as per the manufacturer guidelines. Peptides were dried and stored at −20 °C until injection into the LC instrument.

In-Gel Trypsin Proteolysis

The samples were resuspended in 1× Laemmli buffer with 5% β-mercaptoethanol (Sigma-Aldrich) and heated for 5 min at 95 °C. 40 μL of the lysate was loaded into 12% precast gels (Bio-Rad) in Tris–glycine–sodium dodecyl sulfate (SDS) running buffer at a voltage of 80 V for 15 min. A short SDS-PAGE migration was used to restrict the proteome to a short band before separation.¹⁷ The gel was stained with Coomassie brilliant blue R-250, and the protein band was sliced and washed with excess double distilled water under agitation (700 rpm) at room temperature. The gel was dehydrated with 100% ACN and dried by using a Speedvac vacuum concentrator. The reduction was performed with 10 mM DTT in 50 mM ABC buffer at 56 °C with agitation for 30 min. The excess solution was removed, and the gel was dehydrated again with 100% ACN. The alkylation was performed next with 55 mM IAA in a 50 mM ABC buffer with agitation at 37 °C for 20 min in the dark. The excess solution was removed and replaced with a 50 mM ABC buffer. Dehydration was performed with 50 and 100% ACN, and the product was completely dried in the Speedvac vacuum concentrator. 200 μL of trypsin (12.5 ng/μL) was prepared with 50 mM ABC buffer, and rehydration of the gel was done for 45 min at 4 °C. Next, the excess solution was removed, and digestion was performed overnight at 37 °C with agitation. Peptides were extracted from the gel in two steps by acidification and dehydration of the gel, using 100 μL of 1% TFA (v/v) and then 5% of formic acid (v/v) separately for 15 min under agitation at 37 °C. The combined solutions were then transferred to a new tube. Again, the gel was dehydrated with 100% ACN for 15 min under agitation at 37 °C, and the solution was transferred to the elution tube. The eluate was dried in a Speedvac vacuum concentrator (Thermo Fisher), resuspended in 20 μL of 1% (v/v) formic acid, and sonicated for 5 min at 37 °C. Desalting was performed with C18 zip tips as per manufacturer guidelines. Peptides were then dried and stored at −20 °C until injection into the LC instrument.

In-Solution Proteolysis

The samples were resuspended in lysis buffer 3 [6 M urea, 150 mM Tris, 1× protease inhibitors, and 1% (w/v) ODG] with a sample/lysis buffer 3 ratio of 1:5 (v/v). Samples were subjected to mechanical disruption with glass beads of size 425–600 μm (Sigma-Aldrich) in a ratio of 1:1 (v/v). Five cycles consisting of a 30 s vortex and 60 s in ice were repeated as bead-beating pretreatment. The lysate was carefully removed from the beads and transferred to a new tube, where the reduction was performed with 20 mM DTT at 37 °C for 2 h. In the same tube, alkylation was performed with 100 mM IAA at 37 °C for 2 h in the dark. Precipitation of the proteins was performed with TCA solution (final concentration 10% v/v) for 30 min at 4 °C, followed by centrifugation at 17,500g at 4 °C for 1 h. The protein pellet was washed with 80% and then 100% (v/v) ice-cold acetone, followed by drying using a Speedvac vacuum concentrator. The protein pellet was solubilized in a 50 mM ABC buffer, and digestion was performed with 2 μg of trypsin overnight at 37 °C. Acidification of the peptides was performed with 0.1% (v/v) TFA. Desalting was performed with C18 zip tips as per manufacturer guidelines. Peptides were dried after desalting and stored at −20 °C until injection into the LC–MS setup.

Liquid Chromatography and Mass Spectrometry

Samples were analyzed by nanoLC–MS/MS in the data-dependent acquisition (DDA) high-energy c-trap dissociation (HCD) mode using an RSLCnano UltiMate 3000 System coupled to a nanoESI Q-Exactive or Q-Exactive HF mass spectrometer (Thermo Fisher Scientific).

For the ChipFilter method, the peptides recovered in the trap column were separated on a capillary reverse-phase C18 column Pepmap 75 μm i.d. × 50 cm length (Thermo Fisher Scientific) at 45 °C with a linear 120 min gradient elution from 2.5 to 60% of buffer B [water/ACN/formic acid 10%:90%:0.1% (v/v/v)] in buffer A [water/ACN/formic acid 98%: 2%: 0.1% (v/v/v)] at a fixed flow rate of 220 nL/min.

For mFASP, in-gel, and in-solution methods, the dried peptides were resuspended in 7 μL of 0.1% TFA solution (v/v), and 6 μL was used for single-shot injection. Trapping was done with a C18 Pepmap 300 μm i.d. × 5 mm length column (Thermo Fisher Scientific) and analyzed in a nano-LC MS/MS setup with a 120 min gradient as described earlier.

Mix-1 and Mix-2 sample analyses were performed with a Q-Exactive mass spectrometer operated in the nanoESI mode at 1.7 kV. Full MS survey scans were recorded over the m/z range of 400–2000 with a resolution of 70,000 using an automatic gain control target value (AGC) of 3E6 with a maximum injection time of 100 ms. Up to 15 intense 2⁺–5⁺ charged ions were selected for HCD with a normalized collision energy of 30, with a precursor isolation window at 2 m/z and a resolution set at 17,500 with an AGC value at 1E5 with a maximum injection time of 120 ms. The minimum MS² target value was set at 1E3, and the dynamic exclusion was for 20 s.

The standard gut microbiota samples were analyzed with a Q-Exactive HF mass spectrometer operated in the nanoESI mode (1.6 kV). Full MS survey scans were recorded over the m/z range of 375–1500 with a resolution of 60,000 using an automatic gain control target value (AGC) of 3E6 with a maximum injection time of 60 ms. Up to 20 intense 2⁺–5⁺ charged ions were selected for HCD with a normalized collision energy of 28%, with a precursor isolation window at 2 m/z, a resolution of 15,000, and an AGC value of 1E5 with a maximum injection time of 60 ms. The minimum MS² target value was set at 1E3, and the dynamic exclusion was for 20 s.

Data Analysis

Spectra were processed using Proteome Discoverer v2.4 (Thermo Fisher Scientific). For mixed cells, the Mascot search engine (Matrix Science Mascot 2.2.04) was used against the all-taxonomy SwissProt database (release 2022_03:568,002 sequences; 205,171,419 residues). For the standard gut microbiota, the Mascot search engine was used against a dedicated sequence database (20,303 sequences containing 11,357,410 residues) specifically restricted to the species in the standard gut mix from UniProtKB. A list of 17 species from the standard gut mix includes Faecalibacterium prausnitzii, Roseburia hominis, Bifidobacterium adolescentis, Lactobacillus fermentum, Clostridioides difficile, Methanobrevibacter smithii, Enterococcus faecalis, Clostridium perfringens, Veillonella rogosae, Bacteroides fragilis, Prevotella corporis, Fusobacterium nucleatum, Akkermansia muciniphila, Salmonella enterica, E. coli, Candida albicans,and S. cerevisiae along with Homo sapiens (1,133,353 sequences containing 506,763,197 residues). For the feces samples, a database containing the 50 most prominent bacterial species and their closely related genus protein sequence available in UniProt was built, along with the Mus musculus proteins (SwissProt) and some contaminant proteins. This database included 20,606,901 sequences with 8,689,718 023 residues.

The database search was performed with the following parameters: MS and MS/MS mass tolerance of 10 ppm and 0.02 Da, respectively, trypsin specificity with up to 2 missed cleavages, and partial carbamidomethylation (C), deamidation (NQ), and oxidation (M). Proteins with at least one high-confidence peptide and six amino acids were validated. The target FDR was set at 0.01. To perform label-free quantification in PD, the Minora feature detection tool was used in the processing workflow, with the precursor ions quantifier in the consensus workflow with consideration for both unique and razor peptides as the protein inferring strategy. The abundance value obtained was used to make the scatter plot and determine the correlation using tools in R (v 4.0.3).

The Kyte and Doolittle hydrophobicity index was calculated from a published package.¹⁸

All of the experiments were performed in four replicates for mixed cells and the standard gut mix. Statistical tests to identify significance (unpaired t-test) were performed, and values shown always represent mean ± standard error of the mean. The mass spectrometry data have been deposed on the ProteomeXchange Consortium via the PRIDE partner repository¹⁹ with the data set identifier PXD039581.

Results

Microbial Cell Lysis Can Be Performed in ChipFilter without Pretreatment

In a previous work,¹⁵ we have shown that it is possible to perform proteolysis of eukaryotic cell samples in the ChipFilter. To study microorganisms, it would be advantageous to perform the cell lysis directly in the ChipFilter too. It will help avoid contamination issues and the loss of material. To determine whether it was possible to lyse microbial cells on the chip, 1E6 cells of E. coli, S. cerevisiae, and B. subtilis were introduced separately into the ChipFilter. Lysis was achieved by using a lysis buffer containing 1% (w/v) ODG.

After CFP, peptides were eluted and identified by mass spectrometry (Figure 1) The number of identified proteins ranked from 824 ± 50 (B. subtilis) and 1121 ± 58 (E. coli) to 1339 ± 44 (S. cerevisiae), suggesting that the ChipFilter cell lysis and proteolysis were efficient.

Figure 1.

Whole-cell lysis and bottom-up proteomic sample preparation with ChipFilter. Number of proteins (A) and peptides (B) identified for Escherichia coli, Bacillus subtilis, and Saccharomyces cerevisiae using ChipFilter.

ChipFilter Method Identifies More Proteins and Peptides than Other Methods

To assess the efficiency of ChipFilter for the identification of proteins from different microorganisms, a mixture of 1E6 S. cerevisiae, E. coli, and B. subtilis (Mix 1) was analyzed with CFP and standard proteolysis protocols. The ChipFilter is a miniature of the FASP protocol, for which the fluid transfer is controlled by the laminar flow in the channels under applied pressure; it is important to note that FASP tubes are designed to accommodate SDS-based lysis buffers, but in the present study, we used ODG-based lysis buffers for FASP, called modified FASP (mFASP). This buffer and the lack of pretreatment were chosen to keep it similar to the CFP. Next, an in-gel workflow is considered as it is the most used workflow for metaproteomic sample preparation that allows the usage of SDS-based lysis buffers. Finally, the in-solution method, which introduces a cell lysis pretreatment by bead beating, was included to understand the influence of pretreatment.

The time required for seven steps by each method considered has been considered (Figure S1). The seven steps considered include cell lysis, protein purification, handling step, reduction, alkylation, proteolysis, and peptide cleanup. The digestion time for the ChipFilter is only 2 h as compared to overnight for other methods, as used in most protocols. A shorter digestion time in the ChipFilter method is necessary to avoid the loss of peptides over time that exit the chamber upon digestion.

The performance of the protocols was evaluated by considering the number of proteins and peptides identified from the 3 target species. For the concentration of 3E6 cells, the ChipFilter method achieved a mean protein identification of 1999 ± 54 proteins, whereas mFASP reached 1818 ± 43 proteins, in-gel 1504 ± 43 proteins, and in-solution 1665 ± 30 proteins (Figure 2A). Similarly, at the peptide level, a significantly higher identification number is obtained for the ChipFilter method than for other methods, as shown in Figure 2B. 9770 ± 108 peptides were identified using the ChipFilter method, 8197 ± 435 peptides were identified with mFASP, 8218 ± 262 peptides were identified with in-gel, and 7805 ± 178 peptides were identified with in-solution. Efficient catalysis in a confined space and a reduced loss of materials as compared to other methods can explain the higher identification for the ChipFilter method.

Figure 2.

Performance assessment of the ChipFilter workflow compared to other methods for Mix 1 (3E6 cells/sample). Identification of target proteins (A) and total peptides (B) was carried out by different methods for four replicates. Pearson coefficient for the four protocols (C). The distribution of proteins (D) and peptides (E) was identified at least once by different methods. Number and percentage of proteins identified by each method for E. coli, S. cerevisiae, and B. subtilis (F).

To evaluate the repeatability of the analysis protocols, the Pearson correlation coefficient was calculated between replicates. It varies from 0.771 to 0.840 for the ChipFilter method, 0.712–0.916 for mFASP, 0.859–8.876 for the in-gel protocol, and 0.844–0.867 for the in-solution protocol (Figure 2C).

The distribution of the proteins and peptides among the methods shows that the ChipFilter protocol identified the most distinctive proteins and peptides. At the protein level (Figure 2D), 17.5% (598 proteins) of the total proteins were identified exclusively by the ChipFilter method, whereas the other methods had less than 5.1% of specific proteins. The protein identification common between all of the methods was about 38.5% (1319 proteins). At the peptide level (Figure 2E), 32% (8395 peptides) of the total peptides were identified only in the ChipFilter method.

Interestingly, the peptide common to all of the methods is only 8.8% (2309 peptides) of the total population, indicating that each method characteristically generates different peptide fractions. The ChipFilter allows for the identification of 54.3% of the total peptides. A comparison of the protein origin from three microbial cells (Figure 2F) indicates that the ChipFilter identifies the highest number of Gram-positive bacteria B. subtilis proteins (230 proteins) and Gram-negative bacteria E. coli (738 proteins) in comparison to other methods, whereas the highest number of fungi S. cerevisiae proteins was recorded for the in-solution (1141 proteins) method. Increased fungal proteins in the in-solution method can be reasoned due to the introduction of a pretreatment step, which was lacking in other methods. The in-gel method has poor identification of Gram-positive bacteria. The usage of SDS-containing Laemmli buffer can be the reason as the peptidoglycan cell wall of Gram-positive bacteria was not efficiently denatured by SDS.²⁰ The ChipFilter method can identify all three cell types considerably better than other methods without pretreatment.

To investigate whether the data obtained with the ChipFilter can be used quantitatively, label-free quantification was carried out using PD at the peptide and protein levels. The number of peptides that could be quantified is slightly higher for filter-based methods (ChipFilter and mFASP), and even repeatability is less good for these methods (Figure 3A). The coefficient of variation was calculated at the peptide (Figure 3B) and protein levels (Figure 3C). The distribution of this coefficient of variation shows that the data obtained with the ChipFilter are between those obtained with mFASP and in-gel digestion, with the data obtained after in-solution digestion appearing to be the most repeatable.

Figure 3.

Label-free quantification of the peptides and proteins. Total number of peptides with the abundance value for each method according to the number of identifications in the four replicates (A). Distribution of the coefficient of variation at the peptide (B) and protein (C) levels.

ChipFilter Method Performs Efficiently with a Low Cell Number

The sensitivity of the ChipFilter method was tested for low cell number samples (mix 2:3 × 1E2 cells) to assess the identification and distribution of peptides and proteins like in mix 1. The mean number of proteins and total peptides identified are 367 ± 38 proteins and 1162 ± 126 peptides, respectively (Figure 4A,B). The number of protein identifications is similar for the ChipFilter method as compared to that for the in-solution method (281 ± 42 proteins) and greater compared to that for other methods, and the number of identified peptides is higher than that for every other method.

Figure 4.

Performance assessment of the ChipFilter workflow compared to other methods for Mix 2 (3E2 cells/sample). Identification of target proteins (A) and peptides (B) by different methods was carried out for four replicates. Pearson coefficient for the four protocols (C). The distribution of proteins (D) and peptides (E) was identified at least once across different methods. Number and percentage of proteins identified by each method for E. coli, S. cerevisiae, and B. subtilis (F).

However, the Pearson coefficient varies between 0.580 and 0.679 for the ChipFilter method, 0.746–0.889 for mFASP, 0.747–0.805 for the gel protocol, and 0.–764–0.834 for the solution (Figure 4C). The lower correlation can be due to the decreased intensities and signal-to-noise obtained for low cell numbers. Repeatability is often difficult to achieve in metaproteomics samples but is critical. For the current study, the variations arising during nanoLC–MS/MS are not considered and are attributed to the sample preparation methodology. The correlation coefficient values for the ChipFilter method indicate a high positive correlation between the replicates for high cell numbers (mix 1, Figure 2C). Even though the peptide identifications are significantly high for the ChipFilter method in Mix 2, the correlation between the replicates makes this method useable. The reduction in the correlation between the two cell densities can be due to the reduced intensities obtained at lower cell numbers. The in-solution and in-gel methods exhibit lower standard deviations than the other methods. mFASP generates high variability like the ChipFilter method, and it can be possible that the filter-based methods could contribute to these differences. In conclusion, the ChipFilter method generates replicates that have a positive correlation in Mix 1 and 2, indicating that reproducibility between the different cell numbers is related to each other in the same manner.

As observed at the 3E6 cell density, the coverage of identified proteins is variable depending on the protocol used. For Mix 2, the ChipFilter method identified the highest proportion of proteins (72% of the total) and the highest proportion of exclusive proteins (20.6% of the total target protein identification; Figure 4D). The same trends are obtained at the peptide level. The ChipFilter method identifies the majority of peptides (69.7%) and is the protocol that allows the identification of the largest number of exclusive peptides (45.1% of the total, Figure 4E). The distribution of the identified proteins according to the species is more balanced with the ChipFilter protocol, in particular for Bacillus proteins, which represent 9% of the total (9 proteins) against 2% for the other protocols (less than 4 proteins, Figure 4F). These results demonstrate that the ChipFilter method is an efficient sample preparation method for simultaneously identifying proteins in a single analysis even at low cell numbers.

Physio-Chemical Characteristics of the Peptides Generated by the ChipFilter Method

As the ChipFilter method allows the identification of a large number of specific peptides (Figures 2D and 4D), these peptides were studied to see if they have different physio-chemical characteristics from the peptides obtained by other methods, which could lead to a bias or an advantage in the exploitation of the results. This comparison was done for 3E6 and 3E2 cell samples.

The percentage of missed cleavage was examined for Mix 1 and Mix 2 samples (Figures 5A and S2A). The ChipFilter method has the highest percentages of 1 and 2 missed cleavages with Mix 1 (28 and 8%, respectively) and Mix 2 (23 and 6%, respectively). The percentage is also considerably high in the mFASP method for Mix 1 and Mix 2 (14 and 2%, respectively). The in-solution method is consistent between the two cell numbers with around 10% for 1 missed cleavage and 1% for 2 missed cleavages. The in-gel method has the least. The higher proportion of missed cleavage with ChipFilter can be explained by factors such as a higher density of proteins in a confined volume or a shorter proteolysis duration.

Figure 5.

Physical–chemical characteristics of the peptides generated by different preparation methods for the 3× 1E6 cell sample. The number of missed cleavages generated by each method for all the peptides (A) and for peptides exclusively identified by one protocol (B). Amino acid length distribution. (C) Positive ion mass (MH+) distribution. Kyte–Doolittle hydrophobicity index (E). For (C–E), the mean value is plotted in the graph along with the error bars indicating the standard deviation between four replicates.

As a consequence, the peptides identified by the ChipFilter method are generally longer. Using the ChipFilter method, 42% of Mix 1 peptides have a length of up to 20 amino acids, instead of approximately 80% for other methods (Figure 5B). Similarly, the proportion of Mix 2 peptides shorter than 20 amino acids was 66% with the CFP but 91% for the other methods (Figure S2B). For the ChipFilter method with the Mix 1 sample, 32% (3137 ± 149) of the peptides have a mass below 2000 Da, whereas for other methods, it is approximately 73% (5879 ± 232 peptides) (Figure 5C). For the Mix 2 sample, the percentage of peptides with a mass below 2000 Da was 56% in the ChipFilter (Figure S2C). It is to be noted that the highest mass of the peptides generated was 4988 Da for a ChipFilter cutoff of 10,000 Da. Finally, the hydrophobicity of the peptides generated was studied using the Kyte–Doolittle hydrophobicity index²⁰ (Figure 5D). Most of the peptides were accumulated between −1 and +1, which is indicative of a neutral mixture with slightly hydrophobic or slightly hydrophilic peptides. This trend was observed for all of the methods, including the ChipFilter method for both cell mixtures.

ChipFilter Allows Efficient Proteomic Analysis of a Standard Gut Microbiome

To assess the efficiency of ChipFilter proteolysis on a more complex sample, a commercial mixture of 17 species, including bacteria (Gram-negative and Gram-positive), fungi, and archaea, was studied after in-solution and ChipFilter proteolysis. This gut standard mixture is nonuniform in cell number and contains species commonly found in the intestinal microbiota of humans. A mean of 2629 ± 190 proteins belonging to the 17 expected species has been identified with a total of 8016 ± 1374 peptides and 44,373 ± 3906 PSMs (Table 1). The proteomic data were used to estimate the biomass of each species.²¹ The results are compared with the cell number, 16S and 18S, and genome copy number information provided by the manufacturer (Figure 6).

Table 1. Number of Proteins, Peptides, and PSM Identified According to the Species of the Gut Standard^a.

species name	classification	cell number (%)	identified proteins	identified peptides	identified PSMs
Veillonella rogosae	Gram −ve	6.00 × 107	196 ± 10	671 ± 168	3543 ± 464
Faecalibacterium prausnitzii	Gram +ve	4.45 × 107	413 ± 47	1389 ± 236	9764 ± 870
Roseburia hominis	Gram neutral	3.74 × 107	365 ± 143	1424 ± 411	10,015 ± 1330
Lactobacillus fermentum	Gram +ve	2.91 × 107	268 ± 23	830 ± 57	5341 ± 119
Bifidobacterium adolescentis	Gram +ve	2.66 × 107	94 ± 26	189 ± 54	559 ± 135
Escherichia coli	Gram −ve	2.62 × 107	329 ± 46	855 ± 157	2825 ± 419
Bacteroides fragilis	Gram −ve	2.51 × 107	199 ± 15	515 ± 50	2266 ± 153
Fusobacterium nucleatum	Gram −ve	2.27 × 107	164 ± 19	568 ± 99	3571 ± 352
Prevotella corporis	Gram −ve	1.88 × 107	80 ± 8	176 ± 12	583 ± 27
Akkermansia muciniphila	Gram −ve	4.86 × 106	22 ± 4	34 ± 5	81 ± 10
Clostridioides difficile	Gram +ve	3.30 × 106	51 ± 10	117 ± 16	479 ± 30
Methanobrevibacter smithii	archaea	5.10 × 105	0 ± 0	0 ± 0	0 ± 0
Candida albicans	fungi	4.80 × 105	121 ± 18	307 ± 87	1119 ± 193
Saccharomyces cerevisiae	fungi	4.80 × 105	299 ± 29	826 ± 196	3487 ± 393
Salmonella enterica	Gram – ve	1.95 × 104	20 ± 2	99 ± 22	687 ± 137
Enterococcus faecalis	Gram + ve	3.30 × 103	3 ± 0	6 ± 1	23 ± 4
Clostridium perfringens	Gram + ve	2.70 × 102	6 ± 1	11 ± 2	31 ± 6

^aThe five E. coli strains are grouped under the same species. The cell number is given by the manufacturer.

Figure 6.

Biomass estimations using proteomic approaches (in-solution and ChipFilter proteolysis) for next-generation sequencing methods and cell number for species present in the standard gut mix. Cell number, genome copy, 16S and 18S, and 16S data were provided by the manufacturer.

Analysis of Gut Microbiome Samples Is Possible Using the ChipFilter

To check whether the matrix of complex biological samples was compatible with the use of the ChipFilter, mouse feces samples were analyzed in the device. Starting from about 4 mg of the sample, a mean of 687 ± 172 proteins have been identified with a total of 2612 ± 602 peptides and 6885 ± 1275 PSMs (Table 2). Proteins mainly belong to the firmicute, Bacteroidetes, actinobacteria, and proteobacteria phylum (Figure 7).

Table 2. Number of Proteins, Peptides, and PSM Identified in the Fecal Samples According to the Phylum.

	identified proteins	identified peptides	identified PSMs
Proteobacteria	69 ± 74	208 ± 233	367 ± 339
Firmicutes	446 ± 192	1597 ± 530	3644 ± 1019
Bacteroidetes	86 ± 69	325 ± 265	616 ± 443
Actinobacteria	12 ± 4	41 ± 16	130 ± 67
others	75 ± 27	443 ± 100	2128 ± 198

Figure 7.

Number of proteins identified according to the phylum (A). Phylogenetic relationships between the identified phylum determined using Unipept Desktop v2.0.1 software (B).

Discussion

In the previous work, we have shown the usefulness of the ChipFilter device for the preparation of eukaryotic cells for bottom-up proteomic analysis.¹⁵ In the present study, this system was adapted to the analysis of microbial cells having different phenotypes and cell wall compositions. We have shown that it is possible to perform cell lysis directly into the ChipFilter device, before other preanalytical steps are performed. On a mixture of three species, we identified more proteins and peptides than standard protocols compared in the study. We extended the use of this ChipFilter device to the analysis of a standard gut microbiome. Proteins from almost all species were identified. Moreover, these results were used to estimate the contribution of each species to the biomass, giving results that are in line with those determined by nucleic acid techniques.

To use the ChipFilter device for whole microbial cell sample preparation, we made two types of changes. The first concern is the sample loading. Our earlier work mainly focused on the usage of cellular lysates that was introduced to the ChipFilter by using the same microfluidic capillaries used for the other reagents. When this setup was used for microbial cells, we observed cross-contamination between the samples. To overcome this problem, samples were introduced in the device using an external piston syringe and pump and not through capillaries. Second, the lysis buffer containing the ODG was adapted to be more efficient in microbial cell lysis with the addition of lysozyme in the case of complex mixtures comprising Gram-positive bacteria.

Detergents are the commonly preferred reagents for cell lysis, which are often enhanced by the use of physical or mechanical lysis methods to break the cell wall and membranes. Commonly used lysis buffers use the anionic detergent (SDS) in the range of 1–5% (m/v). SDS is highly effective in cell lysis but is generally associated with several drawbacks including the requirement for urea washing to neutralize, reduction in trypsin activity, and influence in mass spectrometry detection.²³ Hence, we use an alternative lysis buffer without SDS that contains the nonionic detergent ODG. It allowed direct cell lysis on the microfluidic device without any pretreatment. Simple washing with ammonium bicarbonate buffer was sufficient to remove the detergent from the device.

Particular attention was paid to the physio-chemical characterization of the peptides produced after CFP. In CFP, variations like peptides generated can be caused by peptide adsorption on PDMS, continuous fluid flow, and protein processing conditions provided. Peptides generated by the ChipFilter method were novel as they had a higher percentage of peptides with long amino acid chains, higher mass, and more missed cleavage than other methods included in the study. This increase in the number of missed cleavages largely explains the other physico-chemical characteristics observed. Since, in the ChipFilter, digestion is realized under a continuous flow process, peptides smaller than the membrane cutoff can escape from the digestion chamber, even if their digestion is incomplete. It has already been established that increasing the number of missed cleavages can be beneficial in terms of the number of identifications, sequence coverage, and identification of PTM.²⁴⁻²⁶ However, these partial digestions can also increase the diversity of forms in which a sequence will be identified, inducing a bias for quantitative aspects. Logically, the quantitative data obtained with the ChipFilter are no better than those obtained with other methods.

One of the reasons for the generation of distinct peptides can be the proteolysis condition provided inside the ChipFilter. Mechanisms of proteolysis can be broadly classified into in-gel,²⁷ in-solution,²⁸ and in-solid phases,²⁹ while the mechanism of catalysis in the device can be argued to be a hybrid between the in-solution and in-solid phases. The chamber offers the space for interaction between the proteins and the trypsin similar to the in-solution phase, whereas the nitrocellulose membrane and PDMS adsorb the trypsin, thereby acting as the solid phase to enable catalysis.

Also, during the catalysis, a small flow rate (0.5 μL/min) of the ABC buffer was provided to increase the exchange of the reactants. These dual digestion mechanisms offered by the device can also be influencing factors in the generation of distinct peptides. It is well-known that a shorter digestion time can generate missed cleavages. Further, shorter digestion times have been reported in past studies to be advantageous.²⁶ The ChipFilter also identified most of the peptides identified by other methods. All this suggests that even though the number and size of peptides are greater after ChipFilter proteolysis, it does not seem to induce an identification bias compared to other methods.

The ChipFilter method was used to study a commercial mixture of pure isolates obtained from a gut sample consisting of 17 different species with different abundance. This mixture, which has the advantage of being of known composition and representative of the phylum present in the intestinal microbiome, is classically used to assess the performance of sample preparation for genomic experiments. To increase the proteome coverage, this sample was analyzed on a Q-Exactive HF mass spectrometer, which was presumed to be more sensitive than the Q-Exactive spectrometer used in the first part of this work. However, the results obtained with the Q-Exactive HF mass spectrometer are never directly compared to those obtained with the Q-Exactive mass spectrometer. Our proteomic data allowed us to identify at least 10 proteins specific for 14 of these species. Proteins support the vast majority of metabolic functions of living organisms. Knowing the protein composition of a community allows one to distinguish the nature of the microorganisms that generate it and is necessary to describe functions like cell-to-cell interactions, the contribution of cells in the biochemical pathways operating within a community like the breakdown of food by gut communities,³⁰ or identifying biomarkers.^31,32 In the present work, functional analysis of identified proteins has not been carried out due to the use of a standard mixture. The advantage of metaproteomic approaches over other screening approaches is that they can provide information on the functional state of a microbial community, for example, by identifying expressed enzymes and their post-translational modifications. Obtaining the best possible coverage of the proteome reinforces interest in these approaches. Moreover, as a proof of concept, the contribution of each species in the overall biomass was quantified according to Kleiner et al.²² for in-solution and ChipFilter proteomic data. In this model, the quantification data are summed based on the taxonomic assignment of inferred proteins and not based on the taxonomic assignment of peptide identifications because peptides are frequently associated with multiple proteins from different taxa. For both approaches, our results suggest that, in addition to the qualitative data, proteomic data can be used to estimate the contribution of each species to biomass. Our results are in line with those proposed by other sequencing (16S, 18S, or shotgun sequencing) results provided by the manufacturers. This aligns the proteomic finding with the nucleic acid results obtained externally. It also shows the feasibility of the ChipFilter method for proteomic sample preparation.

ChipFilter allows several advantages other than automation, including sample collection, handling of low sample density, and effective cell lysis of a wide range of bacteria and fungi. Additionally, the ChipFilter can be used offline allowing for the collection of the peptides before LC–MS. It should be noted that at this point, the multiplexing/parallelization requires the usage of additional ChipFilters and pumping devices that are similar to the robotic arms-based workstation,³³ increasing the installation cost and operation complexity. Further, the development of microfluidic systems dedicated to automated sample preparation for proteomics is less with a recent report suggesting the usage of only three parallel samples processing using microfluidic devices.³⁴ Future studies will be undertaken to effectively reduce sample processing time.

The main advantage of the ChipFilter is its ability to analyze very small sample volumes. This could be the case, for example, for research into nosocomial diseases, aerosols, saliva, skin, a gut sample across the digestive tract, layers of a biofilm, or forensic experiments. In the present work, we have shown that the key problem of cell lysis and protein extraction from structurally complex microorganisms, including fungi and Gram-positive bacteria, can be overcome using the ChipFilter. A number of steps still need to be completed before moving on “real-word” samples, including determining the matrix effect. As a proof of concept, we have shown that after solubilization of microorganisms, treatment of a feces sample by centrifugation¹⁶ is enough to eliminate the particles that could obstruct the device. Protocols must be adapted to the biological matrices studied, taking into account the nature of the samples, the concentration of microorganisms, and the physical and biological contaminants. By nature, the ChipFilter is suitable for small sample volumes and could be clogged by insoluble particles, but it has the advantage of allowing the use of a large wash volume in relation to the sample volume. Lysis and proteolysis conditions may need to be adapted to the biological matrix being studied.

In conclusion, the ChipFilter-based sample preparation method for bottom-up proteomics by LC–MS/MS allowed for the identification of microbial proteins from single species to communities, paving the way for its use in metaproteomic analyses. One advantage of confining a low cell number without the loss of proteins during lysis or washing stages is that automation will allow for accelerated sample preparation with better identification. The use of ChipFilter could be extended to study microbiomes with a low sample volume or cell density that require confinement during preanalytical steps.

Glossary

Abbreviations

ABC

ammonium bicarbonate

ACN

acetonitrile

CFP

ChipFilter Proteolysis

DTT

dithiothreitol

IAM

iodoacetamide

kDa

kilo Dalton

LB

Luria–Bertoni

LC–MS/MS

liquid chromatography coupled to tandem mass spectrometry

mFASP

modified filter-aided sample processing

MS

mass spectrometry

ODG

octyl-β-d-glucopyranoside

TFA

trifluoroacetic acid

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jproteome.3c00288.

• Comparative analysis of the different workflows and physical chemical characteristics of the peptides generated by different preparation methods for the 3× 1E2 cell sample (PDF)

Author Contributions

The manuscript was written through the contributions of all authors. All authors have approved the final version of the manuscript.

This project has received funding from the European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie grant agreement no. 754387.

The authors declare no competing financial interest.