A Semiautomated Paramagnetic Bead-Based Platform for Isobaric Tag Sample Preparation

Abstract

The development of streamlined and high-throughput sample processing workflows is important for capitalizing on emerging advances and innovations in mass spectrometry-based applications. While the adaptation of new technologies and improved methodologies is fast paced, automation of upstream sample processing often lags. Here we have developed and implemented a semiautomated paramagnetic bead-based platform for isobaric tag sample preparation. We benchmarked the robot-assisted platform by comparing the protein abundance profiles of six common parental laboratory yeast strains in triplicate TMTpro16-plex experiments against an identical set of experiments in which the samples were manually processed. Both sets of experiments quantified similar numbers of proteins and peptides with good reproducibility. Using these data, we constructed an interactive website to explore the proteome profiles of six yeast strains. We also provide the community with open-source templates for automating routine proteomics workflows on an opentrons OT-2 liquid handler. The robot-assisted platform offers a versatile and affordable option for reproducible sample processing for a wide range of protein profiling applications.

Graphical Abstract

INTRODUCTION

Isobaric tag-based strategies facilitate accurate, high-throughput proteomic analysis of complex peptide mixtures with minimal missing values within a single experiment. The introduction of the SL-TMT (streamlined-tandem mass tag) protocol simplifies isobaric tag-based proteome profiling workflows.¹ The SL-TMT protocol begins with lysates homogenized in 8 M urea from which total protein concentrations are determined. Protein disulfide bonds are reduced and cystines are alkylated prior to protein precipitation. Proteins are then proteolytically digested with LysC followed by trypsin and subsequent TMTpro labeling. Differentially labeled samples are combined and fractionated using high-pH reversed-phase chromatography prior to LC-hrMS2 or LC-MS3 analysis.

Studies have shown that bead-based protein aggregation techniques (e.g., single-pot solid-phase-enhanced sample preparation, SP3) help alleviate caveats associated with chemical-based protein extraction, mainly due to inconsistent manual aspirations.^2-5 SP3-based workflows can increase throughput, reduce sample-to-sample variability, and introduce the potential for automated sample processing. Recently, we have augmented our SL-TMT protocol to improve the efficiency of sample processing by substituting methanol-chloroform precipitation with the more versatile and higher-throughput SP3-based protein aggregation, known as SL-SP3-TMT.⁶

Robotic protein sample processing has frequently been applied to clinical analyses and other biomedical applications.⁷ In terms of mass spectrometry, analyses of biobanked tissues⁸ and gel bands⁹ are often automated by various methods, some of which have been greatly streamlined to avoid repetitive and potentially error prone sample manipulation.¹⁰ As the use of paramagnetic beads has an ever-expanding role in sample processing for proteomics,^3,4,6,11 several workflows have incorporated various types of paramagnetic beads with robotic liquid handlers, including the epMotion 5073m liquid handling platform,¹² the Agilent Bravo,¹³ the Kingfisher Flex,¹⁴ Proteograph,¹⁵ and recently, the Hamilton Vantage.¹⁶ Here we have developed a platform for semiautomated SL-SP3-TMT sample processing using the opentrons OT-2 liquid handler. The OT-2 system is robust and versatile, yet very affordable relative to instruments with similar functionality.

We benchmarked the robot platform against manual SL-SP3-TMT processing by profiling the proteomes of six common laboratory Saccharomyces cerevisiae yeast strains.^17,18 We assembled each TMTpro16-plex experiment using S288C-derived strains BY4730, BY4739, BY4741, BY4742, and BY4743¹⁹ in triplicate. We note that BY4743 is a diploid strain resulting from a cross between BY4741 and BY4742 strains. As the 16th channel, we added the prototrophic strain Y55.^20,21 Whereas the BY series required specific amino acids to be supplemented into the growth media (due to having deletions of certain metabolic enzymes), the Y55 strain can propagate without amino acid supplementation. As such, we suspect the protein abundance profile of this strain to be very different from that of the auxotrophic stains. In all, six TMTpro16-plexes were processed, that is, three by manual SL-SP3-TMT and three using the robotic platform. We compared reproducibility within and across the multiplexed data sets with respect to depth, precision, and accuracy. The data were similar across all metrics examined including depth of proteome coverage and the number of peptides quantified as well as conclusions drawn from gene ontology and pathways analysis of proteins that differ significantly or are consistent in abundance among strains. In total, we have shown that the robotic sample processing platform enables an SP3-based workflow starting from cell lysate and resulting in isobarically labeled peptides that can be performed with minimal user intervention while retaining data integrity.

METHODS

Materials.

Tandem Mass Tag (TMTpro) isobaric reagents, BCA protein concentration kit, protease inhibitor tablets, Pierce C18 tips, and trypsin were purchased from ThermoFisher Scientific (Rockford, IL). StageTip Empore-C18 material was purchased from CDSanalytical (Oxford, PA). Sep-Pak cartridges (100 mg) were from Waters (Milford, MA). Lys-C protease was from Fujifilm Wako (Richmond, VA). Mass spectrometry grade water and organic solvents were from J.T. Baker (Center Valley, PA). The S. cerevisiae strains were from Horizon Scientific (Cambridge, UK). Synthetic complete (SC) media was from Sunrise Science (San Diego, CA). Sera-Mag Speed Beads (cat. nos. 45152105050350 and 65152105050350) were from GE Life Sciences (Marlborough, MA).

The OT-2 robotic liquid handler was purchased from opentrons (New York, NY) and included the magnetic module, the temperature module, and two GEN2 8-channel pipets: 20 and 300 μL. Labware used on the OT-2 included: 12-channel reservoirs (USA Scientific, 1061-8150), 96-well microplate (Bio-Rad, HSP9601), 2 mL deep 96-well microplates (USA Scientific, 1896-2000), and 8-tube strip PCR tubes (Bio-Rad, TLS0801). Also, 8-tube strip PCR tube caps (Bio-Rad, TLS0803) are required for 37 °C incubations. Proteolytic digests were performed on the Jitterbug Heated Microplate Shaker (Boekel Scientific, 130000).

Yeast Growth and Protein Extraction.

S. cerevisiae strains were grown at 30 °C in synthetic complete (SC) media (with 2% glucose). Cells were harvested at OD_{600 nm} ≈ 0.8 by centrifugation and washed twice with ice cold water. Cells were lysed by bead-beating in 8 M urea, 200 mM EPPS (4-(2-hydroxyethyl)-1-piperazinepropanesulfonic acid), pH 8.5 supplemented with protease inhibitors. Protein concentrations were determined using a BCA assay performed according to manufacturer’s instructions with samples that were diluted at least 1:20 to ensure that urea concentrations were below their noncompatibility limit.

Manual SP3-Based Protein Digestion and TMTpro16 Labeling.

Samples were reduced with 5 mM TCEP for 20 min, alkylated with 10 mM iodoacetamide for 20 min, and quenched with 10 mM DTT for 20 min, all at room temperature. Single-Pot Solid-Phase-enhanced Sample processing (SP3) as described previously⁵ was used during protein isolation and digestion. Reactions were performed in a 96-well plate (200 μL maximum volume) using an 8-channel pipet. In brief, 2 μL of each bead type was added to 40 μg of protein in 40 μL total volume, as prepared above. Neat ethanol was added to a final concentration of 50%. The beads were carefully triturated 10 times. The samples were held to the magnet for 2 min, and the supernatant was aspirated. The beads (with bound protein) were washed three times with 80% ethanol in the same manner. For protein digestion, we added 50 μL of 200 mM EPPS pH 8.5 and Lys-C overnight at room temperature, followed by trypsin for 6 h at 37 °C on an orbital shaker (Jitterbug Heated Microplate Shaker). Both enzymes were added at a 1:100 protease-to-peptide ratio in the presence of beads. Following digestion, we added a final volume of 30% acetonitrile to the eluted peptides and labeled the 40 μg of peptide with 80 μg of TMTpro16 reagent directly into the digestion mixture (still in the presence of beads) for 1 h. The reaction was quenched with a final concentration of 0.5% hydroxylamine. Following quenching, the samples were held to the magnet for 2 min after which the supernatant was collected. The beads were then washed with 40 μL of 200 mM EPPS pH8.5, which was combined with the initial elution. The samples are mixed at a 1:1 ratio across all channels.

Robot-Assisted SP3 Protein Processing and TMTpro16 Labeling.

Workflow development and sample processing strategies using the OT-2 are described and discussed in detail in the Results and Discussion section.

Sample Pooling and Off-Line Fractionation.

For both the robot-assisted and manually processed workflows, the pooled, multiplexed sample was desalted using a 100 mg SepPak cartridge. The ~300 μg of peptide was fractionated with basic pH reversed-phase (BPRP) HPLC, collected in a 96-well plate, and concatenated down to 24 fractions prior to desalting and subsequent LC-MS/MS processing.^22,23

Mass Spectrometry Analysis.

Mass spectrometric data were collected on an Orbitrap Eclipse mass spectrometer coupled to a Proxeon NanoLC-1200 UHPLC instrument. The 100 μm capillary column was packed with 35 cm of Accucore 150 resin (2.6 μm, 150 Å; ThermoFisher Scientific). The scan sequence began with an MS1 spectrum (Orbitrap analysis, resolution 60 000, 350–1400 Th, automatic gain control (AGC) target set as standard, maximum injection time was set as auto). Data were acquired using the FAIMS Pro interface with three compensation voltages (−40, −60, and −80 V) with each scan cycle set as a 1 s TopSpeed method. MS2 analysis consisted of collision-induced dissociation (CID), quadrupole ion trap analysis, automatic gain control (AGC) set as standard, NCE (normalized collision energy) 35, q-value 0.25, maximum injection time 35 ms, and isolation window at 0.7 Th. Following acquisition of each MS2 spectrum, we collected an MS3 spectrum in which multiple MS2 fragment ions were captured in the MS3 precursor population using isolation waveforms with multiple frequency notches. MS3 precursors were fragmented by HCD and analyzed using the Orbitrap (NCE 55, AGC 500%, maximum injection time 200 ms, resolution 50 000 at 400 Th). The Thermo real-time search (RTS) node was used, and the closeout was set to two peptides per protein.

Spectra were converted to mzXML via MSconvert.²⁴ Database searching included all S. cerevisiae entries from UniProt (downloaded August 2020). Each database was concatenated with one composed of all protein sequences for that database in the reversed order. Searches were performed using a 50 ppm precursor ion tolerance for total protein level profiling. The product ion tolerance was set to 0.9 Da. These wide mass tolerance windows were chosen to maximize sensitivity in conjunction with Comet searches and linear discriminant analysis.^25,26 TMTpro tags on lysine residues and peptide N termini (+304.207 Da) and carbamidomethylation of cysteine residues (+57.021 Da) were set as static modifications, while oxidation of methionine residues (+15.995 Da) was set as a variable modification. Peptide-spectrum matches (PSMs) were adjusted to a 1% false discovery rate (FDR).^27,28 PSM filtering was performed using a linear discriminant analysis, as described previously,²⁶ and then assembled further to a final protein-level FDR of 1%.²⁸ Proteins were quantified by summing reporter ion counts across all matching PSMs, as described previously.²⁹ Reporter ion intensities were adjusted to correct for the isotopic impurities of the different TMTpro reagents according to manufacturer specifications. The signal-to-noise (S/N) measurements of peptides assigned to each protein were summed, and these values were normalized so that the sum of the signal for all proteins in each channel was equivalent to account for equal protein loading (column normalization). Finally, each protein abundance measurement was scaled to a percent of the total, such that the summed signal-to-noise for that protein across all channels equaled 100, thereby generating a relative abundance (RA) measurement. Data analysis and visualization were performed in Microsoft Excel or R.

Data Access.

We have included protein lists with TMT relative abundance measurements in Table S1. In addition, we provided peptide lists with TMT relative abundance measurements in Table S2. RAW files will be made available upon request; in addition, data have been deposited to the ProteomeXchange Consortium via the PRIDE³⁰ partner repository with the data set identifier PXD024282.

RESULTS AND DISCUSSION

A Robotic Liquid Handler Facilitated Sample Processing from Lysate to Quenched TMT-Labeled Peptides with Minimal User Intervention.

The SL-SP3-TMT protocol¹ when performed in a 96-well microplate format is well-amenable to automation. Here, we present a semiautomated SL-SP3-TMT workflow on an affordable, yet versatile and robust, liquid handler (the opentrons OT-2), which may accelerate the incorporation of automated sample handling for virtually any laboratory. Below we describe the semiautomated SP3 workflow which we have outlined in Figure 1.

Figure 1.

Workflow for semiautomated paramagnetic bead-based sample preparation for isobaric tagging experiments on a robotic platform. The OT-2 instrument can (1) perform a BCA assay, (2) process the sample from lysate to labeled peptides using a semiautomated SL-SP3-TMT workflow, and (3) desalt with C18-embedded pipet tips. The steps performed on the robot are highlighted in a solid light orange. As no oscillator is currently available on the OT-2, the proteolytic digests were performed on a Jitterbug shaker (highlighted in the figure with a diagonal white and light orange pattern).

In addition to the OT-2 liquid handler, the robot platform required 8-channel P20 and P300 pipettes, a temperature module and a magnet module, as well as an independent heated microplate shaker (The Jitterbug) as a companion accessory for the proteolytic digests. The starting deck required an 8-well reservoir with the SP3 reagents, specifically 12 mL each of 100% ethanol (EtOH), 80% EtOH, and 0.2 M EPPS pH8.5 (Figure S1). The temperature module was set at 4 °C. The reagents in the temperature module had been prealiquoted into 8-well PCR strips. Each strip contained one of the following reagents: 0.5 M TCEP (25 μL), 0.5 M iodoacetamide (IAA) (25 μL), 1 M DTT (15 μL), SP3 beads (12 μL), Lys-C (1 μg/μL, 15 μL), trypsin (1 μg/μL, 15 μL), anhydrous acetonitrile (25 μL), TMTpro reagent (5 μg/μL, 6 μL), and hydroxylamine (25 μL) in each well. As many of the reagents were temperature-sensitive and a considerable amount of time was often spent between steps, we incorporated pauses (which may be bypassed) in the protocol to allow reagents to be placed on the temperature module shortly before use. The starting deck also included a 2 mL 96-well microplate with cleared lysate at ~1 mg/mL (as determined by the BCA, or equivalent, protein concentration assay). We suggest a lysis buffer that consists of 8 M urea, 0.2 M EPPS, pH 8.5 with 1× protease and phosphatase inhibitors as outlined in the SL-TMT protocol.¹ We also recommend using an optional 96-deep well (2 mL) plate to collect liquid waste rather than depositing it in the tip trash bin.

The workflow began with cleared lysate, as different sample types (e.g., cultured cells, tissue, plasma) required specialized lysis protocols or equipment (e.g., for sonication or homogenization) which were currently not commercially available on the OT-2. We used 16 samples that occupied the first two columns of a 96-well microplate. The entire lysate was then reduced with 5 mM TCEP for 20 min, alkylated with 10 mM IAA for 20 min, and quenched with 10 mM DTT for 20 min. The samples were mixed after each step. The robot transferred a fraction (40 μL) of the prepared lysate to a 96-well microplate on the magnetic module (magnet was disengaged). The robot dispensed the paramagnetic carboxylate-coated SP3 beads (Sera-Mag Speed Beads) to this aliquot of lysate which was mixed by trituration. In line with the SP3 protocol, the sample was diluted with a final concentration of 50% ethanol and mixed. After a 5 min incubation, the magnet was engaged for 2 min after which the unbound liquid was aspirated. The magnet was then disengaged, and 80% ethanol was added to wash the beads by trituration. The magnet was once again engaged for 2 min and the unbound liquid was aspirated. The 80% ethanol wash was repeated once. At this stage, the samples were ready for digestion and 40 μL of 200 mM EPPS, pH 8.5 were added to the beads (with bound proteins) as was 0.5 μg of Lys-C protease. The samples were mixed until homogeneous.

At this point in the workflow, some user intervention was required. The samples were now manually capped and placed on a Jitterbug Heated Microplate Shaker (set at speed 1 and at room temperature) overnight. The next morning, the microplate was placed back on the robot which dispensed 0.5 μg of trypsin protease. We noted that PCR strips containing the reagents (e.g., trypsin, Lys-C, TMT) that were placed on the temperature module (set at 4 °C) were thawed shortly before use so as not to impair reagent integrity. The plate was now incubated for 6 h on the Jitterbug (set at speed 1 and 37 °C). After cooling to room temperature, the microplate was transferred to the disengaged magnet. The robot added acetonitrile to a final concentration of 30% and TMT reagent at a 2:1 (TMT/peptide) molar ratio. After 1 h, a short labeling check sample may be manually prepared for MS/MS analysis to ensure labeling efficiency,¹ or the workflow can be finalized with the addition of a final concentration of 0.5% hydroxylamine. At 15 min after the addition of the hydroxylamine, the magnet was applied for 2 min after which the supernatants were transferred into available wells on the “lysate plate.” Another 40 μL of 200 mM EPPS was added to the beads and mixed. The magnet was applied again for 2 min after which the supernatants were transferred into the same wells corresponding to the first elution. At this point, the sample was combined at a 1:1 ratio, fractionated off-line (e.g., by basic pH reversed phase), and the concatenated superfractions were desalted prior to mass spectrometric analysis, as described previously.¹

The robot excelled at liquid transfer, as we did not observe drops or cross-contamination from pipet tips as they traversed the deck. However, several default parameters were not appropriate for the needs of our applications. The most challenging parameters to optimize involved aspiration and mixing in the presence of beads. In this sequence, the magnet engages and the paramagnetic beads accumulate on the inner sides of the labware. During manual preparation, care was needed to avoid inadvertently aspirating beads during washes; likewise, for automated preparation, the deck and pipets must be properly calibrated for precise tip positioning, aspirating, and dispensing. Many SP-3 protocols recommended mixing by tube inversion or oscillation; however, this option was not available on the OT-2. Here, we must rely on mixing by trituration. As such, the time of magnetic field application, pipet tip height, as well as aspirating and dispensing speeds were key to successful mixing. The optimal parameters were determined by visual inspection to maximize the heterogeneity of the mixed sample. For these parameters, none of the default settings were satisfactory, the magnet height was adjusted to 7 mm, the pipet height was set to 1 mm, the aspirate speed was set to 30 μL/s, and the dispense speed was set to 300 μL/s. We also noted that certain 96-well microplates facilitated the resuspension of beads better than other plates tested. As such, one must ensure that a given plate is appropriate for a specific application. The settings we suggested worked well for 40 μg of protein in a 40 μL sample volume, but may need some adjustment if different protein concentrations and/or volumes were to be used.

The Number of Quantified Peptides and Proteins Were Similar When Using Either the Robotic Platform or Manual Sample Processing.

We evaluated the robotic platform against the standard SL-SP3-TMT which was done manually on a 96-well plate using 8-channel pipets and a magnetic base (Figure 2A). We performed three replicate TMTpro16-plex³¹ experiments each for the robot-assisted and manually processed sample workflows. Each TMTpro16-plex experiment consisted of six yeast (S. cerevisiae) strains. The strains BY4730, BY4739, BY4741, BY4742, and BY4743 (Table 1) were arranged in triplicate and the 16th channel consisted of the prototrophic strain Y55^20,21 (Figure 2B). Samples in each TMTpro16-plex experiment were pooled at a 1:1 ratio across all channels and fractionated off-line using basic pH reversed phase (BPRP) chromatography. Twelve samples were analyzed for each of the TMTpro16-plex experiments across a 90 min gradient. Mass spectrometer data were acquired on an Orbitrap Eclipse mass spectrometer³² with the FAIMS Pro interface that reduced the chemical noise and enhanced the dynamic range of quantification.^33,34 In addition, our data acquisition strategy used real-time database searching (RTS-MS3), which increased the number of quantified peptides^33,35 and improved the accuracy of isobaric tag-based quantification.^36,37

Figure 2.

Benchmarking the robotic platform for SL-SP3-TMT sample processing. (A) Overview of the SL-SP3-TMT workflow highlighting the OT-2 and manual sample processing strategies. (B) Experimental layout of the TMTpro16-plex experiment designed to profile the yeast parental strains that were used to benchmark the robot. Triplicate TMTpro16-plex experiments were performed both using the robot and manually. Created in part with BioRender.com.

Table 1.

Genotypes of S. cerevisiae Used in the TMTpro16-plex Experiments

background	genotype
BY4730	leu2Δ0 met15Δ0 ura3Δ0
BY4739	leu2Δ0 lys2Δ0 ura3Δ0
BY4741	his3Δ1 leu2Δ0 met15Δ0 ura3Δ0
BY4742	his3Δ1 leu2Δ0 lys2Δ0 ura3Δ0
BY4743	4741/4742
Y55	HO/HO

Ultimately, data from the six experiments were combined into a single data set at a 1% false discovery rate. No significant differences (p-value < 0.01) were observed between the robot-assisted and manually processed samples in terms of the number of proteins (Figure 3A), as well as unique and total peptides (Figure 3B). Across all experiments, the tallies were similar, as the number of proteins ranged from 4261 to 4318, the number of unique peptides ranged from 25 881 to 28 329, and the number of total peptides ranged from 34 018 to 38 165. When considering nonredundant proteins and peptides across triplicate TMTpro16-plex experiments, the robot-assisted experiments quantified 4613 proteins and 109 437 peptides while the manually processed sample experiments yielded similar numbers with 4626 proteins and 107 897 peptides. Both methods quantified over 80% of the verified S. cerevisiae proteome in 18 h. In total, we quantified 4739 proteins that were represented by over 217 000 peptides across 6 yeast strains.

Figure 3.

Comparison of the robot-assisted and manually processed data sets. (A) Bar chart illustrating the number of proteins quantified in each TMTpro16-plex experiment. (B) Bar chart illustrating the number of unique and total peptides quantified in each TMTpro16-plex experiment. (C) Upset plot showing proteins overlapping across data sets. (D) Venn diagram illustrating the overlap of nonredundant proteins in the robot-assisted and manually processed data sets. The three replicates are collapsed into a single data set.

We then investigated the overlap between the two sample processing methods. We assessed the similarity among the six yeast strains using an upset plot (Figure 3C). We noted the 3653 proteins were identified in all six experiments and slightly over 400 more were identified in 5 of the 6 experiments. In fact, the number of proteins found in 4 or fewer of the 6 experiments were less than 30 per subset (Figure S2A). In addition, we illustrated the overlap of the nonredundant proteins for all three replicates of the robot-assisted versus manually processed sample sets using a Venn diagram. A total of 4505 proteins were quantified in both methods, corresponding to 95% of the total proteins quantified (Figure 3D). This percentage was comparable to that obtained when comparing any two replicates prepared with the robot (Figure S2B) or manually (Figure S2C). These data demonstrated no bias with respect to sample processing method on the number or the identity of the proteins or peptides quantified.

Protein Abundance Profiles Were Reproducible Both within and across TMTpro16-plex Experiments Regardless of Sample Processing Strategy.

Next, we addressed if quantitative differences existed with respect to TMT relative abundance (TMT RA) measurements. We averaged the triplicate TMT RA values for each auxotrophic yeast strain within each TMTpro16-plex experiment. We then used these values for all six data sets, along with the value of the Y55 strain for each experiment and performed hierarchical clustering on the average TMT RA values using Euclidian distance and Ward’s linkage (Figure 4A). The auxotrophic yeast strain samples clustered as anticipated, first by replicate, second by method, and finally by strain. The prototrophic Y55 strain clustered furthest away from the other stains, and as expected, was the most different among the yeast strains investigated. This result illustrated that no batch effect was observed as the samples did not cluster by experiment. Likewise, we observed no evidence suggesting that experimental errors were propagated by the robot.

Figure 4.

Comparison of yeast strain TMT relative abundance (RA) profiles between the robot-assisted and manually processed samples. (A) Heat map of the average TMT RA values for yeast strains in the robot-assisted and manually processed data sets. Hierarchical clustering was generated using Euclidian distance and Ward’s linkage. (B) Violin plots showing the distribution of the average fold change of a given peptide’s relative abundance ratio of the robot-assisted versus manually processed samples. (C) Correlation matrix of the TMT RA values for proteins across all strains with respect to the sample preparation strategy. Correlation plots are illustrated on the lower left of the diagonal and the corresponding Pearson correlation (R²) values are on the upper right. Scatter plot and R² values for the same strains with different sample preparation strategies are highlighted with a colored border corresponding to a given strain. TMT RA, tandem mass tag relative abundance.

We next attempted to explore the differences more quantitatively between the robot-assisted and the manually processed samples by calculating the log₂ ratios of the average TMT RA value of a given protein in the robot-assisted sample data set versus that in a manually processed sample data set for each strain. We illustrated the data in a series of violin plots (Figure 4B). The distribution of the log₂ ratios of the robot/manual processing showed very little difference between sample processing methods for all strains tested. In fact, over 90% of the ratios were between −0.1 and 0.1, thereby showing very small quantitative differences were observed between the two sample processing workflows.

To evaluate further the quantitative similarities of these data sets, we generated a correlation matrix of TMT relative abundance (RA) for all strains and both sample processing strategies (Figure 4C). In this matrix, Pearson correlation (R²) values were listed to the right of the diagonal and corresponding correlation plots were illustrated to the left of the diagonal. The robot-assisted samples were compared in upper left quadrant and manually processed samples were compared in the lower right quadrant. We highlighted scatter plots in the lower left quadrant with colored borders that corresponded to the same strain processed with the different methods. These colored borders were used in the upper right to highlight the corresponding R² values for the same strains with different sample processing workflows. We noted very high correlation between sample processing methods (R² ≥ 0.9) for all strains investigated. In addition, the R² values between different strains were similar regardless of method (e.g., the R² values between strains BY4730 and BY4739 were 0.11 and 0.09 for the robot-assisted and manually processed samples, respectively). These data strengthen further our confidence in the quantitative similarity between sample processing strategies.

Protein Abundance Profiles of Genotype Markers Were Similar Irrespective of Sample Processing Strategy.

The auxotrophic parental yeast strains can be selected using different combinations of known markers represented by null mutations of the genes encoding HIS3, LEU2, LYS2, MET17, and URA3. These proteins are enzymes required for the syntheses of essential metabolites which may be depleted from the media for purposes of selection. HIS3 is an enzyme involved in histidine biosynthesis.³⁸ LEU2 is a β-isopropylmalate dehydrogenase that is implicated in leucine synthesis.³⁹ Likewise, LYS2 is an enzyme that governs the biosynthesis of lysine.⁴⁰ MET17 is an O-acetyl homoserine-O-acetyl serine sulfhydrylase that is required for the biosynthesis of both methionine and cysteine.⁴¹ Finally, URA3 is an orotidine-5′-phosphate decarboxylase that participates in the biosynthesis of pyrimidines.⁴² All five of these proteins are expressed in the Y55 strain. We noted here that the cells were cultured in synthetic complete (SC) media which included these molecules, so all cells can properly propagate. Considerable TMT reporter ion signal in these null mutants can be indicative of procedural abnormalities, akin to the TKO standard that we developed to understand and attempt to limit interference.^43,44 We recognized that ideally the TMT reporter ion signal should be absent for proteins which have been deleted in a given strain. However, we have also shown that interference in isobaric tagging persists even with FAIMS and RTS.³⁷ As such, we do anticipate some signal in channels where the protein is expected to be deleted, which we presume will be similar regardless of sample preparation workflow.

For each knocked out protein, we plotted the relative abundance profiles using the average across the three replicates for each sample processing workflow (i.e., the robot and manual processing) for each strain. We noted substantial reduction of HIS3 levels in BY4741, BY4742, and BY4743 (Figure 5A). Likewise, MET17 levels were reduced nearly completely in BY4739 and BY4741 and to approximately 50% in BY4743 (which was generated from a cross between BY4741 and BY4742) (Figure 5B). Conversely, the deletion of LYS2 was noted in BY4739 and BY4741, while the LYS2 level in BY4743 again dropped by half (Figure 5C). Finally, negligible amounts of LEU2 (Figure 5D) and URA3 (Figure 5E) were measured in the auxotrophic strains (as consistent with their genotype), with strain Y55 representing over 90% of the TMT signal. Our data supported the expected results regarding the deleted genes (Table 1), for which the protein abundance profiles were similar for both sample processing strategies and for all deletion strains. For these plots, we can also approximate the interference (as denoted by TMT reporter ion signal in channels with the null mutations) to be similar regardless of the sample processing method used. As such, it follows that we observed no major sample processing related differences in the protein abundance profiles of the knocked-out proteins across strains. Such a result supported further our claim that quantitative accuracy was not lost when using the robot-assisted platform.

Figure 5.

Genotype markers of the deletion strains. (A) HIS3, (B) LYS2, (C) MET17, (D) LEU2, and (E) URA3. TMT RA, tandem mass tag relative abundance. Error bars represent ± one standard deviation (n = 9 for auxotrophic strains, n = 3 for Y55).

Similar Conclusions Could Be Drawn from the Protein Abundance Profiles Using Either Sample Processing Strategy.

In addition to the deleted proteins, we also compared the protein abundance profiles of the top approximately 2% of highly altered and unchanging proteins in our data set. We calculated the coefficient of variation (CV) for the five auxotrophic yeast strains using the average of nine replicates (three samples for each of three TMTpro16-plex experiments) for each sample processing method. We noted that the distribution of CV values was nearly identical between the two methods (Figure S3). We considered the CVs of highly changing proteins across strains to be >25%, while those for the less altered proteins to have CVs < 2.5%. With this threshold, 123 and 126 proteins were considered highly changing in the robot-assisted and manually processed data sets, respectively, while the corresponding numbers of relatively unchanged proteins were 177 and 169.

We queried these proteins using the DAVID bioinformatics database⁴⁵ to determine if similar conclusions, with respect to gene ontology and pathway annotations, could have been drawn when using either the robot-assisted or manually processed data sets (Table S3). We noted that categories among those with the highest percentage of proteins (both highly variable and unchanging) were similar across the two sample processing methods. Specifically, our data showed that the most highly altered proteins were metabolism-related, as well as cell membrane related. These classifications agreed well with the genotype deletion differences in these strains. The deleted genes all participated in some aspect of metabolism. Moreover, the inability of a given stain to synthesize these metabolites required transport from the environment (generally supplemented into the growth media). We highlighted some highly altered proteins including CSS1 (Condition Specific Secretion) which is an extracellular protein of unknown function⁴⁶ (Figure S4A), GAD1 (GlutAmate Decarboxylase) which converts glutamate into gamma-aminobutyric acid (GABA) during glutamate catabolism⁴⁷ (Figure S4B), and MUP1 (Methionine Uptake) that is a high affinity methionine permease that participates in methionine and cysteine uptake⁴⁸ (Figure S4C). We noted that the protein profiles were nearly identical between the robot-assisted and manually processed experiments. In addition, proteins with relatively unaltered levels among the five auxotrophic stains were classified mainly as those associated with the cytoplasm (e.g., proteosome subunits), cytoskeleton, or nucleus. Thus, these proteins were related less to metabolism and more to structure/organization of the cell. We also highlighted some relatively unaltered proteins including PRE5 (PRoteinase yscE) which is a subunit of the 20S proteasome⁴⁹ (Figure S4D), PDA1 (Pyruvate Dehydrogenase Alpha) which is a subunit of the pyruvate dehydrogenase (PDH) complex⁵⁰ (Figure S4E), and ARP2 (Actin-Related Protein) which is required for the motility and integrity of actin patches⁵¹ (Figure S4F). These data showed that similar conclusions can be drawn with respect to protein classifications and protein abundance profiles of highly altered and relatively unaltered proteins regardless of sample processing method. For further data exploration, we designed an interactive R Shiny application “Six Yeast Strain Protein Abundance Profile Viewer,” which can be accessed at http://wren.hms.harvard.edu/OT2yeast/ for browsing the relative abundance of the proteins in our data sets (Figure S5).

Additional Workflows Associated with Isobaric Tagging Sample Preparation Can Be Performed on the Robotic Platform.

Although our aim here was to develop and showcase SP3-based sample processing on the robotic liquid handler, we have used several other workflows to automate common tasks associated with the SL-TMT protocol.¹ More specifically, we also developed methods on the robot that can be used to prepare samples for concentration estimates (i.e., a bicinchoninic acid (BCA) or other colorimetric/fluorometric assay) (Figure S6) as well as a C18-embedded tip-based desalting strategy (analogous to a StageTip⁵²) (Figure S7). We also adapted our semiautomated SP3 protocol for high-throughput processing (Figure S8). Here, an entire 96-well plate of samples can be processed yielding up to six TMTpro16-plex experiments. We briefly described these three workflows, which are available to download at http://wren.hms.harvard.edu/OT2yeast/, in the Supporting Information Supplemental Methods section.

Limitations.

As with any new workflow, some limitations exist which typically can be circumvented. One such limitation is the inability to reload pipet tips for certain applications that require multiple transfer and/or mixing steps. We encountered this limitation when designing the high throughput 96-well SP3 protocol (Figure S8). Here, more pipet tips were required than could be accommodated by the deck. We addressed this issue by designing five distinct methods to be run sequentially during which tips may be reloaded between methods. We speculated that such user intervention will be necessary unless a robotic arm is developed that can replenish tips and/or remove empty tip wafers. Another limitation was related to the high cost and efficient use of reagents such as TMT and proteases. We used reservoirs for less costly reagents, such as the EPPS buffer and methanol, for which excess may be recollected and reused. However, this practice is impractical for labeling reagents and proteases. As such, we aliquoted such reagents in narrow-bottom PCR tube strips and added an additional 10% to compensate for absorptive losses. We recommend the prealiquoting of labeling reagents and enzymes into PCR tube strips which are then only thawed and placed on the prechilled temperature module shortly before use. In addition, visual inspections should be performed frequently to ensure proper pipetting and mixing throughout the protocol. We recommend installing small, inexpensive cameras (“webcams”) for recording, monitoring, and remote troubleshooting of robotic sample processing. Moreover, as a guide for novice users, we have included two subsections in the Supporting Information Supplemental Methods entitled (1) Preparation for OT-2 Semiautomated Paramagnetic Bead-Based Platform Protocol and (2) QuickStart Guide for the OT-2 Semiautomated Paramagnetic Bead-Based Platform.

CONCLUSIONS

We have developed a semiautomated workflow for SL-SP3-TMT sample processing using a relatively affordable liquid handling system and a companion orbital shaker/heater. We benchmarked the robotic platform against manual SL-SP3-TMT by profiling the proteomes of six common parental laboratory yeast strains in triplicate TMTpro16-plex experiments. Our data showed similar numbers and a high overlap of peptides and proteins, as well as excellent quantitative correlation between the two sample processing methods resulting in similar conclusions being drawn from the data. Along with the semiautomated SP3 protocol, we also developed, and made publicly available, open-source templates for protein concentration estimations, C18-embedded tip desalting for the OT-2 robot, and a workflow consisting of five sequential methods that enable the simultaneous processing of 96 samples via SP3. Moreover, the protocols developed herein can be adapted seamlessly to other paramagnetic bead- or tip-based applications. For instance, the robotic platform can enrich phosphopeptides using paramagnetic TiO₂ beads or tip-embedded TiO₂ or other immobilized metal affinity chromatography (IMAC) beads (e.g., iron(II)- or zirconium(IV)-IMAC). Moreover, we anticipate that applications related to protein stability profiling may also be ported to this robotic liquid handler.^53,54 We expect that the expanded use of robotics in establishing streamlined, high-throughput sample processing workflows will better overcome the challenge of performing large-scale proteome profiling while minimizing manual sample preparation time and maximizing data output in virtually any laboratory setting.