Evaluating linear ion trap for MS3-based multiplexed single-cell proteomics

Abstract

There is a growing demand to develop high throughput and high sensitivity mass spectrometry methods for single-cell proteomics. The commonly used isobaric labeling-based multiplexed single-cell proteomics approach suffered from distorted protein quantification due to co-isolated interfering ions during MS/MS fragmentation, known as ratio compression. We reasoned the use of MS3-based quantification could mitigate ratio compression and provide accurate quantification. However, previous studies indicated reduced proteome coverages in the MS3 method, likely due to long duty cycle time and ion losses during multi-level ion selection and fragmentation. Herein we described an improved MS acquisition method for MS3-based single-cell proteomics by employing a linear ion trap to measure reporter ions. We demonstrated linear ion trap can increase the proteome coverages for single-cell-level peptides and higher gains were obtained for MS3 methods. The optimized real-time search MS3 method was further applied to study the immune activation of single macrophages. Among the total of 168 single cells, over 1200 and 1000 proteins were quantifiable when 50% and 75% valid values were required, respectively. Our evaluation also revealed several limitations of the low-resolution ion trap detector for multiplexed single-cell proteomics and suggested experimental solutions to minimize their impacts on single-cell analysis.

INTRODUCTION

With the rapid development of microfluidic sample preparation and advanced mass spectrometers, single-cell proteomics (scProteomics) has become one of the most active research areas (1-6). Compared with single-cell genomics and transcriptomics, the capability to directly measure single-cell proteomes and their post-translational modifications has great potential to reveal functional phenotypes of cell populations, to predict disease progression (7, 8), and to identify cell type-specific surface markers and therapeutic targets (9). Two different workflows have been developed and widely used for scProteomics, including label-free (5, 10-16) and isobaric-labeling-based quantification methods (3, 17-22). Although the label-free method can provide more accurate measurement of protein abundance, the analysis throughput is relatively low as every single cell requires a single liquid chromatography-mass spectrometry (LC-MS) run at a time scale of >30 min (11, 16).

With the demand to access a large number of single cells, there is a growing interest in multiplexed scProteomics based on the isobaric labeling method (23). Single Cell ProtEomics by Mass Spectrometry, or SCoPE-MS was developed for high throughput scProteomics by tagging single cells with tandem mass tags (TMT) (3). A carrier channel containing 100 to 200 cells was included in each TMT experiment to improve proteome coverage and reduce sample loss. Since it was developed, SCoPE-MS method has experienced significant improvements, including high-recovery sample preparation (18, 20, 22, 24), advanced MS data acquisition (17, 19, 25), and data processing (21, 26). Currently, the method has enabled the identification of over 1500 proteins at a throughput of >150 single cells per day. However, despite many advantages, the use of isobaric labeling can lead to distorted protein quantification due to co-isolated interfering ions during MS/MS fragmentation (27). The quantification interference or ratio compression can cause an underestimation of protein abundance changes. The co-isolation interference was commonly addressed by performing prefractionation to reduce peptide complexity (28). Unfortunately, offline fractionation is not favorable for scProteomics due to high sample loss and dramatically-reduced throughput.

Alternatively, the ion interference can be efficiently minimized by performing an additional level of gas-phase selection, where multiple MS2 fragments are isolated to perform MS3 fragmentation (29, 30). The MS3 reporter ions contain much less amount of interference signals, resulting in accurate protein quantification. The recent integration of real-time search with MS3 (RTS-MS3) further improved the duty cycle of MS3 acquisition and enabled deep proteome coverages without offline fractionation (31-33). We reasoned that RTS-MS3 would be an ideal method for multiplexed scProteomics. However, in the previous studies by Tsai et al. (17) and Furtwangler et al. (34), significantly-reduced proteome coverages and quantification precision were observed in MS3 method. The reduced performance was mainly attributed to long duty cycle time and ion losses during multi-level ion selection and fragmentation.

This study aims to evaluate the use of linear ion trap (LIT) to measure the reporter ions for RTS-MS3-based scProteomics analysis. One of our motivations to explore LIT for scProteomics is its superior sensitivity due to the use of electron multiplier-based ion detection. Many previous studies have demonstrated electron multiplier detector has the sensitivity to detect single ions or single electrons (35). In contrast, Orbitrap (OT) relies on image current for signal detection, which requires much higher amounts of ions or charges (36). Makarov et al. estimated ~20+ charges were required to generate a detectable signal in OT (37), indicating its sensitivity is ~10-fold lower than linear ion trap. Indeed, a couple of recent studies have demonstrated LIT can greatly improve low-input proteomics based on data-independent acquisition (DIA) (38, 39), parallel reaction monitoring (PRM) (40), and synchronous precursor selection (SPS) MS3 (41). In the SPS-MS3 study (41), Liu and coauthors used 250 ng, TMT 6-plex labeled peptides to compare the proteome coverages and quantification performance between LIT MS3 and OT MS3. The study demonstrated LIT-MS3 can increase protein identification by 66% with minor loss in precision and accuracy. Despite the promise, it is still not clear if the LIT-MS3 method can benefit scProteomics, in which the sample inputs are commonly in picogram scale; 1000-fold lower than the samples tested in Liu’s study (41). To make the evaluation, we comprehensively compared LIT and OT for scProteomics using two different sample inputs, e.g., 100 pg for high input and 25 pg for low input to mimic large and small-sized single cells, respectively (Figure 1). We compared the performance of MS2 and RTS-MS3 in terms of proteome coverages and quantification dynamic ranges. Overall, our study confirmed LIT can greatly improve the sensitivity of scProteomics. The gains were much higher for low-input (25 pg) samples and MS3-based quantification method. Finally, the optimized RTS-MS3 with LIT quantification was applied to study the proteome changes of single macrophages during immune activation.

Figure 1. Experimental design for the comparison of LIT and OT for scProteomics.

Only C or N variants of TMTpro 16plex were used due to low resolution of LIT detector. Two different sample inputs including 100 pg (high) and 25 pg (low) were tested. A same amount of boost peptide (5 ng) was used.

Methods

Reagents and chemicals

Urea, n-dodecyl-β-D-maltoside (DDM), Tris 2-carboxyethyl phosphine (TCEP), Iodoacetamide (IAA), 2-Chloroacetamide (CAA), Ammonium Bicarbonate (ABC), Lipopolysaccharide (LPS), Triethylammonium bicarbonate (TEAB), Trifluoroacetic acid (TFA), anhydrous acetonitrile (a-ACN), and Formic acid (FA) were obtained from Sigma (St. Louis, MO, USA). The enzymes including trypsin (Promega, Madison, WI, USA) and Lys-C (Wako, Japan) were dissolved in 100 mM TEAB before usage. TMTpro 16plex, 50% hydroxylamine (HA), Acetonitrile (ACN) with 0.1% of FA, and Water with 0.1% of FA (MS grade) were purchased from Thermo Fisher Scientific (Waltham, MA, USA).

Cell culture

RAW 264.7 (a mouse macrophage cell line) was grown in Dulbecco’s Modified Eagle’s Medium (DMEM) and C10 (a mouse respiratory epithelial cell line) was grown in Roswell Park Memorial Institute (RPMI) 1640 Medium. Both cell culture medium were supplemented with 10% fetal bovine serum and 1× penicillin-streptomycin (Sigma, St. Louis, MO, USA) and the incubator was operated at 37 °C and 5% CO2. Cells were harvested at 70%-80% confluency. For single cell proteomics analysis, RAW 264.7 cells were maintained in DMEM supplemented with 10% FBS followed to be stimulated with 100 ng/ul of LPS for 24 hr. Both non-treated and LPS-treated RAW264.7 cells were collected. The LPS-induced macrophage activation was verified by nitric oxide (NO) assay using a Griess reagent kit (Thermo Fisher Scientific, Waltham, MA, USA).

Fabrication of the nested NanoPOTS chips

Nested nanoPOTS chips (N2 chip, 3 × 3 cluster in 3 × 11 array) with a well diameter of 500 μm and nanoPOTS chip (4 × 12 array) with a well diameter of 1.2 mm were used for single-cell proteomics. The chips were fabricated on glass slides using standard photolithography, wet etching, and silane treatment approach as described previously (4, 15, 22).

Sample preparation for single-cell-level proteomic experiment

The single-cell level samples were generated from the bulk peptide samples prepared as previously described (22). Briefly, the cultured cells were harvested and washed thrice by 1× PBS. After cell counting, ten million cells were lysed in a buffer containing 8 M urea in 50 mM ABC. Protein concentration was measured with BCA assay. After being reduced and alkylated by DTT and IAA, protein sample was incubated for 4 hrs at 37 °C after adding Lys-C (enzyme-to-protein ratio of 1:40). Then, trypsin (enzyme-to-protein ratio of 1:20) was added and incubated at 37 °C overnight digestion. The digested tryptic peptides were acidified with 0.1% TFA, desalted by C18 SPE column, and completely lyophilized to remove the acidic buffer. The peptides were resuspended with 50 mM HEPES (pH 8.5) solution, and their concentration was measured with BCA assay.

To evaluate the sensitivity improvement at high and low input levels, we generated two TMTpro-labeled samples containing 100 pg and 25 pg peptides per single-cell channel as high and low input sample, respectively. The pooled peptide prepared by mixing peptides from two cell lines at 1:1 ratio was used as boost/carrier. Because of unit resolution of ion trap, we used moiety of C or N heavy isotopes (Table S1). In our study, boost/carrier peptide sample was labeled with 126 and five single cell-level peptide samples were labeled with 129N, 130N, 131C, 132N, and 133N, respectively (Table S1). For the labelling, TMTpro 16plex reagents dissolved in 100% a-ACN were added to each corresponding sample. A TMT-to-peptide ratio of 4:1 (w/w) was used to maintain high labeling efficiency. After 1-hr incubation at room temperature, the labeling reaction was quenched by adding 5% HA and incubating for 15 min. The TMT-labeled peptides were acidified with 0.1% FA and desalted with C18 stage tips. The peptide samples were diluted in 0.1% FA buffer containing 0.1% DDM (w/v), then mixed in two ways; 5 ng of boost/carrier with 100 pg sample peptides per channel for high input and 5 ng of boost/carrier with 25 pg per channel for low input (Figure 1).

Sample preparation for LPS-treated single cell analysis

For single cell proteomics of RAW264.7 cells, two types of microchips including N2 chip (22) that contains single cells in nested nanowells and nanoPOTS chip (4, 15) that contains 50 sorted cells (25 untreated and 25 LPS-treated cells) were processed separately for single cells and boost/carrier, respectively.

The cellenONE system (cellenONE F1.4, Cellenion, France) was used for both cell sorting and sample preparation. First, single cells were sorted into the 0.5-mm-diameter nested nanowells according to experimental design in Table S2. Next, 10 nL lysis buffer containing 0.1% DDM, 2 mM TCEP, and 5 mM CAA in 100 mM HEPES was dispensed into each nanowell. The chip was incubated at 70 °C for 30 min in a humidity box to achieve complete cell lysis, protein reduction, and protein alkylation. Next, proteins were digested to peptides by adding the mixture of 0.075-ng Lys-C and 0.25-ng-trypsin into the nanowells and incubating for 10 h. For isobaric labeling, we added 50 ng of TMT reagent dissolved in DMSO into each of the corresponding nanowells according to experimental design (Table S2). After 1-hr incubation at room temperature, the remaining TMT reagents were quenched by 2 nL of 5% HA. The samples were incubated at RT for 15 min and acidified by adding 5 nL of 5% FA.

For the carrier/boost sample, 25 control RAW264.7 cells and 25 LPS-treated RAW264.7 cells were sorted in the 2.2-mm-diameter well of nanoPOTS chip. The boost/carrier chip was processed in the same manner, but the peptides were labeled with 134N TMTpro. After being acidified, both samples were desiccated to dryness until MS analysis.

LC-MS analysis

For the single cell-level sample analysis, all the samples were analyzed by an LC-MS system which consisted a nanoACQUITY UPLC (Waters, Milford, USA) and a Orbitrap Eclipse Tribrid MS (Thermo Fisher Scientific, Xcalibur Ver. 4.5.445.18). The peptide samples were separated on a capillary LC column (75 μm i.d. × 20 cm) packed with 1.9-μm ReproSil C18 particle and heated at 50 °C with a heater (Analytical Sales and Services Inc., Flanders, NJ, USA). A 100-minutes linear gradient from 8% to 45% solvent B (0.1% FA in ACN) and a flow rate of 200 nL/min were used. Peptides were ionized by applying a voltage of 2200 V at the electrospray source. Ionized peptides were collected into an ion transfer tube heated at 250 °C. Precursor ions within the range from 450 to 1800 m/z were scanned at a 120,000 resolution with an ion injection time (IT) of 50 ms and an automatic gain control (AGC) target of 1E6. Only precursor ions with >2 positive charges and >1E4 intensities were isolated with a window of 0.7 m/z for MS/MS analysis. Advanced peak determination was disabled to reduce precursor co-isolation.

For MS2 analyses, the isolated ions were fragmented by a higher energy collisional dissociation (HCD) level of 34%. The fragments were scanned in Orbitrap at 60,000 resolution or ion trap in enhanced scan mode. The maximum injection time and AGC target for both analyses was 150 ms and 1E5, respectively. For MS3 analyses, synchronized precursor selection (SPS) coupled with real-time search (RTS) strategy was used. The fragment ions produced by collision-induced dissociation (CID) level of 35% were scanned in ion trap with maximum injection time of 100 ms and AGC target of 1E4. Real-time search was configured with a full enzyme specificity, a max variable modification of 2, a max missed cleavages of 1, and a max search time of 100 ms. Static modifications included TMTpro on n-terminal and lysine residue (304.207 Da) and carbamidomethylation (57.02 Da) on cysteine residues. Oxidation (15.9949 Da) on methionine was set as variable modification. Scoring threshold was set to 1.4 XCorr, 0.1 dCn, and 10 ppm precursor tolerance. After 10 notches were selected, these fragment ions were subjected to HCD fragmentation in a level of 45%. The generated ions were scanned in the range of 100 to 150 m/z by Orbitrap at 60,000 resolution or ion trap with zoom scan mode. The MS3 maximum injection time and AGC target for both analyses were 250 ms and 1E5, respectively. The entire cycle time was set to 3 s for all the experiments.

Single RAW264.7 cells as well as boost/carrier samples were analyzed with a nanoPOTS autosampler equipped with a C18 SPE column (100 μm i.d., 4 cm, 300 Å C18 material, Phenomenex) and an LC column (50 μm i.d., 25 cm long, 1.7 μm, 130 Å, Waters) (11). The LC column was heated at 50 °C using an AgileSleeve column heater (Analytical Sales and Services Inc., Flanders, NJ, USA). The home-built autosampler was programmed to sequentially load both single-cell samples from N2 chip and boost/carrier samples from nanoPOTS chip. The loaded single-cell and boost samples were desalted and trapped on the SPE column. The peptides were separated using a 120-min LC gradient from 8% to 45% Buffer B (0.1% FA in ACN) and a 100 nL/min flow rate. Orbitrap Eclipse Tribrid MS with SPS-RTS method was used for the data acquisition. Peptides were ionized by applying a voltage of 2200 V and subjected into an ion transfer tube at 250 °C. Precursor ions within the range from 400 to 1800 m/z were scanned at a 120,000 resolution with an ion injection time (IT) of 50 ms. The precursor automatic gain control (AGC) target was 1E6 and only precursor ions with >2 positive charges and >1E4 intensities were isolated with a window of 0.7 m/z. The fragment ions produced by collision-induced dissociation (CID) level of 35% were scanned in ion trap with maximum injection time of 100 ms and AGC target of 1E4. The same RTS parameters were used as described above. After 10 notches were selected, these ions were subjected to HCD fragmentation with an energy level of 45%. The generated ions were scanned by ion trap with zoom scan mode. The maximum injection time for MS3 scan and the AGC target were 250 ms and 1E5, respectively. The entire cycle time was set to 3 s.

Database searching

All the raw files generated from the MS were processed by FragPipe (Ver. 17.2, build 24) powered by a MSFragger (Ver. 3.5) (42, 43) search engine and a Philosopher toolkit (Ver. 4.2.1) (44) for downstream data processing. The tandem mass spectra search was performed against the UniProtKB protein sequence database of Mus musculus species (downloaded on 02/22/2022 containing 17,084 reviewed protein sequences). The database search was conducted according to the target-decoy search strategy. Due to the difference in resolving power between Orbitrap and ion trap, different mass tolerance parameter for fragment ion and reporter ion were used in the sequence database search. For MS2-based protein quantification method, a MS2 fragment ion mass tolerance was 20 ppm and 0.5 Da for Orbitrap and ion trap analysis, respectively. In case of MS3 analysis, MS2 fragment ion mass tolerance was 0.5 Da for database search. For protein identification, MS3 reporter ion mass tolerance was set to 20 ppm and 0.3 Da for Orbitrap and ion trap, respectively. All other search parameters are common under all conditions: full enzyme digest using trypsin (After KR/−) and Lys-C (After K/−) up to 2 missed cleavages; a precursor ion mass tolerance of 20 ppm (monoisotopic mass); static modifications of 304.207 Da on n-terminal and lysine residue for TMTpro and 57.02 Da on cysteine residue for carbamidomethylation; and variable modifications of 42.01 Da for protein N-term acetylation and 15.99 Da for methionine oxidation. Confidence criteria were set to a false discovery rate (FDR) of less than 1% at both the peptide and protein level.

Statistical analysis and bioinformatics

Perseus (Ver. 1.6.15.0) was utilized for the data processing and statistical analyses (45). For the single cell-level proteomic analysis, protein intensities were log2-transformed and batch-corrected. In LPS-treated single cell analysis, only the proteins that had expression value at least 50% at least one sample group was used for downstream analysis. Sample data were imported into R and outliers were removed based on the median of their respective intensity distributions. Imputation was performed with k-nearest neighbors method (KNN) on proteins with less than 50% missing values across samples from each group (LPS and control). Median normalization was applied prior to TMT batch correction with ComBat (46).

TMT channel effects were considered covariates within each condition and also corrected for with ComBat. A UMAP was embedded into Euclidean space using the first six principal components before visualization with ggplot. R script was provided as supporting document. Two-way t-tests were performed for the pairwise comparison between the control and LPS-treated cellular proteomes utilizing the threshold of Benjamini-Hochberg FDR < 0.05 and S0=0.1. Gene ontology analysis was performed in DAVID bioinformatic tools (database version 6.8, https://david.ncifcrf.gov/) (47).

Results

Study design

We employed Orbitrap Eclipse Tribrid Mass Spectrometer for the study because it has both a linear ion trap (LIT) and an Orbitrap (OT), thus minimizing instrument-associated variations. The Eclipse MS also provided real-time search (RTS) algorithm to allow us to test the performance of RTS-MS3. Because of the resolution of LIT is not sufficient to resolve the C/N pairs of TMTpro report ions, we only chose one variant of TMTpro 16plex tags (Table S1). In details, we generated two sets of TMT-labeled peptides, including a high-input sample and a low-input sample containing 100 pg and 25 pg peptides in each sample channel, respectively. A 5-ng boost/carrier peptides labeled with TMT126 were added in both samples. The use of high and low input samples allowed us to evaluate the impact of LIT and OT detections on typical-size (e.g., cultured cancer cells) (5, 10, 48) and small-size single cells (e.g., cultured immune cells, primary cells) (12, 15).

We performed four different experiments in the evaluation study, including (1) The comparison of LIT and OT for MS2-based scProteomics; (2) The comparison of LIT and OT for MS3-based scProteomics; (3) The impact of AGC level on both MS2 and MS3-level scProteomics quantification; (4) The application of RTS- LIT-MS3 in the study of macrophage activation.

The comparison of LIT and OT for MS2-based analysis.

We first compared the performance of LIT and OT in MS2-based protein identification and quantification. When at least 1 valid value was required in single-cell channels, the number of detected proteins was slightly higher in OT method (Figure 2A and Figure S1) for high input (100 pg) samples, which could be attributed to the higher identification rate using high-resolution MS/MS spectra. For low input (25 pg) samples, the more sensitive LIT method gave slightly higher number of proteins. Encouragingly, when 100% valid values were required, LIT method outperformed OT method in both high and low input samples. The improvement was much greater in low-input samples. The detected protein numbers of low-input samples were increased from an average of 1244 with OT method to an average of 2232 with LIT method, corresponding to ~79.4% improvement (Figure 2B). The enhanced sensitivity of LIT also significantly reduced the missing data for low-input samples. For individual TMT sets, an average of 96.8% of the detected proteins had no missing data in all the sample channels with LIT method, while OT method only retained an average of 51.5% of the detected proteins without missing data (Figure 2C and 2D).

Figure 2. The comparison of LIT and OT for MS2-based scProteomics.

(A) The number of detected proteins in single-cell channels with at least 1 observation and (B) with 100% observations. The percentages of data missingness in single-cell channels for high input and (D) low input samples. (E) Pair-wise Pearson correlations of Log2-transformed protein intensities between OT-OT and OT-LIT methods for high input C10 samples and low input C10 samples. (F) The distributions of protein coefficient of variations (CVs) for C10 samples obtained from OT and LIT method. Protein CVs were calculated among different TMT batches after batch effect corrections. In A to D, data are shown as means with error bars represent standard deviations from triplicates. In F, center lines show the medians.

To assess the quantification performance of LIT and OT methods, we performed pair-wise Pearson correlations of log2-transformed protein intensities among these different methods. For high input C10 samples, high correlation coefficients from 0.96 to 0.99 were observed between the same methods (Figure 2E). However, for low input sample, the correlation was slightly lower of LIT-generated data. The medians of correlation coefficients were 0.94 and 0.96 for LIT and OT methods, respectively, which indicated LIT exhibited lower precision in quantification of low input samples. Although two distinct detectors were used, good agreement in protein quantification with correlation coefficients from 0.92 to 0.94 was observed for high input C10 samples. Similarly, the cross-correlation between OT and IT methods became worse for low-input C10 samples with coefficients ranged from 0.88 to 0.91 (Figure 2E, Figure S2A, and S2B). In addition to correlation analysis, we also calculated the protein coefficient of variations (CVs) (Figure 2F and Figure S2C). For both high and low input samples, significantly higher protein CVs were obtained with LIT methods. The median protein CVs were 10.09% and 15.59% in OT method for high and low input C10 samples, while the corresponding CVs increased to 13.19% and 29.49% in LIT method.

The high technical variations in LIT-based protein quantification were also observed in previous studies (40, 41). One main cause is signal interference from low-mass MS/MS fragments. The low-resolution LIT can not distinguish the interference with reporter ions if the m/z of a fragment is close to that of TMT reporter ions. Indeed, during manual examination of OT MS2 spectra, we occasionally observed such signal interference (Figure S3A). We next estimated the interference levels by averaging all the MS2 or MS3 spectra. As shown in Figure S3B-E, all the single-cell channels were impacted and TMT133 was impacted the most. Because these small peptide fragments were mainly contributed by carrier sample, their impact was more severe for low-input samples, which agreed with our data (Figure S3B and Figure 2F). We also observed reduced signal interference from MS3 spectra (Figure S3D and S3E), which could be attributed to RTS-SPS-based fragment selection, where low-mass fragments were excluded. Additionally, the high variations may be partially attributed to the low copy number of peptide ions and the detection noise in LIT detector. Compared with OT method where average of 1244 proteins were quantified, the LIT method increased the quantifiable protein numbers to average of 2232. We observed that majority of these additionally proteins were low abundant ones.

The comparison of LIT and OT for MS3-based scProteomics

We next evaluated the performance of LIT and OT in MS3-based protein quantification. Real-time search (RTS) in LIT was employed to improve the efficiency of MS3 acquisition and eventually increased the number of quantifiable proteins. Although RTS was used, we still obtained a smaller number of detected and quantified proteins with MS3 method in comparison with MS2 method (Figure 2 and 3). When at least 1 observation was required, an average of 1371 and 1072 proteins were detected in OT method for high and low input samples, respectively, while the use of LIT method improved the coverages to 1518 and 1502 (Figure 3A and Figure S4). When 100% observations were required, LIT significantly outperformed OT method by increased the quantifiable proteins by 28.4% and 114.8% for high and low input samples, respectively (Figure 3B). In term of data completeness, an average of 79.6% proteins had no missing data for low-input samples when LIT was used to collect MS3 spectra, while only 39.1% proteins had no missing data in OT-based MS3 acquisition method (Figure 3C and 3D).

Figure 3. The comparison of LIT and OT for MS3-based scProteomics.

(A) The number of detected proteins in single-cell channels with at least 1 observation and (B) with 100% observations. The percentages of data missingness in single-cell channels for (C) high input and (D) low input samples. (E) Pair-wise Pearson correlations of Log2-transformed protein intensities between OT-LIT-OT and OT-LIT-LIT methods for high input C10 samples and low input C10 samples. (F) The distributions of protein coefficient of variations (CVs) for C10 samples obtained from OT and LIT method. Protein CVs were calculated among different TMT batches after batch effect corrections. (G) The measured fold changes between C10 and Raw cells based on MS2 and MS3 quantification methods. (H) The volcano plots showing the differentially abundant proteins between C10 and Raw cells using MS2 and (I) MS3-based protein quantification method. For plot G,H,I, low input (25 pg) samples and LIT detection method were used. In A to D, data are shown as means with error bars represent standard deviations from triplicates. In F, center lines show the medians.

Pair-wise Pearson correlations of protein intensities showed similar distributions with MS2-based quantification methods (Figure 3E, Figure S5). For high input C10 samples, the correlation coefficients were in the range of 0.96 to 0.99 between the same MS methods (OT or LIT) and in the range of 0.91 to 0.94 across OT and LIT methods. The correlation coefficients became lower for low-input samples, which were in the range of 0.92 to 0.97 between the same MS methods (OT or LIT) and in the range of 0.87 to 0.91 across OT and LIT methods. The protein CVs for both high and input samples were also comparable with the data obtained with MS2-based quantification methods (Figure 3F and Figure S5C). These MS3 quantification data supported our conclusion that the technical variations in LIT-based protein quantification was from low-mass fragment interference.

MS3 improves the dynamic ranges of protein quantification

The main motivation of using MS3 for scProteomics analysis is to reduce ratio compression and improve dynamic range of quantification. We first compared protein fold changes between C10 and RAW cells obtained from LIT-based MS2 and MS3 methods. As shown in Figure 3G, large log2-fold changes with an average range from −3.8 to 4.0 were obtained with MS3 method, which is more than 99% higher than that obtained from MS2 method with an average range from −3.1 to 3.6. The improved dynamic range of MS3 method also allowed us to detect larger number of differentially abundant proteins (Figure 3H and 3I). Although the total quantifiable number of proteins were 45.9% lower in MS3 method, the number of differentially abundant proteins were 37% higher than MS2 method.

The impact of AGC level on LIT-based scProteomics

Many studies have suggested the MS acquisition parameters (AGC target level and ion injection time) need to be well adjusted to improve the sampling efficiency of single cell ions (17, 21, 23). Most of these evaluation studies were performed with Orbitrap detectors. Therefore, we next assess the impact of two different level of AGC target settings (3E4 and 1E5) on both the proteome coverages and quantitative reproducibility when an LIT detector was used. As shown in Figure 4A and Figure S6A, we observed slightly higher numbers of total identified proteins with low AGC target setting for both MS2 and MS3 acquisition approaches, which can be attributed to the increased MS duty cycles. When 100% valid values were required for all the single-cell channels of each TMT dataset, the average protein numbers were comparable between two AGC target settings for MS2-based acquisition approach (Figure S6B).

Figure 4. The impact of AGC level on protein identification and quantification in MS3 analysis.

The effect of two AGC level (3E4 and 1E5) on protein identification (A and B) and quantitative reproducibility (C) for OT-LIT-RTS-LIT MS3 analysis. Protein CVs for C10 samples were calculated among different TMT batches after batch effect corrections. In A to B, data are shown as means with error bars represent standard deviations from triplicates. In C, center lines show the medians.

However, significantly higher protein numbers were observed at high AGC target for MS3-based acquisition approach, especially for low input samples (Figure 4A). For example, the average protein numbers were increased from 928 to 1207 when the AGC target levels increased from 3E4 to 1E5 for low input peptides, which can be ascribed to increased ion sampling efficiency with high AGC target setting (17, 23).

We next compared the quantitative performance between high and low AGC target setting. In the case of MS2 analysis, there was no significant difference in quantitative reproducibility between the two AGC target settings (Figure S6 and S7). In the case of MS3 analysis, the median CVs decreased from 25.9 to 17.3 and from 30.2 to 29.3 for the high input and low input sample, respectively (Figure 4C). This improvement in reproducibility was also observed in the RAW264.7 samples (Figure S7A).

Optimal experimental setting for LIT-based multiplexed scProteomics

The above evaluation demonstrated significantly improved sensitivity when LIT was used for multiplexed scProteomics, especially for MS3-based quantitation. The use of RTS-LIT-MS3 allowed for efficiently mitigating ratio compression and improving the accuracy of protein quantification. Our evaluation also validated high AGC level at 1E5 should be used to improve ion sampling efficiency of multiplexed scProteomics, where a high-amount carrier peptides were commonly spiked in each TMT set. The evaluation also revealed several technical limitations when employing LIT for multiplexed scProteomics. Due to its low revolution, LIT can not efficiently resolve low-mass fragment interferences with reporter ions. These interferences could significantly reduce the accuracy and precision of protein quantification, as well as increase false discovery rate for protein identification. Because low-mass fragments were mainly contributed by 50× or 200× carrier peptides, the most efficiently way to reduce their interference was to use minimal amount of carrier. We identified the 133 channel of TMTpro was most impacted by the fragment and thus this channel should be avoided to use for single cell labelling (Figure S3).

Together, we suggest to use TMTpro channels including 126, 127, 128, 129, 130, 131, and 132 for labelling single cells and channel 134 for labelling carrier peptides. To minimize the low-mass interference, the carrier-sample ratio should be lower than 50. Additionally, a high-level AGC setting of 1E5 should be used for LIT-based MS3 data acquisition.

Application to study immune activation of single macrophage cells

To demonstrate the RTS-LIT-MS3 method for scProteomics, we applied it to study the immune activation of single macrophage cells. We chose RAW 264.7 as a model cell line and employed LPS treatment to activate immune response, which is verified by NO assay (Figure S8). Compared with typical somatic cell lines (HEK293 or HeLa), RAW 264.7 has relatively smaller cell size and lower protein content (15, 22), which we reasoned as a suitable model to evaluate the sensitive LIT detection method. As shown in Figure 5A, we analyzed total 168 single cells containing 72 untreated cells (control) and 96 LPS-treated cells. After cell sorting, we employed a 9-well nested nanoPOTS (N2) chip to perform cell lysis, protein reduction, alkylation, digestion, and TMT labelling. To minimize the signal interference of low-mass fragments, a mixed pool of only 50 RAW cells (25 control and 25 LPS-treated cells) was used as the carrier proteome.

Figure 5. ScProteomics study of macrophage activation with RTS-LIT-MS3-based data acquisition approach.

(A) Illustration of scProteomics workflow including cell culture, LPS treatment, cellenONE-based single-cell isolation, sample preparation in nested nanoPOTS chip, and RTS-LIT-MS3 analysis. (B) The number of detected proteins in single cells with and without LPS treatment. Center lines show the medians; box limits indicate the 25th and 75th percentiles; whiskers indicate the minimum and the maximum value. (C) The number of quantifiable proteins at different percentages of valid values across all single cells. (D) Two-dimensional UMAP projection showing the clustering of single cells based on cell populations (control and LPS treated). (E) Heatmap with hierarchical clustering showing 210 significant proteins between the two cell populations based on two-way T-test. (F) Violin plots showing the differential distribution of protein abundances. *p <0.05, ***p < 0.001, ****p < 0.0001, by two-tailed T-test. Center lines show the medians.

Our analysis identified total 1941 proteins in 24 TMT sets, of which 1905 proteins had at least one valid value across 168 single cells. In average, 1185 and 1191 proteins were identified from control and LPS-treated single cells, respectively (Figure 5B). When more stringent criteria of > 50% and >75% valid values were used, the quantifiable protein number were 1263 and 1064, respectively (Figure 5C). The proteome coverages were ~30% lower than our previously reported scProteomics analysis using MS2 approach (17, 22), which can be mainly attributed to reduced duty cycle of MS3 approach. Despite this, the capability to consistently quantify >1000 proteins in single cells verified the great sensitivity of LIT-MS3 method for scProteomics.

To evaluate the quantitative performance of the RTS-LIT-MS3-based scProteomics, we first performed a dimensional reduction analysis based on UMAP. As shown in Figure 5D, the two cell populations can be clearly segregated and each population was clustered together based on total 1263 protein abundance information (50% valid values across single cells). Next, we performed a two-way T-test with an FDR cut-off of 0.05 to identify the differentially abundant proteins between the two cell populations (Figure 5E). Total 210 proteins were identified as significantly regulated proteins, including 118 upregulated proteins and 92 downregulated proteins in LPS-treated cells. As expected, functional enrichment analysis indicated upregulated proteins were mainly involved in Intrinsic apoptotic signaling pathway (P = 1E-3), Positive regulation of response to stimulus (P = 8E-3), and response to ROS (P = 1E-2). For the downregulated proteins, the enriched pathways were Regulation of cell death (P = 9E-8), Regulation of cell adhesion (P = 7E-5), and Endocytosis (P = 1E-4).

In addition to enrichment analysis, our analysis also revealed key proteins involved in immune activation process (Figure 5F). For example, Aconitate decarboxylase 1 (Acod1) is a protein that has recently gained interest in immunometabolism research. This protein was reported to be up-regulated in LPS-treated immune cells in many previous studies (49, 50). Superoxide Dismutase 2 (Sod2) is a well-known protein that play a role in scavenging the cytotoxic consequences of reactive oxygen species (ROS) (51). As shown in Figure 5F, LPS-treated cells showed increased expression of Sod2. Although the exact mechanism of Sod2 is not well discovered, our result suggests that Sod2 play a role in cell survival as a reaction to oxidative stress. In addition, the BH3 interacting domain death agonist (Bid) is known to have a pro-apoptotic function (52) and its increased expression was observed in LPS-treated cells. Next, the increased level of Sequestosome 1 (Sqstm1), which is known for its association with autophagy-mediated apoptosis (53), was also observed. Interestingly, Heat shock protein 1, also called chaperonin, is known to play complex roles depending on cell type in apoptosis (54, 55). It was observed to be up-regulated in LPS-treated macrophages. The increased abundance in these proteins observed at the single cell level suggests that LPS induces apoptosis in macrophages and several apoptotic markers can be detected.

Additionally, Annexin A1 (Anxa1) is known to be involved in the reprogramming of macrophages to maintain homeostasis when they undergo inflammatory responses (56). The association of Anxa1 with LPS-induced cellular inflammatory response or apoptosis should be further explored. In summary, we were able to identify proteins that played major roles in immune response or apoptosis in LPS-treated single cells, suggesting that LIT-MS3-based scProteomics can help finding novel marker candidates.

DISCUSSION

In this study, we quantitatively compared the performance of LIT and OT for multiplexed scProteomics based on MS2 and MS3 protein quantification methods. We demonstrated LIT can improve scProteomics sensitivity to allow for the detection of low abundant proteins with reduced missing data. In particular, the use of LIT for RTS MS3 acquisition can significantly improve the number of quantifiable proteins by 28.4% and 114.8% for high and low-input samples, respectively. The developed RTS-LIT-MS3 method was applied to study immune response of macrophage cells under a treatment of LPS. Among 168 single cells, ~1000 proteins can be quantified when 70% non-missing data was required. The high proteome coverage allowed us to identify pathways and protein markers during macrophage activation process. With the increased detection sensitivity and quantification dynamic range, we expect the RTS-LIT-MS3 method can be broadly applied to study subtle changes of single-cell proteomes, such as drug treatment, cell development, and cell cycle.

Our study also revealed the limitation of low-resolution LIT for multiplexed scProteomics. In addition to the reduced multiplexities, the TMT reporter ions can be greatly contaminated by low-mass fragments, leading to high signal variation and false discovery. After comparing the signal interference at high and low-input samples, where the carrier-sample ratios were 50 and 200 respectively, we concluded the interference fragments were mainly originated from the high amount of carrier samples. We suggested a low carrier-sample ratio should be used to reduce the interference and minimize its impact in protein quantification. Currently, we used a 50× carrier to obtain good proteome coverage in the RTS-based protein identification and quantification. We suggest the carrier-sample ratio should be further reduced to <10 with the sensitivity improvement of RTS algorithm. In addition to sensitivity and quantification accuracy, other technical improvements such as analysis throughput could be achieved by using higher multiplex isobaric reagents (57) or combined precursor isotopic labeling and isobaric tagging (cPILOT) approach (58, 59).

DATA AVAILABILITY

Mass spectrometry raw data and FragPipe output tables have been deposited to the ProteomeXchange Consortium via the MassIVE partner repository with the dataset identifier MSV000090178 or ftp://massive.ucsd.edu/ (User name: MSV000090178; Password: Trap4251).