Evaluation of Protein Identification and Quantification by diaPASEF Method on timsTOF SCP

Abstract

Accurate and precise quantification is crucial in modern proteomics, particularly in the context of exploring low amount samples. While the innovative 4D-data independent acquisition (DIA) quantitative proteomics facilitated by timsTOF mass spectrometers gives enhanced sensitivity and selectivity for protein identification, the diaPASEF (parallel accumulation-serial fragmentation combined with data-independent acquisition) parameters have not been systematically optimized, and a comprehensive evaluation of quantification is currently lacking. In this study, we conducted a thorough optimization of key parameters on timsTOF SCP, including sample loading amount (50 ng), ramp/accumulation time (140 ms), isolation window width (20 m/z), and gradient time (60 min). To further improve the identification of proteins in low amount samples, we utilized a column with reduced length, inner diameter and particle size of packing materials combined with the introduction of 0.02% of n-Dodecyl-β-D-maltoside (DDM) in sample reconstitution solution, resulting in a remarkable 19-fold increase in protein identification at the single-cell equivalent level. Moreover, a comprehensive comparison of protein quantification using tandem mass tag (TMT)-reporter, complement TMT ions (TMTc), and diaPASEF revealed a strong correlation between these methods. Both diaPASEF and TMTc have effectively addressed the issue of ratio compression, highlighting the diaPASEF method’s effectiveness in achieving accurate quantification data compared to TMT reporter quantification. Additionally, an in-depth analysis of in-group variation positioned diaPASEF between TMT-reporter and TMTc methods. Therefore, diaPASEF quantification on the timsTOF SCP emerges as a precise and accurate methodology for quantitative proteomics, especially for samples with a small quantity.

Introduction

Mass spectrometry (MS) – based proteomics has emerged as a widely employed platform for proteome profiling in recent decades. The success of proteomics studies is evaluated not only by the total number of identified proteins, but also by the quality of quantitative information,^1–2 often with more than 10,000 proteins identified within a single day.^3–5

Two predominant data acquisition approaches, data-dependent acquisition (DDA) and data-independent acquisition (DIA) are extensively utilized in MS-based proteomics studies.^6–8 In the DDA approach, the top N most abundant precursor ions detected in an MS1 survey scan are sequentially chosen for MS2-based peptide fragmentation and protein identification.^9–10 Tandem Mass Tag (TMT) is a prevalent isobaric chemical labeling strategy commonly employed in conjunction with the DDA approach.^11–13 This labeling strategy offers several advantages over label-free quantification, such as increased throughput,^14–15 enhanced MS1 signal for low abundance peptides,¹⁶ and improved precision and accuracy of quantification.¹⁷ Following the MS data acquisition of TMT-labeled peptides with DDA method, either TMT-reporter or complement TMT ions (TMTc) can be used for protein quantification.^18–19 While TMT-reporter quantification allows the quantification of more proteins, TMTc quantification provides more accurate results due to the alleviation of the ratio compression generated by the co-eluting peptides occurring mostly in TMT-reporter quantification.^20–22

In contrast to the DDA strategy, DIA approach obtains fragment ion information for all peptides ions within a specific mass-to-charge ratio (m/z) window, sequentially covering the entire relevant mass range.^23–24 DIA method gained more popularity with the introduction of SWATH (sequential window acquisition of all theoretical mass spectra) workflows in the mid-2000s.^25–28 Label-free quantification (LFQ) is commonly employed in DIA method due to its experimental simplicity but its inherently incompatible with isobaric labeling quantification, such as TMT.^29–30 While reducing sample/spectral complexity is essential to enhance protein identifications, it is unsuitable in DIA-LFQ method to achieve this by increased sample fractionation as applied in DDA-TMT method.^31–32 However, the feasibility of peptide separation based on distinct collisional cross-sections has been realized with the introduction of various ion mobility spectrometers, such as drift-tube ion mobility spectrometry (DTIMS), travelling-wave ion mobility spectrometry (TWIMS), field-asymmetric ion mobility spectrometry (FAIMS) and trapped ion mobility (TIMS).^33–35 With embedded TIMS, the timsTOF Pro mass spectrometry shows significantly improved sensitivity and throughput in peptide identification and is able to detect over 4,000 proteins from 10 ng HeLa peptides in diaPASEF method.³⁶

Moreover, the detection sensitivity has been further improved recently on timsTOF SCP mass spectrometry with the use of a brighter ion source which enables more ions entering the analyzer.³⁷ When loading 1 ng of HeLa peptides onto the timsTOF SCP, Mann’s group identified over 550 proteins in DDA method and over 3,000 proteins in DIA method combined with low flow rate chromatography.³⁷ Pandey’s group further optimized the parameters of DDA-PASEF method and was able to detect 1,116 proteins from 10 sorted human primary T cells.³⁸ The timsTOF SCP serves as an excellent platform for proteomics research involving limited sample material. However, to the best of our knowledge, there has been no thorough parameter optimization study for diaPASEF method, and only a limited number of studies have been focused on enhancing the sensitivity of timsTOF SCP and the evaluation of protein quantification when data acquired in diaPASEF method.

In this study, we systematically optimized the parameters for the diaPASEF approach on timsTOF SCP. These parameters included sample loading amount, ramp/accumulation time, isolation window width and gradient time. Subsequently, with low amount sample input, we found that reducing column length, inner diameter and particle size of packeting materials and adding n-Dodecyl-β-D-maltoside (DDM) in sample reconstitution solution further increased protein identification. Specifically, 1,567, 2,590 and 3,868 proteins were identified from 0.1, 0.3 and 1 ng HeLa peptides respectively. Furthermore, using nine human brain samples, we compared the diaPASEF quantification results with TMT-reporter and TMTc quantification results. The quantification correlation was high, and diaPASEF demonstrated similar quantification accuracy to TMTc without obvious ratio compression.

Methods

Sample preparation.

The HeLa protein digest standard (Pierce, Cat. 88328) was used for the optimization of timsTOF SCP. Human postmortem brain tissue samples provided by Brain and Body Donation Program at Banner Sun Health Research were used for the comparison of DDA and diaPASEF methods.³⁹ The study included 5 cases of Alzheimer’s disease (AD) and 4 controls without dementia (Table S1). The brain samples were lysed in a denaturing buffer (50 mM HEPES, pH 8.5, 8 M urea, 0.5% sodium deoxycholate, 10 μL buffer per mg tissue) with a Bullet Blender.⁴⁰ The protein concentration was determined by Pierce BCA Assay. Four mg proteins from each sample were proteolyzed with Lys-C (Wako, 1:100, w/w) for 3 h at 21 °C, diluted 4-fold with 50 mM HEPES and further digested by trypsin (Promega, 1:50, w/w) overnight at 21 °C followed by Cys reduction and alkylation.^41–42The digestion was terminated by the addition of trifluoroacetic acid to 1%. After centrifugation, the supernatant from each of the 9 samples was desalted and then divided into two aliquots. One aliquot was dried and used as label-free samples. The other aliquot was labeled with TMTpro reagents (Thermo Fisher Scientific) respectively (TMTpro: 127C, 128C, 129C, 130C, 131N, 131C, 132C, 134C and 135N) to comply with 9 channel limitation of TMTc quantification. All nine TMT labeled samples were equally pooled, desalted and fractionated by an offline basic pH reversed-phase liquid chromatography (RPLC) on an Acquity BEH C18 column (3.0 × 150 mm, 1.9 μm particle size, Waters) with a 15%−45% buffer B gradient in 120 min (buffer A: 10 mM ammonium formate, pH 8; buffer B: buffer A plus 95% acetonitrile). A total of 80 fractions were collected and dried. The lysis, digestion and desalting procedures for the preparation of yeast peptide used for label-free analysis were similar as the preparation of human brain samples, and 3 ng and 10 ng of yeast peptides were spiked in 25 ng HeLa peptides, respectively.

LC-MS/MS analysis with diaPASEF method.

Different injection amounts and gradients were tested to achieve optimized results. Initially, HeLa peptides (≥ 1 ng) were separated on a C18 column (15 cm x 75 μm, 1.9 μm particle size, P/N number 1842621, Bruker Daltonics, heated at 55 °C) using a nanoElute 2 liquid chromatography system (Bruker Daltonics, Bremen, Germany) by 5%−26% buffer B gradient in 30 min (buffer A: 0.1% formic acid in water, buffer B: 0.1% formic acid in acetonitrile, flow rate of 0.25 μL/min). All the gradients examined during optimization were listed in Table S2. For low amount loading (≤ 1 ng), a C18 column (10 cm x 50 μm, 1.5 μm particle size, P/N number 1895802, Bruker Daltonics, heated at 55 °C) was used to enhance the sensitivity with a 5%−26% buffer B gradient in 30 min while the gradient was extended to 1 h (4%−24% buffer B) for the label-free 9 human brain samples. The separated peptides were ionized by a CaptiveSpray nano-electrospray ion source, introduced into the timsTOF SCP mass spectrometer (Bruker Daltonics, Bremen, Germany) and analyzed by a diaPASEF approach. The singly charged precursor ions were excluded with the polygon filter. The capillary voltage was set to 1500 V and the dry gas was at 3.0 L/min with a dry temperature of 200 °C. While different isolation window widths (10, 15, 20, 25 and 30 m/z) and ramp times (100, 120, 140, 160, 166, 180 and 200 ms with 100% duty cycle) were examined, the mass range was set as 100–1700 m/z and ion mobility range was 0.70–1.30 1/K0. The collision induced dissociation energies were linearly ramped as a function of ion mobility ranging from 20 eV (1/K0=0.6 Vs cm-2) to 59 eV (1/K0=1.6 Vs cm-2). The diaPASEF isolation window tables were listed in Table S3 to show the detailed method.

The separated peptides were ionized using a captive spray source and introduced into the timsTOF SCP mass spectrometer (Bruker Daltonics, Bremen, Germany). The diaPASEF data was acquired with singly charged precursor ions excluding with the polygon filter. The capillary voltage was set to 1500 V and the dry gas was at 3.0 L/min with a dry temperature of 200 °C. While different isolation window widths and ramp times (100% duty cycle) were examined, the mass range was set as 100–1700 m/z and ion mobility range of 0.70–1.30 1/K0. The collision energy was linearly ramped as a function of ion mobility ranging from 20 eV (1/K0=0.6 Vs cm⁻²) to 59 eV (1/K0=1.6 Vs cm⁻²).

LC-MS/MS analysis with DDA method.

Peptides from each fraction of the TMT labeled human brain samples were separated on a C18 column (20 cm x 75 μm, 1.7 μm particle size, Catalog number HEB07502001718I, CoANN technology, heated at 65 °C) over a 30 min gradient from 15%−36% buffer B (buffer A: 0.2% formic acid, 3% DMSO; buffer B: buffer A plus 67% acetonitrile, flow rate of 0.25 μL/min). Liquid chromatography was coupled to Orbitrap Exploris 480 (Thermo Fisher Scientific) for the analysis of TMT labeled peptides in a data-dependent acquisition (DDA) method. The MS settings included MS1 scans (60,000 resolution, 500–1100 m/z scan range, 1 × 10⁶ AGC, and 50 ms maximal ion time) and 20 data-dependent MS2 scans (60,000 resolution, starting from 120 m/z, 1 × 10⁵ AGC, 110 ms maximal ion time, 1.0 m/z isolation window with 0.2 m/z offset, and 34% HCD collision energy).

Protein identification and quantification of diaPASEF data.

The raw timsTOF data (. d folders) were imported into DIA-NN (version 1.8) and searched against in silico library. To generate the in silico spectral library, human (proteome ID: UP000005640) and yeast (proteome ID: UP000002311) protein sequences were downloaded from UniProt database. The protein FASTA files were then used for in silico library generation with the following settings: Trypsin/P with maximum 2 missed cleavage; Carbamidomethylation on Cysteine as fixed modification; Oxidation on Methionine as variable modification; maximum number of variable modifications set to 2; peptide length from 7 to 30; precursor charge 1–4; precursor m/z from 300 to 1,800; fragment m/z from 200 to 1,800. The search parameters of DIA-NN were set as follows: precursor FDR 1%; mass accuracy at MS1 and MS2 both set as automatic inference; scan window set to 0; isotopologues and MBR turned on; no shared spectra enabled; protein inference at gene level; heuristic protein inference enabled; quantification strategy set to Robust LC (high precision); neural network classifier single-pass mode; cross-run normalization turned off; library generation set to smart profiling; speed and RAM usage set to optimal results. The search results were further filtered with precursor q value <0.01 at the library level and protein group q value <0.01. For quantification, the precursor quantities were first obtained by DIA-NN through summing the intensities of the top six fragments (ranked by their library intensities) for each precursor. Precursors corresponding to unique proteins were then used for protein-level quantification and the intensities of protein groups were obtained using the MaxLFQ algorithm implemented in the iq r package.⁴³ Cross-run normalization was then performed based on the median protein intensity of each sample.

Protein identification and quantification of DDA data.

DDA data analyses were performed by our JUMP software suite.^44–45 Briefly, the acquired MS/MS spectra were converted into mzXML format and searched against a target/decoy protein database to estimate the false discovery rate (FDR).⁴⁵ The target database were generated by combining Swiss-Prot, TrEMBL and UCSC databases with redundancy removed (Human: 83,955 entries) while the decoy database consisted of the reversed target protein sequences. The search was performed with 15 ppm mass tolerance for precursor, 20 ppm for product ions, full trypticity, and two maximal missed cleavages. TMTpro labeling on Lys residues and peptide N-termini (+304.20715 Da) and the carbamidomethylation of Cys residues (+57.02146 Da) were set as static modifications and Met oxidation (+15.99492 Da) was used as a dynamic modification. The resulting PSMs were filtered by mass accuracy, minimal peptide length, and then grouped by precursor ion charge state followed by the cutoffs of JUMP-based matching scores (Jscore and ΔJn) to reduce FDR below 1% for proteins. When the same peptide was derived from numerous homologous proteins, the peptide was typically assigned to the canonical protein form first according to the rule of parsimony. If no canonical protein form was available, the peptide was assigned to the protein with the highest PSM number. For each PSM, TMT reporter ion- and complement (TMTc) ion-based quantification was performed respectively. PSMs that exhibit zero intensity in any of the reporter channels were excluded in reporter ion-based quantification. For TMTc-based quantification, which employs an optimization method to estimate the best possible ratios among channels, PSMs with a minimum difference (minDiff) value of less than 0.005 were kept for further analysis.¹⁸ The resulting PSM-level intensities were then transformed to log2-scale to stabilize variances. To correct the loading bias among samples, the mean centered PSM intensities across samples were calculated and converted to a ratio used for normalization. The protein quantification was obtained by first summarizing the relative rations of all PSMs corresponding to a protein and then multiplying this ratio by the average intensities of the three most abundant PSMs for that protein.^{22, 46}

Results and Discussion

Initial LC-MS/MS parameter optimization for timsTOF SCP.

To explore the optimal condition for protein identifications on timsTOF SCP instrument, we fine-tuned multiple LC-MS parameters using HeLa tryptic peptides. We first investigated the impact of loading amount on protein identification using the default LC-MS/MS method. Titration of HeLa peptides with three replicates from 1 ng to 50 ng resulted in an increase in identified precursors/proteins from 3,388/888 to 53,381/5,867, respectively. A further increase of loading amount to 100 ng resulted in a decrease of identified precursors to 47,638 and a slight increase in identified proteins to 5,966. The titration results suggest that the LC-MS/MS system was saturated around the point of 50 ng (Figure 1A). Secondly, we explored the ramp time (the same as accumulation time with duty cycle of 100%) of TIMS which played an important role in ion separation concerning mobility dimension, MS sensitivity and cycle time. Starting with the default settings of 166 ms, we examined the ramp time from 100 ms to 200 ms in a 20 ms increment (Figure 1B). Except 200 ms, the difference in the count of identified precursors and proteins was always less than 10%. Ultimately, 140 ms was determined to be the optimal setting and used as the default value for the following studies since it allowed for the identification of the highest number of proteins.

Figure 1. Initial optimization of loading amount, gradient time, isolation window width and ramp time for the identification of proteins using HeLa cell digest (n = 3).

The default parameters were configured as follows: 50ng loading amount, 30min gradient time, 166ms ramp time and a 30 m/z isolation window width. The percentage of identified precursors and proteins were assessed under varying conditions, including (A) loading amount, (B) ramp time of Tims, (C) isolation window width of DIA-PASEF and (D) gradient time length of liquid chromatography. 100% equivalents to 53,381 precursors and 5,966 proteins in Fig 1A, 56,508 precursors and 5,771 proteins in Fig 1B, 58,363 precursors and 5,717 proteins in Fig 1C and 66,113 precursors and 6,145 proteins in Fig 1D, respectively.

Furthermore, the isolation window width which affected both the cycle time and spectral complexity was inspected starting with a relatively broad range (20–100 m/z) and 20 m/z resulted in the highest number of the identified proteins (Figure S1).⁴⁷ However, an additional isolation window width evaluation around 20 m/z (10–30 m/z) with a step of 5 m/z did not significantly contribute to an increase in the number of identified precursors and proteins. As shown in Figure 1C, the variation in identification remained below 10%. With the consideration of the cycle time and identification number, the isolation window width was fixed at 20 m/z. Lastly, the effect of gradient time on protein identification was assessed considering its influence on the peptide separation.^48–49 Gradient time of 8, 15, 30, 60 and 120 min were used as shown in Figure 1D. Numbers of identifiable precursors and proteins reached a plateau at the 60 min gradient, suggesting 60 min is an optimal gradient time for maximum protein identification when other parameters were optimized as mentioned above. Thus, our recommended settings included loading amount of 50 ng, ramp/accumulation time of 140 ms, isolation window width of 20 m/z and gradient time of 60 min. For the optimized method, there was a high correlation among three replicates with an average R² as 0.85 (Figure S2), indicating good quantification reproducibility.

Impact of column parameters and n-Dodecyl-β-D-maltoside on increasing sensitivity.

TimsTOF SCP was primarily designed for prioritizing high sensitivity to address the challenges in the analysis of low amount samples, especially at single cell level.³⁷ Following initial LC-MS/MS parameters optimization, the HeLa peptides was used to investigate methods for increasing sensitivity at low amount sample by utilizing a column with reducing the column length, inner diameter and particle size of packing materials⁴⁸ and incorporating DDM^50–51 in the sample reconstitution buffer (Figure 2).

Figure 2. The impact of column parameters and n-Dodecyl-β-D-maltoside (DDM) on timsTOF SCP sensitivity.

The default parameters were configured as follows: 30min gradient time, 166ms ramp time and a 20 m/z isolation window. (A) Percentage of identified precursors and (B) percentage of identified proteins with different loading amount of HeLa cell digest (n = 3). 100% equivalents to 33,434 precursors in Fig 2A and 3,868 proteins in Fig 2B.

Different low amounts of HeLa peptide digest suspended in 5% formic acid (0.1, 0.3 and 1 ng) were injected into a C18 column (50 μm x 10 cm, 1.5 μm, Bruker Daltonics) to assess the impact of different column on sensitivity. As expected, the identified precursors increased from 498 to 550 for 0.1 ng loading, 1,128 to 3,847 for 0.3 ng loading and 4,992 to 13,062 for 1 ng loading when compared with the previous C18 column (75 μm x 15 cm, 1.9 μm, Bruker Daltonics). Similarly, the identified proteins increased from 0 to 178, 135 to 771 and 937 to 2,364 for the loading amount of 0.1, 0.3 and 1 ng, respectively. For a 75 μm x 15 cm column, the average full width at half maximum (FWMH) was 0.09 min, with the highest pressures reaching 311 bar for pump A and 275 bar for pump B. In contrast, for a 50 μm x 10 cm column, the FWMH was 0.07 min, with the highest pressures reaching 547 bar for pump A and 502 bar for pump B. Clearly, the column with reduced length, inner diameter and particle size of packing contributed to enhancing the sensitivity of low amount peptide injections. Small columns have the potential to increase peak intensity by reducing peak broadening and absorption, which is especially crucial for low amount loading in protein identification. This advantage could outweigh the benefits of improved peak resolution by a longer column. Moreover, opting for a longer or smaller inner diameter column may further enhance protein identification, but it comes with the trade-off of increased system pressure, potentially compromising system robustness. Using the 50 μm x 10 cm C18 column, the flow rate could be set at 0.25 μL/min while still maintaining robustness of the system. In addition, the smaller particle size of packing material may also contribute to increased identification by increasing resolution efficiency.

DDM, an MS-compatible nonionic surfactant, can substantially decrease sample loss without affecting peptide detection.⁵⁰ With 0.02% of DDM in the sample loading buffer, the identified precursors continued to increase to 9,130, 18,240 and 33,434 for the loading amount as 0.1 ,0.3 to 1 ng while identified proteins reached 1,567, 2,590 and 3,868, respectively. Across three replicates, the standard error of the identification count was consistently less than 0.7% for all the low amount injections of HeLa peptides. Through the optimization of column setting and the inclusion of DDM, the number of identified proteins ultimately increased by 19-fold at the single-cell equivalent level. DDM was also included to investigate whether there would be a continued increase in protein identification with higher loading amounts. When the loading amount reached 30 ng, the addition of DDM did not result in a significant further increase in protein identification (Figure S3). To further optimize the diaPASEF method for trace samples, 0.3 ng Hela peptides were injected to optimize the gradient time, ramp time and isolation window width. Figure S4 illustrated that 30 minutes gradient time and 20 m/z isolation window width remained optimal, while ramp time didn’t significantly impact the identification result.

Identification and quantification comparison among TMT-reporter, TMTc and diaPASEF methods.

To better understand and evaluate the diaPASEF data generated from timsTOF SCP, we sought to compare the diaPASEF identification and quantification results with DDA data acquired from the commonly used Orbitrap Exploris 480 instrument. The summarized workflow of the sample preparation and data acquiring steps was illustrated in Figure 3A. Approximately 300 ng of peptides from each fractionation were injected for DDA data acquisition, whereas 50 ng of peptides were injected for each individual DIA run. We compared the proteome of post-mortem human brain tissues from five Alzheimer’s disease (AD) brains and four control brains using TMT DDA LC/LC-MS/MS and label free DIA LC-MS/MS methods. In the TMT method, only 9 channels were used to simplify both TMT-reporter and TMTc quantification. From the DDA analysis, a total of 9,677 and 8,994 proteins were quantified by TMT-reporter and TMTc respectively from 80 fractions. In contrast, the DIA analysis on timsTOF SCP (three replicates/sample) had 7,108 proteins quantified with 546 of them excluded due to missing values. Eventually, 5,958 proteins were quantified as overlapping among 3 quantification methods (Figure 3B) and the quantification data was listed in Table S4. Given that another dimension of separation was applied in DDA method (fractionation with basic pH HPLC), it was reasonable that DDA acquired a deeper proteome compared with diaPASEF. If the instrument time was held constant, the diaPASEF quantification showed the advantage by quantifying more proteins (Figure S5). For all the 5,958 overlapping proteins, the protein fold changes (FC) between AD and control groups were calculated and transformed using log 2. Following Gaussian curve fitting, three probability density distribution curves were represented in Figure 3C, revealing that the standard deviation (SD) was similar between diaPASEF and TMTc quantification methods, but larger than TMT-reporter quantification method. This result indicated the ratio compression of TMT-reporter was much larger than that of TMTc and diaPASEF, which exhibited similar patterns. To illustrate the correlation between protein abundance and FC ratios, the distribution curves of FC ratios were depicted along with the protein abundance after removing outliers with log₂FC > 6 in these 3 methods. With increasing protein abundance, the protein FC from TMT-reporter remained consistent, and that was reasonable as ratio compression existed in TMT-reporter quantification which would reduce the FC ratios (Figure 3D).²⁰ In comparison to TMT-reporter, the ratio compression observed in TMTc quantification was much less, yet still relatively consistent (Figure 3E). This suggests the FC ratios were not significantly affected by the protein abundance for TMTc quantification. In Figure 3F, the FC ratios were larger when the protein abundance was low and much smaller for proteins with high abundance. Interestingly, the trend in diaPASEF quantification showed more similarity with TMTc when the protein abundance was low but was more akin to TMT-reporter when the protein abundance was high, which indicated the slight ratio compression of diaPASEF may be caused by the saturation of the high abundance proteins.

Figure 3. Protein quantification by TMT-reporter, TMTc and diaPASEF.

(A) Schematic diagram of sample preparation, data acquisition, identification, and quantification. (B) The overlapped proteins quantified by TMT-reporter, TMTc and diaPASEF quantification methods. (C) The probability density distribution of protein fold changes (AD/CTL) quantified in TMT-reporter, TMTc and diaPASEF quantification methods (n = 5958). (D) The protein fold changes (AD/CTL) across protein abundance distributions by TMT-reporter quantification (n = 5958). (E) The protein fold changes (AD/CTL) across protein abundance distributions by TMTc quantification (n = 5958). (F) The protein fold changes (AD/CTL) across protein abundance distributions by diaPASEF quantification (n = 5958).

To better compare the ratio compression and the quantification accuracy among the 3 methods, linear regression analysis was applied to the overlapped proteins (Figure 4A and 4B). With the linear regression analysis of 5,958 proteins, TMT-reporter and diaPASEF quantification were highly correlated, showing an R-squared value of 0.70. Since TMT-reporter quantification accuracy was affected by the ratio compression caused by coelution, the slope of the equation relating TMT-reporter and diaPASEF quantification was 0.57, indicating that the ratio compression from diaPASEF is smaller than that from TMT-reporter. When diaPASEF quantification was compared with TMTc quantification, the slope of the equation relating diaPASEF and TMTc quantification was 0.99, with an R-squared value of 0.60, indicating that the quantification accuracy was quite similar between TMTc and diaPASEF and little ratio compression was observed from diaPASEF when considering TMTc quantification as reference.

Figure 4. The linear regression analysis of protein fold change (AD/CTL) between different quantification methods.

(A) The linear regression analysis of the overlapped protein fold changes between TMT-reporter and diaPASEF quantification methods (n = 5958). (B) The linear regression analysis of the overlapped protein fold changes between TMTc and diaPASEF quantification methods (n = 5958). (C) The linear regression analysis of differential expressed protein fold changes between TMT-reporter and diaPASEF quantification methods (n = 101). (D) The linear regression analysis of differential expressed protein fold changes between TMTc and diaPASEF quantification method (n = 101).

Differential expressed proteins (DEPs) from TMT-reporter data were further identified by moderated t-test in LIMMA (R package), and 102 proteins were considered as DEPs with an adjusted p-value < 0.05 (Table S5). Some known upregulated proteins in AD were found to be upregulated, such as SMOC1 and Amyloid beta, as reported.⁴⁰ The intra-group SD was further calculated among these 3 methods as 0.12 for TMT-reporter, 0.29 for TMTc and 0.23 for diaPASEF, respectively. The high intra-group variability of TMTc could potentially lead to fewer DEPs, especially when the sample size was small. In contrast, diaPASEF demonstrated a smaller intra-group SD than TMTc. After eliminating the unchanged proteins between AD and control groups and outliers with log₂FC > 6 in three methods, the R-squared values for both linear regression analyses increased to 0.91 and 0.89 in Figure 4C and 4D, respectively. Additionally, the slope of TMT-reporter and diaPASEF increased to 0.71, while the slope of TMTc and diaPASEF increased to 1.20. This implied that diaPASEF exhibited quantification accuracy more closely to TMTc than TMT-reporter, even for the DEPs. However, there was an observed increase in ratio compression for DEPs in diaPASEF when compared to all the overlapped proteins. To further evaluate the quantification accuracy of diaPASEF data, we spiked 3 ng and 10 ng of yeast lysate in 25 ng HeLa peptides. In total, 1,476 proteins and 5,927 proteins were quantified from the yeast and human without missing values, respectively. The quantification accuracy, as illustrated in Figure S6, reached 85% after loading amount normalization using HeLa proteome data, demonstrating a close resemblance to the 90% accuracy reported in another study.³⁶

Conclusions

We have optimized the LC and MS parameters for timsTOF SCP. Utilizing a standard column and diaPASEF method, we successfully identified more than 6,000 proteins with 50 ng of HeLa peptides, employing 60 min gradient time, 140 ms ramp/accumulation time and 20 m/z isolation window width. To enhance sensitivity at a low amount level, we implemented a column with reduced length, inner diameter and particle size of packing materials and introduced DDM to minimize sample loss. The modification resulted in a remarkable 19-fold increase in sensitivity at single cell equivalent level, allowing the identification of 3,888 proteins with 1 ng HeLa peptides. The detection sensitivity can be further improved using the newly released timsTOF Ultra instrument.⁵²

While diaPASEF did not achieve the same depth of proteome coverage as the methodology involving TMT-labeled and offline fractionation using DDA method, the identification numbers were reliable and acceptable, particularly considering the instrument efficiency. Quantification analyses demonstrated a high correlation between TMT-reporter, TMTc, and diaPASEF. Notably, the diaPASEF quantification exhibited similar behavior to TMTc quantification, in contrast to TMT-reporter quantification, which displayed more significant ratio compression. In cases of small sample sizes, the in-group variation of TMTc exceeded that of diaPASEF and TMT-reporter, leading to higher p-values when exploring DEPs. This could be attributed to the relatively lower intensities of TMTc ions in comparison to those of TMT-reporter ions.

In light of our optimized instrument parameters and a comprehensive comparison of the three quantification methods, we suggest that DIA quantification with the diaPASEF method not only provides accurate results without significant ratio compression and substantial in-group variation, but also excels in the analysis of low amount samples. These findings underscore the potential of our optimized approach to significantly contribute to advancing proteomics research, particularly in the realm low amount sample analysis.

Supplementary Material

Supporting information

The Supporting Information is available.: Initial optimization of isolation window width; quantification reproducibility evaluation; DDM impact for high loading amount; diaPASEF method optimization for trace sample; quantification efficiency evaluation; quantification accuracy evaluation; sample information; specific gradient design; isolation window tables of diaPASEF; overlapped quantified proteins; list of DEPs.