Strategies based on sum of peak intensities may better reflect protein expression than peak area integration in iTRAQ protein expression measurement by LC-MS/MS using a TripleTOF 5600+ platform

In the field of quantitative proteomics, the Isobaric Tags for Relative & Absolute Quantitation (iTRAQ) technology has demonstrated efficacy for proteome monitoring despite its lack of a consensus for data handling. In this study, after peptide and protein identification, we compared the widespread quantitation method based on the calculation of MS/MS reporter ion peaks areas ratios (ProteinPilot) to the alternative method based on the calculation of ratios of the sum of peak intensities (jTRAQx (Quant)) and we processed output data with the in-house Customizable iTRAQ Ratios Calculator (CiR-C) algorithm. Quantitation based on peak area ratios displayed no significant linear correlation with Western blot quantitation. In contrast, quantitation based on the sum of peak intensities displayed a significant linear association with Western blot quantitation (non-zero slope; Pearson correlation coefficient test, r = 0.2962, p = 0.0099 **) with an average bias of 0.08747 ± 0.5004 and 95% Limits of Agreement from - 0.8932 to 1.068. We proposed the Mascot-jTRAQx-CiR-C strategy as a simple yet powerful data processing adjunct to the iTRAQ technology.


INTRODUCTION
The recent and already widespread large-scale omics technologies enabled the discovery of unexpected mechanisms in the field of physiology, pathophysiology and pharmacology. These techniques investigate DNA (genomics, epigenetics), mRNAs or microRNAs (transcriptomics), proteins (proteomics), lipids and small molecules (metabolomics). When employed in parallel onto these different targets, large-scale omics techniques help seize the many layers of cell responses to pathophysiological stimuli or to drugs, e.g., regulation and transcription of genes, handling of transcripts and translation into proteins. In pharmacology, beyond the first known drug targets, they are major tools to comprehensively explore all intracellular pathways modified by the drug. This enables a better understanding of cellular side effects of drugs.
A number of MS-based high-throughput proteomics or "shotgun proteomics" technologies are compatible with relative protein quantitation and offer variable performances in terms of proteome and sequence coverage, precision, accuracy and reproducibility of quantitation or versatility of sample application (1). Using Label-free Quantification (LFQ), either protein abundance correlates with the measure of peptide precursor ion MS signal intensities or is obtained from the counting of peptide fragment ion MS/MS spectra (spectral counting) (2).
Isotope-coded Affinity Tags (ICAT) was the first labeling technique, which was based on biased protein labeling through tagging of the non-universal residue cysteine with heavy or light tags (3). Stable Isotope Labeling of Amino Acids in Cell Culture (SILAC) is one of the proteomic approaches using a metabolic labeling based on the introduction of heavy or light amino acids during protein biosynthesis (4), hence a high-level reliability and robustness in terms of labeling stability, precision and accuracy (5). Isobaric Tags for Relative & Absolute Quantitation (iTRAQ) was the first chemical labeling technique (before Tandem Mass Tag (TMT) and the . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint mTRAQ variant) developed to multiplex comparison between protein sets issued from different biological samples, as obtained from a single tandem mass spectrometry run. In iTRAQ, digested peptides, from up to eight different conditions, are labeled by isobaric tags. This allows characterizing peptides, in a condition-independent way, using the first mass spectrometry filter (MS mode) and measuring their relative abundance between the different conditions using the second stage of the tandem mass spectrometer (MS/MS mode) (6), upon the detection of the condition-specific reporter ions of distinct masses in the low m/z region of the MS/MS spectrum.
Although extended multiplexing is the major advantage of iTRAQ, its use benefits from: high reproducibility, precision and accuracy compared to LFQ (like all stable isotope labeling versus LFQ), to the cost of a wider dynamic range, better proteome coverage and faster sample preparation and analysis (7,8); better sensitivity and proteome coverage compared to ICAT (9); wide sample applicability, faster sample processing and better proteome coverage compared to SILAC (10); as well as a valuable 'toolbox' that has been built over the past decade thanks to the literature addressing technique drawbacks and methodological solutions to overcome them (11).
The best method to quantify the isobaric tags, hence the relative peptide abundances, is still under scrutiny. The commercially available software used the ratios of peak areas (RPA) based on the initial description stating that the abundance of a collision-released mass reporter ion appeared to be proportional to the trapezoidal integration of peaks at the theoretical mass-tocharge ratio (m/z). Alternatively, the ratios of sum of peak intensities (RPSI) was shown to result in higher sensitivity and more reliable quantitation (12)(13)(14). In this case, the abundance of a mass reporter ion is directly related to its ion counts -height of the peaks -at the theoretical m/z. To the best of our knowledge, performances of the two methodologies (RPA and RSPI) have never 7 been studied and compared in the light of classical molecular biology approaches for protein quantitation (e.g. Western blot).
In iTRAQ, a correct peptide and protein quantitation needs a correct interpretation of MS/MS spectra, which depends on a trustworthy peptide and protein identification. The identification and quantitation can invariably be performed by commercial built-in algorithms integrated to companion software packages; they fully retreat data generated by the manufacturer's mass spectrometers (e.g. Paragon -ProteinPilot for ABSciex mass spectrometers).
Alternatively, analyses may be split into separate stages through a composite suite of commercial or free algorithms integrated to manufacturer-independent software (e.g. Mascot in Mascot Server, jTRAQx). It is noteworthy that, although tools to compute RPA are widely available, the computation of the ratios of RSPI is not supported by any available companion software, and so, it requires an alternative suite.
The first aim of our work was to compare these two strategies of quantitation (RPA and RSPI) to the classical Western blot technique, commonly used as a non-MS validation technique for iTRAQ-based quantitation. The second aim was to further develop an all-in-one protocol, from sample preparation to result reporting, based on the best strategy of quantitation followed by in-house data processing. This study was carried out on the respective effects of the CalciNeurin inhibitors (CNI) Cyclosporine A (CsA) and Tacrolimus (Tac, a.k.a. FK506) on renal proximal tubular cells, used as a study model. iTRAQ was combined to nano-scale liquid chromatography online with Q-Q-TOF tandem mass spectrometry on proximal tubular cells to investigate whether CsA and Tac nephrotoxicity results from the inhibition of calcineurin or from the modulation of other intracellular pathways targeted by immunophilins.
. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint In this work, we validated the RSPI methodology and we built an automated data processing algorithm called Customizable iTRAQ Ratio Calculator (CiR-C) to refine critical parameters related to peptide confidence and selection. The composite suite made up of the Mascot algorithm, for peptide and protein identification, end-to-end with the jTRAQx software for computation of RSPI at the peptide level and the CiR-C algorithm for data integration and definitive protein quantitation turned out to be a successful combination. It has provided a great improvement compared to already available solutions for iTRAQ-based high-throughput quantitative proteomics and multiplexed analysis of biological systems.
. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint bicarbonate, 100 UI.mL -1 Penicillin / Streptomycin and 50 nM Sodium selenite. LLC PK-1 cells were cultured between passage 7 and passage 25. LLC PK-1 were seeded in four 60 mm Petri dishes (one per condition) and expanded up to sub-confluence in the routine cell culture medium.
Differentiated LLC PK-1 cells were exposed for 24 hours to four different conditions: certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint temperature, to alkylate/block cysteine residues and eventually were digested for 24 h at 37 °C with mass-spectrometry grade trypsin (V5280, Promega) at a 1: 50 enzyme: substrate ratio.
After digestion, samples were incubated with iTRAQ tags (iTRAQ Reagents Multiplex kits, 4-plex, #4352135, Sigma-Aldrich) -one tag per drug exposure condition for 1 h at room temperature -and then mixed together. The tags were interchanged between the five independent experiments (biological replicates) to circumvent tag-related bias ( Table 1) Mascot raw data files (*.dat, 1 per experiment) were saved for further isobaric tags-based peptide and protein quantitation with the Java implementation of the Quant algorithm, jTRAQx (version 1.13, (16)). Reporter mass tolerance was set at 0.05 Da while iTRAQ . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint correction factors were implemented as provided by ABSciex. This tool generated one .jpf file (tab-delimited text file) for each series.
. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. Quantitation was computed via the ImageLab software (Bio-Rad).

STATISTICAL ANALYSIS
Western blot quantitation was compared to RSPI-and RPA-based quantitation using the Bland-Altman comparison method (17,18) and Pearson correlation coefficient test as implemented in the GraphPad Prism software (version 5.04).
. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under  The CiR-C shell script excluded irrelevant data according to criteria summed up in Table   2: i) identification confidence: peptides are retained if the probability that the observed positive match is a random match is below 5 % (p < 0.05, Mascot score > 30; Paragon confidence score > 95); ii) quantification confidence: peptides are retained if all iTRAQ ratios have been . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint successfully calculated i.e., peptides with 0.0 ratios or uncalculated ratios (null ratios) are discarded; iii) peptides related to 'Fragment'-and 'REVERSED'-annotated proteins are discarded. After irrelevant data removal, CiR-C drew up an exhaustive catalogue of identified peptide sequences with their associated Swiss-Prot or TrEMBL accession IDs. Peptides were assigned to a frequency index of positive matches (identification in {1;2;3;4;5} out of 5 independent experiments (biological replicates)) and CiR-C drew a second catalogue of peptides with the highest frequency index (n=5). Protein ratios were calculated as both overall and seriesspecific median values of peptide ratios associated with a given accession ID and frequency index.
. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint Table 1. Summary of experimental design. Proteins from five independent replicates of four distinct experimental conditions were extracted from LLC PK-1 cells. Peptides were labeled with iTRAQ 4-plex reagents and analyzed by nano-HPLC MS/MS. 4-plex distribution rotated between independent experiments (biological replicates) to circumvent tag-related bias.
. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint Table 2. Data processing summary. The MS/MS data from the five replicate experiments were submitted to identification (Mascot versus Paragon), primary quantification (jTRAQx versus ProteinPilot) and data refining, statistical analysis (CiR-C). This table sums up data inclusion criteria, identification output and data refining yield, as numbers of identified, discarded of retained peptides.
. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint

COMPARISON BETWEEN THE TWO QUANTITATION STRATEGIES
The results of the two quantitation strategies (RPA from the Paragon -ProteinPilot -CiR-C data processing and RSPI from the Mascot Server -jTRAQx -CiR-C data processing, Scheme 1B) were compared to the Western blot analysis of five proteins picked from the 370protein final list ( Figure 1A). The selection comprised the cytoskeleton-structuring β-Actin, cytoplasmic Cofilin-1, alpha 1 subunit of membrane-attached Na + -K + ATPase, cytosolic Cyclosporine-complexing Cyclophilin A and Galectin-1, i.e., a panel of both spatially scattered and differently expressed proteins ( Figure 1B).  Concerning the quantitative results obtained from the Mascot Server -jTRAQx -CiR-C strategy based on RSPI, a statistically-significant linear association with Western blot quantitation was observed (non-zero slope; Pearson correlation coefficient r = 0.2962, p = 0.0099 **) ( Figure 3A). The average difference was 0.08747 ± 0.5004 (approx. 9 % of a given ratio).
The 95% Limits of Agreement were closer, i.e., -0.8932 and 1.068 ( Figure 3B). In parallel, the two data processing strategies were applied to the analysis of MS/MS data of UPS1 spiked-in, iTRAQ-labeled LLC PK-1 protein extracts, for MS-related benchmarking ( Figure S1).
The Paragon -ProteinPilot -CiR-C data processing pipeline resulted in the monitoring of 24 out of 48 UPS1 proteins ( Figure S1B). No significant differences (One-way ANOVA, . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint p=0.3532) were observed between DOWN:N1, UP:N1 and N2:N1 ratios. The global expression profile failed to reflect the differential spike-in of UPS1 proteins. Ratio compression towards ratio=1.00 was observed.
. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint

REFINMENT OF THE CiR-C ALGORITHM
The CiR-C script uses two thresholds for peptide selection: a peptide significance threshold (given by the Mascot score) and a peptide frequency threshold (defined as the minimum number of experiments in which a given peptide must be identified and quantified). To establish the best compromise between reliability of iTRAQ quantitation and time/cost efficiency, we explored how threshold tuning may affect (strengthen or weaken) CiR-C results.
The RSPI calculated after tuning the first threshold to either a permissive (Mascot score > 20) or a stringent (Mascot score > 40) cut-off value were compared to Western blot standards ( Figure 4). When opening to more peptides (lower Mascot score) the linear correlation was lost (zero slope; Pearson correlation coefficient test, r = 0.06231, p = 0.5954) ( Figure 4A); doing so, the mean difference compared to Western blot was slightly reduced (-0.06338 ± 0.5367, 95% Limits of Agreement 0.9886 and -1.115) ( Figure 4B). When restricting to less peptides (higher Mascot score) the linear correlation was not improved (non-zero slope; Pearson correlation coefficient r = 0.2737, p = 0.0175 *) ( Figure 4C); again, the difference compared to Western blot was slightly reduced (-0.07496 ± 0.5098, 95% Limits of Agreement 0.9243 and -1.074) ( Figure   4D). . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint The RSPI computed after adjustment of peptide frequency threshold to lower levels were plotted against Western blot results ( Figure 5). Using peptides from four out of five independent experiments (biological replicates) resulted in a significant improvement of correlation (non-zero slope; Pearson correlation coefficient r = 0.3466, p = 0.0067 **) ( Figure 5A) and a greater difference compared to Western blot (-0.05458 ± 0.5186, 95% Limits of Agreement 0.9618 and -1.071) ( Figure 5B). Conversely, using peptides from only three out of five independent experiments (biological replicates) led to a loss of correlation (zero slope; Pearson correlation coefficient r = 0.06591, p = 0.6671) ( Figure 5C) and a greater difference compared to Western blot (-0.09232 ± 0.6069, 95% Limits of Agreement 1.097 and -1.282) ( Figure 5D). strategies and tools to improve quantitation precision and accuracy (21)(22)(23)(24). However, there is still no consensual technique in sample preparation and analysis or data processing.
iTRAQ results in rich and complex MS/MS datasets which require thorough processing and solid statistics to reach relevant conclusions. In this respect, the choice of a tag ratio calculation method is crucial. Two potential strategies have been proposed: either RPA or RSPI.
The two methods are not equivalent since peak area measurement suffers from a major bias originated from the way reporter ion signals are processed, as described by Boehm et al. (13). It is worth noting that peak intensities are proportional to reporter mass ion counts, whereas peak areas are not (Scheme 2). Therefore, the RSPI is more likely to provide a reliable rendering of the actual ion count detected by mass spectrometer. Numerous studies reported the necessity of RPSI-based quantitation workflows to obtain robust, precise, accurate and sensitive when using high resolution MS platforms (Orbitraps and TripleTOF 5600) (25)(26)(27) We confirmed the superiority of RSPI since the commercially available RPA strategy failed to report significant biological variations assessed by Western Blot analysis of LLC-PK-1 protein extracts and iTRAQ-nanoLC-MS/MS benchmark analysis of UPS1 spiked-in LLC-PK-1 protein extracts. In contrast, with the lowest bias and best correlation, the RSPI from Mascot -jTRAQx -CiR-C data processing strategy provided the most reliable set of quantitation ratios when compared to Western Blot performance. As compared to the Western blot 'quality control', the median RPA showed dramatic differences leading to a global misrepresentation of changes in protein .
CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint expression. Conversely, the median RSPI for the set of proteins of interest were well correlated and showed moderate and unbiased differences compared to Western Blot. Similar observations were made monitoring the protein abundance of spiked-in UPS1 proteins after differential spikein. Scheme 2. RSPI fit better than RPA to report tag signature ion counts. Western blot was chosen as a non-MS-based quality control for iTRAQ because the technique has already been successfully used to validate iTRAQ. Either alone (28,29), in parallel with other classical molecular biology techniques such as RT-qPCR (30,31), or, in parallel with MS-based targeted methods like selected reaction monitoring/multiple reaction monitoring (SRM/MRM) (32,33), Western blot always managed to validate and confirm results from both iTRAQ and other, more performant, validation techniques. Most importantly, the technique was the first to assess performance issues when discrepancies between Western blot and iTRAQ results highlight ratio compression and underestimation (34). The observed difference between iTRAQ ratios and Western blot ratios may be explained by the intrinsic ratio compression due to background noise at the low m/z end, by co-elution of peptides with close m/z, and to a lesser extent by tag purity and inter-contamination (35). Even if correction factors are provided to take tag 'impurity' into account, the background mass spectrometry noise brings the ratio towards unity. However, the custom-made strategy using the RSPI appeared to minimize these differences with respect to the commercially available strategy.
Differences between iTRAQ and Western blot may also be explained by the technical evolution of LC-MS-based quantitative proteomics when compared to antibody-based . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint approaches. On one hand, Western blot remains a qualitative and semi-quantitative technique with an inherent variability in analytical sensitivity, specificity and reproducibility, especially because of the multifactorial (e.g. variability of antibodies specificity, non-linearity of chemiluminescent reaction) non-linearity between protein abundance and signal intensity (36,37). On the other hand, the use of high-performance liquid chromatography (such as nanoLC) online with high-resolution mass spectrometers (such as TripleTOF 5600+ QqTOF mass spectrometer) has outperformed Western blot with greater sensitivity, selectivity, specificity and reproducibility, wider linear dynamic range, increased accuracy and precision, high-throughput of in-depth information (38)(39)(40)(41)(42). Despite these considerations, Western blot quality control alone highlighted the superiority of the RSPI-based Mascot -jTRAQx -CiR-C pipeline over the RPA-based Paragon -ProteinPilot -CiR-C pipeline, later confirmed thanks to the UPS1 spike-in experiments.
A major point of iTRAQ data processing is the upstream tolerance for peptide characterization and identification. More peptides were identified with Mascot Server than with Paragon, mainly because mass tolerances were more stringent with the latter, due to the use of non-customizable manufacturer's parameters optimized for data generated by TripleTOF 5600+ mass spectrometers. Mass tolerances with Mascot Server were chosen to be sufficiently permissive to provide enough data for further selection. The choice of stringent parameters for identification is in complete agreement with the RPA quantitation strategy, as implemented in ProteinPilot, where peak width is essential to reliable quantitation. Conversely, our approach postpones the application of stringent conditions to after quantitation.
Another major point of iTRAQ data processing is the downstream management of irrelevant data. It can be done manually but it is time-consuming and error-prone due to the huge . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint amount of information at the peptide level as well as the number of replicates. Just like with every high-throughput approach, the analysis of iTRAQ results needs to be automated. The CiR-C algorithm was designed to be the simplest in terms of data elimination and transformation. Likely to exacerbate inherent issues -such as ratio-compressing variance-stabilizing normalization -heavy data transformation was not used after jTRAQx or ProteinPilot data processing. The biological significance of the results arose from: i) a focus on peptide selection thanks to the probabilistic Mascot score or its Paragon counterpart (Confidence score); ii) the restriction to selected peptides based on their occurrence among multiple biological replicates; iii) the calculation of median and weighted mean ratios for each set of peptides obtained from a given protein. As mentioned above, upstream data processing was willingly permissive to sustain this statistical approach. It resulted in less loss of information and good data fitting.
Paradoxically, stringent identification criteria also resulted in increased output, including a large amount of irrelevant data: when parsing peptide summaries, it appeared that irrelevant data were essentially null ratios, i.e., the technical impossibility to provide peptide quantitation from the MS2 mass spectra.
The confidence threshold for the Mascot score was a compromise between the generation of aberrant information and the loss of information. This threshold highly depends on the number of peptides characterized by mass spectrometry. The probability of a random match was first set to 1 out of 1000 (s = 30, p = 0.001) and the test of scores 10-fold apart from our initial choice (s = 20, p = 0.01) confirmed how close it is from a potential optimum for the size of our datasets.
Indeed, this threshold highly depends on the number of peptides characterized in MS/MS.
The occurrence threshold was set to the largest possible values -the more peptides, the better -hence the restriction to ubiquitous peptides, i.e., peptides retrieved from all five . CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint biological replicates. The main source of variations when using iTRAQ is of biological origin (± 25 %) (43). The number of biological replicates was set to five independent experiments (biological replicates), corresponding to the use of a complete given set of iTRAQ tags.
However, as a matter of time and cost efficiency, the possibility to reduce the number of independent experiments (biological replicates) had to be addressed. As expected, considering biological variations are the most impactful source of variability of iTRAQ, we showed that using four independent replicates is the lowest limit to obtain reliable results, five replicates appearing optimal under our experimental conditions.
In summary, this work demonstrated that RSPI outperform the commercially available RPA in quantifying biological modifications using iTRAQ. Furthermore, we propose a Mascot -jTRAQx -CiR-C strategy as a simple yet powerful answer to the need for an all-inclusive suite for iTRAQ data processing.
. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint Figure S1. Protein spike-in benchmarking of compared iTRAQ data processing strategies. A. Summary of UPS1:LLC PK-1 spike-in ratios and iTRAQ 4-plex distribution, UP, 25 % increase in UPS1 protein abundance, DOWN, 25% decrease in UPS1 protein abundance, N1/N2, reference UPS1 protein abundance, B. Protein abundance of UPS1 spiked-in proteins as computed by the Mascot-jTRAQx-CiR-C pipeline, C. Protein abundance of UPS1 spiked-in proteins as computed by the Paragon-ProteinPilot-CiR-C pipeline, DOWN:N1 expected ratio 0.75, N2:N1 expected ratio 1;00, UP:N1 expected ratio 1.25. One-way ANOVA plus Dunnett's post-test, n=3, *p<0.05, **p<0.01.
. CC-BY 4.0 International license a certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under The copyright holder for this preprint (which was not this version posted March 13, 2019. ; https://doi.org/10.1101/577254 doi: bioRxiv preprint