Most cellular processes are regulated by post-translational modifications (PTMs). Given the sensitivity of current mass spectrometric analyses, even non-functional basal PTMs can be identified by mass spectrometry. Thus, an understanding of the stoichiometry and changes thereof are crucial to understanding function. Numerous bioinformatics tools that enable unbiased and large-scale identification of modified peptides and localization of the modification site such as MSFragger (Kong et al., 2017), MetaMorpheus (Solntsev et al., 2018), or TagGraph (Devabhaktuni et al., 2019) have been developed. With steadily increasing qualitative information about the identity of PTMs (Grimsrud et al., 2010; Choudhary et al., 2014; Kim et al., 2011; Melo-Braga et al., 2014; Udeshi et al., 2013), tools to investigate the modification quantity, that is the proportion of the protein carrying PTMs, become increasingly important. Thus far, analysis of the PTM extent has mainly focused on quantifying the modified peptides, for example using PTM enrichment methods (Pinkse et al., 2004; Rush et al., 2005; Zhou et al., 2011; Lundby et al., 2012; Lundby et al., 2013), heavy isotope-labeled synthetic peptides (Gerber et al., 2003) or by enzymatically removing (Yamagata et al., 2002; Wu et al., 2011) PTMs. These approaches require prior knowledge of the PTMs and have been primarily applied to the phosphoproteome.

The development of FLEXIQuant (Full-Length-Expressed Stable Isotope-labeled Proteins for Quantification) (Singh et al., 2009; Singh et al., 2012a), for the first time, enabled the quantification of the extent of modification across a whole protein without prior knowledge of modification identities using an indirect approach of quantifying only the unmodified peptides. FLEXIQuant relies on the principle that the total number of molecules of a given protein is conserved in a sample, thus the number of molecules of each unique peptide derived from that protein will be equal. If a peptide is modified chemically by a PTM this would reduce the abundance of its unmodified cognate in the peptide pool. In FLEXIQuant, this idea is realized by analyzing the peptide pool of the sample with an added unmodified heavy isotope-labeled full-length reference protein. The extent of modification is calculated by comparing the quantities of labeled reference peptides derived from the standard to the unlabeled cognate peptides from the sample, that is the unlabeled endogenous protein. The advantage of this approach is the precise and accurate quantification of the degree of abundance reduction due to a modification as defined by Wold, 1981 or amino acid substitutions of each quantified unmodified peptide and the ability to calculate the absolute concentration. FLEXIQuant and derivatives thereof have been applied to study for example Tiki1 modification in head formation in frogs (Zhang et al., 2012), to gain insights into the phosphorylation dynamics of GSK3β-dependent phosphorylation of DCX (Singh et al., 2012b), the cell-cycle-dependent phosphorylation of KifC1 (Singh et al., 2014) or to investigate PTM of Tau in Alzheimer’s disease (Mair et al., 2016). However, the limitation of this method is the requirement of a purified, unmodified, labeled reference protein and thus, it is labor intensive and has largely been used to investigate individual proteins. Recently, a PTM analysis pipeline without the need for an internal reference protein has been published (Bagwan et al., 2018). Although this is encouraging, the quantification relies on isobaric labeling approaches which are expensive to conduct due to the high cost of the labeling reagents and is therefore not widely applicable, especially not for large-scale experiments. This highlights the need for bioinformatic tools suitable for the analysis of label-free experiments.

Here, we introduce FLEXIQuant-LF as an unbiased, label-free computational tool to indirectly detect modified peptides and to quantify the degree of modification based solely on the unmodified peptide species building upon the FLEXIQuant (Singh et al., 2009) idea developed in our lab. We developed this approach to identify and elucidate differential protein modification extent in commonly analyzed time series or case-control studies. In these studies, a timepoint or control group is used as reference to enable a FLEXIQuant-like modification analysis. The key requirement of FLEXIQuant-LF is the unbiased, comprehensive and highly reproducible quantification of peptides, therefore, we use a data-independent acquisition (DIA) strategy (namely, SWATH [Gillet et al., 2012]). Here, we demonstrate the utility of FLEXIQuant-LF by applying our method to interrogate the peptide-resolved modification dynamics of the anaphase-promoting complex/cyclosome (APC/C) during mitosis (Singh et al., 2009; Steen et al., 2008).

The objective of our study was to establish a workflow to elucidate peptide-resolved modification dynamics in an unbiased manner without the need for heavy isotope-labeled reference proteins or peptides. Our strategy focused on profiling unmodified peptides, as any variation in the extent of modification of a peptide results in the reduction of the amount/intensity of the remaining unmodified peptide species. This consideration also applies to modifications on or close to proteolytic cleavage sites that interfere with the proteolytic cleavage. In such cases, the affected proteolytic cleavage yield shows a reduction which is equivalent to the modification extent. This interference with the proteolytic cleavage results in a proteolytic missed cleavage peptide, which in turn leads to a change in the amount/intensity of the two unmodified, fully cleaved peptide species. Thus, variation in the amount of an unmodified peptide can be used to identify and quantify peptides whose modification state changes relative to a reference point in a time series or in a dose-dependent manner. The idea of focusing on the quantity of the unmodified peptide to identify and quantify modified peptides was implemented in the original FLEXIQuant approach, which uses stable isotope-labeled reference samples (Singh et al., 2009; Singh et al., 2012a). Here, we introduce a new, label-free (LF) implementation of this approach: FLEXIQuant-LF.

FLEXIQuant-LF enables unbiased indirect identification of differentially abundant peptides resulting from PTM in label-free mass-spectrometry data. Besides identifying differentially modified peptides the tool can also quantify the extent of these modifications. Our tool only requires raw peptide intensities as input data making it widely applicable. FLEXIQuant-LF determines differentially abundant unmodified peptides using RANSAC-based robust linear regression in combination with a sophisticated normalization and score calculation procedure (Figure 1B). The RANSAC algorithm iteratively determines outliers in the data and fits the regression line only to observations not classified as outliers (inliers). This ensures that the regression line is fitted to the most robustly quantified peptides and facilitates the quantification of the modification extent based on the distance to the regression line.

We benchmarked FLEXIQuant-LF by analyzing the anaphase promoting complex/cyclosome (APC/C) during mitotic arrest. Application of FLEXIQuant-LF to this complex for proof of concept had two advantages: (1) excellent enrichment protocols for isolation of the complex by targeting the cell division cycle protein 27 (CDC27) are well established (Singh et al., 2009) and (2) the cell cycle-dependent phosphorylation dynamics of APC/C have been qualitatively and quantitatively described (Singh et al., 2009; Steen et al., 2008). We studied the APC/C after treating thymidine-synchronized HeLa cells with nocodazole for 4 hr, 8 hr, and 10 hr to induce prometaphase arrest. The APC/C isolated from S-phase cells served as a control (workflow in Figure 1A).

The software is open source and available in two versions: an easy-to-use graphical user interface (GUI) version as well as a command line interface (CLI) version to allow for integration into bioinformatics analysis pipelines.

Benchmarking FLEXIQuant-LF with CDC27 and APC5

For the benchmarking of FLEXIQuant-LF, we focused on CDC27 and APC5. Cell cycle-dependent modification dynamics have been extensively studied for both proteins (Singh et al., 2009; Steen et al., 2008). These pre-existing published data enabled us to validate our peptide classifications.

We quantified 33 peptides corresponding to CDC27 (sequence coverage: 55%), of which nine peptides were classified as likely differentially modified within the time series (Figure 2A, magenta bars) by our FLEXIQuant-LF method. Four peptides were classified as possibly differentially modified (Figure 2A, blue bars). One peptide (₃₁LYAEVHSEEALFLLATCamYYR₅₀) was classified as outlier based on its raw scores and removed before RM score calculation.

Figure 2

Download asset Open asset

We first checked for modified peptide species within our DDA dataset. In total, we identified six CDC27-derived phosphorylated peptides in the DDA data (Figure 2A, orange stars). The FLEXIQuant-LF analysis classified the unmodified versions of these six phosphopeptides as follows: five likely differentially modified peptides and one possibly differentially modified peptide. For the other four peptides identified as likely differentially modified as well as for two out of the three remaining peptides classified as possibly differentially modified by FLEXIQuant-LF, phosphorylations, and/or ubiquitinations were described (Steen et al., 2008) or deposited online at http://www.phosphosite.org (Hornbeck et al., 2015) (yellow stars). In summary, FLEXIQuant-LF demonstrated an excellent performance in detecting potentially modified peptides. We found evidence of modification for all peptides identified as likely modified in our study and for three out of four possibly differentially modified peptides.

In addition to the detection of potentially modified peptides, we designed FLEXIQuant-LF to also quantify the degree of differential modification. Relative to time point 0 hr (S-phase) in our time series, the N-terminal region of CDC27 (amino acids 170–441) was increasingly modified after 4 hr (27% to 54%, average: 36%) and reached its maximum modification extent after 10 hr (between 55% and 81%, average: 65%) (Figure 2A). The peptide spanning amino acids 442 to 456 (₄₄₂ISTITPQIQAFNLQK₄₅₆) also showed an increasing degree of modification but to a lower extent (up to 48% modified after 10 hr).

For APC5, two peptides (₄₉₇FPPNSQHAQLWMLCamDQK₅₁₃ and ₃₅IAVLVLLNEMSR₄₆) out of 37 quantified peptides (sequence coverage: 59%) were classified as outliers based on their raw scores and were excluded from further analysis. Out of the 35 remaining peptides, four were classified as likely differentially modified across the time points (Figure 2B, magenta bars), while five peptides were considered as possibly differentially modified (Figure 2B, blue bars). We identified in our DDA data the phosphorylated cognates for two of the four likely modified peptides (Figure 2B, orange stars). We found evidence of modifications within the literature for the other two peptides identified as likely differentially modified in our FLEXIQuant-LF analysis as well as for four of the five peptides classified as possibly differentially modified (yellow stars) (Hornbeck et al., 2004). Interestingly, we did not find evidence for modification for one possibly differentially modified peptide (₉₉LMAEGELK₁₀₆). This could be interpreted either as a false positive classification or that this peptide has a hitherto undescribed modification on the methionine, glutamic acid or lysine residues.

Overall, these findings are in excellent agreement with the original label-based FLEXIQuant analysis of CDC27 and APC5 (Singh et al., 2009), which identified the same protein regions as being modified to a comparable extent. It was noted that compared to the original FLEXIQuant study the extent of modification was lower in our novel FLEXIQuant-LF analysis. A highly probable explanation for this phenomenon could be that there was a basal modification level of up to 20% for some peptides in S-phase, as shown with FLEXIQuant (Singh et al., 2009) which determines the absolute extent of modification. Thus, we emphasize that the FLEXIQuant-LF approach provides the extent of modification relative to the given reference time point.

The peptide-resolved modification dynamics of other APC/C complex components

FLEXIQuant-LF has an advantage over the conventional FLEXIQuant implementation in that the analysis can easily be performed on all proteins within a sample (provided a minimum of five peptides was quantified to allow for linear regression). Here, we extended our analysis to all other APC/C proteins of which sufficient peptides were reliably quantified in our data, namely APC1 (Figure 3) and APC4, APC7, APC10, CDC16, CDC20, and CDC23 (Figure 3—figure supplement 1). By applying FLEXIQuant-LF, we classified between zero (APC4, APC10) and seven peptides (APC1) per protein as likely differentially modified and between zero (APC4, APC10, CDC20) and four peptides (APC1) per protein as possibly differentially modified (summary in Table 1 and Table 2).

Figure 3 with 2 supplements see all

Download asset Open asset

Table 1

			FLEXIQuant-LF analysis				Evidence
Protein	Quant. peptides	Seq. cov.	Likely diff. modified	Possibly diff. modified	Excluded		DDA	Lit.	No evidence
APC1	54	39%	7	4	4		8	9	2
APC4	28	43%	0	0	2		-	-	0
APC5	37	59%	4	5	2		2	8	1
APC7	27	52%	4	2	1		1	4	2
APC10	9	69%	0	0	3		-	-	0
APC15	1	30%	0	0	0		-	-	0
APC16	3	30%	0	0	0		-	-	0
CDC16	23	41%	3	2	3		1	3	2
CDC20	7	22%	1	0	1		0	1	0
CDC23	33	52%	3	3	4		3	5	1
CDC27	33	55%	9	4	1		6	12	1
in total	248	AVG 47%	30 12%	20 8%	20 8%		21 42%	41 82%	9 18%

Table 2

Protein	Peptide	Start	End		RM score
					4 hr	8 hr	10 hr
APC1	GDSPVTSPFQNYSSIHSQSR	311	330		0.71	0.37	0.27
	SPSISNMAALSR	341	352		0.71	0.32	0.29
	FSEQGGTPQNVATSSSLTAHLR	285	306		0.69	0.34	0.30
	AHSPALGVHSFSGVQR	353	368		0.69	0.35	0.30
	NFDFEGSLSPVIAPK	680	694		0.80	0.43	0.36
	LLQLCamQR	1070	1076		0.57	0.51	0.45
	ARPSETGSDDDWEYLLNSDYHQNVESHLLNR	696	726		0.78	0.61	0.48
	LHDSLYNEDCamTFQQLGTYIHSIR	566	588		0.75	0.42	0.50
	SLCamLSPSEASQMK	727	739		0.83	0.61	0.51
	TMALPVGR	1077	1084		0.65	0.59	0.56
	LHDSLYNEDCamTFQQLGTYIHSIRDPVHNR	566	594		0.69	0.63	0.57
APC5	EELDVSVR	190	197		0.64	0.42	0.35
	EEEVSCamSGPLSQK	198	210		0.59	0.39	0.40
	ALTPASLQK	230	238		0.57	0.49	0.43
	CamQVASAASYDQPK	670	682		0.52	0.47	0.46
	KTVEDADMELTSR	168	180		0.76	0.64	0.50
	LILTGAESK	281	289		0.67	0.53	0.52
	LMAEGELK	99	106		0.70	0.58	0.54
	DSDLLHWK	404	411		0.58	0.58	0.57
	TVEDADMELTSR	169	180		0.82	0.63	0.59
APC7	VRPSTGNSASTPQSQCamLPSEIEVK	116	139		0.85	0.46	0.41
	AYAFVHTGDNSR	242	253		0.60	0.47	0.47
	DMAAAGLHSNVR	43	54		0.61	0.52	0.49
	YTMALQQK	100	107		0.70	0.47	0.54
	ALTQRPDYIK	468	477		0.67	0.55	0.57
CDC16	QTAEETGLTPLETSR	573	587		0.86	0.34	0.34
	CamYDFDVHTMK	544	553		0.60	0.55	0.45
	SSICamLLR	130	136		0.65	0.60	0.49
	IYDALDNR	139	146		0.71	0.56	0.52
	DPFHASCamLPVHIGTLVELNK	261	280		0.44	1.00	0.60
CDC20	VLSLTMSPDGATVASAAADETLR	446	468		0.51	0.32	0.34
CDC23	NQGETPTTEVPAPFFLPASLSANNTPTR	558	585		0.82	0.27	0.13
	RVSPLNLSSVTP	586	597		0.71	0.30	0.20
	VSPLNLSSVTP	587	597		0.82	0.36	0.28
	AALYFQR	350	356		0.67	0.57	0.52
	LWDEASTCamAQK	525	535		0.74	0.62	0.56
	NTSAAIQAYR	380	389		0.70	0.55	0.56
CDC27	EVTPILAQTQSSGPQTSTTPQVLSPTITSPPNALPR	341	376		0.46	0.31	0.19
	LDSSIISEGK	432	441		0.66	0.33	0.27
	LFTSDSSTTK	381	390		0.58	0.29	0.30
	YSLNTDSSVSYIDSAVISPDTVPLGTGTSILSK	224	256		0.56	0.32	0.33
	LNLESSNSK	215	223		0.69	0.33	0.33
	GGITQPNINDSLEITK	416	431		0.73	0.47	0.41
	QPETVLTETPQDTIELNR	197	214		0.71	0.42	0.41
	SVFSQSGNSR	331	340		0.66	0.35	0.42
	FTSLQNFSNCamLPNSCamTTQVPNHSLSHR	170	196		0.73	0.46	0.45
	ISTITPQIQAFNLQK	442	456		0.77	0.54	0.52
	ASVLFANEK	710	718		0.53	0.48	0.56
	HYNAWYGLGMIYYK	634	647		0.53	0.76	0.59
	DAVFLAER	23	30		0.62	0.61	0.59

From APC1, we quantified 54 peptides (sequence coverage: 39%) in total, of which four peptides were classified as outlier based on their raw scores and were excluded from further analysis (₉₃₀LVVWMTNVGFTLR₉₄₂, ₉₄₃DLETLPFGIALPIR₉₅₆, ₁₅₄₆TGGEMNYGFHLAHHMALGLLFLGGGR₁₅₇₁ and ₁₆₁₈LLVPVDVDTNTPCamYALLEVTYK₁₆₃₉). Out of the remaining 50 peptides, seven peptides were classified as likely differentially modified and four peptides as possibly differentially modified (Figure 3, magenta and blue bars, respectively). We identified phosphorylated peptides in the DDA data for five of the seven likely and for three out of four possibly differentially modified peptides (Figure 3, orange stars).

We also classified the peptide ₃₁₁GDSPVTSPFQNYSSIHQSR₃₃₀ as likely differentially modified by our FLEXIQuant-LF analysis, although no modified cognate peptides were identified in the DDA data. However, multiple modifications, that is phosphorylations on S313, T316, S317, Y322 and S324, have been reported in previous studies (Hornbeck et al., 2015) (indicated by a yellow star in Figure 3). For one likely (₁₀₇₀LLQLCamQR₁₀₇₆) and one possibly differentially modified peptide (₁₀₇₇TMALPVGR₁₀₈₄), we could not find published evidence of PTMs within the sequence of those two peptides. Interestingly, these are consecutive peptides with the latter one starting with a threonine residue (AA 1077). A hitherto undiscovered phosphorylation at this threonine residue could lead to a highly increased missed cleavage which would be consistent with the observed intensity reduction of both peptides.

A more detailed analysis of the peptide-resolved modification dynamics observed in APC1 identified two regions within the protein that appear to have different modification kinetics (Figure 3). Peptides in the N-terminal part ranging from residues 285–368 (four peptides), showed a degree of modification of 29–31% after 4 hr, 63–68% after 8 hr and between 70% and 71% after 10 hr. In contrast, the peptides in a more C-terminal domain spanning residues 680–739 (three peptides) showed slower modification kinetics and a lower degree of modification at 10 hr. After 4 hr, 17–22% change in modification status was observed. The extent of modification changed between 39% and 57% after 8 hr, and 49–64% after 10 hr. These data demonstrate that tools to study the peptide-resolved modification dynamics in a quantitative manner will greatly improve the current understanding of biological processes.

Testing and validating reproducibility of FLEXIQuant-LF

We tested the reproducibility of FLEXIQuant-LF by analyzing a large inhouse MRM data set consisting of 81 samples of which 26 were reference group samples. Over 1000 runs FLEXIQuant-LF yielded ambiguous results for 6 out of 81 samples when using one RANSAC initiation. In all six cases, two distinct outcomes were observed. To improve the reproducibility, we initiated the RANSAC algorithm multiple times within a single run (1, 10, 20, 30, 40, 50, 60, 70, 80, 90, and 100) and selected the best model (based on r2 score). This was again tested by running each set of initiations 1000 times and analyzing the slopes of the resulting best models. Initiating RANSAC 10 times resulted in six ambiguous samples, 20 initiations in 5, 30 in 2 and 40 in 3. Between 50 and 100 initiations, the number of ambiguous samples was oscillating between 1 and 2 (Figure 4A). Furthermore, as illustrated by means of sample 30, the frequency of different outcomes within a sample decreased with increasing number of RANSAC initiations (Figure 4B). With one initiation the suboptimal outcome prevailed (88%), whereas with 10 initiations the optimal outcome was predominant (67.2%). Using 30 initiations the fraction of suboptimal outcomes was 2.2% and further decreased to 0.5% with 50 initiations.

Figure 4

Download asset Open asset

Overall, using 30 RANSAC initiations and selecting the best model resulted in 44 suboptimal outcomes in 81,000 runs of FLEXIQuant-LF (0.05%). These results confirm the benefits of multiple RANSAC initiations and demonstrate a very high reproducibility of the developed approach. Based on these finding, the default number of RANSAC initiations in FLEXIQuant-LF was set to 30 but can be changed by the user.

We also tested the quality of FLEXIQuant-LF classification and quantification on an independent DIA dataset (Bruderer et al., 2017). After stringent filtering 122 proteins without detected modifications and with unique and unambiguous associations to peptides were selected for in silico modification, that is intensities of single peptides were reduced to simulate the presence of modified peptide species. For each protein, four peptides were randomly chosen, and their intensity adjusted by a random change factor for each replicate resulting in 1574 total FLEXIQuant-LF runs (Supplementary file 3). RM score changes showed a high correlation with the in-silico intensity change factors (Pearson’s r = 0.98, Figure 4C). The overall near perfect correlation of the in-silico and RM score changes was also reflected in the classification results. The classes of likely not differentially modified and likely differentially modified peptides also showed near optimal performance with sensitivity and precision values at 97.5% and 99.1%, and 100.0% and 96.1%, respectively, while the class of possibly differentially modified peptides with hard-to-define borders showed reduced performance with a sensitivity of 77.2% and precision of 88.6% (Figure 4D). This decrease in accuracy likely reflects more on the definition criteria of the middle group than on the performance of the method. Additionally, of interest was the robustness of FLEXIQuant-LF results in relation to the number of peptides available for each protein. Overall, quantification errors were low with a mean error of 0.020 (SD = 0.058,<0.1 for 94.7% of cases, Figure 4E). Quantification errors in proteins with fewer peptides (n = 5–7: μ = 0.027, SD = 0.072) did not deviate from proteins with more peptides (n = 8–10: μ = 0.010, SD = 0.020; n = 11–13: μ = 0.013, SD = 0.026; n = 14–17: μ = 0.006, SD = 0.012) and were within a very small error margin of modification extent of <5% (Figure 4F). Outliers are rare (<5%) and can be attributed to the random selection and simulation effects of the assessment approach.

Our FLEXIQuant-LF is based on robust linear regression, as implemented in the RANSAC algorithm. In general, linear regression requires a certain number of data points to yield meaningful results and becomes more accurate with increasing number of data points. Thus, FLEXIQuant-LF becomes more powerful with increasing numbers of unique peptides that can be identified for the proteins of interest. We set the minimum number of unique peptides required for a FLEXIQuant-LF analysis to five, although a larger number of peptides will provide more robust results. Addressing this limitation, we introduce the ‘superprotein’ approach.

In the presented proof-of-concept study, we applied FLEXIQuant-LF to a time series experiment. However, FLEXIQuant-LF can also be applied to study protein modification differences between different experimental conditions or cohorts in general, e.g. comparing diseased vs. healthy individuals or testing the effect of different drugs. As FLEXIQuant-LF only determines the extent of differential modification of a peptide relative to the reference based on the unmodified counterparts, quantification of the absolute abundance of the modification still requires an isotopically labeled internal standard. However, using a reference time point or condition has the advantage of highlighting modifications that exhibit dynamic regulation in the context of interest. Furthermore, FLEXIQuant-LF can be used to repeatedly analyze the same data set with varying reference samples, providing complementary perspectives on the peptide-resolved modification dynamics. This consideration is particularly important for the current implementation of FLEXIQuant-LF which only quantifies modifications extents that are increasing relative to the reference sample. This is due to our normalization method used for RM score calculation, which excludes peptides with decreased modification. FLEXIQuant-LF reports these peptides in the ‘_removed_peptides.csv’ output file and raw score information can be retrieved from the ‘_raw_score.csv’ output file. We reasoned that the reference sample is normally the least modified protein. In case this assumption does not hold, a second FLEXIQuant-LF can be carried out after selecting another sample as the reference.

While FLEXIQuant-LF, similar to the initial FLEXIQuant approach, indirectly identifies peptides that are differentially modified, identifying the specific type and site of modification is not the objective of this analysis. The detailed characterization and localization of the modifications require either more in-depth data analysis or follow-up experiments. Nevertheless, this analysis method can be easily applied in any affinity purification or global proteomics experiment and the results can further be used to identify proteins to be synthesized as reference standards, if more in-depth experiments are needed. In the original FLEXIQuant approach, changes in the modification extent had to be above 20% in order to be reliably detected. For our proof of concept study, we chose 40% which is consistent with stoichiometries commonly observed in cell cycle-focused or signaling pathway activation-related large-scale phosphoproteomics studies, which reported serine and threonine phosphorylation extents exceeding 75% and tyrosine phosphorylation extents above 50% (Sharma et al., 2014). However, our FLEXIQuant-LF approach enables the selection of a minimum change in modification extent deemed to be interesting and/or compatible with the precision of the dataset; clearly the usual trade-off considerations between specificity and sensitivity apply.

Even though APC/C is a moderately abundant complex within the cell, we opted for affinity enrichment to ensure a high number of unique peptides per protein of the complex components are measured. For highly abundant protein classes such as the ribosome and proteasome (Beck et al., 2011), a direct FLEXIQuant-LF analysis on the whole cell lysate is most likely a viable option. However, we expect that future technological developments in LC/MS instrumentation and data analysis approaches will make less abundant proteins and protein complexes such as signaling proteins/complexes amenable to FLEXIQuant-LF analysis without further enrichment of the proteins/protein complexes of interest.

A premise of the FLEXIQuant-LF method is, that the modification state of the majority of peptides in a protein does not change between experimental conditions or over the time period considered in a study. This is likely to hold true for most proteins. However, in some cases with more modified than unmodified peptides as in Tau for example (Mair et al., 2016), FLEXIQuant-LF might not be applicable.

The combination of affinity purification of the protein complex and DIA analysis facilitated the precise and reproducible peptide quantification. Currently, DIA is the most suitable acquisition method due to its precise peptide quantification capabilities with fewer missing values. However, the FLEXIQuant-LF approach is also applicable to other label-free quantification methods.

MS data are available via ProteomeXchange under accession code PXD018411 and PXD005573. All data generated or analysed during this study are included in the manuscript and supporting files. FLEXIQuant-LF is available with graphical user interface (GUI) as standalone executable EXE file. Additionally, a command line interface (CLI) version is available as PY file. Both versions as well as the Python source code can be downloaded via GitHub: https://github.com/SteenOmicsLab/FLEXIQuantLF (copy archived at https://archive.softwareheritage.org/swh:1:rev:4ea3945f86ba477227c89e9ced75fc23751355ac/).

1. 1. Schlaffner CN
   2. Kahnert K
   3. Muntel J
   4. Chauhan R
   5. Renard BY
   6. Steen JA
   7. Steen H

   (2020) ProteomeXchange

   ID PXD018411. FLEXIQuant-LF: Robust Regression to quantify protein modification extent in label-free proteomics data.

   http://proteomecentral.proteomexchange.org/cgi/GetDataset?ID=PXD018411

1. 1. Bruderer R
   2. Bernhardt OM
   3. Gandhi T
   4. Xuan Y
   5. Sondermann J
   6. Schmidt M
   7. Gomez-Varela D
   8. Reiter L

   (2017) ProteomeXchange

   ID PXD005573. Optimization of Experimental Parameters in Data-Independent Mass Spectrometry Significantly Increases Depth and Reproducibility of Results.

   http://proteomecentral.proteomexchange.org/cgi/GetDataset?ID=PXD005573

1. 1. Bagwan N
   2. Bonzon-Kulichenko E
   3. Calvo E
   4. Lechuga-Vieco AV
   5. Michalakopoulos S
   6. Trevisan-Herraz M
   7. Ezkurdia I
   8. Rodríguez JM
   9. Magni R
   10. Latorre-Pellicer A
   11. Enríquez JA
   12. Vázquez J

   (2018) Comprehensive quantification of the modified proteome reveals oxidative heart damage in mitochondrial heteroplasmy

   Cell Reports 23:3685–3697.

   https://doi.org/10.1016/j.celrep.2018.05.080

   • PubMed
   • Google Scholar
2. 1. Beck M
   2. Schmidt A
   3. Malmstroem J
   4. Claassen M
   5. Ori A
   6. Szymborska A
   7. Herzog F
   8. Rinner O
   9. Ellenberg J
   10. Aebersold R

   (2011) The quantitative proteome of a human cell line

   Molecular Systems Biology 7:549.

   https://doi.org/10.1038/msb.2011.82

   • PubMed
   • Google Scholar
3. 1. Bruderer R
   2. Bernhardt OM
   3. Gandhi T
   4. Miladinović SM
   5. Cheng LY
   6. Messner S
   7. Ehrenberger T
   8. Zanotelli V
   9. Butscheid Y
   10. Escher C
   11. Vitek O
   12. Rinner O
   13. Reiter L

   (2015) Extending the limits of quantitative proteome profiling with data-independent acquisition and application to acetaminophen-treated three-dimensional liver microtissues

   Molecular & Cellular Proteomics 14:1400–1410.

   https://doi.org/10.1074/mcp.M114.044305

   • PubMed
   • Google Scholar
4. 1. Bruderer R
   2. Bernhardt OM
   3. Gandhi T
   4. Xuan Y
   5. Sondermann J
   6. Schmidt M
   7. Gomez-Varela D
   8. Reiter L

   (2017) Optimization of experimental parameters in Data-Independent mass spectrometry significantly increases depth and reproducibility of results

   Molecular & Cellular Proteomics 16:2296–2309.

   https://doi.org/10.1074/mcp.RA117.000314

   • PubMed
   • Google Scholar
5. 1. Choudhary C
   2. Weinert BT
   3. Nishida Y
   4. Verdin E
   5. Mann M

   (2014) The growing landscape of lysine acetylation links metabolism and cell signalling

   Nature Reviews Molecular Cell Biology 15:536–550.

   https://doi.org/10.1038/nrm3841

   • PubMed
   • Google Scholar
6. 1. Cox J
   2. Mann M

   (2008) MaxQuant enables high peptide identification rates, individualized p.p.b.-range mass accuracies and proteome-wide protein quantification

   Nature Biotechnology 26:1367–1372.

   https://doi.org/10.1038/nbt.1511

   • PubMed
   • Google Scholar
7. 1. Devabhaktuni A
   2. Lin S
   3. Zhang L
   4. Swaminathan K
   5. Gonzalez CG
   6. Olsson N
   7. Pearlman SM
   8. Rawson K
   9. Elias JE

   (2019) TagGraph reveals vast protein modification landscapes from large tandem mass spectrometry datasets

   Nature Biotechnology 37:469–479.

   https://doi.org/10.1038/s41587-019-0067-5

   • PubMed
   • Google Scholar
8. 1. Escher C
   2. Reiter L
   3. MacLean B
   4. Ossola R
   5. Herzog F
   6. Chilton J
   7. MacCoss MJ
   8. Rinner O

   (2012) Using iRT, a normalized retention time for more targeted measurement of peptides

   Proteomics 12:1111–1121.

   https://doi.org/10.1002/pmic.201100463

   • PubMed
   • Google Scholar
9. Software

   1. Espoo F

   (2019) QT Designer, version 5.13.0

   The QT Company.

   https://doc.qt.io/qt-5/designer-to-know.html
10. 1. Fang G
    2. Yu H
    3. Kirschner W

    (1999) Control of mitotic transitions by the anaphase–promoting complex

    Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences 354:1583–1590.

    https://doi.org/10.1098/rstb.1999.0502

    • Google Scholar
11. 1. Fischler MA
    2. Bolles RC

    (1981) Random sample consensus

    Communications of the ACM 24:381–395.

    https://doi.org/10.1145/358669.358692

    • Google Scholar
12. 1. Gerber SA
    2. Rush J
    3. Stemman O
    4. Kirschner MW
    5. Gygi SP

    (2003) Absolute quantification of proteins and phosphoproteins from cell lysates by tandem MS

    PNAS 100:6940–6945.

    https://doi.org/10.1073/pnas.0832254100

    • PubMed
    • Google Scholar
13. 1. Gillet LC
    2. Navarro P
    3. Tate S
    4. Röst H
    5. Selevsek N
    6. Reiter L
    7. Bonner R
    8. Aebersold R

    (2012) Targeted data extraction of the MS/MS spectra generated by data-independent acquisition: a new concept for consistent and accurate proteome analysis

    Molecular & Cellular Proteomics 11:O111.016717.

    https://doi.org/10.1074/mcp.O111.016717

    • PubMed
    • Google Scholar
14. 1. Grimsrud PA
    2. Swaney DL
    3. Wenger CD
    4. Beauchene NA
    5. Coon JJ

    (2010) Phosphoproteomics for the masses

    ACS Chemical Biology 5:105–119.

    https://doi.org/10.1021/cb900277e

    • PubMed
    • Google Scholar
15. 1. Hornbeck PV
    2. Chabra I
    3. Kornhauser JM
    4. Skrzypek E
    5. Zhang B

    (2004) PhosphoSite: a bioinformatics resource dedicated to physiological protein phosphorylation

    Proteomics 4:1551–1561.

    https://doi.org/10.1002/pmic.200300772

    • PubMed
    • Google Scholar
16. 1. Hornbeck PV
    2. Zhang B
    3. Murray B
    4. Kornhauser JM
    5. Latham V
    6. Skrzypek E

    (2015) PhosphoSitePlus, 2014: mutations, PTMs and recalibrations

    Nucleic Acids Research 43:D512–D520.

    https://doi.org/10.1093/nar/gku1267

    • PubMed
    • Google Scholar
17. 1. Hunter JD

    (2007) Matplotlib: a 2D graphics environment

    Computing in Science & Engineering 9:90–95.

    https://doi.org/10.1109/MCSE.2007.55

    • Google Scholar
18. Software

    1. Jones E

    (2001)

    SciPy: Open Source Scientific Tools for Python

    SciPy: Open Source Scientific Tools for Python.
19. Software

    1. Kahnert K
    2. Steen Lab

    (2020) FLEXIQuant-LF, version swh:1:rev:4ea3945f86ba477227c89e9ced75fc23751355ac

    Software Heritage.

    https://archive.softwareheritage.org/swh:1:rev:4ea3945f86ba477227c89e9ced75fc23751355ac;origin=https://github.com/SteenOmicsLab/FLEXIQuantLF;visit=swh:1:snp:b5333babb41e87a963ae8d0ef8d64b3a038eb906
20. 1. Kim W
    2. Bennett EJ
    3. Huttlin EL
    4. Guo A
    5. Li J
    6. Possemato A
    7. Sowa ME
    8. Rad R
    9. Rush J
    10. Comb MJ
    11. Harper JW
    12. Gygi SP

    (2011) Systematic and quantitative assessment of the ubiquitin-modified proteome

    Molecular Cell 44:325–340.

    https://doi.org/10.1016/j.molcel.2011.08.025

    • PubMed
    • Google Scholar
21. 1. Kong AT
    2. Leprevost FV
    3. Avtonomov DM
    4. Mellacheruvu D
    5. Nesvizhskii AI

    (2017) MSFragger: ultrafast and comprehensive peptide identification in mass spectrometry-based proteomics

    Nature Methods 14:513–520.

    https://doi.org/10.1038/nmeth.4256

    • PubMed
    • Google Scholar
22. 1. Lundby A
    2. Secher A
    3. Lage K
    4. Nordsborg NB
    5. Dmytriyev A
    6. Lundby C
    7. Olsen JV

    (2012) Quantitative maps of protein phosphorylation sites across 14 different rat organs and tissues

    Nature Communications 3:ncomms1871.

    https://doi.org/10.1038/ncomms1871

    • Google Scholar
23. 1. Lundby A
    2. Andersen MN
    3. Steffensen AB
    4. Horn H
    5. Kelstrup CD
    6. Francavilla C
    7. Jensen LJ
    8. Schmitt N
    9. Thomsen MB
    10. Olsen JV

    (2013) In vivo phosphoproteomics analysis reveals the cardiac targets of β-adrenergic receptor signaling

    Science Signaling 6:rs11.

    https://doi.org/10.1126/scisignal.2003506

    • PubMed
    • Google Scholar
24. 1. Mair W
    2. Muntel J
    3. Tepper K
    4. Tang S
    5. Biernat J
    6. Seeley WW
    7. Kosik KS
    8. Mandelkow E
    9. Steen H
    10. Steen JA

    (2016) FLEXITau: quantifying Post-translational modifications of tau protein in Vitro and in Human Disease

    Analytical Chemistry 88:3704–3714.

    https://doi.org/10.1021/acs.analchem.5b04509

    • PubMed
    • Google Scholar
25. Conference

    1. McKinney W

    (2010)

    Data structures for statistical computing in Python

    Proceedings of the 9th Python in Science Conference. pp. 51–56.

    • Google Scholar
26. 1. Melo-Braga MN
    2. Schulz M
    3. Liu Q
    4. Swistowski A
    5. Palmisano G
    6. Engholm-Keller K
    7. Jakobsen L
    8. Zeng X
    9. Larsen MR

    (2014) Comprehensive quantitative comparison of the membrane proteome, Phosphoproteome, and sialiome of human embryonic and neural stem cells

    Molecular & Cellular Proteomics 13:311–328.

    https://doi.org/10.1074/mcp.M112.026898

    • PubMed
    • Google Scholar
27. 1. Pedregosa F

    (2011)

    Scikit-learn: machine learning in Python

    Journal of Machine Learning Research 12:2825–2830.

    • Google Scholar
28. 1. Pinkse MW
    2. Uitto PM
    3. Hilhorst MJ
    4. Ooms B
    5. Heck AJ

    (2004) Selective isolation at the femtomole level of phosphopeptides from proteolytic digests using 2D-NanoLC-ESI-MS/MS and titanium oxide precolumns

    Analytical Chemistry 76:3935–3943.

    https://doi.org/10.1021/ac0498617

    • PubMed
    • Google Scholar
29. Software

    1. PyInstaller Development Team

    (2019) PyInstaller, version 3.6

    PyInstaller.

    https://www.pyinstaller.org/
30. Software

    1. Riverbank Computing Limited

    (2019) PyQT, version 5.13.2

    PyQT.

    https://www.riverbankcomputing.com/software/pyqt/download5
31. 1. Rush J
    2. Moritz A
    3. Lee KA
    4. Guo A
    5. Goss VL
    6. Spek EJ
    7. Zhang H
    8. Zha XM
    9. Polakiewicz RD
    10. Comb MJ

    (2005) Immunoaffinity profiling of tyrosine phosphorylation in Cancer cells

    Nature Biotechnology 23:94–101.

    https://doi.org/10.1038/nbt1046

    • PubMed
    • Google Scholar
32. 1. Sharma K
    2. D'Souza RC
    3. Tyanova S
    4. Schaab C
    5. Wiśniewski JR
    6. Cox J
    7. Mann M

    (2014) Ultradeep human phosphoproteome reveals a distinct regulatory nature of tyr and ser/Thr-based signaling

    Cell Reports 8:1583–1594.

    https://doi.org/10.1016/j.celrep.2014.07.036

    • PubMed
    • Google Scholar
33. 1. Singh S
    2. Springer M
    3. Steen J
    4. Kirschner MW
    5. Steen H

    (2009) FLEXIQuant: a novel tool for the absolute quantification of proteins, and the simultaneous identification and quantification of potentially modified peptides

    Journal of Proteome Research 8:2201–2210.

    https://doi.org/10.1021/pr800654s

    • PubMed
    • Google Scholar
34. 1. Singh S
    2. Kirchner M
    3. Steen JA
    4. Steen H

    (2012a) A practical guide to the FLEXIQuant method

    Methods in Molecular Biology 893:295–319.

    https://doi.org/10.1007/978-1-61779-885-6_19

    • PubMed
    • Google Scholar
35. 1. Singh SA
    2. Winter D
    3. Bilimoria PM
    4. Bonni A
    5. Steen H
    6. Steen JA

    (2012b) FLEXIQinase, a mass spectrometry-based assay, to unveil multikinase mechanisms

    Nature Methods 9:504–508.

    https://doi.org/10.1038/nmeth.1970

    • PubMed
    • Google Scholar
36. 1. Singh SA
    2. Winter D
    3. Kirchner M
    4. Chauhan R
    5. Ahmed S
    6. Ozlu N
    7. Tzur A
    8. Steen JA
    9. Steen H

    (2014) Co-regulation proteomics reveals substrates and mechanisms of APC/C-dependent degradation

    The EMBO Journal 33:385–399.

    https://doi.org/10.1002/embj.201385876

    • PubMed
    • Google Scholar
37. 1. Solntsev SK
    2. Shortreed MR
    3. Frey BL
    4. Smith LM

    (2018) Enhanced global Post-translational modification discovery with MetaMorpheus

    Journal of Proteome Research 17:1844–1851.

    https://doi.org/10.1021/acs.jproteome.7b00873

    • PubMed
    • Google Scholar
38. 1. Steen JA
    2. Steen H
    3. Georgi A
    4. Parker K
    5. Springer M
    6. Kirchner M
    7. Hamprecht F
    8. Kirschner MW

    (2008) Different phosphorylation states of the anaphase promoting complex in response to antimitotic drugs: a quantitative proteomic analysis

    PNAS 105:6069–6074.

    https://doi.org/10.1073/pnas.0709807104

    • PubMed
    • Google Scholar
39. Software

    1. The Pallets Projects

    (2020) Click 7.0

    Click 7.0.

    https://palletsprojects.com/p/click/
40. Software

    1. The QT Company

    (2019) QT 5.9.7

    QT 5.9.7.

    https://www.qt.io/download-open-source
41. 1. Udeshi ND
    2. Svinkina T
    3. Mertins P
    4. Kuhn E
    5. Mani DR
    6. Qiao JW
    7. Carr SA

    (2013) Refined preparation and use of anti-diglycine remnant (K-ε-GG) antibody enables routine quantification of 10,000s of ubiquitination sites in single proteomics experiments

    Molecular & Cellular Proteomics 12:825–831.

    https://doi.org/10.1074/mcp.O112.027094

    • PubMed
    • Google Scholar
42. 1. van der Walt S
    2. Colbert SC
    3. Varoquaux G

    (2011) The NumPy array: a structure for efficient numerical computation

    Computing in Science & Engineering 13:22–30.

    https://doi.org/10.1109/MCSE.2011.37

    • Google Scholar
43. Software

    1. Waskom M

    (2017)

    Mwaskom/Seaborn: V0.8.1

    Mwaskom/Seaborn: V0.8.1.
44. 1. Wold F

    (1981) In vivo chemical modification of proteins (post-translational modification)

    Annual Review of Biochemistry 50:783–814.

    https://doi.org/10.1146/annurev.bi.50.070181.004031

    • PubMed
    • Google Scholar
45. 1. Wu R
    2. Haas W
    3. Dephoure N
    4. Huttlin EL
    5. Zhai B
    6. Sowa ME
    7. Gygi SP

    (2011) A large-scale method to measure absolute protein phosphorylation stoichiometries

    Nature Methods 8:677–683.

    https://doi.org/10.1038/nmeth.1636

    • PubMed
    • Google Scholar
46. 1. Yamagata A
    2. Kristensen DB
    3. Takeda Y
    4. Miyamoto Y
    5. Okada K
    6. Inamatsu M
    7. Yoshizato K

    (2002) Mapping of phosphorylated proteins on two-dimensional polyacrylamide gels using protein phosphatase

    Proteomics 2:1267–1276.

    https://doi.org/10.1002/1615-9861(200209)2:9<1267::AID-PROT1267>3.0.CO;2-R

    • PubMed
    • Google Scholar
47. 1. Zhang X
    2. Abreu JG
    3. Yokota C
    4. MacDonald BT
    5. Singh S
    6. Coburn KL
    7. Cheong SM
    8. Zhang MM
    9. Ye QZ
    10. Hang HC
    11. Steen H
    12. He X

    (2012) Tiki1 is required for head formation via wnt cleavage-oxidation and inactivation

    Cell 149:1565–1577.

    https://doi.org/10.1016/j.cell.2012.04.039

    • PubMed
    • Google Scholar
48. 1. Zhou H
    2. Low TY
    3. Hennrich ML
    4. van der Toorn H
    5. Schwend T
    6. Zou H
    7. Mohammed S
    8. Heck AJ

    (2011) Enhancing the identification of phosphopeptides from putative basophilic kinase substrates using ti (IV) based IMAC enrichment

    Molecular & Cellular Proteomics 10:M110.006452.

    https://doi.org/10.1074/mcp.M110.006452

    • PubMed
    • Google Scholar

Author details

1. Christoph N Schlaffner

   1. F.M. Kirby Neurobiology Center, Boston Children’s Hospital, Boston, United States
   2. Department of Neurology, Harvard Medical School, Boston, United States

   Contribution

   Conceptualization, Data curation, Formal analysis, Supervision, Validation, Investigation, Visualization, Methodology, Writing - review and editing

   Contributed equally with

   Konstantin Kahnert and Jan Muntel

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-2717-3406

2. Konstantin Kahnert

   1. Department of Pathology, Boston Children’s Hospital, Boston, United States
   2. Bioinformatics Unit (MF1), Robert Koch Institute, Berlin, Germany
   3. Department of Medical Biotechnology, Institute of Biotechnology, Technische Universität Berlin, Berlin, Germany

   Contribution

   Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing - original draft

   Contributed equally with

   Christoph N Schlaffner and Jan Muntel

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-8454-4894

3. Jan Muntel

   1. Department of Pathology, Boston Children’s Hospital, Boston, United States
   2. Department of Pathology, Harvard Medical School, Boston, United States

   Present address

   Biognosys AG, Schlieren, Switzerland

   Contribution

   Investigation, Visualization, Writing - original draft

   Contributed equally with

   Christoph N Schlaffner and Konstantin Kahnert

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-2320-5829

4. Ruchi Chauhan

   F.M. Kirby Neurobiology Center, Boston Children’s Hospital, Boston, United States

   Present address

   Grousbeck Gene Therapy Center, Gene Transfer Vector Core, Massachusetts Eye and Ear Infirmary/Harvard Medical School, Boston, United States

   Contribution

   Investigation

   Competing interests

   No competing interests declared

5. Bernhard Y Renard

   1. Bioinformatics Unit (MF1), Robert Koch Institute, Berlin, Germany
   2. Data Analytics and Computational Statistics, Hasso-Plattner-Institute, Faculty of Digital Engineering, University of Potsdam, Potsdam, Germany

   Contribution

   Conceptualization, Supervision, Funding acquisition, Validation, Methodology, Writing - review and editing

   Competing interests

   No competing interests declared

6. Judith A Steen

   1. F.M. Kirby Neurobiology Center, Boston Children’s Hospital, Boston, United States
   2. Department of Neurology, Harvard Medical School, Boston, United States

   Contribution

   Conceptualization, Resources, Funding acquisition

   Competing interests

   No competing interests declared

7. Hanno Steen

   1. Department of Pathology, Boston Children’s Hospital, Boston, United States
   2. Department of Pathology, Harvard Medical School, Boston, United States
   3. Precision Vaccines Program, Boston Children’s Hospital, Boston, United States

   Contribution

   Conceptualization, Resources, Supervision, Funding acquisition, Validation, Methodology, Project administration, Writing - review and editing

   For correspondence

   Hanno.Steen@childrens.harvard.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-0179-6648

• 1,070

  views

• 138

  downloads

• 2

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.