Quantifying protein interaction dynamics by SWATH mass spectrometry: application to the 14-3-3 system
Summary (5 min read)
Introduction
- Cellular functions are rarely attributed solely to a single molecule.
- The 14-3-3 ligand motif therefore facilitates direct physical interactions but in many cases the ligand itself is complexed to other proteins, and represents the attachment point to a multi-subunit complex.
- As such, the number of AP-MS studies which have attempted to detect changes in protein-protein interaction networks in perturbed systems remains small, at least with respect to projects engaged in large scale interaction mapping 13, 14, 22 .
- A recent development has seen the application of targeted proteomics via SRM (selected reaction monitoring) to differential interaction proteomics in an attempt to address some of these limitations 28, 29 .
- SWATH-MS provides SRM-like performance in terms of quantitative accuracy, data completeness and dynamic range without specifying target peptides prior to data acquisition.
Characterization of the 14-3-3β signaling system
- The authors chose IGF1 stimulation to perturb the 14-3-3 interactome because prior data showed that substrates of AKT kinase, a central mediator of insulin-IGF1 signaling, frequently bind 14-3-3 proteins 5 .
- As serum starvation was not sufficient to abolish AKT activation, the authors included a 60 minute pretreatment with the reversible PI3K inhibitor LY294002 to achieve a PI3K inactive ground state.
- Having established the 14-3-3β expression system, in which the bait was expressed ~3-4 times below the endogenous level (Supplementary Fig. 1b ), and confirmed that the stimulation was effective, the authors performed APs of 14-3-3β from the time course.
- The authors characterized these samples by shotgun mass spectrometry leading to the identification of 31,509 unique peptide sequences (FDR 0.2 %) corresponding to 2,532 unique proteins (FDR 0.3%) across the entire experiment.
Generation and analysis of AP-SWATH maps
- The 18 affinity purified 14-3-3β samples from 6 conditions in the time course (Fig. 1 ), plus 9 GFP control APs, were subjected to SWATH-MS (Online Methods).
- The authors analysed the AP-SWATH data using OpenSWATH, an open source software for automated targeted analysis of DIA datasets (unpublished data H. R., G. R.; www.openswath.org).
- The targeted analysis workflow consists of extracting ion chromatograms for groups of fragment ions from a given peptide precursor in the appropriate MS2 SWATH map, and scoring peak groups detected in these chromatograms with respect to prior knowledge from peptide MS2 spectral libraries (Supplementary Fig. 2 and 3 and Online Methods).
- The result was an essentially complete quantitative data matrix, displayed as a heat map (Fig. 2 -source data is provided in the supplementary material), for all 1,967 proteins in all conditions (see Supplementary Results for information on quantitative reproducibility).
- This data matrix formed the basis for downstream quantitative comparisons relating firstly to filtering of non-specific contaminants and identifying high confidence 14-3-3β interactions, and secondly for examining the dynamics of 14-3-3β interacting proteins over the perturbation.
Time-resolved quantification of a perturbed interaction network
- The completeness of the measurements enabled facile calculation of statistics 38 to determine the remodeling of the 14-3-3β interactome after IGF1 stimulation (see Supplementary Table 4 ).
- The temporal profiles of 14-3-3β binding to 6 selected representative proteins are shown (Fig. 4a-f ).
- Consistent with this, the authors found a dramatic reduction of AKTS1 binding to 14-3-3β after treatment with a PI3K-AKT inhibitor and an immediate and sustained regain of binding after 1 minute of treatment with IGF1.
- This, combined with the absence of mTORC1 specific subunits and radically different time course profiles of AKTS1 and mTOR suggests that the bulk of mTOR kinase in the 14-3-3β.
- This observation is based on previous literature knowledge 39 in addition to AP-SWATH data, however, the authors would emphasize that such literature corroborative conclusions could not have been made using static AP-MS maps.
Dynamic range of 14-3-3β interactome
- Given the large number of 14-3-3β interacting proteins detected, the authors elected to estimate the range of their abundances via the 'best flyer peptide' approach 43 .
- As SWATH-MS data processed via targeted extraction has the same structure as SRM data, in addition to being very highly correlated quantitatively with SRM 35, 44 , the authors used an implementation of this strategy recently validated for SRM 45 .
- The authors plotted the estimated log10 protein abundances for the 14-3-3β interacting proteins ordered from high to low abundance (Fig. 6 ., data in Supplementary Table 6 ).
- While in their study the abundance of the interacting proteins containing the 14-3-3 binding motif spanned 4+ order of magnitude (Fig. 6 ), the matched 14-3-3β literature interactions showed a clear bias towards the upper 2.5 orders of abundance.
- This analysis suggests that their method is extremely sensitive in detecting interactions, particularly in highly complex scaffold protein APs.
Discussion
- The authors have developed a strategy for quantifying time-resolved remodeling of protein-protein interactions in APs.
- The method provides quantitative data for confidently identifying true protein-protein interactions via comparisons with control APs, and notably, for following dynamic changes in protein-protein interactions in perturbed systems.
- Secondly, there is no limit on the number of peptides that can be analyzed via AP-SWATH, whereas AP-SRM studies will struggle beyond ~100 proteins.
- Finally, as AP-SWATH data contains an MS2 record for essentially the entire tryptic peptide space, there is potential for re-mining of the data post-acquisition using additional spectral libraries from sources such as synthetic peptides 35, 49 , deep shotgun characterization, or enrichment of PTM containing peptides (Supplementary Results on phosphopeptide analysis).
- This type of iterative reanalysis is unique to SWATH or DIA data and provides a compelling argument for further development in this area.
Figure 2 -Complete quantitative data matrix from AP-SWATH
- Quantitative data for 1,967 proteins was extracted from AP-SWATH data for 27 samples including 18 x 14-3-3β AP samples and 9 x GFP control AP samples.
- The proteins were clustered in the vertical direction hierarchically using Minkowski distance.
- This data matrix formed the basis for all downstream quantitative comparisons including filtering of non-specific contaminant proteins (Fig. 3 ) and time course protein interaction dynamics (Fig. 4 ).
Figure 3 -14-3-3β protein interactions are confidently identified by enrichment over control APs
- (a) A volcano plot showing log2 fold change is plotted against -log10 adjusted p-value for 14-3-3β.
- AP samples versus samples generated from an irrelevant bait (GFP).
- Data points highlighted in red in the upper right section represent proteins which display and enrichment in 14-3-3β.
- The bait abundances (black circles) of the negative control kinase group spanned the same abundance range as the set of detected 14-3-3β interactors.
- The number of replicates is indicated in brackets and the error bars represent +/-1 standard deviation.
Figure 6 -Dynamic range of 14-3-3β interacting proteins
- The log10 abundance of 567 high confidence 14-3-3β interacting proteins calculated using the 'best flyer peptide' approach is plotted (black circles).
- The same set of proteins is re-plotted with an offset 4 additional times to highlight 14-3-3 protein isoforms (purple circles), proteins containing a 14-3-3 ligand motif (blue circles), 14-3-3β protein interactions previously detected in the literature (turquoise circles), and the set of protein kinases verified by reciprocal AP-MS (orange circles).
Cell line generation
- HEK293 cells inducibly expressing n-terminally SH tagged (tandem streptavidin binding peptide tag + hemagglutinin tag) 14-3-3β were generated essentially as described previously 51 .
- (Cells have not been recently tested for mycoplasma contamination).
- The 14-3-3β ORF, obtained in pDONR223 from the Gateway compatible human orfeome library (hORFeome v5.1, Open Biosystems), was recombined into the pcDNA5/FRT/TO/SH/GW destination vector using the LR Clonase II kit (Life Technologies).
- Flp-In T-REx HEK293 cells (Life Technologies) containing a single genomic FRT site and stably expressing the tet repressor were cultured in DMEM (10 % FCS, 50 µg/ml penicillin, 50 µg/ml streptomycin), and co-transfected using Fugene 6 transfection reagent with the 14-3-3β expression plasmid and pOG44 vector (Life Technologies) for coexpression of the Flp-recombinase.
Time course treatment of cells
- Expression of the bait in 40 % confluent cells was induced for 24 hours with media containing 1.33 µg/ml doxycycline .
- The cells were then washed in PBS, and then serum starved overnight with serum-free DMEM containing doxycycline.
- The time course treatment was carried out in biological triplicate, using 2 x 15 cm dishes per condition, in serial passages of the HEK-Flp-1433B cells.
Affinity purification
- The 14-3-3β protein complexes were obtained by single step AP via the tandem streptavidin binding peptide sequence included in the SH tag.
- APs for phosphoenrichment were as above, except that forty culture plates of 15 cm diameter were used as starting material.
- Dynamics for 6 selected 14-3-3β interactors was verified by Western blotting .
SWATH mass spectrometry
- SWATH mass spectra were acquired using an AB Sciex 5600 TripleTOF mass spectrometer interfaced to an Eksigent NanoLC Ultra 2D Plus HPLC system essentially as previously described 54 .
- For SWATH-MS-based experiments, the mass spectrometer was operated in a looped product ion mode.
- The instrument was specifically tuned to allow a quadrupole resolution of 25 m/z mass selection.
- Using an isolation width of 26 m/z (containing 1 m/z for the window overlap), a set of 32 overlapping windows was constructed covering the precursor mass range of 400-1200 m/z.
- The collision energy for each window was determined based on the calculation for a charge 2+ ion centered upon the window with a spread of 15.
Shotgun mass spectrometry
- In addition to SWATH-MS acquisition, every sample was also analysed using classical shotgun data acquisition with a TripleTOF 5600.
- The chromatographic parameters were as above for SWATH-MS.
- The 20 most intense precursors with charge state 2-5 were selected for fragmentation and MS2 spectra were collected in the range 50-2000 m/z for 100 ms, and precursor ions were excluded from reselection for 15 seconds.
- Reciprocal APs of protein kinases were analysed using an Orbitrap Elite with MS1 resolving power set to 240,000 and the top 10 precursors selected for MS2 in the ion trap after CID activation.
Spectral library and target assay construction
- Profile mode wiff files from shotgun data acquisition were centroided and converted to mzML using the AB Sciex Data Converter v1.3, and further converted to mzXML using ProteoWizard 55 MSConvert v3.04.238.
- Mass spectra were queried against the canonical UniProt complete proteome database for human (July 2011) appended with common contaminants and reversed sequence decoys (40,548 protein sequences including decoys) using XTandem with kscore plugin 56 , and additionally XTandem with native scoring 57 .
- Carbamidomethyl was set as fixed modification for cysteines, and methionine oxidation and phosphorylation on serine/threonine/tyrosine were set as variable modification.
- A non-redundant consensus spectral library was then constructed from the redundant library using SpectraST.
SWATH-MS targeted data extraction
- SWATH-MS wiff files were first converted to profile mzXML using Proteowizard msconvert 55 v3.0.3316.
- As the precursor isolation window scheme was wrongly retrived by msconvert, a custom python script (fix_swath_windows.py) was used to correct this.
- For easier file access, the mzXML files were split into 33 individual files (32 SWATH MS2 files + 1 MS1 file) using the custom python script (split_mzXML_intoSwath.py).
- Peak groups from the extracted fragment ion chromatograms were formed and scored based on their elution profiles, similarity to the target assay in terms of RT and relative fragment ion intensity, as well as features from the full MS2 SWATH spectrum extracted at the chromatographic peak apex .
- Identification of phosphopeptides from AP-SWATH data using phosphopeptide specific libraries was also performed using OpenSWATH.
Data analysis and statistics
- Peptide features (i.e. peptides in a given charge state) which met the 1 % FDR threshold in 3 out of 3 biological replicates for any experimental condition were retained, and intensities for these peptides/transitions across the experiment were used for further analysis.
- For comparisons between 14-3-3β and GFP control purifications, data were not normalized.
- Protein abundances were estimated by summing the 2 most intense transitions from the 3 most intense peptides for a protein as previously described for SRM data 65 .
- The authors used a non-strict filter on the quantitative data to remove proteins which showed no change (-0.5 > log2 FC > 0.5, adjusted pvalue < 0.05 at any time point compared with t=0 minutes), and then performed time series clustering using the open source R package Mfuzz 67 using median fold change as input.
- Literature protein-protein interactions were retrieved from the PINA 70 and iRefWeb 71 interaction databases.
Did you find this useful? Give us your feedback
Citations
1,074 citations
Cites background from "Quantifying protein interaction dyn..."
...MS-based methods cover more than four orders of magnitude of interaction stoichiometry (Collins et al., 2013), and our low-stringency biochemicalworkflow ideally harnesses this sensitivity....
[...]
...However, in practice, measured stoichiometries between preys and baits span orders of magnitude (Collins et al., 2013; Hauri et al., 2013)....
[...]
833 citations
728 citations
Cites methods from "Quantifying protein interaction dyn..."
...Targeted proteomics (SWATH; Collins et al., 2013) was used to orthogonally quantify the acetylated peptide with synthetic peptides as the internal standard....
[...]
613 citations
Cites background from "Quantifying protein interaction dyn..."
...…studies (Liu et al, 2015; Okada et al, 2016; Williams et al, 2016), clinical drug/perturbation studies (preprint: Litichevskiy et al, 2018; Tan et al, 2017; Keam et al, 2018) or exploratory basic research (Collins et al, 2013; Lambert et al, 2013; Parker et al, 2015a; Schubert et al, 2015b)....
[...]
536 citations
References
31,015 citations
4,861 citations
3,794 citations
3,604 citations
2,975 citations
Related Papers (5)
Frequently Asked Questions (16)
Q2. What is the effect of phosphorylation of AKTS1 on mTORC?
It is thought that phosphorylation of AKTS1 by AKT induces 14-3-3 binding of AKTS1 leading to dissociation of AKTS1 from mTORC1 and subsequent reduction of mTORC1 inhibition.
Q3. Why did the authors choose to stimulate the 14-3-3 interactome?
The authors chose IGF1 stimulation to perturb the 14-3-3 interactome because prior data showed that substrates of AKT kinase, a central mediator of insulin-IGF1 signaling, frequently bind 14-3-3 proteins5.
Q4. How many unique proteins were identified across the entire experiment?
The authors characterized these samples by shotgun mass spectrometry leading to the identification of 31,509 unique peptide sequences (FDR 0.2 %) corresponding to 2,532 unique proteins (FDR 0.3%) across the entire experiment.
Q5. What was the way to estimate the false discovery rate?
the optimal separation between true and false peak groups was achieved using a linear model training and the score distribution from the shuffled decoy assays was used to estimate the false discovery rate (mProphet62).
Q6. What could be used to help define protein complexes in the cell?
Combination with orthogonal information, such as might be provided from improvements in native protein complex separation strategies47, or proximity labeling48, could prove instrumental toward this end.
Q7. How was the collision energy determined for each window?
The collision energy for each window was determined based on the calculation for a charge 2+ ion centered upon the window with a spread of 15.
Q8. What are the corresponding plots for GFP purifications?
The corresponding plot area for GFP purifications is essentially devoid of proteins other than the control bait itself (GFP), 2 zinc finger proteins (Q9H5H4 and O60290) and another DNA binding protein (Q15424) which are very consistently enriched in control purifications for unknown reasons.
Q9. How many peptide features were identified in the AP-SWATH dataset?
From the 14-3-3 AP-SWATH dataset the authors identified 19,123 peptide features (corresponding to 1,967 proteins) at 1% FDR in 3 out of 3 biological replicates from at least one experimental condition (Supplementary Fig. 4).
Q10. What was the correct python script to use to correct this?
As the precursor isolation window scheme was wrongly retrived by msconvert, a custom python script (fix_swath_windows.py) was used to correct this.
Q11. What did the authors do to characterize the dynamics of the 14-3-3 interactome?
Having used the AP-SWATH quantitative data to identify 567 high confidence 14-3-3β interacting proteins, the authors went on to characterize their dynamics.
Q12. What is the largest interactome for a single bait?
To their knowledge, this study represents the largest reported interactome for a single bait, and further indicates that at least 2.8 % of the proteome can be engaged by 14-3-3 dimers containing the 14-3-3β isoform.
Q13. what are the proteins in panels (a) – (h)?
The 5 selected proteins in panels (a) – (e) are typical of the 5 clusters found in panels (g) – (h). IGF1R shown in panel (f) is nominally part of cluster 2, however, the time course behavior is not completely consistent as IGF1R shows no response to the PI3K inhibitor unlike the other proteins shown in cluster 2. Panel (i) shows the enrichment of known or predicted AKT substrates in each of the 5 clusters compared to the background UniProt human proteome.
Q14. How did the authors determine the significance of the 14-3-3 interactor?
The authors found 14-3-3β as an interactor in 19 out of 21 kinases, whereas, no 14-3-3β was detected in APs from GFP or from 21 additional kinases selected as negative controls (Fig. 3b).
Q15. How many protein interactions were detected in the shotgun?
APs of log2 fold change > 2 and adjusted p-value < 0.01. (b) A subset of the detected protein interactions were verified by reciprocal AP-MS. 14-3-3β (red circles) was detected as a protein interaction in 19 of the 21 protein kinases analysed.
Q16. What was the python script used to split the mzXML files?
For easier file access, the mzXML files were split into 33 individual files (32 SWATH MS2 files + 1 MS1 file) using the custom python script (split_mzXML_intoSwath.py).