Super-resolution mass spectrometry enables rapid, accurate, and highly-multiplexed proteomics at the MS2-level

In tandem mass spectrometry (MS2)-based multiplexed quantitative proteomics, the complement reporter ion approaches (TMTc and TMTproC) were developed to eliminate the ratio-compression problem of conventional MS2 level approaches. Resolving all high m/z complement reporter ions (∼6.32 mDa spaced) requires mass resolution and scan speeds above the performance levels of Orbitrap™ instruments. Therefore, complement reporter ion quantification with TMT™/TMTpro™ reagents is currently limited to 5 out of 11 (TMT) or 9 out of 18 (TMTpro) channels (∼1 Da spaced). We first demonstrate that a Fusion™ Lumos™ Orbitrap™ can resolve 6.32 mDa spaced complement reporter ions with standard acquisition modes extended with 3-second transients. We then implemented a super-resolution mass spectrometry approach using the least-squares fitting (LSF) method for processing Orbitrap transients to achieve shotgun proteomics-compatible scan rates. The LSF performance resolves the 6.32 mDa doublets for all TMTproC channels in the standard mass range with transients as short as ∼108 ms (Orbitrap resolution setting of 50 000 at m/z 200). However, we observe a slight decrease in measurement precision compared to 1 Da spacing with the 108 ms transients. With 256 ms transients (resolution of 120 000 at m/z 200), coefficients of variation are essentially indistinguishable from 1 Da samples. We thus demonstrate the feasibility of highly-multiplexed, accurate, and precise shotgun-proteomics at the MS2 level.


Introduction
Multiplexed quantitative proteomics based on tandem mass spectrometry (MS2) is most commonly realized using peptides covalently labeled with isobaric tandem mass tags (iTRAQ, TMT, TMTpro). [1][2][3][4] It may simultaneously probe up to 18 sample-specific biological conditions. 5 The analytical benefits include the inherently high measurement precision, omission of missing values, and reduced experimental time per proteome. The corresponding applications have been exploited for basic and applied biological research, including translation regulation, 6 embryology, 7 breast cancer treatment, 8 and lung cancer metastasis. 9 The TMT isobaric reagents consist of a reactive moiety that covalently binds to a peptide and reporter and normalization parts that uniquely encode several quantitation channels with different numbers of heavy isotopes -13 C and 15 N. 2,3,10 The reporter parts are detected as singly charged reporter ions in the low, m/z 110 -140, mass range. The abundances of the reporter ions quantify peptides and associated proteins in each proteome. The reporter ion cluster contains channels separated by ~1 Da from other channels and doublets containing two channels encoded by the 13 C 14 N vs. 12 C 15 N isotopes, which differ by only ~6.32 mDa. 11 To baseline resolve the 6.32 mDa doublets in this (low m/z) mass range, a resolution value of about 60 000 is sufficient.
This level of resolution performance is readily provided by modern Orbitrap TM Fourier transform mass spectrometers (FTMS) at their resolution setting of 50 000 at m/z 200. 4 Despite its attractiveness, the MS2-level multiplexed proteomics suffers from measurement artifacts due to co-isolation and co-fragmentation of coeluting peptides when applied to complex samples. The MS3-level tandem MS approaches have been developed to address this limitation. [12][13][14] Adding one more MS step mitigates the peptide co-isolation issue but reduces sensitivity and throughput. 15 An alternative approach to tackle peptide co-isolation and the associated reporter ion ratio-compression problem is MS2-level complement reporter ion quantification. [16][17][18][19] Following this method, the species to be monitored are the complementary reporter ions -the remainders of the precursor ions labeled with the normalization parts and the reactive groups.
For TMTpro/TMT reagents, the complementary ion clusters will have one charge less than the isolated precursor ions and thus will be detected in an even higher mass range, up to 2 000 m/z, and sometimes beyond. The Orbitrap mass analyzer baseline resolves the complementary ion channels that differ by ~1 Da at a proteomics-grade scan rate. However, this reduces the number of encodable channels from 11 to 5 when using TMT and from 18 to 9 when using TMTpro. 18,19 Adding complementary reporting channels (TMTc/TMTproC clusters) with the 6.32 mDa doublets to the analysis increases the number of quantification channels but requires resolution performance that is not attainable even with the state-of-the-art Orbitraps. For example, to baseline resolve the 6.32 mDa doublets of the complementary reporter ions at m/z 2 000, a resolution value of 600 000 is necessary (which corresponds to a virtual resolution setting of 2 000 000 at m/z 200). However, this capability is not 5 available for general Orbitrap users. 20,21 On the other hand, user access to extended-duration transients can be readily provided using external highperformance data acquisition (DAQ) systems. [22][23][24] However, even if available, such long transients would drastically reduce the MS scan rates and are thus prohibitive for routine shotgun high-throughput proteomics applications using liquid chromatography (LC)-MS analysis. In addition, extended periods of ion oscillation in the mass analyzer may result in the loss of ion motion coherence, manifesting itself through pronounced transient decay and frequency shifts. 21,25,26 Finally, the FT-type signal processing methods, being the spectral estimators (as they yield profile-type data with the characteristic peak shape), are prone to frequency (m/z) and amplitude (ion abundance) errors of peak maxima. 27 The super-resolution (SR) algorithms for transient data processing were introduced for FTMS to overcome the abovementioned limitations. 28 Compared to FT-based approaches, a SR method is characterized by a less strict uncertainty principle for the resolution performance as a function of the detection period and may provide the same resolution as the FT methods using multiple-fold shorter transients. 27 Additionally, the difference in the uncertainty principles of the SR vs. FT methods reduces the peak m/z and amplitude measurement errors; this may translate to improved accuracy of quantitation, particularly in the TMTc/TMTproC doublets analysis.
Previously, we implemented the least-squares fitting (LSF) of time-domain signal processing as a powerful SR method for FTMS. 29 By definition, the LSF method is a targeted method that requires information on the (estimated) frequency of interest (m/z values) as an a priori knowledge. The latter matches 6 the case of multiplexed quantitative proteomics for both low m/z reporter and complementary ion detection. The output of LSF processing contains the amplitudes and refined frequencies of ion signal components in the transient (and, thus, the abundances and m/z values of peaks in the peaklist-type mass spectrum). Here, we combined the LSF capabilities to resolve the nearly isobaric ion signals using shorter time-domain transients with the highlyplexed TMTc/TMTproC workflow's ability to overcome the inherent MS2-level TMT/TMTpro strategy limitations.

Methods
Sample preparation. Yeast peptides from Saccharomyces cerevisiae cell lysates were prepared as described previously 18,30 and labeled with selected TMTpro reagents to yield 12 complement reporter ion channels (four 6.32 mDa doublets and four 1 Da singlets) or with selected TMT reagents to yield four channels (one 6.32 mDa doublet and two 1 Da singlets). In the case of TMTproC labeling, only 12 out of 18 channels were employed due to the degeneration (overlap) of the complement reported ions. Using only four channels for the TMTc approach validation was sufficient to evaluate method performance and applicability. durations, up to 3 seconds, were acquired using an external high-performance 7 DAQ system (FTMS Booster X2, Spectroswiss, Lausanne, Switzerland) in parallel to RAW file acquisition. The FTMS Booster X2 is an add-on to diverse Orbitrap models, with the previously reported implementation on the Q Exactive Orbitraps. [22][23][24] Notably, the employed DAQ system enables the acquisition of phased transients, i.e., transient signals whose sinusoidal components induced by the individual m/z ion rings trapped in the mass analyzer are all nearly in-phase.
In order to receive the analog transient signals from the Orbitrap preamplifier's output, the external DAQ system was interfaced through its two high-impedance analog inputs (i.e. a single differential input) to the two outputs of the Orbitrap pre-amplifier using T-splitters ( Figure S1, Supporting Information), as described elsewhere. 23,24 To increase resolution beyond conventional FTMS, the acquisition of longer transients was enabled by introducing "dummy" scans into the experimental logic without hardware or operational software modifications ( Figure S2, Supporting Information). 23,24 As a result, long time-domain transients, up to 6 s in this work, could be acquired.
Data processing. Assignment of MS 2 spectra was performed using the SEQUEST algorithm 31 by searching the data against the Uniprot Saccharomyces cerevisiae proteome. A peptide-level MS 2 spectral assignment false discovery rate of 1% was obtained by applying the target decoy strategy with linear discriminant analysis as described previously. 32 Super-resolution processing, absorption-mode FT, and calculations of reference m/z values for TMTc and TMTproC ions were performed using the Peak-by-Peak software 8 package (Spectroswiss) running on one or several 8-core desktop computers, eachequipped with 32 GB RAM and a compute unified device architecture (CUDA)-compatible graphics card (GPU) of the compute capability index ranging from 5.2 to 6.1. The super-resolution data processing was performed using the LSF approach similarly to the previously described fundamentals and implementation. 29 The basis functions of the LSF method correspond to the TMTproC (or TMTc) cluster and compose of the individual channels (~1 Da spacing) and doublets (~6.32 mDa spacing), Figure S3   Exactive HF instruments), that could provide sufficient quantitative precision at the acceptable scan speed. We also estimate that an additional 10%-20% time-domain transient period increase may be essential to tackle the real-life proteomic samples that present a wider range of the TMTc/TMTproC channel intensities and added matrix effects.
Several experimental factors may limit the LSF (and FT) performance of the TMTc/TMTproC approach. Coulombic interaction of the ions of interest is one example. Indeed, ion-ion interaction may result in peak interference, leading to frequency shifts and, eventually, peak coalescence. 39 These interactions can be particularly strong when ions are separated only by several mDa. 40 Therefore, a certain care should be taken to avoid the coalescence, especially for the doublets in TMTc/TMTproC clusters, by restricting the number of elementary charges for precursor ion isolation, via the automatic gain control (AGC) setting. 41 On the other hand, it has been shown that, thanks to the precursor ion signal split between many product ions, the coalescence may be less pronounced in a regular bottom-up proteomics experiment. 42 Further distribution of the total charge, due to multiplexing in TMTc/TMTproC, may help to loosen the coalescence threshold. Additionally, the risks of the detrimental effects in question could increase or reduce depending on the manufacturing quality and factory tuning of the mass analyzer that is installed in a particular Orbitrap instrument.
Another limitation of the LSF/FT performance for TMTc/TMTproC, as well as other similar multiplexing strategies, is the low signal-to-noise ratios of the peaks of interest. Processing time-domain transients with low intensity ion signals may result in CVs too high for accurate protein quantitation. A 16 related example is a TMT/TMTpro-labeled sample with a very broad range of concentrations over the (distinguishable) TMTc/TMTproC channels, so that quantitation results for the low abundance channels can be unacceptable.
Finally, high sample complexity resulting in interfering peaks in the TMTc/TMTproC cluster can be a potential limitation for the presented LSF approach when the interfering peaks are located close to the TMTc/TMTproC channels (presumably, nearby the doublets and at a distance that is less than the doublet split) or when the interfering peaks are of very large amplitude (presumably, many-fold) compared to the TMTc/TMTproC ions.

Conclusions
Resolving the 6.32 mDa doublets in the TMTc/TMTproC clusters across the entire mass range offers a valuable increase in the multiplexing capacity of the complementary ion approach, currently from commercially available 9 to 13 channels. In addition, by modifying the distribution of heavy isotopes, it would be possible to encode 21 channels without changing the TMTpro structure. 18 The described application of LSF in TMTc/TMTproC quantitative proteomics is a welcome addition to the arsenal of LSF-enabled strategies, such as targeted drug monitoring in mass spectrometry imaging. 43 The described approach is compatible with other Orbitrap instruments as well, e.g. Q Exactive HF and Exploris. The LSF algorithm applied here and other SR algorithms, such as the filter-diagonalization method, 44,45 phase-constrained spectral decomposition method, 46