The Future of Proteomics is Up in the Air: Can Ion Mobility Replace Liquid Chromatography for High Throughput Proteomics?

Abstract

The coevolution of liquid chromatography (LC) with mass spectrometry (MS) has shaped contemporary proteomics. LC hyphenated to MS now enables quantification of more than 10,000 proteins in a single injection, a number that likely represents most proteins in specific human cells or tissues. Separations by ion mobility spectrometry (IMS) have recently emerged to complement LC and further improve the depth of proteomics. Given the theoretical advantages in speed and robustness of IMS in comparison to LC, we envision that ongoing improvements to IMS paired with MS may eventually make LC obsolete, especially when combined with targeted or simplified analyses, such as rapid clinical proteomics analysis of defined biomarker panels. In this perspective, we describe the need for faster analysis that might drive this transition, the current state of direct infusion proteomics, and discuss some technical challenges that must be overcome to fully complete the transition to entirely gas phase proteomics.

Graphical Abstract:

1. INTRODUCTION

The overall function of a biological system, such as pumping blood or converting sugar to ATP, arises from the exact molecular composition of that system. In response to perturbations, such as drugs, toxins, or infection, the composition of molecules in a biological system change to meet the challenge and maintain overall function. If a biological system is pushed too far, however, disease states with diminished or altered function appear. When the mechanisms driving such disease states are unknown, clues can be obtained from measuring how the quantities of molecules in the system change. So called “omics” approaches attempt to measure complete layers of the biochemical dogma: the genome, the transcriptome, the proteome, or the metabolome.

Proteomics involves the comprehensive study of proteins, including their composition, structure, functions, and interactions within biological systems. Proteomics is an attractive choice among the omics approaches because proteins carry out most functions in cells, especially catalysis and structural roles. Proteomics data provides a molecular view of the functional state in a biological system. Modern proteomics has enabled quantification of over 10,000 proteins from human samples,¹ discovery of protein interactions,^2–4 discovery of post-translational modifications^5–7 including site specific stoichiometry,⁸ and even measurement of protein structure changes at the proteome scale.^9,10

The development of modern proteomics involved a significant amount of experimentation and refinement. Throughout the 1970s and 1980s there was a fervor of research effort attempting to unlock the potential of mass spectrometry (MS) for protein and peptide analysis. Attempts at fast atom bombardment,^11,12 secondary ion MS,¹³ or peptide derivatization coupled to gas chromatography MS^14,15 enabled some ionization of labile polypeptides. One of the key driving technologies for modern proteomics was electrospray ionization.^16,17 This not only allowed “elephants to fly” but was also compatible with LC, and therefore enabled separation of peptides with online detection by MS. This led to data-dependent acquisition (DDA) to collect discrete tandem mass spectra from peptides, and database search to identify peptides computationally.¹⁸ Another major milestone came from improved peptide separation before mass spectrometry. Multidimensional protein identification technology (MudPIT) uses the idea that strong cation exchange and reversed phase liquid chromatography can be carried out on one set of peptides sequentially using a single column.¹⁹ This enabled identification of over 1,000 proteins from one sample injection. Since then, the rate of proteins we can analyze per hour has been increasing exponentially due to improvements in LC and MS (Figure 1).^20–26

Figure 1.

Protein identifications per hour versus year showing selected milestone papers. Data acquisition method, MS type, and separation type are illustrated. Protein numbers include redundant identifications in subsequent samples, for example if a method can identify 5,000 proteins from one sample in half an hour, that is plotted as 10,000 proteins per hour.

2. A NEED FOR HIGH THROUGHPUT PROTEOMICS

“More is different.” This phrase popularized by Philip W. Anderson²⁷ expressed the idea that as you add more units to a system, at some point, new properties and behaviors emerge that are not predictable from the properties at smaller scales. This idea has become a fundamental concept in the study of complex systems and significant benefits of “more” have been demonstrated in multiple disciplines including physics, computer science, and biology. To illustrate this, in the realms of machine learning and artificial intelligence, expansion of training datasets along with advancements in cost-effective and efficient computing have markedly accelerated the transition of these technologies toward practical applications. It is not difficult to imagine how “more” in proteomics will benefit clinical diagnoses and biological studies. It remains an open question of whether more samples or more depth will provide more benefit in proteomics, but likely more of both are needed.

A current roadblock in proteomics is how to acquire “more”. We foresee that a stable, economical, and high-throughput methodology will be a key driver in advancing proteomics applications. We think that given the choice of more depth or more samples, more samples are better, and thus “more” should start with throughput. Increasing throughput will reduce costs and analysis times to open new research horizons that require tens of thousands of omics profiles, such as clinical trials, drug screening, structural proteomics, and single cell proteomics.

Many are working to make proteomics faster. This requires innovations in physics, chemistry, mathematics, and computer science. Some technologies are being widely implemented, while others are newly emerging. For instance: 1) Protein microarrays, also known as protein chips, represent a mature and powerful tool in proteomics for studying protein interactions, functions, and expressions. The primary advantage of this technology is its miniaturized, high-throughput design, which requires only a small amount of biological sample to generate extensive data. However, despite all the benefits, this technology faces several challenges including the complex selection and validation of antibodies, high costs, and the occurrence of false positives or negatives due to unstable specificity.²⁸ Another exciting option is nanopore technology, which was first commercialized for DNA and RNA sequencing, and is now emerging as a promising tool for peptide sequencing. Recent achievements include the detection of all 20 proteinogenic amino acids, real-time analyses of post-translational modifications (PTMs),²⁹ and peptide sequencing by sequentially cleaving amino acids from the C-terminus.³⁰ Although this technology currently struggles with large-scale sequencing under complex conditions, it has undoubtedly holds promise as a rising star. For example, millions of nanopores could be arranged on a chip to enable read depth comparable to MS. Finally, many are working to increase the throughput and standardization of LC-MS, which is valued for its high throughput, cost-efficiency, deep coverage, wide dynamic range, and precise quantification, making it an indispensable tool in modern proteomics, and remaining the centerpiece of this scientific banquet. It is noteworthy that the rise of artificial intelligence is empowering these technologies and accelerating their iteration and upgrade.

LC-MS typically requires minutes or even hours to collect data from one sample. There are approximately 10,000 min in a week (10,080 to be exact). That means we can think of proteomic method throughput in terms of every additional minute requiring an additional week to collect data from 10,000 samples. A method that takes a total of 60 min from sample to sample would take 60 weeks, or over one year, to complete data collection from 10,000 samples. Such a long time to finish a cohort brings more problems like batch effects due to needs for MS instrument maintenance, MS cleaning, LC column degradation, or LC system clogging. A 5 min analysis would require ~ 5 weeks to complete 10,000 samples. We think that even this throughput is too slow to achieve throughput required for many applications, such as single cell proteomics and clinical cohorts. We need to be thinking about how to collect data 100x faster, in 3 s total instead of 300 s. Three seconds per sample would enable 28,800 samples per day (assuming any overhead due to sample introduction time is parallelizable).

There have been numerous technical advancements and developments to improve the throughput of MS-based proteomics (reviewed by Xiao et al.³¹) related to sample preparation, LC separation, and data acquisition. We briefly cover some of these advances here.

2.1. Sample Preparation

Sample preparation does not determine the data acquisition speed of proteomics but can nonetheless be a limiting factor because it usually takes several hours to days to complete. Sample preparation includes protein extraction, purification, concentration, digestion, desalting and perhaps chemical labeling. Reproducibility, robustness, consistency, and speed are key goals but also significant challenges in proteomic sample preparation. These challenges are especially magnified when preparing samples for large-scale cohort proteome analysis. The main reasons include: 1) Sample complexity and variability: Differences between proteomic samples can be significant due to biological variation among individuals and sources (such as different tissues, cell types, or disease states). This variability affects how samples respond to specific processing strategies, making it difficult to design a universal method for sample processing. 2) Multiple processing steps: Proteomics involves several crucial steps, including protein extraction, digestion, and desalting, which require substantial amounts of reagents and consumables. Maintaining complete consistency in these reagents and consumables over extended periods is challenging. Additionally, the numerous processing steps increase the risk of operational errors and the potential for introducing errors. 3) Human interference: Variations in technical skill levels, experience, and operating habits among operators significantly impact outcomes. Results may vary even for the same operator under different spatiotemporal conditions. To minimize the impact of these systematic differences, there is a growing trend toward automating sample preparation using robotic liquid handlers for all or part of the sample processing workflow, for example using positive pressure devices or solid-phase-enhanced sample-preparation (SP3) for magnetic beads.^32–34

2.2. LC Separation

Faster LC separations, down to a few minutes per sample, have now become routine. For example, the Evosep system uses preformed gradients and sample elution from disposable tips to commoditize chromatography in a positive way and achieve up to 500 samples per day.³⁵ A second recent idea is microflow LC, which limits issues with nanoflow and increases throughput at the expense of higher sample loading requirements.³⁶ Other approaches try to limit the overhead between injections using dual traps^37,38 or dual columns.^39,40 With all these approaches, choosing between proteome depth and gradient length remains an inevitable compromise.

2.3. Data Acquisition and Hardware

Both new mass spectrometry hardware and new scan sequences have improved throughput. Newer Orbitrap MS platforms with optimized data collection methods continue to push throughput.^23,41–43 The introduction of the trapped ion mobility spectrometry (TIMS) for ion storage to improve sensitivity and ion utilization efficiency.⁴⁴ This coupled with additional hardware improvements led to the timsTOF SCP,⁴⁵ and data collection using variations of DIA improve sensitivity and data completeness, including DIA parallel accumulation-serial fragmentation (diaPASEF),⁴⁶ synchroPASEF,⁴⁷ slicePASEF,⁴⁸ and midiaPASEF.⁴⁹ Further innovative ideas to improve DIA used scanning quadrupole windows instead of static windows to improve precursor selectivity, called scanning SWATH.²⁵ More recently, the new Astral analyzer has shown to be a fast and sensitive TOF variant that enables high throughput.²⁶ A narrow window DIA approach that blurs the lines between DDA and DIA was recently shown beneficial on both the Orbitrap and Astral mass analyzers.^26,50 Finally, an approach based only on MS1 measurement called directMS1 was shown to enable fast sample analysis.⁵¹

3. DO WE STILL NEED LC?

Despite these advancements, due to the need for LC, current untargeted proteomics with fast gradients still requires at least 1 min of data collection²⁵ but more commonly at least 5–30 min. On top of that minute of analysis time, there is always a need for column washing and regeneration, meaning the true time cost of LC is usually higher. Although LC has enabled modern proteomics and dominates this field, like any analytical method, limitations exist, especially when it comes to high-throughput proteomic analysis. LC-related issues such as clogging, carryover, limited peak capacity, high costs, and gradual column degradation are the most likely causes of data quality problems or instrument failure in the proteomic workflow. These challenges become more pronounced in large-scale sample analyses. Conducting analyses on large cohorts typically involves thousands of hours of continuous operation and numerous sample injections, by which time the chromatography column significantly deteriorates from its original condition. This degradation severely impacts protein identification and the consistency of label-free quantification results. Even replacing the column with a new one from the same manufacturer may not completely resolve these issues. More critically, in large cohort analyses, the stable tracking and accurate quantification of specific protein groups become crucial, often more so than the depth of protein identification, as key protein biomarkers tend to be hidden within subtle variations in protein content. Further, despite normalization attempts, each LC system across laboratories (or even within the same laboratory) will be slightly different due to changes in solvent delivery, mobile phase composition, or column quality. Thus, there would be several benefits to excluding LC, but without LC we lack the separation required to perform the proper MS-based identification necessary to achieve deep coverage. An alternative mode of separation to address these shortcomings is welcome.

Compared to LC, which is based on chemical interactions of the analytes with the solid phase, the qualities of ion mobility separation are more of a physical process that can be more easily calibrated to have the same effect for every experiment. Using IMS instead of LC, we should be able to remove chemical variation, at least for the analyte separation step, through reproducible engineering processes. IMS across laboratories should be much more stable and comparable, this is extremely important for the consistent detection and quantification of proteins, especially in large-scale studies. Also, IMS is much faster than LC, with entire IMS separations requiring 10⁻² to 10⁻³ seconds instead of instead of the 10²–10³ seconds needed to complete entire LC separations.⁵² Current IMS technologies, such as field asymmetric ion mobility spectrometry (FAIMS), trapped ion mobility spectrometry (TIMS), and structures for lossless ion manipulation (SLIM), have gradually matured in their integration with other mass spectrometry technologies and are commercially available. These ideas led us to develop a new subfield of proteomics called direct infusion shotgun proteome analysis (DISPA) based on the use of IMS instead of LC for higher throughput.

4. THE CURRENT STATE OF DIRECT INFUSION PROTEOMICS

DISPA is the concept of replacing LC with faster separation by IMS (Figure 2). This is achieved by using flow injection or direct infusion and ESI of a complex peptide mixture. To simplify the highly complex ion mixture entering the MS, we use IMS as a first dimension of separation. IMS alone is not enough, so a somewhat orthogonal separation using small quadrupole selection windows is subsequently used to filter ions based on m/z. To spread out the precursors and produce more unique signals for peptide identification and quantification, we operate the MS in DIA mode, fragmenting any ions that are doubly selected by mobility and m/z.

Figure 2.

Overview of the DISPA concept including data analysis. (A) Concept and workflow diagram of DISPA. (B) Method settings and execution details of FAIMS and MS for implementing DISPA. (C) Schematic diagram of DISPA data analysis (identification and quantification).

The original DISPA paper described the concept and workflow using FAIMS plus quadrupole slices in DIA mode with MS/MS detection.²⁴ We will refer to this original paper as the DISPA-SILAC approach. In this seminal publication, we showed how various data collection parameters influenced observable depth; in general more selective precursor windows and longer ion accumulation times improved identifications. We also established a peptide identification workflow based on the projected spectrum approach used in mixture-spectrum partitioning using libraries of identified tandem mass spectra (MSPLIT-DIA).⁵³ Quantification was achieved using SILAC labeled proteome standards and relative fragment ion signal in MS/MS. Up to 500 human proteins were monitored in only a few minutes of direct infusion. We demonstrated how this approach can be applied to quickly enable complex multifactorial experimental designs, such as multiple genotypes of human cells grown in different nutrients and stressed with different mitochondrial toxins. We achieved over 45,000 quantifications in only 4.4 h of MS data collection time at a rate of over 170 complete protein profiles per minute. Notably, we also demonstrated the application of this approach to presimplified samples such as mitochondria. Despite the reduced depth, we found that this approach can reflect known biology while also revealing novel biology.

There are cases where the biological goal requires quantification of only a few proteins from many samples. In such cases, the DISPA approach is a perfect option for high throughput feedback. For example, in metabolic engineering, we may want to know if we have upregulated a pair of protein targets important for biofuel production. We combined the DISPA approach with parallel reaction monitoring (PRM) to target only a specific set of prototypic peptides, which we refer to as DISPA-PRM.⁵⁴ We assessed multiple combinations of MS data collection including Q1 isolation, injection time, and Orbitrap resolution to optimize detection of two proteins in E. coli. We found that spike in of stable isotope labeled peptides can produce consistent ratios using a dilution series and we were able to quantify the induction of protein expression over time from 72 samples very quickly and accurately relative to LC-MS.

The original data analysis method for DISPA was highly cumbersome, requiring multiple tools including scripts implemented in both R and python. To facilitate adoption of this approach, an easy data analysis solution was required. We wrote a software tool called csoDIAq that includes a graphical user interface to make DISPA data analysis approachable.⁵⁵ This software includes spectrum-spectrum match scoring, FDR calculation, protein inference, and protein quantification. We found that we could achieve the data analysis much more quickly using a spectrum pooling approach like the fragment ion indexing used by MS-Fragger. csoDIAq increased IDs while simultaneously increasing usability. We recently rewrote and released version 2 called zoDIAq, which incorporates software development best practices such as modularity and extensibility to encourage more community contributions (https://github.com/xomicsdatascience/zoDIAq/releases/tag/v2.1.1).

The first two iterations of DISPA relied on introduction of stable isotope labeled standard peptides for quantification, which is an expensive requirement that also reduces our data collection bandwidth and depth by wasting time on the heavy ion signal. We wanted to determine to what extent we could use DISPA for label free quantification (DISPA-LFQ).⁵⁶ We found that using DISPA with csoDIAq excluding stable isotopes, we could quadruple our identifications to over 2,000 proteins from cultured human cells. We also doubled the quantifiable proteins to ~1,000 from no more than 1 μg of input material. We achieve quantification using the sum of the fragment ion signals from a single scan where the peptide was identified. Remarkably, we found that we could identify 35 peptides from a single scan, and we plotted all spectrum-spectrum matches in the supplement for anyone to examine. We demonstrated that DISPA-LFQ could reliably produce protein quantities that reflect known toxin activities, such as a dose-dependent increase in glycolysis proteins induced by deferoxamine treatment. This approach expands the utility of DISPA for any sample, but a key prerequisite is that the nanoelectrospray stability must be high across samples to achieve LFQ.

Another group adopted the DISPA approach and the csoDIAq software to improve the performance. They applied what they dub repeat enhancing featured ion-guided stoichiometry (RE-FIGS).⁵⁷ In this study, they added two concepts: first, they append a linear discriminant analysis (LDA) classifier at the end of the csoDIAq processing; second, they applied repeated cycles of DISPA in one injection. They found that, overall, the peptide identifications were increased by up to 30%. They also introduced a new concept of how to obtain quantification from the fragment ions using the slope of all ions relative to the library spectrum, which they found could reliably produce quantification of multiple species mixture sample benchmarks. In general, these developments are exciting new ideas about how to leverage the DISPA concept.

One exciting extension of the DISPA concept is that because we remove the requirement for multiple types of chromatography that are often used for metabolomics, lipidomics, and proteomics, we can instead directly infuse multiomics samples simultaneously. We recently described proof of concept for this idea called Simultaneous Multi-Omics Analysis by Direct infusion (SMAD).⁵⁸ Using a FAIMS-Orbitrap MS platform we achieve identification of around 1,000 human proteins while also detecting thousands of metabolite and lipid features from single samples. We applied this concept to study macrophage polarization and high throughput screening of 293T cells in resposne to drugs and showed how SMAD provides a multiomic view of these biological changes in only 5 min of direct infusion per sample.

Finally, in our most recent study, we integrated DISPA with nanoparticle (NP) protein corona enrichment for high throughput and efficient plasma proteomic profiling.⁵⁹ We identified over 280 plasma protein groups in only 1.4 min of data collection time, and 405 unique protein groups with five different NPs combined in approximately seven total minutes of data collection. This result demonstrated the potential and advantages of combining DISPA with specific sample preparation methods that simplify the sample before injection.

5. IDEAS TO IMPROVE DISPA

Although we have made great progress toward fast direct infusion proteomics, several challenges remain. For example, in DISPA we must deal with extensive ion competition for both ionization and detection. We also consistently see compressed signal response in dilution series leading to imperfect quantification curves. Further, we need to fully leverage the potential speed of IMS-MS to go much faster. There are several potential ways that DISPA can improve to achieve better quantification and ultimately enable a 100x increase in throughput to 3 s per sample. These ideas relate to five different areas: ion mobility, sample handling, electrospray ionization efficiency, mass analyzer sensitivity and speed, and software.

5.1. Ion Mobility

The original attempts at DISPA all used FAIMS as the ion mobility device. This is preferable because it acts as a filter that allows the subsequent ion traps to accumulate sub populations of ions for analysis. However, there are emerging types of ion mobility that warrant exploration as potential routes to improve the sensitivity and throughput of DISPA. These alternatives all act as continuous elution devices more like traditional chromatography and therefore may improve DISPA, for example by using signal integrated from the extracted ion mobiligram (EIM) for quantification. Another attractive element of these continuous mobility devices is they are often coupled to fast scanning high resolution TOFs, which pair well with the goal to increase throughput 100x while providing more measures across a single ion mobility pass. Key examples include structures for lossless ion mobility (SLIM) and trapped ion mobility spectrometry.

5.1.1. Structures for Lossless Ion Manipulation (SLIM) Introduction.

SLIM was developed by researchers at Pacific Northwest National Laboratory to overcome sensitivity and resolution limitations of existing IMS techniques.^60–62 Manufactured using robust and inexpensive conventional printed circuit board (PCB) technology, SLIM uses electric fields generated via arrays of mirror-image electrodes patterned on closely spaced surfaces to define a path in the space between the surfaces where ions can be moved or trapped.^63–66 The novel electrode design also allows for abrupt turns in the ion path, such that ions can be made to follow serpentine paths, resulting in very long path lengths up to 13 m in a single pass device and multiple kilometers in multipass devices.^62,64,67 In this way, SLIM provides access to higher resolution IM separations than any other technology.⁶⁸ SLIM-based IMS enables complex mixtures of analytes to be quickly separated for characterization and quantification in a lossless manner up to 3 orders of magnitude faster than traditional LC-MS methods. To date, SLIM has been integrated with various mass analyzer types including Time-of-Flight,⁶⁹ Triple Quadrupole,⁷⁰ and Orbitrap mass spectrometry platforms. In 2021, the first commercial SLIM-based high resolution ion mobility (HRIM) device was released by MOBILion Systems, the company which has licensed the SLIM technology from PNNL. This product, the MOBIE platform, is compatible with Agilent Q-TOF systems. In the context of proteomics, SLIM-based HRIM has been used to accelerate routine PTM monitoring workflows⁷¹ and reveal protein dynamics to enable therapeutic drug development. Given the recent advent of this technology, this list of applications and capabilities is sure to grow.

To test how SLIM might be useful for DISPA, we applied this platform with direct infusion of peptides that are overlapping in mass with highly similar chemical structure due to deamidation of D/E or citrullination of R. Since both deamidated and citrullinated proteins only differ by 0.984 Da from the native forms, the mass-to-charge (m/z) ratios of the isotopes from the unmodified species overlap and the modified states. Low levels of these modifications hinder typical MS/MS identification. Without LC separation of the modified and unmodified peptides, identification, and quantification of these modifications would be highly challenging. Previously, these different species would be resolved with nano-LC prior to HRMS.^72,73 Direct infusion without further separation would result in both the modified and unmodified forms being transmitted within the quadrupole isolation window, thus complicating MS/MS interpretation. This peptide mixture therefore represents the most extreme test for IMS in direct infusion proteomics.

We infused an equimolar mixture of four standard peptides that are all variations of the same amino acid sequence: no modification, one citrullination, one deamidation, and both citrullination and deamidation. As expected, their overlapping precursor mass spectra cannot be distinguished (Figure 3A). The composite MS1 spectrum exhibits an aberrant isotopic distribution where the second isotopic peak is the most abundant. Tandem mass spectrometry analysis is required to differentiate the presence of citrullination or deamidation, but differentiating fragment ions may not always be observed, especially with direct infusion. HRIM allows for baseline separation of these peptides before MS detection in the absence of chromatographic separation (Figure 3B). Peak-to-peak resolution (Rpp) > 1 was achieved for each peptide pair. The reproducibility of HRIM separation was demonstrated via the determination of CCS for each target peptide (relative standard deviation <0.1% across all peptides for HRIM for six replicates). Like extracted ion chromatograms, the EIMs can be used to obtain accurate and reproducible relative quantification. Thus, if SLIM can resolve these highly similar peptides, it is likely to be more applicable to broad distributions of peptides from proteome digestion, and SLIM is therefore a promising technology that may improve DISPA in the future.

Figure 3.

Flow injection analysis of an equimolar mixture of unmodified, deamidated, citrullinated, and doubly modified SAVNARSSVPGVR peptides. (A) MS1 spectra and (B) extracted ion mobiligram (EIM). Peptides ID are indicated above the EIM.

5.1.2. Trapped Ion Mobility.

Compared to traditional IMS where ions traverse some region encouraged by an electric field, trapped ion mobility^44,74 uses a much smaller cell where ions are propelled forward by a gas flow but detained by an opposing electric field. Gradually decreasing the trapping electric field allows selective elution of the trapped ions. This concept was developed by Bruker and commercially incorporated in the timsTOF line of Q-TOF mass spectrometers. This approach can improve sensitivity by enabling accumulation of ions from the source before they are released together for MS/MS analysis.⁷⁵ Like SLIM, the continuous nature of the elution from a TIMS device may improve the detection options for directly infused peptides. We explored this idea by directly infusing peptides from whole HeLa proteome standard into the timsTOF Pro 2. Data was collected using 72 m/z and ion mobility (IM) windows, covering a mass range from 420 to 1200 m/z and an IM range from 0.6 to 1.5 1/K0 in the multicharged ion area (Figure 4A). With manual data inspection using alphaTims⁷⁶ we found clear patterns of multiple peptides eluting over TIMS space with constant quadrupole (Q1) selection ranges (Figure 4B). Full exploration of this approach will require a future software development to systematically interrogate the data while simultaneously testing optimization grids of instrument parameters.

Figure 4.

Example of multiple peptide signals separated by direct infusion TIMS-TOF using 80 ng/μL trypic peptides from whole HeLa proteome on the timsTOF Pro 2. (A) AlphaTims visualization of detected m/z and ion mobility data points on the MS1 level from a 2 min direct infusion analysis of Hela standard proteome. DIA method and corresponding ion mobility regions are indicated in the figure. Sample concentration: 80 ng/μL. (B) Heatmap visualization of m/z and ion mobility on the MS2 level of a single DIA window.

5.1.3. Other Types of Ion Mobility.

One potential challenge with continuous ion mobility separation devices is that they are often used before any other ion filtering, which means they must accommodate all the ions from the source and low abundance peptide species of interest may get literally crowded out by the higher abundance peptides, thus limiting dynamic range. Another promising example of ion mobility that may be beneficial includes the cyclic ion mobility system from Waters. This system is particularly interesting because it allows quadrupole selection before the mobility separation, which may be beneficial to prevent overloading the ion mobility trap. We expect that additional exploration of all alternative IMS options is required to determine the optimal hardware for 100x throughput DISPA.

5.2. Sample Introduction

Assuming that we can improve the above factors to collect deeper data more quickly, we need to think about how to introduce samples to the MS more quickly. There are several commercial options for this, but additional robotics may need to be developed. For example, the Agilent Rapid Fire enables high throughput sample infusion in as little as 2 s per sample.⁷⁷ Another option for fast sample introduction is based on acoustic liquid transfer, such as acoustic mist ionization⁷⁸ or acoustic ejection mass spectrometry⁷⁹ or the Echo commercialized by Sciex.

5.3. Ion Source Brightness or Spray Efficiency

The lack of presimplification of peptide mixtures by LC results in a highly complex mixture being ionized together. This means that ion suppression is extreme between peptides where higher abundance and easier to ionize peptides hide the less abundance or more difficult to ionize peptides. For these reasons, increasing the depth of DISPA requires modifications of the electrospray ionization efficiency. Possible routes to improve the brightness of the source are well-known, such as mobile phase modifiers,⁸⁰ lower flow rate,⁸¹ and multinozzle emitters.⁸²

5.4. Mass Analyzer Sensitivity and Speed

The scan speed and sensitivity of a mass analyzer influence the cycle time and the depth of protein identification for DISPA, respectively. Currently, leading mass analyzers such as the Orbitrap, Astral, Fourier Transform Ion Cyclotron Resonance (FT-ICR), and Time-of-Flight (TOF) must make a trade-off between scan time and mass resolution. The cycle times for the Orbitrap and FT-ICR are at the 10s to 100s of milliseconds level, while that of a TOF is at the 100s of microseconds level. Using a mass analyzer that achieves high resolution (>100k) in 1 ms or less, it would be possible to complete the scan of all mobility windows within our stated goal of three seconds. Similarly, increasing sensitivity of the mass analyzer will improve detection of extremely low-abundance ions after very narrow quadrupole window selection or multiple gas-phase separations by different ion mobilities. This could compensate for the shortcomings in the coverage and depth of DISPA. However, progress in this area is currently very challenging and may require breakthroughs in mechanical engineering.

5.5. Software

The software for data analysis we use is relatively simple.⁵⁵ Given the large number of peptides entering the MS at any time, there are likely to be fragments from many peptides in each scan. We use spectral libraries of previously identified peptides from DDA and compute spectra–spectra match scores. We leverage only the potentially matchable fragments using the projected spectrum approach, which looks in the query spectra for only the fragments that can possibly match the library spectra within some ppm mass tolerance. We then compute a shape of the match using the cosine score and then multiply that by the fifth root of the number of matched fragments to generate a composite match count and cosine (MaCC) score. Then simple false discovery rate analysis is used to determine the statistically significant spectra–spectra matches, which are rolled up to peptides and proteins. The RE-FIGS paper added a layer of LDA to this process to improve IDs.⁵⁷ We posit that more advanced models for spectra–spectra matching, rescoring matches using machine learning, and more advanced approaches to obtain quantification will likely improve sensitivity of DISPA. Especially when we think about using multiple scans over the EIM elution profile with alternative continuous IMS approaches, there will be much need for more advanced algorithms that extract the most information from DISPA.

6. SUMMARY AND PROSPECTS

As IMS and MS hardware continue to improve, we expect to see more adoption of the DISPA concept in the proteomics community. Based on the current technology with limited depth, ideal applications would include mixtures that are already simplified, such as subcellular organelle preparations or immunoprecipitations. In the future, as the speed and sensitivity increase to approach the 100x improvement to require only a few seconds per sample using the approaches mentioned above, we expect that LC-free proteomics will be applied to large scale drug perturbation studies, clinical cohorts, structural proteomics, and even single cell proteomics. We cannot make these advances alone and welcome everyone to contribute to this nascent field.

7. METHODS

7.1. FIA-HRIM-MS Data Acquisition and Processing

Synthetic peptides with arginine modified to citrulline or asparagine deamidation modifications at preselected sites were procured from Vivitide (formerly New England Peptide). Individual peptides and mixtures (at 2 pmol/μL) were analyzed using a 2 min automated flow injection acquisition with a constant flow solvent of 0.1% formic acid in 1:1 methanol/acetonitrile at a carrier flow rate of 100 μL/min using an Agilent 1290 Infinity II binary pump and chilled multisampler held at 4 °C. HRIM data acquisition was carried out on a MOBIE instrument (MOBILion Systems) coupled to a 6545XT QTOF (Agilent Technologies). In all cases, the ESI source was operated in positive-ion mode using the following conditions: nebulizer pressure, 35 psi; sheath gas flow rate, 12 L/min; sheath gas temperature, 275 °C; drying gas flow rate, 13 L/min; drying gas temperature, 325 °C; capillary voltage, 4000 V; entrance nozzle voltage, 500 V. The SLIM boards were operated at ca. 2.5 Torr. Traveling wave-based separation was performed using a wave speed of 40 kHz and a peak-to-peak wave amplitude of 90 Vp-p. SLIM parameters were as follows: fill time, 12 ms; trap time, 0.3 ms; release time, 3.2 ms; IMS frame length, 500 ms; on-board accumulation fill frequency, 15000 Hz; on-board accumulation fill amplitude, 20 Vp-p; on-board accumulation release frequency, 15000 Hz; on-board accumulation release amplitude, 40 Vp-p; on-board accumulation t-wave frequency, 15000 Hz; on-board accumulation t-wave amplitude, 30 Vp-p; Waste T-Wave Speed, 15000 Hz, Waste T-Wave Amplitude, 40 Vp-p; Offset, 0 V; Separation T-Wave Speed, 40000 Hz; Separation T-Wave Amplitude, 90 Vp-p; Offset, 0 V; Waveform Shape, SIN; Funnel In, 175 V; Funnel Exit, 100 V; Funnel Conductance Limit, 95 V; On-Board Accumulation (Short Path) Gate, 80 V; SLIM Bias, 90 V; SLIM Guard, 10 V; Deflector, 5 V; Exit Conductance Limit, 50 V; Quad Bias, 40 V; SLIM RF Amplitude, 250 Vp-p. These separation parameters were chosen based on optimal conditions for high resolving power in the m/z range of the synthetic peptides at [M+2H]2+ charge state. Data was acquired via MassHunter Acquisition (Agilent Technologies) and EyeOn software (MOBILion Systems). For CCS calibration, data for the tune mix was acquired in a separate experiment using identical instrument parameters (i.e., external CCS calibration). The QTOF stage was operated in the 3200-mass range (m/z 300–3200), ion slicer operated at high sensitivity, and the digitizer operated at 2 GHz extended dynamic range. Accurate mass, isotope spacing, ion mobility arrival time distribution, and Collision Cross Section (CCS) determination were used to identify modified/unmodified peptide mixtures, with data acquired in at least triplicates. Data processing, analysis, relative quantification, and visualization were achieved using HRIM Data Processor, PNNL Preprocessor, and adapted onto Skyline.

7.2. timsTOF Pro 2 Data Acquisition

The Hela proteome standard was diluted to a concentration of 80 ng/μL and directly injected into the MS (timsTOF Pro2, Bruker Daltonics, Bremen, Germany) using a syringe. timsTOF Pro2 equipped with a TIMS-Pro 2 cartridge. Data acquisition was performed in positive mode and dia-PASEF acquisition was applied, featuring an ion accumulation time of 200 ms and an MS1 scan range that extended from m/z 200 to 1700. For sequential MS2 fragmentation, 72 isocratic m/z and ion mobility (IM) windows were chosen, covering a mass range from 420 to 1200 m/z and an IM range from 0.6 to 1.5 1/K0, respectively. The data collection period lasted for 2 min.