Absorption Mode Fourier Transform Ion Mobility Mass Spectrometry Multiplexing Combined with Half-Window Apodization Windows Improves Resolution and Shortens Acquisition Times

Abstract

Fourier transform multiplexing enables the coupling of drift tube ion mobility to a wide array of mass spectrometers with improved ion utilization and duty cycles compared to dual-gate signal averaging methods. Traditionally, the data generated by this method is presented in the magnitude mode, but significant improvements in resolution and the signal-to-noise ratio (SNR) are expected if the data can be phase corrected and presented in the absorption mode. A method to simply and reliably determine and correct phase shifts in Fourier transform ion mobility mass spectrometry data using information readily available to any user is presented and evaluated for both small molecule and intact protein analyses with no modification to instrument hardware or experimental procedures. Additionally, the effects of apodization and zero padding are evaluated for both processing methods, and a strategy to use these techniques to reduce acquisition times is presented and evaluated. Resolution is improved by an average factor of 1.6, the SNR is improved by an average factor of 1.2, and acquisition times are reduced by up to 80% through the application of absorption mode processing combined with apodization and zero padding.

Graphical Abstract

INTRODUCTION

Drift tube ion mobility (DT-IM) coupled to mass spectrometry (MS) enables analysis of the size and shape (collision cross section (CCS)) of gas-phase ions, thus enabling applications ranging from separation of isomers to characterization of tertiary structures of proteins.^1–5 To date, the large majority of reported DT-IM-MS systems have been based on time-of-flight (TOF) mass analyzers owing to their high scan rates enabling real-time sampling of ions exiting the drift tube.¹ In these instruments, an electrostatic ion gate at the entrance injects short packets of ions into the drift tube and they are sampled in real time as they exit. While some IM technologies, such as trapped ion mobility spectrometry (TIMS)⁶ and field asymmetric ion mobility spectrometry (FAIMS),⁷ can operate independently of the mass spectrometer scan rate, they lack the simplicity and ability to calculate collision cross sections (CCS) based on first principles (i.e., without the need for calibration with an ion of known CCS) afforded by DT-IM. Coupling DT-IM to slower mass analyzers such as ion traps or Fourier transform (FT)-based analyzers, though appealing due to their enhanced mass resolution and tandem MS capabilities, requires the ability to sample ions at shorter timescales than allowed by the scan rate of the analyzer. This mismatch has been bridged by the addition of a second ion gate to control ions exiting the drift tube.^8–10 A time delay between the first and second gate allows only ions with drift times that match the delay to be detected, and by iteratively sweeping this delay over the drift time range of interest, a full arrival time distribution (ATD) can be generated.¹⁰ The otherwise low duty cycle of this method can be significantly improved through the utilization of Hadamard or FT multiplexing methods, which sample multiple drift times simultaneously and require post-acquisition processing.^11–13

FT multiplexed acquisition is performed by opening and closing both gates simultaneously at a gradually increasing frequency (f_g) using a symmetric square waveform typically swept from 5 Hz to 5–10 kHz over the course of several minutes.^8,11,14 This frequency-gradient gating method encodes ion drift times as oscillations in the recorded ion current. As the value of f_g is swept across the frequency range, ions with drift times (t_d) satisfying eq 1a are transmitted at maximum intensity (I_max), while ions satisfying eq 1b are transmitted at minimum intensity (I_min), creating a periodic oscillation between I_max and I_min in each ion signal over the course of the sweep.¹⁴

$$I_{\text{max}}\text{ when: }{f}_{\text{g}}=\frac{1}{t_{\text{d}}},\frac{2}{t_{\text{d}}},\frac{3}{t_{\text{d}}}$$ (1a)

$$I_{\text{min}}\text{ when: }{f}_{\text{g}}=\frac{1}{2t_{\text{d}}},\frac{3}{2t_{\text{d}}},\frac{5}{2t_{\text{d}}}$$ (1b)

This periodic signal is then Fourier transformed to generate a frequency spectrum, which in turn is converted to an arrival time distribution (ATD) by dividing the frequency vector by the sweep rate used in the acquisition. While this last step may appear counterintuitive, it reflects the fact that the frequency of the time domain signal is a function of the drift time as illustrated in eqs 1a and 1b. By creating extracted ion chromatograms (XICs) for each m/z of interest, any ion signal can be assigned an individual ATD and thus be resolved in both m/z and mobility domains.

Traditionally, FT-IM data is presented in the magnitude mode (mFT), which utilizes the absolute value of the sum of real and imaginary parts to determine spectral intensity, as follows.

$$\begin{gathered} F(\omega )=\text{FT}(f(t))=\int f(t){\text{e}}^{-i\omega t}\text{d}t=\text{Re}[F(\omega )] \\ +i\text{Im}[F(\omega )] \end{gathered}$$ (2a)

$$\text{mFT}=\text{abs}(F(\omega ))=\sqrt{Re{[F(\omega )]}^2+\text{Im}{[F(\omega )]}^2}$$ (2b)

where f(t) is the time domain XIC, ω = 2πf is the angular frequency of each component of F(ω), and Re[F(ω)] and Im[F(ω)] are the real and imaginary parts of F(ω), respectively. A primary advantage of mFT is that all frequency signals are represented as positive values regardless of their initial phases at the start of data acquisition. However, two significant limitations to mFT processing of finite, damped, and periodic time domain signals highlight the need for more sophisticated approaches:

1. Standard mFT processing of rapidly decaying time domain signals produces peaks that exhibit long tails, which interfere with the detection of neighboring peaks and contribute to overall broadening of the peak shape, reducing the observed resolution of the spectrum.¹⁵

2. The use of symmetric apodization functions such as the Hanning window has been shown to reduce or eliminate these tails but has the unfortunate effect of suppressing signal intensity at the beginning of the transient leading to significant reductions in signal-to-noise ratios.¹⁶

An alternative approach known as absorption mode FT (aFT), which utilizes only the real portion of the FT, is known to improve spectral resolution and accuracy in nuclear magnetic resonance (NMR) spectroscopy and FT ion cyclotron resonance (FT-ICR), Orbitrap, and FT electrostatic linear ion trap (ELIT) mass spectrometers.^16–19 A related method termed the “enhanced Fourier transform” (eFT) that utilizes both the real and imaginary components of the FT combined with the magnitude spectrum also produces significant gains in resolving power and is routinely employed for processing Orbitrap data.²⁰

$$\text{aFT}=\mathit{\text{Re}}[F(\omega )]$$ (3)

The use of aFT also reduces tailing and improves peak shapes when used in conjunction with asymmetrical “half-window” apodization functions compared to mFT processing. Compared to symmetrical apodization functions, half-window functions have the advantage of preserving the beginning of the time domain signal, where the highest intensity portion of rapidly decaying signals is typically found.^17,21,22 However, a requirement for aFT processing is that the initial phase of each frequency component must be zero. In most real-world applications, frequency signals have non-zero phases, which causes absorption mode frequency peaks to appear asymmetrical and contain a mixture of both positive and negative values. In order to realize the advantages of aFT, the sources of the phase shifts must be identified and corrected by first determining a phase function for the spectrum (φ(ω)) and then using it to “rotate” F(ω) to create a new phase-corrected spectrum termed G(ω).

$$\phi (\omega )={\text{tan}}^{-1}\left(\frac{\text{Im}[F(\omega )]}{Re[F(\omega )]}\right)$$ (4a)

$$G(\omega )=F(\omega ){\text{e}}^{-\text{i}\phi (\omega )}$$ (4b)

In FT-ICR mass spectrometry, phase shifts are primarily attributed to nonsimultaneous ion excitation during the linear frequency sweep of the excitation waveform used to form coherent ion packets (ions with different m/z values are excited at different times, causing them to accumulate various amounts of phase shifts during the excitation event proportional to their m/z) and the delay between excitation and the start of detection.¹⁷ This manifestation of phase shifts presents a complex problem that has spurred the development of several methods to determine φ(ω) by modeling the excitation waveform,²³ sequentially calculating φ in small but densely populated regions of the spectrum,²⁴ or a combination of the two.¹⁷ Importantly, calculating φ(ω) directly from a spectrum is complicated by extensive phase wrapping^17,25 (i.e., φ_max > 2πN radians, where N is the number of phase wraps), so spectra containing a high density of peaks throughout the m/z range are typically needed, thus requiring external calibration of φ(ω) for routine application of aFT processing of sparse spectra.²²

Orbitrap and ELIT mass spectrometers use a distinctly different ion excitation mechanism than FT-ICR mass analyzers, which leads to a simpler method for determining φ(ω).^19,20 In these electrostatic traps, all ions are excited simultaneously regardless of m/z by a rapid, high amplitude DC pulse that accelerates ions into the analyzer. The phases of all ions, regardless of m/z, are aligned at the instant that the pulse is applied, so φ(ω) can be theoretically derived if the time delay between excitation and the start of detection (Δt) is known to sufficient precision using eq 5.

$$\phi (\omega )=-\omega \Delta t$$ (5)

Though it is tempting to think that zero-phase spectra can be generated directly by simply synchronizing excitation and detection on the instrument, this approach has proved to be difficult due to limitations of the electronics involved and additional phase shifts arising from field imperfections and space charging.¹⁹ Instead, various approaches to determine Δt using external calibration²⁰ (built into the routine user calibrations of modern Orbitraps) or by synchronizing excitation and detection as much as possible and determining the remaining small Δt as the slope of a plot of φ (determined using eq 4a) vs ω.¹⁹

While the dynamics responsible for oscillations in time domain FT-IM data are clearly different than those in FT-MS, it is possible to determine φ(ω) in a manner analogous to the method used for Orbitraps and ELITs. To illustrate this concept, one must recall that the key to determining φ(ω) in these types of analyzers is first to identify a time point when the phases of all ion signals are aligned (and ideally zero) and then quantify the time difference between that point and the start of data acquisition (Δt). Though it is not necessarily obvious from eqs 1a and 1b, when both gates are held open indefinitely, which can be thought of as f_g = 0 Hz, all ions will be transmitted at maximum intensity regardless of t_d. With this requirement satisfied, all that remains is to determine Δt as the time that would be required for f_g to increase from 0 Hz to the actual starting frequency of the sweep (typically 5 Hz), which is readily calculated using the sweep rate, and address any additional delays resulting from imperfect synchronization between the IM and MS electronics. With this information, φ(ω) can easily be calculated for the full spectrum using eq 5.

In addition to the well-documented improvements in resolution, aFT promises an additional benefit to FT-IM data processing through the use of half-window apodization functions.²¹ While a typical FT-IM experiment might involve sweeping the gating frequency from 5 Hz to 5 kHz over the course of 5 min, it is often the case that detectable oscillations only persist for the first 1–2 min of sweep, with the remainder of the signal dominated by random noise produced by spray instability and electronic noise in the MS system. While this suggests that conducting a shorter sweep, say 1 Hz to 1 kHz in 1 min, would both reduce the time needed for the experiment and reduce noise without sacrificing any frequency information; longer sweeps are still routinely used to produce sufficient data points to adequately represent peak shapes in the frequency domain. This issue can be remedied by simply extending the length of the time domain signal by appending it with an array of zeros, a technique known as zero padding.^16,26,27 However, doing so requires the use of an appropriate apodization function to avoid spectral artifacts that would arise from the abrupt drop-off in signal intensity at the union of the acquired signal and appended zeros. While the symmetrical apodization functions used in mFT processing are often incompatible with FT-IM data due to excessive suppression of the beginning of the time domain signal, the half-window functions used with aFT inherently preserve this part of the signal and are thus an excellent fit for these strategies.

Herein, we present a simple and robust method to determine φ(ω) based only on knowledge of user-defined parameters and demonstrate that aFT-IM combined with half-window apodization and zero filling improves resolution and the SNR while allowing reduction of acquisition times by up to 80%. This method is demonstrated for acquisition of arrival time distributions using an atmospheric pressure drift tube coupled to a linear ion trap and Orbitrap mass spectrometers.

METHODS

Materials.

Tetraalkylammonium bromide (TXA) salts where X is the number of carbons in each alkyl chain (5, 6, 7, 8, 10, or 12) and bovine ubiquitin were purchased from Sigma-Aldrich (St. Louis, MO) and used without further purification. TXA salts were dissolved in a 40:40:20 water:-methanol:isopropyl alcohol mixture at a concentration of 10 μM. Ubiquitin was dissolved in a 1:1 water:methanol mixture with 0.1% formic acid at a concentration of 10 μM.

Instrumentation.

The atmospheric drift tube used in this study was designed and constructed as described previously.²⁸ Briefly, the instrument is fabricated from stacked printed circuit board (PCB) electrodes separated by spacers and connected through a chain of 1 MΩ resistors. A three-grid ion shutter divides the drift tube into an ~10 cm desolvation region and an ~10 cm drift region, and a second shutter located at the end of the drift region controls the flow of ions into the mass spectrometer. A potential of 10 kV is applied across the resistor chain to create a uniform electric field strength of 500 V/cm. The gates are operated by supplying a supplemental potential of 65 V (front gate) or 80 V (back gate) above the standard potential at their respective locations using a pair of FET pulsers²⁹ built by GAA Custom Engineering (Kennewick, WA). The pulsers are controlled by a square waveform generated in a custom program written in LabView and supplied by a National Instruments PXI 6289 digital to analog converter (National Instruments, Austin, TX). The LabView GUI allows the user to specify the starting and final frequencies of the gating frequency sweep and the duration of the sweep. A schematic of the drift tube and mass spectrometer is provided in Figure 1.

Figure 1.

Instrument schematic. Ions are generated at the ESI emitter and travel into the drift tube. Gates 1 and 2 are controlled by the waveform generator, modulating the ion current through the ~10 cm drift region. Ions then enter the mass spectrometer through the standard ion transfer tube and are mass analyzed in the Orbitrap.

The drift tube is mounted in place of the standard ion source on an Orbitrap Elite mass spectrometer (Thermo Fisher Scientific, San Jose, CA) using a 3D-printed mounting system. Mass spectra of the tetraalkylammonium cations were acquired using ion trap detection with a fixed ion injection time (IIT) of 20 ms. Ubiquitin spectra were acquired using Orbitrap detection with a resolution setting of 30,000 at m/z 400 and a fixed IIT of 100 ms. Ions were produced by electrospray ionization using Au/Pd-coated borosilicate emitters pulled and coated in house. The potential applied to the emitters was 0.8–1.2 kV above the potential at the entrance of the desolvation region (10.8–11.2 kV relative to ground).

Data Processing.

Data were processed with a custom program written in Matlab. Thermo .raw files were converted to .mzXML files using MSConvertGUI³⁰ and imported using the mzxmlread.m function. Extracted ion chromatograms (XICs) were generated automatically based on a list of user defined m/z values and optionally apodized and zero padded prior to Fourier transformation. mFT spectra were generated by taking the absolute value of the complex value FT. aFT spectra were generated by first phase-shifting the complex value FT as shown in eq 4b and then plotting only the real part of the spectrum. Baseline correction of aFT ATDs is performed using the method described by Xian et al.³¹ Noise was calculated as the root-mean-square (RMS) of ~100 data points near the beginning of each ATD using the rms.m function, and peak widths were calculated using the findpeaks.m function. Unless otherwise noted, all ATDs were smoothed using a cubic spline interpolation to increase the density of data points by a factor of four. A version of the processing script written in Python along with a sample data file are included in the Supporting Information.

RESULTS AND DISCUSSION

Two samples, a mixture of tetraalkylammonium bromide salts (MW, 295–690 Da) and intact ubiquitin (MW, ~8500 Da), were selected to evaluate the feasibility and utility of aFT processing over a range of masses, CCS values, and charge states. aFT is expected to improve peak resolution by a factor of the square root of 3 to 2 and the SNR by a factor of the square root of 2.^26,32,33 ATDs generated using either mFT or aFT processing were evaluated on the basis of resolution, SNR, and ability to identify known features in peak shapes related to protein conformational states.

Initially, aFT processing was applied to mass spectra generated from the mixture of TXA cations without phase correction. The resulting ATDs (Figure S1) showed a characteristic peak shape consisting of a substantial shift in the center of the peak accompanied by large negative value side lobes indicative of improperly phased absorption spectra. Close examination of these ATDs reveals that the extent of the phase shift increases with the increasing drift time, which is consistent with the theoretical model used to explain systematic phase shifts in aFT-IMS experiments. This model, illustrated in Figure 2, predicts that the observed phase shift is fully explained by the nonzero starting frequency of the sweep. Because longer drift times are encoded as higher frequencies in the XIC, a constant value of Δt produces a larger proportional phase shift compared to shorter drift times that are encoded as lower frequencies.

Figure 2.

Illustration of the method for estimating Δt using the starting frequency (fstart, typically 5 Hz) and the rate of the gating frequency sweep. Δt is defined as the time that would have been needed for the gate frequency to reach the starting frequency had the sweep started precisely at 0 Hz and increased at the same rate used in the remainder of the sweep. The t_d values of each theoretical drift time are shown next to each triangle wave.

For a standard five-minute 5 Hz to 5 kHz sweep, a Δt value of 0.3 s is obtained by dividing the starting frequency (5 Hz) by the sweep rate (16.65 Hz/s). Applying this value to eq 4b eliminates most, but not all, of the observed phase shift as illustrated in Figure S2. The remaining phase shift is assumed to be an artifact of imperfect synchronization between the start of the frequency sweep and the start of the mass spectrometer data acquisition. Adding an additional shift of 0.1 s to Δt to compensate for the acquisition delay was found to consistently produce well-phased aFT spectra for all experiments conducted using the same sweep rate and thus was used to generate all the ATDs shown in the remainder of this work.

Application of the above phase shift approach with baseline correction that produced the ATDs of the tetraalkylammonium cations is shown in Figure 3. As expected, the widths of each peak (FWHM) are reduced considerably while preserving the height and center of each peak. Close examination of the T8A peak shown in the inset reveals that the FWHM reduced from 0.67 s in the mFT ATD to 0.37 s in the aFT ATD, and the long tails observed around the mFT peak are eliminated. Table 1 shows a comparison of FWHM, resolution (t_d/FWHM), and the signal-to-noise ratio (SNR) for the six peaks shown in Figure 3. Notably, aFT resolution was found to increase by 1.4-fold to 1.8-fold over mFT, with the average improvement of 1.6-fold in good agreement with reported improvements from aFT-MS and theory.^26,32,33 SNR was also improved by 1.1 to 1.3-fold, again in good agreement with observations from FT-ICR studies.¹⁸ While this improvement is modest, it highlights an additional potential benefit of aFT processing that could be particularly useful for atmospheric pressure IM-MS systems that suffer from reduced ion transmission. This effect is believed to result from the fact that the phase of noise peaks is independent from the phase of analyte peaks and is expected to be random, so most peaks would be expected to display lower amplitudes similar to the unphased aFT peaks shown in Figure 1. The source of this effect and potential to capitalize on it for further SNR improvements is the subject of ongoing investigation.

Figure 3.

Magnitude and absorption mode processing of tetraalkyl ammonium cations (1+, TXA where X indicates the length of the alkyl chain) following phase correction. Peak widths are noticeably decreased with minimal side lobes or other artifacts. Although each species is plotted individually, the spectra were acquired from a mixture containing all six species in a single experiment, and the same phase correction function was applied to all ATDs. The individual plots reflect the fact that these data are inherently resolved in both mobility and m/z domains.

Table 1.

Comparison of Peak Full Width Half-Maximum (FWHM), Resolution (t_d/FWHM), and Signal-to-Noise Ratio (SNR) for ATDs of Tetraalkylammonium Cations (1+). aFT Improves Resolution by an Average Factor of 1.6 and the SNR by an Average Factor of 1.2

		FWHM (milliseconds)		resolution			signal-to-noise ratio
TXA	drift time (milliseconds)	mFT	aFT	mFT	aFT	ratio (aFT/mFT)	mFT	aFT	ratio (aFT/mFT)
T5A	20.7	0.52	0.37	40	56	1.4	41.8	54.3	1.30
T6A	23.3	0.55	0.35	42	67	1.6	46.8	52.0	1.11
T7A	25.8	0.62	0.38	42	68	1.6	29.3	32.2	1.10
T8A	28.1	0.67	0.38	42	74	1.8	25.2	31.0	1.23
T10A	32.1	0.79	0.43	41	75	1.8	21.0	27.7	1.32
T12A	35.1	0.89	0.53	39	66	1.7	8.18	9.59	1.17
average		0.67	0.41	41	68	1.6	28.7	34.5	1.21

To further investigate the utility of aFT-IMS, several charge states of ubiquitin (~8.5 kDa) were processed using both mFT and aFT methods. aFT processing consistently produced narrower and better-defined peaks while eliminating peak tailing and reducing baseline noise, as seen in Figure 4. In particular the 8+ charge state, which is known to contain at least two conformations and produce a bimodal ATD under the denaturing solution conditions, used here³⁴ (50% MeOH w/ 0.1% formic acid) only displays this characteristic peak shape when processed in aFT mode. This result suggests that aFT processing has significant advantages for intact protein analysis with IM-MS, which can be particularly challenging owing to the large collision cross sections of protein ions combined with numerous charge states and conformations that dilute limited ion signals leading to a low SNR. Future work will explore the application of this method to larger proteins and protein complexes.

Figure 4.

ATDs of selected charge states of denatured ubiquitin processed in magnitude and absorption modes following phase and baseline correction. Peak widths are reduced, shapes are improved, and the noise level is reduced. XICs from three sweeps were averaged to produce the ATDs shown here. Although each species is plotted individually, the spectra were acquired from a mixture containing all six species in a single experiment, and the same phase correction function was applied to all ATDs. The individual plots reflect the fact that these data are inherently resolved in both mobility and m/z domains.

The use of apodization functions and zero padding was examined with the aim of reducing acquisition time in FT-IMS workflows. The acquisition time required for an FT-IMS experiment is determined by two factors: the range over which the gating frequency is swept and the rate of the sweep in Hz/second. The maximum sweep rate that can be used in a given experiment is ultimately determined by the Nyquist limit with respect to the frequency encoding each drift time (f_en) that is directly proportional to the sweep rate (f_en = sweep rate/drift time) and must be kept at or below half the scan rate of the mass spectrometer. In other words, for a given set of scan parameters such as mass analysis time (i.e., the transient duration on an FT-MS instrument), ion injection time, and any additional scan overhead, the maximum usable sweep rate is determined by the encoding frequency of the longest drift time to be measured. In principle, the gating frequency range only needs to be large enough to capture the full duration of oscillations in the XIC, which typically decays at gating frequencies above 1–3 kHz (Figure S3). However, most FT-IMS sweeps use final frequencies in the range of 5 to 10 kHz to assure a sufficient density of data points to fully capture the shapes of peaks in ATDs, even though the oscillations that encode drift times typically decay long before the final frequency is reached, meaning that several minutes of the sweep are often “wasted” collecting data with no usable frequency information. Zero padding (appending a series of zeros to the time domain data to simulate a longer acquisition) is an attractive option that promises to increase the density of data points in the ATD without the need for longer acquisition times, but it necessitates the use of an appropriate apodization function to smooth the transition from real data to appended zeros. Standard apodization functions, such as the Hanning window,^21,35 typically used in conjunction with mFT processing drive both the beginning and end of a time domain signal to zero and thus suppress the most information-rich early portion of FT-IMS XICs, making it a poor fit for this type of data. On the other hand, half-window functions commonly used with aFT processing that preserve the beginning and only suppress the end of a time domain signal are an excellent fit with this data as they closely resemble the natural decay profile of the signal while still smoothing the transition to appended zeros.

Figure 5 shows the application of both types of apodization windows to the XIC of tetraoctylammonium cations (T8A) and their respective effects on both the mFT and aFT peak shapes. Following full-Hanning window apodization, the mFT peak shape is noticeably improved and the long tails seen in the unapodized ATD are eliminated, both of which were anticipated. However, the intensity of the peak is reduced by more than 80% owing to the suppression of the beginning of the time domain signal, and the peak displays significant broadening. The full-Hanning window apodized aFT peak shows an even greater reduction in intensity in addition to significant negative side lobes, which is also expected. Both observations demonstrate the inappropriateness of using full-window apodization for these data. Half-window apodization, which is typically used in conjunction with aFT processing,^22,35 is expected to produce significant peak tailing when used with mFT processing. However, in FT-IMS data, signal intensity decreases sharply over the course of the sweep, creating a profile that closely resembles the half-window function even when no apodization is applied. Therefore, it is reasonable to assume that the characteristic artifacts expected from mFT processing of half-window apodized data will also be present in unapodized data, which is observed in all the data presented here. Indeed, comparison of the unapodized and half-Hanning-window-apodized mFT peaks reveals only minimal differences in peak shape and extent of tailing. In contrast, aFT processing of the half-Hanning-window-apodized data produces a well-defined symmetrical peak with no noticeable tailing or side lobe artifacts. Based on these observations, combining half-window apodization functions with aFT processing is ideal when apodization of FT-IMS data is desired, though mFT processing of half-window-apodized data may still produce acceptable results.

Figure 5.

Illustration of effects of the apodization window choice on mFT and aFT spectra of tetradecylammonium T6A (1+). Top row: XICs (a) with no apodization, (b) following full Hanning window apodization, and (c) following half-Hanning window apodization. Bottom row (d–f): ATDs resulting from Fourier transformation of each XIC from the top row, respectively.

Using aFT processing with half-window apodization, we designed an experiment to evaluate the use of shorter duration frequency sweeps with lower final frequencies for the 10+ charge state of ubiquitin. Intact proteins are an attractive target for this strategy as their frequency signals decay significantly faster than smaller molecules (Figure S3), so the potential exists to substantially reduce acquisition times without sacrificing any frequency information. Starting with a standard 5 min, 5 Hz to 5 kHz sweep, the sweep time was reduced in 1 min increments while simultaneously reducing the end frequencies in 1 kHz increments to maintain a nominally constant sweep rate of ~16.6 Hz/second (exact parameters for each sweep are shown in Table S1). Each of these five data sets was then zero padded between one and nine times to simulate sweep durations between 10 and 12 min with final frequencies between 10 and 12 kHz and processed without interpolation or smoothing. Fourier transformation and aFT processing of these XICs (shown in Figure S4) produced the ATDs that are displayed in Figure 6. All five options produced narrower peaks with higher SNRs compared to mFT processing of a standard 5 min sweep, indicating that the use of proper apodization functions and zero padding offers the potential to reduce acquisition times by up to 80% without sacrificing resolution. Future work will explore the application of this technique to other types of ions varying in sizes and charge states.

Figure 6.

Comparison of ATDs of ubiquitin (10+) processed in the magnitude mode (blue traces) or absorption mode (red traces) with apodization and zero padding using sweeps with different end frequencies and durations. All ATDs were apodized with a half-Hanning window and zero padded between one and nine times to simulate sweeps lasting 10–12 min. Even the shortest sweep (1 min) processed using aFT and apodization with zero padding shows improved resolution and SNR compared to standard mFT processing of a full 5 min sweep.

CONCLUSIONS

A method to generate absorption mode arrival time distributions from Fourier transform ion mobility mass spectrometry has been presented and validated. Average resolution improvements of 1.6-fold and SNR improvements of 1.2-fold were demonstrated with no modifications to hardware or experimental parameters. Full- and half-window apodization functions were evaluated for both mFT and aFT processed data, and the half-window function combined with aFT processing was shown to produce the best peak shape with minimal artifacts, peak broadening, and loss of sensitivity. Combining apodization with zero padding allows data acquisition times to be shortened by up to 80%, allowing a full experiment to be carried out in 1 min or less with minimal loss of information. Future work will focus on realizing additional improvements in the SNR and further reducing acquisition times to expand the application of Fourier transform multiplexing to a wider array of sample types and more high-throughput workflows.