Synchronized Stepped Frequency Modulation for Multiplexed Ion Mobility Measurements

Abstract

Implementation of frequency encoded multiplexing for ion mobility spectrometry (e.g. Fourier Transform Ion Mobility Spectrometry, (FT-IMS)) has facilitated the direct coupling of drift tube ion mobility instrumentation with ion trap mass analyzers despite their duty cycle mismatch. Traditionally, FT-IMS experiments have been carried out utilizing continuous linear frequency sweeps that are independent of the scan rate of the ion trap mass analyzer; thus, creating a situation where multiple frequencies are swept over two sequential mass scans. This in turn creates a degree of ambiguity in which the ion current derived from a single modulation frequency cannot be assigned to a single datapoint in the frequency modulated signal. In an effort to eliminate this ambiguity, this work describes a discrete stepwise function to modulate the ion gates of the IMS while synchronization between the generated frequencies and the scan rate of the linear ion trap is achieved. While the number of individual frequencies used in the stepped frequency sweeps is less than in continuous linear modulation experiments, there is no loss in performance and maintain high levels of precision across differing combinations of terminal frequencies and scan lengths. Furthermore, the frequency-scan synchronization enables further data processing techniques such as linear averaging of the frequency modulated signal to drastically improve signal-to-noise ratio for both high and low intensity analytes.

Introduction

Adding a method of separation such as ion mobility to mass spectrometry allows for identification of isobaric compounds such as enantiomers or structural isomers that contain features not seen in the mass domain.^1–4 While there are benefits to this particular combination of instrumentation, there are challenges to be overcome such as the duty cycle mismatch between both instruments. Typically, signal averaged drift tube ion mobility spectrometry experiments (DT-IMS) allow admission of one packet of ions into the drift region at a time. Separation of ions within the packet is based on differences in velocity under a weak electric field and a counterflow of a neutral drift gas; a metric known as its mobility coefficient. This process is repeated many times and occurs at an interval in the order of tens of milliseconds. In contrast, mass spectrometers such as linear ion traps have an acquisition rate in the order of one mass scan per hundreds of milliseconds; a rate too slow for nested mass information within the arrival time distribution. This challenge has been overcome by implementation of Fourier or Hadamard -based multiplexing methods to this hyphenated technique (IMS-MS). ^5–8 Compared to the <1% duty cycle of signal averaged IMS experiments, Fourier based multiplexed IMS experiments reach upwards of 25% duty cycle and drastically decreases the amount of sample loss from continuous ionization sources such as the commonly used electrospray ionization.⁹ In addition to duty cycle improvements, there are inherent throughput and signal-to-noise (SNR) ratio improvements in multiplexed experiments attributed to both the Fellgett and Jacquinot advantages, respectively.^10–12 In order to mitigate ion losses and realize better ion utilization, approaches such as ion accumulation through the use of ion traps,¹³ ion funnels,⁹ PASEF,¹⁴ TWAVE,¹⁵ and SLIM onboard accumulation¹⁶ have been implemented. The caveat to these approaches is they require radially confining RF fields and lower pressures (i.e. <10 Torr) to function. In contrast, multiplexing approaches can be implemented using DT-IMS operating at atmospheric pressure which simplifies implementation.

Fourier based multiplexing experiments (FT-IMS) involve the simultaneous opening and closing of two ion gates at each end of the IMS drift region. The modulation of these gates is accomplished by using a predefined frequency sweep typically starting at a low but non-zero frequency (e.g. 5 Hz) and ending at a desired terminal frequency (e.g. 8005 Hz).⁸ For a given frequency, only ions with specific mobilities are able to traverse through the drift region and continue past the second gate while it is still in the open state. Following the second gate, ions arrive at the vacuum interface of the mass spectrometer for subsequent mass analysis and detection. The sweeping of these frequencies and continuous generation of ions into the IMS generates an oscillating signal for all masses within the user-defined scanning range of the mass spectrometer. Fourier transform of this oscillating signal will generate a frequency domain representation which can then be directly converted into an arrival time distribution (ATD) by dividing the frequency (Hz) axis by the sweep rate (Hz/s).⁵

While Fourier-based multiplexing of ion mobility experiments has seen success,^6,17–22 there are ambiguities that exist within the experiment, particularly the assignment of frequencies at a given time point within the frequency modulated signal. In short, the traditional method of performing FT-IMS experiments requires synchronizing only the start and end times of the frequency sweep and first and last scans of the mass spectrometer. When this approach is taken, frequencies are swept through in a continuous manner whereas mass scans are acquired in semi-repetitive fashion according to user-defined scan parameters (e.g. mass range). Though not immediately obvious to the casual user, the target scan rate in many commercial ion trap mass analyzers is not rigorously followed and often can vary by tens to hundreds of milliseconds owing to instrumental overhead (e.g. resetting of electronics and automatic gain control). Consequently, the continuous linear frequency sweeps in the traditional FT-IMS experiment and the semi-regular mass scans establish conditions where uncertainty emerges as to which ion gating frequency is contributing to the ion current observed in the mass spectrum. More specifically, this ambiguity arises when a recorded mass scan captures ion signal from more than one frequency (i.e. all the frequencies swept over since the previous mass scan). Ameliorating this particular issue, we have modified the FT-IMS experiment to instead use a discrete stepped frequency sweep that allows for synchronization of the scan rate of the ion trap mass spectrometer with the sweeping of frequencies (illustrated in Figure S1 in Supporting Information). This approach removes any ambiguity of which mobilities are admitted into the mass spectrometer at a given data acquisition point. Using common metrics such as resolving power and signal-to-noise ratio, this work evaluates the performance of the stepped frequency sweeps across differing sweep parameters. Despite the stepped sweeps going over a lower number of frequencies than the continuous linear sweeps, there are no losses in performance and high levels of precision are maintained. Furthermore, due to the unambiguous designation of frequencies at each data acquisition point, linear averaging of the frequency modulated signal becomes possible and therefore allows for drastic increases in signal-to-noise ratio in the arrival time distribution for both high and low intensity analytes.

Experimental Methods

Atmospheric Pressure Dual-Gate Drift Tube Ion Mobility Linear Ion Trap Mass Spectrometer

Ion mobility experiments were performed using an atmospheric pressure dual-gate PCB drift tube system operating under ambient conditions (~690–700 Torr, ~25° C). The drift region of the IMS was 17.4 cm and bracketed by two sets of ion gates following the tri-grid shutter principles outlined by Reinecke et al. and Langejuergen et al. The ion gates were driven by a set of open-source ion gate pulsers described previously.^23–25 Prior to the drift region, ions are desolvated in a 10 cm region containing the same electric field gradient as the drift cell. Voltages applied to the drift region were ~490 V cm⁻¹ for all experiments. The ion mobility spectrometer was coupled to a linear ion trap mass spectrometer (LTQ, Thermo Scientific, San Jose, CA). A counterflow of dry and purified nitrogen drift gas was introduced into the drift region at 2.5 L min⁻¹ at the interface between the IMS and LTQ. It is worthy to note that the conductance of the inlet capillary is approximately 2 L min⁻¹ leaving ~500 mL min⁻¹ as a counter current flow.

Chemicals and Reagents

Analytes used to evaluate the performance of each of the sweeps included tetraalkylammonium salts (Sigma-Aldrich, St. Louis, MO), morphine (Cerrilant, Round Rock, TX), cocaine (Cerrilant, Round Rock, TX), and leucine-enkephalin (Sigma-Aldrich, St. Louis, MO). For experiments where continuous linear and stepped frequency sweeps are compared, a mixture of 500 nM tetrabutylammonium (T4A), tetraoctylammonium (T8A), tetradodecylammonium (T12A), and 10 μM leucine enkephalin (leu-enk) were used. All subsequent experiments used a mixture of 2 μM morphine, 500 nM cocaine, 500 nM T8A, and 10 μM leu-enk. All solutions were made in 95:5 mixture of acetonitrile and water with 0.1% formic acid and electrosprayed into the desolvation region of the IMS. A custom glass capillary emitter with a nominal ID of 75 μm was held at +2500 V relative to the first electrode on the IMS.

Frequency Sweep Generation and Data Analysis

Custom python scripts were developed to generate the frequency sweeps used to modulate the ion gates of the ion mobility spectrometer. All sweeps started at 5 Hz with terminal frequencies of 2505, 5005, 7505, and 8005 Hz. The lengths of these sweeps were 500, 1000, 1500, 2000 scans for the stepped frequency sweeps, which in turn determined the frequency step size throughout the duration of the experiment. Due to the lack of sweep-scan rate synchronization in the continuous linear sweep experiments, their duration was matched to the scan length duration of the stepped experiments in order to maintain an equivalent sweep rate. For reference, a 2000 scan experiment is generally ~7 minutes and may vary depending on the mass range of interest. To generate the stepped sweeps, a 200 MHz Arduino-compatible microcontroller (Wio Terminal, Seeedstudio) running code developed in-house was used to modulate the ion beam within the drift cell. The changes in the frequency output by the waveform generator (Analog AD9850) were synchronized with the beginning of each mass scan of the linear ion trap using the TTL output from the LTQ. As detailed in the assembly instructions (see below), a voltage level shifter was used to accommodate the different signal levels from the LTQ and the 3.3 V logic levels used for the microcontroller. For comparison experiments, continuous linear frequency sweeps were generated using the previously referenced python script and streamed to the ion gates using IGOR Pro 8.0 (Wavemetrics, Lake Oswego, OR) and a NI-Multifunction DAQ data acquisition system (USB-6351, National Instruments, Austin, TX). For these latter experiments only the start of the waveform generation was synchronized with the ion trap using a contact closure mechanism. The relevant python scripts, firmware code, and bill of materials for the stepped frequency generator are included as part of the Supporting Information and at the following github repository (https://github.com/bhclowers/Stepped-Pulser).

To process the respective ion modulation approaches datasets were converted from the LTQ “.RAW” file format into the “.mzML” file format using the “msConvertGui” tool from ProteoWizard.²⁶ Using Python scripts developed in-house, mass spectra and extracted ion chromatograms (frequency modulated signals) were generated. Each of the frequency modulated signals was extracted from a user defined mass window, in which for all cases, the mass window included the monoisotopic peak of interest and its associated isotopic peaks. The first step in processing the frequency modulated signal required making a new time axis vector of evenly spaced data points that closely matched the original. A spline object of the data was subsequently created and evaluated at the new time axis vector. Creating this spline with evenly spaced points was necessary due to small variations in time between scans across multiple replicates for the ion trap data acquisition cycle. It is important that this spline is created using the same number of points as the original dataset in order to maintain the original sampling frequency of the LTQ and to ensure the Nyquist frequency was not erroneously changed. The newly splined signal was then Fourier transformed where the magnitude of the signal was used. To recover the arrival time distribution, the frequency axis (Hz) obtained by the Fourier transform was divided by the sweep rate (Hz/s) and multiplied by 1000 in order to create an axis represented in milliseconds. Further details on code and python packages used to process the datasets can be found in the Supporting Information and on the previously referenced github repository.

Results and Discussion

Stepped and Continuous Linear Sweep Resolving Power Comparison

In order to ensure there was no loss in performance when implementing the stepped frequency sweeps, resolving power between both stepped and continuous linear methods were compared. For this comparison and all other resolving power calculations in this work, the definition of resolving power used was R_p = t_d/FWHM, where t_d is the drift time of the analyte and FWHM is the width of the peak at half its maximum intensity. Analytes for this comparison were the 1+ charge state monomer of leu-enk, T4A, T8A, and T12A. Sweep parameters for these comparisons were as follows: starting frequency for all experiments was 5 Hz with varying terminal frequencies of 2505, 5005, 7505, and 8005 Hz. Scan lengths were kept constant at 2000 scans for all stepped frequency experiments. As for the continuous linear sweep experiments, due to lack of scan-frequency synchronization, sweep time was chosen to match the length of the stepped sweeps in order to maintain an equivalent sweep rate (Hz/s) and therefore only start and end of experiments were matched.

Resolving power values and their standard deviations are shown in Figure 1, with each resolving power value being the average of five replicates. Across all analytes used in this comparison, resolving power was similar for stepped and continuous sweeps across all terminal frequencies used, with the stepped sweeps showing a slight improvement across almost all comparisons. This result was expected and essential for deployment of a new waveform generator and modulation technique. For both sweep types, resolving power reaches values ~80–90 at terminal frequencies as low as 5005 Hz. In general, as terminal frequency is increased, resolving power also increases, albeit with small gains after 5005 Hz. This trend is in agreement with previous FT-IMS experiments conducted by Morrison et al.⁶ Another important observation from this dataset is that the stepped frequency sweeps often displayed higher degrees of precision than results from the continuous sweeps. The most drastic improvements were seen at terminal frequencies of 2505 and 5005 Hz as shown in Figure 1. We attribute this trend to the unambiguous nature of the frequency designation at each data point. This in turn yields ion signal resonant with a single frequency and therefore allowing for more uniformity in the frequency modulated signal across replicates.

Figure 1.

Resolving power comparison between continuous linear and stepped frequency experiments at 2000 scans for tetraalkylammonium salts, T4A, T8A, T12A, and leu-enk singly charged monomer. With highest resolving power values achieved at higher terminal frequencies, both methods show similar performance across all sweep parameters. Each bar value is an average of five replicates and error bars show standard deviation.

Resolving Power and SNR Evaluated Using Different Sweep Parameters

To further evaluate the performance of the stepped frequency sweeps, resolving power and signal to noise ratio were measured at differing combinations of terminal frequencies and scan lengths to determine whether scan length or terminal frequency had a greater impact on both of these metrics. For these experiments, cocaine, T8A, and leu-enk were used. For the purposes of discussion surrounding the observed leucine enkephalin species, the doubly charged homodimer (i.e. [2M+2H]²⁺) and monomeric forms [M+H]⁺ correspond to the the abbreviated peaks leu-enk #1 and leu-enk #2, respectively. It should be made clear that while extensive work on species identification for each peak in the leu-enk ATD was not performed, peak assignments were based on collisional cross section (CCS) calculations for the [M+H]⁺ monomer showing agreement with reported CCS values²⁷ and relative mobility trends between monomers and doubly charged homodimers showing agreement with previous reports.²⁸ Scan parameters used for these experiments were as follows: starting frequency for all experiments was 5 Hz, terminal frequencies at 2505, 5005, 7505, and 8005 Hz, and the sweeps carried out at scan lengths of 500, 1000, 1500, and 2000 scans.

Figure 2 shows the resolving power for all sweep parameters tested. Across these parameters, the highest resolving power values are achieved at higher terminal frequencies; in agreement with the results shown in Figure 1. This trend was maintained regardless of scan length. In some cases resolving power values exceed 100, such as for the leu-enk homodimer (leu-enk #1) at 1500 scans and a terminal frequency of 5005 Hz. However, resolving power across most scan-length experiments continues to show a plateauing at this frequency and slightly later at 7505 Hz for the 500 point datasets. This indicates that lower terminal frequencies in combination with shortened scan lengths will yield similar resolving power to experiments utilizing higher terminal frequencies and experimental times. It is important to note that for the 500 scan subplot in Figure 2, leu-enk peak #2 and T8A were intentionally excluded. This was done due to the combination of sweep rate and Nyquist frequency at terminal frequencies greater than 7505 Hz limiting the maximum drift time in the arrival time distribution (ATD) that could be recorded (the relationship between drift times that can be captured with a given set of frequencies is described by the work by Knorr and Siems).⁵ This cutoff results in the Fourier transform of the signal to give incorrect frequencies from this aliased signal and therefore giving incorrect drift times for these particular analytes. Due to the direct relationship between resolving power and drift time, these were removed from Figure 2 due to erroneous data and comparisons.

Figure 2.

Resolving power measured for different combinations of sweep lengths and terminal frequencies for cocaine, T8A, leu-enk dimer (leu-enk #1), and leu-enk monomer (leu-enk #2). Across all swep lengths, the highest resolving power values are achieved at higher terminal frequencies. Each point is an average of five replicates with error bars showing standard deviation.

For all signal to noise ratio measurements, signal was the maximum of the peak of interest and a region of 10 ms (e.g. 10–20 ms) in the ATD with little to no ion signal was chosen for the noise region. Using this region, the average of the baseline intensity plus 3x its standard deviation was used as the noise value. Figure 3 shows SNR values across the different sweep lengths and terminal frequencies used for cocaine, T8A, and both leu-enk homodimer (leu-enk #1) and monomer (leu-enk #2). Across all terminal frequencies used, the signal to noise ratio increases are seen as the number of scans per experiment is increased. SNR values for the leu-enk homodimer are much lower than all other analytes in the comparison due to low relative abundance relative to the monomer species, as shown in the arrival time distribution in Supporting Information (Figure S2). A SNR comparison between the leu-enk homodimer and signal extracted from lower intensity isotopic signals can also be seen in Supporting Information (Figure S3 and Figure S4). The high variability for T8A SNR values for 2505 Hz subplot in Figure 3 is due to the chosen noise region interval. In order to maintain consistency the same noise region in their respective ATD used for all analytes. For this particular region, the T8A showed the highest variation in baseline intensity and higher levels of precision could be attained if a more appropriate region was chosen. However, although variability was higher for T8A, it showed high SNR values and followed the same SNR trends as all other analytes, where values increase with longer scan lengths.

Figure 3.

Signal to noise ratio (SNR) comparison for cocaine, T8A, leu-enk homodimer (leu-enk #1), and leu-enk monomer (leu-enk #2). SNR measured at differing combinations of terminal frequencies and scan lengths. Longer scans produce higher signal to noise ratio values at each terminal frequency used. Each point is an average of five replicates and error bars show standard deviation.

This dependence of SNR on scan length as shown in Figure 3 is attributed to the increased number of points across the span of the frequency modulated signal for a given terminal frequency; allowing for a more complete representation across oscillations and thus capturing maximum ion current where full frequency resonance occurs. This allows for a cleaner and more uniform frequency modulated signal and reduced noise levels post-Fourier transform. Furthermore, another contributing factor is the raw number of bins that allow for noise to spread out into and therefore increasing the benefit from the multiplex advantage.

SNR Improvements from Signal Averaging

The unambiguous designation of frequencies at each data point allows for the assumption that the ion signal gathered at each point is from ion-current resonant with a single frequency. As a consequence of this, frequency modulated signals across multiple replicates are highly uniform, examples of which are shown in the left column of Figure 5. However, even though high signal to noise ratios are achieved at the acquisition level, they are not identical. Fluctuations in experimental conditions such as pressure, temperature, or instrumental noise contribute to variation between replicates, and therefore higher quality arrival time distributions can be reconstructed by linear averaging of the frequency modulated signal pre-Fourier transform.

Figure 5.

Interpolated frequency modulated signals for morphine shown in the left column taken over (a) 500 and (b) 1000 scans and plotted up to the first 5 and 10 seconds, respectively, to show detail in the aligned oscillating signal. Linear averages of these signals are shown in the center column and their Fourier transform shown on the right column. Importantly, as the sample rate diminishes a larger degree of ambiguity arises when converting the data to an ATD that distorts peak shape. A balance is required between the total experimental time and terminal sweep frequency in order to preserve the capacity to signal averaging the data.

Spectra derived from continuous frequency sweep FT-IMS experiments are often the result of a single frequency sweep. As noted previously, the slight variations in instrumental overhead for commercial ion trap mass analyzers make direct signal averaging under such conditions difficult. Alternatively, the stepped frequency approach provides a foundation to directly average separate experiments. For the signal averaging experiments using the synchronized frequency sweeps, data handling involved random sampling from a population of 16 intraday datasets acquired to minimize experimental conditions. For each number of averages, random sampling (without replacement) of the datasets was repeated five times, effectively creating five different replicates. Scripts performing the random sampling of the datasets included code to prevent any duplicate sets from being considered. It is worth noting that averaging was done after mass extraction of the frequency modulated signal for individual analytes. The compressed time span over which data were acquired was necessary to minimize the impact of environmental fluctuations on the observed signal to noise ratio. It is recognized that the observed measurement uncertainty may be biased through this exercise of random resembling of the technical replicates, but does provide a reasonable approach to evaluate the trends with signal to noise ratio.

Figure 4 shows the signal to noise ratio trends for all analytes in a mixture of morphine, cocaine, T8A, and both the singly charged monomer and doubly charged dimer of leu-enk. Implementation of signal averaging yielded large signal to noise ratio improvements and continued to increase as the number of averaged signals was also increased. To be clear, due to the lack of exactly periodic cycling of the ion trap, prior efforts at averaging using the continuous sweep conditions and FT-IMS introduce a level of noise that degrades the expected SNR gains. The sweep conditions used for the stepped frequency experiments shown in Figure 4 were 500 and 1000 scans using stepped frequency sweeps starting at 5 Hz with a terminal frequency of 7505 Hz. It’s worth noting that morphine was the lowest intensity analyte peak and about 1/7th of the intensity of the tallest peak in the mass spectrum. It had the most significant improvements in signal-to-noise ratio, showing ~5x improvement in the 500 scan sweeps. A representative mass spectrum showing relative intensities can be seen in Supporting Information Figure S5. Examples of the averaged signals for morphine and the resulting Fourier transform can be seen in Figure 5 center and right columns, respectively.

Figure 4.

Signal to noise ratio (SNR) comparison for the linear averages of (a) 500 scan and (b) 1000 scan-length frequency modulated signals for morphine, cocaine, T8A, and both leu-enk monomer and dimer, each column labeled accordingly. Each point is the average of 5 replicates and error bars show standard deviation. Significant signal to noise ratio increases are seen for all analytes as the number of averages increases.

An important consideration for conducting signal averaging of stepped FT-IMS experiments is that averaged signals should be recorded within a reasonable amount of time from each other. As mentioned previously, analyte drift times will vary depending on fluctuations in experimental conditions, as a consequence of this, drift times will have differing resonant frequencies and cause shifting in the oscillations within the frequency modulated signal. These shifts will cause broadening when averaged and therefore degrade the quality of the reconstructed arrival time distribution. A trade-off that must be considered for these experiments is whether a longer sweep with a low number of averages or shorter sweeps with a greater degree of averaging is desired.

The left column in Figure 5 shows 16 frequency modulated signals overlaid on top of each other. In order to show detail of the signal oscillations, the signal was only plotted to the first 5 and 10 seconds for 500 and 1000 scan datasets, respectively. Time intervals shown for the frequency modulated signals were chosen to show a direct comparison in the same number of oscillations but at differing number of points between each; 5 seconds at 500 scans and 10 seconds at 1000 scans with the same terminal frequencies will produce the same number of oscillations in the signal due to the relationship between their effective sweep rates. The main differentiating factor here is the point density across the signal for 1000 scans is two times higher than for 500 scans across the same time interval and therefore producing more uniform peaks. Linear averaging of the signals in the left column of Figure 5 produces the signal shown in the center column. Fourier transform of this averaged signal produces the arrival time distribution shown in the right column. Even with the direct synchronization with the LTQ, the data sets from experiments with a reduced number of experimental points (Figure 5, Top Row) illustrate that an adequate number of frequency steps must be included to fully capture the oscillations in the frequency domain. Failure to adequately account for this behavior can induce aliasing and spectral broadening in the ATD due to the poorly resolved maxima and minima in the recorded signal.

Conclusion

Supplanting continuous linear sweeps with discrete stepped frequency waveforms does not adversely impact the resolving power, signal to noise ratio, or precision of ion mobility measurements made in combination with an ion trap mass analyzer. Furthermore, the unambiguous designation of frequencies in the frequency modulated signal generated by the stepped sweep approach allows for greater confidence in peak assignment in the frequency domain and, by extension, confidence in drift times in the reconstructed arrival time distribution. By minimizing any ambiguity in the modulation frequency delivering ion current to the mass analyzer, additional signal processing techniques are possible. Due to small fluctuations in ion trap acquisition rate, the use of the continuous linear frequency sweeps largely precluded traditional approaches to signal averaging. The synchronized, stepped frequency waveforms, on the other hand, is compatible with direct signal averaging between runs which drastically increases the signal to noise ratio for both high and low-intensity analytes. However, trade-offs between number of averaged signals and length of experiment must be taken into consideration in order to maintain near identical experimental conditions (e.g. temperature and pressure) as to not lose precision in the frequency modulated signal’s oscillations. Through the use of open-source code and easily attainable hardware for both building the PCB DT-IMS²⁹ and frequency generation, the presented solution proves to be a low cost, highly flexible approach to implementing FT-IMS experiments. The concept of efficiently coupling drift tube IMS systems with ion trap mass analyzers is a relatively new development within the field with noticeable improvements in speed and resolution. To realize the ultimate potential of multiplexing in ion mobility for experiments employing ion trap mass analyzers, this report represents a critical, foundational step towards data-driven and customized modulation waveforms which find extensive utility in other analytical chemistry techniques including NMR and FT-ICR.^30–33