Enhanced Charge-Sensitive Amplifier Performance Leads to Substantially Reduced CD-MS Measurement Times for Charge State Resolution

Abstract

In charge detection mass spectrometry (CD-MS) ions are trapped in an electrostatic linear ion trap (ELIT) where they oscillate back and forth through a conducting cylinder. The oscillating ions induce a periodic charge separation that is detected by a charge sensitive amplifier (CSA) connected to the cylinder. The resulting time domain signal is analyzed using short-time Fourier transforms to give the mass-to-charge ratio and charge for each ion, which are then multiplied to give the mass. For ions to be assigned to the correct integer charge states with a low error rate, the charge should be measured with a precision of <0.2 e (elementary charges). Electrical noise reduces the precision of the charge measurement. However, the effect of the noise can be ameliorated by signal averaging, and the measurement time can, in principle, be increased to achieve a precision of <0.2 e. Previously, through optimized ELIT design and improvements to the CSA, the measurement time (with a cryogenically cooled input JFET) required to achieve a charge precision of <0.2 e was reduced by a factor of 2, from 3 s to 1.5 s. In this study, further improvements in JFET selection, capacitance matching, and cryogenic cooling has allowed us to further reduce the electrical noise so that the target precision of <0.2 e can now be achieved for mAb MSQC4 in 600–700 ms with the input JFET cryogenically cooled, and in 900–1000 ms with the input JFET at room temperature. This performance upgrade cuts the overall time for high-resolution charge measurements by more than another factor of 2. For a measurement time of 100 ms, the charge RMSD is 0.51 e with cryogenic cooling. The results presented here further cements CD-MS with an ELIT as the fastest and most accurate approach to single ion MS measurements.

Graphical Abstract

INTRODUCTION

In conventional mass spectrometry (MS) the m/z (mass to charge) ratio is measured for an ensemble of ions and then used to determine their masses. Initially, MS measurements were performed on small volatile molecules where z was usually 1 and the mass equaled the measured m/z. Subsequently, electrospray increased access to higher mass samples^1–4 but introduced a new problem.⁵ High mass ions are generated by electrospray in a broad range of charge states, with z much greater than 1. The charge states must then be determined to convert the measured m/z to mass. There are several strategies available to determine the charge states, including the use of isotopic spacing in high resolution spectra, and the separation between m/z peaks with the same mass but different charge states. However, these strategies become untenable as ion size and heterogeneity increase and the m/z spectra become uninterpretable due to complexity from overlapping peaks.

Adopting a single ion approach and directly measuring the charge and m/z for each ion overcomes the spectral congestion problem.⁶ In the first implementation of this strategy, charge detection mass spectrometry (CD-MS), the ions pass through a conducting cylinder and when inside the cylinder, they induce a charge that is sensed by a charge sensitive amplifier (CSA). If the cylinder is long enough, the induced charge equals the charge on the ion. The m/z ratio can be deduced from the flight time through the cylinder and the charge is obtained from the signal amplitude. This approach was first applied to electrospray ions in 1995.⁷ A CSA with a junction field effect transistor (JFET) at the input was employed and the resulting signals were amplified and processed by Gaussian pulse shaping. The ions passed through the detection cylinder once (single pass CD-MS). Electrical noise limits the performance and results in a high limit of detection (~250 e) and a large uncertainty in the charge (~50 e) (where e = elementary charge).

The signal-to-noise ratio can be improved by signal averaging, either by using a detector array,^8–12 a linear array of detection cylinders one after the other, or by embedding a detection cylinder in an electrostatic linear ion trap (ELIT) so that the ions bounce back and forth through the same cylinder many times.^13–15 The ELIT yields the best performance and it is currently the preferred approach, though the array has higher throughput. An ELIT consists of two opposing end-caps. The potentials on the end-caps can be changed to switch them from transmission mode to reflection mode where they act as ion mirrors. The oscillating ions generate periodic signals that are digitized and sent to a computer for storage and analysis. The time domain signals are analyzed by fast Fourier transforms (FFTs). Instead of using a single FFT of the full trapping event, short time FFTs (STFT) are stepped through the time domain data to ensure that ions remain trapped until the end of the trapping event (an approach that has recently been called frequency chasing).^14,16 The signal generated by an oscillating ion is a rounded square wave and the FFT contains a series of peaks due to the fundamental and harmonics. The harmonic relative intensities depend on the duty cycle, the fraction of the ion’s oscillation period spent in the detection cylinder. With a 50% duty cycle, the odd numbered harmonics vanish, and the relative intensity of the fundamental is maximized.¹⁷ The ion’s m/z ratio is determined from the fundamental frequency, f:

$$\frac{m}{z}=\frac{C}{f^2}$$ (1)

where C is a constant that depends on the ELIT geometry and the ion energy.¹⁴ The charge is determined from the magnitude of the fundamental. Parabolic interpolation is used to minimize scalloping loss.^14,16 The signals are processed in real time (i.e., each trapping event can be analyzed in less time than required to collect the data).¹⁶ Real time analysis facilitates optimization of the experimental conditions.

The mass resolving power is given by m/Δm, where Δm is the full width at half-maximum of the peak in a histogram composed of masses measured for an ensemble of ions of identical mass.¹⁸ The imprecision in the m/z determination and the imprecision in the charge determination both contribute to width of the mass peak. However, charge is quantized and if the charge is measured accurately enough that each ion can be assigned the correct charge state with a low error rate, the imprecision in the charge vanishes leaving the mass resolution solely determined by the m/z resolution. To accomplish this goal the charge root-mean-square deviation (RMSD) needs to be less than 0.2 e.

The precision of the charge measurement is determined by measuring the charge for an ensemble of ions of known identical charge and then evaluating the RMSD. Such a measurement can be performed for a sample with resolved charge states in the m/z spectrum. In this case, ions in different m/z charge states can be separated and the RMSD determined for each charge state. When m/z charge states are not resolved, ions can be separated into different charge states using the charge measurement. However, this can only be performed if the precision of the charge measurement is around 0.2 e or better, where the charge states are well-resolved in the charge spectrum. If the charge RMSD is significantly larger than 0.2 e a significant fraction of the ions will be assigned to the wrong charge state.

Noise reduces the precision of the charge measurement; there are two main sources: microphonic (from vibrations) and electrical noise. Electrical noise covers the whole frequency range, while increasing at lower frequencies. Microphonic noise occurs at specific frequencies so it can be isolated in an FFT spectrum of the noise. Microphonic noise in the frequency range of interest has been largely eliminated by careful design of the vacuum chamber, ELIT mount, and particularly the mount for the detection cylinder within the ELIT. Thus, the main factor limiting the precision of the charge measurement is electrical noise and the precision can be improved by signal averaging. In principle, the measurement time can be extended to achieve the target RMSD:

$$\text{charge RMSD}\propto \frac{\text{noise}}{\sqrt{\text{measurement time}}}$$ (2)

Initially, a measurement time of 3 s was required to achieve a charge RMSD of <0.2 e with a cryogenically cooled JFET at the input of the CSA (an Amptek A250).^19,20 Subsequent improvements to the ELIT geometry¹⁷ and CSA circuit design²¹ reduced this value by a factor of 2 to 1.5 s.²² A critical feature enabling high resolution charge measurements is the implementation of dynamic calibration where an antenna radiates a small signal that is picked up by the charge detection cylinder and used to calibrate the short and long-term response of the CSA to negate any drift in its gain.²³

Here we describe further performance upgrades that have allowed us to achieve a charge RMSD < 0.2 e in 600–700 ms for mAb MSQC4 with a cryogenically cooled input JFET. The same precision can be achieved in 900–1000 ms with an uncooled JFET.

OPTIMIZATION OF JFET PERFORMANCE

Figure 1 shows a circuit diagram for a generic low noise CSA with feedback capacitor (C_F) that sets the gain, and feedback resistor (R_F). The feedback resistor is a significant source of thermal noise. It can be removed, but then an alternative method must be provided to reset the circuit and provide a stable operating point. A variety of optoelectronic and transistor resets have been developed.^24–26 However, the additional circuit elements invariably induce noise. We adopted the approach of Bertuccio and co-workers where removal of the feedback resistor is made possible by allowing the JFET to be biased by its gate leakage current, forward biasing the JFET’s gate-channel junction.²⁷ This approach is limited to JFETs with the appropriate leakage current. We have tried other strategies to reset the amplifier, but we were unable to surpass the performance of the resistor-less self-biasing approach.

Figure 1.

Generic circuit diagram for a low noise CSA.

The input JFET (see Figure 1) is a critical component in low noise CSAs. In previous work we had found that hand selection of the JFET could be used to optimize the performance of the CSA and improve the signal-to-noise ratio (S/N). We tested over 100 SMP5484 JFETs (InterFET) using the Atlas DCA75 Semiconductor Analyzer (Peak Electronic Design Ltd.) to measure I_dss (the drain current with 0 V on the gate). However, there was not a good correlation between the drain current and the JFET performance in a circuit. Thus, a more rigorous testing strategy was developed where the JFETs were tested on the bench in a CSA virtually identical to the one used in CD-MS instruments. The amplifier circuit board was modified so that JFETs could be tested without soldering them onto the board. Power was supplied by filtered Rigol DP832 power supplies. The input signal was supplied by a Tektronix AFG3252C function generator, as a 50% duty cycle square wave with amplitude of 100 mV_pp and frequency of 12 kHz. The signal was attenuated by 40 dB before being applied to the gate of the JFET. For S/N measurements, a capacitor (3.2pF) was attached to simulate the capacitance between the detection cylinder and grounded shield of the ELIT. The capacitance was estimated using Ansys Electronics Desktop. The output signal was analyzed using a Rohde & Schwarz RTO2044 oscilloscope set to a Gaussian FFT window. Each JFET was measured for at least 5 min to properly average the output signal and allow the circuit to reach stability. As before, the Atlas DCA75 was used to measure I_dss for each JFET that was tested. Other general component characteristics were obtained from the JFET master Table provided by InterFET (InterFET.com). Initially, ten surface mount (SOT-23) JFET models were evaluated: SMP5484, IF140AST3, SMP4340, IFN160C, SMPJ210, SMPJ305, IFN202, IFN160C, SMPBF511, SMP4340 (all supplied by InterFET). After an initial evaluation, the number was narrowed down to five: SMP5484, IF140AST3, SMP4340, IFN160C and SMPJ210. This list includes the SMP5484 used in our initial implementation of the resistor-less CSA design.²¹ Some of the reported characteristics of these JFETs are given in Table 1.

Table 1.

JFET Characteristics from the JFET Master Table Provided by InterFET

JFET Model	Idss (mA)	Noise (nV/√Hz)	Ciss (pF)
SMPJ210	2–15	2.5	4
SMP5484	1–5	4	5
SMP4340	1.2–3.6	6	7
IFN160	1–6	3	4
IF140ST3	5–15	4	3

Figure 2 shows the measured S/N ratios plotted against the I_dss values determined for the five JFETs mentioned above. The range of I_dss values plotted in Figure 2 for the SMPJ210s JFETs is much narrower than the range quoted in Table 1. In order to maintain a stable operating point, the optimum value for I_dss was thought to be around 2–3 mA and the SMPJ210 JFETs were preselected to lie within this range. The I_dss ranges for the other JFET models were not preselected. The I_dss values for the IF140ST3 JFETs are significantly larger than for the other JFETs, and the values and they show a much wider dispersion. The corresponding S/N values also show a larger dispersion with many of the IFN140ST3 JFETs showing by far the worst S/N performance. The best performance for IFN140ST3 JFETs is obtained with lower values of I_dss, a result that is consistent with the expectation that the optimum value for I_dss is in the 2–3 mA range. Overall, the SMPJ210 JFETs are the best performers with S/N values that are significantly better than the SMP5484 model originally chosen for the amplifier. The better performance of the SMPJ210 versus the SMP5484 is probably related to the lower intrinsic noise and lower input capacitance of the former (see Table 1). Results for the SMP4340 continue this trend: the SMP4340 has a lower S/N than SMP5484 while also having a higher intrinsic noise and higher input capacitance (see Table 1). Note, however, that the IFN160C has signal-to-noise ratios comparable to those for the SMP4340 JFET while having a lower noise and a smaller input capacitance, according to the manufacturer’s specifications.

Figure 2.

Plot of S/N plotted against I_dss for the JFETs discussed in the text. The colors of the points show the JFET model as indicated by the keys in the plots. All JFETs were in the SOT-23 package and were provided by InterFET. For the SMPJ210 JFETs, some of the best devices were used in amplifiers before their I_dss values were measured. The S/N values are included in the plot for completeness with random values for I_dss.

JFET PACKAGING

All models tested above were in the SOT-23 (surface mount) package. The tested JFETs are also available in the more commonly used TO-92/TO-72 package. Both types of package can be accommodated on the amplifier board. To determine whether or not the packaging style affected the performance we compared the two packaging styles for the 5484 JFET: the SMP5484 (SOT-23) and 2N5484 (TO-92). Both of the tested JFETs had a drain current of around 3.9 mA. The benchtop S/N measurement for the SMP5484 was 18.62 dB, and for the 2N5484 it was 19.50 dB (see Table 2). The JFETs were then evaluated under vacuum in a test system that simulated a working CD-MS instrument (but without the electrospray source and differentially pumped chambers). A test signal, consisting of a 12 kHz square wave with a 50% duty cycle was applied to an antenna located in the grounded shield of the ELIT. The antenna is in line-of-sight to the detection cylinder and the resulting induced charge simulates an ion signal. The amplitude of the test signal was calibrated to yield an induced charge corresponding to 200 e. The signal was recorded for the equivalent of 5000 100 ms trapping events and then analyzed by the code used to process real CD-MS data. The results were then compiled to calculate the RMSD of the charge measurement. Measurements were performed for both the SMP5484 and 2N5484 JFETs (installed one after the other into the same amplifier board) and the results are listed in Table 2. The JFET in a TO-92 package has a lower S/N ratio when measured on the bench, but it has the higher charge RMSD when installed into an amplifier board.

Table 2.

Comparison of 5484 JFET in TO-92 and SOT-23 Packages

JFET Package	S/N Ratio Ambient Pressure Room Temperature	Charge RMSD Vacuum Room Temperature	Charge RMSD Vacuum Cryogenically Cooled
2N5484 (TO-92)	19.50 dB	0.94 e	0.85 e
SMP5484 (SOT-23)	18.62 dB	0.80 e	0.68 e

It is necessary to provide a way to manage the temperature of the JFETs when in vacuum. The metal connections on the SOT-23 package are much thicker and shorter than the thin wires on the TO-92 package. Thus, the heat transfer from the SOT-23 package to the circuit board is much better than for the TO-92. To manage the thermal load from the TO-92 we attached a metal heat sink to its case that was connected to the exterior of the vacuum chamber through a copper rod. For the SOT-23 package the thermal load was managed by attaching a heat sink to the circuit board. Measurements with a hand-held infrared thermometer indicated that the operating temperature of the TO-92 package was 29.3 °C versus 25.2 °C for the SOT-23. The higher operating temperature of the TO-92 package is at least partly responsible for its decreased performance in vacuum versus the SOT-23 package. However, that is not the whole story. Another factor that needs to be considered is that adding the heat sink to the TO-92 package increases the capacitive coupling to the JFET and this degrades the performance.

It is possible to cryogenically cool both JFET packages by using liquid nitrogen to cool the copper rod that transfers heat to the exterior. The temperature of the rod just before coupling to the circuit board (SOT-23 package) or the heat sink on the JFET (TO-92 package) was around −140 °C in both cases. When cryogenically cooled the performance difference persisted, with the SOT-23 package yielding a lower charge RMSD than the TO-92 package (see Table 2).

In summary, the SOT-23 package yields better performance than the TO-92 package because it is easier to manage the JFET thermal load with the SOT-23 package.

CSA INPUT CAPACITANCE

For a CSA, the capacitive load from the detector directly impacts the output signal.²⁵ To evaluate the sensitivity to the capacitive load we uncoupled the CSA from the CD-MS detection cylinder and coupled the amplifier input (the JFET gate) to a capacitor to ground (which provided the capacitive load) and to a capacitor chain coupled to a signal generator (to provide an input test signal). Measurements were performed using an SMPJ210 JFET for capacitive loads of 0.0, 1.2, 2.9, 4.1, and 6.9 pF. Signals were recorded for 5000 100 ms events which were analyzed using the code employed to process real CD-MS signals. The results were compiled and then the charge RMSD was calculated. The measurements were initially performed on the bench, but we found that much more reproducible results could be obtained by performing the measurements in a vacuum chamber, under vacuum. The results are shown in Figure 3, which shows a plot of charge RMSD against the capacitive load at the input of the CSA.

Figure 3.

Plot of charge RMSD against capacitive load at the input of the CSA. The points are the measured values, and the line is a guide.

Figure 4a shows a cross section through the Gen6 ELIT which is described in ref 17. The Gen6 ELIT is configured to provide a 50% duty cycle where oscillating ions spend half their time in the detection cylinder and half in the end-caps. With a 50% duty cycle, the even harmonics in the Fourier transform vanish and the odd harmonics have fixed ratios.¹⁷ This maximizes the precision of the charge determination and makes it easier to identify ion signals when multiple ions are trapped simultaneously. The gap between the detection electrode and the grounded electrode in the Gen6 ELIT is only 0.38 mm (Figure 4a). The capacitance of the detection cylinder to ground through the grounded electrode and grounded shield (as indicated by the red and purple arrows in Figure 4a) make the main contribution to the input capacitance of the CSA. Other contributions include any capacitance from the addition of a heat sink (discussed above) and capacitance due to the wire that connects the detection cylinder to the input JFET. To minimize capacitance, the wire should be as short and thin as possible, with the constraint that vibration will introduce noise if the wire is too flimsy. The capacitance from the detection cylinder to ground was calculated to be 3.1 pF using the Ansys Electronic Workbench. According to Figure 3, a charge RMSD of 0.62 e is expected with an input capacitance of 3.1 pF. A slightly lower value for the charge RMSD (0.60 e) was obtained when the SMPJ210 was connected directly to the detection cylinder.

Figure 4.

Cross section through (a) Gen6 ELIT and (b) Gen6E ELIT (from geometry reoptimization of Gen6 ELIT). The red arrows indicate capacitive coupling to the grounded electrode and the purple arrows indicate capacitive coupling to the grounded shield.

As discussed above, the m/z ratio is determined from the ion’s oscillation frequency (eq 1). The oscillation frequency also depends on the ion energy. The ELIT geometry can be optimized to virtually eliminate the kinetic energy dependence of the oscillation frequency over a narrow energy range.²⁸ We recently described a reoptimization of the Gen6 ELIT to achieve this goal. The geometry reoptimization required an increase in the spacing between the end-caps. The length of the detection cylinder also needed to be increased to maintain a 50% duty cycle. As part of the reoptimization, we took the opportunity to minimize the capacitance between the detection cylinder and the grounded electrode. The result is shown in Figure 4b which shows a cross section through the modified ELIT (now called Gen6E). With the geometry change, the grounded electrode was much thicker, and material could be removed from the grounded electrode to minimize its capacitance to the detection cylinder (see arrows in Figure 4b). This modification led to the nozzles extending from the grounded electrode toward the detection cylinder. With the changes evident in Figure 4b, the capacitance of the detector cylinder to ground was reduced to 2.4 pF (calculated with Ansys Electronic Workbench). According to the results in Figure 3, the charge RMSD should decrease to 0.58 e. A slightly lower value (0.56 e) was obtained when the SMPJ210 was coupled directly to the detection cylinder. While the decrease in the charge RMSD on going from the original Gen6 ELIT design to the optimized Gen6E design is relatively small (around 7%), it is evident from eq 2 that the reduction in the measurement time scales as the reduction in the noise squared, so the measurement time is reduced by around 13%.

CD-MS MEASUREMENTS

Ions are aerosolized by electrospray and enter the CD-MS instrument through a metal capillary. The supersonic jet from ambient gas flow through the capillary is broken up in the FUNPET interface (an ion funnel-ion carpet hybrid).²⁹ The ions then pass through an RF only hexapole where the ion energy is set and then travel through an RF-only segmented quadrupole. Ions that exit the quadrupole are focused into a dual hemispherical energy analyzer which transmits a band of kinetic energies that largely encompasses the kinetic energy distribution of the ion beam. The transmitted ions then enter the Gen6E ELIT where the end-cap potentials switch between transmission and trapping modes. Trapped ions oscillate between the two end-caps, inducing a charge on the detection cylinder as they pass through. The induced charge is detected by the CSA and the resulting signal is amplified, digitized, and transferred to a computer. The signals are analyzed using short-time Fourier transforms. The oscillation frequency is related to the m/z and the magnitude of the first harmonic is proportional to the charge.

All samples were electrosprayed with 250 nm emitters pulled from quartz capillaries (Sutter 2000). The samples employed in this study were SILuLite SigmaMAb (MilliporeSigma, MSQC4) and Hepatitis B Virus (HBV) virus-like particles (VLPs) assembled from truncated core protein (Cp149). The samples were buffer exchanged and desalted using Micro Bio-Spin columns (Biorad, 7326221) into neutral 100 mM and 20 mM ammonium acetate solutions, respectively.

EFFECT OF CRYOGENIC COOLING ON CHARGE RMSD

As noted above, thermal noise is reduced significantly when the input JFET is cryogenically cooled. In the case of the SMP5484 JFET, cryogenic cooling reduced the charge RMSD by around 20%. To determine the improvement that occurs with the SMPJ210 a series of experiments were performed with trapping times ranging from 100 to 1000 ms with the JFET temperature regulated to 25 °C and −140 °C. The temperature was monitored on the copper heat transfer rod (the cooling arm) close to the JFET. For each temperature, at least 5000 ions were measured for each trapping time. The optimum temperature for cryogenic cooling depends on the competition between noise reduction (which improves as the temperature is lowered) and JFET transconductance which diminishes at low temperature as the fraction of ionized dopants in the JFET decreases (due to the decrease in thermal energy available to promote ionization). While we did not do a detailed study, it appears that at −140 °C the signal-to-noise ratio has leveled-off and probably peaked, though we cannot rule out that small additional improvements may be realized at slightly lower temperatures than −140 °C. Charge histograms measured for MSQC4 mAb with the SMPJ210 JFET with a trapping time of 600 ms and nominal JFET temperatures of a) 25 °C and b) −140 °C are shown on the left of Figure 5. The corresponding mass distributions are shown on the right. The mass distributions show peaks for the MSQC4 monomer, dimer, trimer, and tetramer at around 150 kDa, 300 kDa, 450 kDa, and 600 kDa. Differences in the relative abundances of the multimers in Figure 5c and d are due to the measurements being made with different samples. Charge state envelopes for the monomer (at around 22 e), dimer (at around 32 e), and higher order multimers (at around 37 e) are evident in the charge distributions. The charge state resolution is much better for the data measured at −140 °C, consistent with a smaller charge RMSD expected at the lower temperature.

Figure 5.

CD-MS measurements for mAb MSQC4. Charge spectra measured with a trapping time of 600 ms with the JFET nominally at (a) 25 °C and (b) −140 °C are shown on the left. The corresponding mass distributions are shown on the right.

The charge RMSD is given by

$$\text{RMSD}=\sqrt{{\frac{\sum \left(q_{\text{meas}}-q_{\text{int}}\right)}{n}}^2}$$ (3)

where q_meas is the measured charge, q_int is the integer charge, and n is the number of individual ion measurements. There is however an issue with identifying the integer charge for each ion. If the charge RMSD is small, say <0.1 e, then it is reasonable to assume that q_int is given by rounding to the nearest integer charge. For example, with a charge RMSD of 0.1 e, an ion with a measured charge of 91.2 e almost always has an integer charge of 91 e. However, if the charge RMSD is large, say 0.5 e, then there is a significant probability that the actual integer charge of the ion is not the nearest integer charge to the measured charge. In this case, if an ion has a measured charge of 91.2 there is a substantial probability that it corresponds to an integer charge of 91 e, but there is a significant probability that it corresponds to an integer charge of 92 e. If eq 3 is implemented by simply rounding q_meas to the nearest integer charge to obtain q_int, the resulting RMSD will be artificially low. To avoid this problem, we isolated the ions from the abundant +21 charge state in the m/z spectrum for the MSQC4 sample and used just these ions to determine the charge RMSD (using 21 e for the integer charge). For the measurements shown in Figure 5 this strategy led to charge RMSD values of 0.19e at −140 °C and 0.23 e at 25 °C.

The charge RMSDs for the other trapping times are shown in Figure 6, where red points indicate measurements at 25 °C and blue points indicate results at −140 °C. The lines are least-squares fits of the function

$$\text{RMSD}=\frac{N}{t^{1/2}}$$ (4)

to the measured points. In eq 4, N is a constant and t is the trapping time. At both temperatures the points track the expected dependence. Cryogenic cooling of the SMPJ210 JFET lowered the charge RMSD by around 16%. As described above, a charge RMSD < 0.2 e is required to assign charge states with a low error rate. The green line in Figure 5 shows the target charge RMSD of 0.2 e. According to the results in Figure 5, the target of <0.2 e is achieved with a trapping time of 600–700 ms with a cryogenically cooled JFET, and 900–1000 ms with the JFET at room temperature.

Figure 6.

Plot of charge RMSD against trapping time for the SMPJ210 JFET at temperatures of 25 °C and −140 °C (measured on the copper cooling rod close to the JFET). The red points and line are for data measured at 25 °C and the blue points and line are for −140 °C. The points are the experimental measurements, and the lines are least-squares fits to the points assuming a t^−1/2 dependence (where t is the trapping time). The green line indicates a charge RMSD of 0.2 e. For a charge RMSD < 0.2 e, charge states can be assigned with a low error rate.

Figure 7 shows charge distributions measured for HBV T = 4 VLPs which have an expected mass of 4.04 MDa. CD-MS measurements were performed with nominal JFET temperatures of 25 °C and −140 °C, using trapping times of 1000 and 700 ms, respectively. In this study, charge states were not resolved in the m/z spectrum. This is a common problem with larger and more heterogeneous samples. Because charge states were not resolved in the m/z spectrum it is not possible to use the strategy described above for MSQC4 where the m/z charge is used to set the integer charge state used to determine the charge RMSD. As an alternative approach we used the charge distributions in Figure 7 to determine the charge RMSD and found values of 0.214 e for 25 °C/700 ms and 0.213 e for −140 °C/1000 ms. As noted above, this approach leads to an underestimate of the charge RMSD because ions with measured charges that are more than 0.5 e away from the true integer charge are misassigned to the nearest integer charge state. Figure 8 shows the error introduced by this approach for a Gaussian charge distribution. The orange points show the estimated charge RMSD (determined directly from the charge distribution) plotted against the true RMSD. The estimated charge RMSD tracks the true RMSD for charges up to around 0.2 e and then starts to deviate significantly and approach 0.289 asymptotically (0.289 is the value for a flat distribution, devoid of any charge state resolution). The orange points in Figure 8 show the deviation plotted against the true RMSD.

Figure 7.

Charge spectrum measured for HBV T = 4 VLPs with (a) a nominal JFET temperature of 25 °C with a trapping time of 1000 ms and (b) a nominal JFET temperature of −140 °C and a trapping time of 700 ms.

Figure 8.

Error introduced by determining the charge RMSD directly from the charge distribution. The blue points show the charge RMSD estimated directly from the charge distribution plotted against the true charge RMSD. The black line shows a linear relationship between estimated RMSD and the true RMSD. The orange points show the deviation between the estimated RMSD and the true RMSD. The dashed blue and orange lines are polynomial fits to the points.

If the estimated charge RMSD is less than around 0.25 e it can be corrected to account for the deviation described above, assuming that the charge distribution is Gaussian. For the values estimated above, 0.214 e with 25 °C/1000 ms and 0.213 e with −140 °C/700 ms, the corrected values are 0.223 and 0.221 e, respectively. These values can be used to determine the trapping time required to achieve a charge RMSD of <0.2 e: 1240 ms at 25 °C and 870 ms at −140 °C. A slightly longer trapping time is required to realize a charge RMSD of <0.2 e for HBV VLPs than for MSQC4 because the HBV VLP ions have larger m/z values and hence lower oscillation frequencies. The lower oscillation frequencies have slightly larger electrical noise due to an increased contribution from 1/f noise.

SUMMARY AND CONCLUSIONS

The dominant factor limiting the precision of the charge measurement in CD-MS is electrical noise. While the effect of electrical noise can be overcome by signal averaging, and the averaging time can be made long enough to achieve a charge RMSD of <0.2 e, this leads to long measurement times. The measurement time can be reduced by decreasing the electric noise. In this work we revisited JFET selection. The SMPJ210 outperforms the SMP5484 used previously. We also improved the strategy employed to hand pick the best performing JFETs. Modifications to the design of the ELIT allowed the input capacitance to be decreased, which also contributes to a lower noise level. Overall, the measurement time required to achieve a charge RMSD < 0.2 e was reduced by more than a factor of 2. For mAb MSQC4, a charge RMSD of <0.2 e can be achieved in 600–700 ms with a cryogenically cooled JFET. The measurement time increases to 900–1000 ms when the JFET is held at room temperature. When high precision charge measurements are not required, the measurements are often performed with a trapping time of 100 ms. The charge RMSD for mAb MSQC4 with a 100 ms trapping time is 0.51 e with cryogenic cooling to −140 °C and 0.60 e with the input JFET at 25 °C.