Anal. Chem. 2006, 78, 372-378
Selective Ion Filtering by Digital Thresholding: A Method To Unwind Complex ESI-Mass Spectra and Eliminate Signals from Low Molecular Weight Chemical Noise Steven A. Hofstadler,* Jared J. Drader, and Amy Schink
Ibis Division of Isis Pharmaceuticals, 1891 Rutherford Road, Carlsbad, California 92008
In this work, we present a simple method by which to preferentially detect either high molecular weight or low molecular weight ions generated by electrospray ionization. This approach, termed selective ion filtering by digital thresholding (SIFdT) is demonstrated on a commercial ESI-TOF instrument that employs a fast digitizer coupled to a microchannel plate detector. The digital representation of each individual scan is digitally filtered prior to spectral coaddition. As larger, more highly charged ions induce a more intense response than low molecular weight singly charged species, a digital threshold can be set that precludes the detection of singly charged species yet permits the efficient detection of larger, more highly charged species. In this work, we demonstrate the applicability of this approach to eliminate low molecular weight chemical noise in ESI-TOF spectra of oligonucleotide and protein ions, demonstrate improved dynamic range for analyte solutions containing high levels of low MW constituents, and show that spectra acquired at different digital thresholds can be subtracted to yield spectra of low molecular weight constituents with improved mass measurement accuracies. A notional scheme is presented in which an alternative digitization approach is employed using multiple differentially thresholded data streams to allow improved internal mass calibration and higher resolution ion partitioning.
In recent years, the performance of commercially available mass spectrometers has seen significant improvement due, in part, to the availability of improved core components including more stable power supplies, faster digitizers, and more sophisticated fabrication methods for ion optic elements. Particularly noteworthy are the newest generation ESI-TOF mass spectrometers which, from several vendors in a variety of configurations, are now routinely yielding mass measurement errors less than 10 ppm with resolution exceeding 10 000 (fwhm), performance previously attainable only on high-end sector or FTICR-based platforms. As such, the use of such benchtop instruments by the bioanalytical community continues to expand as these instruments are increas* To whom correspondence should be addressed. E-mail: shofstad@ isisph.com. Fax: (760) 603-2599.
372 Analytical Chemistry, Vol. 78, No. 2, January 15, 2006
ingly being made available to scientists and technicians with a broad range of analytical needs. Accordingly, a number of progressively more sophisticated automation schemes are emerging, many incorporating some form of liquid chromatography (LC) as an on-line sample purification step to support high-throughput QC or drug-screening activities. While there are a number of applications in which some form of LC is a requisite step that facilitates the analysis of very complex mixtures, it is also used frequently as a generic desalting/purification protocol to prepare relatively pure analyte fractions for MS analysis. Low MW chemical noise is often the limiting factor in overall MS performance as the presence of high levels of low molecular weight interferences (polymers, buffer constituents, etc.) can drastically limit the spectral dynamic range and adversely affect mass accuracy. While LC is often used to reduce the adverse affects of such backgrounds, constraints on sample throughput and issues associated with solvent usage/disposal must be considered as part of the laboratory work flow. Consequently, there is an increasing need for simple methods to reduce the chemical noise floor and render less than “pristine” samples amenable to mass spectrometric analysis. A few approaches to mitigate chemical noise in ESI-MS analyses using ion manipulation techniques have appeared in the literature including the combination of broadband collisional activation with resonant ejection in an ion trap1 and, more recently, high-field asymmetric waveform ion mobility spectrometry.2 Studies on a sector instrument by Loo and co-workers exploited the fact that highly charged ions could be efficiently detected at relatively low microchannel plate voltages while singly charged species went undetected at the same detection voltage.3,4 In this work, we exploit both the system architecture of a mass spectrometer that directly digitizes the analog signal from a multichannel plate and the observation that ions of different charge states but at the same m/z induce a significantly different detector response. As will be shown below, the detector response from a large multiply charged ion is significantly greater than that of a singly charged low molecular weight ion of the same m/z. By digitally filtering each scan such that only ion detection events (1) Ramsey, R. S.; Goeringer, D. E.; McLuckey, S. A. Anal. Chem. 1993, 65, 3521-3524. (2) Purves, R.; Guevremont, R. Anal. Chem. 1999, 71, 2346-2357. (3) Loo, J. A.; Pesch, R. Anal. Chem. 1994, 66, 3659-3663. (4) Loo, J. A.; Loo, R. R. O. J. Am. Soc. Mass Spectrom. 1995, 6, 1098-1104. 10.1021/ac051816r CCC: $33.50
© 2006 American Chemical Society Published on Web 12/13/2005
above a certain threshold are recorded, it is possible to render low MW singly charged species undetectable without significantly affecting the detection efficiency of the higher molecular weight (more highly charged) analyte ions. Here we present a novel approach, termed selective ion filtering by digital thresholding (SIFdT) as a method to eliminate signals from low molecular weight chemical noise. We further demonstrate how spectra acquired at different digital thresholds can be subtracted from one another to yield improved mass measurement accuracy. Expanding upon this scheme, we propose that, with minor modifications to the data acquisition scheme, it will be possible to detect specific MW “slices” from a complex mixture resulting in a new mode of multidimensional MS analysis. MATERIALS AND METHODS ESI-TOF Mass Spectrometry. A Bruker Daltonics (Billerica, MA) MicroTOF ESI time-of-flight (TOF) mass spectrometer was used in this work. Ions from the ESI source undergo orthogonal ion extraction and are focused in a reflectron prior to detection. Ions are formed in the standard MicroTOF ESI source, which is equipped with an off-axis sprayer and glass capillary. For operation in the negative ion mode, the atmospheric pressure end of the glass capillary is biased at 3500 V relative to the ESI needle during data acquisition. A countercurrent flow of dry N2 is employed to assist in the desolvation process. External ion accumulation is employed to improve ionization duty cycle during data acquisition5 and to enhance sensitivity in the m/z range of interest. In this work, each 75 µs scan is comprised of 75 000 data points (a 37.5 µs delay followed by a 37.5 µs digitization event at 2 GHz). For each spectrum, 660 000 scans are coadded. The digital threshold (digitizer noise suppression) was adjusted as described below. All aspects of data acquisition were controlled by the Bruker MicroTOF software package 1.0 running on a 2.4-GHz dual-processor Intel Xeon computer. Postprocessing of data was performed using the standard Bruker MicroTOF software and Bruker Xmass 7.0.4. PCR Conditions. All PCR reactions were assembled in 50-µL reaction volumes in a 96-well microtiter plate format using a Packard MPII liquid-handling robotic platform and M.J. Dyad thermocyclers (MJ research, Waltham, MA). The PCR reaction mix consists of 4 units of Amplitaq Gold, 1× buffer II (Applied Biosystems, Foster City, CA), 1.5 mM MgCl2, 0.4 M betaine, 800 µM dNTP mix, and 250 nM primer. Detailed thermocycling parameters are given elsewhere.6 PCR Product Purification. Following PCR amplification, amplicon mixtures were rigorously desalted using a protocol based on the weak anion-exchange method published elsewhere.7 Amplicons are bound to a weak anion-exchange resin directly from the PCR reaction buffer. Unconsumed dNTPs, salts, and other low MW species that might interfere with subsequent ESI-MS analysis are removed by rinsing the resin with solutions containing volatile salts and organic solvents. Elution of the final purified/ (5) Senko, M. W.; Hendrickson, C. L.; Emmett, M. R.; Shi, S. D. H.; Marshall, A. G. J. Am. Soc. Mass Spectrom. 1997, 8, 970-976. (6) Hofstadler, S. A.; Sampath, R.; Blyn, L. B.; Eshoo, M. W.; Hall, T. A.; Jiang, Y.; Drader, J. J.; Hannis, J. C.; Sannes-Lowery, K. A.; Cummins, L. L.; Libby, B.; Walcott, D. J.; Schink, A.; Massire, C.; Ranken, R.; White, N.; Samant, V.; McNeil, J. A.; Knize, D.; Robbins, D.; Rudnik, K.; Desai, A.; Moradi, E.; Ecker, D. J. Int. J. Mass Spectrom. 2005, 242, 23-41. (7) Jiang, Y.; Hofstadler, S. A. Anal. Biochem. 2003, 316, 50-57.
desalted PCR products is accomplished by rinsing the resin with an aliquot (typically 25 µL) of a high-pH buffer solution. The final electrospray buffer contains 35% MeOH and 25 mM piperidine/ imidizaole.8 Some samples were intentionally spiked with poly(propylene glycol) (PPG) obtained from Sigma-Aldrich (St. Louis, MO), which was used as received without additional purification. RESULTS AND DISCUSSION TOF Detection Principles. In time-of-flight mass spectrometry, ions are separated based on differences in their velocities as they traverse a flight tube.9 As ions strike the detector, their arrival times are recorded and subsequently converted to m/z based on the specific configuration of the spectrometer (length of flight path, accelerating voltage, geometry, etc.). It is generally accepted that, for singly charged species, detector response is inversely proportional to molecular weight (velocity), and, for example, in the case of MALDI, higher MW species induce a smaller detection signal than lower MW species.10 As will be described below, it is also the case that lower charge states (i.e., lower velocity species) induce a smaller signal than do the higher charge states (i.e., high-velocity species) under the same accelerating voltages.11,12 The reduced response of high MW “slow” ions can be partially ameliorated by the use of postacceleration methods in which ions are accelerated to very high kinetic energies immediately prior to detection. Studies on a sector instrument by Loo and Pesch exploited the fact that highly charged ions could be efficiently detected at relatively low microchannel plate voltages while singly charged species went undetected at the same detection voltage.3 Accordingly, while in the TOF mass analyzer, ions of the same m/z have the same velocity; ions of different molecular weight and charge do not have the same momentum or kinetic energy and thus do not induce the same signal on the detector. In an attempt to optimize detection efficiency of large oligonucleotide ions, and to better understand the relationship between ion arrival statistics and mass accuracy, we undertook a detailed systematic study of detector response as a function of MW, m/z, and charge state at the individual ion level. It became immediately apparent that ions of the same nominal m/z but different molecular weights induced significantly different detector responses. The heavier, more highly charged ions consistently produced detector responses several times greater than those of their singly charged counterparts at the same m/z. This phenomenon is readily illustrated by looking at spectral response as a function of the digital threshold employed to acquire mass spectra of species covering a range of molecular weights, vide infra. Under the acquisition conditions we routinely employ to characterize PCR products,6 individual scans are acquired and coadded at a rate of 13.3 kHz. Thus, for a typical 49.5-s acquisition, each spectrum is composed of 660 000 coadded individual scans. To reduce the shot/white noise in the coadded spectrum, the MicroTOF electronics allow one to set a digital filter threshold (digitizer noise suppression) such that white noise from the (8) Greig, M.; Griffey, R. H. Rapid Commun. Mass Spectrom. 1995, 9, 97102. (9) Guilhaus, M.; Mlynski, V.; Selby, D. Rapid Commun. Mass Spectrom. 1997, 11, 951-962. (10) Brown, R. S.; Gilfrich, N. L. Anal. Chim. Acta 1991, 248, 541-552. (11) Syage, J. A. J. Physics B: At., Mol. Opt. Phys. 1991, 24, L527-L532. (12) Syage, J. A. Phys. Rev. A 1992, 46, 5666-5679.
Analytical Chemistry, Vol. 78, No. 2, January 15, 2006
373
Figure 1. Schematic representation of the SIFdT concept. (a) A notional raw single scan in which a singly charged ion (ion1) strikes the detector at T1 and a large multiply charged ion (ion2) strikes the detector at time T2. Under “unfiltered” spectral conditions, both white noise from the digitizer and detector response from ion detection events are digitized and coadded, resulting in an electronic noise floor that grows linearly with the number of scans acquired resulting in the spectrum depicted in (b). The spectrum depicted in (c) results from setting the digital threshold above the white noise level but low enough to capture the detection of singly charged ions. Setting the digital threshold above the signal level of singly charged ions but low enough to capture multiply charged ions, the singly charged species are “invisible” resulting in the spectrum depicted in (d).
detector at the single or low-bit ADC count is zeroed out of each scan and only detector responses consistent with ion detection events are passed to the data system to be coadded. This concept is shown schematically in Figure 1. Figure 1a depicts the raw ADC output from a theoretical single scan in which a singly charged ion (ion1) strikes the detector at T1 and a large multiply charged ion (ion2) strikes the detector at time T2. During the time intervals in which neither ion1 nor ion2 is striking the detector, the ADC is picking up and digitizing detector noise generally corresponding to 1-5 bits. Because of the fast acquisition rate of the TOF and the finite ion capacity of the source, each scan is typically composed of relatively few ion detection events and for any given ion channel (i.e., discrete arrival time) it is very unlikely that an ion will be detected in each scan. Thus, coadding large numbers of unfiltered scans such as those depicted in Figure 1b would result in a noise floor that increases linearly with the number of scans and a mass spectrum in which the ultimate dynamic range would be limited by the relatively high electronic noise floor. To minimize the deleterious effects of coadding low-bit white noise, the MicroTOF electronics allow the user to set a cutoff voltage (digitizer noise suppression) that has the net effect of zeroing-out low-level signals that are attributed only to detector noise. As illustrated in Figure 1c, this approach (ideally) does not affect the ADC counts for signals consistent with a singly charged ion but digitally filters each scan on-board prior to coadding such that detector white noise is zeroed out and not coadded with the same efficiency as detector ion response. As illustrated in Figure 1d, this concept can be taken a step further by setting the digital filter threshold such that ADC counts derived from detector noise and singly charged ions striking the detector are zeroed out prior to coadding. Thus, with the digital threshold set at the level depicted in Figure 1d, a singly charged ion striking the detector is “invisible” in the postfiltered ADC output and the net result is a “high-pass” molecular weight (charge) filter in which low MW 374 Analytical Chemistry, Vol. 78, No. 2, January 15, 2006
Figure 2. Signal response as a function of charge and MW. a) The signal response for a series of near isovelocity analyte ions (at approximately m/z 1233) as a function of digital threshold setting is consistent with higher charges states producing a more intense detector response. The response isopleths in b) illustrate the detector response for ions of the same mass but different charge states (and velocities).
(charge) species are not detected but high MW (charge) species are still detected. Detector-ADC Response as a Function of MW (Charge). Unlike MALDI of large biomolecules, the multiple charging phenomenon inherent to the ESI process generally produces mass spectra in which the majority of the signals are in the same m/z range. Molecular ions from moderate to large biomolecules (1100 kDa) are typically detected in the 500-2000 m/z range, and it is thus not at all uncommon for complex mixtures to yield spectra in which peaks of many different masses are detected at the same nominal m/z. To characterize detector response as a function of MW (charge), solutions containing single analytes with molecular masses of 1.2, 3.7, 11.8, 21.5, and 43.2 kDa were analyzed at a range of digital voltage thresholds. For each series, a single charge state at or near m/z 1233 was used to gauge detector response. The resulting MW isopleths are plotted in Figure 2a. Importantly, at low cutoff voltages, signals from the singly charged PPG ions are extinct at significantly lower cutoff voltages than do the higher molecular weight (charge) species. For example, at a cutoff voltage of 9 mV, the signal of the PPG ions at m/z 1233 is attenuated to nondetectable levels while the 43-kDa PCR product at m/z 1233 is still detected at >80% of the initial response. There is a definite trend in cutoff voltages as a function of MW (charge state) suggesting that one can select a threshold to selectively detect (or not detect) species of interest.
Furthermore, it is clear that the difference in signal response is not a function of velocity as the isopleths in Figure 2a were acquired from species of approximately the same m/z (and therefore approximately the same velocity). The data in Figure 2a suggest that the detector response is not velocity dependent but instead either mass or charge dependent. As illustrated in Figure 2b, ions of the same mass but different charge states (albeit with different velocities) exhibit a chargedependent detector response. Thus, the differential detector response is not due merely to differences in the mass of the ions striking the detector, but rather is more likely related to the bulk charge striking the detector. Referring to the data plotted in Figure 2b, while the different charge states of the 38-mer oligonucleotide are traveling at different velocities (higher charge states traveling faster), there appears to be a charge-dependent detector response as the data in Figure 2a clearly demonstrate that the detector response is not velocity dependent. Chemical Noise Removal (High-Pass Filtering). A key challenge in the analysis of large biopolymers by ESI-MS is sample purification. Low molecular weight contaminants in biopolymer solutions can have deleterious effects on the quality of ESI-MS spectra and can significantly limit the dynamic range and accuracy of the measurement. In some cases, these low molecular weight “contaminants” are actually required additives as components of on-line separations. Such additives include ampholytes used in capillary isoelectric focusing,13,14 phosphate salts commonly used as components of buffers used in capillary zone electrophoresis,15,16 and solution modifiers used to promote micelle formation in micellar electrokinetic chromatography.14,17 Similarly, electrosprayincompatible additives such as glycerol and polymers (e.g., poly(ethelene glycol), poly(propylene glycol)) are often used to stabilize enzymes to be used in biochemical processes. These compounds often make their way through an entire biochemical process and end up in the mass spectrometer. A key example of the latter type of “contaminant” that posed a significant challenge to our research efforts was the presence of high levels of poly(ethelene glycol) and poly(propylene glycol) polymers in the Taq polymerase, which is used as an integral component of the polymerase chain reaction (PCR). While typically only 1-2 µL of Taq are used in each 50-µL PCR reaction, the relatively high concentration of polymer in the presence of the relatively low concentration of PCR products (typically 10-100 nM), coupled with the fact that such polymers are ionized with high efficiency, led to a significant chemical noise issue. While this issue was significantly addressed by the development of a high-throughput PCR purification protocol based on weak anion exchange7 and the use of Taq from vendors that used relatively low levels of polymer stabilizers, variations in enzyme lots and biochemical protocols led to variations in low MW chemical noise backgrounds. Figure 3a illustrates an example of an ESI-TOF spectrum of a 106-mer PCR product into which a contaminating amount (∼0.1 (13) Tang, Q.; Harrata, A. K.; Lee, C. S. Anal. Chem. 1995, 67, 3515-3519. (14) Yang, L.; Lee, C. S.; Hofstadler, S. A.; Pasa-Tolic, L.; Smith, R. D. Anal. Chem. 1998, 70, 3235-3241. (15) Olivares, J. A.; Nguyen, N. T.; Yonker, C. R.; Smith, R. D. Anal. Chem. 1987, 59, 1230-1232. (16) Thorne, J. M.; Goetzinger, W. K.; Chen, A. B.; Moorhouse, K. G.; Karger, B. L. J. Chromatogr., A 1996, 744, 155-165. (17) Fridstrom, A.; Markides, K. E.; Lee, M. L. Chromatographia 1995, 41, 295300.
Figure 3. Chemical “denoising” of oligonucleotides using SIFdT in the negative ionization mode illustrated in ESI-TOF spectrum of a 106-mer PCR product into which a contaminating amount of PPG (X) was spiked along with relatively high levels of singly charged calibrants (C). The spectrum in (a), which is inundated with peaks from low MW interferences, was acquired under normal acquisition conditions (digitizer noise suppression cutoff voltage of 3 mV) while the spectrum in (b) was acquired from the identical solution at a digitizer noise suppression cutoff voltage of 15 mV under otherwise identical conditions.
µM) of PPG was spiked along with relatively high levels of singly charged peptides (which serve as internal mass standards). The signal from the charge-state envelope of the multiply charged strands of the PCR amplicons is confounded by the presence of the intense signal arising from the low MW species. This spectrum was acquired using a “normal” digital threshold setting in which the white noise from the digitizer is filtered out but the threshold is set low enough to ensure that signals from singly charged ions are captured. This spectrum is exemplary of a common situation in which a large biopolymer is analyzed in the presence of a significant chemical noise background arising from low molecular weight contaminants. As shown below, such interferences can adversely affect the mass accuracy of the measurement and result in reduced spectral dynamic range. In contrast, the ESI-TOF spectrum in Figure 3b was acquired on the same spectrometer from the identical solution using the identical ESI source parameters and acquisitions conditions with the important exception that the spectrum in Figure 3b was acquired at a digitizer noise suppression cutoff voltage of 15 mV while the spectrum in 3a was acquired moments earlier at a digitizer noise suppression cutoff voltage of 3 mV. It is clear form these spectra, and the data presented in Figure 2, that the 15-mV cutoff setting precludes the detection of the singly charged species in the solution yet facilitates the detection of the larger, more highly charged PCR amplicons. It is evident from the spectra in Figure 3 and the cutoff profiles in Figure 2 that the intensity of the amplicon peaks are reduced by ∼30%; importantly, the peaks from the singly charged polymer and calibrant ions are not present in the spectrum acquired at the higher cutoff voltage and the spectrum in Figure 3b has significantly improved signal-tochemical noise characteristics. It is worthwhile to emphasize that no other instrument, solution, or data processing parameters were changed between collecting the spectra in Figure 3; the only difference was the digital threshold setting. These data suggest that in some high-throughput screening and QC applications a Analytical Chemistry, Vol. 78, No. 2, January 15, 2006
375
Figure 4. Chemical “denoising” of a protein analyte using SIFdT in the positive ionization mode illustrated in the ESI-TOF mass spectrum of 150 nM human carbonic anhydrase containing 0.001% Tween 80 (X). The spectrum in (a), which is of relatively low quality due to the preponderance of signals from the detergent clusters, was acquired under normal acquisition conditions (digitizer noise suppression cutoff voltage of 3 mV) while the spectrum in (b) was acquired at a digitizer noise suppression cutoff voltage of 13 mV under otherwise identical conditions.
less rigorous sample purification protocol might be employed and chemical noise can be removed via the digital filtering approach described above. Additionally, this approach serves to significantly reduce adverse effects from low MW chemical noise (i.e., species that are directly detected as low MW ions) but is not a panacea that obviates the need for all sample purification methods. In the case of the PCR amplicons in Figure 3, if the analyte solution were high in cationic salt ions (e.g., Na+ or K+), the adducted species would not be filtered out by SIFdT and the cation-adducted molecular peaks would still be detected. Importantly, this approach allows ESI-MS analysis of large biomolecules (or noncovalent complexes) from solutions that might otherwise contain too much chemical noise to produce interpretable spectra. Figure 4 illustrates the removal of interfering signals from a contaminant frequently observed in protein analyses: detergent clusters. As detergents are often used in biochemical processes to stabilize or solubilize proteins, they are often present at relatively high levels. This, combined with the fact that most detergents are ionized with very high efficiency, poses significant challenges for subsequent analysis by ESI-MS. While a number of detergent removal columns, kits, and tips are now commercially available to mitigate this phenomenon, each requires an additional biochemical processing step and an additional per-sample expense. The spectra in Figure 4 were acquired from a solution containing 150 nM carbonic anhydrase in the presence of 0.001% Tween 80 detergent. The spectrum in (a) was acquired using a “normal” digital threshold setting that facilitates the detection of both singly and multiply charged analytes while the spectrum in (b) was acquired at a digital threshold setting that precludes the detection of singly charged species. The signals from the multiply charged carbonic anhydrase ions are significantly obscured in (a) owing to the presence of multiple peaks corresponding to detergent clusters and sodium/ potassium adducted detergent clusters. The SIFdT spectrum in 376 Analytical Chemistry, Vol. 78, No. 2, January 15, 2006
Figure 5. Expanded region of the ESI-TOF spectra from Figure 3 illustrating the enhanced dynamic range of the SIFdT spectrum in (b) relative to the spectrum in (a) acquired by employing a normal digital threshold. Note in particular the improved signal-to-noise ratios of the higher charge states.
(b) is free from low molecular weight chemical noise and represents a spectrum of significantly higher quality. As in Figure 3 above, the only difference between the spectrum shown in (a) and (b) is the digital threshold setting. Dynamic Range Enhancement. The original concept behind the digital filtering approach employed on the MicroTOF platform was to reduce the electronic noise floor by digitally filtering out analog signals from the detector that were consistent with detector noise and inconsistent with ion detection events. While this approach significantly improves the electronic noise floor, it does nothing to attenuate the chemical noise floor and it is often the chemical noise floor that limits the dynamic range of bioanalytical mass spectrometry measurements. By reducing/eliminating the chemical noise floor in addition to reducing the electronic noise floor, significant improvements in dynamic range and spectral quality are attainable. This concept is demonstrated in Figures 5 and 6. Shown in Figure 5 is an expanded region of the ESI-TOF spectra from Figure 3 in which the relatively low abundance highcharge states of the PCR amplicon are detected. Note that the signals from the (M - 41H+),41- (M - 40H+),40- and (M 39H+)39- charge states are barely visible in the unfiltered spectrum (top) but clearly visible in the filtered spectrum (bottom). The effective signal-to-noise of the spectrum in Figure 5a is defined by the signal-to-chemical noise ratio, while the effective signalto-noise of the spectrum in Figure 5b is defined by the signal-toelectronic noise ratio. For example, for the (M - 41H+)41- charge state (∼800 m/z) of the amplicon, the signal to (chemical) noise in the spectrum acquired at the low-cutoff threshold is ∼2 while the signal to (electronic) noise of the spectrum acquired at the higher cutoff threshold is ∼12. Additionally, signals from charge states (M - 43H+)43- and (M - 42H+)42- are not readily discernible from the chemical noise in Figure 5a but clearly visible in Figure 5b. The improvement in effective dynamic range afforded by SIFdT is further illustrated in Figure 6 in which a solution containing ∼0.5 nM PCR product in the presence of 500 nM PPG was characterized at high and low threshold settings. At the normal threshold setting, the spectrum is dominated by highly abundant
Figure 6. Spectral dynamic range enhancement using SIFdT for detection of PCR products in the presence of an excess of low MW contaminant. The spectrum in (a) was acquired under normal threshold conditions (electronic noise discrimination only) while the spectrum in (b) was collected at the higher threshold to discriminate against low molecular weight ions. The inset illustrates that the low concentration PCR product (0.5 nM) is not detected in the presence of the highly abundant (500 nM) low molecular weight PEG contaminant. Under high threshold conditions employed in (b), the PCR product is readily observed and characterized. See text.
singly charged polymer ions and the very low level PCR products are not observed. As shown in the inset, the top spectrum is also inundated with other chemical noise components and the peakat-every-mass background precludes the detection of the low-level PCR products. When the SIFdT threshold is set such that signals from singly charged species are not detected, a distinct signature for the low-level amplicon is detected. This attribute has the potential to significantly improve the detection of low-concentration biomolecules in solution as it is frequently the presence of lowlevel, ubiquitous contaminants introduced from buffer impurities, plasticware, and sample handling that define the chemical noise floor of the mass spectra and limit the applicability of ESI-MS to complex biological systems. Implications for Accurate Mass Measurement. In addition to reducing the useful dynamic range of a mass spectrum, chemical noise and low molecular weight contaminants can have adverse affects on accurate mass measurements. As described above, ESI-MS spectra often have overlapping peaks that result from species of different molecular weights but the same m/z. This is particularly problematic for large biopolymer ions, which generally produce somewhat congested spectra in which multiple charge states are observed in the 500-2000 m/z range. Because low MW species are isotopically resolved and species above ∼10 kDa are generally not, it is quite common to see a low MW contaminant peak “step on” and distort an otherwise analytically useful analyte peak. An example of this is shown in Figure 7, in which the signal from the (M - 3H+)3- charge state of a 12-mer oligonucleotide is observed at the same m/z as the (M - 35H+)35charge state of a much larger 140-mer PCR product. In this case, the smaller oligonucleotide is intended to serve as an internal mass standard, but as is illustrated in Figure 7, the colocation of these signals is deleterious to both signals. First, at the 7-mV threshold, it is not immediately apparent that there are two species at m/z 1233 as peaks from the isotopically resolved 12-mer mask the presence of the larger unresolved amplicon peak. Additionally,
Figure 7. Overlapping signals from the (M - 3H+)3- charge state of a 12-mer oligonucleotide and the (M - 35H+)35- charge state of a 140-mer PCR product separated by SIFdT. (a) Spectrum acquired at a digital threshold of 7 mV. The presence of the unresolved 140mer distorts the peak shapes and centroids of the isotopically resolved 12-mer peaks resulting in average mass measurement error across the isotope distribution of 5.9 ppm (internally calibrated). (b) The spectrum, acquired at a digital threshold of 11 mV, does not allow detection of the isotopically resolved (M - 3H+)3- 12-mer. Subtraction of this spectrum from (a) results in the difference spectrum in (c). The average mass measurement error was reduced from 5.9 to 1.5 ppm (internally calibrated) following spectral subtraction. See text.
the presence of the unresolved amplicon peak distorts the peak shapes and centroids of the isotopically resolved 12-mer peaks such that the mass accuracy is compromisedsin this case resulting in an average mass measurement error across the isotope distribution of 5.9 ppm (internally calibrated). When the digital threshold is set to 11 mV, the contribution to the peak from the triply charged 12-mer is eliminated and the presence of a high molecular weight unresolved peak is apparent. Importantly, because the aggregate signal (i.e., 12- and 140-mer) is captured at the 7-mV digital threshold level, and the contribution to the signal from the 140-mer can be measured at a higher digital threshold level (11 mV in this example), the signal from the 12mer can be derived by subtracting the spectrum acquired at 11 mV from the spectrum acquired at 7 mV. The resulting spectrum exhibits a notably improved isotope distribution, and perhaps more importantly, the centroided peaks yield a reduced mass measurement error across the isotope distribution. In this example, the average mass measurement error was reduced from 5.9 to 1.5 ppm (internally calibrated) following the spectral subtraction. Spectral Subtraction, New Calibration Approaches, and Ion Partitioning. The data in Figures 3-6 illustrate that the digital thresholding method described above allows one to detect large multiply charged biomolecular ions in such a manner as to render low molecular weight species “invisible” (based on digital thresholding) while the data presented in Figure 7 illustrate a method by which low molecular weight species can be analyzed in such a manner as to make large multiply charged biomolecular ions “invisible” (by digital thresholding and spectral subtraction). The results from the relatively simple subtraction described above lay the foundation for more sophisticated digital thresholding schemes in which multiple “slices” of a complex ion population can be analyzed simultaneously with the effective result being a multidimensional detection configuration in which ions are Analytical Chemistry, Vol. 78, No. 2, January 15, 2006
377
Figure 8. Schematic representation of (a) existing single channel threshold filtering configuration used in this work and (b) notional multichannel threshold configuration in which multiple differentially thresholded spectra are acquired concurrently from the same digitizer. This approach should allow digital thresholding of a spectrum acquired from a very complex mixture and evaluation of a range of molecular weights (charges) independent of other, potentially interfering, ion populations. This approach should allow perfect spectral subtraction and the use of low MW internal calibrants that do not overlap with analytical peaks from high MW species. See text.
simultaneously measured based on both their mass-to-charge ratios and their charge states. In this work, all high-threshold/low-threshold comparisons were made by multiple measurements of the same analyte solution acquired under identical instrument conditions except the digital threshold was varied. This was done out of necessity because, as illustrated in Figure 8a, the basic system architecture of the Bruker MicroTOF consists of a single data stream from the detector to the digitizer for which a single threshold level is applied to the data stream prior to coadding of scans. While the signal is very steady over the course of the measurements, it is virtually impossible to generate spectra with identical peak intensities and spectral characteristics, thus limiting the applicability of the spectral subtraction approach as presented here. Furthermore, as sample throughput is a key driver in many laboratories, requiring each sample to be analyzed two (or more) times at different digital thresholds is generally not be feasible. We propose an alternative digitization scheme illustrated in Figure 8b in which output from the ADC is split to multiple parallel data streams, each of which is subjected to a different digital threshold. By subtracting spectra acquired at different digital thresholds, one could obtain a mass spectrum for any “slice” of the ion population. This would allow one to perform digital thresholding on a very complex mass spectrum and evaluate a range of molecular weights (charges) independent of other, potentially interfering, ion populations. Furthermore, having multiple variably thresholded mass spectra derived from the identical digitization event would guarantee perfect subtraction of spectral features and would eliminate potential artifacts that may arise from spectral drift over the course of acquiring multiple spectra. Importantly, this also means that one could introduce low molecular weight internal mass standards to very accurately calibrate the entire m/z axis (e.g., the PPG series in Figure 3a) but derive accurate mass
378
Analytical Chemistry, Vol. 78, No. 2, January 15, 2006
measurements of biomolecular analytes from peaks that are never “stepped on” by low molecular weight species (e.g., the digitally thresholded spectrum in Figure 3b). It is clear from the cutoff profiles presented in Figure 2 that at most cutoff voltages there is significant overlap in detection thresholds among the molecular weight range of the analytes used in these studies. Furthermore, the data presented in this preliminary work illustrate that the detector, digital filtering, and signal processing employed in SIFdT are not perfectly uniform and the technique is not as ideal as depicted schematically in Figure 1. However, just as TOF performance has drastically improved in recent years due primarily to the availability of more stable power supplies, faster digitizers, and more precise ion optics, it is likely that continued improvements in detectors, ion optics, and digital signal processing will enhance the functionality of the SIFdT methodology that will both enhance the resolution of the digital thresholding and enable novel SIFdT-related methods that will further enhance the functionality of ESI-MS instrumentation. Future work will focus on experimental verification of the parallel data stream scheme illustrated in Figure 8 and on detailed study of the effect of SIFdT on unresolved isotope peak shapes and mass measurement accuracy. CONCLUSIONS In this work, we have presented preliminary studies in which we exploit the observation that ions of the same velocity but different charge induce a significantly different detector response which, when digitally filtered on a scan by scan basis prior to coadding, allows preferential filtering of low molecular weight ions. Using this SIFdT approach, we were able to detect 33-kDa PCR amplicons in the presence of very high levels of low molecular weight contaminants (in the negative ionization mode) and multiply charged carbonic anhydrase ions in the presence of a high concentration of interfering detergent (in the positive ionization mode). The SIFdT approach yielded overall improved spectral quality for the large biopolymer ions and extended the dynamic range of the mass spectrum facilitating the detection of very low analyte levels in the presence of a vast excess of low molecular weight interences. It was also demonstrated that multiple spectra acquired at different digital thresholds could be subtracted from one another yielding spectra of low molecular constituents with improved peak shapes and improved mass measurement errors. We postulate that, in addition to providing a practical measure to ameliorate the adverse affects of low MW chemical noise by operating in a “high-pass” thresholding mode, there is significant potential to expand this approach using multiple differentially thresholded data streams that will enable a new multidimensional detection scheme for MS.
Received for review November 16, 2005. AC051816R
October
10,
2005.
Accepted
