Filter Diagonalization Method-Based Mass Spectrometry for Molecular

Feb 27, 2012 - Franklin E. Leach , Andriy Kharchenko , Gleb Vladimirov , Konstantin ... B. O'Connor , Eugene Nikolaev , Ron M.A. Heeren , I. Jonathan ...
0 downloads 0 Views 2MB Size
Download PDF
Article pubs.acs.org/ac

Filter Diagonalization Method-Based Mass Spectrometry for Molecular and Macromolecular Structure Analysis Anton N. Kozhinov and Yury O. Tsybin* Biomolecular Mass Spectrometry Laboratory, Ecole Polytechnique Fédérale de Lausanne, 1015 Lausanne, Switzerland S Supporting Information *

ABSTRACT: Molecular and macromolecular structure analysis by high resolution and accurate mass spectrometry (MS) is indispensable for a number of fundamental and applied research areas, including health and energy domains. Comprehensive structure analysis of molecules and macromolecules present in the extremely complex samples and performed under time-constrained experimental conditions demands a substantial increase in the acquisition speed of high resolution MS data. We demonstrate here that signal processing based on the filter diagonalization method (FDM) provides the required resolution for shorter experimental transient signals in ion cyclotron resonance (ICR) MS compared to the Fourier transform (FT) processing. We thus present the development of a FDM-based MS (FDM MS) and demonstrate its implementation in ICR MS. The considered FDM MS applications are in bottom-up and top-down proteomics, metabolomics, and petroleomics.

H

resolvent transform, or parameter estimators, e.g., linear prediction, the Prony method, and the filter diagonalization method (FDM).12−15 The FDM is a SR parameter estimator recently developed to be devoid of numerical limitations which other SR methods normally suffer from.15 Historically, the FDM was first introduced for quantum mechanics and further developed for nuclear magnetic resonance (NMR) spectroscopy.15−22 Currently, along with the standard FDM,18−20 there are several FDM-related methods, e.g., decimated signal diagonalization (DSD) and regularized resolvent transform (RRT).23,24 DSD-related methods were examined in FT-ICR MS earlier;25,26 however, they have an inferior performance compared to the FDM, primarily due to the fundamental limitations of the DSD-related methods. Recently, Aizikov and O’Connor applied the FDM to trace frequency modulation effects in experimental transient signals in FT-ICR MS and demonstrated super-resolution mass spectra obtained by FDM for simulated transients.27,28 Here, we introduce FDM-based mass spectrometry (FDM MS) to accelerate high resolution FT-ICR MS data acquisition and demonstrate its performance on the experimental data obtained for common classes of molecules and macromolecules.

igh performance mass spectrometry (MS) provides structural information of organic, inorganic, and biological molecules by accurately measuring the masses of the ions of interest and their fragments.1 Mass spectrometers with signal processing based on Fourier transform (FT) deliver superior analytical performance among the palette of MS instruments and are employed for the comprehensive structural analysis of molecules and macromolecules, both isolated and contained in the extremely complex mixtures.2 However, to provide high resolution and high mass accuracy mass spectra, FT signal processing employed in the modern MS platforms requires long data acquisition time, typically on the order of hundreds of milliseconds. Thus, the advancement of the FTMS-based application areas including bottom-up,3 middledown, and top-down proteomics,4 as well as metabolomics,5 petroleomics,6 and imaging,7 is severely hindered since these applications not only require high resolution but also set the very tight time constraints for the MS experiment. To overcome these limitations and advance molecular structure analysis, the high resolution MS data must be acquired faster. The two main FT-based mass spectrometers today, ion cyclotron resonance (ICR) and Orbitrap, measure the masses by employing the relation between m/z ratios and abundances of ions with the frequencies and amplitudes of components of the time-domain transient signal. 8−11 In general, the frequencies and amplitudes of the transient signal components can be obtained by either spectral or parameter estimation. The FT signal processing is arguably the most robust solution, albeit, being a spectral estimator, it has several drawbacks, including the resolution limitation. Super-resolution (SR) signal processing methods have been developed to overcome the FT limitations.12−14 The SR methods are either non-FT spectral estimators, e.g., the maximum entropy method and regularized © 2012 American Chemical Society

■

EXPERIMENTAL SECTION Sample Preparation. Cisplatin, substance P, and equine myoglobin were obtained from Sigma-Aldrich (Buchs, Switzerland). LC-MS grade water and acetonitrile were obtained from Fluka (Buchs, Switzerland). Formic acid was obtained from

Received: December 20, 2011 Accepted: February 27, 2012 Published: February 27, 2012 2850

dx.doi.org/10.1021/ac203391z | Anal. Chem. 2012, 84, 2850−2856

Analytical Chemistry

Article

Figure 1. The main steps in the suggested here FDM-based mass spectrometry (FDM MS) methodology. The FDM processing is the standard FDM algorithm described elsewhere.15,18−20 Other steps are realized in the custom-written PyFTMS package.

conventional magnitude-mode spectral representation. Finally, calibration procedure based on the relation between frequency and m/z values is performed to provide the mass spectrum.30 The FDM MS was performed as described in the main text. To find the monoisotopic mass of substance P from the experimental data, a standard deconvolution procedure was performed on the basis of the m/z of the monoisotopic peak and m/z spacing between the peaks in the mass spectra. To find the experimental monoisotopic mass of myoglobin, the isotopic ion distributions in the mass spectra were deconvoluted using the THRASH algorithm.31

Merck (Zug, Switzerland). Solution of hydrated cisplatin was prepared in water. Solutions of substance P and myoglobin were prepared in 1:1 (v/v) water/acetonitrile solvent mixtures containing 1% (v/v) of formic acid. Mass Spectrometry. The MS experiments were performed on a hybrid 10 T electrospray ionization linear ion trap Fourier transform ion cyclotron resonance (ESI LTQ FT-ICR) mass spectrometer (Thermo Scientific, Bremen, Germany). The instrumental parameters and operation were controlled by standard data acquisition software. Acquisition of transient signals was performed in MIDAS data format. Transient signal of a European crude oil was acquired on a custom 9.4 T FTICR MS (NHMFL, FL, USA).6 Data Analysis. Data analysis was performed using the inhouse developed Python-based software. The calculations were carried out on a standard desktop computer with a quad-core processor. The FTMS was performed following the standard algorithm.29 First, the transient signal is submitted to the procedures of apodization with Hann’s window and single zerofilling. Second, Fourier transformation is employed to convert the transient signal into the frequency domain spectrum with

■

RESULTS AND DISCUSSION

Methodology of Filter Diagonalization Method-Based Mass Spectrometry. The FDM-based MS methodology as employed here is described in Figure 1. Briefly, the harmonic inversion problems are solved for the selected frequency (massto-charge) window and chosen number of the basis functions following the standard FDM algorithm.15 The frequency window selection can be performed without (data-independent approach) or with (data-dependent approach) use of the 2851

dx.doi.org/10.1021/ac203391z | Anal. Chem. 2012, 84, 2850−2856

Analytical Chemistry

Article

preliminary FTMS step performed to find the location of the frequency regions with peaks of interest. In the latter case, since FT processing does not necessarily resolve the peaks of interest, FTMS is used only to reveal the corresponding frequency regions for further analysis. The transient signal is subjected to the FDM calculations for the selected frequency window. The FDM calculations are carried out with a user-defined number of basis functions. The number of basis functions is a constraint of the implemented numerical algorithm of FDM. For the appropriate results, the number of basis functions should exceed the actual number of ion peaks to be resolved in a given m/z (frequency) window, whereas the upper limit of the number of basis functions is restricted by the available computational resources. The number of basis functions can be selected on the basis of the particular sample being analyzed, MS experimental conditions, or stability of the calculations performed for a set of different numbers of basis functions. In the calculations performed here, the number of basis functions was between 15 and 240 per a frequency window, vide infra. In addition, the control calculations for the sets of different numbers of basis functions were carried out to verify the calculations stability and to confirm the obtained results. The FDM processing provides the solutions as FDM table that contains information on frequency, amplitude, phase, decay, and roughly estimated complex frequency error as a figure of merit for each solution. In the following step, the possible spurious solutions (false positives) are filtered out on the basis of the threshold values specified for user-defined acceptable peak parameters. If the applied threshold is not sufficiently strong, the spurious peaks can remain in the spectra and, vice versa, if the threshold is too strong, the real ion peaks can be lost. Therefore, method optimization should be performed for the particular experimental conditions and type of samples employed. Further, on the basis of the frequencies and amplitudes from the FDM table, the bar plot in the frequency scale is created and then reconstructed into the spectrum with Lorentzian-mode spectral representation with user-defined peak width. Finally, the bar plot and the spectrum in the frequency scale are subjected to the standard frequency-to-m/z conversion.30 Note, compared to the typical procedure in the FDM NMR,15 here, the spectral representation with a given peak width is performed for visual convenience only. The m/z ratios and abundances of interest are parameters to be found from the transient signal. Therefore, the reported m/z ratios and abundances are taken directly from the final bar plot in the m/z scale as the results of the considered parameter estimation problem. The methodology described in Figure 1 can be similarly applied to provide analysis of the broadband mass spectra, which would be considered as a number of adjacent windows. Application of Filter Diagonalization Method-Based Mass Spectrometry to Isolated Compounds. The performance comparison between the FTMS and the FDM MS as a function of data acquisition time is shown in Figure 2. The presented examples correspond to the MS application areas of metabolomics (small molecule analysis),5 bottom-up proteomics (peptide analysis),3 and top-down proteomics (protein analysis).4 Figure S1, Supporting Information, provides the theoretical isotopic distributions of the selected compounds. The corresponding ICR transient signals for all compounds analyzed are shown in Figure S2, Supporting Information. The FTMS and FDM MS accuracies of the monoisotopic masses and m/z ratios, obtained with an external

Figure 2. Filter diagonalization method (FDM) MS versus magnitude mode Fourier transform (FT)-based ICR MS for molecular structural analysis as a function of data acquisition time (transient length, T). (Top) analysis of a singly charged Pt-based compound; (Middle) analysis of an 11 amino acid long peptide, substance P, doubly charged; (Bottom) analysis of a 16.9 kDa protein, myoglobin, carrying 21 charges. The particular advantage of FDM MS application here is the potential ability to resolve the isotopic distributions of molecules and macromolecules in the experiments with (i) time constraints, e.g., in the online liquid chromatography-MS and (ii) finite lifetimes of transient signals, e.g., in a top-down proteomics.

calibration, are given in Tables S1−S3 (Supporting Information). In the structural analysis of a singly charged hydrated derivative of a typical metal-based anticancer therapeutic compound, cisplatin,32 FDM MS resolves the isotopic distribution even at an acquisition time of 0.7 ms, Figure 2 top. Here, the ion abundances and peak widths on the panel 1 (T = 100 ms) differ from those on the panels 2 (T = 2.2 ms) and 3 (T = 0.7 ms) since the isotopic fine structures are resolved only for T = 100 ms case (see Figure S3, Supporting 2852

dx.doi.org/10.1021/ac203391z | Anal. Chem. 2012, 84, 2850−2856

Analytical Chemistry

Article

(bottom) to a crude European oil sample. Importantly, FDM MS confirms the molecular ion peaks suggested by FTMS, including those with low abundances, Figure 3 insets. In addition, molecular ion peaks differing by only 1.1 mDa are resolved in the FDM MS mass spectrum. Thus, FDM MS confirms its applicability even in the case of high spectral complexity of analyzed transient signals. Analytical Characteristics of Filter Diagonalization Method-Based Mass Spectrometry. Figure 4 shows the

Information, for more details). For a given acquisition time, the FDM MS-provided m/z accuracies are normally superior to those obtained by FTMS for the same acquisition times, especially for the short ones. In the structural analysis of a doubly protonated 11 amino acid long peptide substance P (H-RPKPQQFFGLM-NH2), all FDM MS mass spectra shown, even the one for the 5 ms-long transient signal, reveal the isotopic ion distribution, in contrast to FTMS, Figure 2 middle. For all FDM MS mass spectra, the determination of the monoisotopic mass resulted in mass accuracies better than 1.3 ppm (external calibration). Finally, all FDM MS mass spectra of a protein shown in Figure 2 bottom, even the one for a transient signal containing only one isotopic beat (Figure S2, Supporting Information), clearly reveal the isotopic distribution of myoglobin charged with 21 protons. In contrast, FTMS requires at least 4 isotopic beats of the transient signal to yield baseline resolution in the magnitude mode (Figure S4, Supporting Information). The deconvolution of the FDM MS mass spectra of myoglobin provided the correct protein charge state of 21 and the monoisotopic masses with the accuracies of 1.6 ppm (external calibration), whereas the FT-based mass spectrum from 3 or more isotopic beats transient was required. Comparable values of mass accuracies in both methods indicate that the mass calibration error prevailed over the errors of the signal processing methods and the deconvolution procedure. Thus, for all cases considered in Figure 2, compared to FTMS, FDM MS presents superior performance along with acceptable mass accuracy. Application of Filter Diagonalization Method-Based Mass Spectrometry to a Complex Mixture of Compounds. The ultimate test for the analytical performance of a mass spectrometer and the associated method of data treatment involves the structural analysis of extremely complex mixtures of molecules as contained, for example, in crude oil.6 Figure 3 shows a comparative application of FTMS (top) and FDM MS

Figure 4. Comparison of filter diagonalization method (FDM) and fast Fourier transform (FFT)-based signal processing in ICR MS. The linear dependences for magnitude (mFFT, baseline resolution m/Δm0% and crossover regime m/Δm100%) and absorption (aFFT, crossover regime m/Δm100%) modes were calculated for a 10 T magnetic field (see Figure S6, Supporting Information). The resolution levels R0 = 100k and 12.5k represent acquisition times typically employed in FTMS-based applications. The FDM MS results are given with the panel numbers that correspond to the numbers given to the mass spectra in Figure 2.

dependence of the required acquisition time of the transient signals on the desired resolution for molecules in a wide range of molecular weights and different charge states. The required acquisition time is determined by the frequency spacing between the ion peaks to be resolved (see Figure S5, Supporting Information, for further details). In FTMS, the required acquisition time depends on the smallest frequency spacing between the ion peaks in a given m/z ratio range to be resolved and on the desired degree of resolution (see Figure S6, Supporting Information, for further details). Up to twice shorter acquisition times may be required to yield a comparable resolution in FTMS when the absorption mode of spectral representation is used instead of the typically employed magnitude mode. 33−36 Furthermore, detection of ICR frequency harmonics or multiples may reduce the required acquisition times in FTMS by several times.37 Importantly, and in contrast to the FTMS, in the FDM MS, the minimum transient duration required to resolve the ion peaks of interest is determined by the average peak spacing in the considered m/z (frequency) narrow window (see Figure S5, Supporting Information, for further details). This time is shorter or even significantly shorter than the one required in FTMS, depending on the particular peak distribution within the considered m/z (frequency) spectrum window.17,18,38 The obtained FDM MS

Figure 3. Filter diagonalization method (FDM) MS versus magnitude mode Fourier transform (FT)ICR MS for structural analysis of very complex mixtures of organic molecules, such as a European crude oil.6 Length of the transient signal employed was 2.86 s. The particular features of FDM MS here are the potential abilities to reveal low abundance ions and resolve closely spaced ion peaks that could not be separated by conventional FT-ICR MS at the same acquisition times. 2853

dx.doi.org/10.1021/ac203391z | Anal. Chem. 2012, 84, 2850−2856

Analytical Chemistry

Article

Table 1. Parameters of the Employed FDM MS Methodology: Frequency Windows, Numbers of Basis Functions, and the Corresponding FDM and FDM MS Calculation Times FDM parameters analyte cisplatin

substance P

myoglobin

crude oil

a

length of transient signal, ms 0.7 2.2 100 5 11 29 667 195 185 290 2 860

m/z window

frequency window, kHz

calculation time, s

number of basis functions

FDM

FDM MS