Article pubs.acs.org/ac
Ultrahigh-Resolution Differential Ion Mobility Separations of Conformers for Proteins above 10 kDa: Onset of Dipole Alignment? Alexandre A. Shvartsburg*,† Biological Sciences Division, Pacific Northwest National Laboratory, P.O. Box 999, Richland, Washington 99352, United States S Supporting Information *
ABSTRACT: Biomacromolecules tend to assume numerous structures in solution or the gas phase. It has been possible to resolve disparate conformational families but not unique geometries within each, and drastic peak broadening has been the bane of protein analyses by chromatography, electrophoresis, and ion mobility spectrometry (IMS). The new differential or field asymmetric waveform IMS (FAIMS) approach using hydrogen-rich gases was recently found to separate conformers of a small protein ubiquitin with the same peak width and resolving power up to ∼400 as for peptides. The present work explores the reach of this approach for larger proteins, exemplified by cytochrome c and myoglobin. Resolution similar to that for ubiquitin was largely achieved with longer separations, while the onset of peak broadening and coalescence with shorter separations suggests the limitation of the present technique to proteins under ∼20 kDa. This capability may enable one to distinguish whole proteins with differing residue sequences or localizations of post-translational modifications. Small features at negative compensation voltages that markedly grow from cytochrome c to myoglobin indicate the dipole alignment of rare conformers in accord with theory, further supporting the concept of pendular macroions in FAIMS. ost biomacromolecules including proteins are flexible and may assume numerous 3-D conformations, depending on the charge state (z), temperature, and environment (e.g., pH or ionic strength in solution).1−4 There also is substantial memory, with the structures determined by history as well as current conditions. For example, the geometries of desolvated gas-phase protein ions generated by electrospray ionization (ESI) are strongly influenced by the solvent chemistry and mostly track folding in solution.5,6 As the biological activity of macromolecules is controlled by structure, understanding their conformational propensities and transitions as a function of physical parameters is paramount to life sciences and has been the object of vigorous experimentation and modeling.7,8 In particular, the critical role of multiple in vivo protein morphologies in life processes (e.g., misfolding disorders) is increasingly accepted in biomedicine.9,10 Investigations of isolated ions with a given z and cross-section selected by mass spectrometry (MS) and ion mobility spectrometry (IMS) enable one to explore the geometries and especially the thermodynamics and kinetics of their interconversion with far finer detail than solution studies,3,11 while the absence of a matrix simplifies computational modeling and thus allows a tighter connection to theory. As analytical approaches, MS and IMS/MS are much more sensitive and rapid than condensedphase alternatives and can deal with (even minor) components of complex mixtures rather than necessarily pure substances. Methods for macromolecular structure characterization are challenged by conformational multiplicity, which broadens the observed features and therefore decreases the actual resolving power (R) and specificity. This effect, widening the peaks by an order of magnitude or more relative to those for smaller species including peptides, is omnipresent for analyses in both liquids
M
© 2014 American Chemical Society
(chromatography, electrophoresis, and NMR)12−18 and gases (IMS).3,19 This has made raising the instrumental R of drifttube IMS nearly moot for proteins, where conformational ensembles within the structural families have limited3,19 the measured R to ∼20 while that for smaller ions increased from ∼10 to ∼200. Lately, novel IMS designs have dramatically augmented the resolution for macroions. Overtone mobility spectrometry (OMS), the frequency-domain form of linear IMS,20,21 has raised R for unfolded proteins with high z up to ∼200. Differential or field asymmetric waveform IMS (FAIMS) with H2/N2 buffers22 containing ∼80% H2 has recently attained23 R up to ∼400. This technique24,25 exploits that the mobilities (K) of ions in gases depend on the field intensity (E) and extracts the difference between K values at higher and lower E utilizing a periodic asymmetric field and superposed fixed “compensation field” (EC). The measured K(E) derivative is correlated to ion size and thus mass (m) weaker than absolute K, rendering FAIMS more orthogonal to MS than IMS is.26,27 Hence, FAIMS generally distinguishes isomers better28 than linear IMS with equal R, and >10 protein conformers have been baseline-resolved for some charge states.23 The peak widths in those FAIMS spectra and their scaling with the charge state and separation duration (t) were consistent with a largely instrumental (not conformational) limitation, suggesting the resolution of individual protein geometries rather than ensembles.23 Received: June 29, 2014 Accepted: October 3, 2014 Published: October 23, 2014 10608
dx.doi.org/10.1021/ac502389a | Anal. Chem. 2014, 86, 10608−10615
Analytical Chemistry
Article
Figure 1. FAIMS spectra for Cyt ions (z = 8−20, as indicated) measured using H2/N2 buffers with 0%, 50%, 70% (for z = 10), and 84% H2 v/v (color-coded on top) and Q = 2 L/min. Selected smaller features are magnified as marked, with the original spectrum deleted. Features found at EC < 0 in N2 and their derivatives in H2/N2 mixtures are plotted as dotted lines. Significant peaks are labeled, relating the data across gas compositions. The labels for features listed in Table S1, Supporting Information, are underlined.
kDa as discussed below) while the peak widths (w) scale as K−1/2 and thus slightly broaden.32,33 Hence, at some size, the conformational diversity must defeat even the impressive resolving power of the present FAIMS method,23 with the peaks for apparently specific geometries merging into wide features seen at lower R. The primary goal here is to gauge the extent of new “super-resolution” FAIMS capability (allowed by hydrogen-rich buffers) for heavier proteins. Cytochrome c (Cyt, 12.2 kDa) and myoglobin (Myo, 17.0 kDa) are ubiquitous models in MS and IMS research. Earlier investigations of Cyt and Myo ions generated by ESI using linear IMS1,3,34−40 and FAIMS3,41,42 have found the typical unfolding with an increasing charge state. Multiple conformers were encountered for many z, often exhibiting subpopulations in H/D exchange that reveal unresolved geometries.41 This paper demonstrates that FAIMS in hydrogen-rich buffers can (for most z) fully resolve Cyt and Myo conformers with peak widths similar to those for Ub. A few species have positive K(E) derivatives, ascribed to the alignment of protein dipoles in strong fields. This observation, the first for proteins under ∼30
With both OMS and FAIMS, this performance ostensibly reflects the elimination of less stable conformers that evolve (and thus change mobility or EC) during filtering, leaving only those in deep wells on the potential energy surface.23,29 This effect, termed “self-cleaning” of species structurally compromised by FAIMS,23,29 is parallel to the complete loss of excited metastable ions within selection quadrupoles, where mass filtering removes all fragments with the wrong m/z. Conversely, dispersion methods such as drift-tube or traveling-wave IMS30 do not destroy conformers that isomerize in the course of separation, instead averaging the cross section over all instantaneous geometries31 (which contributes to the peak broadening because the isomerization trajectories are never identical for all ions). Both OMS21 and FAIMS23 results above have been obtained for ubiquitin (Ub), a common small protein (8.6 kDa). The number of conformations exhibited under any conditions should go up for larger proteins. Meanwhile, the FAIMS peak capacity stays about constant or somewhat decreases as the separation space remains approximately the same (up to ∼30 10609
dx.doi.org/10.1021/ac502389a | Anal. Chem. 2014, 86, 10608−10615
Analytical Chemistry
Article
Figure 2. Same as Figure 1 for Myo ions (z = 10−27).
an ion funnel. The asymmetric waveform with peak amplitude (dispersion voltage, DV) of 5.4 kV and added scanned compensation voltage are produced by a high-definition generator.43 The H2 and N2 gases are filtered, metered by flow controllers (MKS Instruments), and mixed for delivery to the FAIMS stage. The flow rate was Q = 1−2 L/min, corresponding to t = 0.4−0.2 s. The ∼4 μM solutions of bovine
kDa, is in line with theory and supports the paradigm of pendular states for macromolecules in strong-field IMS.
■
EXPERIMENTAL METHODS This work has employed the FAIMS/MS system optimized for protein analyses.23 Briefly, a planar FAIMS device with a 1.88 mm gap is interfaced to a modified Thermo LTQ ion trap via 10610
dx.doi.org/10.1021/ac502389a | Anal. Chem. 2014, 86, 10608−10615
Analytical Chemistry
Article
Unlike for Ub, there are small but reproducible peaks at EC < 0 for Cyt with z = 14−16 (Figure 1) and Myo with z = 16−25 (Figure 2). This means that K increases at high E rather than decreases as is normal for m > ∼400 Da.42 This pattern was previously observed only for proteins above ∼30 kDa [bovine carbonic anhydrase II (CA, 29 kDa), equine liver alcohol dehydrogenase (40 kDa), bovine serum albumin (BSA, 66 kDa), and transferrin (78 kDa), as well as the BSA dimer (133 kDa)], with the fraction of total ion count at EC < 0 growing for larger masses. 42,44 That has been rationalized 42 as a manifestation of reversible alignment of permanent electric dipoles of macroions in a dynamic field, where pendular locking is tighter in the high-E half-cycle. Indeed, significant alignment in N2 at ambient temperature and pressure theoretically requires the dipole moment (p) of ∼400 Debye (D), and the p values for native geometries (pnat) trend up (with local oscillations) for larger proteins, crossing that threshold near 30 kDa. (For instance, pnat is 189 D for Ub, 287 D for Cyt, 225 D for Myo, 318 D for CA, and 1.1 kDa for BSA.)42 However, the protein ions processed by FAIMS are not native conformers but are extensively unfolded by Coulomb repulsion (particularly for higher z) and then by field heating in the gap.3,23 In simulations, the ensuing conformational ensembles cover wide segments of p that span pnat, especially at higher z where the deviations from native geometries are greatest.44 The p values for Ub ions derived from replica-exchange molecular dynamics range44 from 0 to ∼250 D, overall increasing at higher z. However, even the maximum p (for a tiny slice of conformations) are too low for dipole locking, in agreement with experiments.23 Analogous simulations for Cyt and Myo ions remain to be performed, but the range of p could be estimated through scaling the p distributions for Ub by the ratio of pnat values. Then, the highest p for Cyt and Myo ions would be ∼300−400 D, on the cusp of alignment. That is in line with Figures 1 and 2, showing only a few small peaks with moderately negative EC. Although Myo has a lower pnat than Cyt, this characterizes the native geometry and not directly the unfolded conformers with high z, where the p values tend to scale with the protein size and charge state by first-principles physics. It then fits that the features at EC < 0 are much more prevalent for Myo than Cyt in terms of both the number of pertinent charge states (10 versus 3) and abundances relative to the base peak for each (∼0.3−6% vs ∼0.1−0.3%). As anticipated, adding H2 up to 50%, 70%, and finally 85% uniformly transposes FAIMS spectra to higher EC, while narrowing the peaks and thus enhancing the resolving power and resolution (Figures 1 and 2). At lower z (∼8−11 for Cyt and ∼10−15 for Myo), the stronger field heating at higher H2 concentrations induces unfolding and high-E C isomers diminish. For Cyt with z ∼ 8−9 and Myo with z ∼ 10−13, broad features persist up to the maximum H2 fraction: apparently, the unfolding is incomplete and diversity of conformers overwhelms the available peak capacity. The signal for Cyt vanishes above ∼50% H2 for z = 8 and 9 and ∼70% H2 for z = 10, perhaps because of self-cleaning upon unfolding.23,29 For higher charge states, the signal loss upon H2 addition is moderate: almost nil between 0% and 70% H2 and ∼2−3 times between 70% and 84% H2. The pattern for Myo is similar. These losses are much less severe than those for peptides,22,43 reflecting slower diffusion of heavier protein ions.32 For higher z, the proteins are largely unfolded with N2 and adding H2 enhances specific conformers at lower EC (e.g., Cyt13+ d1, Cyt17+ d, Myo17+ e1) or higher EC (e.g., Cyt18+ g) or
cytochrome c (12 229 Da) or sperm whale myoglobin (16 950 Da) in 50:49:1 methanol/water/acetic acid were infused to the ESI emitter at ∼0.4 μL/min.
■
RESULTS AND DISCUSSION As is usual with denaturing solvents, a broad envelope of protonated ions was observed, here with z = 8−20 for Cyt and
Figure 3. Summary separation metrics for Cyt and Myo conformers depending on the charge state, in comparison with Ub:23 (a) number of resolved species at t = 0.2 s (solid circles) and 0.4 s (empty circles) with quadratic regressions (solid and dashed lines, respectively); (b) mean peak widths at t = 0.2 s (solid circles, for Cyt only) and 0.4 s (empty circles) with linear regressions (dashed lines) and z−1/2 dependences (solid lines). The regressions for Cyt and Myo scaled vertically by 2/3 are marked by the dotted (t = 0.2 s) and dash-dot (t = 0.4 s) lines.
10−27 for Myo. The data at t = 0.2 s are presented first. There is substantial separation already with the N2 medium, with ∼5− 7 distinct (though some partly merged) features for Cyt with z = 8−18 (Figure 1) and ∼5−11 such features for Myo with z = 13−26 (Figure 2). This is markedly more than the 2−3 features for most Ub charge states under identical conditions,23 as expected for larger proteins affording a greater number and diversity of conformers. These spectra are also much more structured than those obtained using the commercial cylindrical FAIMS device,3 as evidenced by separation of the major features d and e for Cyt13+, splitting of the base peak (into d and e) and near-baseline separation of the secondary peak c for Cyt14+, and clear splitting of the dominant peak into three features (d, e, f) for Cyt15+. The improvements for Myo are even greater.42 These gains follow from the inherent resolution advantage of planar gap shape,32,33 where a homogeneous field allows for the equilibration of only one species at any EC value. 10611
dx.doi.org/10.1021/ac502389a | Anal. Chem. 2014, 86, 10608−10615
Analytical Chemistry
Article
Figure 4. Same as Figure 1 for z = 11−20, measured using H2/N2 with 81% H2 and Q = 1 L/min. In the insets, selected segments are compared to those in Figure 1 (shifted to compensate for a slightly differing gas composition).
affects the isomer populations little (e.g., for Cyt15+, Myo18+, or Myo19+). The w and R values for well-shaped Cyt peaks at 0%, 50%, and 84% H2 are listed in the Table S1, Supporting Information. (These were not extracted for Myo because of coalescence for some intermediate z even at 84% H2, Figure 2.) As for Ub, sharp peaks emerge for higher z. The gain of resolving power between 0% and 85% H2 is an order of magnitude, specifically 12 times on average for Cyt with z = 16−20 where the six well-shaped peaks at 0% H2 permit accurate comparisons: R increases from 17 to ∼150−220 for Cyt16+ (base peak e vs e1, e2″, e3), from 19 to ∼150−240 for Cyt17+ (base peak f vs f1, f3, f4, f5), from 18−25 to ∼170−270 for Cyt18+ (base peak b vs b2, b3 and g vs g2), from 25 to ∼220−230 for Cyt19+ (base peak c vs c2″, c2‴), and from 24 to ∼460 for Cyt20+ (sole peak a). The separation powers with 1:1 He/N2 and 1:1 H2/N2 buffers are similar; hence, the gain at 85% versus 50% H2 emulates the benefit of substituting H2 for He (where ∼50% is the maximum avoiding the electrical breakdown at the present DV). For the 29 features of Cyt with z = 11−20 well-defined at 50% H2 (Table S1, Supporting Information), that gain is 1.3 to 5 times and 3.0 times on average. As with smaller ions,22,23 the gains are split equally between the EC increase and peak narrowing: each contributes a mean improvement of 3.4 times between 0% and 85% H2 and 1.7 times between 50% and 85% H2. The ion signal decreases at higher H2 fractions as usual (mainly because ions diffuse faster in lighter gases) but is still substantial at 85% H2 where the dynamic range is ∼103 (e.g., for Cyt16+ and Myo20+). By 85%
H2, the number of distinct features for intermediate z increases to ∼10−15 for Cyt and ∼10−20 for Myo, close to that23 for Ub (Figure 3a). The features for both proteins coming at EC < 0 in N2 shift with others upon H2 addition and thus move to EC > 0 at significant H2 fractions. That is in concord with the theory: the increased field heating of ions at higher H2 fractions includes rotational heating that obstructs alignment.44 This effect is moderate but may well eliminate loosely pendular states at slightly negative EC. The increases of resolving power by roughly 10-fold between 0% and 85% H2 and 3-fold between 50% and 85% H2 mirror the findings for Ub and hint at a separation similarly governed by mostly instrumental factors rather than conformational multiplicity.23 This can be probed closer by examining the peak width statistics for Cyt (Figure 3b). The resolved features are, on average, somewhat wider than Ub, with mean w of 1.16 V/ cm for all 65 peaks across charge states (Table S1, Supporting Information) vs 1.02 V/cm for all 53 Ub peaks. The difference is much greater, ∼50%, at the same z in the overlap region (Figure 3b). The peaks should fundamentally broaden for larger proteins at equal z because w is proportional32,33 to ∼K−1/2 and, as the mobility for homologous ions scales as ∼m−2/3, to ∼m1/3. However, that would translate into only ∼13% broadening from Ub to Cyt (that is heavier by 43%), close to the value derived from mobilities measured for unfolded conformers1,45 that differ by ∼30%. Hence, most of the effect ought to be due to other reasons, such as greater 10612
dx.doi.org/10.1021/ac502389a | Anal. Chem. 2014, 86, 10608−10615
Analytical Chemistry
Article
Figure 5. Same as Figure 4 for Myo ions (z = 13−25).
features for Cyt across charge states are baseline-resolved (Figure 4). Some of these (e.g., d1, d2, and d3 for z = 11; c1″ and c2 for z = 14; e2′, e2″, and e2‴ for z = 16; d, e, f1, f2′, and f2″ for z = 18; and c2″ and c2‴ for z = 19) were resolved in Figure 1 cleanly but not baseline, while others were separated as a shoulder (e.g., e1 and e2 for z = 14; d3 and d4 for z = 15; and e1′ and e1″ for z = 16) or not at all (e.g., d2′ and d2″ for z = 13; e1′ and e1″ for z = 13; d1 and d2 for z = 14; and c2‴ and c2⁗ for z = 19). The total number of peaks for all z is nonetheless almost conserved, as some weak features seen in Figure 1 vanish (Table S1, Supporting Information). The resolution similarly improves for Myo (Figure 5), with many broader features (such as h1 for z = 16; h2 for z = 18; f4 for z = 19 and 20; e1, e1′, and e2 for z = 21; e and g for z = 22; and f1 for z = 24) splitting into multiple peaks. However, the resolution is still not baseline for intermediate charge states (z = 18−22): the technique is reaching its peak capacity limitations even with extended separation. The average resolving power gain for Cyt is 1.8 times, exceeding the 21/2
conformational multiplicity. Unlike for Ub, the mean peak width for Cyt scales with z materially stronger than the z−1/2 dependence for diffusional broadening (Figure 3b). That is consistent with a growing role of multiplicity that is attenuated at higher z by better protein annealing. Still, the mean w for z = 13−20 (where Cyt is essentially unfolded) is 1.10 V/cm, above the analogous w = 0.84 V/cm for Ub with z = 10−14 and 0.7− 0.8 V/cm for the narrowest peaks for multiply charged peptides in the same regime. Together, these comparisons indicate the conformational broadening emerging on top of a largely instrumentally limited resolution. As with other separations, the resolving power of planar FAIMS devices scales33,46 as t1/2 and thus can be raised by decreasing the gas flow. Ion diffusion inevitably reduces the signal for longer t, but slower diffusion for larger ions such as proteins mitigates the problem, allowing extended filtering even at the maximum H2 fraction. At the near-minimum Q = 1 L/ min (t = 0.4 s), the signal drops about 10-fold and the dynamic range is thus compressed to ∼100, but virtually all significant 10613
dx.doi.org/10.1021/ac502389a | Anal. Chem. 2014, 86, 10608−10615
Analytical Chemistry
Article
capability may serve to separate and assign protein isomers with differing post-translational modification (PTM) sites or order of residues on the backbone. Such separations are readily implemented for peptides employing the same device and surprisingly are as easily achieved in the middle-down (∼3−4 kDa) as in the bottom-up (
