Localization of Post-Translational Modifications in Peptide Mixtures by High-Definition Differential Ion Mobility Separations with Electron Transfer Dissociation

Abstract

Precise localization of post-translational modifications (PTMs) on proteins and peptides is an outstanding challenge in proteomics. While electron transfer dissociation (ETD) has dramatically advanced PTM analyses, mixtures of localization variants that commonly coexist in cells often require prior separation. Although differential or field asymmetric waveform ion mobility spectrometry (FAIMS) achieves broad variant resolution, the need for standards to identify the features has limited the utility of approach. Here we demonstrate full a priori characterization of variant mixtures by high-resolution FAIMS coupled to ETD and the procedures to systematically extract the FAIMS spectra for all variants from such data.

Graphical Abstract

Introduction

Many proteins incorporate post-translational modifications (PTMs) that govern crucial biological functions [1,2]. For example, the phosphorylation pattern of human τ protein is thought to control the aggregation of paired helical filaments in Alzheimer’s disease by influencing the cis/trans ratio of prolyl bonds adjacent to T²³¹ and S²³⁵ and thereby the protein conformation[3]. Phosphorylated sites also have distinct regulatory functions for non-amyloid proteins, such as ERK and CDK family members and adapter proteins like SRC and Fak[4]. The same PTMs often attach in different locations on the backbone, creating “localization variants” that coexist in vivo and have different and even opposite biological activities [5–8]. For instance, demethylation of H3 histone is associated with transcription activation when on the K4 residue but with heterochromatin and repression when on the closest lysine (K9) [6].

As the technologies for identifying and quantifying primary protein sequences mature, the proteomics frontier is moving to the characterization of PTMs and their roles in disease states [1,2,9]. That advance is hindered by the open-ended diversity of PTMs (with hundreds discovered to date) and sub-stoichiometric modification[9]. Many PTMs are labile, including perhaps the most biologically consequential - the phosphorylation considered here[2,5,7,8,10]. Ergodic methods such as collision-induced dissociation (CID) usually eject them from the peptide and/or shift them to another residue before severing the backbone, obliterating or falsifying the attachment site information [2,10]. Thus comprehensive tandem mass spectrometry (MS/MS) analysis of PTMs requires a direct fragmentation mechanism, such as the electron capture dissociation or electron transfer dissociation (ETD) that cut the peptide backbone at every residue without abstracting PTMs [2,5,11,12]. However, MS/MS methods are fundamentally unable to distinguish some variants in a mixture of three or more due to the absence of unique product masses [10]. This can in theory be overcome by subsequent fragmentations, but (unlike with CID in MSⁿ protocols) successive ETD application is impractical in view of (i) charge reduction in the 1^st step, which produces few or no multiply-charged ions that could be precursors for the 2^nd step, and (ii) low ETD yield into each fragment (normally ~0.1 – 1% because of limited ETD efficiency and non-specific dissociation along ~10 – 100 channels).

Hence peptide variants generally need to be separated prior to the MS/MS step. While proteolytic digests are routinely fractionated by liquid chromatography (LC) prior to MS/MS, it often fails to resolve the localization variants, especially those with nearby alternatively occupied sites that also produce the most similar fragmentation patterns [7,13].

A growing substitute to LC is ion mobility spectrometry (IMS), where ionized compounds are driven through gas by electric field and fractionated depending on the transport properties [14]. The original approach of linear (drift tube) IMS is based on the absolute ion mobility (K) at moderate fields and achieves a resolving power (R) up to ~150 [15]. With a respectable R ~ 80, this technique can partly separate the phosphopeptide and unmodified peptide domains and resolve some variants[16–18]. Unfortunately, those could not be assigned using ETD: the extended times needed for a suitable yield of reaction between the ions and ETD reagent (commonly ~100 – 300 ms) greatly exceed the temporal peak width of ion packets upon IMS separation (typically ~0.5 ms) and thus implementing ETD on the fly after IMS would obliterate the established resolution. Whereas the IMS resolving power can be increased somewhat to pull apart the variants better, that would counteract matching the IMS peak width and the ETD timescale.

A newer approach of differential or Field Asymmetric waveform IMS (FAIMS) leverages the nonlinear ion motion in fields of high intensity (E) to sort ions by the mean derivative of K(E) function over a certain range[19,20]. That quantity is elicited employing an asymmetric waveform of some amplitude (dispersion voltage, DV) set up across a gap between two electrodes, through which ions are carried by the gas flow. Ions injected into the gap are pushed to either electrode, but a given species can be balanced (and thus pass and be detected) by superposing a particular fixed compensation voltage (CV) on the waveform. Scanning CV, commonly expressed as the compensation field (E_C) to normalize for the gap width, reveals the spectrum of species present. As the K(E) derivative is correlated to ion mass substantially weaker than the mobility itself, FAIMS is much more orthogonal to MS than linear IMS [21–23]. The gain depends on the analyte nature and is about fourfold for peptides [24]. Hence FAIMS tends to resolve isomers finer than linear IMS with equal R metric.

The resolution in FAIMS strongly depends on the gap shape, maximizing in planar gaps where homogeneous electric field allows equilibrating only one species at a time [25]. It is also sensitive to the gas composition. The He/N₂ buffers generally work well because of higher ion mobilities in lighter gases and prominent non-Blanc effect in mixtures of molecules with disparate masses [20,26]. The optimized buffers and high-fidelity waveform generators have recently raised the attainable R for multiply-charged peptides up to ~500 [27]. That performance has allowed fully separating all variants tried so far for sequences up to ~3 kDa, including those with smaller PTMs such as acetylation and methylation [28,29]. Those experiments utilized an ion trap MS system without ETD capability, and the spectral annotation relied on pure standards. With real protein digests, synthesizing those for all potentially relevant variants is often prohibitive.

Rather, one wishes to identify the features disentangled by FAIMS via ETD. That has been shown and utilized in proteomic analyses, but the commercial cylindrical-gap FAIMS device provided only marginal variant separation [12, 30–32]. Here we describe a new platform integrating high-definition FAIMS with ETD and deploy it to resolve a mixture of PTM localization variants baseline with a priori assignment for all.

Experimental Methods

Our FAIMS/MS platform is an upgrade on the earlier instrument [27]. The custom planar FAIMS unit remains as reported, with the gap width of 1.88 mm and length of ~50 mm. The bisinusoidal waveform with the 2:1 harmonics ratio and DV = 4 kV is delivered by a novel high-definition generator (Heartland Mobility, Wichita, KS) protected from arc discharge to permit operation near the breakdown threshold without risking equipment damage. The electrospray (ESI) emitter is biased at ~3 kV above the curtain plate of FAIMS inlet.

That system is mounted on a Thermo LTQ XL ion trap with ETD capability, enabling ETD of species filtered by FAIMS. A major present challenge for high-resolution FAIMS/ETD analyses is low sensitivity due to poor ion utilization in both stages. We have ameliorated that problem by effective FAIMS/MS coupling via an electrodynamic ion funnel interface preceded by a slit aperture that maximizes the overlap with ion beams exiting the FAIMS gap [25].

For the ETD stage, the temperatures were 160 ^oC at the source, transfer line, and restrictor, and 110 ^oC at the reagent (fluoranthene) vial. The pressure of chemical ionization gas (He) was delivered to the mass spectrometer at 10 PSI. Under those conditions, the reagent ion signal was ~10⁶. During ETD, we scanned CV at the rate of 0.5 V/min to acquire the E_C spectra or kept it constant to obtain the fragmentation pattern for a given feature. Most ETD spectra were obtained with the 120 – 140 ms reaction time followed by supplemental collisional activation at the potential of 25 V, which has balanced the sensitivity with coverage. All fragments were singly charged.

The N₂ and He/N₂ buffers were formulated from UHP components by digital flow meters (MKS Instruments) controlled from a PC, purified by an Agilent filter (RMSHY-4), and delivered at the “standard” rate of 2 L/min. This pilot study targets the τ 226 – 240 segment (VAVVRT²³¹PPKS²³⁵PS²³⁷S²³⁸AK) monophosphorylated on T²³¹ (1), S²³⁵ (2), S²³⁷ (3), or S²³⁸ (4) with the monoisotopic mass of m = 1602.9 Da previously separated by FAIMS [28]. Their equimolar mixtures (~5 μM each) were dissolved in 50:49:1 water/methanol/acetic acid and infused to the emitter at ~0.5 uL/min. This peptide set was chosen for the inordinate difficulty for differentiation by MS/MS ensuing from the two prolines between T²³¹ and S²³⁵, one proline between S²³⁵ and S²³⁷, and adjacent S²³⁷ and S²³⁸ residues. As the C-terminal side of proline is protected from cleavage, the fragments of 1 and 2 differ by just two pairs, and those of 2 and 3 or 3 and 4 by one pair only.

Results and Discussion

These peptides turn into protonated ions with the charge states (z) of 2 (m/z = 802.5) and 3 (m/z = 535.3), but both FAIMS separation and ETD tend to work much better for z = 3 [28]. Possible c- and z- fragments for all four variants (with either z = 2 or 3) needed to extract the E_C spectra were cataloged using the Protein Prospector software and compared to identify the unique m/z values (Table 1) [33]. As usual, we start from the N₂ buffer and add He in steps to improve resolution. We first look at a binary mixture (1 + 2) without the problem of non-unique fragments, then employ FAIMS to unlock that problem for the quaternary mixture.

Table 1.

Theoretically expected ETD fragments of the τ 226 – 240 peptides monophosphorylated at T²³¹, S²³⁵, S²³⁷, or S²³⁸.

		C Fragments						Z Fragments
		T231	S235	S237	S238			T231	S235	S237	S238
V	1	117.10	117.10	117.10	117.10	V	15	---	---	---	---
A	2	188.14	188.14	188.14	188.14	A	14	1488.77	1488.77	1488.77	1488.77
V	3	287.21	287.21	287.21	287.21	V	13	1417.73	1417.73	1417.73	1417.73
V	4	386.28	386.28	386.28	386.28	V	12	1318.66	1318.66	1318.66	1318.66
R	5	542.38	542.38	542.38	542.38	R	11	1219.60	1219.60	1219.60	1219.60
T	6	---	---	---	---	T	10	1063.49	1063.49	1063.49	1063.49
P	7	---	---	---	---	P	9	---	---	---	---
P	8	917.50	837.53	837.53	837.53	P	8	---	---	---	---
K	9	1045.59	965.63	965.63	965.63	K	7	688.38	768.34	768.34	768.34
S	10	---	---	---	---	S	6	560.28	640.25	640.25	640.25
P	11	1229.68	1229.68	1149.71	1149.71	P	5	---	---	---	---
S	12	1316.71	1316.71	1316.71	1236.74	S	4	376.20	376.20	456.16	456.16
S	13	1403.74	1403.74	1403.74	1403.74	S	3	289.16	289.16	289.16	369.13
A	14	1474.78	1474.78	1474.78	1474.78	A	2	202.13	202.13	202.13	202.13
K	15	---	---	---	---	K	1	131.09	131.09	131.09	131.09

The E_C spectrum of (1 + 2) in N₂ has two major features, and the ETD pattern of 3+ precursor comprises all unique fragments (UF) for 1 (c₈ at m/z of 917.5, c₉ at 1045.6, z₆ at 560.3, and z₇ at 688.4) and 2 (c₈ at 837.5, c₉ at 965.6, z₆ at 640.2, and z₇ at 768.3), as well as numerous non-distinguishing products (Fig. 1a). All four UF for either variant have identical spectra, but those for 1 and 2 drastically differ (Fig. 1 b, c). This illustrates the self-consistency and resolution of our procedure to extract the spectra for peptide variants from those for UF. The spectrum for each precursor is reconstructed by aggregating those for all four fragments (Fig. 1d): 1 is readily separated from 2, but not vice versa because the higher-E_C peak (features 1 B - D) almost coincides with 2. The sum of two traces matches the measured total spectrum (that also includes all non-distinguishing fragments), further validating the protocol.

Fig. 1.

FAIMS/ETD data (with N₂ buffer) for the pT²³¹/pS²³⁵ mixture: (a) total ETD spectrum with c/z fragments labeled (unique fragments color-coded); (b, c) normalized spectra for all UF of pT²³¹ and pS²³⁵ with the m/z values as marked; (d) measured trace (line) with the scaled spectra for both species produced by summing the spectra for all UF of each (above), and the mixture spectrum reconstructed by adding the two (dotted line).

In He/N₂ buffers, the E_C spectra for four UF of either 1 or 2 remain close and their reconstructed traces also sum to the measured spectrum (Figs. S1, S2). Upon He addition, the 1D feature rapidly disappears while the minor 1B and 1C grow between the major peaks 1A and 2C (Fig. 2). The evolution of 2 is simpler: the front ledge (2A and 2B) vanishes. The variant separation becomes perfect by ~30% He.

Fig. 2.

Same as Fig. 1d with He/N₂ buffers.

With the robustness of approach thus confirmed, we progress to the mixture of all four variants. Besides all ETD products in Fig. 1, we see (Fig. 3a) the distinguishing fragments at m/z = 1149.7 (c₁₁), 369.1 (z4), and 456.2 (z5). The c12 fragment (m/z = 1236.7) was observed, but swamped by undetermined interference. Therefore, the only marker for 4 was the tiny z4 feature.

Fig. 3.

FAIMS/ETD data (with N₂ buffer) for the four-variant mixture: (a) total ETD spectrum with c/z fragments labeled; (b) normalized spectra for the precursors or their groups derived by summing the spectra for all distinguishing fragments; (c) measured trace (line) with the scaled spectra for individual species derived from above data, and the mixture spectrum obtained by adding them all (dotted line).

The lack of UF for 2 and 3 can be overcome by accounting for both presence and absence of specific products at particular FAIMS features and exploiting the experimental trace to properly scale the E_C spectra for various variants and their combinations elicited from ETD data. [As the yields of different c/z pairs are grossly unequal (Fig. 3a), the heights of raw E_C spectra correlate with the precursor abundances only weakly).

The E_C spectrum in N₂ contains three major peaks (Fig. 3). We first use UF to find the spectrum for 1 (Fig. 3b), which naturally copies that in Fig. 1. The set 2/3/4 has four UF (c₈, c₉, z₆, z₇) that sum to the spectrum overlapping with peaks B and C, but not A. Hence A must be solely due to 1, and we can scale the trace for 1 to fit that peak in the measured spectrum (Fig. 3c). The spectrum for 1/2 obtained from UF c₁₁ (1229.7) and z₄ (376.2) has the dominant peak coinciding with C but little intensity at B, and the opposite nearly holds for the spectrum for 3/4 derived from parallel UF at 1149.7 and 456.2 (Fig. 3b). That allows us to scale the spectrum for 3/4 to fit B, then subtract it from the measured trace to scale the spectrum for 1/2 to fit C (Fig. 3c). We finally subtract from that the spectrum for 1 to deduce the one for 2. The normalized spectrum for 4 is extracted from its sole UF (Fig. 3b). Whereas the lack of a pronounced feature in the measured trace at its peak position precludes confident scaling and thus subtraction from the spectrum for 3/4, clearly B is due to 3 rather than 4. The sum of spectra for 1, 2, and 3/4 reasonably tracks the measured trace (Fig. 3c). Overall, FAIMS has fractionated a mixture of four components into A (1), B (predominately 3 + 4), and C (mostly 1 + 2). This partial separation is nearly enough for full characterization using ETD, which can deal with binary mixtures as explained above.

Much higher resolution is available with He/N₂ buffers. By 30% He, the three peaks separate nearly baseline, with B featuring a shoulder B1 on the left and tail B3 on the right (Fig. 4a). The trace is interpreted via the outlined process (Fig. S4): the spectra for 1 and 2/3/4 indicate that A consists exclusively of 1. As the spectrum for 1/2 has a peak at C but those for 1 and 3/4 do not overlap with C, it must be solely due to 2. Thus the spectra for 1 and 1/2 can be accurately scaled, and 2 recovered by subtracting 1 from 1/2. The spectrum for 3/4 again peaks at B and is scaled to it (the spectrum for 1/2 hardly overlaps with B). The peak for 4 has moved to the low-E_C side of B, where it engenders B1. It can now be scaled by subtracting the spectrum for 1/2/3, obtained from UF c₁₂ (1316.7) and z3 (289.2) and scaled to fit B, from the trace in the appropriate region. The spectrum for 3 is derived by subtracting the scaled spectra for 2 and 4 from that for 2/3/4. Within the logic of above solutions, one can alter the details. For example, in the last step one can instead subtract the spectrum for 1/2 from 1/2/3. These variations do not affect the outcome. Again, the spectra for all four variants add up to the measured trace.

Fig. 4.

Same as Fig. 3c with He/N₂ buffers.

The separation further improves at 40% He (Fig. 4b). Here, the structures B1 and B3 split into well-defined peaks. Analysis along the above path (Fig. S5) tells that B1 comprises 4 (which continues transposing to lower E_C relative to 3) with some 1, while B3 is made from 1 and some 2. This conclusion agrees with the results for (1 + 2) mixture in Fig. 2.

The fraction of He in FAIMS is limited by electrical breakdown, presently to under ~65% (v/v). However, a common constraint for peptides is that some disappear largely or completely upon He addition, presumably upon “self-cleaning” where the species that changed the CV by more than the baseline peak width upon isomerization inside the gap are filtered out [34]. This can occur in any medium, but is more likely at greater He concentrations because of both stronger heating and higher resolution that decreases the magnitude of geometry shift needed for elimination. Accordingly, the best variant separations were often achieved at He fractions significantly below the breakdown threshold, such as ~40 – 60% [29]. The ion counts required for MS/MS and especially ETD exceed those for MS detection. This shifts the resolution/sensitivity balance toward the latter, and the optimum He fractions for FAIMS/ETD are lower than those for FAIMS/MS: here ~30 – 40%.

For maximum signal, all above mass spectra were collected in the “turbo mode” with low MS resolving power. That sufficed for the present exemplary mixture, but many realistic samples require higher MS resolution and mass accuracy. Those can be reached in the mode resembling multiple reaction monitoring (MRM), where the CV during ETD is switched between discrete spectral peaks selected from the initial MS-only scan. This permits extended data accumulation at a given E_C that compensates for lower signal (even at the highest He fractions) at least in the “normal” MS mode (Fig. 5). For one, that enables baseline resolution of the c₁₂ informative and z₁₂ non-informative fragments partly merged in the turbo mode.

Fig. 5.

ETD spectrum measured at the peak B2 of Fig. 4b in the normal mode.

These peptides are also somewhat separated by FAIMS for z = 2. However, ETD of the 2+ ions from quaternary mixture produced no fragments below m/z = 1220 despite the reaction time up to 350 ms (Fig. 6). This has limited the UF to c₁₁ and c₁₂ that would not elucidate the picture even had FAIMS resolved all the variants. Hence we did not apply FAIMS/ETD to 2+ ions. However, those may help for other peptides that exhibit more informative fragments (e.g., because of PTMs closer to either terminus that would turn large-mass ions into UF) and/or better separation in the 2+ state.

Fig. 6.

ETD spectrum of the four-variant mixture (selecting 2+ precursors)

Conclusions

We have integrated the highest-resolution FAIMS system with ETD (on a Thermo LTQ XL ion trap) and demonstrated the ability of new platform to fully characterize the mixtures of peptide localization variants not amenable to ETD alone due to the lack of unique fragments (UF). The key aspects of data interpretation are the (i) summation of counts for all UF of each variant, (ii) scaling of so reconstructed precursor spectra to the measured trace, and (iii) use of the variant groups that have UF, the information regarding both present and missing fragments, and signal subtraction to derive the spectra for each variant. While we have addressed four species, the approach should work for more complex samples given the facility of ETD to deal with binary mixtures and the flexibility of FAIMS separations controlled by changing the dispersion voltage and gas composition. Real bioanalyses would benefit from higher mass resolution and accuracy, and we are working to couple high-definition FAIMS to the Orbitrap mass spectrometer.