Liberation of Membrane Proteins from Detergent Micelles via Infrared Irradiation

We previously described an instrument capable of identifying the molecular composition of small molecule ligands and lipids bound to membrane protein complexes using multi-stage tandem mass spectrometry (MSⁿ) on an Orbitrap Eclipse.¹³ Here, we describe a further modification in which we interface a continuous wave IR CO₂ laser to the back of the QLIT manifold, allowing for 943 cm⁻¹ (10.6 μm) photons to be incident onto ions stored in the high-pressure cell of the QLIT (Figure 1A). In this arrangement, ionized proteomicelles, generated by nanoelectrospray ionization, are transferred to the high-pressure cell. The entire population of ions is then subjected to irradiation with IR photons prior to transfer to the Orbitrap for detection or m/z isolation and subsequent fragmentation. Utilizing the IR-based approach to remove detergent micelles offers control over two tunable parameters: laser output power (up to 60 Watts) and irradiation time (milliseconds to seconds). Therefore, we obtain greater control over the liberation of membrane proteins from proteomicelles, ensuring the preservation of native complexes while fully removing membrane mimetics from the complex. We initiated our study by benchmarking the utility of the IR laser for ejecting the intact trimeric ammonia channel AmtB from DDM micelles (Figure 1B). Low laser output power (3.0 W) at an irradiation time of 200 ms produces a poorly resolved charge state series consisting primarily of detergent-bound protein ions with m/z between 6000–8000. Increasing the laser output power to ≈3.3 W produced two well-resolved charge state series, the most abundant of which centered at the 20⁺ charge state near m/z≈6500, corresponding to the intact trimer with a mass of 126,794±1 Da, which matches well with the theoretical mass of 126,791 Da. Adjacent to the main charge state series are peaks corresponding to mass increments of ≈700–730 Da, which are signatures of phospholipid binding (Figure 1C). Up to five bound phospholipids can be clearly observed (Figure S1) and have been assigned previously as highly stabilizing phosphatidylglycerols.²¹ The other series, around the 11+ charge state at m/z≈3700, corresponds to AmtB monomers. The satellite features adjacent to the monomer peaks are assigned as integer multiples of DDM detergent molecules bound to the monomer. Amid this distribution of peaks, phospholipids can be observed still associated with the monomeric protein, albeit at a low abundance.

Preservation of non-covalent interactions between proteins and their effectors (e.g., drugs, lipids, or other proteins) is a defining feature of native MS. To assess the ability to maintain non-covalent interactions, we systematically surveyed combinations of IR laser output power and irradiation times (Figure S2) to determine optimal micelle-removal conditions. At low laser output power (≈3.6 W) and low activation times (10 ms), we detected primarily detergent-bound protein ions. A slight increase in laser output power to ≈4.5 W led to a well-resolved charge state distribution of trimers (m/z≈6500, z_avg≈20⁺). Under these conditions, we also observed several satellite peaks corresponding to lipid-bound states. Indeed, higher laser output power (≈5.4 W) caused the complex to dissociate into subunits. This phenomenon has been observed in previous studies which have attributed the emergence of monomeric subunits to concurrent gas-phase unfolding and dissociation.²² Expectedly, longer irradiation times reduce the threshold for the IR laser power required to efficiently remove micelles from trimers (3.6 W for 200 ms versus 5.4 W for 10 ms). Thus, we find that the IR parameters for ejecting membrane protein ions from detergent micelles are highly tunable, and careful optimization can lead to settings that both maximize detergent removal and preserve membrane protein-ligand interactions.

We next investigated the laser output power required to effect the removal of a range of native MS-compatible detergents from trimeric AmtB in vacuo.^{10b, 23} The detergents were compared based on their ease of removal in vacuo, which was defined previously as the ability to resolve protein peaks in the mass spectrum as a function of imparted activation energy,^5a and here, described as a function of laser output power. We first considered a glycol ether (C₈E₄), which typically requires low levels of collision energy, a synthetic, first-generation, dendritic oligoglycerol detergent (G1), and the non-ionic maltoside (DDM), known to require high levels of collision energy (Figure 2). At a fixed irradiation time of 200 ms and at low laser output power (2.1 W), we could readily discern charge state series for the intact trimer embedded in C₈E₄, centered around the 16+ charge state at m/z≈7925, albeit with high levels of detergent adducts. An increase in laser output power to 2.4 W generated a well-resolved charge state series. Similarly, with G1 detergent, the m/z peak series corresponding to AmtB trimers displayed adduct peaks when irradiated with a laser output power of 2.4 W that were absent when irradiated with 3.0 W. By contrast, for DDM, a poorly resolved series of charge states for the trimer were observed following irradiation with an output power of 3.0 W; 3.6 W led to a well resolved series centered at a 21+ charge state (m/z≈6038) with satellite peaks consistent with bound phospholipids. A peak series corresponding to monomers of AmtB was also observed at m/z<5000. In summary, IR activation is an efficient means of liberating membrane proteins from commonly used detergents and produces mass spectra that are comparable to collision-based micelle removal. Similar to collision-based micelle removal, different detergents will require optimization of the laser output power and irradiation time to yield comparable spectra.

Across the three different detergents tested, C₈E₄ required the least IR laser output power (2.4 W) to generate well-resolved charge states. G1 and DDM required a minimum output power of 3.0 and 3.6 W respectively, to obtain comparable spectra. These trends follow those previously observed for collision-based micelle removal.^5a All together, we find that laser output powers <3.6 W are ideal for liberating membrane proteins from detergent micelles while still preserving non-covalent interactions. This represents a relatively low level of activation energy as it remains well-below the energy inputs typically required to induce bond cleavage along the protein backbone by IRMPD (discussed below).²⁴ While there are many confounding factors which influence the stability of a membrane protein in a micelle, and therefore the ease of detergent removal in vacuo,^{5a, 6, 10a, 14b} the relatively abrupt transition between detergent adducted and liberated membrane protein ions, between 3.0 W and 3.3 W was unexpected. The laser energy that is deposited into the proteomicelle is potentially withdrawn via collisional cooling with the helium gas which fills the trap at a pressure of 6 mTorr. We surmise that, if the mechanism were solely dependent on bulk heating through translation of laser energy into vibrational modes, we would observe a gradual resolution of protein peaks in the mass spectrum with increasing laser output power. These considerations prompted us to consider additional factors which may make photon-based micelle removal more effective over traditional collision-based approaches.

We explored the hypothesis that IR photons are preferentially absorbed by chemical moieties in the detergent molecules/micelles. It is well-established that in CO₂-driven IRMPD, good absorption occurs for CH₃ rocking, O−H bending, as well as P−O, P−O−C, and P−O−P stretching.²⁵ As the detergents described here carry O−H moieties, we hypothesized the detergents could carry distinct vibrational signatures which are on-resonant with the IR laser used here. This would allow us to affect the preferential activation of detergent. To test this hypothesis, we recorded gas-phase IR spectra of protonated detergent ions using cryogenic IR action spectroscopy (Figure S3). Ionized detergents were captured in superfluid helium droplets, which constitute an IR-transparent, cryogenic spectroscopic matrix.²⁶ Upon absorption of multiple IR photons, the intact ions are released from the droplets and detected by time-of-flight mass spectrometry. The appearance of released ions is then plotted as a function of wave number to yield a vibrational spectrum. For C₈E₄, six sharp bands were observed in the vibrational spectra (Figure 3). Abundant bands near ≈1200, 1100, and ≈1000 cm⁻¹ are consistent with C−O single bond stretching, which are predominant in the chemical structure of C₈E₄.

The gas-phase IR spectra for G1 and DDM contain additional features not as apparent in C₈E₄; new peaks emerge near 940 cm⁻¹ that are proximal to the anticipated profile of the IR beam (943 cm⁻¹, ≈5.3 μm linewidth) (Figure 3). While the gas-phase spectra here illustrate the presence of absorption features in certain regions, we note that vibrational modes are strongly coupled and can be affected by anharmonic shifts ranging hundreds of wavenumbers.²⁷ For protonated detergent ions, hydrogen bonding networks and proton sharing within oxygen-rich detergents can alter the observed vibrational frequencies. To demonstrate this, we computed the harmonic vibrations of the headgroup of protonated C₈E₄ using density functional theory (DFT) calculations. The computed harmonic IR spectrum (Figure S4) reveals 21 bands between 800 and 1800 cm⁻¹, and most notably, a high intensity band at 945 cm⁻¹ was observed. Indeed, the harmonic at 945 cm⁻¹ results from the coupled contributions of O−H bending, C−C stretching, C−H bending, and C−O stretching (Figure S4 and Supplementary Movie S1).

We also considered this effect would be greater for detergent micelles which envelop the proteins; we hypothesized that new features in the IR spectra would also emerge upon self-assembly. We therefore sought to draw qualitative comparisons between the presence of oscillators in assembled vs monomeric detergent molecules. As this is technically challenging to achieve by gas-phase action spectroscopy, because of the effects described above, we opted to record solution phase IR spectra for C₈E₄, G1, and DDM (Figure 3 and Figure S5.). We found that the solution phase spectrum for C₈E₄ micelles contains a prominent band which overlaps with the laser profile. This observation suggests that the multitude of hydrogen bonds in the micelle result in a stronger oscillator that overlaps with the profile of the beam. By comparison, the equivalent band is less intense, or absent, in the solution phase IR spectra of G1 and DDM, the major and minor absorption bands are offset from the profile of the laser beam (Figure 3). The differences between the detergent micelles highlight the effect of hydrogen bonding networks on observed vibrational signatures. Adding to this, within the mass spectrometer, where there is pumping of energy during IR irradiation, further broadening of the vibrational modes occurs. Overall therefore we anticipate some spectral overlap of various oscillators with the IR laser in our experiments, than predicted from gas phase IR spectra.

Taken together, the oscillators that emerge upon detergent self-assembly generate vibrational signatures that can be targeted by IR photons with a wavenumber of ≈943 cm⁻¹. The presence of strong vibrational signatures on-resonant with the laser profile allow us to rationalize the low energy deposition required to liberate membrane proteins from each the detergents described above. While protein ions also contain IR-active moieties, they are poor absorbers of IR photons in vacuo.^{25, 28} Therefore, we find it likely that the detergent clusters that envelop the membrane proteins are the principal absorbers of the IR photons. It is also possible that vibrationally-excited protein ions transfer energy to the weakly bound detergents during irradiation as a means of “evaporative cooling”. But in any case, we find the low and short doses of 943 cm⁻¹ photons deposit sufficient vibrational energy to dissociate the micelles that surround the membrane protein ions, without disrupting the non-covalent interactions between protein subunits and bound lipids. Thus, we find this to be a particularly useful means for liberating membrane protein ions from detergent micelles in vacuo.

Infrared Multiphoton Dissociation of Membrane Proteins

The gentle removal of detergent micelles by absorption of IR photons results in liberated intact protein ions that are then amenable to tandem MS experiments. To generate sequence-informative ions of a high molecular weight protein, we isolated the 19+ charge state of trimeric AmtB at m/z 6674 and subjected the ions to short irradiation (5 ms) with high laser output power (≈9 W). The region between m/z 1750 and 4000 in the MS² spectrum is densely populated with highly charged fragment ions (Figure 4A and Figure S6). Interestingly, we do not observe ejected monomers in the MS² spectrum, suggesting that covalent fragmentation occurs faster than protein unfolding and subunit ejection. This effect has also been observed for a range of soluble and membrane protein complexes, where the direct fragmentation of complexes has been shown to reveal fragment ions which report on higher-order structure.^14d Many fragment ions could be assigned as b-type or y-type ions resulting from a single backbone cleavage and, as a result, 26 % sequence coverage was obtained (Figure S7). Some of the assigned ions correspond to repeat fragments, as in, the fragment ions originate from the same cleavage site but are detected as having multiple charge states. This is notable, as we often do not observe the same number of charge states for cleavages at adjacent sites, suggesting that there may be preferred sites of fragmentation (Figure S8).

To investigate these preferred cleavage sites, the fragment ion intensities were normalized by charge state²⁹ allowing us to evaluate the propensity for cleavage at preferred sites based on adjacent amino acid pairs (Figure S9). We observe high intensity sites of fragmentation resulting from traditional charge-remote X|P and D|X cleavages, namely T|P and I|P, which are known high propensity fragmentation sites for low-charge density precursors (i.e. proteins under native conditions).³⁰ The remaining amino acid pairs resulting in the most intense fragments occur at A|G, F|G, and V|G, suggesting that there is an increased preference for fragmentation N-terminal to glycine. Despite the identification of these fragmentation “hotspots”, a correlation plot of cleavage with respect to amino acid pairs demonstrates that correlating high intensity fragment ions to sites of cleavage cannot be predicted by amino acid content alone (Figure S10). In addition, there is little correlation with previously observed fragmentation of soluble proteins via charge remote pathways under native conditions.^{30, 31} This observation prompted us to explore the location of observed fragmentation relative to the regions of secondary structure within the protein sequence.

To explore this, we constructed a plot of the relative abundance of each fragment with respect to the point of dissociation along the amino acid sequence of the protein (Figure 4B). Fragment ions with multiple charge states were summed after normalizing their intensities by charge as described above. Overlaid on the bar plot is the architecture of secondary structures of AmtB, which allowed us to determine correlations between regions of fragmentation, secondary structure, and amino acid position. Most of the secondary structure of AmtB is α-helical (11 α-helices), which are connected by five ordered periplasm loops and five unstructured cytosolic loops. The N-terminus is periplasmic, and the C-terminus is in the cytosol. Focusing first on the regions of the protein which reside outside of the membrane, few associated fragment ions were observed. Basic residues in these regions, particularly arginine, are expected to sequester the already sparse mobile protons, leading to charge-remote dissociation pathways cleaving primarily C-terminal to acidic residues or N-terminal to prolines. Indeed, these fragments in the periplasmic and cytoplasmic regions arise from charge-remote fragmentation (e.g., at residues 160, 191, and 310), which are C-terminal to Asp (D|X) and Glu (E|X), as well as N-terminal to Pro (X|P), respectively. These observations are largely consistent with previous fragmentation studies of natively folded proteins.³¹

Turning our attention to the fragment ions that originate from bond cleavage in the membrane-spanning α-helices, we found a high propensity of successive cleavage of adjacent amino acids to enable the assignment of sequence tags. In addition, we found that certain helices fragmented with higher propensity than others. High intensity fragments originate from helices 4, 6 and 8, while we observe little to no fragmentation originating from helices 3, 5, 9, 10, and 11 (Figure 4B). This atypical fragmentation pattern, where some transmembrane helices are “skipped,” led us to consider further the sites of fragmentation as well as their potential origin. The highest intensity fragment in the transmembrane regions corresponds to cleavage at residue 273 (helix 8) which corresponds to cleavage N-terminal to proline (T|P). Expectedly, this site of cleavage is comprised of many repeat fragments, with four unique fragments of varying charge states identified (Figure S8). The fragments with even greater repetition (i.e. five unique charge states) occurred at residues 217 (A|G) and 220 (A|G) in helix 6. The remaining high intensity fragments originating from helix 6 occur at residues 210 (I|G), 213 (F|G). In addition, we observe complementary b-type and y-type fragments in helix 8, with the most intense y-type fragment originating from cleavage at position 270 (V|G). From these data, it is tempting to speculate that fragmentation occurring N-terminal to glycine must be a preferred cleavage site in native top-down MS of membrane proteins, however many other helices with a greater number of glycines (helix 9, 10, and 11, containing 6, 5, and 3 glycine residues, respectively) appear impervious to fragmentation by IRMPD. This observation implies that specific amino acid pairs are not good predictors of fragmentation propensity. Next, we mapped the relative abundance of the fragments onto the structure of the AmtB trimer. We observed that high intensity fragments typically do not originate from amino acids which face the lipid bilayer but instead are mostly in the core or at subunit interfaces (Figure 4C top). This was intriguing, as most native top-down MS studies have observed bond cleavage primarily at solvent exposed regions in quaternary assemblies.^14d

We attempted to rationalize why fragment ions would form in internal regions of the protein assembly by following changes to the AmtB assembly using an all-atom MD simulation at increasing temperatures of 300 K, 400 K and 500 K. After a 100 ns simulation at 300 K, we heated the protein to 400 K (for 100 ns), and then subjected the final structure to an additional 100 ns of high temperature (500 K) simulation. The structure remains relatively stable for the entire 300 ns simulation, as the protein exhibited limited (<10 nm) fluctuations, indicating the overall topology of the complex was maintained (Figure S10). Inspecting snapshots of secondary structure across the trajectory, we find helices 5, 9, 10 and 11, which appear impervious to fragmentation, all lose the majority of their helical structure at elevated temperatures (Figure 4C bottom and Figure S11). By contrast, the α-helical character of helices 4, 6, and 8 is maintained in the regions that are readily fragmented. It has been demonstrated previously that helices in vacuo are stabilized when a charge-carrying amino acid “caps” the helix,³² forming electrostatic interactions with the dipoles of the peptide bonds comprising the helix.³³ Indeed, at least one charge remote fragment is found originating from a nearby loop, adjacent to a fragmented helix. Thus, it would appear that successive cleavage of adjacent amino acids arises from charge-induced dissociation driven by mobile protons along regions where the α-helix is stabilized through this mechanism. However, there are still regions which retain localized structure but resist fragmentation; this can be reconciled given the extreme stabilization of α-helices in the gas phase,³⁴ as well as differences in van der Waals interactions and hydrogen bonding networks which may contribute to individual stability, even at elevated temperatures. Still, such a peculiar fragmentation pattern suggests that residual influence from the original protein structure is reflected in the fragment ions observed.

To explore the extent that fragment ions can be used to recover details of the antecedent structure, we extended our investigation to include two class A GPCRs, the beta-1-adrenergic receptor (β₁AR) and the adenosine A_2A receptor (A_2AR). All class A GPCRs share a common architecture,³⁵ and both receptors maintain near complete conservation of tertiary structure, but only share <35 % sequence homology. Comparison of their fragmentation pathways would therefore allow us to separate structural variation from amino acid sequence effects. β₁AR and A_2AR consist of long N-terminal extracellular domains followed by seven transmembrane helices separated by alternating cytoplasmic and extracellular loops. We first optimized laser output power and irradiation time to liberate these protein ions from ionized detergent micelles (Figures S12 and S13). The native mass spectra yield measured masses of both β₁AR (40976±0.5 Da) and A_2AR (46527±0.5 Da). These masses are lower than the anticipated sequence masses (40980 Da and 46535 Da) and are consistent with the formation of two and four disulfide bridges, respectively. The three most abundant charge states (16+, 17+ and 18+) of each GPCR were isolated and subjected to fragmentation via IRMPD using a laser output power of 7.8–8.4 W for 5 ms. Despite the lack of sequence homology (<35 %), similar sequence coverage for β₁AR (28 %) and A_2AR (19 %) was observed (Figures S14 and S15).

The normalized intensities of the fragments were then plotted against the assigned site of cleavage (Figure 5A and C). To first explore connections between amino acid pair and fragmentation propensity, the normalized intensities of the 119 unique b- and y-type fragments for β₁AR were used to generate a heat map of fragment intensity relative to cleavage at specific amino acid pairs. Again, we observed high propensity for fragmentation at A|G residue pairs followed by several preferential cleavages at L|X (Figure 5E). The two most abundant fragments in β₁AR originate from position 60 (A|G) and position 279 (L|P), with normalized intensities of 100 % and 50 %, respectively. For A_2AR, the intensities and positions of the 72 unique b- and y-type fragment ions were used to generate a similar heat map (Figure 5F). Apart from high propensity X|P bond cleavage, the two heatmaps are markedly different, indicating that there does not appear to be a preference for amino acid cleavage for these two receptors.

By contrast, we observe highly similar localization of fragmentation relative to key secondary structural domains. Considering first β₁AR, we observe highly intense fragments corresponding to successive cleavage along the peptide backbone within helices 1, 2, 6, and 7, while only a few low abundance fragments are observed from the soluble extracellular and cytoplasmic domains (Figure 5A). There is a marked absence of fragment ions originating from helices 3, 4, and 5, which can be attributed to the presence of two disulfide bonds. Turning to A_2AR, we observe similar fragmentation patterns as above with highly abundant, successive fragmentation in helices 1 and 6, as well as some lower abundance fragments localized to helix 7 (Figure 5C). The absence of fragments originating from the innermost helices (3 through 5) is again attributed to extensive disulfide bonding networks. When the sites of fragmentation are overlaid onto the structures of each GPCR, the similarities in secondary structure-specific fragmentation patterns are evident (Figure 5B and D).

To compare to traditional collision-based approaches, we subjected the same isolated charge states of both GPCRs to fragmentation by IRMPD versus higher-energy collisional dissociation (HCD) in back-to-back experiments (Figures S16 and S17). We note that, on this instrument, we found that the acceleration potential into the ion-routing multipole was too low to induce fragmentation of AmtB trimers by HCD. For β₁AR, fragment ions generated by IRMPD were of higher average molecular weight of 8866 Da compared to 5843 Da for HCD. Consequently, the fragment ions produced by IRMPD also retained a higher average charge (4.7+ vs 3.6+ z). Ultimately, fragmentation by IRMPD resulted in higher sequence coverage of β₁AR at 28 % compared to 17 % for HCD (Figure S16). Similar trends were observed for A_2AR, with 19 % coverage by IRMPD compared to 9 % for HCD (Figure S17). Although there are qualitative differences in the fragment ions generated by IRMPD versus HCD for the GPCRs, it should be noted that the sites at which fragment ions originate are similar across both modalities. The framework for peptide and protein dissociation via mobile protons is well established,³⁶ and IRMPD and HCD are both expected to fragment according to this mechanism.³⁷ Improvements gained by IRMPD can be attributed to (i) enhanced migration of protons, afforded by the slow pooling of energy in IRMPD, and (ii) the paucity of charged side chains in membrane proteins compared to soluble proteins which allows protons to migrate further. Together these two factors result in larger and more highly charged fragment ions.

Again, we investigated the influence of structure on the fragmentation for both GPCRs using molecular dynamics simulations. At the highest temperature of 500 K, both GPCRs exhibit changes associated with the loss of initial structure and protein collapse in vacuo (Figure S18). Interestingly, snapshots taken from the final structure at 500 K reveal regions in both GPCRs which retain α-helical character and correlate well with observed sites of cleavage (Figure S19). These observations further support our proposal that structural features persist and enable fragmentation by IRMPD. Together these results highlight that stabilized transmembrane helices, likely derived from nearby protonation sites, render them prone to a higher frequency of backbone cleavages.