Elucidating protein-ligand interactions in cell lysates using high-throughput hydrogen deuterium exchange mass spectrometry with integrated protein thermal depletion

Abstract

Hydrogen deuterium exchange coupled with mass spectrometry (HDX-MS) is a powerful technique for the characterization of protein-ligand interactions. Currently, the need for breakthroughs in the application of HDX-MS analysis to protein-ligand interactions in highly complex biological samples such as cell lysates is rapidly growing. However, HDX-MS analysis in such systems suffers from extreme spectral complexity as a result of high sample complexity and limited LC separation power due to the traditional use of a short LC gradients. Here, we introduced protein thermal depletion (PTD) to reduce protein complexity in E. coli cell lysate to our subzero temperature long gradient UPLC-HDX-MS platform (PTD-HDX-MS) to facilitate high-throughput analysis of protein-ligand interactions in cell lysates. We spiked bovine carbonic anhydrase II (CaII) and its inhibitor acetazolamide (AZM) into E. coli cell lysate as a model system in our study. We demonstrated that PTD at 60 °C greatly reduces protein complexity in cell lysates while the AZM-targeted CaII complex still remains in solution due to improved thermal stability upon binding. Using both PTD to reduce sample complexity and subzero-temperature long gradient UPLC to boost LC separation power, we successfully elucidated the interaction site between AZM and CaII in E. coli cell lysate from the high-throughput HDX-MS analysis of thousands of deuterated peptides from hundreds of proteins. Our results highlight the great promise of the PTD-HDX-MS platform for the identification of ligand targets and characterization of protein-ligand interactions in highly complex biological samples such as cell lysates.

Graphical Abstract

Hydrogen deuterium exchange coupled with mass spectrometry (HDX-MS) is a powerful protein footprinting approach for the characterization of protein-ligand interactions¹. This technique is based on the dependence of the exchange rate of protein amide hydrogens with deuterium on solvent accessibility and hydrogen bonding when the protein is exposed to deuterium oxide ^2–3. Consequently, reduced solvent accessibility following ligand binding protects the amide hydrogens involved in the binding interface against exchange while the same amide hydrogens will undergo rapid exchange without ligand binding. In the typical HDX-MS workflow, deuterium labeling is initiated by the dilution of the protein into a deuterium oxide-prepared buffer, so that proteins maintain native state. The deuterium labeling reaction is then quenched by decreasing the pH to 2.5. After quenching, proteins are digested using an acid-stable protease and subjected to LC separation and MS detection^4–5. Over the recent decades, improvements in several aspects of HDX sample handling including protein reduction and digestion^6–9, separation^10–12, MS detection¹³, and HDX data analysis software^14–16 have enabled the HDX-MS analysis of a wide size range of proteins under native conditions where other traditional methods for the characterization of protein-ligand interaction may be of limited use¹⁷. Currently, the HDX-MS community seeks to investigate protein-ligand interactions in native-like biological environments such as cell lysates^{16, 18–19}.

Back exchange, or the loss of deuterium after quenching is the central confounding issue in conventional HDX-MS analysis²⁰. Back exchange can reduce sensitivity and make it difficult to distinguish HDX variation between protein and protein-ligand complexes which can bias the identification of sites of interest. Thus, a short LC gradient is typically implemented (e.g., 5–10 minutes) to reduce spectral complexity and minimize back exchange¹⁷. However, short LC gradients can hinder the application of HDX-MS analysis to highly complex protein matrix environments, such as cell lysates, as the deuterium labeled peptides in a mixture of all digested proteins will coelute together in a short LC gradient, resulting in extreme spectral complexity that compromises the HDX data analysis^{13, 19, 21}. Furthermore, as with data dependent acquisition-based shotgun proteomics, sequence coverage of the protein(s) of interest will be decreased in the presence of a large amount of co-eluted peptides from nontarget proteins^22–23. Immobilization ^24–25 or affinity capture²² of nontargeted proteins and the integration of size exclusion chromatography (SEC) ²¹ prior to LC-MS analysis have been developed for HDX-MS analysis to address high spectral complexity. However, reported strategies for protein enrichment, including immobilization and biotinylation-based affinity capture of proteins, may introduce steric effects resulting in changes to protein-ligand interactions. Short SEC separations may suffer from low-resolution for highly complex biological samples such as cell lysates. Moreover, these strategies are typically only applicable to purified protein systems with known target proteins and are unsuitable for HDX-MS analysis of unknown protein-ligand interactions in complex samples such as cell lysates. To improve separation power, ion mobility²⁶, UPLC-based separation^10–11, and subzero temperature-based long gradient separations^{18–19, 27} have been introduced into the HDX-MS workflow. Previously, our group optimized a long gradient (e.g., 90 minutes) subzero-temperature UPLC separation method using Escherichia coli (E. coli) cell lysate digest, which significantly boosts LC separation power to enable the high-throughput analysis of thousands of peptides with low back exchange in a single HDX-MS analysis for complex biological samples ¹⁹. Yet, despite great progress, HDX-MS analysis is still not well-adapted for the characterization of protein-ligand interactions in native-like biological environments such as cell lysates.

Shifts in protein thermal stability upon ligand binding have been widely used to study protein-ligand interactions^28–29. The temperature at which a protein denatures (melting point, T_m) depends on the thermal stability of the protein. Upon exposure to heat at the melting point (T_m) or higher temperatures, proteins will denature and precipitate; precipitated proteins can be removed from the sample via filtration or centrifugation. However, it is possible that target proteins in the cell lysate will become more stable upon ligand binding and resist thermal denaturation at temperatures which more unstable proteins aggregate. It is reported that protein T_m is widely distributed between 45 °C to 80 °C for the E. coli proteome and 40 °C to 65 °C for the human proteome (Homo sapiens).³⁰ Therefore, protein thermal depletion (PTD) by heating ligand-treated cell lysate at temperatures lower than the target protein T_m allows the removal of nontargeted proteins with lower T_m, providing a potential strategy to reduce sample complexity for high-throughput HDX-MS analysis in cell lysates. Here, we aim to integrate PTD with our subzero-temperature UPLC-HDX-MS platform (PTD-HDX-MS) for the high-throughput analysis of protein-ligand interactions in cell lysate.

As a proof-of-concept for the proposed PTD-HDX-MS platform, E. coli cell lysate spiked with bovine carbonic anhydrase II (CaII) and its ligand acetazolamide (AZM) was used as the pairing model in our study. To determine the appropriate temperature for PTD, we heated both −AZM and +AZM aliquots to temperatures from 45 °C to 95 °C in increments of 5 °C. Our preliminary results suggested CaII remained in solution at 60 °C and denatured at 65 °C in the −AZM set (Supplementary Figure 1A) while CaII remained in solution at 65 °C and denatured at 70 °C in the +AZM samples (Supplementary Figure 1B). This T_m shift indicates that CaII was stabilized upon AZM binding. Our label-free, two-temperature point bottom-up TPP results also verified that AZM targeted CaII in E. coli cell lysate (Supplementary Figure 2). We then performed triplicate PTD at 60 °C, 65 °C, and 70 °C. Subsequent SDS-PAGE analysis (Figure 1A) showed an identical protein distribution between triplicate PTD runs for each temperature, indicating the reproducibility of the PTD process. Compared with the protein pattern at 37 °C, some protein bands were diminished or disappeared completely following PTD while CaII still remained in solution at 60 °C −AZM and at 65 °C +AZM. These results indicate that the protein complexity was greatly reduced by the removal of some untargeted proteins in PTD. It has also been reported that free CaII still remains functional for catalytic activity after heating at 60 °C for 5 minutes³¹. Therefore, −AZM and +AZM E. coli samples exposed to PTD at 60 °C were used for the following characterizations. BCA measurement suggested that approximately 40% of the protein was removed in 60 °C-depleted E. coli cell lysate (Figure 1B). We then further evaluated the identification of CaII peptides in both E. coli cell lysate without PTD (37 °C) and E. coli cell lysate with PTD at 60 °C following typical HDX-MS workflow (Figure 1C, Supplementary Figure 3). 66 ± 2 unique CaII peptides were identified in 60 °C-depleted E. coli cell lysate and 35% fewer unique CaII peptides (49 ± 2) were identified in E. coli cell lysate without PTD. 91.5 ± 0.5% sequence coverage of CaII was achieved in 60 °C-depleted E. coli cell lysate, approximately 7 % higher than the sequence coverage obtained in E. coli cell lysate without PTD. Moreover, 55% of the additional unique peptides less than 16 amino acids in length were identified in 60 °C-depleted E. coli cell lysate compared to those peptides identified in E. coli cell lysate without PTD. These shorter peptides are essential for increasing the resolution of the characterization of the protein-ligand binding site in native-like conditions. Overall, our results suggest that protein thermal depletion greatly reduces protein complexity in cell lysates, enabling more peptide identification and improved sequence coverage of the targeted protein. The improved results will be advantageous for the binding site characterization in the following HDX-MS analysis.

Figure 1.

Evaluation of the PTD application for reducing sample complexity. (A) SDS-PAGE evaluation of the reproducibility of PTD process. (−) AZM indicates E. coli cell lysate without AZM treatment while (+) AZM indicates the E. coli cell lysate with AZM treatment. (B) BCA evaluation of protein concentration of E.coli cell lysate (-AZM) before PTD (37 °C) and after PTD at 60 °C. (C) Comparison of the number of identied CaII peptides in E. coli cell lysate before PTD (37 °C) and after PTD at 60 °C following the HDX-MS workflow.

Increasing proteome coverage and peptide identification count are essential for drug target identification and binding site elucidation in HDX-MS analyses. Using E. coli digest, we previously demonstrated that subzero temperature, long gradient UPLC separation can be applied in HDX-MS analysis for improved peptide identification and proteome coverage with low deuterium back exchange¹⁹. Here, before performing HDX-MS analysis on 60 °C-depleted E. coli cell lysate, we optimized our gradient length (15 min to 45 min) for subzero temperature UPLC separation using the 60 °C-depleted E. coli cell lysate −AZM to obtain higher E. coli proteome coverage and peptide identification count. The base peak chromatographs using different gradient lengths are presented in Supplementary Figure 4. With the 45 min gradient, we identified 2010 peptides from 169 proteins, approximately 2 times the number of identified peptides in the 15 min gradient LC run. Increasing the LC gradient length also enabled the identification of additional CaII peptides.

We then performed differential HDX-MS for the elucidation of both the AZM target and mechanism-of-action using 60 °C-depleted E. coli samples and a 45-min LC gradient. We characterized 1220 peptides from 134 proteins in the deuterated samples (Supplementary table 1). A previous HDX-MS study on CaII and troglitazone, another CaII inhibitor, reported small deuterium uptake differences between free CaII and CaII bound with troglitazone due to the small binding interface³². Therefore, a deuterium uptake difference greater than 0.3 Da and p-value less than 0.01 were used as the cutoff in the volcano plot for statistical analysis of the differential deuterium uptake for the characterized E. coli peptides. As shown in Figure 2A, 6 peptides exhibited significantly reduced deuterium uptake after AZM treatment. These 6 peptides were all identified from CaII and were involved in the region 189–203. It was previously reported that the amino acids Leu198, Thr199, and Thr200 are involved in the hydrophobic pocket (Leu 198) and hydrophilic face (Thr199 and Thr200) of the binding cavity of CaII, regions which are reported to be vital for CaII ligand binding³³. Examples of MS spectra of the peptides involved in the region 189–203 are shown in Figure 2C and Supplementary Figure 5. Overall, we obtained 91.54% sequence coverage of CaII in the E. coli sample (Figure 2B and Supplementary Figure 6). A few examples of HDX-MS analysis for other E. coli proteins are shown in Supplementary Figure 7. The T_m of alkyl hydroperoxide reductase C was reported to be 63.1 °C and the T_m of 60 kDa chaperonin was reported to be 68.3 °C³⁰. In 60 °C-PTD sample, an 88.24% sequence coverage and a 72.10% sequence coverage were obtained for alkyl hydroperoxide reductase C and 60 kDa chaperonin, respectively. However, no significant differences in deuterium uptake for their characterized peptides following AZM treatment were detected (Figure 2A, Supplementary table 1). These results suggest that CaII is the target of AZM. Particularly, our results demonstrate that AZM targets the region 189–203 of CaII to inhibit catalysis. These findings are consistent with an X-ray crystallography study of the CaII-AZM interaction which reported that one hydrogen bond is formed between AZM and Thr199 on CAII³⁴. These results suggest that our PTD-HDX-MS platform is highly sensitive and can distinguish the minor deuterium uptake differences resulting from the small binding interface between a protein and its ligand. Moreover, this is the first reveal of the interaction between CaII and AZM in a complex biological environment by the analysis of thousands of peptides from hundreds of proteins. Our findings highlight the great promise of the application of the PTD-HDX-MS platform for the identification of unknown protein targets and the characterization of the binding site between the unknown target and ligand in complex biological samples.

Figure 2.

PTD-HDX-MS for the elucidation of the AZM target and mechanism-of-action in 60 °C-depleted cell lysate. (A) Volcano plot showing deuterium uptake differences of E. coli peptides with/without AZM treatment under different HDX labeling conditions (0.3 Da deuterium uptake difference and p value < 0.01 are set as cutoffs). (B) Structural interpretation of HDX-MS analysis of CaII (PDB:1V9E). The cyan color indicates no significant deuterium uptake difference in this region. The red color indicates significant deuterium uptake differences for peptides characterized in this region. The grey color indicates no coverage in the region. (C) MS spectra of CaII (189–203) in triplicate under different HDX labeling conditions.

Next, to further confirm the binding site between CaII and AZM, we performed HDX-MS analysis using pure CaII and AZM. As shown in the Woods’ plot (Figure 3), a 91.5% sequence coverage was achieved from the pure CaII sample. Three peptides covering the region 189–205 showed significantly less deuterium uptake upon AZM binding. The MS spectra of the three peptides are presented in Supplementary Figure 8. The HDX-MS results in the pure CaII samples are highly consistent with the PTD-HDX-MS results from the complex E. coli samples. We then evaluated the conformational differences between CaII heated to 37 °C and 60 °C to assess the conformation of CaII involved in our PTD-HDX-MS analysis. As shown in Supplementary Figure 9, no significant difference in deuterium uptake was detected for CaII heated at 37 °C or 60 °C. In addition, AZM binding resulted in similar isothermal titration calorimetry (ITC) profiles for CaII regardless of treatment temperature. These results indicate that the conformation of CaII after heating to 60 °C and cooling was the same as its native form.

Figure 3.

Woods’ plot from the HDX-MS analysis of pure CaII bound with AZM. The dashed and full lines indicate 98% and 99% confidence, respectively. Peptides with significant shifts in deuteration are indicated using red bars; and grey bars indicate peptides without significant differential deuterium uptake.

Generally, two steps are involved in protein thermal denaturation: the first step is reversible conformational change, and the second step is global unfolding^35–37. Once the denaturation conditions are removed, proteins partially unfolded in the first step can intrinsically recover and refold to their native states given sufficient time as demonstrated by Anfinsen’s dogma that the native conditions typically represent global thermodynamic minimum^{36, 38–39}. Many studies on diverse proteins indicate that proteins refolded to their native structures after moderate heating^{37, 40–43}. Upon exposure to temperatures ≥ T_m, proteins will denature and aggregate as a result of the association of exposed hydrophobic regions. A potential limitation of PTD is the possibility that some proteins may become trapped in metastable states in which conformation is misfolded to attain a local free energy minimum during refolding^{36, 44}. While this misfolded conformation may not affect differential HDX-MS analysis results when protein and protein-ligand complex are both in the same metastable states, implementation of appropriate heating temperature can help minimize local conformational changes to reduce the possibility for misfolded local conformation. Lower protein concentration in PTD can also increase the probability of proteins refolding to their native conformations prior to HDX-MS analysis³⁶. Overall, our results confirm that our PTD-HDX-MS platform coupled with subzero-temperature long gradient UPLC separation can accomplish high-throughput HDX-MS analyses for both unbiased identification of ligand targets as well as elucidation of the mechanism of protein-ligand interactions in cell lysate.

Here, we have integrated protein thermal depletion with subzero-temperature long gradient UPLC-HDX-MS (PTD-HDX-MS) to form a novel platform and demonstrated the implementation of the PTD-HDX-MS platform for the unbiased characterization of the interaction of CaII and AZM in E. coli cell lysate. Our results suggest that heating complex biological samples such as cell lysates can help deplete untargeted proteins, thereby greatly reducing sample complexity while the targeted protein remains in solution. Coupled with subzero temperature long gradient UPLC to boost LC separation power, our PTD-HDX-MS platform enables the high-throughput analysis of thousands of peptides from hundreds of proteins in single HDX-MS analysis. These results highlight the great potential and utility of the PTD-HDX-MS platform for the identification of ligand targets and characterization of protein-ligand interactions in highly complex biological samples. We are currently making improvements and optimizations of the PTD-HDX-MS platform to achieve even higher proteome and protein sequence coverage; for example, protease digestion in cell lysate is complicated due to the presence of diverse structures of cell lysate proteins. Optimization of cell lysate-based digestion in the HDX-MS workflow is important to obtain higher proteome sequence coverage. Additionally, although more than 13000 mass features were deconvoluted in the 45-minute LC gradient run, only 2010 mass features could be identified by the data-dependent acquisition method. The implementation of data-independent acquisition methods may help boost the peptide identification to achieve higher proteome sequence coverage. Finally, improvements in HDX software for handling highly complex samples may also improve PTD-HDX-MS analysis efforts in cell lysates.

As protein thermal shift assay-based methods such as cellular thermal shift assay (CETSA)^{29, 45} and thermal proteome profiling (TPP)^46–47 have been widely applied for protein target identification, known thermal stability of target proteins can be considered in the application of our PTD-HDX-MS strategy to directly characterize interactions of the ligand and the known target proteins in complex biological samples. When targeted proteins are unknown, determining and implementing appropriate depletion temperatures can be challenging. In this case, ligand-targeted MS detection or ligand-specific spectroscopy analysis (e.g., monitoring ligand concentrations in temperature-gradient samples) can be performed to select the optimal depletion temperature. Alternatively, TPP approaches can be implemented for protein target identification before the application of PTD-HDX-MS to cross-validate the identified targets as well as remove random targets from non-specific binding. This can be advantageous for PTD-HDX-MS over other prefractionation methods (e.g., ion exchange chromatography and SEC) for HDX-MS analysis. Still, PTD-HDX-MS may have limitations on detecting targeted proteins with low thermal stability due to inefficient depletion under low depletion temperatures. In this case, SEC-based prefractionation may be employed to solve this problem.

Optimization of protein digestion in cell lysates and implementation of DIA acquisition in HDX-MS may help boost peptide identification to achieve higher proteome sequence coverage for more complex samples (e.g., mammalian cell lysates). However, it is typically challenging for HDX-MS to distinguish differential uptakes due to direct ligand binding, allosteric changes, or ligand-induced protein-protein interaction. PTD-HDX-MS may also suffer in this regard. In this case, TPP methods or activity-based protein profiling (ABPP)⁴⁸ can be used to cross verify the identified protein targets and eliminate false positive results; complementary methods such as chemical cross-linking and mutagenesis can be employed to reveal the mechanism of action of ligands in cell lysates⁴⁹. In addition, thermal proximity coaggregation (TPCA) ⁵⁰ can be paired with our PTD-HDX-MS to discern induced protein-protein interactions as the result of overexpressing a protein in solution from protein-ligand interactions to see if the change in deuterium uptake relates to coaggregating proteins.

Overall, we envision that, by benefiting from both reduced sample complexity and improved LC separation power, our PTD-HDX-MS platform will promote the routine characterization of protein-ligand interaction in cell lysates and other complex samples such as serum and plasma to facilitate direct HDX-MS analysis of clinical samples.

SUPPORTING INFORMATION

• Evaluation of the melting points of CaII in E.coli cell lysate, validation of AZM target using two-temperature-points thermal proteome profiling, map of CaII peptides identified in E.coli cell lysate before and after 60 °C PTD, LC-HDX-MS analysis of CAII spiked E.coli cell lysate digest with different gradient lengths, MS spectra of CaII involved in binding site peptides in triplicate, Peptide map of CaII under 45 minutes gradient length, examples of structural interpretation of PTD-HDX-MS analysis of other E. coli proteins, MS spectra of CaII involved in binding site in triplicate under 120 s HDX labeling conditions in pure protein complex, evaluation of conformations of CaII under two conditions: CaII at 37 °C and CaII cooled down after heating at 60 °C for 5 minutes, materials and methods (PDF)