Multidimensional cross-linking and real-time informatics for multi-protein interaction studies

Abstract

Chemical cross-linking combined with mass spectrometry is a technique to study protein structures and identify protein complexes. Traditionally, chemical cross-linkers contain two reactive groups allowing them to covalently bond a pair of proximal residues, either within a protein or between two proteins. The output of a cross-linking experiment is a list of interacting site pairs that provide structural constraints for modeling new structures and complexes. Due to the binary reactive nature of cross-linking reagents, only pairs of interacting sites can be directly observed, and assembly of higher order structures typically requires prior knowledge of complex composition or iterative docking to produce a putative model. Here we describe a new tetrameric cross-linker bearing four amine-reactive groups, allowing it to covalently link up to four proteins simultaneously, and a real-time instrument method to facilitate the identification of these tetrameric cross-links. We applied this new cross-linker to isolated mitochondria and identified a number of higher-order cross-links in various OXPHOS complexes and ATP synthase, demonstrating its utility in characterizing complex interfaces. We also show that higher-order cross-links can be used to effectively filter models of large protein assemblies generated using Alphafold. Higher-dimensional cross-linking provides a new avenue for characterizing multiple protein interfaces even in complex samples such as intact mitochondria.

Introduction

There are a variety of analytical techniques for identifying protein-protein interactions, and they typically have a tradeoff between the rate at which interactions can be profiled and what level of structural information is obtained. High-throughput methods of interaction profiling, such as yeast two-hybrid¹ or affinity-purification mass spectrometry, can yield very large numbers of interacting proteins, but provide little to no structural information about the interaction. Alternatively, methods such as cryo-EM or x-ray crystallography can provide high-resolution structural information on complexes with hundreds of proteins in a single experiment but are comparatively low-throughput. As a technique, chemical cross-linking coupled with mass spectrometry offers a compromise of throughput and structural information, offering low-resolution distance constraints from many hundreds of pairs of proteins in an experiment.

Chemical cross-linking coupled with mass spectrometry is a broadly applicable technique to obtain spatial information about proteins and protein complexes. The knowledge gained from an identified cross-link includes the identity of two proximal residues that were linked to each other, which indicates the presence of a conformation where the residues are sufficiently close to be cross-linked through a solvent accessible distance. This distance constraint is a function of the cross-linker structure, and has an impact on what cross-links are possible to form². When formed between two different proteins, cross-links provide information about the identity of proteins in complex with each other and can provide sufficient information to assemble complex structures^3–6. Within a single protein, intra-links can provide information about conformational states of a protein and provide a proxy by which the relative populations of conformers can be measured⁷.

Cross-linking reagents vary dramatically in their structures and the chemistries they employ, as recently described⁸. The most common cross-linkers employed for whole proteome studies contain amine-reactive esters for targeting lysines and incorporate labile bonds in their backbone that break under CID or HCD activation. These labile bonds reduce the problem of cross-link assignment from a quadratic search to a pair of linear searches⁹. Beyond these initial parameters, cross-linkers have been designed to target carboxylic acids^4,10 or thiols¹¹, or alternatively incorporate a novel affinity tag like phosphonic acid for IMAC enrcihment¹². We explore a new direction in cross-linker design by increasing the number of reactive groups on the cross-linker, allowing it to react with up to four proteins simultaneously. This not only increases the level of structural constraints of model protein structures, but provides unique indication of the existence of interactions among more than two protein partners in a complex sample.

Here we present the synthesis of a tetrameric cross-linker and describe an associated real-time instrument method for automated targeting of released peptides to facilitate their identification. The molecule and method were initially tested by cross-linking bovine serum albumin (BSA). BSA has been used extensively for benchmarking a variety of cross-linkers and acquisition methods¹³. Finally, we applied this new method to mitochondria isolated from mouse hearts and explored the use of Alphafold2¹⁴ to model a trimeric interface in combination with a higher-order cross-link.

Methods

Cross-linker Synthesis

The peptide backbone of the tetrameric cross-linker, Bisby, was prepared on 0.05 mmol of low-loading rink amide protide resin (CEM) using Fmoc-based solid phase synthesis on a CEM Liberty Lite peptide synthesizer, with the sequence Lys-Lys₂-Pro₄-Asp₄-succinate₄. Following the succinylation, the resin was transferred to a Poly-prep column (Biorad) and incubated in 12-fold molar excess of TFA-NHP in pyridine for 20 minutes to yield the four terminal NHP-esters. The pyridine was then removed by vacuum filtration, and the resin was washed with 3x10 mL of DMF, allowing the final wash to incubate with the resin for 10 minutes before removal. Three additional washes were performed each with 10 mL of DCM to remove excess DMF. After washing, the esterified product was cleaved from the resin by incubation in a 2 mL cleavage solution comprised of 95% TFA in DCM for 3 hours at room temperature with gentle rocking. The linker was precipitated by adding the cleavage solution to 35 mL of cold diethyl ether and was further washed using 4x20mL of cold ether. The resulting pellet was dried to completion by vacuum centrifugation.

Bisby purification

Reverse phase chromatography was carried out on an Agilent 1200 series HPLC system using a 250mm x 9.4mm Partisil 10μm ODS-3 column (Whatman) equipped with a 20mm x 2.1mm BetaBasic C18 Javelin guard column (Thermo Scientific). The mobile phases consist of 0.1% TFA in water (Solvent A) and 0.1% TFA in acetonitrile (Solvent B). Crude product was resuspended at a concentration of 20 mg/mL in 0.1% TFA in 50% acetonitrile, and 10 mg were loaded per injection. Product was separated at a flowrate of 0.5 mL/min for 15 min with the following gradient: 50%B at 0 min, 70%B at 10 minutes, 50%B at 10.01 min until completion. Fractions were collected at 1-minute intervals starting at 4 minutes, and fraction 7 was retained (Supplemental Figure 1). Fraction 7 was pooled across all injections and dried to completion by vacuum centrifugation before being resuspended in DMSO for use. The purity of the stock solution was estimated by direct infusion (Figure 1) to be 35% for fully intact Bisby, increasing to 67% when considering the first and second hydrolysis products which can still generate productive cross-links.

Figure 1.

(A) Outline of synthesis of Bisby. Primary amine-reactive NHP esters are highlighted in blue and mass spectrometry cleavable DP bonds are highlighted in red. (B) Direct infusion spectrum of Bisby showing the intact species and partial hydrolysis products. Additional peaks corresponding to water loss (−18 Da) or TFA adducts (+97 Da) are also present. (C) CID fragmentation spectrum of Bisby (m/z = 1116) showing distribution of product ions formed by fragmentation of one to four DP bonds. CID activation predominantly leads to the fragmentation of one or two DP bonds.

BSA Cross-linking

One mg of Bovine serum albumin (BSA) was dissolved in 1mL of 170 mM Na₂HPO₄ pH 8.0. Bisby was then added to a final concentration of 0.1 mM, and the reaction was allowed to proceed for 30 minutes at 27°C with constant shaking at 600 rpm in a thermomixer. After cross-linking, urea was added to a final concentration of 8M, and then tris(2-carboxyethyl)phosphine (TCEP) was added to a final concentration of 5 mM to reduce disulfides, allowing the reduction to continue for 30 minutes at 27°C at 600 rpm on a thermomixer. Following reduction, iodoacetamide (IAA) was added to a final concentration of 15 mM, and cysteines were alkylated by mixing for 30 minutes at 27°C at 600 rpm on the thermomixer. After alkylation, the solution was diluted 8-fold with 100mM ammonium bicarbonate pH 8.0, trypsin was added at a 1:200 ratio of trypsin to protein, and digestion was carried out overnight at 37°C shaking at 600 rpm on a thermomixer. The resulting digest was desalted using Waters C18 sep-paks on a vacuum manifold, and the desalted peptides were then fractionated using peptide size exclusion chromatography.

Animal model

All protocols concerning animal use were approved by the Institutional Animal Care and Use Committee at University of Washington. This study utilized six wild-type mice, strain C67Bl6/NCrl (IMSR_CRL:27). Adult (>10 weeks old, weighing 22-24g) male mice were chosen randomly. Mice were housed in a vivarium with a 12-hr light/dark cycle at 22°C. Mice were maintained on ad libitum standard rodent diet and water.

Mitochondria isolation from cardiac tissue

Hearts were excised from mice and the aortas and atria were removed. Heart tissues were rinsed briefly in ice-cold mitochondria isolation medium (MIM: 70mM sucrose, 220mM mannitol, 5mM MOPS, 1.6mM carnitine hydrochloride, 1mM EDTA, 0.025% fatty acid-free BSA, pH 7.4 with 5M KOH), to remove residual blood. Tissues were minced on ice and resuspended in fresh MIM, followed by trypsin digestion (10μg/ml) and incubated on ice for 10 min. Trypsin digestion was stopped by the addition of trypsin inhibitor (0.5mg/ml) and additional BSA (1 mg/ml) to MIM. The suspension was centrifuged for 1 min at 1,500 x g at 4°C, and the supernatant was discarded. The tissue pellets were resuspended in fresh MIM containing 1mg/ml BSA, and transferred to a Teflon-glass tube and homogenized on ice with a Teflon pestle. The homogenates were centrifuged for 10 min at 800 x g at 4°C. The supernatants were collected and centrifuged for 10 min at 8,000 x g at 4°C. The supernatant was discarded, and the mitochondrial pellets were resuspended in MIM to wash. The resuspension was centrifuged for 10 min at 8,000 x g at 4°C, and the supernatant discarded. The mitochondrial pellet was used for the cross-linking reaction.

Mitochondria Cross-linking

Six total mice were used to generate four mitochondrial samples for cross-linking. The first sample contained half the mitochondria gathered from a pool of three mice hearts, while the remaining three samples are each generated from a single mouse. Each sample was resuspended in 100 μl of 170 mM Na₂HPO₄ pH 8.0, and Bisby was added to a final concentration of 10 mM. The solution was mixed at 600 rpm in a thermomixer at 27°C for 45 minutes. Samples were centrifuged at 8000g to pellet the mitochondria, and the supernatant was removed. Each sample was resuspended in 500 μl of 100 mM ammonium bicarbonate pH 8.0 and pelleted again at 8000g to wash. The samples were then resuspended in 100 μl of ammonium bicarbonate pH 8.0 containing 48mg of urea. The mitochondria are then lysed by sonication using a GE-130 ultrasonic processor. The samples were reduced with 5 mM TCEP for 30 min at 600rpm in a 27°C thermomixer and then alkylated with a final concentration of 15 mM iodoacetamide for 45 minutes in the dark. The samples were diluted to a final volume of 800 μl with 100 mM ammonium bicarbonate pH 8.0 and digested overnight at 37°C with 1:200 ratio of trypsin. Following digestion, samples were desalted with a Waters C18 sep-pak, dried to completion by vacuum centrifugation, and subjected to peptide size exclusion chromatography.

Size Exclusion Chromatography

Cross-linked peptides from both the BSA and mitochondrial samples were fractionated by peptide size-exclusion chromatography (SEC) using an AKTA Pure system (GE) equipped with a Superdex Peptide 10/300 GL column (GE). Desalted peptides were resuspended in 0.5 mL of 30% ACN/0.1% TFA in water. Peptides were fractionated using an isocratic flow at a rate of 0.5 mL/min consisting of 70% solvent A (0.1% TFA in water) and 30% solvent B (0.1% TFA in ACN) for 1.1 column volumes (CVs). 1 mL fractions were collected during elution, starting at 0.2 CV. For BSA, only fraction A5 was carried forward for LC-MS analysis. The A6 fraction for BSA was not carried forward for further analysis as it historically contains predominantly binary linkages. For the mitochondrial samples, fraction A5 and 200 μl of A6 were carried forward for LC-MS analysis, while the remaining 800 μl of fraction A6 were further fractionated by SCX.

Strong Cation Exchange Chromatography

Fraction A6 from the peptide SEC from each of the mitochondrial samples was further fractionated by SCX, as it contains a significant number of binary linked peptides that separate efficiently by SCX. SCX was carried out on an Agilent 1200 series HPLC system equipped with a 250 x 10 mm column packed with Luna 5 μm 100 Å particles (Phenomenex). The mobile phases consisted of 7 mM KH₂PO₄, 30% acetonitrile pH 2.8 (Solvent A) and 7 mM KH₂PO₄, 350 mM KCl, 30% ACN pH 2.8 (Solvent B). Fraction A6 was resuspended in 0.5 mL of solvent A prior to injection. Peptides were fractionated at a flow rate of 1.5 mL/min for 97.5 min with the following gradient: 0% B at 0 min, 5% B at 7.5 min, 60% B at 47.5 min, 100% B at 67.5 min, 100% B at 77.5 min, 0% B at 77.51 min to completion. Fractions were collected at 5-minute intervals starting at 17.5 minutes and were combined into 5 final pools consisting of fractions 6-7,8,9,10, and 11-14. Fractions were concentrated to approximately 1 mL final volume by vacuum centrifugation to remove acetonitrile, and then were desalted using Waters C18 sep-paks.

LC-MS Analysis

Peptides from all fractions were analyzed using a Waters NanoAcquity UPLC coupled to a Thermo Velos Fourier-transform ion cyclotron resonance mass spectrometer¹⁵ (Velos FT). Samples were fractionated over a 60cm x 75 μm inner diameter fused silica analytical column packed with ReprosSil-Pur C8 (5 um diameter, 120 A pore size) by applying a linear gradient from 88% solvent A (0.1% formic acid in water), 12% solvent B (0.1% formic acid in acetonitrile) to 70% solvent A, 30% solvent B over 240 minutes at a flow rate of 300 nL/min. The Velos-FT was operated using a real-time strategy developed for the analysis of tetra-linked peptides. A top 1 DDA method was used in which a high resolution (50000 mass resolving power at 400 m/z) MS1 scan from 400 to 2000 m/z is taken, followed by a high resolution (50000 mass resolving power at 400 m/z) MS2 scan on the most abundant ion of charge 4+ or greater not currently in dynamic exclusion. The effective low-mass cutoff from the m/z and charge preclude identification of cross-links comprised of small peptides because short peptides are often not unique to a single protein from large-scale databases and therefore, represent less useful expenditure of limited real-time MSⁿ capabilities. Each MS2 scan was processed in real-time to determine if a set of 4 peaks fulfilling a mass relationship are present, and if they were then 2 low resolution ion trap scans were taken for each valid target. MS1 scans used an AGC target of 5x10⁵, MS2 scans used an AGC target of 2x10⁵, and MS3 scans used an AGC target of 1x10⁵. MS1 and MS3 scans used a maximum injection time of 500 ms, while MS2 scans used a maximum injection time of 1500 ms. These ion targets and maximum injection times are the same employed in previous Real-time Analysis of Cross-linked peptides Technology (ReACT) analyses.

Real-time Instrument method

ReACT4 is an updated version of the original ReACT¹⁶ implementation for real-time targeting and sequencing of cross-linked peptides, and was implemented in ion trap control language (ITCL), the native control language used on LTQ series Thermo Scientific mass spectrometers. Whenever an MS2 spectrum was acquired, its full peak list was pulled from the acquisition system. First, this peak list was scanned to determine if any peak corresponds to the precursor mass – 215.042987, as this suggests that at least one arm of the cross-linker was hydrolyzed. This peak list was then deisotoped to remove redundant peaks from downstream consideration. In an initial pass of the deisotoped list, the peak list was analyzed to annotate all pairs of complement ions, any pair of neutral masses that add up to the precursor mass. Following the detection of complement ions, all ions with a charge of 3 or less were carried forward for relationship detection. A peak of neutral mass = 215.042987 was prepended to the peak list, which corresponds to the mass of a hydrolyzed arm. This reduced peak list was then used to construct an array of all pairwise sums of peaks in the spectrum. The array of pairwise sums was sorted by the summed neutral mass of its two constitutive peaks. Due to memory limitations on the instrument this procedure is limited to the 62 most abundant peaks in the reduced peak list, and the sort was performed by sorting up to four 500 element chunks which were then merged to produce the final sorted array. After the array was sorted, all solutions that fulfill the mass relationship

$$m_{precursor}\cong m_{reporter}+{\sum }_{i=1}^4{m}_{peak\left(i\right)}$$ (eq. 1)

were identified in a single pass of the array. In equation 1, ${\text{m}}_{\mathrm{precursor}}$ is the neutral mass of the precursor, ${\text{m}}_{\mathrm{reporter}}$ is the neutral mass of the reporter ion (789.525 for Bisby), and ${\text{m}}_{\mathrm{peak}\left(\text{i}\right)}$ is the neutral mass of some peak in the spectrum. Solutions to the mass relationship were found with a tolerance of 20 ppm. Once all solutions were determined they were ranked first by the number of unique complement ions observed for the set of four peaks, and then the solution with the highest summed intensity of the set of four peaks was selected for scheduling. If a peak corresponding to the precursor mass-215.042987 was detected, then solutions with at least one hydrolyzed arm were prioritized. For each peak, two ion-trap MS3 scans were scheduled, one targeting the observed peak that triggered the solution, and a second that targets either the same peak again or the calculated 2+ charge state of the ion if a 1+ ion was targeted for the first scan. Peaks corresponding to the mass of a hydrolyzed arm had no scans scheduled. If three or more peaks in the selected solution corresponded to hydrolyzed arms, then no scans were scheduled.

Data searching

The MS3 spectra were searched using Comet¹⁷ version 2019.01.5 using the default parameters for analysis of low-resolution spectra with a high-resolution precursor mass with the following modifications: isotope_error=5, allowed_missed_cleavage=5, ms_level=3; variable modifications: 15.9949 at M, 42.010565 at protein N-termini; required modification: 197.032422 at an internal K. The BSA search additionally included a variable modification of −17.02655 at peptide N-terminal C. BSA samples were searched against the uniprot bovine database, while mitochondria samples were searched against the MitoCarta 2.0 database¹⁸. Both databases were supplemented with reverse protein sequences for all entries to serve as decoys during the search. After searching with comet, peptide-spectrum matches (PSMs) were processed using PeptideProphet¹⁹ and iProphet²⁰ to assign probabilities to all PSMs, as well as ProteinProphet²¹ to perform protein inference. PSMs were filtered to a 1% FDR based off their iProphet probabilities, and then they were assembled into their appropriate grouping of tetra-/tri-/binary-links based off their parent MS2 scan. To account for the accumulation of decoys and corresponding increase in FDR associated with grouping, we performed an entrapment search (Supplemental Methods) to estimate the FDR at the cross-link level, which is estimated to be 4.4% (Supplemental Figure 2).

Structural Modeling

Colabfold²² was used to generate models of the ATPA/ATPG/ATPE hetero-trimer. Uniprot mouse sequences (Q03265, Q91VR2, P56382 for ATPA, ATPG, ATPE respectively) with the annotated mitochondrial signaling peptide removed from ATPA and ATPG were used as input. To generate models, we used 32 random seeds with 5 models each to generate a pool of 160 models. These models were clustered using Calibur²³, and clusters with more than 5 members were retained yielding 5 final clusters. The representative model from each cluster was then compared against an identified inter-protein tri-link between the three modeled proteins to evaluate model consistency with an observed tri-link.

Data Availability

Raw files for all samples are available on PRIDE²⁴ with the dataset identifier PXD032222 with username reviewer_pxd032222@ebi.ac.uk and password gFGy1m6P.

Results and Discussion

Linker Synthesis

Bisby is a tetrameric cross-linker prepared using solid-phase peptide synthesis (SPPS) employing traditional fmoc-based protection chemistry. SPPS provides the advantage of fast prototyping of molecules while enabling diverse functionalities to be incorporated and has been previously used in the synthesis of various protein interaction reporter²⁵ (PIR) cross-linkers. These previous PIR cross-linkers have included photocleavable groups²⁶, affinity enrichment tags¹⁶, and recently isobaric reagent sets for quantitative interactome studies^27,28. The primary distinguishing feature of Bisby from previous cross-linkers is the inclusion of four reactive NHP-esters, allowing it to react with up to four proximal primary amines during a cross-linking experiment. As with other members of the peptide-based Protein Interaction Reporter (PIR) family of cross-linkers, Bisby contains a CID-cleavable aspartyl-prolyl (DP) bonds in each arm that preferentially fragment during CID-activation (Figure 1). The inclusion of CID cleavable bonds enables facile MS3 targeting of released peptides by making use of a mass relationship connecting the precursor to its released peptides (equation 1).

Following low-PH reverse phase purification, we obtained Bisby and some partial hydrolysis products. CID-activation (NCE=35) of the 2+ charge state of Bisby yields a fragmentation spectrum with a distribution of DP bond cleavage events. The most abundant fragmentation product is formed by cleavage of two DP bonds, but fragments resulting from the cleavage of one, three, or all four DP-bonds are also observed. For each of these major fragmentation products, we also observe a second peak corresponding to water loss, likely from one of the aspartic acids. In general, we expect to observe a distribution of DP-bond cleavage events under CID activation, as ions fall out of resonance after fragmentation, causing some labile bonds to go unbroken.

Instrument method

Real-time mass spectrometry methods have become prominent and sophisticated in recent years, as they facilitate deeper and more efficient ways to interrogate samples. Real-time Comet searches enable improved quantitation with shorter gradients for enhanced²⁹ throughput, while augmented deconvolution and multiple precursor targeting³⁰ for top-down studies has been shown to improve sequence coverage. ReACT4 is a modified version of the original ReACT¹⁶ platform for real-time targeting of cross-linked peptides generated by cleavable cross-linkers and is the first implementation of a cross-linking strategy to identify cross-links composed of more than two peptides. The overview of the targeting strategy of ReACT4 is outlined in Figure 2. First, a high-resolution MS1 spectrum is acquired. The high-resolution is required to resolve the high mass and charge that tetra-linked peptide quartets typically carry. From this spectrum, an MS2 target of charge 4+ or higher is selected using traditional DDA logic and obeying dynamic exclusion, and a high-resolution MS2 spectrum is acquired. Charge state generally increases with the number of attached peptides, and a charge of 4 or higher facilitates sampling of both binary links and higher order links during a single run. This MS2 spectrum is used as input to ReACT4, wherein a set of 4 peaks fulfilling the mass relationship are determined using equation 1.

Figure 2.

Schematic representation of ReACT4 for real-time targeting of tetra-linked peptides. MS2 spectra are processed in real-time, and MS3 scans targeting each released peptide are scheduled only if a valid mass relationship was detected. Hydrolyzed arms are added to the peak list for all MS2 spectra to enable detection of binary, ternary, and quaternary cross-links during a single experiment, but are skipped for MS3 targeting.

Beyond the ability to detect tetra-linked peptides, ReACT4 includes several additional features to improve runtime and accuracy. Input peaks from the MS2 spectrum are deisotoped prior to conversion to neutral mass, and peaks with charges greater than three are not considered for the relationship detection. Both heuristics reduce the number of candidate peaks, and due to sort-related memory limitations on the instrument only 62 candidate peaks can be considered for each spectrum. Reducing the number of candidate peaks makes the most of limited memory while also improving runtime, as the number of solutions to the mass relationship to evaluate is quadratic with respect to the number of input peaks as solutions are found by iterating through a sorted list of all possible pairs of ions. Additionally, ReACT4 considers the presence of complement ions in the MS2 spectrum when selecting a peak set to target for MS3. Bisby contains 4 MS-cleavable bonds, and during CID activation, it is very common that only one of the bonds breaks to yield a released peptide but also yields the corresponding complement ion. As such, when determining the final set of peaks to target, solutions with the greatest number of unique observed complement ions are preferentially selected. Finally, ReACT4 can detect relationships resulting from partially hydrolyzed cross-links containing zero to four waters, enabling simultaneous identification of 2-, 3-, or 4-armed cross-links from a single run.

An MS3 based identification strategy was selected over a chimeric spectrum identification strategy for characterizing tetra-links. Chimeric spectrum identification strategies^9,31–32 are often favored over MS3 based strategies due to the faster duty cycle allowing a greater number of precursors to be sampled during an LC-MS run, while also not requiring MS3 capable hardware or real-time targeting methods. The primary downside to chimeric spectrum identification is that MS2 spectra resulting from the co-fragmentation of released peptides can yield spectra dominated by fragments from one of the peptides, making identification of the cross-linked species difficult or impossible³³. Intuitively, this downside is exacerbated with up to four co-fragmenting peptides in a single cross-link, as fragments need to be observed from all four peptides to fully identify a tetra-link. An MS3 based strategy avoids this difficulty, as only the released peptides and not their fragments need to be observed in the MS2 spectrum to enable MS3 targeting. MS3 isolation and fragmentation of each released peptide occurs serially and independently, improving the ability to obtain fragments from all four released peptides, at the cost of a slower duty cycle.

Cross-linking a purified protein

For initial testing of the new cross-linker, we cross-linked Bovine serum albumin (BSA), a staple protein standard for testing cross-linkers and cross-linking methods¹³. We utilized peptide size exclusion chromatography for enrichment of quaternary and ternary cross-links, which has been demonstrated previously as an efficient strategy for enriching binary cross-links³⁴. A detailed example of an identified tetra-link is shown in Figure 3. A high-mass high-charge ion (m/z = 931.786) was selected from the MS1 (Figure 3A), which generates a rich fragmentation spectrum after activation (Figure 3B). Candidate released peptides for this spectrum were detected at a sub-3ppm mass error according to equation 1. All four of these released peptides, SLGK*VGTR, LSQK*FPK, CCTK*PESER, and CASIQK*GFER, also have a complement ion corresponding to the precursor minus one of the released peptides (Figure 3B). These complement ions are used as the primary method to rank solutions when multiple mass relationships are detected in a single spectrum. Additional complement ions resulting from the loss of any combination of two released peptides can also be observed (Figure 3B), but they are not currently used for ranking potential solutions.

Figure 3.

(A) MS1 signal with inset showing isotope envelope selected for MS2 analysis, the peak selected for MS2 is indicated by a red asterisk. (B) Fragmentation spectrum of selected tetra-linked species with released peptides and complement ions formed from one and two peptide losses annotated. (C) Highest scoring MS3 spectrum from each released peptide annotated in panel B. (D) Cross-linked residues corresponding to this identified tetra-link mapped onto BSA (pdb: 3V03) with all pairwise C_α-C_α distances annotated.

Following the successful detection of a relationship from this MS2 spectrum, 8 MS3 spectra are scheduled, 2 for each target with the first scan targeting the observed peak fulfilling the mass relationship and the second targeting at least a 2+ charge of the same species, to increase the chance of obtaining an identifiable fragmentation spectrum as 1+ ions often fragment poorly. The higher scoring of each of those pairs is shown in Figure 3C, and it is this set of spectra that is used to fully identify the tetra-linked peptide species. This tetra-link can be mapped onto a crystal structure of BSA (pdb: 3V03, Figure 3D), which yields C_α-C_α distance ranging from 13.5 to 33.0 Å, within the expected maximum span of the linker of 45 Å.

Beyond this tetra-link, we additionally identified 30 other quaternary links, 105 ternary links, and 91 binary links. In total, these links can be represented as 103 unique lysine-lysine site linkages within a monomer of BSA. The number of unique lysine pairs identified is comparable to the average of 78 found across a diverse number of cross-linkers and workflows¹³, albeit with an additional technical replicate contributing additional unique pairs. While these cross-links can be predominantly mapped to BSA monomers, there are several links that contain unambiguous homo-dimer associations, and a small number of links that contain unambiguous homo-trimers of a peptide. Interestingly, in all cases where a homo-trimeric link is detected, the same peptide (SLGKVGTR) appears as the homo-trimeric peptide, suggesting some consistent assembly of a higher order homo-oligomer. BSA is known to form dimers and potentially trimers³⁵ depending on its concentration, so the observation of unambiguous homo-trimers seems physically plausible. All these links provide information about the spatial relationship of at least two lysines, yielding 2 to 6 distance constraints depending on the number of peptides identified in the cross-link. While quaternary links supply 6 distance constraints, they typically have reduced spectral quality for some released peptides compared to ternary and binary links, likely caused by reduced efficiency in producing released peptides to fragment for MS3. In contrast to traditional binary cross-linkers, not all arms of Bisby need to be assigned to a peptide sequence to provide structural information. In cases with hydrolysis of some arms or poor spectral evidence for some released peptides, structural information is still obtained if at least two arms can have a peptide sequence assigned.

Cross-linking isolated mitochondria

Bisby was further applied to cross-link mitochondria isolated from mouse hearts to explore its performance in a complex sample. Mitochondria serve as an ideal system due to the many well characterized complexes and super-complexes that are critical to mitochondrial function. In our initial application of Bisby to mitochondria, we identified 39 unique quaternary links, 186 unique ternary links, and 801 binary links after FDR filtering. In total, these can be flattened to 691 unique lysine-lysine site pairs involving 249 protein pairs and can be viewed in the MitoBXP_mouse_mixed_Bruce table on XLinkDB. The comparison of C_α – C_α in various mitochondrial cross-linking experiments (Supplemental Methods) reveals a correlation between the observed distance distribution and cross-linker spacer-arm length, with Bisby forming links between lysines that are farther apart on average compared to a smaller cross-linker like DSSO (Supplemental Figure 3). These links are concentrated heavily in electron transport chain complexes, large chaperones, and members of the TCA cycle. For example, we identify an unambiguous homo-tetrameric link between copies of CH10 (Supplemental Figure 4), consistent with its known heptameric assembly (PDB: 4PJ1). Overall, the depth of site pair coverage obtained by Bisby is less than commonly used binary cross-linkers, but unlike binary cross-linkers Bisby provides the benefit of unambiguous identification of three and four protein interfaces. Greater depth could likely be obtained through more extensive fractionation or the inclusion of an affinity tag such as a biotin tag¹⁶, azide group³⁶, or phosphonic acid¹² as with previous cross-linkers, at the tradeoff of a more difficult synthesis.

One of the tetra-links identified forms a link between ATPA, ATPB, and ATIF1. ATPA and ATPB are catalytic subunits of ATP synthase, also called Complex V (CV), and the ATPA-ATPB interfaces form the catalytic pockets for ATP synthesis. This quaternary link represents the first unambiguous identification of an interface formed from three different proteins from a single cross-link. ATIF1 is a known CV inhibitor that functions by halting rotation of the central stalk thought to occur during conditions with low matrix pH to prevent reverse rotation and consumption of ATP by CV. This tetra-link, as well as all the other links between ATIF1 and ATP synthase are shown in Figure 4, consistent with the cryo-EM structure of ATIF1 in complex with ATP synthase. Further reinforcing confidence in this tetra-link assignment, all four potential constitutive ternary links and all six potential binary links were also observed. Although already known from 6J5K, the tetra-links and tri-links in Fig 4B conclusively identified the ternary ATPA-ATPB-ATIF1 interaction that was present in mitochondria at the time of cross-linker application. This conclusion is not generally possible with consideration of binary links only and offers new opportunities for mapping multi-subunit complexes in mitochondria or other complex systems.

Figure 4.

(A) Positional context for interaction between ATIF1 and ATP Synthase from half of the tetrameric structure. (B) Zoomed in inset highlighting an identified tetra-link (red) between two residues of ATIF1, and one residue each from ATPA and ATPB. Also shown are all the ternary (blue) and binary links (yellow) identified between ATIF1 and ATPA or ATPB. The tetra-link identified here is supported by all four possible ternary links and all six possible binary links between the four linked lysines.

One of the new functionalities enabled by Bisby is the unambiguous identification of protein complex interfaces comprised of more than two proteins. Traditionally, multi-protein interfaces are assembled from binary cross-linking data using observed cross-links combined with prior knowledge of complex composition and iterative docking methods to assemble complexes protein by protein. Iterative docking to assemble complexes of more than two proteins using constraints provided by multiple binary cross-links can be problematic, as some complex members could be mutually exclusive. Binary interaction data connecting three different proteins does not provide the ability to discriminate, for example, between a set of three heterodimers or a single heterotrimer without additional information. Bisby helps resolve this ambiguity by facilitating identification of up to four proteins at an interface, which can only occur if none of those identified proteins are mutually exclusive with each other. These multi-protein interfaces can now be modeled in a single step usingAlphafold¹⁴ which enable simultaneous modeling of an arbitrary number of proteins in complex³⁷. This method of modeling has the added benefit of allowing conformational rearrangements associated with complex formation that is not captured by rigid-body docking, which may improve accuracy of the resulting complex³⁸.

To explore this combination of a tetrameric cross-linker and Alphafold2 modeling of protein complexes, a single tri-link to model the three-protein interface between ATPA, ATPG, and ATPE from ATP synthase was used. Preliminary results suggest that agreement with cross-linking distance constraints correlates with higher confidence dimer models from Alphafold2³⁹. A potential challenge for this assembly for traditional iterative docking is that ATPA and ATPE share no interface in any rotational state, making it challenging to determine their relative orientation in the absence of additional constraints. A single wholly inter-protein tri-link provides two important pieces of information. It provides unambiguous evidence that three proteins were close to one another as well as supplying a set of three distance constraints that must be fulfilled in some conformation. These two pieces of information provide both a target to model, and information by which to filter resulting models for accuracy.

Colabfold²² was used to generate 160 models of these three proteins in complex to capture a breadth of potential complex conformations. These models were then clustered using Calibur to find representative models, and clusters with more than 5 members were evaluated against the tri-link. Two clusters had representative models that were consistent with a tri-link between ATPA:K454, ATPG:K39, and ATPE:K37 (Figure 5) were identified. These models are very similar to cryo-EM structures obtained for ATP synthase⁴⁰ (PDB: 6J5K, Figure 5), and differ with varying degrees of rotation of ATPA about the central axis of ATPG. In contrast, models that are inconsistent with this tri-link have ATPA rotated perpendicular to ATPG, or are rotated too far around the central axis with respect to ATPG. In these cases, one or more distance constraints supplied by the trimeric cross-link are violated, allowing us to reject the model. The models that have all three constraints satisfied have excellent agreement with known structures, especially considering the small interface shared by ATPA and ATPG, and the lack of interface between ATPE and ATPA.

Figure 5.

(A) Representative models from clusters that are consistent with an observed ternary link between ATPA (red), ATPG (cyan), and ATPE (blue), aligned against the same proteins from one rotamer of ATP synthase (grey; pdb: 6J5I; chains: A,G,I). A ternary link yields 3 pairwise distance constraints that must be consistent with some assembly. (B) Representative models from clusters that can be rejected due to the presence of at least one unfulfilled distance constraint derived from the same ternary link.

Summary

XL-MS provides a method for identifying protein-protein interactions from complex systems, which can provide new biological insights. Previously, XL-MS has provided information about numerous binary interactions between or within proteins, which can be used to identify and potentially assemble complex structures. Here, traditional cross-linking chemistry and methodologies were extended to facilitate the linking and identification of up to four proteins simultaneously with the synthesis of the tetra-reactive cross-linker, Bisby. We demonstrated this technology with a tetrameric cross-linker, but the framework is conceptually scalable to an arbitrary number of reactive groups. The primary limitation is likely to be in the successful synthesis of a larger cross-linker, as incomplete reaction products will quickly complicate the product mixture. Solving the mass-relationship scales exponentially with half the number of reactive groups (n for 2-arms, n² for 4-arms, n³ for 6-arms, etc.), but the presence of complement ions can be used to greatly reduce the number of peaks carried forward as a significant time-saving heuristic. While it would be difficult to solve this calculation on the duty cycle time-scale in ITCL for more than 4-arms due to language specific challenges, it is likely feasible to solve at least the 6-arm mass relationship in a C# implementation with the current iAPI. Beyond the synthesis and computational complexities, increasing the number of released peptides in a single spectrum will result in signal dilution that would further complicate their identification. Further studies would be required to determine the practical bottleneck of higher-order cross-linkers and characterize what additional utility they would provide.

A real-time strategy for the analysis of tetra-linked products based on the original ReACT strategy was developed and used to facilitate multi-linked peptide identification, where a mass relationship was used to determine the constitutive released peptides and then schedule MS3 targeting. As an initial application, Bisby was used to cross-link BSA to demonstrate that tetra-linked peptides were formed and ReACT4 was useful for identification. Next, Bisby was applied to a more complex system of isolated mitochondria, where generation and identification of tetra-links and tri-links from chaperones and OXPHOS complexes in their native environment was demonstrated. While the initial implementation of ReACT4 was performed within ITCL on a Velos-FTICR, it could be similarly implemented using the API on Orbitrap-tribrid instruments for broader use.