Reassembling protein complexes after controlled disassembly by top-down mass spectrometry in native mode

Abstract

The combined use of electrospray ionization run in so-called “native mode” with top-down mass spectrometry (nTDMS) is enhancing both structural biology and discovery proteomics by providing three levels of information in a single experiment: the intact mass of a protein or complex, the masses of its subunits and non-covalent cofactors, and fragment ion masses from direct dissociation of subunits that capture the primary sequence and combinations of diverse post-translational modifications (PTMs). While intact mass data are readily deconvoluted using well-known software options, the analysis of fragmentation data that result from a tandem MS experiment – essential for proteoform characterization – is not yet standardized. In this tutorial, we offer a decision-tree for the analysis of nTDMS experiments on protein complexes and diverse bioassemblies. We include an overview of strategies to navigate this type of analysis, provide example data sets, and highlight software for the hypothesis-driven interrogation of fragment ions for localization of PTMs, metals, and cofactors on native proteoforms. Throughout we have emphasized the key features (deconvolution, search mode, validation, other) that the reader can consider when deciding upon their specific experimental and data processing design using both open-access and commercial software.

Introduction

An emerging trend in structural biology involves using tools and methods from proteomics, particularly mass spectrometry (MS), to investigate the composition, stoichiometry, assembly, and topology of protein complexes [1–8]. One technique, top-down MS run in “native mode” (nTDMS), employs a three-tiered strategy for compositional analysis that employs non-denaturing electrospray ionization (ESI) to preserve non-covalent protein assemblies and determine their intact mass, followed by subunit disassembly and further primary sequence characterization by tandem MS [9–12] (Fig. 1A). There are three levels of mass information operative in the nTDMS analysis of a protein complex — the intact complex (MS¹), its subunits (MS²), and the subunits’ N/C-terminal fragment ions, such as b/y and c/z ion types (MS³) [9, 13]. The fragmentation step uniquely provides “proteoform” resolution that reveals subtle changes to a protein’s canonical sequence that may be critical to structure and function (e.g., coding single nucleotide polymorphisms, mutations, splice variants, post-translational modifications (PTMs), endogenous proteolysis, or allosteric modulators and co-factors) [10]. Importantly, MSⁿ is an instrument-centric term that describes the levels of tandem MS performed on the analyte. In the case of a monomeric protein, the MS² describes the fragmentation step; in contrast, the fragmentation of a subunit ejected from a complex is denoted by the 3 in MS³ to denote the third stage of MS analysis.

Figure 1.

Depictions of nTDMS data processing (A) and of the Bioinformatics and Validation of the data (B).

A recently published resource [14] provides extensive details on how to perform intact and tandem MS experiments on a standardized set of native-state proteins and complexes, including discussion on key parameters related to the data acquisition and the interpretation of the MS¹ and MS² data using readily available deconvolution software, such as Unidec and MagTran [15, 16]. Recently, progress has also been made on the integration of intact mass data (MS¹ and MS² of protein complexes) with fragmentation (MS³ of ejected subunit), or alternative omics approaches (e.g., bottom-up), for a comprehensive “native-proteomics” pipeline for large-scale characterization of native-state protein complexes [17, 18]. For example, MetaOdysseus, is an R-based software tool that integrates bottom-up with native top-down protein analysis for the characterization of metal–protein complexes (i.e., metal–protein stoichiometry, identification of metal-binding sites, or characterization of metal-coupled folding events)[18]. Similarly, the Search Engine for Multi-Proteoform Complexes was created as an online search strategy for identification of monomeric proteins and complexes with proteoform resolution. This platform [19, 20] utilizes search routines that combine protein interaction information from the CORUM database of human protein complexes [21] with comprehensive PTM, isoform, and cofactor information from the UniProtKB database (e.g., ligands, metals, etc.) to allow precise identification and scoring of subunits, along with bound non-covalent cofactors and covalent PTMs.

Subunit fragment data emulates MS/MS proteomics datasets while retaining the heterogeneity introduced by variable complexation to proteins, cofactors, and adducts. This retained heterogeneity makes traditional data processing challenging, as in conventional TDMS software, all modifications are considered as covalent and not transient. For example, a metal cofactor may or may not be retained post-fragmentation, resulting in traditional software annotating two separate “proteoforms” despite only the bound form being present (Fig. 1B) [8, 14, 17, 22, 23]. Moreover, subunits and their fragment ions can partially retain non-covalent cofactors in combination with other PTMs, complicating the analysis. For these reasons, the assessment of both raw and processed fragmentation data, along with manual validation of results, is critical for confident proteoform assertions and relies on steps that are highly time-consuming and often incompatible with the automated search engines used in proteomics [14]. More advanced bioinformatics analysis for specialized hypothesis-testing of fragments most often relies on the creative re-purposing of tools made for other proteomics applications, presenting a significant barrier to entry for new researchers [24–28]. To help address this issue, we outline strategies for the analysis of fragmentation data with examples for zinc-bound Carbonic Anhydrase (CA), a modified histone ejected from a nucleosome (Nuc), and copper-bound and oxidized amyloid beta (Aβ). This report deploys a decision-tree (Fig. 2) to guide the analysis of this data using open-access, freeware and commercial resources (Table S1) to (1) identify the subunits of protein-complexes in mixtures, (2) derive proteoform-level insights (e.g., sequence variants or PTMs), and (3) map non-covalent cofactors. Throughout, we highlight many shared features of existing open-access and commercial software (e.g., deconvolution approaches, informatic search modality, validation resources) to inform the reader how to best design their unique experimental and data processing pipelines (Table S1). The examples provided were obtained on Orbitrap hybrid instruments in high mass modes; however, the software resources discussed are also amenable to mass spectrometers that selectively activate targets, such as quadrupole time-of-flight (QTOF) or Fourier Transform ion cyclotron resonance (FT-ICR) [29, 30]. Finally, in a supplemental discussion we briefly introduce a new nTDMS software resource, ProSight Native, that is built for seamless navigation across the various levels of tandem MS through the integration of mass deconvolution, native fragmentation data, proteoform identification, and stoichiometry calculations into a single resource (see Supplemental Discussion 4).

Figure 2.

Decision tree to guide the identification and characterization of unknown proteoforms, including those harboring multiple post-translational modifications or non-covalent cofactors. The name of select software tools used at each step are in italics, with an expanded list of common open-source and commercial resources detailed in Table S1. Much of this decision tree has been incorporated into a new nTDMS resource, ProSight Native, that provides a platform for the analysis of native proteoforms and for determining the composition of protein complexes (see Supplemental Discussion 4 (SD.4)). Other topics discussed in more detail in the supplement include: (SD.1) Deconvolution strategies of fragmentation data; (SD.2) the calculation of a precise mass shift from the fragmentation data; (SD.3) interrogation of histone H3 datasets for fragments containing and acetylation site; (SD.5) and introductions to the discovery/validation resource, TDVALIDATOR; (SD.6) an introduction to a proteoform database generation tool, PROTEIN ANNOTATOR; (Table S1) and an overview of existing open-access, freeware and commercial resources commonly used for nTDMS.

An important distinction within the described workflow is the concept of subunit “identification” vs. “characterization”. Here, different confidence metrics can often be used to describe the quality of the identified protein in a database (e.g., P-score, E-score, Q-score) [31] and describe the selectivity of the proteoform characterized (e.g., C-score) [26]. While the goal of any informatics approach is to arrive at a solution that achieves absolute confidence via automated search engines independent of user intervention, asserting the existence of a modified proteoform is challenging in practice. In our experience, many datasets need researcher intervention to manage the complexities of the fragmentation data quality, such as the simultaneous fragmentation of isobaric proteoforms, the complete or partial ejection of labile PTMs or non-covalent cofactors (e.g., phosphorylation, glycosylation, metals), charge-state dependence of protein-metal ion interactions [32], or possible rearrangement of binding sites upon activation in the gas-phase [33].

Protein Identification and Proteoform Characterization

Fragmentation datasets are made up of distinct – though often overlapping – isotopic distributions for the fragment ions in the mass-to-charge (m/z) domain. Typically the observed fragment m/z values are converted into monoisotopic masses by various deconvolution algorithms that are available as part of the manufacturer’s software packages (e.g., Xtract, FreeStyle, etc.) or through open-source/free tools like mMass [34], Mash Explorer, etc.; Table S1). These tools use assorted algorithmic strategies with varied mass determination approaches for resolved isotopic distributions that result in subtly different fragment mass and intensity outputs (Supplemental Discussion 1). The most important variable at this level is the signal-to-noise ratio (SNR) thresholds for deconvoluting fragment ions, wherein a lower value often enables detection of low-abundant fragments that can improve characterization metrics above acceptable threshold limits (e.g., C-score ≥ 40) [26]. However, improved characterization can come at the expense of generating a noisier output that can generate false positive hits that negatively influence the identification confidence scores typically employed (e.g., P-Score ≤ 0.01) [35]. Therefore, for identification of unknown subunits in a complex, emphasis must first be placed on improving identification scores; this often means that searches should ideally be performed with robust, high-quality fragment ion populations (e.g., SNR ≥ 3 is a good starting point, but can vary from 10:1 down to 1.2:1 depending on the application).

Automated identification of unknown proteins in complexes can be achieved through the repurposing of top-down MS search engines (Fig. 2 and Table S1; e.g., ProSight PC, ProSight PD, BIGMASCOT, TopPIC, TDPortal [36, 37]). These tools produce a list of protein candidates, graphical fragment maps showing sequence coverage, and various statistics associated with the confidence of the identification between candidate proteins vs. candidate proteoforms for a given protein. In most cases, these resources utilize knowledge of the subunit mass to preselect candidate proteins to test against the collected fragmentation data [13, 24, 26, 35]. In cases where the identity of the protein is known, tools like ProSight Lite or Mash Explorer can be useful for visualization of the fragment ions directly on the sequence. This type of analysis can be used to derive confidence protein/proteoform scores (e.g., P-score, C-score) independent from database searches [24, 25]. These tools also allow the user to determine if observed fragments support user-defined sequences containing PTMs, cofactors, or sequence variants (e.g., SNPs, mutations, or proteolytic truncations) at different positions [23].

As with any -omics approach, fidelity of the search is highly dependent upon the content of the database. Resources such as UniProt [38] provide prepopulated databases for an individual organism’s proteome that can be added to search engines. Moreover, these databases can be further annotated to account for common events closely tied to gene-translation (e.g., cleavage of initiator methionine, signal peptide removal, or N-terminal acetylation). To support accurate proteoform assignment, Protein Annotator (Fig. S7) helps amend databases to include candidate proteoforms that result from removal of signal peptides, known sequence variations, PTMs and cofactors available in UniProt or UniMod for a more targeted approach. However, unlimited restrictions on the size of proteoform families in a database can have deleterious consequences both on search speed as well as confidence metrics linked to database size (e.g., expectation scores). In such instances, it is best to separate the identification from the characterization step.

Characterization of Unknown Events

Asserting the existence of a proteoform necessitates a high level of informatic rigor to discern the true form of the protein present in a complex. Specific sequence events (i.e., sequence variants, PTMs, or cofactors) can affect the intact mass shift and create a confusing fragmentation landscape (e.g., bound-complex detected by MS¹ and/or MS²) [14]. For example, proteins with intra-molecular disulfide bridges – treated as covalent PTMs – can fragment in unconventional ways given their rigid structure and inaccessible termini [17]. Additionally, proteolysis or modifications at one or both termini, affinity tags [39], and the generation of internal fragments can complicate the analysis and shroud the presence of added PTMs or cofactors. To overcome these challenges, different fragmentation techniques may provide complementary results (e.g., HCD, CID, EThcD, and UVPD) that corroborate and improve sequence coverage [40]. However, not all fragmentation techniques are universally supported by available software, such as UVPD [41]. While the annotation of internal fragments can significantly improve sequence coverage, this is not commonly supported in existing tools. Testing for diverse internal fragments (ax, ay, az, bx, by, bz, cx, cy, and cz) exponentially increases search times and adversely impacts most scoring metrics via high false discovery rates [42]. However, TDValidator and the new tool ClipsMS can assign both terminal and internal fragments for TDMS [43].

While there are no established baseline criteria for confidently asserting the existence of a proteoform, a general recommendation is to consider a proteoform “characterized” if the observed and theoretical masses match and there is adequate fragmentation coverage to localize unique proteoform-specific features. Various statistical metrics (C-score) can also help to establish confidence in identified proteoforms. If the observed and theoretical masses for a given protein are not in agreement, this is a clear indication that the analyte is an unknown proteoform or contains a cofactor. To elucidate the unknown, an informative first approach is to perform a de novo PTM identification; this entails shortlisting the chemical formula of a modification based on the precise mass shift — preferably within hundreds of a Dalton — directly calculated from the fragmentation evidence. For additional information on how to calculate the exact mass shift from fragmentation data, see the Supplemental Discussion 2 on this topic.

Another useful approach to interrogating unresolved mass shifts is to identify gaps in fragmentation coverage that may exist across the sequence. As a starting point, the mass shift is added to the N- or C-terminus using a sequence gazer (e.g., ProSight Lite), revealing the nearest terminus to the sequence event in terms of sequence coverage – more fragments may support a site near the correct terminus (the “delta-M” search in ProSight PC automates this function). To find the exact location, the PTM or cofactor can be iteratively added to possible sites downstream of the “correct” terminus as the fragmentation quality is monitored (i.e., with metrics like the numbers of matched fragments and characterization scores).

In the challenging case where the mass residual entails a combination of smaller modifications, there is currently no straightforward solution. If the fragmentation data is deemed to be of high quality, then one can justify the effort put into splitting the mass shift into smaller “hypothetical” modifications and testing different parts of the sequence with the new hypothetical masses. Relying on prior knowledge to define the most logical explanations for each protein will facilitate such hypothesis-driven proteoform interrogation. Here, tools such as mMass [34], Mash Explorer [44], MetaOdysseus [18], or TDValidator [28] can then help determine if additional fragmentation coverage due to multiple modifications stems from bona fide signals in the data.

Given the labile nature of many PTMs (e.g., glycosylation) and non-covalent cofactors (e.g., metals), the intact mass data on the complex or ejected subunit may reveal a mass shift for which the fragmentation data shows only weak (or no) evidence (reflected in poor fragmentation or scoring metrics) [8, 23, 29, 30]. This can be the case when the collisional energy ejects the labile species prior to subunit dissociation or during backbone cleavage, or because multiple residues from disparate parts of the protein are involved in the binding/coordination of a cofactor/metal. In the latter case, it is important to clarify that localizing labile cofactors, such as metals, actually involves finding evidence that they are retained on a subset of fragment ions [33, 45].

If the fragmentation data has sufficient SNR, it can be helpful to first treat the metal as a “known PTM” and manually iterate through the protein sequence on a sequence gazer (e.g., ProSight Lite) or automatically using Protein Annotator and ProSight PC to find possible binding sites. Anecdotally, larger fragments at higher charge states seem likelier to contain labile cofactors, perhaps because they sterically shield the cofactor/binding region from disruptive collisions. Given that these tools rely on the deconvoluted data, it is important to return to the raw mass spectrum and manually validate the quality of the fragmentation.

Validation

The final analytical step – the validation of the fragment assignments – constitutes a rigorous determination of data quality for a given characterized proteoform (see NOTE 1). While validation can be time-consuming, it is highly complementary to any informatics workflow and bolsters researchers’ confidence in both protein identification and proteoform characterization. Once a suspected sequence event has been found and its position narrowed down to a single (or few) residue(s), it is important to assess (and often report) the spectral quality of the diagnostic ions critical to validating a proteoform’s identity within the broadband mass spectrum. Similarly, comparing the intensities of the diagnostic fragments is useful to determine the ratio of the positional isomers that were fragmented simultaneously [46].

Various software tools available support the validation process that typically involves generating theoretical fragments that can be qualitatively compared to the original broadband spectrum in the m/z domain (Table S1) (Fig. 1B). For example, TDValidator automatically searches the m/z spectrum for fragments corresponding to a proteoform of interest; the researcher can then manually compare the raw data to the theoretical isotopologue distributions for validation. A complementary approach that is useful for hypothesis-driven fragment interrogation – i.e., looking for evidence that a specific sequence event is present – involves fragmenting the proteoform in silico and comparing observed and theoretical fragment ions directly within the spectrum [17, 23]. For this purpose, mMass can generate a list of all possible fragment ions and their charge states for a given proteoform with a defined modification state or cofactors. The user can then select a fragment of interest and overlay its theoretical isotopic distribution on the raw data to confirm its presence.

There are some recommended strategies for choosing which fragments to investigate: (1) structural and spectroscopic data in the literature can reveal potential sites/residues in similar proteins or for the specific metal ion; (2) D/E/G/P sites are highly prone to HCD fragmentation and generate high signal intensity fragments [47, 48]; and (3) if intense but unbound fragment ions are detected in the data, these fragments may have unassigned satellite peaks with a mass shift corresponding to a cofactor.

To put this framework to practice and walk readers through the data analysis workflow, we provide our analyses of three protein systems: (1) localizing zinc-binding on Carbonic Anhydrase (CA); (2) determining the acetylation site on a histone ejected from a whole nucleosome (Nuc); and (3) localizing PTMs and copper-binding on the amyloid beta peptide (Aβ).

Example Datasets

Metal-bound Carbonic Anhydrase

Protein structure is often reliant upon binding of small ligands (such as metals) whose conserved interactions contribute to stability of tertiary and quaternary structures. Though an active area of research, it is possible to localize ligand binding sites using nTDMS data by asserting either the regions or the specific residues that make up the binding pocket. We highlight this process using bovine Carbonic Anhydrase II (CA, 29 kDa) as an example, a well-studied zinc metalloenzyme that catalyzes the reversible hydration of carbon dioxide in vivo [49]. Given that a published crystal structure clearly captures CA binding to Zn(II) via residues H93, 95, and 118 [49], we chose it as a model system for developing tips and best practices in nTDMS analysis related to localizing ligands bound through multiple positions via non-covalent bonding [14, 23, 33, 50–52].

First, the MS¹ analysis demonstrates that the predominant species is N-terminally acetylated CA (theoretical average mass of 29,024.32 Da) complexed with a Zn(II) ion (63.18 Da) at 1:1 stoichiometry (Fig. 3A, see NOTE 2), and shows a minor signal for sodium-bound CA + Zn(II) as well as non-covalent adducts present at low relative abundance (e.g., bicarbonate) [14]. To maximize sequence coverage for localization of binding sites, we used multiple fragmentation techniques to characterize CA and its metal binding site (i.e., HCD, EThcD and UVPD, Fig. 3B). We show iterative hypothesis testing of the metal site by placing it on multiple candidate residues throughout the protein, including the termini, and monitoring for the sites that yield the greatest number of matched fragments and ultimately the lowest P-score (Fig. 3C). In particular, there is an extensive number of unbound-fragments on the N- and C-terminal ends of the protein that are not sufficiently represented in a bound-form, indicating that the zinc modification is likely in the core of the primary sequence where bound-form coverage is maximized and unbound-form coverage is minimized (for a visualization of unbound- vs. bound-fragment intensities as a function of the CA sequence, see Fig. S5 and Supplemental Discussion 4 on ProSight Native).

Figure 3.

nTDMS analysis of the bovine Carbonic Anhydrase II in complex with Zn(II). (A) Broadband mass spectrum indicates CA is complexed with a Zn(II) ion at 1:1 stoichiometry. (B) Representative HCD, EThcD, and UVPD fragmentation spectra for CA+Zn(II) complex. (C) ProSight Lite fragment maps showing P-score minimization strategy as Zn(II) cofactor is placed at N- and C-termini, at H63, and at H93. (D) UVPD and EThcD fragment maps for Zn(II) at H93. (E) TDValidator helps localize the Zn(II) ion to H93. Color boxes highlight diagnostic fragments depicted in (F). (F) Examples of theoretical fragment isotopic distributions (red triangles) overlaid on top of experimental fragment isotopic distributions. The spectral window is wide enough to visualize Zn-bound and unbound fragments.

We tested Zn(II) on H65 and H93 as two possible binding resides using the HCD data and found that their P-scores are comparable. Alternative fragmentation techniques like EThcD and UVPD (Fig. 3D) were also used to obtain additional fragments that support metal localization at or near H93 with their respective P-scores. Nonetheless, the P-score metric alone on ProSight Lite is not sufficient to confidently assert the Zn(II) site given the labile nature of metal cofactors. Specifically, the P-score indicates the probability of matching ‘n’ ions (or more) by chance alone. Because the proteoform [CA + Zn(II)] can formally be composed of positional isomers with different metal binding, “chance alone” is not the only factor to consider when comparing P-scores. Therefore, examination of the raw data using tools such as TDValidator, Mash Explorer or mMass is imperative in order to evaluate the quality of the “localizing” fragments identified by ProSight Lite and determine the likeliest Zn-coordinating residues.

Given that the metal or a given non-covalent cofactor may not always be retained upon fragmentation, it is important to manually validate both the unbound- and bound-fragments that occur at the same fragmentation site (Fig. 3E). To do this, we used TDValidator to carefully inspect and validate the spectral signals that give rise to the “fragments” in the sequence maps of Zn-bound CA (Fig. S6). This software aligns the raw data with theoretical isotopic distributions for Zn-bound and unbound fragments, producing a list of “validated” fragment ions and the corresponding scoring metrics plus a graphical map. The examination of each fragment in turn enabled examination of their metal content, painting a more complex picture than expected (Fig. 3F). First, the b40 and y67 ions revealed little Zn(II) binding, confirming that the major binding site is internal. The y179 fragment showed clear evidence that Zn(II) binding occurs at or near H93/95; nonetheless, the b80 and b70 fragments indicate that there may be other Zn(II) binding sites N-terminal to Asp80 (such as H63). Looking up the CA entry in Uniprot (P00921) as well as examination of the crystal structure confirms that the Zn(II) cofactor binds at H93/95 [49]. Interestingly, H63 is found in the secondary coordination sphere of the Zn(II), consistent with the fragmentation evidence indicating metal binding near that residue.

Finally, there is moderate fragmentation evidence of Zn(II) binding near H118 but the localizing y-type fragments are too low in intensity to fully differentiate this event from binding at H93/95. To manually investigate these localizing y-type fragments in a hypothesis-driven fashion, mMass produces the theoretical isotopic distributions of any desired fragment and overlays them on the raw spectrum. In contrast, if the TDValidator algorithm does not detect a target fragment due to low SNR, the software will not enable visualization of that fragment’s theoretical isotopic distribution. Hence, mMass can provide additional flexibility to examine the presence of low intensity fragments that help localize a given cofactor or metal.

H3K14ac ejected from a nucleosome

Eukaryotic chromatin is comprised of repeating structural units called nucleosomes (Nucs), which are involved in gene regulation and compaction of genetic material [54, 55]. The Nucs consist of globular protein octamers that contain two copies each of histones H2A, H2B, H3, and H4 that are wrapped around by 147 bp of DNA [55]. The N-terminal tails of the histones undergo various PTMs, which serve as epigenetic marks and drivers of critical cellular processes [56–60]. Recently, we reported the development of Nuc-MS, the three-tiered tandem MS approach based on the native top-down workflow that measures intact nucleosome particles and displays the modification status of their constituent histones in the same spectrum [61]. Here, we demonstrate how a Nuc-MS experiment can assert the positions of a specific histone PTM on synthetic Nucs acetylated by a recombinant acetyl transferase in vitro (pCAF) [62, 63].

Deconvolution results from the MS¹ analysis of synthetic Nucs yielded an intact mass (199,947.8 Da) that was within Δm = 4.1 Da (20.5 ppm) of the theoretical mass (Fig. 4A). The MS² results on the histones ejected from the intact Nuc indicated the presence of a mixture of H3.1 proteoforms that were either unmodified, acetylated and/or oxidized (Fig. 4B). To characterize the site of acetylation, the acetylated H3 was isolated from the unmodified H3 and fragmented by HCD. In order to first narrow down the sites of acetylation out of the many possible lysine residues within the sequence of H3.1 (Fig. 4C), a database of 16,384 H3.1 “candidate proteoforms” was created in Protein Annotator (Fig. S7) and subsequently used in a search of the H3-acetyl fragmentation data in ProSightPC. For the four top-scoring H3.1 proteoforms, the mass accuracy observed between the theoretical intact mass and the observed precursor mass (−0.21 ppm) supports a single acetylation event on the target H3.1 proteoform. Even though the top-scoring proteoforms indicate that the acetylation was localized near the N-terminus, the C-score = 0 highlights that there was insufficient fragmentation data from the Xtract deconvolution process to differentiate the four possible acetylation sites (Fig. 4D).

Figure 4. The new process of ‘Nuc-MS’ uses nTDMS to measure Histone H3 acetylated at K14 ejected from an intact nucleosome (Nuc) [53].

(A) Mass spectrum of Nuc complex containing up to 2 acetylations and a smaller Nuc species denoted by an asterisk that contains a shorter DNA strand. (B) Deconvoluted mass spectrum of ejected histone H3 and acetylated proteoform. (C) A sequence map of histone H3 with all theoretical sites of acetylation highlighted. The map serves as an input for Protein Annotator which results in a database of 16,384 differentially acetylated proteoforms of H3. (D) ProSightPC output demonstrating automated localization using curated H3 proteoform database of acetylation event near N-terminus (aa 1–17). (E) TDValidator helps to localize acetylation to K14. (F) Selected fragments are shown with theoretical isotopic distributions (red triangles) demonstrating confident localization. (G) Examples of fragments not detected by TDValidator but manually identified using mMass.

To overcome this challenge, a tool like TDValidator can be used to search for additional low intensity fragments that help to successfully localize the acetylation on K14 (Fig. 4E). Specifically, by analyzing the raw data directly we successfully assigned seven low SNR fragments that were not picked up by the Xtract algorithm (b7 through b13), four of which clearly distinguished the acetylation position at K14 (Fig. S3, Fig. 4F). To further bolster confidence in the localization, mMass was used to manually validate the fragmentation found by TDValidator, and to search for additional fragmentation not detected by TDValidator in a “hypothesis-driven fashion” (Fig. 4G). Through the combination of various tools, we are able to fully characterize the acetylated histone proteoform as H3.1K14ac, consistent with the enzymatic activity of pCAF, and can assert that no other positional isomer exists at >5% abundance [53].

Amyloid-Beta

nTDMS is emerging as a powerful approach to characterize the presence of covalent PTMs and non-covalent ligands on amyloidogenic proteins, such as the 42 amino acid peptide amyloid beta (Aβ_1–42) [64]. Aβ_1–42 is an intrinsically disordered protein commonly found in the brain of Alzheimer’s disease patients where it has propensity to form aggregates thought to disrupt synaptic pathways [65–67]. Aggregation of the peptide follows the formation of oligomeric species and structurally rich β-sheet fibrils, the latter of which will continue to develop into thick plaques [68–70]. In recent years, native ion mobility MS has been used to investigate the amyloidogenic assemblies of synthetic Aβ_1–42 and its fragments to help gain an understanding of the misfolding pathway [71–77]. Interestingly, emerging clinical proteomics data suggests that canonical Aβ_1–42 represents only one of many Aβ proteoforms in the AD brain that have been linked to toxic functions (e.g., pyroglutamylation, di-tyrosine crosslinking, or modification by pleiotropic effectors) [78, 79].

In addition to PTMs, the Aβ landscape in vivo is further muddied through non-covalent interactions between Aβ proteoforms (i.e., aggregation) and non-covalent cofactors that can contribute to formation of toxic reactive oxygen species targeting Aβ [80]. In the example shown here, native-mode ESI of a synthetic, monomeric Aβ_1–42 peptide revealed the presence of multiple proteoforms, with the two dominant species each exhibiting at least two putative oxidations events (+16 and +32 Da) (Fig. 5A, inset). Fragmentation data from the most abundant Aβ proteoform (m/z = 1128.8) followed by sequence interrogation with ProSight Lite confidently identified Aβ_1–42 with 29 matched b/y-ions and 70% sequence coverage (P-score = 2.4x10⁻⁴⁶). Similar analysis of the tandem MS dataset on the +16 Da proteoform (m/z = 1132.8) also showed that Aβ_1–42 was oxidized at M35 (P-score = 4.4x10⁻³⁹, Fig. 5A–C). In this case, overlay of theoretical and observed isotopologues in mMass for the fragments y8 (1+), and y6 (1+) confirms the localization of the oxidation at M35 (Fig. 5F, see NOTE 3). Interestingly, the presence of methionine sulphone was not detected in the fragmentation data, indicating that the +32 Da proteoform (m/z = 1136.8) derives from a more complicated mixture of oxidation positional isomers. To confirm this, we searched for other positional isomers in ProSight Lite through a termini Delta mass test by manually adding +16 Da to both the N- and C-termini. This test on the tandem MS dataset for the mono-oxidized proteoform revealed oxidation on 11 diagnostic b-ions ending at F19/20 (Fig. 5C). Again, the results were manually confirmed via mMass and TDValidator, including the highly diagnostic fragments b21 and b26 (Fig. 5F). While the mechanism of Aβ oxidation could be attributed to sample handling, nTDMS may provide an avenue to establish evidence for further hypothesis testing related to biologically meaningful oxidation (e.g., metal-induced oxidation) targeting Aβ [80]. For example, many of the remaining Aβ proteoforms detected in the intact spectrum indicated that the proteoforms Aβ_1–42, Aβ_1–42ox, Aβ_1–42ox(x2) each had a +60.93 Da shift. This Δm was found to correspond to a Cu(II) (monoisotopic mass: 60.92 Da; average mass: 61.5 Da) —a known Aβ ligand in vivo and in vitro [81]. A Delta mass test in TDValidator confirmed the presence of the cofactor near the N-terminus which harbors histidine residues at positions 6, 13, 14. Iterative hypothesis testing – i.e., examination of each residue in turn for copper binding – localized the cofactor to H13/14, a known binding site for Cu(II) on Aβ [81] (Fig. 5D). To bolster confidence in this localization, b13, b14, b21 and b27 bound-fragments were manually validated using TDValidator (Fig. 5E, see NOTE 4).

Figure 5.

nTDMS analysis of Aβ_1-42. (A) Mass spectrum of native Aβ_1-42. Highlighted are masses corresponding to Aβ 1-42 (yellow), Aβ_1-42+Ox (blue) and Aβ_1-42+Cu (green). (B) MS/MS spectra of the corresponding selected m/z. (C) ProSight Lite fragment maps of base Aβ_1-42 (1128.8 m/z) and oxidized Aβ_1-42 (1132.8 m/z) with the oxidation placed at M35 or F19. (D) Δmass of theoretical and observed b ions of Aβ_1-42+Cu (II). ProSight Lite graphical map of fragments relating to Aβ_1-42+Cu(II) and Aβ_1-42+Cu(I) are overlaid. Fragments seen only for Aβ_1-42+Cu(I) are indicated (*). (E) TDValidator detection of b13 (2+), b14 (2+), b21 (2+), and b27 (2+). Monoisotopic peaks relating to bound Cu(II) (green triangles) and Cu(I) (purple squares) are highlighted. (F) Fragment masses not found through ProSight Lite and TDValidator were manually searched through mMass. Monoisotopic peaks relating to Aβ_1-42 oxidized at M35 (blue circles) and oxidized at F19 (red stars) are highlighted.

Conclusion

We offer a framework and some tips for analyzing and reporting fragmentation data that result from nTDMS analysis on proteins and their complexes. The landscape of strategies and software tools for fragment analysis is evolving and complex, with no apparent one-size-fits-all, fully automated approach currently available. Our hope is that the decision tree, the example datasets and publicly available raw data in this tutorial will provide a baseline of guidance for new and advanced researchers endeavoring to fragment and characterize native proteoforms that may be relevant in broader structural biology investigations of composition, stoichiometry, assembly, interactions, and topology of proteins and their complexes.

Methods

Carbonic Anhydrase experiments

The specific MS experimental conditions and analytical workflows for MS¹ and MS² data on CA can be found in our recent protocol for performing native mass spectrometry and nTDMS experiments [14]. High-resolution fragmentation data were processed using Xtract (Signal-to-Noise threshold ranging from 1-30, Thermo Fisher Scientific), mMass 5.5.0 (www.mmass.org), ProSight Lite 1.4 [25] (precursor mass type: average; fragmentation method: HCD; fragmentation tolerance: 10-15 ppm), and TDValidator 1.0 [82] (Proteinaceous, max ppm tolerance: 25 ppm; cluster tolerance: 0.35; charge range: 1-10; minimum score: 0.5; SNR cutoff: 3; BRAIN algorithm; minimum size: 2) to assign recorded fragment ions to the primary sequence of the protein.

Histone experiments

The nucleosome synthesis and modification protocol, the specific MS experimental conditions, and analytical workflows for MS¹ and MS² data, can be found in the Nuc-MS publication [53]. High-resolution fragmentation data were processed using Xtract (Signal-to-Noise threshold ranging from 1-30, Thermo Fisher Scientific), mMass 5.5.0 (www.mmass.org), ProSight Lite 1.4 [25] (precursor mass type: average; fragmentation method: HCD; fragmentation tolerance: 10-15 ppm), and TDValidator 1.0 [82] (Proteinaceous, max ppm tolerance: 25 ppm; cluster tolerance: 0.35; charge range: 1-10; minimum score: 0.5; SNR cutoff: 3; BRAIN algorithm; minimum size: 2) to assign recorded fragment ions to the primary sequence of the subunits.

Amyloid-beta experiments

Aβ1-42 (California Peptide) was resuspended in dymethylsulfoxide, diluted to 100 µM in phosphate-buffered saline and then exchanged into 150 ammonium acetate at a final concentration of 10 µM. The sample was directly infused into a Q Exactive-HF Extended Mass Range (Thermo Fisher Scientific) with a nano ESI source [9]. All resulting spectra were deconvoluted with Xtract (S/N:3, Fit factor:45, Remainder:80) in Xcalibur (Thermo Fisher Scientific) followed by a manual calibration (12 ppm). Calibrated masses were input into ProSight Lite and TDValidator (Proteinaceous, max ppm tolerance: 10 ppm; cluster tolerance: 0.35; charge range: 1-5; minimum score: 0.5; SNR cutoff: 3; BRAIN algorithm; minimum size: 2), with an average mass error of 0.97±1.496 Da.

MS Data Availability

Fragmentation spectra presented in the manuscript are available online in the MassIVE database under accession code MSV000086404 as .RAW files and can be visualized with Thermo Qual Browser.

Data Acquisition

Data were collected on either a custom Thermo Fisher Q Exactive Orbitrap HF MS with Extended Mass Range (QE-EMR) [8] or a commercially available Thermo Fisher Q Exactive Orbitrap MS with Ultra High Mass Range (UHMR). Critical details of nTDMS data collection were previously outlined in our manuscript on the development of standardized best-practices for collection of nTDMS data. This standard operating procedure included extensive discussion on recommended reagents, materials, suggested values for key instrument tuning and data acquisition parameters, and step by step instructions on the collection of MS¹, MS², and MS³ data [14].