Complete Post-Translational Modification Mapping of Pathogenic N. meningitidis Pilins Requires Top-Down Mass Spectrometry

Abstract

In pathogenic bacteria post-translationally modified proteins have been found to promote bacterial survival, replication and evasion from the host immune system. In the human pathogen Neisseria meningitidis, the protein PilE (15–18 kDa) is the major building block of type IV pili, extracellular filamentous organelles that play a major role in mediating pathogenesis. Previous reports have shown that PilE can be expressed as a number of different proteoforms, each harbouring its own set of post-translational modifications (PTMs) and that specific proteoforms are key in promoting bacterial virulence. Efficient tools that allow complete PTM mapping of proteins involved in bacterial infection are therefore strongly needed. As we show in this study, a simple combination of mass profiling and bottom-up proteomics is fundamentally unable to achieve this goal when more than two proteoforms are present simultaneously. In a N. meningitidis strain isolated from a patient with meningitis, mass profiling revealed the presence of four major proteoforms of PilE, in a 1:1:1:1 ratio. Due to the complexity of the sample, a top-down approach was required to achieve complete PTM mapping for all four proteoforms, highlighting an unprecedented extent of glycosylation. Top-down mass spectrometry therefore appears to be a promising tool for the analysis of highly post-translationally modified proteins involved in bacterial virulence.

Introduction

Post-translational modification (PTM) increases the functional diversity of proteins by covalent addition of functional groups, modification of amino acid side chains and proteolysis. PTMs are implicated in almost all aspects of normal cell biology and pathogenesis. In viral or bacterial infection, pathogens often use PTMs to manipulate pathways in the host cell in order to promote their own survival, replication and evasion from the host immune system [1, 2]. For a long time PTM was considered an exclusively eukaryotic process but it is now widely accepted to also occur in bacteria and archaea. Recent evidence supports the hypothesis that acetylation broadly impacts bacterial physiology [3]. Highly phosphorylated bacterial proteins have been described as being potential intermediates of degradative pathways [4, 5] and sulfated proteins have been shown to trigger the host immune system and bacterial cell-cell communication [6]. Bacterial surface structures such as flagella (Pseudomonas aeruginosa and Campylobacter jejuni) and pili (Neisseria spp. and P. aeruginosa) have been found to be particularly rich in post-translationally modified proteins [7]. Indeed studies on these organelles have led to the description of several complete microbial glycosylation models [8]. As many of the proposed bacterial glycoproteins are surface-exposed, these modified proteins have been postulated to play important roles in pathogenicity and antigenicity.

Type IV pili (T4P) of pathogenic Neisseria are hair-like structures that protrude from the bacterial surface and are implicated in a wide variety of processes including bacterial motility and DNA uptake [9]. Since T4P are also required for host-cell adhesion, and thus play a crucial role in colonisation of the host, they are considered as a major bacterial virulence factor. T4P are protein macropolymers predominantly composed of a single protein subunit, the major pilin. This pilin protein is arranged in a helical fashion to create the long and flexible pilus fibre. In Neisseria the major pilin is the protein PilE, which is highly post-translationally modified. It is always N-terminally processed and methylated and carries a pair of oxidised cysteines close to the C-terminus. It is glycosylated by the unusual glycan 2,4-diacetamido 2,4,6-trideoxy α-D-hexose (DATDH) [10] or 2-acetamido 4-glyceramido 2,4,6-trideoxy-α-D-hexose (GATDH) [11], which can be further elaborated by up to two galactose or glucose subunits and may be O-acetylated. In addition, this protein may also harbour a number of phosphoforms such as phosphate (P) [12], phosphoethanolamine (PE) [13, 14], phosphocholine (PC) [13–15] and phosphoglycerol (PG) [16].

PilE has been reported to be concurrently expressed as number of different proteoforms each carrying an array of different PTMs. (The term proteoform has recently been proposed to describe the different molecular forms in which the protein product of a single gene can be found, including changes due to genetic variations, alternatively spliced RNA transcripts and post-translational modification) [17]. This was first suggested from X-ray crystallography data that showed a weak electron density for phosphate around Ser⁹⁴ in N. gonorrhoeae strain MS11, in addition to the phosphate on Ser⁶⁸ [12]. This indicated that a small proportion of the PilE population harboured an additional phosphate group. More recently proteoforms of PilE from both Neisseria meningitidis and Neisseria gonorrhoeae have been evidenced by mass spectrometry in intact mass profiling experiments. In N. gonorrhoeae, pilins modified with both PE and PC have been reported and the ratio of the different phosphoforms expressed has been shown to be dependent on the presence of the glycan and of the minor pilin PilV [13]. In N. meningitidis, proteoform variation also centres around the phosphoform [18]. Mass profiling of the 8013 strain has shown PilE to be expressed as two proteoforms in a 4:1 major/minor ratio [11]. The minor proteoform carries an extra PG on Ser⁹³. In this strain PG has been found to be a regulated PTM that mediates bacterial pathogenesis. Increased modification of PilE with PG at Ser⁹³, several hours after host cell contact, changes the electrostatic surface of the pilus fibre and disrupts pilus/pilus interactions. This is a prerequisite for crossing the epithelial layer and a key step in pathogenesis [19]. The extent and variation of pilin PTMs in highly pathogenic N. meningitidis strains is therefore of particular interest, as is understanding their role in bacterial virulence [20].

The tools available at present to achieve complete PTM mapping of all expressed proteoforms of a protein of interest are extremely limited. Mass spectrometry-based proteomics has proven to be powerful approach for the identification of individual PTMs, leading in some case to the global identification of hundreds to thousands of PTMs within a sample [21]. However peptide-based, bottom-up proteomics fails to provide a complete picture of PTM since the connectivity between peptides and their parent proteoforms is lost. This is particularly important when two or more modifications work together on a single protein. Moreover, bottom-up strategies do not provide proteoform level information i.e. the explicit identities of the proteoforms present in the sample and their relative abundance, which is crucial information to understand proteoform function and in vivo regulation. Top-down mass spectrometry, which is based on the analysis of intact proteins, preserves the proteoforms and thus facilitates their full characterisation including PTMs [22, 23]. Top-down approaches have been successfully applied to the characterisation of various protein PTMs [23–25] and recent improvements in intact protein chromatographic separations and high performance Fourier transform mass spectrometry (FTMS) instrumentation have greatly expanded the observable range of proteoforms in complex samples. However the analytical requirements of top-down MS remain particularly challenging due to the size of the systems under study (proteins and not peptides). Top-down MS is currently limited to high-resolution instrumental platforms (FT-ICR, Orbitrap or high resolution ToF instruments) and separation of intact proteins can be a difficult task, as can efficient protein fragmentation on an LC time scale. Finally the software options available for top-down MS data analysis are rather limited, although recent developments in the field are helping to overcome this [26, 27].

To date the PTM complement of PilE has only been investigated in two different strains of N. meningitidis (8013 and C311) [10, 11, 16, 28] and the diversity of PTMs present in the population of N. meningitidis strains found in human patients remains unknown. Because of the link between PTMs and virulence, the analysis of novel strains is of great interest, but it is limited by the lack of approaches that are sufficiently rapid and amenable to high throughput analysis. To determine the feasibility of typing clinical strains in terms of pilin PTMs, we isolated a strain from a patient hospitalised with meningitis and characterised its pilin PTMs by different approaches. Mass profiling revealed the presence of four major proteoforms of PilE, in a 1:1:1:1 ratio. Initially, a bottom-up strategy, which had proven to be efficient for the analysis of the other strains was used, but it led to an incomplete PTM mapping. We therefore employed top-down mass spectrometry to select and fragment the intact proteoforms individually. Strengths and weaknesses of both approaches for the analysis of these challenging proteins will be discussed. We will show that when more than two proteoforms of a single protein are present, bottom-up alone is fundamentally unable to map all PTMs and top-down is required to achieve this goal. Top-down mass spectrometry therefore appears a promising tool for the analysis of highly post-translationally modified proteins involved in bacterial virulence.

Materials and Methods

Bacterial strain

The strain of Neisseria meningitidis used in this study was isolated from a patient with meningitis at the Limoges hospital in the Haute Vienne County of France in 2006 (strain number 278534). It is a serogroup A capsular serotype and multilocus sequence typing revealed that it is part of the ST-5 complex/subgroup III clonal complex. N. meningitidis was grown on solid GCB Agar (Laboratorios Conda, Spain) containing Kellogg’s supplements [29]. The major pilin gene (pilE) was PCR amplified and sequenced using oligos NG1705 (GTCAAACCCGGTCATTGTCC) and NG1706 (CAGGAGTCATCCAAATGAAAGC) [30].

PilE Preparation

Pili were prepared as described previously [31]. Briefly, the content of 10–12 Petri dishes was harvested in 5 mL of 150 mM ethanolamine at pH 10.5. Pili were sheared by vortexing for 1 min. Bacteria were centrifuged at 4,000 x g, 30 min, 4 °C and the resulting supernatant further centrifuged at 15,000 x g, 30 min, ambient temperature. The supernatant was removed, pili precipitated by the addition of 10 % vol. ammonium sulfate saturated in 150 mM ethanolamine pH 10.5 and allowed to stand for 1 h. The precipitate was pelleted by centrifugation at 4,000 x g, 1 h, 20 °C. Pellets were washed twice with PBS and suspended in 100 μL distilled water.

Bottom-up Mass Spectrometry

Ten μL of crude PilE preparation was suspended in Laemmli buffer and separated by SDS-PAGE. After removal of the Bio-Safe Coomassie stain (Biorad) used for visualisation, the band corresponding to PilE was excised and digested in-gel as described elsewhere [32]. Briefly gel pieces were reduced, alkylated and digested overnight with trypsin (Promega) at 37 °C. After desalting by C₁₈ Ziptip® samples were eluted into 10 μL spray solution of ACN:H₂O:HCOOH (50:50:0.1) for mass spectrometry. The resulting digest fragments were examined in positive ion mode by direct infusion nano-ESI, using a TriVersa Nanomate (Advion Biosciences, Harlow, UK) on an Orbitrap Velos mass spectrometer, equipped with an ETD module (Thermo Fisher Scientific, Bremen, Germany). A full set of automated positive ion calibrations was performed immediately prior to mass measurement, as were the calibrations for reagent ion transfer. All spectra were acquired in full profile mode. For MS experiments ions were accumulated in the ion trap and then transferred to the Orbitrap for high resolution mass measurement. For MS/MS experiments, ions were selected with an appropriate mass window and HCD was performed at normalised collision energies of 15–25 %, with other activation parameters left as default. For ETD the reagent gas was fluoranthene and the interaction time tuned to maximise sequence coverage. Supplemental activation was also applied. The FT automatic gain control (AGC) was set at 1×10⁶ for MS and 2×10⁵ for MSⁿ experiments. Spectra were acquired in the FTMS over several minutes with between one and three microscans and a resolution of 60,000 for MS and 30,000 @ m/z 400 for MS/MS before being processed with Thermo Xcalibur 2.2. Peak picking was performed manually for all spectra and fragmentation maps were generated using a home built package.

Top-Down Mass Spectrometry

Top-Down experiments were performed on a solariX 12T hybrid Qh-FT-ICR (Bruker Daltonics, Billerica, MA) equipped with a hollow dispenser cathode. Crude protein extracts were desalted by C₄ Ziptip® (Millipore) and eluted into 10 μL electrospray solution MeOH:H₂O:HCOOH (75:25:3). Protein was introduced into the mass spectrometer through pulled borosilicate capillaries or by using a TriVersa NanoMate® (Advion Ithaca, NY). The NanoMate was the injection method of choice over pulled capillaries with wider tips, even for samples prone to aggregation and needle (nozzle) clogging. For mass profiling experiments ions were accumulated in the hexapole for 0.1–1 s before being transferred to the Infinity^™ cell for detection. Spectra were accumulated for 50–200 scans. For MS/MS spectra, since the concentration of sample, and therefore single scan intensity, was highly preparation-dependent, ions were accumulated for ≤ 4s in the hexapole in order to reach a threshold precursor ion intensity. Ions were then transferred to the Infinity^™ cell where ECD was performed with a pulse length of 5–10 ms and electron energy of 1.0–1.7 eV. For each experiment, 300–450 scans were accumulated. The number of data points (1 Mega points) was chosen to have near baseline resolution without detrimentally decreasing scan speed. Calibration was performed monthly with clusters of NaI.

Top-Down Data Analysis

Data processing was performed using DataAnalysis 4.0 SP5 (Bruker Daltonics, Billerica, MA). For MS experiments spectra were deconvoluted using the maximum entropy option. This gives much cleaner deconvoluted spectra than the other options available. For MS/MS experiments, acquired spectra were internally calibrated and peak picked using the SNAP 2.0 algorithm; quality factor 0.1, S/N 2, relative intensity 1×10⁻⁵ (%), absolute intensity 0 and a maximum charge state altered to just above the maximum observed charge state. Peak picking results were saved as an XML file and peak assignment was performed by importing this data into a home built package for ion assignment and automated fragmentation map creation. PTM assignment was performed manually with this software on combined peak lists from the 14+ and 15+ charge states.

Results and Discussion

Mass profiling

Purification of PilE from the 278534 strain produced a large amount of protein in high purity (see SDS-PAGE in Suppl. Fig. 1). When measured by FT-ICR MS the crude sample gave a complex MS spectrum exhibiting several different charge state envelopes. Deconvolution of the raw data gave four major peaks with monoisotopic neutral molecular masses (M_r) of 15,146.7058, 15,374.8325, 15,602.9369 and 15,831.0584 in an approximate 1:1:1:1 ratio (Figure 1A).

Figure 1.

To provide a reference point for further investigation the pilE gene from the N. meningitidis 278534 strain was sequenced. Surface expressed PilE is known to be post-translationally processed by the endoprotease pilD which cleaves a short N-terminal leader sequence or prepilin before a conserved phenylalanine residue [33]. Making this modification to the initial 147 amino acid sequence furnished a 140 amino acid protein with the theoretical M_r of 14,524.47 (sequence depicted in Figure 1B). Even when compared to the lowest mass major peak observed in the MS profile this represents a difference of over 620 Da and indicated that PilE from this particular clinical strain could be highly post-translationally modified.

Bottom-up analysis

A MS strategy based on the combination of accurate high resolution intact mass measurement of proteoforms and tandem mass spectrometry experiments on peptides (bottom-up approach) had previously proven useful in identifying PTMs on PilE from N. meningitidis strain 8013 [19, 34]. Therefore a similar approach was initially employed here. A sample of PilE was subjected to SDS-PAGE followed by in-gel tryptic digestion. The digest was then analysed by nanoESI-FTMS on an Orbitrap mass spectrometer (Figure 2).

Figure 2.

Comparison of the experimental masses measured for this tryptic digest and theoretical ones calculated in silico from the PilE sequence revealed the presence of non-modified (naked) peptides spanning the ranges [31–59], [76–112] and [122–138]. A [1–30]+14.016 Da peptide was also observed, confirming the N-terminal methylation of PilE and leading to a sequence coverage of over 80% (Table 1).

Table 1.

Digest product ions of PilE from 278534 strain (from Figure 2).

Digestion Product	Measured m/z	Charge	Measured Monoisotopic Mass [M+H]+	Theoretical Monoisotopic Mass [M+H]+	Error (ppm)
[1–30]+Me	1088.2893	3	3262.8533	3262.8524	0.290
[31–44]	492.9174	3	1476.7376	1476.7363	0.912
[31–44]	738.8725	2	1476.7377	1476.7363	0.964
[31–44]	1476.7381	1	1476.7381	1476.7363	1.219
[31–44]Ox	746.8698	2	1492.7323	1492.7312	0.753
[31–44]Ox	1492.7366	1	1492.7366	1492.7312	3.618
[45–59]	844.4092	2	1687.8111	1687.8326	12.724*
[60–75]+DATDH+PG	928.9366	2	1856.8659	1856.8637	1.197
[60–75]+2DATDH+PG	1042.9924	2	2084.9775	2084.9747	1.354
[76–92]Ox	915.4706	2	1829.9339	1829.9313	1.434
[80–92]	676.8405	2	1352.6737	1352.6726	0.831
[80–92]+DATDH	790.8960	2	1580.7847	1580.7836	0.711
[93–112]+DATDH	739.7317	3	2217.1805	2216.1769	†
[93–112]+DATDH	1108.5933	2	2216.1793	2216.1769	1.094
[99–112]	701.3835	2	1401.7597	1401.7584	0.944
[99–112]	1401.7600	1	1401.7600	1401.7584	1.141
[99–112]+DATDH	815.4392	2	1629.8711	1629.8694	1.057
[113–121]+DATDH	651.2957	2	1301.5841	1301.5831	0.786
[113–121]+DATDH	1301.5850	1	1301.5850	1301.5831	1.460
[122–138]	604.2972	3	1810.8770	1809.8912	-*
[130–138]	555.2745	2	1109.5417	1109.5408	0.832

^*Errors for these ions are greater than expected since these digest products contain asparagine vicinal to glycine residues and are therefore subject to facile deamidation. For the ion at m/z 844.4092 deamidation is incomplete but the monoisotopic peak is very close to the baseline. For the ion at m/z 604.2972 complete deamidation increases the observed mass by approximately 1 Da.

^†The monoisotopic peak for this ion is obscured by the doubly charged [31–44] at m/z 738.8725.

All peptide ions were isolated and subjected to Higher energy Collision Dissociation (HCD) in order to confirm their identity (data not shown). Despite the high sequence coverage obtained with these ions, numerous abundant, multiply charged peaks present in the MS spectrum could not be attributed. In addition, no peptide spanning the [60–75] region could be assigned. Since this region of PilE is almost always post-translationally modified, its absence encouraged us to investigate these multiply charged, non-assigned peaks in the hunt for additional PTM bearing peptides.

When ions observed at m/z 790.8960, 815.4392, 651.2957 and 739.7317 were subjected to HCD, three very abundant fragment ions appeared at m/z 229.118, 211.108 and 169.097 in the resulting MS/MS spectra (Figure 3 A, C, E, G).

Figure 3.

These ions correspond to the oxonium ion of the DATDH glycan, its dehydrated partner and a fragment ion characteristic of the glycan core and indicate that all fragmented peptides contain this sugar moiety. DATDH is a previously described PTM for PilE and these reporter ions have previously been used to identify glycosylated peptides [35, 36]. The analysis of the other fragment ions in the spectra (y/b ions) confirmed the sequence for the four peptides ions as [80–92], [99–112], [113–121] and [93–112] respectively, but could not be used to localise the sites of glycosylation. All four precursor ions were therefore subjected to electron transfer dissociation (ETD) in order to confidently localise the glycosylation sites (Figure 3 B, D, F, H). For the [80–92] peptide, the glycan could be easily localised on Ser⁸³, for [99–112] and [93–112] it was localised exclusively on Ser¹⁰¹ and finally for [113–121], Ser¹¹³ was found to be modified with DATDH. In none of the ETD spectra was any trace of the reporter ions detected at m/z 229.118, 211.108 or 169.097 (see Supplementary Tables for all MS/MS data).

In addition, when fragmented by HCD, two other doubly charged ions, at m/z 928.934 and 1042.999, produced intense reporter ion signals for DATDH. In both cases the [60–75] tryptic peptide was identified suggesting multiple forms to be present, each modified by DATDH. ETD data identified a DATDH on Ser⁶³ for the precursor m/z 928.934, on both Ser⁶³ and Ser⁶⁸ for m/z 1042.99 and showed that both forms were further modified by PG on Ser⁷⁰ (Figure 3 I, J).

Upon inclusion of these modified peptides, the sequence coverage from the digest was extended to 98 % with only the two last amino acids of the sequence, AK, unaccounted for. These results were corroborated but not improved upon using a LC-MS approach. Interestingly, the presence of several peptides in both non-modified and modified forms confirmed that PilE was expressed as multiple proteoforms. Furthermore it appeared that these proteoforms differ in the number of DATDH glycan units. This hypothesis correlated perfectly with the pattern observed by mass profiling, where the mass difference between the four major peaks is 228.11 Da - the mass expected from addition of DATDH. PilE thus appeared to be present in multiple glycoforms each bearing a different number of DATDH subunits.

Armed with mass profiling data and the results from the bottom-up experiments, one can begin to assign peptide combinations, and thus PTMs, to specific proteoforms. Since all peptides are found in non-modified forms apart from [113–121] and [60–75], the lowest mass forms of these two peptides [113–121]+DATDH and [60–75]+PG+DATDH plus the exclusively non-modified peptides, may naturally be attributed to the lowest mass proteoform. This assignment seems satisfactory since it results in the correct 15,146.7058 Da protein mass with an experimental-theoretical mass error of only 0.03 ppm. The combination of peptides giving the protein with the heaviest mass; [60–75]+PG+2DATDH, [80–92]+DATDH, [99–112]+DATDH, [113–121]+DATDH could similarly be assigned to the highest mass proteoform. This also seems correct and is the only peptide combination possible to reach a total mass of 15,831.0584 Da. However a problem arises when considering the two intermediate proteoforms, since various combinations of peptides are possible to achieve the observed protein masses. Here, even with the mass profile as a reference, a bottom-up methodology is intrinsically unable to relate modified peptides to their parent proteoforms without making several important assumptions. One may think that examining the ion intensities of modified peptides and relating them to abundances may help solve this problem, but this is approach is unrealistic since PTM has been well documented to drastically affect peptide ionisation efficiency [37].

To achieve complete proteoform mapping, each proteoform must be investigated separately. Off-line fractionation methods require prior knowledge of the PTMs present on each proteoforms and are often difficult to implement at the protein level. A top-down MS approach allows each proteoform to be easily isolated and fragmented separately in the mass spectrometer. This top-down strategy was therefore applied to PilE. We initially sought to use the Orbitrap Velos and ETD fragmentation for this purpose but rapidly realised that in our hands this instrument was not optimised for the analysis of intact proteins and that it would not give the required sequence coverage. Therefore a 12T FT-ICR mass spectrometer was chosen for all top-down experiments, using ECD (electron capture dissociation) as the fragmentation technique.

Top-down analysis

PilE from N. meningitidis 8013 strain has previously been investigated by top-down ECD MS/MS using an FT-ICR mass spectrometer (unpublished data) and experimental parameters optimised in that study were used as a starting point here (irradiation time, energy of the electrons etc.). The PilE sample examined in this study did however exhibit some differences compared to that purified from the 8013 reference strain (PilE-8013). Clogging of nanoelectrospray needles during spectral acquisition was much more acute; despite centrifugation prior to sample injection and the use of pulled capillaries with a wider tip. A more stable spray was obtained using a Triversa NanoMate and most importantly this injection method allowed the spray signal to be monitored during spectral acquisition and the nozzle to be quickly changed if necessary. The charge state envelope was also different for the 278534 strain. It was shifted to lower charge states, with the maximum observed here 16+ compared to 19+ for the 8013 strain. This is possibly due to the difference in protein size or the different nature and number of PTMs carried by the two proteins. The envelope was also much more complex, the presence of four major proteoforms causing some proteoforms of one charge state to overlap with other proteoforms from another. Proteoforms of the highest exploitable charge states (15+ and 14+) were isolated in the hexapole of the 12T FT-ICR mass spectrometer and single charge states were subjected to top-down ECD MS/MS.

In general the 15+ ions afforded greater sequence coverage for the same precursor intensity but the 14+ charge state was more abundant and, for some proteoforms, furnished even greater sequence coverage when precursor ions were allowed to accumulate. Extensive fragmentation was observed in the top-down spectra especially in the case of the highest mass proteoform (15,831.0584 Da) (Figure 4D). Cleavage maps are also shown for the other three proteoforms (Figure 4 A–C). Interestingly the intact cysteine bridge between Cys¹²⁰-Cys¹³⁷ appeared to strongly inhibit fragmentation in this region for the 14+ charge state but not for the 15+ charge state.

Figure 4.

For proteoform 1, fairly extensive N- and C-terminal fragmentation enabled straightforward identification of the methylated N-terminus and a DATDH glycan on Ser¹¹³. The c₆₂ ion at m/z 1115.4284 (6+), 1339.3085 (5+), c₆₆ ion at m/z 1205.8037 (6⁺), 1446.7672 (5⁺) and c₆₇ ion at m/z 1217.6449 (6⁺) enabled localisation of a second DATDH on Ser⁶³. The presence of the same c ions in spectra obtained from both the 14+ and 15+ protein forms increases confidence in this assignment, as does the presence of multiple charge states for c₆₆ in each spectrum. In addition the C-terminal fragment ions z₇₀, z₇₄ and y₇₃ enabled identification of a phosphoglycerol group on Ser⁶⁸ or Ser⁷⁰ but the absence of ions between these residues precluded definitive localisation. This was however provided by the bottom-up data that showed Ser⁷⁰ to be exclusively modified with phosphoglycerol. For proteoform 1 this gave a PTM complement of two DATDH glycans at Ser⁶³ and Ser¹¹³ and a phosphoglycerol group at Ser⁷⁰ in addition to the expected N-terminal methylation and cysteine bond. The theoretical protein mass for this assignment of 15,146.7017 Da correlated exceptionally well with the 15,146.7058 Da experimental value, giving a + 0.03 ppm error.

For proteoform 2, two DATDH subunits were also easily identified on Ser⁶³ and Ser¹¹³; the former by the c₆₀ and c₆₆ ions and the latter by a large series of z ions from z₂₆ to z₅₂. Despite poor fragmentation in the central region of the protein, the sequence coverage was sufficient to indicate that no PTM was present between residues 71 and 112. The difference in mass between z₇₀ and z₇₄ (698.2519 Da) indicated the presence of both a PG and a DATDH between Ala⁶⁷ and Ser⁷⁰ but as for proteoform 1 the absence of fragments between these residues prevented definitive location. Taking into account the bottom-up data a PG group was assigned on Ser⁷⁰ and therefore a DATDH assigned to Ser⁶⁸. Again the measured mass of 15,374.8325 Da corresponded excellently to the theoretical mass for this PTM assignment (15,374.8122 Da) with an error of +1.3 ppm. In comparison to proteoform 1, proteoform 2 exhibited an additional DATDH subunit at Ser⁶⁸.

For proteoform 3, a better sequence coverage in the regions of interest allowed an easy assignment of the four DATDH groups to Ser⁶³, Ser⁸³, Ser¹⁰¹ and Ser¹¹³. Again a PG could be identified either on Ser⁶⁸ or Ser⁷⁰ and bottom-up data were used to confirm the latter position. As the number of modifications increases so does the number of possibilities for site localisation. Deciding upon the correct PTM assignment therefore becomes increasingly difficult, especially when potential modification sites are close together in the protein sequence. In the case of proteoform 3, the large series of z type ions from z₅₅ to z₇₀ allowed confident assignment of DATDH on Ser⁸³ rather than Ser⁶⁸. Our results indicated that, in this proteoform, two previously unmodified serines are now glycosylated and one which was occupied in proteoform 2 is now non-modified. Similarly to proteoforms 1 and 2, the measured mass of 15,602.9369 Da corresponds excellently to the theoretical mass for this PTM assignment (15,602.9226 Da) with an error of +0.9 ppm.

Finally, for the highest mass proteoform (proteoform 4) the assignment of four DATDH was found to be similar to proteoform 3 and an additional glycan was localised on Ser⁶⁸, as already observed for proteoform 2. This proteoform led to the highest sequence coverage in ECD (62%) and the presence of the c₆₉, c₇₀ and complementary z₇₀ ions mean that for this proteoform no bottom-up data is required to assign PG to Ser⁷⁰. The measured mass of 15,831.0584 Da again corresponds very well to the theoretical mass (15,831.0331 Da) with an error of +1.5 ppm.

These results indicated that in most cases spectra were of sufficient quality and fragmentation sufficiently extensive to allow PTMs to be unambiguously localised on the protein backbone. The only exception was for the location of the PG group which needed information from bottom-up experiments to provide the exact modification site for three of the four proteoforms. This result may be explained by a lower efficiency of electron capture at sites close to PTMs, or even capture of the electron by the modifications themselves. This would likely modify the overall reactivity of the ECD process.

Taken in the context of the mass profiling experiment, the correlation between experimental and theoretical masses for all PTM assignments is excellent, with errors at the protein level consistently below 2 ppm. The complement of post-translational modifications has been explicitly defined for each proteoform and the bottom-up and top-down data are in perfect agreement. All proteoforms are modified with a PG at Ser⁷⁰ and DATDH at Ser⁶³ and Ser¹¹³. The extra sites of glycosylation are Ser⁶⁸, Ser⁸³ and Ser¹⁰¹ but our results also revealed that the glycosylation process does not appear to be completely successive. This was not expected since the glycerotransferase PglO is known to transfer the glycan en-bloc to the PilE substrate and we might therefore expect sequential glycan addition based on decreasing affinity for sites on the protein backbone. Our results suggest that modification at one site may alter the affinity for the others, however the specificity of this enzyme is currently unknown. Most importantly, this level of glycosylation is a hitherto unreported phenomenon for pilin of Neisseria spp.

Strengths and weaknesses of both approaches

First of all, it is important to point out that, when aiming to characterise all PTM present on a protein or its different proteoforms, the mass profiling experiment is a key piece of information. In the case of the strain studied here, mass profiling of PilE indicated the presence of four major proteoforms, each expressed in similar abundance and each modified with phosphoglycerol plus two, three, four and five DATDH subunits respectively. This acted as primary reference for all future experiments.

The bottom-up approach was powerful for identifying post-translationally modified peptides and when coupled with an appropriate fragmentation technique such as ETD, for localising the sites of post-translational modification themselves. However, in order to achieve complete characterisation of a mixture of proteoforms and thus explicitly define the PTM content of each one, it is necessary to link these pieces of information together and to relate modified peptides to their parent proteoforms. Given the homogeneity of proteoform abundance, the strain characterised here was an ideal case to explore whether a bottom-up proteomics approach is able to achieve this goal, or not. Our results clearly show that although it is possible to completely map all PTMs on the lightest and heaviest forms of PilE, information obtained from the bottom-up approach is not sufficient for mapping of the middle forms. On the other hand, by selecting individual proteoforms and subjecting them to top-down ECD MS/MS the connectivity that was lost by the bottom-up approach was retained thus allowing the complete assignment of all glycosylation sites. This is expected to be the case in other mixtures of more than two proteoforms where each proteoform is modified by different numbers of a particular PTM. One must also point out that an important weakness of the top-down approach remains the analysis of data; from peak picking, to PTM assignment and scoring. Although currently available bioinformatics tools are sufficient for proteins with one or two known modifications, the analysis of highly modified proteins remains a challenge and manual interpretation is often needed. Improvement in this field is a requirement for more confident PTM assignment and high throughput analysis.

Conclusion

In this study the protein PilE, purified from a previously uncharacterised strain of Neisseria meningitidis isolated from a patient hospitalised with meningitis, has been analysed by different mass spectrometric approaches. Mass profiling showed that PilE exists as four major proteoforms, differing only by the number of DATDH glycans. A bottom-up strategy was initially chosen to map all PTMs of the four proteoforms. ETD on glycopeptides proved very useful for reliable assignment of glycosylation sites, however bottom-up only proved capable in characterising the PTM content of the lightest and heaviest proteoforms but not those of intermediate mass. Individual proteoforms were therefore selected and subjected to top-down ECD MS/MS on a 12T FT-ICR mass spectrometer. The sequence coverage obtained in each case allowed unambiguous identification of an increasing number of glycan subunits. Combining the top-down and bottom-up data allowed complete PTM characterisation of all proteoforms. The lightest form was found to be N-terminally processed and methylated, to carry a disulfide bridge close to the C-terminus, one phosphoglycerol on Ser⁷⁰ and DATDH at Ser⁶³ and Ser¹¹³. The additional glycosylation sites for the other proteoforms are Ser⁶⁸, Ser⁸³ and Ser¹⁰¹. Such an extent of glycosylation and indeed of PTM has never before been described for PilE and may be linked to increased pathogenicity or antigenicity. In general our results show that, in proteoform mixtures with more than two components, where multiple modifications of the same mass are present, a top-down mass spectrometry approach is necessary for complete proteoform mapping. In this study a 12T FT-ICR mass spectrometer was chosen for the top-down approach but it is probable that in the near future these experiments will be possible on a routine basis on other instrumental platforms such as later generation Orbitrap systems.