Enhanced correction methods for hydrogen exchange-mass spectrometric studies of amyloid fibrils

Abstract

We describe methods for minimization of and correction for artifactual forward and backward exchange occurring during hydrogen exchange–mass spectrometric (HX–MS) studies of amyloid fibrils of the Aβ(1–40) peptide. The quality of the corrected data obtained using published and new correction algorithms is evaluated quantitatively. Using the new correction methods, we have determined that 20.1 ± 1.4 of the 39 backbone amide hydrogens in Aβ(1–40) exchange with deuteriums in 100 h when amyloid fibrils of this peptide are suspended in D₂O. These data reinforce our previous conclusions based on uncorrected data that amyloid fibrils contain a rigid protective core structure that involves only about half of the Aβ backbone amides. The methods developed here should be of general value for HX–MS studies of amyloid fibrils and other protein aggregates.

Hydrogen/deuterium exchange (HX) has been used for several decades to study noncovalent interactions in globular proteins (Englander and Kallenbach 1983; Li and Woodward 1999; Dempsey 2001). Such HX experiments usually involve exposing proteins to a deuterated solvent and measuring the exchange of protons with deuterons over time (Englander and Kallenbach 1983) using various spectroscopic techniques. The method exploits the fact that protons either involved in H-bonded structures such as α-helices and β-sheets or buried in a protein’s core structure exchange more slowly than protons in solvent-exposed and non-H-bonded regions.

Over the past decade, HX has been used in conjunction with mass spectrometry (MS) to study protein structure, interactions, and dynamics (Katta and Chait 1991; Smith et al. 1997; Hernandez and Robinson 2001). We recently developed HX–MS methodology to characterize the secondary structure of amyloid fibrils (Kheterpal et al. 2000), which are the primary protein component of neuritic plaques found in the brain of Alzheimer’s disease patients, and are associated with over 20 other amyloid diseases (Buxbaum and Tagoe 2000; Sipe and Cohen 2000; Prusiner 2001; Westermark et al. 2002). Several biophysical techniques have shown that amyloid fibrils are highly ordered structures rich in β-sheet (Serpell 2000; Sipe and Cohen 2000). However, because of the insoluble, heterogeneous, and noncrystalline nature of the fibrils, a high-resolution structure has not been discernable. Solid-state nuclear magnetic resonance (NMR; Tycko 2000), X-ray diffraction (Serpell et al. 2000), and several indirect methods for probing protein structure (such as photoaffinity cross-linking [Egnaczyk et al. 2001], limited proteolysis [Kheterpal et al. 2001], and hydrogen exchange [Kheterpal et al. 2000; Hoshino et al. 2002; Ippel et al. 2002]) are being used to gain insight into the extent and pattern of individual residues involved in β-sheet formation within the fibril structure.

Our HX–MS studies have focused on the secondary structure of amyloid fibrils associated with Alzheimer’s disease, which are composed of Aβ peptides (39- to 43-residue proteolytic products of the amyloid precursor protein [APP; Selkoe 1994]). We have monitored the kinetics of deuterium exchange into fibrils by a process involving on-line mixing of a deuterated fibril suspension with a quenching/disaggregating/dissolving solution followed by immediate infusion into an electrospray ionization (ES) source. In our earlier work (Kheterpal et al. 2000), we found that Aβ incorporated into fibrils undergoes much slower exchange than monomeric Aβ peptide, with more complex kinetics. Our data indicated that about half of the 39 backbone amide hydrogens in Aβ(1–40) are protected from exchange when this peptide is incorporated into fibrils. However, in such HX studies, accurate and precise assessment of the extent of exchange is compromised by artifactual backward and forward exchange that occurs during quenching and further sample processing prior to injection into the mass spectrometer (Zhang and Smith 1993). Although correction methods to account for such artifactual exchange in globular proteins have been developed (Zhang and Smith 1993) and widely applied (Johnson and Walsh 1994; Wang et al. 1998, 1999; Anderson et al. 2001; Zhang et al. 2001; Tobler and Fernandez 2002), to our knowledge, these methods have not been applied to fibrillar systems like Aβ.

In this paper, we describe optimized protocols for minimizing the extent of artifactual exchange during HX–MS experiments on the Aβ fibrillar system. We also describe a test for the quality of correction for artifactual exchange, and use it to evaluate a series of alternative correction algorithms. The results confirm our earlier estimate that ∼50% of the 39 backbone amides of Aβ(1–40) are protected from exchange within the fibril. These new methods, which lead to significant improvement in the quality of the raw data and the robustness of the correction scheme, may be of general utility in HX–MS studies of amyloid fibrils.

Results and discussion

The back-exchange problem

The protofilaments thought to comprise amyloid fibrils are macroscopic, involving large numbers of monomer molecules; even if they could dissolve while retaining their structure, the aggregate mass would likely be too heavy for direct MS analysis. This is why we have carried out HX–MS experiments on the fibrils in two steps. First, fibrils are suspended in a D₂O solution buffered at pD 7.6 to promote exchange of exposed labile hydrogens with deuteriums in a process similar to that used for globular (soluble) proteins. Following a controlled incubation time, the suspension of partially deuterated fibrils is injected through one arm of a mixing T. Exchange is quenched and fibrils are rapidly disaggregated and dissolved by a low pH (or pD) processing solvent infused through the second arm of the T. The resulting dissolved monomers are immediately injected into the MS through the third arm of the T; the entire quenching, dissolution, and sampling process can be accomplished in <10 sec (Kheterpal et al. 2000). The quenching solvent can be H₂O- or D₂O-based; in either case, artifactual exchange (backward or forward) can take place on dissolution of the fibrils. Even in as little as 10 sec at pH 2.6, protons at acidic and basic terminal and side-chain sites may exchange completely (Wuthrich and Wagner 1979), so that deuterium incorporation in these sites will reflect only the final composition of the processing solvent, and not the deuteration prior to quenching. Additionally, as the fibrils disaggregate, all backbone amide positions become equally accessible; there may therefore be some artifactual exchange and incumbent label scrambling at these critical sites as well.

In our initial studies (Kheterpal et al. 2000), fibrils were incubated in 2 mM d-Tris-DCl buffer (pD 7.6) to exchange exposed labile hydrogens with deuteriums. The suspension (2 μL/min) was mixed with quenching solvent (18 μL/min of 50:50:0.2 [v/v/v] water/acetonitrile/formic acid) to yield a final sample mixture in which 18.2% of the water came from D₂O. (For convenience, we will subsequently refer to this composition as 18.2% D₂O, overlooking the acetonitrile.) Dissolution of Aβ(1–40) fibrils after incubation for 384 h in the deuterated buffer solution generated monomers containing on average 18.6 deuteriums. Rapid exchange of the 27 labile side-chain and terminal Aβ protons on dissolution presumably accounted for 5 of the incorporated deuteriums (18.2% of 27). Subtracting these 5 from the measured total deuterium content provides an estimate of 13.6 deuteriums associated with the amide groups in fibrillar Aβ(1–40). This correction does not, however, account for any artifactual exchange of the amide protons. Such exchange was evident from the fact that a fully deuterated monomer sample contained fewer than 39 amide deuteriums after processing (see following and Kheterpal et al. 2000), consistent with observations elsewhere of back exchange of backbone amide protons during HX–MS of peptides and globular proteins (Johnson and Walsh 1994; Smith et al. 1997; Wang et al. 1998, 1999; Ehring 1999; Resing et al. 1999; Tito et al. 2000; Zhang et al. 2001; Engen et al. 2002).

Method optimization

Prior to attempting correction for artifactual amide exchange, efforts were expended to minimize it. Several experimental parameters that can affect the ES performance and/or the extent of isotope exchange were optimized to obtain the maximum deuterium content for fully deuterated Aβ(1–40) monomer sampled at a total flow rate of 10 μL/min with 10% D₂O in the final solvent mixture. The tested parameters and final values included the emitter voltage (3.5 kV), flow rates of N₂ nebulizing and drying gases (20 and 275 L/h, respectively), source temperature (35°C), and cone voltage (40 V). Among these, the latter two were the most critical, consistent with observations elsewhere (Smith et al. 1997).

One important parameter—the sample temperature—could not be optimally controlled. The rate of HX is minimized in an aqueous solvent of pH ∼2.5 at or below 0°C (Englander and Kallenbach 1983). However, connection of the mixing T directly to the ES probe made it impractical to cool it. The pH and composition of the quenching solvent specified earlier were therefore chosen to minimize artifactual exchange at room temperature while maximizing the signal-to-noise ratio (S/N). As will be seen following, the resulting artifactual exchange was not large. The broadly optimized conditions provided raw data of higher quality than those obtained in our previous study, and were used in most of the tests of correction algorithms for artifactual exchange described as follows.

Correction for artifactual exchange of amide protons

Quenching conditions can be adjusted to minimize, but not eliminate, artifactual exchange. As noted earlier, approaches are available to correct for any residual effects in an effort to determine the extent of amide exchange prior to quenching—the information of most interest. This value should be independent of the quenching protocol. In our experiments, variation of solvent flow rates provided a particularly convenient means of varying the quenching conditions. The total flow rate determines the time available for artifactual exchange, whereas the relative flow rates of the sample and quenching solvent determine the final solvent composition. Because neither of these should affect the prequenching deuteration, the constancy of corrected values obtained at various flow rates should provide a sensitive test for the robustness of a correction method.

The method of Zhang and Smith (1993) was the first evaluated. It has been used widely for correcting back exchange and further forward exchange of amide protons during HX studies on globular proteins (Johnson and Walsh 1994; Wang et al. 1998, 1999; Anderson et al. 2001; Zhang et al. 2001; Tobler and Fernandez 2002). In this approach, nondeuterated and fully deuterated protein samples are analyzed using the same work-up as the incubated (partially exchanged) sample, and the corrected deuterium content (D_corr) in the backbone amides of the target sample is then given by

(1)

where m is the observed molecular mass for a partially deuterated target analyte; m_0% and m_100% are the molecular masses obtained when the procedure is applied to nondeuterated and fully deuterated analytes, respectively; and N is the number of backbone amides (N = 39 for Aβ[1–40]; Zhang and Smith 1993). The denatured state of a globular protein is typically used to obtain a fully deuterated sample. In all the studies presented here, molecular masses are corrected before substitution into equation 1 by subtracting side-chain contributions.

We first evaluated equation 1 with our original data set (Kheterpal et al. 2000), collected before instituting the improved data collection conditions described above. In terms of equation 1, the 13.6 value for amide deuterium content in partially exchanged fibrillar Aβ(1–40) noted earlier corresponds to an m value of 4343.0 Da (13.6 deuteriums added to the unlabeled peptide [measured molecular weight = 4329.4 Da; Kheterpal et al. 2000]). In an attempt to correct this value for artifactual exchange, we obtained an m_100% value by using completely deuterated monomer run under conditions identical to those of the partially deuterated fibrils (analyte in D₂O mixed with protonated processing solvent). The resulting molecular weight (after side-chain correction) was m_100% = 4356.7 Da, corresponding to an addition of 27.3 amide deuteriums (the other 11.7 amide sites having back-exchanged). We then obtained an m_0% value (4333.4 Da after side-chain correction; addition of 4 Da) by injecting protonated monomer in H₂O with a "quenching" solvent containing sufficient D₂O to give 18.2% D₂O after mixing with sample in the T.

Application of equation 1 using these values generates a D_corr value of 16.1 deuteriums, indicating that 41% of the amide hydrogens in fibrillar Aβ(1–40) exchanged with deuterium under these conditions. Comparison with the uncorrected number (13.6) indicates that artifactual back exchange has apparently removed a net ∼2.5 deuteriums. This number is relatively small and reasonable, in light of the fact that H₂O (rather than D₂O) is the primary water component (81.8%) after mixing, so that artifactual back exchange may be expected to be more prominent than forward exchange.

To assess the robustness of this method, we carried out a study of the flow rate dependence of the corrected values as discussed above. Thus, the flow rates of the sample and quenching solvent streams were systematically and proportionally varied to attain final flow rates of 2.5–60 μL/min, corresponding to mixing times that ranged from 45 to 2 sec. The final solvent mixture composition was maintained at 18.2% D₂O. Figure 1A ▶ presents the flow-rate dependence of the measured deuterium content in partially deuterated Aβ(1–40) fibrils ("d-fibrils," obtained by incubation in d-Tris-DCl buffer (pD 7.6) for ∼24 h; m in equation 1), deuterated Aβ(1–40) monomer ("D-monomer"; m_100% in equation 1), and protonated Aβ(1–40) monomer ("H-monomer"; m_0% in equation 1). Each mass is corrected by subtracting artifactual exchange into side-chain and terminal protons (18.2% of 27).

Figure 1.

(A) Measured amide deuterium content for deuterated monomer (squares), partially deuterated fibrils (circles; exchange time 24 h), and protonated monomer (triangles) versus total flow rate. The dotted line indicates the level of labeling that would be expected if the samples were fully equilibrated with the final mixed solvent. Deuterated monomer and fibril samples were generated by incubating these samples in 2 mM d-Tris-DCl buffer for ∼24 h. Deuterated fibrils and monomer in d-Tris-DCl were mixed with a processing solvent containing 50/50/0.2 (v/v/v) H₂O/CH₃CN/HCOOH. Protonated monomer was mixed with a processing solvent containing 50/38.9/11.1/0.2 (v/v/v/v) CH₃CN/H₂O/D₂O/HCOOH. Relative sample:solvent flow rates were 1:9 to obtain 18.2% D₂O in the final solvent mixture in all cases. Other experimental conditions were approximately those from our earlier work (Kheterpal et al. 2000). The data shown here have been corrected for fast exchanging labile terminal and side chain (nonamide) protons. Error bars represent the standard deviation of the mean (σ/√n) of multiple measurements (3–5) made on a single sample within 1 d. Multiple points at any one flow rate represent different samples run under similar conditions on different days. (B) Corrected amide deuterium content using the data of part A and the method described in equation 1 (Zhang and Smith 1993). The error bars were calculated using propagation of errors (Harris 1998).

Cursory inspection of Figure 1A ▶ shows that, as expected, longer mixing times before MS analysis (slower flow rates) lead to increased undesired exchange. For example, the amount of deuterium measured in backbone amides of d-fibril samples varies from ∼7.5 to 14 as the flow rate is changed from 2.5 to 60 μL/min (higher deuterium content corresponds to less back exchange). Similarly, the measured amount of deuterium in backbone amides of H-monomer and D-monomer samples varies from ∼7 to 2 and ∼14 to 34, respectively, as the flow rate is changed from 2.5 to 60 μL/min. These variations in deuterium content are largely due to the artifactual exchange in backbone amide protons during sample processing. For reference, Figure 1A ▶ shows as a dashed line the number of deuteriums expected in the backbone amides of a sample that has completely back exchanged (7.1 deuteriums, 18.2% of the 39 amide protons). It can be clearly seen from the analysis of D-monomer sample (for which all 39 amide sites are deuterated at the start of the experiment) that significant information is retained at all flow rates and can be used to analyze d-fibril data. The extent of artifactual amide back exchange observed in these experiments is ∼13%–65%, a range comparable to that reported elsewhere (5%–60%) for backbone amide protons during HX–MS analysis of peptides and globular proteins (Johnson and Walsh 1994; Smith et al. 1997; Wang et al. 1998, 1999; Ehring 1999; Resing et al. 1999; Tito et al. 2000; Zhang et al. 2001; Engen et al. 2002).

The results of using equation 1 to correct the d-fibril data from Figure 1A ▶ are presented in Figure 1B ▶. D_corr varies from ∼2 to 15 incorporated deuteriums as the flow rate is changed from 2.5 to 60 μL/min. It is clear that equation 1 does not provide consistent, flow-rate-independent results for these amyloid fibrils, at least for flow rates <50 μL/min. Thus, although the percentage of artifactual exchange in our data is within the range of other experiments in the literature, the corrected values fail the flow-rate independence test for robustness.

Unfortunately, flow rates > 50 μL/min cannot be routinely used in these experiments because amyloid fibrils are not completely dissolved at such high flow rates, leading to frequent clogging of the mixing device. Even if clogging were not a problem, it is also possible that at these fast flow rates only the outer sheath of the fibrils is dissolved, which could result in unrepresentative and misleading results. Furthermore, the S/N ratio in mass spectra was poor at high flow rates. Therefore, to completely dissolve fibrils and to obtain good S/N ratios, the optimum flow rate for our setup is ∼10 μL/min.

Figure 2A ▶ presents data similar to Figure 1A ▶, except that these data were generated using the optimized data collection conditions outlined earlier. For these studies, fibrils were incubated in d-Tris-DCl for ∼100 h, flow rates ranged from 2.5 to 38 μL/min, and the final solvent mixture contained 10% D₂O. Despite the higher H₂O content (90%, versus 81.8%), the artifactual amide back exchange for D-monomer under the optimized conditions was only 0%–33%, compared with 13%–65% in Figure 1A ▶. It can be clearly seen from the D-monomer and H-monomer data that there is essentially no artifactual backward or forward exchange at 38 μL/min. Therefore, it is reasonable to conclude that the number of deuteriums measured in d-fibrils at 38 μL/min (19.9 ± 1.1) must also reflect the correct number of amide deuteriums gained during the initial exchange reaction; if samples under these conditions could be routinely processed at 38 μL/min, no correction would be necessary. However, routine analysis at fast flow rates is problematic as noted above. The variation evident at lower flow rates in Figure 2B ▶, although improved relative to Figure 1B ▶, still indicates that the correction method of equation 1 is imperfect for fibrils. Ideally, these plots should be horizontal lines (slope = 0) if the correction were perfect. Furthermore, by the reasoning given earlier, the line should be at a level corresponding to incorporation of 19.9 deuteriums. By comparison, the slope of the least squares line in Figure 2B ▶ is 0.06 ± 0.02, the Y-intercept is 17.0 ± 0.3, and the average value is 17.9 ± 1. These values provide a quantitative means for comparing the performance of correction methods. In the discussion that follows, several such alternatives will be evaluated.

Figure 2.

(A) Measured amide deuterium content for deuterated monomer (squares), partially deuterated fibrils (circles; exchange time 100 h), and protonated monomer (triangles) versus total flow rate. Dashed lines on the top and bottom represent a deuterium content of 39 and 0, respectively. Conditions were optimized as described in the text. (B) Corrected amide deuterium content using the data of part A and the method described in equation 1 (Zhang and Smith 1993). The error bars were calculated using propagation of errors (Harris 1998).

Equation 1 was developed to address exchange experiments on proteins and peptides expected to undergo rapid unfolding under quenching conditions (Zhang and Smith 1993). The exchange behavior of amyloid fibrils during quenching is expected to be much more complex because it will be affected by the kinetics of dissolution. D-monomer and H-monomer samples may therefore not be good standards for assessing artifactual exchange of d-fibrils. As a test of this possibility, experiments were repeated, this time using completely deuterated fibrils (D₂O-fibrils, grown from D-monomer in deuterated phosphate buffer) and protonated fibrils (H₂O-fibrils, with no incubation in D₂O) to obtain, respectively, the m_100% and m_0% control values required for equation 1. This correction method is henceforth referred to here as Method A.

The deuterium contents in D₂O-fibrils and H₂O-fibrils were measured versus total flow rate, as was done for d-fibrils in Figure 2A ▶ (data not shown). The measured amide deuterium content in D₂O-fibrils was 1%–5% higher than in D-monomer at all flow rates, consistent with the expectation that relatively slow disaggregation reduced the time available for artifactual back exchange. No significant difference could be discerned between the deuterium gain of H₂O-fibrils and H-monomer; measured amide exchange into unincubated H₂O-fibrils was 0.1–2.5 (± 0.2) deuteriums compared with 0.3–2.4 (± 0.2) deuteriums in H-monomer. This is not surprising in light of the low absolute magnitude of artifactual forward exchange, which in turn is attributable to the low (10%) D₂O concentration after mixing. The maximum possible extent of forward artifactual exchange in these experiments was just 3.9 Da (10% of the 39 amide protons); a 5% change in these small numbers would constitute an absolute change of only 0.2 Da, which would not be detectable within experimental uncertainty. Figure 3A ▶ shows that the application of equation 1, now using fibril-based m_100% and m_0% values, does not eliminate the flow rate dependence of the corrected data; the D_corr values still vary significantly as the flow rate is changed. Indeed, the slope is higher (0.12 ± 0.01) and the intercept lower (15.8 ± 0.2) than the corresponding data from Figure 2B ▶.

Figure 3.

Amide deuterium content (corrected using various methods) versus total flow rate. (A) Method A, as described in equation 1 (Zhang and Smith 1993), using fibrillar standards (see text). (B) Method B, as described in equations 2–4. (C) Method C, as described in equations 5–6 (Mandell et al. 1998). The parameters needed for use of equation 6 were determined by replotting the d-fibril data of Figure 2A ▶ as a function of mixing time (calculated from flow rate), and then fitting to a single exponential function (equation 5) as described in the text. The best-fit equation was D_meas = 12.6 + 10.3 (e^−0.102×t). (D) Method D, a modification of Method C using equation 7 instead of equation 5 to determine the parameters needed for use of equation 6. The best-fit equation was D_meas = 3.9 + 16.5 (e^−0.0206×t). The error bars were calculated using propagation of errors (Harris 1998).

Alternative correction algorithms

Method B

Equation 1 concedes a minimum artifactual forward exchange (m_0%), and then assesses the fraction of maximum exchange (m_100%) achieved; it does not explicitly account for the possibility that forward and backward exchange may proceed at independent rates. In an attempt to improve on the performance of Figure 3A ▶, we considered an alternative correction scheme whereby the maximum amounts of artifactual forward (F.E.) and backward (B.E.) exchange were calculated at each flow rate as follows:

(2)

(3)

where MW is the measured average molecular weight of the Aβ(1–40) peptide (4329.6 Da), and N, m_0%, and m_100% are as defined earlier. The D₂O-fibrils and H₂O-fibrils were used as the m_100% and m_0% controls, respectively. As earlier, all data were corrected for labile side-chain and terminal protons prior to application of the amide exchange correction. The corrected deuterium content (D_corr) is obtained as described in equation 4 at each flow rate:

(4)

where m is again the measured value for d-fibrils (corrected for statistical exchange of labile terminal and side-chain protons), and the other terms are as defined earlier.

This correction method is based on the assumption that the amount of backward and forward amide exchange in d-fibrils is identical to that observed in D₂O-fibrils and H₂O-fibrils, respectively. This would be the case if, for example, there was a fixed (site-independent) rate for amide forward exchange, and a (different) fixed rate of amide backward exchange. Although this assumption is probably unrealistic, the resulting D_corr values (Fig. 3B ▶) are remarkably constant at all but the slowest (2.5 μL/min) flow rates. The slope of the linear fit of these data is −0.07 (Table 1), improving on that of Figure 3A ▶ (slope = 0.12; Table 1) despite the apparent outlier in Figure 3B ▶ at the lowest flow rate. Moreover, the average D_corr value from Figure 3B ▶ is 19.9 ± 1.3 deuteriums, identical to that obtained in raw data at 38 μL/min, where there is no significant artifactual exchange. It is unclear why this method does not perform well at the lowest flow rate. Flow rates lower than 2.5 μL/min did not provide stable electrospray and therefore this observation could not be further evaluated.

Table 1.

Quantitative analysis of the performance of correction methods

Correction method and % D2O in the final solvent mixture	Average corrected deuterium content	Intercepta	Slopea
Method A (Equation 1) using fibrillar standards for m0% and m100% (Zhang and Smith 1993)
5.26%	19.2 ± 0.9	17.7 ± 0.9	0.08 ± 0.05
10%	17.3 ± 1.4	15.8 ± 0.2	0.12 ± 0.01
10%b	17.6 ± 1.3	15.8 ± 0.3	0.12 ± 0.02
18.18%	16.9 ± 1.2	15.1 ± 0.1	0.15 ± 0.01
Method B (Equations 2–4)
5.26%	20.6 ± 0.3	21.2 ± 0.2	−0.03 ± 0.01
10%	19.9 ± 1.3	20.9 ± 0.6	−0.07 ± 0.03
10%b	19.4 ± 0.5	19.7 ± 0.3	−0.02 ± 0.01
18.18%	18.9 ± 0.5	19.6 ± 0.4	−0.06 ± 0.03
Method C (Equations 5 and 6; Mandell et al. 1998)
5.26%	no exponential fitc
10%	22.8 ± 0.5	22.7 ± 0.3	0.002 ± 0.02
10%b	22.8 ± 0.6	22.9 ± 0.4	−0.007 ± 0.02
18.18%	23.4 ± 0	23.4 ± 2 × 10−4	−3 × 10−5 ± 1.8 × 10−5
Method D (Equations 6 and 7)
5.26%	20.6 ± 0.3	20.2 ± 0.5	0.02 ± 0.03
10%	20.3 ± 1.3	20.5 ± 0.6	−0.01 ± 0.04
10%b	19.9 ± 0.9	19.2 ± 0.5	0.04 ± 0.03
18.18%	19.4 ± 0.5	19.3 ± 0.9	0.01 ± 0.07

Data are for analysis of fibrils incubated in D₂O for 100 h. Conditions were optimized as described in the text.

^a Intercept and slope for linear fit to plots of corrected deuterium content versus flow rate. Ideal values are 19.9 ± 1.1 and 0, respectively.

^b Values obtained excluding the apparent outlier at 2.5 μL/min.

^c Exponential fit did not converge.

Method C

Mandell et al. (1998) have described a correction scheme for back exchange during HX measurements using matrix-assisted laser desorption ionization (MALDI) MS. In this work, several replicate HX samples were simultaneously placed on a MALDI target and dried. Subsequent analysis of each sample took 3–5 min. The measured number of deuteriums (D_meas) decreased for successive samples, tangentially approaching a value B₁ at long sample analysis time (t). An effective rate constant (k) for this decay process was determined by fitting D_meas to a single-exponential curve:

(5)

where B₁ + B₂ is the initial total number of amide deuteriums in the peptide prior to sample work-up. Using values of B₂ and k from the fit, D_corr can be estimated from D_meas at a given analysis time as

(6)

The decay in deuterium content observed by Mandell and coworkers (1998) is analogous to the changes in D_meas observed with increasingly long mixing times (lower flow rates) in data like those of Figure 2A ▶. By replotting these data as a function of mixing time (calculated using the flow rate and the length of the capillary from the mixing T to the ES source) and fitting a single-exponential decay curve (data not shown), it was possible to calculate k, B₁, and B₂ as in equation 5; resulting parameters are listed in the legend to Figure 3C ▶. The corrected amide deuterium content (D_corr) was then calculated at each mixing time using equation 6. For comparison with the other correction schemes, the resulting data are plotted as a function of flow rate in Figure 3C ▶. Method C provides flow-rate-independent results, as evident from the slope of 0.002 for the linear fit (Table 1). However, the average D_corr value is too high: 22.8 ± 0.5, compared with the benchmark of 19.9 ± 1.1 obtained at 38 μL/min (Fig. 2A ▶). Thus, the performance of Method C was deemed to be inadequate for fibril HX.

Method D

In contrast to the conditions in the MALDI experiments of Mandell and coworkers (1998), the final solution conditions in the ES experiments described here are relatively well-defined, so that the ultimate equilibrium value of D_meas (B₁) can be predicted independently of the fitting routine. For example, for the data of Figure 2 ▶, 10% of the water in the final solvent mixture was D₂O, so that at equilibrium (t = ∞), 3.9 amide protons (10% of the 39 amides) will be deuterated (i.e., B₁ = 3.9). Equation 5 can be modified slightly to reflect this added information:

(7)

In a final attempt to derive a robust and accurate correction method, the d-fibril data presented in Figure 2A ▶ were fitted to the exponential decay function described in equation 7. Resulting k and B₂ values (reported in the legend of Fig. 3D ▶) were again used with equation 6 to determine a corrected amide deuterium content for data at each flow rate; results are presented in Figure 3D ▶. It can be seen that this method provides both flow rate independence (slope for the linear fit = 0.01; Table 1) and a good match to the expected average corrected deuterium content ( = 20.3 ± 1.3, consistent with the expected value of 19.9 ± 1.1).

Concentration robustness

The correction schemes described earlier were further evaluated for robustness by assessing their performance using two additional values of final solvent compositions (5.3% and 18.2% D₂O). Fibrils for these tests were incubated for 100 h, and samples of d-fibrils, D₂O-fibrils, and H₂O-fibrils were examined at flow rates ranging from 5 to 30 μL/min. The data are summarized in Table 1, where it can be seen that Method A (using equation 1) and Method C (equations 5 and 6) do not perform well at any composition. Method D performs best, giving slopes for plots of D_corr versus flow rate <0.02 in all cases (ideal = 0) and corrected deuterium values (estimated from the average or from the intercept) varying within the uncertainty (± 1.1) of the expected value of 19.9.

One limitation to Method D is that it requires routine collection of data at several flow rates in order to obtain k and B₂ from equation 7. Furthermore, reliable fits benefit from data at short times (fast flow rates), which are difficult to obtain, as discussed earlier. Method B offers a practical compromise that requires determination of m, m_100%, and m_0% at only one flow rate. The cause of the apparent outlier at the lowest flow (2.5 μL/min) in Figure 3B ▶ remains uncertain. Exclusion of this point improves the performance of Method B slightly; results of fits with and without this point are presented in Table 1. Until or unless the cause of the outlier becomes apparent, we recommend use of Method B only at flow rates ≥4.75 μL/min.

Conclusion

The concentration studies summarized in Table 1 reveal additional evidence of need for further improvements in the correction schemes. Even the two "successful" methods (B and D) show consistently lower corrected values of D_corr as percent D₂O in the final sample mixture was increased (i.e., as the expected amount of back exchange decreased). This may result from the fact that, like any simple correction method, these must to some extent overlook the site specificity of the forward and backward exchange rates; it is not at present feasible to apply site-specific corrections.

Nevertheless, the performance of the various algorithms is good, and the amplitude of the correction is (perhaps fortuitously) small, once the methodology has been optimized. The corrected exchange data obtained using Methods B and D (48%–53% amide hydrogens exchange in 100 h) confirm our previous results (based on the uncorrected data) that about 50% of the amide protons of the polypeptide backbone of Aβ (1–40) reside in highly protected core regions within the amyloid fibril structure. The best value of D_corr is probably that obtained using Method D, and averaging over all flow rates and solvent compositions: 20.1 ± 1.4. As we move ahead with studies targeting more detailed understanding of the fibrillar structure, the correction methods developed here should help provide more accurate and precise measures of exchange in these HX–MS experiments.

Materials and methods

Sample preparation

Aβ fibrils were generated from chemically synthesized Aβ (1–40) monomer (NH₂-DAEFRHDSGYEVHHQKLVFFAEDVGSNKGAIIGLMVGGVV-COOH) (Keck Biotechnology Center, Yale University) as described previously (Kheterpal et al. 2000, 2001). For HX experiments on the monomer, Aβ(1–40) monomer was dried after treatment with trifluoroacetic acid (Sigma) and hexafluoroisopropanol (Acros Organics) to remove residual aggregates (Zagorski et al. 1999). The dried monomer was then dissolved in 2.5 mM Tris-HCl buffer (pH 7.6; prepared in H₂O or D₂O, depending on the experiment). For HX experiments on protonated or deuterated Aβ (1–40) fibrils, fibrillar aggregates were collected by centrifugation. Protonated fibrils were washed once with 2.5 mM Tris-HCl buffer and deuterated fibrils were washed once with deuterated 2.5 mM Tris-HCl buffer (d-Tris-DCl). Fibrils were then resuspended in 2.5 mM Tris-HCl or d-Tris-DCl buffer (pH 7.6). All samples were prepared at an equivalent monomer concentration of 5–15 μM.

ES-MS

A Quattro II (Micromass) triple quadrupole ES mass spectrometer operated in the positive ion mode was used in the experiments described here. The experimental set-up used to dissolve fibrils on-line was similar to the one described previously (Kheterpal et al. 2000). Briefly, the standard electrospray probe was modified to accommodate a Valco "T" (0.25 mm i.d.) fitting in which sample solution (monomer or fibrils in 2.5 mM Tris-HCl or d-Tris-DCl buffer at pH 7.6) was mixed with quenching and dissolving solution (usually 50:50:0.5 [v/v/v] H₂O/CH₃CN/HCOOH) and then electrosprayed into the mass spectrometer. Sample solution and processing solvent were delivered using two 50-μm i.d. fused silica capillaries (Polymicro) connected to Harvard Bioscience model 22 and model 11 syringe pumps. Nitrogen was used for drying and nebulizing gases. The emitter voltage, cone voltage, source temperature, and flow rates of drying and nebulizing gas were optimized to minimize artifactual back exchange and to maximize signal intensity as described in the Results and Discussion.

Data were obtained in multichannel acquisition mode (summing 20–30 individual scans) with m/z ranging from 650 to 1150 at a rate of 2 sec/scan. The spectra were smoothed and centroids of the unresolved isotopic envelopes for the +5 and +6 charge states were used to calculate average molecular mass. Because the solvent contained a mixture of H₂O and D₂O, the ionizing protons were assumed to contain an equilibrium distribution of deuteriums (reflecting the percent D₂O in the mixture). Therefore, molecular masses were calculated as {m/z×z−[z+(z×%D₂O)]}. Resulting mass values derived from both +5 and +6 charge states were averaged to obtain the molecular masses.