Fresenius J Anal Chem (2000) 366 : 3–9
© Springer-Verlag 2000
PA P E R I N F O R E F R O N T
Jianzhong Lu · Renato Zenobi
In-situ monitoring of protein labeling reactions by matrix-assisted laser desorption/ionization mass spectrometry
Received: 3 September 1999 / Revised: 27 October 1999 / Accepted: 27 October 1999
Abstract Taking the labeling reaction of horse heart cytochrome c or ubiquitin with biotinamidocaproate N-hydroxysucchinimide ester (biotin-NHS) as test cases, this report demonstrates the usefulness of matrix-assisted laser desorption/ionization (MALDI) mass spectrometry for insitu monitoring of the labeling process and for determining the composition of the labeled products without the need for prior separation. The effects of pH and starting materials concentration on the labeling process were investigated in detail. Our MALDI MS results show that: (1) labeled products are always mixtures of different conjugates, which may explain peak broadening found in chromatographic studies of labeling reactions; (2) the higher conjugate fractions become more prominent as the labeling reaction proceeds, with a concomitant exponential decline of the lower conjugate fractions; (3) biotinNHS can be incorporated into peptides and protein in a stepwise and controlled manner simply by adjusting the molar ratio of the starting materials.
Introduction Proteins must be labeled in order to detect and/or amplify the initial antibody-antigen reaction in an immunoassay, or to improve the sensitivity of protein analysis by chromatography and electrophoresis. Labeled proteins have been used in many disciplines, including molecular biology, clinical diagnosis, environmental monitoring, food safety, and others [1–4]. Protein labels include radioisotopes; enzymes such as horseradish peroxidase (HRP) and alkaline phosphatase (AP); biotin; and fluorescent dyes.
Jianzhong Lu1 · Renato Zenobi (Y) Department of Chemistry, ETH Zentrum, Universitätstrasse 16, 8092 Zürich, Switzerland e-mail:
[email protected] Present address: leave from Shijiazhuang Medical College, Shijiazhuang 050081, Hebei, China 1 On
Proteins already conjugated to a label are commercially available from many sources. The required purity of the labeled proteins depends greatly on the assay system used and the sensitivity aimed at. For most labeled proteins, the product will perform better if unbound label and native protein are removed. Competitive immunoassays do not normally require further purification of the labeled product, but “sandwich” type noncompetitive assays need a high purity of the labeled proteins. Accordingly, the purity as well as purification methods of proteins have become more important factors in developing these assays. In addition, it is well known that precolumn labeling of protiens often leads to poor electrophoretic resolution. It is widely assumed [1, 2, 5] that this peak broadening is from multiple labeling of the protein, but only circumstantial experimental evidence exists to support this assumption. Craig and Covichi [5] have used capillary zone electrophoresis to follow the labeling reaction of the Aequorea victoria green fluorescent protein with 3-(2-furoyl)quinoline-2-carboxaldehyde. At least in an indirect way, they provided some evidence of multiple labeling by comparing the electropherograms of the reaction products after different incubation periods with that of the native protein. The precise composition of the labeled protein could not be determined since electrophoresis only yields electrophoretic mobilities, and not molecular mass. A method that allows in-situ monitoring of protein labeling reactions and that gives information of the composition of the labeled products would be very useful. The protein-to-label molar ration can greatly influence immunoassay sensitivity, such that a robust and fast method is preferred for monitoring the progress of the labeling reaction on-line. The currently available methods are not sufficiently fast and accurate to provide this important information [1, 2]. Conventionally, only the labelto-protein molar ratio (“label ratio”) of commercial labeled proteins is obtained from the absorbance or fluorescence difference before and after labeling the proteins. Usually, companies only give an approximate label ratios for their labeled products and cannot provide further information about their composition. Finally, in order to de-
4
sign and optimize experimental conditions for a labeling reaction, it should be followed in-situ and reaction kinetics need to be investigated in detail. Since the advent of soft ionization methods, such as matrix-assisted laser desorption/ionization (MALDI) and electrospray ionization (ESI), mass spectrometry has developed into a powerful tool for the analysis of biomolecules and synthetic polymers [6–12]. To date, there are only few examples where soft ionization MS has been used for following chemical reactions in-situ [13–17]. For example, Thurmer et al. [13] monitored the enkephalin synthesis on a soluble polyethylene glycol support with MALDI-MS. With minute amounts of sample, the high mass accuracy permitted detection of side reactions as well as incomplete reaction products. Sauvagnat et al. [14] employed ESI to monitor a liquid organic synthesis step-by-step. Bleczinski et al. [15] followed the hybridization of oligonucleotide mixtures to immobilized complementary DNA by MALDI. Cavelier et al. [16] investigated the cyclization of the tetrapeptide H-Leu-Pro-LeuPro-OH by ESI MS and LC/ESI-MS. Keough et al. [17] studied the conjugation of two small molecules to hen egg white lysozyme and human serum albumin by MALDI. The purpose of this project was to follow a protein labeling reaction in-situ and to monitor the composition of the labeled protein at different reaction conditions. We also sought to obtain direct evidence for multiple labeling of the proteins. Using the labeling reaction of horse heart cytochrome c and ubiquitin with biotinamidocaproate N-hydroxysucchinimide ester as test cases, our results demonstrate the suitability of MALDI MS for this purpose. We show that MALDI is a very informative and powerful tool to characterize the labeling reaction and to investigate the reaction kinetics due to the ease and speed of the analysis.
Experimental Materials. All the materials used in these experiments were obtained from commercial sources. Horse heart cytochrome c, biotinamidocaproate N-hydroxysucchinimide ester (biotin-NHS), ubiquitin, and biotin labeled horse heart cytochrome c were supplied by Sigma (Buchs, Switzerland). Sinapinic acid (SA) and other matrix substances were purchased from Fluka Chemie AG (Buchs, Switzerland). All chemicals were of highest purity available, solvents were of spectrophotometric grade. Aqueous solutions were prepared using distilled water. Intrumentation. All mass spectra were obtained using a home-built 2 m linear time-of-flight mass spectrometer. Desorption and ionization of the samples was performed using the 337 nm output from a pulsed nitrogen laser (model VSL-337 NDT, Laser Science Inc., Franklin, MA). The nitrogen laser was focused onto the sample surface and formed an elliptical spot of 0.1 mm × 0.2 mm. Laser pulse energies in the range of 10–30 µJ were used. For ion extraction, a static acceleration potential of 25 kV was used in the source region. Typically, spectra were obtained by averaging 100 single shots in order to enhance the signal-to-noise ratio of the mass spectra. All the spectra shown were recorded in positive ion mode. A more detailed description of the setup can be found in the paper of Dubois et al. [18].
Labeling reaction and MALDI sample preparation. Unless otherwise stated, a 1 mL aliquot of a 1.6 × 10–4 M solution of horse heart cytochrome c or ubiquitin in 0.1 M sodium bicarbonate buffer was added to 20 µL biotin-NHS solution (1.0 mg in 10 µL N,N-dimethylformamide) and incubated at room temperature for 2, 10, 30, 60, 90, 120, 150, 180, 240, and 300 min. This is a routine procedure for biotin labeling of proteins [19]. Immediately thereafter one part of the sample was mixed with five parts (by volume) of a saturated matrix solution for MALDI analysis. 1 µL of this mixture was deposited on the probe tip, partially dried in a warm air stream and then completely dried in the vacuum. This fast evaporation resulted in a microcrystalline, optically homogeneous sample layer over the whole probe tip. The reproducibility of the MALDI results was checked for several experiments. The relative intensities of the major MALDI signals were reproducible within 10%, of the minor peaks generally within 20%. In the pH dependence experiments, the pH of the sample solution was artificially adjusted using NaHCO3 and HCl.
Results and discussion Characterization of starting materials and products of the labeling reaction For amine-reactive probes like biotin, biotin derivatives, and fluorescein isothiocyanate (FITC), conjugation to proteins is thought to occur almost exclusively through attack of the ε-amino group of lysine residues and of the N-terminal α-amino group. Although cytochrome c contains 18 lysine residues and 1 N-terminal amino group, only 9 sites are accessible on the exterior of the three-dimensional structure of cytochrome c in the folded form as deduced from ESI and MALDI results [22–22]. This means that a maximum of 9 sizes can be labeled provided that cytochrome c does not denature during the labeling reaction. We first recorded MALDI mass spectra of the starting materials and of the labeled product after 60 minutes reaction time using several common matrices. The optimum matrix for cytochrome c and biotin-NHS was found to be sinapinic acid. The MALDI data of the product with this matrix is shown in Fig. 1. Nine peaks corresponding to different conjugates can be clearly observed, indicating
Fig. 1 MALDI TOF mass spectrum of biotin labeled cytochrome c after 60 min reaction. Sinapinic acid was used as a matrix
5
that at least nine conjugates coexist in this product. The intensities of unlabeled cytochrome c (average MW = 12,385 Da) and of conjugates with less than 3 or more than 13 labels are negligible (S/N < 1.5). The most intense peaks are from the 1 : 8 and 1 : 9 conjugates and the maximum number of labels is 12. The mass accuracy is high enough to determine the label ratio without any ambiguity. The number of labels observed is higher than the maximum number of accessible terminal and lysine amino acid residues of the folded cytochrome c. Either sites other than lysine amines can be labeled, or cytochrome c is partially denatured during the labeling reaction. It is well known that for modified peptides such as sialylated glycopeptides [23–25], there is a distinct possibility of loss of the label during the MALDI process, in particular when using high laser energy. However, during experiments with a range of laser fluences, the relative intensities of cytochrome c and its biotin conjugates did not change significantly. Therefore, we belive that no label loss occurs in our case. For a molecule with n accessible amino groups, 2n – 1 reaction products are possible [26]; if 12 sites can been labeled, a total of 212 – 1 = 4095 different conjugates can be formed, many of them isomeric. Some lower conjugated fractions have disappeared after 60 min reaction, and there are only 9 observed in our MALDI spectra. Each of these can therefore be due to many isomeric products with the same number of attached labels that could only be distinguished with MS/MS methods. Figure 1 clearly shows that the labeling reaction leads to multiple products, which explains and extends the earlier results of Craig and Dovichi [5]. By performing electrophoresis on native green fluorescence protein and on the reaction products produced by fluorescence labeling, they found that the unlabeled protein only gave one sharp peak whereas the labeled product yielded several peaks or a very broad peak, depending on the incubation time. They interpreted this to be the result of multiple labeling of the protein. Effect of labeling reagent concentration The number of labels corresponding to the most intense peak in the mass spectrum (“label number”) depends on the relative reactant concentration, as illustrated in Fig. 2. The label number was found to increase with increasing biotin-NHS concentration up to a molar excess of ca. 14. Above this, there was little change and the label number leveled out at 9 or 10. For a 1.38-fold excess of biotinNHS, only four conjugates were observed after 3 h of reaction time (Fig. 3a), with unlabeled cytochrome c always present. Peaks corresponding to higher conjugates were never observed, even after increasing the reaction time. This indicates that once the fourth label is attached to cytochrome c, no new sites will be labeled for a small molar excess of biotin-NHS. When applying a straight baseline (dashed line in Fig. 3a), the relative peak integrals of native, singly, doubly, triply and quadruply labeled cytochrome c were found to
Fig. 2 Label number vs. molar ratio of biotin-NHS and cytochrome c after 180 min reaction. The reaction pH was 9.0
Fig. 3 (a) MALDI TOF mass spectra for cytochrome c and a 1.38fold excess of biotin-NHS after 3 h of reaction time. (b) Deconvolution of peak distribution by fitting each conjugate separately to a Gaussian
be 19%, 37%, 32%, 11% and 3%, respectively. We did not determine the relative ionization/detection efficiencies of different conjugates, but based on previous work from our laboratory [27], we can assume that they are similar. It follows that the resulting label ratio is 0.19 × 0 + 0.37 × 1 + 0.32 × 2 + 0.11 × 3 + 0.03 × 4 = 1.46, somewhat larger than the label-to-protein molar ratio used. The discrepancy is small, and probably due to experimental error. We can also deconvolute the spectrum by fitting each conjugate separately by a Gaussian (Fig. 3b). In this case, the relative peak integrals of native, singly, doubly, triply and quadruply labeled cytochrome c were found to be 15%, 31%, 30%, 17% and 6%, respectively. The resulting label ratio was determined to be 1.66, close to the value ob-
6
tained above. For convenience, we used the first method to determine relative peak integrals. For a 28-fold excess of biotin-NHS, a conjugate with 12 labels can be observed after 3 h of reaction time. Even by greatly increasing the reaction time, at most the conjugate with 14 labels was observed. This shows that cytochrome c is at least partly folded, because we should observe 19 peaks (from 18 lysines and 1 terminal amino acid) if cytochrome c is completely denatured. Of course, from these results no conlcusion can be drawn about whether the tertiary structure is close to cytochrome c’s native structure or not. It should be pointed out, however, that native proteins have been claimed to survive the soft ionization processes used in modern mass spectrometry. Complete denaturation is not expected for a routine labeling procedure. We made some attempts to increase the number of accessible sites of cytochrome c further by applying some common denaturing methods such as heating or mixed organic solvents. A higher label number, up to 17, was observed occasionally, but mostly, we were unable to label more than 14 sites. The reason for this behavior remains unclear. Our data shows that biotin-NHS can be incorporated into cytochrome c in a stepwise manner by adjusting the relative molar ratio of the starting materials. This has an important consequence: the product composition can be controlled. For example, a very low molar ratio produces only two peaks, one from native and the other from singly labeled cytochrome c. Based on this, singly labeled protein can in principle be generated using a simple separation. Singly labeled protein maybe used for the development of highly sensitive affinity probes such as those recently employed by Shimura et al. [28]. Effect of pH on the labeling process In addition to the concentration of the starting materials, pH has an important influence on the reaction. Therefore, labeling reactions at five different pH values were followed in-situ. Figure 4 compares the label number attributable to the highest MALDI peak versus reaction time at different pH values. In each case, the product was mixed off-line with SA, which naturally gave an acidic pH of the sample solution. A lower label number was found for pH ≤ 6.0; the reaction rate was also significantly reduced. Effects of pH on the labeling reaction can be explained by the reaction mechanism. This labeling reaction is a second-order nucleophilic substitution, i.e., the rate is proportional to the concentration of both the nucleophile and the species being attacked. With decreasing reaction pH, the ε-amino groups of lysyl residues and α-amino groups of the N-terminal amino acids are more easily protonated, such that the accessible labeling sites will be reduced, leading to a decrease in the concentration of the nucleophile. In addition, the species being attacked, biotin-NHS is prone to hydrolysis at a low pH. Therefore, the rate of the labeling reaction decreases with decreasing pH.
Fig. 4 The label number of biotin labeled cytochrome c vs. reaction time at different pH for a cytochrome c to biotin-NHS molar ratio of 1 : 27.5. Lines are drawn to guide the eye
At pH 7.0, the label number was always a bit larger than that at pH 9.0, although the reaction rates were quite close at pH 7.0 and 9.0. Due to the absence of disulfide bonds, cytochrome c is quite sensitive to slight variations of the medium. For example, Yang et al. [29] found cytochrome c to exist in a molten globule state at low concentrations of guanidine hydrochloride (GuHCl), in partially denatured conformations at high concentrations of GuHCl, but in a stable conformation in the absence of GuHCl. Similarly, varying the pH between 7.0 and 9.0 may cause minor difference in the tertiary structure of cytochrome c. It is worth to mention that only unlabeled cytochrome c can be observed at pH 2.5. This leads to a particular advantage of MALDI MS for such studies: it is not necessary to add excess hydroxylamine to quench the labeling reaction [30], the acidic MALDI matrix is sufficient for this purpose. Three different sets of experiments were designed to further support the hypothesis that the number of accessible sites for labeling is reduced at lower pH. Firstly, ubiquitin was selected as another model protein and the pH effect on the biotin-NHS labeling reaction was followed in-situ. The label number of the most intense peak and accessible amino sites were also found to decrease with decreasing pH. Secondly, if cytochrome c was reacted with biotin-NHS at pH 4.0, the reaction went to completion in 24 hours as shown in the MALDI spectra. If the biotin concentration was then doubled, the label number of the most intense peak barely changed, indicative of no more available reactive amino groups at this pH. However, if we artificially adjusted the pH using concentrated NaOH, we found that the label number of the most intense peak gradually increased up to a maximum value of 9 or 10, as shown in Fig. 5. These results are in agreement with our interpretation.
7
Fig. 5 Label number of the most intense peak of biotin labeled cytochrome c from each spectrum versus reaction pH. The labeling reaction between 1.6 × 10–4 M horse heart cytochrome c and 4.4 × 10–3 M biotin-NHS was initially monitored at pH 4.0, and the first spectrum was recorded after 24 h reaction. Then the pH was artificially increased to a higher value by using NaOH, and the spectrum was recorded for each pH
Fig. 6 MALDI TOF mass spectra of the labeling products at various times (pH 9.0) using 1.6 × 10–4 M horse heart cytochrome c and 3.2 × 10–3 M biotin-NHS
In-situ monitoring of the labeling reaction The labeling reaction between horse heart cytochrome c and biotin-NHS in 0.1 M sodium bicarbonate solution (pH 9.0) for a 20 : 1 molar excess of biotin-NHS was followed in-situ by MALDI MS, as shown in Fig. 6. Each MALDI mass spectrum represents a snapshot of the labeling reaction’s progress. The label number of the most intense peaks gradually increased with increasing reaction time, reaching a maximum label number of 10, above which it remained constant. As outlined above, the prod-
Fig. 7 Time profiles for some of the peaks in Fig. 6
ucts were always mixtures of different conjugates. These results indicate that at most 13 or 14 sites can be occupied and the label number of the dominant conjugate is 10 at the experimental conditions used, in good agreement with the label number of commercial biotin labeled cytochrome c (the label number is 9 or 10) [27]. It can be concluded that the additional aminocaproyl spacer that was used here does not change the reactivity, although in general, biotinylation reagents incorporating an aminocaproyl spacer can reduce steric hindrance in binding avidin to some biotinylated compounds [31]. Figure 7 shows time profiles for some of the peaks in the MALDI spectra in Fig. 6. If we assume that the concentration of a given conjugate is proportional to its peak integral [32, 33], a kinetic analysis of the data can in principle be made. The time dependent concentration of the conjugate carrying n biotin labels will be of the form [CBn] = a exp (–bt) + c exp (–dt). However, the constants a, b, c and d depend on initial concentrations of native cytochrome c and biotin-NHS and on rate constants in a complicated fashion. It is not possible to deduce individual rate constants from fits of the time dependence of the relative peak integrals. In agreement with the functional form given above, we found that double exponential functions fit the data quite well, as shown in Fig. 7. However, this must be considered as an empirical result only.
Monitoring of the labeling reaction between ubiquitin and biotin-NHS As outlined above, the label number is expected to decrease with decreasing numbers of lysine residues located on the exterior of the three-dimensional structure of a folded protein. This is indeed reflected in the case of the labeling reaction between ubiquitin (7 lysines and 1 terminal amino acid) and biotin-NHS in 0.1 M sodium bicarbonate solution (Fig. 8). Even after 48 hours reaction, the label number of the most intense peak was only 5 and
8
(3) The labeling reaction leads to a complex mixture of different conjugates. Consistent with the interpretation of Craig and Dovichi, we believe that multiple labeling of proteins leads to peak broadening in electrophoresis. (4) Both pH and the concentration of the reactants have a strong influence on the labeling reaction. Biotin-NHS can be incorporated into peptides and proteins in a stepwise manner by adjusting the molar ratio of the starting materials. Acknowledgement Jianzhong Lu gratefully acknowledges a Wilhelm Simon fellowship from the ICSC-World Laboratory. We thank Edda Lehmann for critically reading the manuscript.
References
Fig. 8 MALDI mass spectra of the labeling products between 1.0 × 10–4 M ubiquitin and 1.8 × 10–3 M biotin-NHS in 0.1 M sodium bicarbonate solution (pH 9.0) at various reaction times
the maximum label number was 6. This indicates that at most 4 or 5 lysines can be labeled in our experimental condition. These results are similar to the ESI mass spectrometric results for the native form of ubiquitin [34]. Katta and Chait found that not all of the 7 lysine residues and terminal amino group are easily protonated for ubiquitin in its folded form. Previous work from our laboratory also shows that only six sulfonates can interact with protonated amine groups of ubiquitin in its folded form [18]. In addition, these results show again that the labeling reaction leads to a complex mixture of products and the higher conjugate fraction grows as the labeling reaction proceeds with a concomitant exponential decline of the lower conjugate fraction.
Conclusions We demonstrate that MALDI MS is a useful and easily implemented method for in-situ monitoring of the labeling reaction of proteins with biotin-NHS. Each reaction intermediate can be detected within minutes using very small amounts of sample. The most important findings about the labeling process are: (1) MALDI mass spectra provide molecular weight information about the composition of labeled proteins, although isomers with the same number of attached labels cannot be distinguished. (2) The higher conjugated fractions grow as the labeling reaction proceeds at the expense of the lower conjugated fractions. The lower conjugated fractions disappear exponentially.
1. Hemmila HA (1991) Applications of fluorescence in immunoassays. John Wiley & Sons, New York 2. Price CP, Newman DJ (1997) Principle and practice of immunoassays. Stockton Press, New York 3. Szulc ME, Kruss IS (1994) J Chromatogr 659 : 231–245 4. Liu J, Hsieh YZ, Wiesler D, Novotny M (1991) Anal Chem 63 : 408–412 5. Craig DB, Dovichi NJ (1988) Anal Chem 70 : 2493–2494 6. Hillenkamp F, Karas M, Beavis RC, Chait, BT (1991) Anal Chem 63 : 1193A–1203A 7. Chowdhury SK, Katta V, Chait BT (1990) J Am Chem Soc 112 : 9012–9013 8. Karas M, Bachmann D, Bahr U, Hillenkamp F (1978) Int J Mass Spectrom Ion Proc 78 : 53–68 9. Tanaka K, Waki H, Ido Y, Akita S, Yoshida Y, Yoshida J (1988) Rapid Commun Mass Spectrom 1:151–153 10. Fenn JB, Mann M, Meng LK, Wong SF, Whitehouse CM (1989) Science 246 : 64–71 11. McLuckey SA, Van Berkel GJ, Glish GL (1990) J Am Chem Soc 112 : 5668–5670 12. Liang X, Lubman DM, Rossi DT, Nordblom GD, Barksdale CM (1998) Anal Chem 70 : 498–503 13. Thurmer R, Meisenbach M, Echner H, Al-Qawasmeh RA, Weiler A, Voeter W, Korff U, Schmitt-Sody W (1998) Rapid Commun Mass Spectrom 12 : 398–402 14. Sauvagnat B, Enjalbal C, Lamaty C, Lazaro R, Martinez J, Aubagnaz JL (1998) Rapid Commun Mass Spectrom 12 : 1034–1037 15. Bleczinski CF, Richert C (1998) Rapid Commun Mass Spectrom 12 : 1737–1743 16. Cavelier F, Enjalbl C, El Haddadi M, Martinez J, Sanchez P, Verducci J, Aubagnac JL (1998) Rapid Commun Mass Spectrom 12 : 1585–1590 17. Keough T, Lacey MP, Trakshel GM, Asquith TN (1997) Int J Mass Spectrom Ion Proc 169/170 : 201–215 18. Dubois F, Knochenmuss R, Steenvoorden RJJM, Breuker K, Zenobi R (1996) Eur Mass Spectrom 2 : 167–172 19. Goding JW (1986) Monoclonal antibodies: principles and practice. Academic Press, San Diego 20. Mirza UA, Cohen SL, Chait BT (1993) Anal Chem 65 : 1–6 21. LeBlanc JCY, Beuchemin D, Siu KWM, Guevremont R, Berman SS (1991) Org Mass Spectrom 26 : 831–839 22. Salih B, Zenobi R (1998) Anal Chem 70 : 1536–1543 23. Pitt JJ, Gorman JJ (1996) Rapid Commun Mass Spectrom 10 : 1786–1788 24. Huberty MC, Vath JE, Yu W, Martin SA (1993) Anal Chem 65 : 2791–2800 25. Spengler B, Kirsch D, Kaufmann R (1992) Rapid Commun Mass Spectrom 6 : 105–108 26. Zhao JY, Waldron KC, Miller J Zhang JZ, Harke HR, Dovichi NJ (1992) J Chromatogr 608 : 239–242
9 27. Lu J, Zenobi R (1999) Anal Biochem 269 : 312–316 28. Shimura K, Karger BL (1994) Anal Chem 66 : 9–15 29. Ynag HH, Li XC, Amft M, Grotemeyer J (1998) Anal Biochem 258 : 118–126 30. Banks PR, Paquette DM (1995) J Chromatogr A 693 : 145–154 31. Leary JJ, Brigati DJ, Ward DC (1983) Proc Natl Acad Sci USA 80 : 4045–4049
32. Johnstone RAW, Lewis IAS, Rose ME (1983) Tetrahedron 38 : 1597–1603 33. Young D, Hung H, Liu LK (1997) Rapid Commun Mass Spectrom 11 : 769–773 34. Katta V, Chait BT (1991) Rapid Commun Mass Spectrom 5 : 214–217