Multivariate Curve Resolution Applied to the Analysis and Resolution


Multivariate Curve Resolution Applied to the Analysis and Resolution...

4 downloads 124 Views 315KB Size

Anal. Chem. 2004, 76, 7094-7101

Multivariate Curve Resolution Applied to the Analysis and Resolution of Two-Dimensional [1H,15N] NMR Reaction Spectra Joaquim Jaumot,† Vicente Marcha´n,‡ Raimundo Gargallo,† Anna Grandas,‡ and Roma` Tauler*,§

Department of Analytical Chemistry and Department of Organic Chemistry, University of Barcelona, Martı´ i Franque` s 1-11, E-08028 Barcelona, Spain, and Department of Environmental Chemistry, IIQAB-CSIC, Jordi Girona 18-26, E-08034 Barcelona, Spain

Multivariate curve resolution is proposed for the study of complex chemical reactions monitored by two-dimensional (2D) NMR spectroscopy. In particular, in this work, multivariate curve resolution is applied to the study of the reaction between 15N-labeled cisplatin and the amino acid-nucleotide hybrid (Phac-Met-linker-p5′dG). At several stages of the reaction, 2D [1H,15N] HSQC NMR spectra were acquired and stored in data matrices. In a first step, multivariate curve resolution was applied to analyze individually each one of these 2D spectra, allowing the resolution of the corresponding 1H and 15N onedimensional correlation spectra. In a second step, the whole set of 2D spectra recorded along the reaction were simultaneously analyzed by multivariate curve resolution, allowing the resolution of the kinetic concentration profiles and of the pure 2D NMR spectra of each of the species detected along the reaction. Results finally obtained confirmed previously postulated reaction mechanisms involving the existence of two monofunctional adducts and of two bifunctional adducts, with the structure of one of them not completely resolved. In recent years, chemometric methods have been applied to the exploration, analysis, and resolution of complex data sets obtained not only by traditional spectroscopies such as UV, NIR, IR absorption, or circular dichroism but also by more sophisticated spectroscopic techniques supplying more complex data sets as those obtained in spectroscopic imaging. Spectroscopic images are three-way data arrays giving information in three ways or modes, two spatial and one spectroscopic. When several spectroscopic images are studied simultaneously, an even more complex four-way data array structure may be built up. Examples of these images are related to satellite or microscopic spectroscopic images.1-5 Another type of spectroscopic image available at * To whom correspondence should be addressed. Fax: (34)-932045904. E-mail: [email protected].. † Department of Analytical Chemistry, University of Barcelona. ‡ Department of Organic Chemistry, University of Barcelona. § IIQAB-CSIC. (1) Andrew, J. J.; Hancewicz, T. M. Appl. Spectrosc. 1998, 52, 797-807. (2) de Juan, A.; Tauler, R.; Dyson, R.; Marcolli, C.; Rault, M.; Maeder, M. TrAC, Trends Anal. Chem. 2004, 23, 70-79. (3) Geladi, P. Multivariate Image Analysis in Chemistry and Related Areas: Chemometric Image Analysis, 1st ed.; Wiley Europe: Chichester, U.K., 1996.

7094 Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

present is that obtained by means of magnetic resonance imaging (MRI) techniques, which are especially suitable for medical diagnosis and research. Different methods and approaches have been applied to the analysis and resolution of this kind of images.6-9 One-dimensional (1D) NMR spectroscopy has been widely used for structural elucidation, chemical reaction monitoring, and chemical analysis.10-12 However, few attempts have been described so far trying to resolve 1H NMR data spectra recorded directly from biochemical or biophysical process monitoring.13 Between MRI and 1D NMR, there is still a field for chemometric resolution methods scarcely explored. This is the field of multidimensional NMR spectroscopy14 and includes all the NMR techniques that record more than one spectral dimension per sample analyzed. The use of NMR to monitor the evolution of reaction processes has to overcome some difficulties. On one hand, there is the labile behavior of the species involved in fast equilibria compared to the time of measurement of the recorded NMR signals, which usually is slow. On the other hand, species identification and resolution problems arise because of the presence of complex overlapping nonselective signals, unknown reaction mechanisms, mixtures with unknown ratios of products, or mixtures containing products of unknown structure. With respect to the use of chemometric methods to analyze the results of NMR studies, if equilibria between species are slow (4) Huang, J.; Wium, H.; Qvist, Karsten B.; Esbensen, K. H. Chemom. Intell. Lab. Syst. 2003, 66, 141-158. (5) Martinez, M.; Aragon, A. D.; Rodriguez, A. L.; Weber, J. M.; Timlin, J. A.; Sinclair, M. B.; Haaland, D. M.; Werner-Washburne, M. Nucleic Acids Res. 2003, 31, e18/1-e18/8. (6) Antalek, B.; Hornak, J. P.; Windig, W. J. Magn. Reson. 1998, 132, 307315. (7) Lindon, J. C.; Holmes, E.; Nicholson, J. K. Prog. Nucl. Magn. Reson. Spectrosc. 2001, 39, 1-40. (8) Windig, W.; Hornak, J. P.; Antalek, B. J. Magn. Reson. 1998, 132, 298306. (9) Witjes, H.; Simonetti, A. W.; Buydens, L. Anal. Chem. 2001, 73, 548A556A. (10) Nordon, A.; McGill, C. A.; Littlejohn, D. Analyst 2001, 126, 260-272. (11) Holmes, E.; Antti, H. Analyst 2002, 127, 1549-1557. (12) Lindon, J. C.; Holmes, E.; Nicholson, J. K. Anal. Chem. 2003, 75, 384A391A. (13) Vives, M.; Tauler, R.; Moreno, V.; Gargallo, R. Anal. Chim. Acta 2001, 446, 439-450. (14) Croasmun, W. R.; Carlson, R. M. K. Two-Dimensional NMR Spectroscopy: Applications for Chemists and Biochemists, 2nd ed.; Wiley & Sons: New York, 1994. 10.1021/ac049509t CCC: $27.50

© 2004 American Chemical Society Published on Web 10/23/2004

in relation to the measuring time of the NMR signals (inert signals), bilinear chemometric methods can be easily and safely applied. On the contrary, when equilibria between species are faster than the measuring time (labile signals), the structure of the obtained data is much more complex and difficult to analyze using chemometric bilinear methods and special pretreatment techniques are required.15 In the case of spectra with overlapping signals, the application of chemometric methods can be of great help, allowing the extraction of hidden information from the N-dimensional NMR data sets. Quantitative information in addition to qualitative and structural information can also be obtained. In this work, multivariate curve resolution alternating least squares (MCR-ALS),16-17 a soft-modeling chemometric method, is tested in the study of a set of 2D NMR spectra recorded along the reaction between 15N-labeled cisplatin and the methionineguanine conjugate Phac-Met-linker-p5′dG.18 This reaction is very slow; therefore, no changes in reaction speciation are expected during 2D NMR spectra acquisition. Cisplatin is a highly effective anticancer drug whose therapeutic effect is considered to result from the formation of a chelate with two adjacent guanine bases in a DNA chain.19 The formation of this chelate bends DNA and may hijack transcription factors away from their binding site or protect this lesion from being repaired.20 It has also been suggested that proteins might act as a cisplatin reservoir and mediate its transfer to DNA.21 The investigation of the reaction of cisplatin with the PhacMet-linker-p5′dG conjugate was aimed at providing information on the coordination preferences of cisplatin when exposed simultaneously to nucleic acid and protein binding sites, to draw conclusions concerning the possible transfer of the platinum complex from proteins to nucleic acids.18 The progress of the reaction was followed by HPLC, as well as 1D NMR, and the different species were isolated and fully characterized.18 These studies indicated that when 15N-labeled cisplatin was added to the conjugate, two monofunctional adducts were formed, one in which platinum was coordinated to the S atom of methionine (Pt-S adduct, major product) and one in which the metal was linked to the N7 atom of guanine (Pt-N7 adduct, minor product). These two monofunctional adducts evolved yielding a unique intermediate, the Pt-S,N7 chelate, which was transformed into a new chelate in which the ammine group trans to the S atom had been lost. The reaction between 15N-labeled cisplatin and Phac-Met-linker5′ p dG was also followed by [1H, 15N]-heteronuclear single-quantum coherence (HSQC) NMR spectroscopy, which is especially useful to monitor complexation reactions in a rather simple way.22 In this experiment, the use of HSQC pulse sequences allows the detection of 15N through 1H with high sensitivity. Only protons directly attached to 15N are detected, and all other 1H resonances are eliminated. For the system studied in this work, the power of (15) Jaumot, J.; Vives, M.; Gargallo, R.; Tauler, R. Anal. Chim. Acta 2003, 490, 253-264. (16) Tauler, R. Chemom. Intell. Lab. Syst. 1995, 30, 133-146. (17) Tauler, R.; Smilde, A.; Kowalski, B. J. Chemom. 1995, 9, 31-58. (18) Marchan, V.; Moreno, V.; Pedroso, E.; Grandas, A. Chem.-Eur. J. 2001, 7, 808-815. (19) Lippert, B. Cisplatin. Chemistry and Biochemistry of a Leading Anticancer Drug, 1st ed.; Helvetica Chimica Acta/Wiley-VCH: Zurich/Weinheim, 1999. (20) Jamieson, E. R.; Lippard, S. J. Chem. Rev. 1999, 99, 2467-2498. (21) Reedijk, J. Chem. Rev. 1999, 99, 2499-2510. (22) Bodenhausen, G.; Ruben, D. J. Chem. Phys. Lett. 1980, 69, 185-192.

this technique lies in the fact that the ammine groups coordinated to platinum appear at three different spectral regions, with 15N chemical shifts depending on which is the ligand located at the trans position (S, N, and Cl, or O).14 This study definitely confirmed the formation of complexes with platinum-sulfur linkages and the loss of the ammine trans to sulfur from the PtS,N7 chelate, but the high number of overlapping signals in the spectra recorded over the evolution of the reaction prevented the structures of the different species from being inferred from these data. In this work, the number and structures of the different species involved in the process have been reexamined using multivariate curve resolution alternating least squares. The whole set of 2D NMR spectra acquired along the reaction have been reanalyzed, and the individual 2D NMR spectra of the different species have been elucidated. From these data analyses, kinetic information about the reaction and structural information on the detected species can also be deduced. EXPERIMENTAL SECTION Materials and Methods. Peptide-oligonucleotide hybrid and 15N-labeled cisplatin were synthesized as previously described.18,23 The sample for NMR measurements was prepared by dissolving 10.5 OD254 of hybrid Phac-Met-linker-p5′dG (lyophilized) in 216.3 µL of H2O and 35 µL of D2O (including 3 µL of a dioxane solution at 0.1% in H2O). Next, 1.1 equiv (98.7 µL) of an aqueous solution of a 10 mM aqueous solution of 15N-labeled cisplatin was added, obtaining a final hybrid concentration of 2.56 mM. The reaction mixture was stirred and transferred to an NMR Shigemi tube. Temperature was kept constant at 37 °C in a thermostat bath. At 23 different reaction times, two-dimensional [1H,15N] HSQC NMR spectra were recorded using the standard sequence in a Varian spectrometer operating at 500 MHz and processed with the XWIN NMR software. Water suppression was achieved by including a WATERGATE module in the pulse sequence prior to acquisition.24 Adquisition time for each spectrum oscillated between 5 and 10 min. Some precipitate was observed after 15-h reaction time. The different products formed throughout the reaction process were isolated by HPLC and fully characterized by UV spectroscopy, 1H NMR, and mass spectrometry. A [1H,15N] HSQC NMR spectrum was also recorded for the final chelate (labeled 5 in Scheme 1b). The Mestre-C 3.0 program was used for processing free induction decay data by Fourier transform with exponential and Gaussian apodization, automatic phase correction, and linear prediction filling (1024 points) in both directions of the twodimensional data set.25 A high-pass filter of sensibility 8 was applied at a 1H chemical shift signal of 4.76 ppm in order to remove the water signal. Experimental spectra were then exported as an ascii file and loaded into the MATLAB workspace.26 Selection of the chemical shift regions where signals for the monitoring of the reaction appeared was performed. Only 1H chemical shift signals (23) Marchan, V.; Rodriguez-Tanty, C.; Estrada, M.; Pedroso, E.; Grandas, A. Eur. J. Org. Chem. 2000, 13, 2495-2500. (24) Piotto, M.; Saudek, V.; Sklenar, V. J. Biomol. NMR 1992, 2, 661-665. (25) Cobas, J. C.; Sardina, F. J. Concepts Magn. Reson., Part A 2003, 19A, 8096. (26) Matlab version 6.5., The Mathworks Inc, Natick, MA.

Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

7095

Scheme 1. Reaction Pathwaysa

a (a) Structure of the methionine-guanine hybrid. Arrows indicate the active sites of methionine (S) and guanine (N7). (b) Proposed reaction mechanism: 1, 15N-labeled cisplatin, 2, Pt-N7 monofunctional adduct, 3, Pt-S monofunctional adduct, 4, Pt-S,N7 chelate, and 5, Pt-S,N7 final product.

between 3.8 and 4.50 ppm (136 points) and 15N chemical shift signals between -310 and -275 ppm (751 points) were considered for the further data analysis. HSQC pulse sequences allowed the detection of different NMR chemical shift signal regions. However, the resolution of the spectra recorded at each reaction time was rather poor because of a considerable signal overlapping. Data Treatment. Investigation and resolution of experimental 2D NMR spectra has been performed using MCR-ALS, 16-17 which has been already applied to the analysis of different types of spectroscopic data obtained in biochemical and biophysical processes.27-29 The first step of this data analysis is the estimation of the number (Ns) of independent NMR spectroscopically active components present in a single 2D NMR spectrum. This number may be estimated by singular value decomposition (SVD)30,31 of the corresponding 2D NMR spectrum data matrix. In this estimation, singular values associated with experimental noise are assumed to be significantly lower than those singular values associated with systematic chemical data variance. (a) Procedure for the Analysis of a Single 2D NMR Spectrum. Each one of the 2D NMR spectra gives a data matrix R of dimensions (I,J), where I is the number of measured 15N chemical shifts and J is the number of measured 1H chemical shifts. Assuming a bilinear model, the MCR method performs the decomposition of the experimental 2D NMR spectrum data matrix (R) into the product of two data matrices X of dimensions (I,Ns) and YT of dimensions (Ns,J), where Ns is the number of components or species assumed to be present in the system. Columns of matrix X give the estimation of the pure 1D 1H NMR correlation species spectra, and rows of YT matrix give the (27) Jaumot, J.; Escaja, N.; Gargallo, R.; Gonzalez, C.; Pedroso, E.; Tauler, R. Nucleic Acids Res. 2002, 30, e92/1-e92/18. (28) Navea, S.; de Juan, A.; Tauler, R. Anal. Chem. 2003, 75, 5592-5601. (29) Jaumot, J.; Avin ˜o, A.; Eritja, R.; Tauler, R.; Gargallo, R. J. Biomol. Struct. Dyn. 2003, 21, 267-278. (30) Golub, G.; Loan, C. V. Matrix computations, 3rd ed.; The Johns Hopkins University Press: Baltimore, 1996. (31) Malinowski E. R. Factor Analysis in Chemistry, 3rd ed.; Wiley-VCH: New York, 2002.

7096

Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

estimation of the pure 1D 15N NMR correlation species spectra. Residual or error matrix E of dimensions (I,J) includes data variation not explained by the resolved profiles in X and YT matrices. This data matrix decomposition can be simply written as

R ) X YT + E

(1)

Solving eq 1 requires an initial estimation either of X or YT. In this work, initial estimations of YT were obtained from detection of purest variables.32 Equation 1 is then solved iteratively by ALS, which estimates X and YT matrices that optimally fit experimental data using the proposed number (Ns) of species. During this ALS optimization, non-negativity constraints were applied to restrict solutions to be positive or zero.33-34 All MCR-ALS resolution calculations were performed using in-house MATLAB routines.35 (b) Procedure for the Analysis of the Whole 2D NMR Reaction Spectra. When the whole reaction was considered, multiple data matrices were obtained, each one of them corresponding to a single 2D NMR spectrum acquired at a particular reaction time. A new third measurement dimension, time, is now present. The whole data set may now be represented as a data cube of 2D NMR spectra (see Scheme 2). To apply MCR simultaneously to the whole data set and obtain information about the reaction evolution, a new type of augmented data matrix was built up. This is a very convenient way of matrix augmentation to directly recover reaction concentration profiles.4,36 Each 2D NMR spectrum data matrix is unfolded in a long vector (vectorized) of dimensions I × J, giving all together a new augmented matrix of dimensions (t,I × J), where t is the number of reaction times at (32) Windig, W.; Guilment, J. Anal. Chem. 1991, 63, 1425-32. (33) Lawson, C. L.; Hanson. R. J. Solving Least-Squares Problems, 1st ed.; Prentice Hall: Englewood Cliffs, NJ, 1974. (34) Bro, R.; De Jong, S. J. Chemom. 1997, 11, 393-401. (35) Multivariate Curve Resolution Homepage: www.ub.es/gesq/mcr/mcr.htm. (36) Tauler, R.; Marques, I.; Casassas, E. J. Chemom. 1998, 12, 55-75.

Scheme 2. Unfolding Procedure and MCR Analysis of the Set of 2D NMR Spectra Acquired during the Reaction

which a NMR spectrum is measured. In this case, matrix eq 1 can be described as follows: R is the unfolded data matrix of dimensions (t,I × J). X is now the matrix of kinetic concentration profiles of dimensions (t,Ns). Y is the matrix of spectroscopic information (Ns,I × J), and E is the matrix of residuals not explained by the model (t,I × J). Kinetic concentration profiles and their associated 2D NMR spectra for each one of the species present in the system were resolved by MCR-ALS using the following constraints. For concentration profiles, non-negativity, unimodality, and closure constraints were applied in the traditional way,17,34,37-38 and for spectra non-negativity and selectivity, constraints 17 were applied. From the matrix of resolved concentration profiles (X), kinetic information was obtained by postulation and fitting of a reaction model. This kinetic data fitting was performed using a method similar to that described in previous works.39-40 Unfolded spectra (YT) were refolded to recover the 2D NMR species spectra by applying the reverse transformation of the folding process. Possible application of selectivity constraints in 2D NMR spectra deserves special attention. Once a preliminary resolution of the system was carried out, each one of the resolved 2D NMR spectra were investigated to check their reliability. In some of the preliminary resolved spectra, some spectral features in particular regions of the spectra were wrong because of unresolved rotational ambiguities.17,41 An improvement of this resolution process can be achieved by application of a selectivity threshold inequality constraint in regions of a spectrum where it is known that some signals should not be present. This type of constraint can be implemented during ALS iterations by substitution/update of large signal values in these regions by low threshold values. Signal values in these regions below threshold are not modified. Due to the iterative nature of ALS and to the fact that these constraints are really fulfilled by the data, at (37) Bro, R.; Sidiropoulos, N. D. J. Chemom. 1998, 12, 223-247. (38) de Juan, A.; Van der Heyden, Y.; Tauler, R.; Massart, D. L. Anal. Chim. Acta 1997, 346, 307-318. (39) de Juan, A.; Maeder, M.; Martinez, M.; Tauler, R. Chemom. Intell. Lab. Syst. 2000, 54, 123-141 (40) Bezemer, E.; Rutan, S. C. Chemom. Intell. Lab. Syst. 2001, 59, 19-31. (41) Tauler, R. J. Chemom. 2001, 15, 627-646.

convergence, most of the constraints are not active any longer, and the finally resolved 2D spectra will not show large values in the desired regions. RESULTS AND DISCUSSION Analysis of an Individual Data Matrix. First, MCR-ALS was applied to the analysis of a single 2D NMR spectrum, giving a two-way data matrix. The selected spectrum was that corresponding to the purified final Pt-S,N7 chelate species (labeled 5 in Scheme 1b). This spectrum was selected because it is the only one known to correspond to a single species. As described above, the final product showed two signals in the region of δ1H ) 3.98 ppm/δ15N ) -295.5 and δ1H ) 3.90/δ15N ) d - 296.5 ppm (Figure 1a). The 2D NMR spectrum of this purified product was imported into the MATLAB workspace as a data matrix of 751 rows (corresponding to the 15N chemical shifts from -310 to -275 ppm) and 136 columns (corresponding to the 1H chemical shifts from 3.80 to 4.50 ppm). Data analysis started with SVD of this 2D NMR spectrum to estimate the number of linear components needed to explain the observed data variance. Although the analyzed data matrix corresponded to a single purified product, the number of linear components needed to reproduce the data variance was clearly larger than one; i.e., the chemical rank of this data matrix (rank in the absence of experimental and instrumental noise) was higher than one. This could be a consequence of the presence of nonseparable isomer species for the same final product. But, on the other hand, this might be also a consequence of the intrinsic complexity of the signal, which needs more than one independent linear contribution to explain the presence of more than one independent correlation peak contribution. The experimental data matrix R was decomposed into the product of two matrices: the matrix X of 1D 15N NMR correlation spectra and the matrix YT of 1D 1H NMR correlation spectra. During ALS resolution, non-negativity constraints of 1D 15N NMR and of 1D 1H NMR spectra and normalization of 1D 1H NMR spectra were applied. Resolved 1D NMR correlation spectra are given in the sides of Figure 1b. Each of these spectra can be easily analyzed qualitatively by the position of the peak maxima allowing a rapid identification of the signals. The first resolved component (solid blue line) identifies the more intense signal located at δ1H ) 3.98 and δ 15N ) -295.5 ppm. The second resolved component (solid green line) identifies the less intense signal at δ1H ) 3.90 and δ15N ) -296.5 ppm. In Figure 1b, the reproduced 2D NMR spectrum using these two first contributions is given. Most of the features observed in the experimental spectra (Figure 1a) were reproduced when these two components were considered. On the contrary, if only one component was considered, only one group of the two main signal regions would have been described. The percentage of experimental data variance explained by two components was 79.1%. The fact that two components were needed for the resolution of the pure 2D spectrum of a single final product species should be related with the nonbilinear nature of this spectrum.42 Resolution using more components than two allowed a slightly better fit but explained (42) Massart, D. L.; Vandeginste, B. G. M.; Buydens, L. M. C.; de Jong, S.; Lewi, P. J.; Smeyers-Verbeke, J. Handbook of Chemometrics and Qualimetrics; Data Handling in Science and Technology 20; Elsevier: Amsterdam, 1997.

Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

7097

Figure 1. (a) 2D NMR spectrum of the Pt-S,N7 chelate labeled 5 in Scheme 1b. (b) Reproduction of this 2D NMR spectrum with two components. 1D NMR spectra at the sides are the 1H and 15N correlation spectra. Blue and green lines correspond to the first and second resolved components, respectively.

mostly noise in the spectral regions where no NMR signals were present. Analysis of the Whole Reaction. As it was described in the Experimental Section, the application of MCR to the analysis of the whole reaction 2D NMR spectra required unfolding of each one of the 2D NMR spectrum data matrices into a column vector of 102 136 elements (i.e., 136 1H × 751 15N). Unfolded vectors obtained at different reaction times were merged into a single tubewise augmented data matrix with 23 rows (corresponding to the 23 reaction times at which 2D NMR spectra were recorded) and 102 136 columns (Scheme 2). 7098 Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

Analysis of this augmented data matrix required some data pretreatment to remove noise. In the analysis of the 2D NMR spectrum of the purified product, the effect of noise was not considered relevant because of the relatively high signal-to-noise ratio of the detected signals. However, when considering the spectra recorded at the end of the reaction, signal-to-noise ratios were much poorer than for spectra recorded at the beginning of the reaction, mostly due to the lower intensity of the signals recorded at the end of the reaction as a consequence of product precipitation (see Experimental Section and below). Accordingly, PCA filtering was first applied to the tubewise augmented matrix

Figure 2. Resolved reaction concentration profiles of the whole data set. Inset: resolved concentration profiles for the first 5 h of the reaction. Circles, resolved points; solid line, interpolated concentration profiles with the Curve Fitting Toolbox of the Matlab Software. Blue markers, 15N-labeled cisplatin; green markers, Pt-S monofunctional adduct; red markers, Pt-N7 monofunctional adduct; cyan markers, Pt-S,N7 chelate; magenta markers, final product.

considering eight components to remove part of this noise from the recorded spectra. Percentage of recovered variance after PCA filtering was 84%. SVD analysis of the tubewise augmented data matrix confirmed the possible presence of four or five major components. Initial estimations of unfolded spectra were obtained at those reaction times giving purest spectra.32 Using these initial estimates, ALS matrix decomposition according to eq 1 allowed the estimation of concentration profiles for the different postulated species (matrix X) and of their unfolded 2D NMR spectra (matrix YT). Several constraints were applied during the calculation of these concentration profiles (non-negativity, unimodality, and closure) and of spectra (non-negativity and selectivity). The explained variance considering five components (for the PCA filtered data matrix) was 91.4%. An analysis of the residuals not explained by the model did not show any significant structure that could be indicative of the presence of additional species. The resolved concentration profiles are shown in Figure 2. The set of unfolded resolved spectral profiles in YT matrix were refolded to give the five data matrices corresponding to the five pure 2D NMR species spectra (Figure 3). These results agree with previous studies.18 Finally, resolved species were postulated to be 15N-labeled cisplatin, monofunctional Pt-S adduct, monofunctional Pt-N7 adduct, Pt-S,N7 chelate, and the final Pt-S,N7 chelate that had lost the ammine group in position trans to the S atom. Free 15N-labeled cisplatin (which is the only detectable species at the beginning of the reaction) reacted with the conjugate (first 5 h) to yield two monofunctional adducts. After 5 h, 15N-labeled cisplatin decreased up to a relative concentration of ∼0.1 mM and

remained practically unchanged until the end of the reaction. This fact can be explained because of the small excess of concentration of 15N-labeled cisplatin present at the beginning of the reaction in relation to the initial concentration of hybrid. The resolved pure 2D NMR spectra of 15N-labeled cisplatin (Figure 3a) shows a single very intense signal in the chemical shift region where the ammine trans to Cl or 15N signals should appear(δ1H ) 4.09 ppm/δ15N ) -304 ppm). Analysis of this resolved spectra (in the same way as described previously for the final product) showed that the number of components needed to reproduce adequately the resolved spectrum is also higher than one (chemical rank of the corresponding spectrum data matrix is larger than one). Using just one component, 84.1% variance was explained, and using two components, 94.6% was explained. This again confirms the nonbilinear nature of 2D NMR spectra. The second resolved species was assigned to the Pt-S monofunctional adduct. This assignation was based on the signal resolved at δ 15N ) -280 ppm (Figure 3b). Two signals could be observed in the recovered spectrum at δ1H ) 4.09 ppm/δ15N ) -304 ppm and at δ1H ) 4.15 ppm/δ15N ) -280 ppm. The concentration profile for this species showed an increase of its concentration at the beginning of the reaction and its disappearance when the reaction evolved. The third resolved species was assigned to the Pt-N7 monofunctional adduct. Its resolved spectrum (Figure 3c) showed two signals in the 15N region where only ammonia trans to N or Cl appeared (δ1H ) 4.40 ppm/δ15N ) -296.5 ppm and δ1H ) 4.10 ppm/δ15N ) -304 ppm). This species also appeared at the beginning of the reaction and quickly disappeared (Figure 2). Neither of the two monofunctional adducts was present at Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

7099

Figure 3. Resolved 2D NMR spectra: (a) S,N7 chelate, and (e) final product.

15N-labeled

cisplatin, (b) Pt-S monofunctional adduct, (c) Pt-N7 monofunctional adduct, (d) Pt-

Figure 4. Experimental 2D NMR spectra acquired after 2 h and 10 min of reaction. For assignation of signals see text.

concentrations higher than 0.1 mM after 10 h of reaction. The concentration of the Pt-S nonfunctional adduct was higher than the concentration of the Pt-N7 adduct during the reaction (Figure 7100 Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

2). This fact could be explained by kinetic effects favoring the Pt-S bonding.21 From these two monofunctional adducts, a more stable product was formed corresponding to a fourth component, which can be assigned to a Pt-S,N7 chelate. The maximum of its concentration was reached after 5 h of reaction. Figure 3d shows that the spectrum of this species has one signal at the S region (δ1H ) 4.41 ppm/δ15N ) -280 ppm) and a group of signals at the N region (δ1H ) 3.98 ppm/δ15N ) -295.5 ppm and δ1H ) 3.90 ppm/ δ15N ) -296.5 ppm). This means that this species presented two new different bonds in agreement with the formation of a PtS,N7 chelate. These two signals were different from those observed for the different species previously detected. Finally, a fifth species appeared showing only one of the two signals previously found for the fourth species. This species could only be resolved by forcing spectral selectivity constraints (see Experimental Section) during ALS optimization. This selectivity constraint allowed the differentiation of this species from the Pt-S,N7 chelate species simultaneously formed. Figure 2 shows that this species started to appear at 5 h after the beginning of the reaction and that it was the major product at the end of it. It gave only one group of signals at δ1H ) 3.98 ppm/δ15N ) -295.5 ppm and δ1H ) 3.90 ppm/δ15N ) -296.5 ppm, which could be explained by the loss of the ammine group trans to the sulfur atom of the conjugate. The presence of these two different signals is probably related to the fact that the final product may exist as a mixture of two diastereomers, which result from coordination of platinum to either of the lone pairs of sulfur in the final product. It is interesting to point out that these two different isomers existed also for the Pt-S,N7 chelate and that these two isomers reacted in the same way to yield the two products in which the ammine group trans to the sulfur atom of methionine was lost. The loss

of this group is not unexpected, since it is well known that sulfur atoms show a strong trans effect. In this case, unfortunately, it is not known what ligand, if any, occupies the coordination position of platinum left empty. In the previous study,18 the different steps of the reaction were inferred from HPLC analysis and characterization of the different species isolated and from HSQC NMR data on the ammines trans to sulfur. The problem with these spectra, as previously stated, was the overlap between different signals in the same spectral regions. This can be observed, for example, in the spectrum shown in Figure 4. The group of signals of the ammines trans to sulfur were easily identified. The main one corresponded to the monofunctional Pt-S adduct 2 and the minor peak to chelate 4. With respect to the other signals, the 15N chemical shifts only allowed to infer that they corresponded to ammines trans to either nitrogen or chlorine. The data obtained from multivariate curve resolution alternating least squares, as indicated above, now allow all of these signals to be identified as belonging to one of the different species present in the reaction medium at the reaction time. As mentioned above, some precipitation was observed after 15-h reaction time. This fact affects the intensity of the acquired spectra, since the concentration of species in solution decreases. The above-described results were obtained by applying a mass balance closure constraint to the ALS-resolved concentration profiles. Therefore, the experimentally observed decrease of signal intensity was transferred to the ALS-resolved 2D NMR spectra of the species formed after the 15 h of reaction and not to their ALSresolved concentration profiles. As shown in Figure 3, the intensity of the resolved signals for the bifunctional adducts (species 4 and 5) is lower than for the free 15N-labeled cisplatin and monofunctional adducts (species 2 and 3). From MCR-ALS-resolved concentration profiles (Figure 2), a kinetic model can be postulated and overall reaction rates can be obtained fitting ALS-resolved concentration profiles. Only concentration values at reaction times below 15 h were considered in this calculation: (43) Hambley, T. W. J. Chem. Soc., Dalton Trans. 2001, 19, 2711-2718. (44) Lempers, E. L. M.; Reedijk, J. Adv. Inorg. Chem. 1991, 37, 175-217. (45) Bancroft, D. P.; Lepre, C. A.; Lippard, S. J.. J. Am. Chem. Soc. 1990, 112, 6860-6871.

cisplatin + hybrid f S-adduct cisplatin + hybrid f N7-adduct N7-adduct f S, N7 chelate S-adduct f S, N7 chelate S, N7 chelate f final product

k1 ) 0.4 M-1 s-1 k2 ) 0.3 M-1 s-1 k3 ) 2.8‚10-4 s-1 k4 ) 2.1‚10-4 s-1 k5 ) 0.9‚10-5 s-1

These reaction rate values were in agreement with previously estimated values for the binding of 15N-labeled cisplatin to methionine and guanine to yield both mono- and bifunctional adducts.19,43-45 CONCLUSIONS Application of multivariate curve resolution alternating least squares has proved to be a very powerful tool to understand the evolution of complex chemical reactions monitored by 2D NMR spectroscopy. Multivariate curve resolution has been applied to the analysis and resolution of 2D NMR experimental spectra acquired for the reaction between 15N-labeled cisplatin and a peptide-nucleotide hybrid, affording qualitative, structural, and quantitative kinetic information from the resolved spectra and concentration profiles. Resolution of the spectra has allowed the identification of different chemical species formed during the reaction. Resolved concentration profiles allowed the postulation of a model for the mechanism of the reactions under study, and the kinetic reaction rates of these reactions were estimated. From these results, information about cisplatin binding to proteins and nucleic acids through Pt-S and Pt-N7 bond has been obtained. ACKNOWLEDGMENT J.J. acknowledges a Ph.D. scholarship from the Spanish Ministerio de Ciencia y Tecnologia (MCYT). This research was supported by the Spanish MCYT (BQU2003-00191 and BQU20013693-C02-01) and the Generalitat de Catalunya (2001-SGR00049, 2001-SGR00056 and Centre de Referencia de Biotecnologia). Received for review March 29, 2004. Accepted September 10, 2004. AC049509T

Analytical Chemistry, Vol. 76, No. 23, December 1, 2004

7101