Multivariate Resolution of Multidimesional Spectroscopic Data

Joaquim Jaumot, Raimundo Gargallo and Rom Tauler* Department of Analytical Chemistry Vicente Marchn, Anna Grandas Department of Organic Chemistry University of Barcelona, Spain
e-mail: roma@apolo.qui.ub.es

MOTIVATIONS OF THIS WORK

Multidimensional NMR is a very powerful tool for species structural elucidation and identification and may be also used for reaction and process monitoring in R+D activities.

Frequently, however, species identification and resolution problems arise because of complex overlapping non-selective signals, unknown reaction mechanisms and/or unknown mixtures.

MOTIVATIONS OF THIS WORK

Application of self- and soft-modelling multivariate curve resolution methods may extend considerably the monitoring and modelling capabilities of multidimensional spectroscopic methods.

In particular Multivariate Curve Resolution (MCR) is proposed for the investigation of complex chemical processes and reactions monitored by multidimensional spectroscopic methods.

Scope:

1. Description of the problem to solve

2. Multivariate Curve Resolution 3. Results and Discussion

4. Summary and Conclusions

Multidimensional Spectroscopic Data 2D [1H,15N]-HSQC NMR correlated spectrum

1H,15N

couplings information about bonded N-H atoms

Multidimensional Spectroscopic Data 2D [1H,15N]-HSQC NMR mixture correlated spectrum

Study of the cis platination reaction between:

O O NH NH O P O
-

NH

O O

NH2

dG Methionine
S OH

H3C

Cl
H315N Pt
15NH 3

oligopeptide methionine-guanine conjugate Phac-Met-linker-p5dG Phac = phenylacyl Cl

+
cisplatin [15N]-cis-dichlorodiammineplatinum(II)

V.Marchn, V.Moreno, E.Pedroso and A. Grandas Chem. Eur. J., 2001, 7 (4), 808-15

2D-NMR reaction monitoring

[1H,15N]-HSQC NMR correlated spectroscopy Cl H315N Experimental conditions of the reaction

Met

dG

+ cisplatin

Pt
15NH 3

Cl

15N

labelled cisplatin 751 15N chemical shifts 156 1H chemical shifts

- reaction time: hours (slow reaction) - at several times one 2D-NMR spectrum - 23 - 2D NMR correlated spectra were measured

D (23,156,751)

2D-NMR reaction monitoring

Only the d 15N ppm spectral window where correlations with 15N labelled NH3 groups of cisplatin in position trans were observed, is studied Observed d 15NH3 values vary as a function of the bonded ligands at trans In the 2D NMR plots, two main regions are present: d 15NH3 trans to S (methionine) at higher d 15N ppm d 15NH3 trans to N and/or Cl at lower d 15N ppm

d15NH3 trans to S (methionine) dregion

R
H315N
d5NH3 trans to Cl and/or N region

Pt
15NH 3

Final Product cisplatin Monofunctional Pt-S adduct

N or Cl trans adducts

Chelate Pt-S,N7

Chelate Pt-S,N7

2D-NMR reaction monitoring

Main information gathered from NMR experiments: 1) Determination of signals corresponding to methionine S groups trans to labeled 15N NH3 groups of cisplatin: 1a. Formation of monofunctional adducts 1b. Formation of chelate complexes

2) Loss of NH3 groups with time

751 d 15N 156 d 2D-spectrum 1H

Data structure:
751 d 15N

t1

156 d 2D-spectrum 1H

t2

multiple 2D data matrices at different reaction stages

751 d 15N
156 d 2D-spectrum 1H

Evolving reaction

t21

751 d 15N

156 d 2D-spectrum 1H

t22

751 d 15N

156 d 2D-spectrum 1H

t23

S adduct cisplatin

N adduct guanine

S adduct cisplatin

N adduct guanine

chelate

chelate

S adduct cisplatin

N adduct guanine

chelate

chelate

final product

cisplatin

chelate

chelate

final product

cisplatin

chelate

chelate

final product

cisplatin

chelate

chelate

final product

cisplatin

chelate

chelate

Multidimensional Spectroscopic Data Interpretation of overlapped 2D-NMR signals in evolving mixtures of chemical compounds (e.g. in reactions) is difficult using conventional methods of univariate data vanalysis

Multivariate methods of analysis

Multivariate Curve Resolution

Scope:

1. Description of the Problem to Solve

2. Multivariate Curve Resolution 3. Results and Discussion

4. Summary and Conclusions

Multivariate Curve Resolution

(reaction/process modelling)
NC
1 0.8 0.6 0.4 0.5 0.2 0 0 0 0 1.5

NC

ST C
NR
20 40 60 80 100

ST

E
NR

10

20

30

40

NC
1.5

D
NR
10 20 30 40 50 60 70 80 90

0.5

N d c s e ij k1 ik kj ij

0 0

Bilinearity!

Multivariate Curve Resolution

(HPLC modelling)
2 x 10
-5

x 10 3.5

1.5

3 2.5 2

ST

ST

1.5
0.5 1 0.5 0 0 20 40 60 0 0 20 40 60 80 100

C
NR

E
NR

NC
1.2 1 0.8

NC

D
NR
10 20 30 40 50 60

0.6
0.4 0.2 0 -0.2 0

N d c s e ij k1 ik kj ij

Bilinearity!
LC-DAD coelution problem

Spectroscopic Imaging Data 2D Raman data

Number of pixels (x x y)

y x
Pixel Scanned surface

2000

2000

1000

1000

000

0 0

50

100

150

200

250

3000 2000 2000 1000 1000 000 0 0

50

100

150

200

250

D Data set

Chemical measurement

Multivariate Curve Resolution

(spectroscopic image systems modelling)

pure spectrum

=
C D
((m x n) x p) unfolded/matricized ((m x n) x s)

(s x p)

600 000 400 300 200 100

reshape as sampling surface

0 0

50

100

150

200

250

60 50 40 30 20

distribution map
(m x n)

10 5 10 15 20 25 30 35 40 45 50 55 60

Multidimensional Spectroscopic Data

751 d 15N
15N
0.14

NMR correlated spectrum

0.12

plot of columns

0.1

156 d 1H

0.08

2D-NMR correlated spectrum data matrix

0.14

0.06

1H

NMR correlated spectrum

0.04

plot of 751 1H correlated NMR spectra

0.02

20

40

60

80

100

120

140

160

plot of rows
0.12 0.1

0.08

0.06

0.04

plot of 156 15N correlated NMR spectra

0.02

100

200

300

400

500

600

700

800

Contour plot: heteronuclear correlated spectra

1D NMR 15N correlated pure spectra (SNT)

Multivariate Curve Resolution

(2D-spectra modelling)
SNT

1D NMR 1H correlated pure spectra (SHT)

SHT

NR

E
NR

NC

2D-heteronuclear correlated NMR spectrum

D
NR

N d s s e ij k1 ik kj ij

Bilinearity!

Multivariate (Soft) Self Modeling Curve Resolution

Multivariate Curve Resolution (MCR) methods have been shown to be powerful self-soft-modeling tools able to investigate complex chemical systems with a minimum number of assumptions.
Alternating Least Squares (ALS) has become a popular method for Multivariate Curve Resolution (MCR) due to its flexibility in constraint implementation during the optimization of resolved profiles.

Flowchart of MCR-ALS
PCA EFA

D
1 purest
FSMWEFA

2
Constraints:
Natural Selectivity Local Rank Shape Equality Correlation Hard model .......... Quantitative Information N.components Initial eatimates Local Rank

ALS
5

ST
Qualitative Information

Fit and Diagnostics

Spectra selectivity constraint

selectivity ST*
0.8

STc < threshold values

0.8 0.7

0.7

0.6

0.6

0.5

0.5

0.4

0.4

0.3

0.3

0.2

0.2

0.1

0.1

100

200

300

400

500

600

700

800

100

200

300

400

500

600

700

800

MCR-ALS applied to a 2D NMR data matrix

1D NMR 15N correlated spectra (SNT)

MCR-ALS
Non-negativity constraints

2D-heteronuclear NMR correlated spectrum 1D NMR 1H correlated spectra (SHT)

MCR-ALS applied to multiple correlated data matrices (three-way data analysis)

- Improvement of solutions (higher sobredetermination of the system of linear equations) - Improvement of the unique resolution conditions (decrease of rotational ambiguities effects )
-Resolution of rank deficiency problems associated to the analysis of chemical reaction and closed data systems

MCR-ALS applied to multiple correlated 2D NMR data matrices (three-way data analysis)
Dt(23,751x156) C (23,4) ST (4,751x156) MCR-ALS 23 23 .......... = times 15 751 d N 156 d 1H times
X

751 ............................................ 751 d 15N 751 d 15N X 156 d 1H

refolded 2D-NMR pure species correlated spectra 156

0.9

C (23,4)
Kinetic concentration profiles

0.8

0.7

0.6

0.5

156
0 5 10 15 20 25

0.4

0.3

0.2

0.1

401

401

23

ST (4,751x156)

Scope:

1. Description of the Problem to Solve

2. Multivariate Curve Resolution 3. Results and Discussion

4. Summary and Conclusions

Noise filtering by PCA

Percent Variance Captured by PCA Model Principal Eigenvalue % Variance % Variance Component of Captured Captured Number Cov(X) This PC Total --------- ---------- ---------- ---------1 6.82e-004 57.81 57.81 2 2.09e-004 17.75 75.57 3 7.27e-005 6.16 81.73 4 4.43e-005 3.76 85.48 5 2.60e-005 2.21 87.69

0.12 140 0.1 120

100

0.08

80

experimental
0.06 0.04

60

40

20

0.02

100

200

300

400

500

600

700

0.05 140 0.06 140 0.04 0.05

120

120

0.03

100

0.04

100

80

PCA filtered

noise filtered

0.02

0.01

0.03 80 0 0.02

60 0.01 40 0 20 -0.01 100 200 300 400 500 600 700

60

-0.01

-0.02 40 -0.03 20 -0.04

100

200

300

400

500

600

700

Example of MCR-ALS resolution of a single 2D NMR spectrum to recover the original 1D NMR correlated spectra

Is this really a pure compound? 3 or 4 components are needed to reproduce the observed data variance!

Recovery of 1D NMR correlated spectra from 2D NMR correlated spectrum of the final product by MCR-ALS N spectra
15

MCR-ALS 2D NMR 1H, 15N spectrum

1H

spectra

1D NMR 15N correlated spectra

1D NMR 1H correlated spectra

spectral non-linearities? impure product? isomers?

Noise!

Mostly Noise!

Multidimensional Spectroscopic Data 2D NMR data structure

unfolding

unfolding

Evolving reaction

unfolding

unfolding

751 d 15N

Dc(156x23,751)
401 d 15N 156 d 2D-spectrum 1H 156 d 2D-spectrum 1H 156 X 23 times 156 d 2D-spectrum 1H

156 d 1H

2D-spectrum
751 d 15N 156 d 1H

D(23,156,751

Dr(401x23,62)
156 d 1H 751 d 15N

2D-spectrum

751 X 23 times

Dt(23,751x156)

23 times 751 ............................................ 751 d 15N

751 X 156 d 1H
7511 d 15N

time = 23, 1H spectra = 156 channels , 15N spectra = 751 canals Row-wise Augmented Data Matrix 1 d 1H Column-wise Augmented Data Matrix 2 d 15N

15N

Dr

d 1H

Dc
Dimensions 751 x 3588

Dimensions 17273 x 156 Tube-wise Augmented Data Matrix 3 time

Dt
Dimensions 23 x 117156

SVD analysis of 2D NMR spectra Dt(23,751x156)

3 or 4 components?

Recovery of kinetic information MCR-ALS analysis of tube-wise augmented matrix Dt(23,751x156) MCR-ALS C (23,4)
23 times 751 ............................................ 751 d 15N 751 d 15N X 156 d 1H

ST (4,751x156)
751 d
15N

23 times

.......... 1
X

156 d H

refolded 2D-NMR pure species spectra 156

0.9

C (23,4)
Kinetic concentration profiles

0.8

0.7

0.6

0.5

156
0 5 10 15 20 25

0.4

0.3

0.2

0.1

401

401

23

ST (4,751x156)

R2 0.9951 Fitting filtered PCA reproduced data R2 0.9287 Fitting experimental data
1 0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

10

15

20

25

Recovery of kinetic information MCR-ALS analysis of tube-wise augmented matrix usign spectra selectivity constraint Dt(23,751x156)
23 times 751 ............................................ 751 d 15N 751 X 156 d 1H

C (23,4)

ST (4,751x156)

23 .......... times 751 X 156 d 1H constrain refolded two last 2D-NMR species pure species spectra spectra 156

C (23,4)
Kinetic concentration profiles

0.9

0.8

0.7

0.6

0.5

156 751 751

0.4

0.3

0.2

0.1

10

15

20

25

23

ST (4,751x156)

0.9

Using spectra selectivity constraint R2 0.9370 Fitting PCA filtered reproduced data R2 0.8950 Fitting experimental data

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

10

15

20

25

Interpolated concentration lines kinetic constants rate law constraints

S1

Cl H3N Pt NH3 Cl

S2

Met H3N Pt NH3 Cl

dG

+
Met
Cl

dG
Pt NH3

NH3

S3

Met

dG

Pt
H3N NH3

S4

Met Pt H3N

dG

Met

dG

1 H3N

Met Pt NH3 Cl

dG

Met Cl

dG Pt NH3

2 NH3

+ cisplatin

Proposed reaction mechanism

Reaction: 1
2 NH3 Trans Met Monofuntional NH3 Trans dG Monofuntional NH3 Trans Metal Chelate Final Product

Met 3 H3N Pt

dG metal chelate NH3 Trans to Met NH3

Met
4 H3N Pt

dG
Final Product

3
4

MCR-ALS analysis of pure 2D-NMR resolved spectra

1st species

2nd species

3rd species 4th species

MCR-ALS Analysis of 1st species 2D NMR correlated spectrum Recovery of 1D NMR correlated spectra of 1st species

cisplatin

non-lineraities? shifting?

MCR-ALS Analysis of 2nd species 2D NMR correlated spectrum Recovery of 1D NMR corrrelated spectra of 2nd species

Cisplatin, N-adduct guanine

S-adduct

MCR-ALS Analysis of 3rd species 2D NMR correlated spectrum Recovery of 1D NMR correlated spectra of 3rd species

S chelate

10-3

N chelate

MCR-ALS Analysis of 4th species 2D NMR correlated spectrum Recovery of 1D NMR correlated spectra of 4th species

final product

10-4

Noise, Non-linearities impurities

Scope:

1. Description of the Problem to Solve

2. Multivariate Curve Resolution 3. Results and Discussion

4. Summary and Conclusions

Summary of results
Dt(23,751x156)
Tube-wise augmented data matrix containing the 23 2D NMR correlated spectra at the 23 times of reaction
751 ............................................ 751 d 15N 1 cisplatin 2 mixture of S- and N- adducts 3 S-N chelate 4 loss of NH3 group

751 X 156 d 1H

MCR-ALS

3 3 4

Conclusions:
1. The reaction* between 15N-labeled cisplatin, [15N]-cisdichlorodiammineplatinum(II), and a methionine-guanine conjugate, Phac-Met-linker-p5'dG (Phac = phenylacetyl), could be investigated using [1H,15N]-HSQC NMR spectroscopy (giving a set of 2D NMR correlated spectra obtained at the different stages of the reaction) and MCR-ALS of the augmented tube-wise (time direction) matrix. 2. MCR-ALS of this data set provided estimates of A) kinetic concentration profiles for each one of the four detected components of the reaction, from which reaction mechanisms and reaction kinetics can be deduced B) 2D 'pure' NMR correlated spectra of these four components from which their identification and structure can be deduced C) separate contributions of each of the individual 1D NMR spectra signals assigned to particular chemical groups.