NATURAL ANTIOXIDANTS AND HIGH PERFORMANCE LIQUID CHROMATOGRAPHY (HPLC) HYPHENATED SCREENING TECHNIQUES A thesis submitted in accord with the requisites of the Degree of Doctor of Philosophy (Science) By Mariam Mnatsakanyan School of Natural Sciences University of Western Sydney New South Wales, Australia May 2010
194
Embed
NATURAL ANTIOXIDANTS AND HIGH …Multidimensional High Performance Liquid Chromatography in Hyphenated and Alternative Methods of Detection in Chromatography. Editor R.A. Shalliker,
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
NATURAL ANTIOXIDANTS AND HIGH
PERFORMANCE LIQUID CHROMATOGRAPHY
(HPLC) HYPHENATED SCREENING TECHNIQUES
A thesis submitted in accord with the requisites of the Degree of Doctor of
Philosophy (Science)
By
Mariam Mnatsakanyan
School of Natural Sciences
University of Western Sydney
New South Wales, Australia
May 2010
“Happiness lies in the joy of achievement and the thrill of creative effort”
F. D. Roosevelt
i
TABLE OF CONTENTS
Table of Contents .......................................................................................................... i
Statement of Authentication ................................................................................ vi
Acknowledgements ................................................................................................... vii
Publications Arising From This Thesis ....................................................... viii
List of Abbreviations ................................................................................................. ix
List of Tables ................................................................................................................... xi
List of Figures ................................................................................................................ xii
Abstract .......................................................................................................................... xviii Preface .................................................................................................................................1
2.5 Data Analysis ............................................................................................................... 42
Chapter 3
High performance liquid chromatography with two simultaneous on-line antioxidant assays: Evaluation and comparison of espresso coffees ................................................................................................................................. 44
A Discussion on the Process of Defining Two-Dimensional Separation Selectivity ................................................................................................ 57
The Assessment of Selective Stationary Phases For Two-Dimensional HPLC Analysis of Foods: Application to the Analysis of Coffee ................................................................................................................................. 76
The Analysis of Café Espresso using Two-Dimensional Reversed Phase-Reversed Phase High Performance Liquid Chromatography with UV-Absorbance and Chemiluminescence Detection .................... 104
The work presented in this thesis is, to the best of my knowledge and belief, original
except as acknowledged in the text. I hereby declare that I have not submitted this
material, either in full or in part, for a degree at this or any other institution.
Mariam Mnatsakanyan
May 2010
vii
ACKNOWLEDGEMENTS
I take this opportunity to express my gratitude to many people for their support and
friendship during the time I spent at the University of Western Sydney.
First and foremost, I wish to sincerely thank my principal supervisor, A/Prof.
Andrew Shalliker, who generously and tirelessly offered his advice, assistance and
inspiration through the period of this research. This work could not been succeeded
without Andrews encouragement and help. I have learned a lot from him.
I want to appreciate Prof. N.W Barnett and Dr X.A. Conlan from Deakin University
(School of Life and Environmental Sciences) for their expertise provided during this
work.
I appreciate the support of my panel supervisors, Dr Rosalie Durham and A/Prof.
Kaila Kailasapathy. I want to thank late A/Prof. Geoff Skurray for the interesting
conversations. Special regards to Dr Michael Phillips for his assistance with any
questions that I might have had.
I am grateful to Paul Stevenson for his friendship and for writing an algorithm that
enhanced the data analysis during this research.
My deepest appreciation goes to the International Postgraduate Research
Scholarship (IPRS) scheme of University of Western Sydney, for providing me with
the necessary financial assistance. Special regards to Tracy Mills, for the given
opportunity.
I would like to express my gratitude to all my colleagues and friends; Arianne, Kirsty,
Coleen, David for their friendship and support throughout the period of my studies.
I’m grateful to Mr Steven MacJohn for always being there for me. I am indebted to
my parents for their endless love.
viii
PUBLICATIONS ARISING FROM THIS THESIS
Refereed Journal Publications
1. Mnatsakanyan M., Goodie T.A., Conlan X.A., Francis P.S., McDermott G.P., Barnett N., Shock D., Gritti F., Guiochon G., Shalliker R.A., High performance liquid chromatography with two simultaneous on-line antioxidant assays: evaluation and comparison of espresso coffees. Talanta 81(3) (2010); 837.
2. Mnatsakanyan M.; Stevenson P.G.; Shock D.; Shalliker R.A.: The
Assessment of Stationary Phases for Natural Products. Talanta 82(4) (2010); 1349.
3. Mnatsakanyan M., Stevenson P.G., Conlan X.A., Francis P.S., Goodie T.A.,
McDermott G.P., Barnett N.W., Shalliker R.A., The Analysis of Café Espresso using Two-Dimensional Reversed-Phase High Performance Liquid Chromatography with UV-Absorbance and Chemiluminescence Detection. Talanta 82(4) (2010); 1358.
4. Stevenson P.G., Mnatsakanyan M., Francis A.R., Shalliker R.A., A
Discussion on the Process of Defining Two-Dimensional Separation Selectivity. Journal of Separation Science 33 (2010); 1.
5. Stevenson P.G., Mnatsakanyan M., Shalliker R.A., Peak Picking from 2D-
G.: Application of Power Functions to Chromatographic Data for the Enhancement of Signal to Noise Ratios and Separation Resolution. Journal of Chromatography A 1217(36) (2010); 5693.
7. McDermott G.P., Noonan L.K., Mnatsakanyan M., Shalliker R.A., Conlan
X.A., Barnett N.W., Francis P.S., The on-line HPLC-DPPH• antioxidant assay: methodological considerations and application to highly complex samples. Analytica Chimica Acta 675 (2010); 76.
Book Chapters
1. Milroy C., Stevenson P.G., Mnatsakanyan M. and Shalliker R.A., Multidimensional High Performance Liquid Chromatography in Hyphenated and Alternative Methods of Detection in Chromatography. Editor R.A. Shalliker, Publishers Taylor and Francis, (2010); (In submission).
ix
LIST OF ABBREVIATIONS
2D HPLC: Two-Dimensional High Performance Liquid Chromatography
Table 2.1 Limits of Detection (LoD) and Limits of Quantification (LoQ) of
standard antioxidants by DPPH detection. Table 2.2 Intermediate precision of the retention time and peak area of standard
antioxidant compounds with a DPPH detector within 3 days interval (n = 3).
Table 3.1 Key peaks in the chromatograms for the Ristretto coffee sample
obtained using UV-absorbance, DPPH and chemiluminescence modes of detection.
Table 4.1 Summary of the statistical measures of the peaks separated with the
different thresholds and in the different zones. Table 5.1 Preliminary assessment of 2D HPLC separation performance during
solvent selectivity studies. Table 5.2 GAFA calculations for the 2D HPLC separations and in each of the
quadrants. Table 5.3 Mass Spectra data of protonated (a) and deprotonated (b) 21
compounds in Ristretto and their retention times on the first (CN, PFP, PHX) and second (C18) dimensions.
Table 6.1 Number of peaks detected for each café espresso flavour for both UV-
absorbance and chemiluminescence detection. Table 7.1 Caffeine second dimension retention times in three coffees and it‟s
Mean and StDev at 95% confidence level. Table 7.2 Reproducibility of the first and second dimensional retention times in
the segmented area between 3.6 to 6.6 min, represented as the Mean of three injections ± StDev.
Appendix I
Table I.1 Simulated results of a single peak when applied to smoothing and polynomial functions. Rt and variance were calculated with the peak moments method.
xii
LIST OF FIGURES
Figure 1.1 Effective non-orthogonal, two-dimensional retention space where the
peak spreading angle is Figure 1.2 Illustration of sample dimensionality (a) polycyclic aromatic
hydrocarbons (size and shape), and (b) diastereoisomers of n-butyl polystyrene oligomers with n = 2 to n = 5 styrene repeating units.
Figure 1.3 ODS C18 column embedded with (a) carbamate and (b) amide polar
groups. Figure 1.4 Phenyl functional group in FluoroSep RP Phenyl column. Figure 1.5 Diagrams of (a) Heart-cutting and (b) Comprehensive switching valves. Figure 1.6 Surface plot of 2D HPLC separation of a mixture of 35 alkyl benzenes
by Ikegami et al., [185]. Figure 2.1 Diagram of the flow system for antioxidants screening based on UV,
DPPH and CL detection. Figure 2.2 Separation and identification of antioxidants in apple flesh (Granny
Figure 2.3 Chemical structures of (a) caffeic acid, (b) gallic acid, (c) catechol, (d)
ascorbic acid. Figure 2.4 Linear dependence of the peak areas on the tested concentrations
detected by DPPH assay. Figure 3.1 Chromatograms for the Ristretto sample, separated on (a) SphereClone
and (b) Kinetex columns. Response for UV-absorbance detection and
DPPH assay shown. Figure 3.2 Caffeine standard at 1 mg/mL on SphereClone C18 column with UV-
absorbance and DPPH detection response shown. At 5% min-1 aqueous/methanol gradient going from 0 to 100% methanol.
Figure 3.3 Chromatograms for separation on Kinetex column and UV-absorbance
detection, of Ristretto, Volluto and Decaffeinato café espresso samples.
xiii
Figure 3.4 (a) Chromatograms for separation of Ristretto coffee with (A) UV-absorbance detection, (B) DPPH decolourisation assay, and (C) acidic potassium permanganate assay. Figure 3.4(b) and Figure 3.4(c): as above, with close up view of 0-10 min and 10-20 min respectively.
Figure 3.5 Chromatograms for Ristretto, Volluto and Decaffeinato samples: (a)
acidic potassium permanganate assay and (b) DPPH decolourisation assay.
Figure 4.1 One-dimensional chromatograms Ristretto on (a) Cyano and (b) C18
stationary phases. Both columns 150 4.6 mm; 5 m Pd, mobile phase was aqueous/methanol going from 100% water to 100% methanol at a gradient rate of 5% min-1. Flow rate of 1 mL/min. Detection at 280 nm.
column, 2nd Dimension: C18 column. Both dimension separations employed aqueous/methanol gradient elution going from 100% water to 100% methanol at a rate of 10% min-1. Flow rates in both dimensions was 1 mL/min. Injection volume in the first dimension was 100 µL, detection at 280 nm.
Figure 4.3 Scatter plots detailing the location of peak maxima across the two-
Figure 4.4 Scatter plot illustrating the correlation in retention time data in (a) Zone
3 and (b) Zone 1. Figure 4.5 Scatter plot of the retention times of the peaks contained in Zones 1 and
3. Figure 5.1 One-dimensional separations of Ristretto on (a) Cyano, (b) Phenyl-
Hexyl, (c) Pentafluoro-Phenyl, (d) Synergi-Hydro C18 and (e) C18 phases. Mobile phase was aqueous/methanol, going from 100% water to 100% methanol at a gradient rate of 10% min-1. All flow rates were 1 mL/min and injection volumes were 100 µL.
Figure 5.2 Two-dimensional separations of Ristretto. First dimension (a) Cyano, (b)
Phenyl-Hexyl, (c) Pentafluoro-Phenyl and (d) Synergi Hydro-C18 and second dimension C18 phases. In both dimensions mobile phase was aqueous/methanol, going from 100% water to 100% methanol.
xiv
Figure 5.3 Scatter plots for the 2D HPLC separations with (a) Cyano, (b) Phenyl-Hexyl, (c) Pentafluoro-Phenyl and (d) Synergi Hydro-C18 first dimension columns. The quadrants are defined by the red dashed lines.
Figure 5.4 Heart-cut segment separation of Ristretto on C18 phase at 3.2 min. Figure 5.5 Structures of some compounds identified in Ristretto. Figure 5.6 Components identified in the Cyano/C18 system: Note, for the purposes
of illustration the location of the components on the 2D plot represents only the generalised location, and not the exact 2D retention time. a) caffeic acid b) malic acid c) quinic acid d) fumaric acid e) catechin f) a procyanidin dimer g) feruloylquinic acid h) ferulic acid i) 3-(4,5)-ο-caffeoylquinic acid j) 3,4-dicaffeoylquinic acid k) trigonelline l) nicotinic acid m) sucrose n) caffeine o) caffeoylquinic acid p) rutin q) acetylated hexose based oligosaccharide r) oligosaccharide containing anhydrohexose s) acetylformoin hexose based oligosaccharide t) caffeoylshikimic acid u) nicotinamide.
Figure 5.7 Components identified in the Phenyl-Hexyl/C18 system a) caffeic acid
b) malic acid c) quinic acid d) fumaric acid e) catechin f) a procyanidin dimer g) feruloylquinic acid h) ferulic acid i) 3-(4,5)-ο-caffeoylquinic acid j) 3,4-dicaffeoylquinic acid k) trigonelline l) nicotinic acid m) sucrose n) caffeine o) caffeoylquinic acid p) rutin q) acetylated hexose based oligosaccharide r) oligosaccharide containing anhydrohexose s) acetylformoin hexose based oligosaccharide t) caffeoylshikimic acid u) nicotinamide.
Figure 5.8 Components identified in the Pentafluoro-Phenyl/C18 system a) caffeic acid b) malic acid c) quinic acid d) fumaric acid e) catechin f) a procyanidin dimer g) feruloylquinic acid h) ferulic acid i) 3-(4,5)-ο-caffeoylquinic acid j) 3,4-dicaffeoylquinic acid k) trigonelline l) nicotinic acid m) sucrose n) caffeine o) caffeoylquinic acid p) rutin q) acetylated hexose based oligosaccharide r) oligosaccharide containing anhydrohexose s) acetylformoin hexose based oligosaccharide t) caffeoylshikimic acid u) nicotinamide.
Figure 5.9 2D surface plot of Synergi Hydro-C18/C18 system represented in four
quadrants. Figure 6.1 Two-dimensional separations of (a) Ristretto, (b) Decaffeinato and (c)
Volluto café espresso. First dimension Cyano and second dimension C18 phases. In both dimensions mobile phase was aqueous/methanol, going from 100% water to 100% methanol, at 10% min-1 gradient.
xv
Figure 6.2 One-dimensional separation of Ristretto (undiluted) on Cyano column. Mobile phase was aqueous/methanol, going from 100% water to 100% methanol at a rate of 10% min-1. Flow rate at 1 mL/min. Detection at 280 nm.
Figure 6.3 Heart-cut segment separation of (a) Ristretto, (b) Decaffeinato and (c)
Volluto samples on C18 column at 3.2 min. Figure 6.4 Overlay of the separations of (a) Ristretto, (b) Deccaffeinato and (c)
Volluto café espresso on C18 column heart-cut at 7.6 min. Peaks A, B, and C marker peaks in this region.
Figure 6.5 Chemiluminescence detection plots of (a) Ristretto, (b) Decaffeinato
and (c) Volluto samples. Figure 6.6 UV-absorbance and chemiluminescence detection response of heart-cut
fractions of (a) Ristretto, (b) Decaffeinato and (c) Volluto samples at 3.2 min. Blue and red lines represent UV-absorbance and chemiluminescence (CL) response, respectively.
and (d) Decaffeinato café espresso. First dimension Cyano and second dimension C18 phases. In both dimensions mobile phase was aqueous/methanol going from 100% water to 100% methanol.
Figure 7.2 Overlay of second dimension retention times of the segmented area
between 3.6 to 6.6 min in the first dimension (n = 3). Figure 7.3 Overlay of the second dimensional retention times of two independent
runs of the Ristretto. Figure 7.4 Figure 7.4(a) is the 1D separation of apple peel on a CN column using
aqueous/THF mobile phase gradient. Inset is an expanded view of the retention between 12.9 and 14.7 minutes. Figure 7.4(b) is the second dimension separation (C18 column with aqueous/MeOH) of the cut at 13.4 minutes (between 17 and 20.2 minutes as the baseline is largely flat before this section). Figure 7.4(c) represents a stacking of the 8 cuts from the expanded first dimension separation (from 13.0 to 14.4 minutes in 0.2 minute increments (200 µL cut volumes)).
Figure 7.5 Two-dimensional separations of Red Delicious apple peel methanol
extract. First dimension Cyano and second dimension C18 phases. In first dimension mobile phase was aqueous/THF going from 100% water to 100% THF at 10% min-1 gradient. In the second dimension mobile
xvi
phase was aqueous/MeOH, going from 100% water to 100% methanol at 10% min-1 gradient. Detection at 280 nm.
Figure 7.6 Two-dimensional separation of red wine using Cyano (1st dimension)
and SphereClone C18 (2nd dimension) stationary phases, with UV absorbance detection. Mobile phase composition was aqueous/THF going from 100% water to 100% THF at 10% min-1 gradient. Detection at 280 nm.
Figure 7.7 1D separations of Penfold‟s Rawson‟s Retreat red wine on (a) Luna 100
Ǻ CN column (150 × 4.60 mm × 5 M Pd). Experimental conditions: A: water; B: THF at 5% min-1 linear gradient. Flow rate 1 mL/min, injection volume 100 µL, UV/Vis at 280 nm. (b) SphereClone ODS
column (150 × 4.66 mm × 5 M Pd). Experimental conditions: A: water; B: MeOH at 5% min-1 gradient. Flow rate 1 mL/min, injection volume 100 µL, UV/Vis at 280 nm.
Figure 7.8 (a) Two-dimensional separation of Penfold‟s Rawson‟s Retreat
Cabernet sauvignon using CN (1st dimension) and C18 (2nd dimension) stationary phases, with permanganate chemiluminescence detection. (b) Enlarged and re-scaled section containing peaks for dominant antioxidant compounds.
Appendix I
Figure I.1 (a) represents a 3D surface plot of apple flesh 2D comprehensive (off-line) heart-cut separation (0.2 minute increments, 200 µL cut volumes) using 1st D CN (aqueous-THF) and 2nd D C18 (aqueous-MeOH) gradient at 10% min-1 (segment from the first dimension between 13.0 to 14.4 minutes). The z-axis scale has been restricted so the less absorbing peaks can be observed. Figure I.1(b) is a contour plot of the same data. The dark regions represent 2D peaks with the darker regions having a greater detector response.
Figure I.2 An example of how derivatives of peaks can be used to determine
retention times and peak regions from a chromatogram. Figure I.2(a) is the chromatogram to be analysed. Figure I.2(b) is an expanded section of the smoothed chromatogram; thrh2 is represented by the horizontal dashed line. The dots represent the retention time and peak height of the detected peaks. Figure I.2(c) is the first derivative of Figure I.2(b). The horizontal dashed lines represent εfd and the outer vertical lines represent the peak region define in Figure I.2(b). Figure I.2(d) is the second derivative of the chromatogram with εsd represented by the horizontal dashed line.
xvii
Figure I.3 The same separation displayed in Figure I.1 after the data has been applied to the peak detection algorithm. The blue points represent the 2D peak maxima and the red points are peaks that were detected in adjacent cuts that were deemed to be the same compound as those connected to it by a red line.
Figure I.4 (a) Gaussian peak with simulated noise that is smoothed in (b) and
applied to a polynomial function in (c). Figure I.5 (a) is an illustration of a 2D HPLC separation of an apple flesh extract.
I.5(b) to (d) are expanded regions of this separation where the response has undergone different degrees of polynomial enhancement. I.5(b) emphasises the importance of selecting appropriate thresholds when analysing the data.
Figure I.6 Expanded region of Figure I.5(a). White points represent peak
maximum and red lines join 1D peaks deemed to belong to the same component.
Figure I.7 Geometric approach to factor analysis when applied to the apple
separation.
xviii
ABSTRACT
Reactive intermediates in the oxidative processes are among the major sources of
primary catalysts that initiate oxidation in vivo and in vitro; a process that is one of
the major causes of food quality deterioration and generation of chronic diseases.
Antioxidants, both natural and synthetic, have been widely used at legal limits in
food, medical and personal care industries. However, recently, the use of synthetic
antioxidants has been severely restricted because of the evidence that suggests they
may be harmful to human health. As such, the importance of screening naturally
occurring alternatives, which are presumably safe, effective as dietary supplements
or as processing aids, is increasing. Natural products are well known for their
molecular biodiversity and are of great interest as potential source for novel
antioxidant molecules. Due to the chemical complexity of natural products,
determining their antioxidant content can be a formidable task, but at the same time
likely to be worthwhile task. This thesis investigates the hyphenation of one- and
multi-dimensional chromatography towards to profiling antioxidant content in
natural products.
The chromatographic antioxidant profiles of various espresso cafés, as complex
samples of natural origin, were used to illustrate on-line simultaneous screening and
detection of sample antioxidant components based on two rapid DPPH (2,2'-
diphenyl-1-picrylhydrazyl) and Chemiluminescence (CL) (acidic potassium
permanganate) assays. Some differences in detection selectivity were observed that
underlined the importance of multi-dimensional modes of detection in profiling
antioxidant content of multi-component natural products. Results revealed that
information gained from the on-line separation and antioxidant detection analysis can
be limited by the peak capacity of the chromatographic separations.
The separation and subsequent analysis of samples derived from natural origin
requires high peak capacity separation techniques, such as two-dimensional high
performance liquid chromatography (2D HPLC). When dealing with complex
samples, optimisation of separation selectivity or in actual fact measuring the
selectivity of potential one-dimensional (1D) systems for optimised coupling into a
two-dimensional (2D) mode is a difficult task. Standard compounds must be selected
xix
that represent accurately the sample, as selectivity differences are solute dependent.
The results in this work further verified that the performance of the 2D separation
system is highly dependent on the compounds present within the chemical matrix of
the sample. Standard compounds representative of the sample would likely yield
inaccurate measures of performance, even with careful selection. Thus to measure
the selectivity differences between the two separation dimensions the sample itself
i.e., café espresso, has been used, as using the sample was the best measure of the 2D
separation performance. Optimisation in the separation process was best undertaken
using an „incremental heart-cutting‟ approach to 2D HPLC, with a constant second
dimension separation environment to identify selectivity changes as a function of the
first dimension chromatographic environment.
Intensive selectivity studies were undertaken incorporating a total of 17 stationary
and mobile phase combinations. 2D chromatographic performance measurements i.e.,
number of separated compounds, correlation, spreading angle, use of separation
space and practical peak capacity was undertaken across the separation plane using a
Geometric Approach to Factor Analysis (GAFA). Overall the best performing 2D
HPLC system was that of Cyano/aqueous-methanol in the first dimension coupled
with C18/aqueous-methanol in the second dimension.
High Resolution Screening Technique (HRS) based on a new approach of
combination of 2D HPLC separation on-line with acidic permanganate CL detection
was applied to profile the total antioxidant content of various espresso café flavours.
Such a combination provided unprecedented resolution for antioxidant detection and
assessing the reactivity of individual components in complex samples. Detailed
information regarding the identity of individual antioxidant compounds responsible
for the total antioxidant profile of café espressos could be drawn based on mass
spectrometry (MS) analysis. Overall phenolic acids and their adducts seem to
dominate in the antioxidant profile of these samples. Such screening techniques
could offer great fingerprinting potential in food science and drug discovery with
respect to sensitivity, specificity, and time-efficiency, for the rapid identification of
radical scavengers in complex samples.
xx
Given the high content of information generated through the 2D HPLC high peak
capacity separations, and the sensitivity of CL assay, the 2D HPLC-CL technique
was applied in a brief study to profile the antioxidants of the red wine.
In conclusion the 2D HPLC separations could fundamentally improve the screening
of natural products and on-line antioxidant assays could enhance the information
content that the analyst can draw on the antioxidant profiles of the samples. This
aspect of natural product discovery is the most demanding and with such an
analytical approach new molecules may be rapidly discovered. Such screening
techniques have the potential of being beneficial to not only antioxidants and in drug
discovery processes, but also to quality control in various consumer related sectors
such as food industry.
1
PREFACE
For thousands of years, natural products have played an important role in treating
and preventing human diseases. Exploration of the biodiversity of natural products
continues to provide novel chemicals with useful bioactivity [1, 2]. It is estimated
that almost half of the drugs currently in clinical use were derived from various
natural source materials [2]. Antioxidants that have largely been derived from natural
products are an example of (bio) functional molecules used for medicinal purposes
and are continually gaining attention for their preventative and health fortifying
activity [3, 4]. The idea of „antioxidant functionality‟ reflects a major shift in
attitudes to the relationship between diet and health. But protection from hazardous
reactive oxygen species is not only of nutritional and health relevance. The
customer‟s awareness for synthetic ingredients in products has also been extended to
the cosmetic industry, specifically to add anti-inflammatory benefits against UV
irradiation on human skin [5, 6].
Natural products are characterised by extremely complex chemical matrices, where
bioactive phytochemicals of interest coexist with thousands of other compounds,
sometimes at trace concentrations. At present, application of the analytical
techniques traditionally involved in natural product screening appears limited, as we
tend to explore large numbers of complex samples of various natural origins. This
thesis addresses and investigates the practical potential of 2D HPLC as a hyphenated
screening technique to the benefit of profiling food samples for antioxidant activity.
Chapter 1 discusses the current literature on the fundamentals in the field of natural
antioxidant research. Discussion has been extended into the basic conventional
antioxidant testing methodologies involved in the antioxidant analysis, emphasizing
particularly DPPH and CL assays as the most commonly used detecting assays. An
underline has been put forward regarding the informational limitations that the
analyst can draw from such batch-type assays i.e., less specificity towards to
particular antioxidant active molecules. The hyphenated on-line techniques based on
combined application of analytical separation and antioxidant detection assays
contribute considerably to the antioxidant screening analysis and has been
acknowledged by various research groups [7, 8, 9]. Nevertheless, there is an urge for
2
high peak capacity separations to benefit from such analysis. The theoretical and
practical concepts of both one- and two-dimensional chromatography are then
discussed, underlining especially the separation limits of the former technique and
the separation power of latter technique and its suitability in natural products analysis.
Chapter 2 details general experimental protocols. Specific details of chromatographic
separations and assay methodologies are discussed in each relevant chapter.
Chapter 3 investigates the combination of UV, DPPH and CL (acidic potassium
permanganate) detectors, used simultaneously for on-line post-column antioxidant
screening analysis. Despite the similarities in the results, each detector responded
specifically to different analytes within a sample suggesting that the results generated
by the different methods cannot be directly correlated, that is being complementary
one to another. Such combinatory methods are sensitive, specific, and time-efficient
and have high potential for the rapid identification and fingerprinting of antioxidant
compounds in complex samples. The approach provides much greater information
than total antioxidant conventional measurements, but is often hindered by
insufficient resolution of chromatographic peaks due to the samples chemical
complexity. The peak capacity in liquid chromatography can, however be
substantially increased by conducting the separation across two different dimensions.
If these dimensions are orthogonal, and the sample is comprehensively analysed, the
total peak capacity of the system is the product of the individual dimensions,
ultimately increasing experimental information. This chapter is therefore the
introduction to more power separation techniques that employ multidimensional
HPLC.
Selecting an appropriate set of compounds that accurately reflect a sample‟s
chemical matrix is of critical importance, as the separation optimisation parameters
are clearly a function of the solutes contained within the complex sample. Standard
selection, with respect to accurate representation of the sample is detailed in Chapter
4. Metrics were used to assess localised measures of component distributions within
the two-dimensional separation plane. The results of this analysis of data showed that
the measure of separation quality varied markedly, depending on the elution zone for
which the test was undertaken. If the separation was optimised based on the set of
3
compounds chosen, such as, based on the elution within discrete regions of the
chromatographic space, then there significant differences between the expected and
the real separation may be realised, when real sample is tested. The study concluded
that if standards cannot be obtained that adequately describe the entire sample matrix
the sample itself should be used, and that the separation should be optimised for
regions of interest, not necessarily the separation as a whole.
Through Chapter 5 selectivity studies undertaken on total 17 stationary phase and
mobile phase 2D combinations is discussed for optimal 2D HPLC separations of café
espressos. Selectivity testings have been carried out on the first dimension of a two-
dimensional system, while the second dimension was solely a „selectivity detector‟
performing within a constant stationary/mobile phase chromatographic environment.
The entire separation space was divided into quadrants and the orthogonality was
assessed based on selectivity differences between regional quadrants of the two-
dimensional separation plane accordingly for each quadrant. The results revealed that
chromatographic behaviour of each phase system was entirely sample dependent and
that it would be practically impossible to develop an „absolute‟ orthogonal 2D HPLC
separations for complex samples like natural products. Most importantly, the
interpretation of the outcome depends on the expectations and the goal of the analyst.
Overall, a 2D HPLC system combining a Cyano stationary phase operating with
aqueous/methanol mobile phase on the first dimension and a C18 stationary phase
operating with aqueous/methanol on the second dimension, both with gradient
elution, offered „the best‟ separation environment to meet the scope of the current
work, that is, to have the separation conditions that could be the most informative
about the sample‟s chemical heterogeneity.
In Chapter 6 hyphenation of „the best‟ performing 2D HPLC system with antioxidant
CL assay predominantly by using acidic potassium permanganate reagent, was
investigated. The outcome was that the analytical power of 2D separation and the
sensitivity and the speed of acidic potassium permanganate CL detection
discriminated between the antioxidants and many other compounds that possess a
suitable chemiluminophore i.e., indicating antioxidant reactivity of sample analytes.
Such hyphenated screening tests are useful for the rapid pre-selection of „marker‟
antioxidant compounds of the sample.
4
Chapter 7 explored the feasibility of a 2D system comprising a CN/aqueous-
methanol phase combined with a C18/aqueous-methanol phase as alternative to
currently used fingerprinting techniques. The information obtained
chromatographically reflected chemical diversity of the given samples, with high
reproducibility suggesting the reliability of CN/C18 system coupled with on-line
post-column antioxidant assays (2D HPLC-CL). 2D HPLC separations were carried
out on apple peel samples generating chemical „maps‟ that could potentially be
useful in understanding the types of chemical functionality that compounds present
in apple peel are posing, that is, for example, hinting on the polarity or
hydrophobicity of certain compounds. In 2D HPLC-CL mode a study was conducted
on red wine, as they are known for the high antioxidant content „wrapped‟ within a
complex chemical matrix. Information generated although not quantitative could
serve as a qualitative „guide‟ for antioxidant profiling of the wine samples.
Chapter 8 concludes the thesis. Overall the issues addressed by the current study are
timely and the outcome of it can be of great importance to the research focusing on
the natural products and their antioxidant contents by facilitating and empowering
the information content from the experiments.
5
CHAPTER 1
Introduction
6
1.1 Defence Against Oxidants: Antioxidants
Reactive intermediates in oxidation processes, particularly free radicals, are at
present receiving increased attention in biology, medicine, food chemistry, various
industrial and environmental areas [10]. An extensive amount of data infers the
important role of reactive oxygen species (ROS) and reactive nitrogen species (RNS)
in the pathogenesis of various diseases, including cancer, cardiovascular diseases,
diabetes, tumours, and the toxicity of numerous compounds [11, 12, 13]. This has
stimulated research on the potential of intervening in these oxidation processes with
antioxidants; defined as “any substance that when present at low concentrations
compared to that of an oxidisable substrate significantly delays or inhibits the
oxidation of that substrate” [14]. Consequently, the role of compounds, capable of
acting as antioxidants or of inducing antioxidant‟s protective mechanisms, offering
protection against the damaging effects of ROS/RNS generation, has received
increased attention.
Depending on the scientific discipline, the scope and protection target of antioxidants
are different. In food science, antioxidants have a broader scope, in that they include
components that prevent quality deterioration of products and maintain their
nutritional value by interrupting the chain of free radicals, decomposing
hydroperoxides, or as chelating agents [15, 16]. Further to this dietary antioxidants
include a substance in foods that significantly decreases the adverse effects of
reactive species, such as ROS and RNS, on normal physiological function in humans
[15]. Within biological systems, modern theories of reactive intermediates have
revealed that they play a dual role [17]. Firstly; they are involved in the organism‟s
vital activities including phagocytosis, regulation of cell proliferation, intracellular
signalling and synthesis of ATP (adenosine triphosphate) and other biologically
active compounds [18]. Secondly, as natural by-products of our own metabolism
they attack the cells, trying through cellular membranes to react with the nucleic
acids, proteins, and enzymes present in the body [19]. Under intense influence of
environmental and endogenous radical-initiating factors i.e., hard ultraviolet
radiation, mineral dust, pollutants, autooxidation reactions, mitochondrial leak etc.,
such attacks can cause oxidative stress by initiating cells to lose their structure,
function and eventually destroying them [14, 20]. Today, oxidative stress is
7
increasingly becoming an important hypothesis of the generation of various
pathologies, including neurodegenarative disorders, cancer, atherosclerosis, heart
diseases and aging [21, 22, 23]. Consequently, the role of antioxidants within the
living organisms in counteracting the effects of oxidative stress include keeping the
antioxidant status balance between the antioxidant system and prooxidants by
eliminating excess production of free radicals during physiological processes [24, 25,
26]. Dietary components are important junctions of the antioxidant system in various
biological fluids especially after oxidative depletion of endogenous source of
antioxidants, such as superoxide dismutase and glutathione peroxidase enzymes, by
contributing to the improvement of antioxidant status [27, 28, 29], and diminishing
damaging effects of oxidative stress [30].
1.2 Antioxidants and Their Mechanisms of Action
Despite the difference in the overall scope of antioxidative protection either as food
or cosmetic quality preservatives or as dietary supplements with potential health
benefits, radical chain reaction inhibitors are commonly regarded as antioxidants that
largely induce protective effects, and also, to-date they have been the most
extensively studied antioxidants [15].
Both exogenous and endogenous antioxidants can be classified into two main types:
primary (chain-breaking) and secondary (synergistic) and in biological systems the
third group, that is, antioxidant enzymes, has been introduced [14, 31]. Primary
antioxidants (hydroquinones, tocopherols, etc.) react directly with free radicals
before they react with other molecules converting them to more stable products.
Secondary antioxidants act by mechanisms such as binding transition metal ions
between each dimension. Normalisation may, for example, be undertaking using the
process described in Equation 1.10:
0
0
ttf
ttia RR
RRX
(1.10)
where Xa is the normalised retention time, Rti is the retention time of the component i,
Rto is the retention time of the least retained solute and Rtf is the retention time of the
final solute in the sample.
19
1.7.1.1 Geometric Approach to Factor Analysis (GAFA)
Factor analysis is useful for examining large data sets and for determining the
orthogonality and practical peak capacity of two-dimensional chromatographic
systems [144]. Correlation matrices can be constructed from the scaled retention
times of solutes from each of the dimensions and in this way the practical peak
capacity is able to be visualised. The correlation matrix (C) is calculated according to
Equation 1.11 [144]:
''
11
MMN
C T
(1.11)
where N is the number of scaled retention times, M’ is the matrix of scaled retention
times and M’T is the transposed matrix of the scaled retention times. This yields a
square correlation matrix in the form of Equation 1.12 [144]:
C= 12
21 11 CC (1.12)
where C12 = C21 and is a measure of the correlation between two sets of retention
time data and the orthogonality of a two-dimensional system. Complete correlation
exists in a chromatographic system when C21 = unity. When C21 = zero a totally
orthogonal chromatographic system is evident. The product of the peak capacities of
the individual dimensions theoretically predicts the peak capacity of a two-
dimensional system in truly orthogonal systems. However, the practical peak
capacity is much smaller than the theoretical value when some degree of correlation
is present and can be quantified by the spreading angle (), which is the
characteristic of non-orthogonal two-dimensional separations [142].
The geometric plot (Figure 1.1), created by calculating the region of the correlation,
and then the spreading angle ( is subtracted from the product of the theoretical
peak capacity of each dimension. The region external to the vectors separated by the
angle ( illustrates the two-dimensional retention space that cannot be utilised due
to correlation between dimensions. The practical peak capacity is given by Equation
1.13 [144]:
)( CANN TP (1.13)
20
where NP is the practical two-dimensional peak capacity, NT the theoretical two-
dimensional peak capacity and A and C are the unavailable areas in Figure 1.1 due to
correlation. The smaller the angle ( the greater the degree of correlation and the
less space that can be utilised. Values of the spreading angle range between zero and
ninety degrees. A spreading angle of ninety degrees indicates a maximum peak
capacity in which true orthogonality exists for the two-dimensional system. A
spreading angle of zero indicates a highly correlated two-dimensional system
equivalent to that of a one-dimensional system [144].
Figure 1.1 Effective non-orthogonal, two-dimensional retention space where the peak spreading angle is ].
1.7.2 Sample Dimensionality
Realisation of 2D separation potential requires that ideally the two separation
dimensions display orthogonal selectivity or retention behaviour. While achieving
this condition practically is very rare, in order to fully utilise the power of a two-
dimensional separation, it does require system design to be undertaken with due
consideration to the nature of the sample [111]. That is, each of the dimensions
within the separation system should ideally be selectively and uniquely oriented
towards each specific sample attribute or dimensionality (n), targeted for separation
[111]. This enhances ordered displacement of sample components across the 2D
space. Examples of sample dimensionality, include, but are not limited to, carbon
number, molecular weight, pKa, functionalities, tacticity and chirality [145].
Two examples of sample dimensionality are found for example for the samples of: (1)
polynuclear aromatic hydrocarbons (PAHs) and (2) low molecular weight polymers
(see Figure 1.2(a and b)). In the case of the PAHs if we consider the members of the
homologous series, naphthalene, anthracene, 2,3-benzanathracene and pentacene,
21
then each increases in size through the addition of a single aromatic ring – the first
sample dimension. If then the structural isomers of the four ring homologue are
considered (chrysene, pyrene, 2,3-benzanthracene and benz[a]anthracene), a second
sample dimension is realised. Ultimately, selection of the various separation
dimensions can be made by considering the nature of the sample and subsequently
the separation can be essentially tuned to the various sample attributes, leading to
very high selectivity. However, Giddings‟ recommendation was that in practice, a
system comprising (n) dimensions for (n) sample attributes is impossible due to the
physical limitations of the number of system dimensions that may be coupled in the
design of real separation processes [111]. Therefore, finding fully non-correlated
selectivity for each dimension in a two-dimensional system is rare [144]. This makes
the development, or at least the optimisation, of a two dimensional separation much
more complex.
(a)
22
(b)
Figure 1.2 Illustration of sample dimensionality (a) polycyclic aromatic hydrocarbons (size and shape), and (b) diastereoisomers of n-butyl polystyrene oligomers with n = 2 to n = 5 styrene repeating units.
Overall, when designing 2D separation systems there are two basic aspects that
should be considered. The first is the nature of the sample or sample dimensionality
and the second is the selection of the phase systems that are most suitable for
coupling, and that yield the most divergent retention behaviour for that particular
sample.
1.8 The Practical Criteria of 2D HPLC Applications (Natural Products)
Chemical profiling analysis of complex samples often requires pre-treatment
strategies, such as fractionation and precipitation, in order to reduce the complexity
of the sample and thus work within the limitations of the unidimensional peak
capacity [146]. Such experiments can be not only labour intensive, but also can cause
degradation of the active components within the chemical matrix. Given its
separation power, 2D HPLC has gained greater recognition in the analysis of foods
and biological samples. Depending on the scope of the analysis, 2D separations
could be fine-tuned to target, for instance, specific compounds by maximising
resolution in the region of interest, or it could be employed as a fingerprinting
technique. Consequently, there is much scope in the design of 2D systems and
method development, with ultimately the needs of the analyst dictating the end result,
C
H
H
CH2
C H
CH2
C H
CH2
C H
CH2
C H
CH2
CH2CH2CH2CH3
C
CH2
H
CH2
C H
CH2
CH2CH2CH2CH3
C H
CH2
CH
CH2
C H
H
C
CH2
H
CH2
C H
CH2
CH2CH2CH2CH3
C H
CH2
CH
CH2
C H
H
C
CH2
H
CH2
C H
CH2
CH2CH2CH2CH3
CH
CH2
CH
CH2
C H
H
C
CH2
H
CH2
CH
CH2
CH2CH2CH2CH3
CH
CH2
CH
CH2
C H
H
C
CH2
H
CH2
CH
CH2
CH2CH2CH2CH3
C H
CH2
CH
CH2
C H
H
C
CH2
H
CH2
C H
CH2
CH2CH2CH2CH3
CH
CH2
C H
CH2
C H
H
C
CH2
H
CH2
C H
CH2
CH2CH2CH2CH3
C H
CH2
C H
CH2
C H
H
C H
H
CH2
C H
CH2
CH
CH2
CH
CH2
CH2CH2CH2CH3
C
CH2
H
CH2
C H
CH2
CH2CH2CH2CH3
CH
CH2
C H
H
C
CH2
H
CH2
C H
CH2
CH2CH2CH2CH3
C H
CH2
C H
H
C
CH2
H
CH2
CH
CH2
CH2CH2CH2CH3
CH
CH2
C H
H
C
CH2
H
CH2
C H
CH2
CH2CH2CH2CH3
C H
H
C
CH2
H
CH2
C H
CH2
CH2CH2CH2CH3
C H
H
C
H
H
CH2
C H
CH2
CH2CH2CH2CH3
n = 2 n = 3 n = 4 n = 5
23
i.e., assessing the relationship between overall separation power, with respect to time
and the amount of information required from the separation.
1.8.1 Sample Dimensionality Selection
Designing efficient two-dimensional separation systems demands that at the very
least an understanding with respect to the behaviour of the solutes in each dimension.
The search for orthogonal differences in selectivity for natural product samples is
complex, because; firstly the sample dimensionality increases as the complexity of
the sample increases, and secondly, the sample dimensionality may be impossible to
deduce without first having detailed knowledge about the sample components. The
latter of course is difficult to obtain for a truly unknown sample. The question is then,
how to select suitable compounds that truly represent the retention behaviour of the
sample so that operational performance of potential 2D systems can be measured?
While numerous researchers have employed commercially available standard
materials [147, 148], the question is nevertheless how accurately does the retention
behaviour of these limited number of standard components relate to the real sample
as a whole. Hence, deducing the „real‟ 2D system performance may be difficult. In
order to gain information that reflects the nature of the compounds, isolation of
specific components may be required, hence hyphenated methods of analysis are
then important. Several hyphenated techniques; HPLC with UV (photodiode) [149], -
fluorescence [150], -electrochemical [151], or -MS [152] detection have been used
for the analysis of complex samples. In practice, however, many factors may hinder
on-line detection and structure determination of an unknown plant or food metabolite
and often only partial structural information will be obtained [153]. The number of
components present in the sample may be such that it is practically impossible to
gain sufficient information that allows a useful description of the sample to be
established. Ultimately, due care should be given to the compounds used to
determine the selectivity of the phase, as it must match that of the sample for an
accurate measure of performance. If the number is too small, or their distribution
does not reflect that of the actual sample separation, then their reliability as a
measure of selectivity performance may be questionable.
24
1.8.2 Selectivity Studies-Stationary Phase
As has been discussed prior, in order to gain the benefits of two-dimensional
separation, the chromatographic system should be developed in the way that the
divergence between the two retention mechanisms is provided, to generate ideally
orthogonal separations. Reversed phase LC and capillary electrophoresis (CE) for
peptide separation [154], liquid-chromatography and gas-chromatography (LC - GC)
for separation of environmental pollutants [155], 2D strong cation-exchange
chromatography (2D-SCX)/RPLC for proteomics research [156], are examples of
systems that yield significant divergence due to the different separation mechanisms
of the two dimensions. Employing detection methods, such as mass
spectrophotometry and infrared spectroscopy, with hyphenated techniques such as
liquid chromatography, can resemble multidimensional systems [152, 157].
Nevertheless, designing two-dimensional systems based on different techniques can
be challenging as the practical problems, e.g., opposing operating conditions, could
generate broadened peaks in the second dimension because of solvent mismatch [130,
131]. Uniting separation systems with similar operating environments, such as LC x
LC, can enhance the compatibili ty between the two dimensions, albeit with some
loss in system divergence [142]. Nevertheless, some degree of correlation can be
useful as it increases predictability in peak displacement [142].
In multidimensional chromatography programming column selectivity is particularly
important as it is necessary to achieve independent retentive separations, and even
more important for 2D RPLC separation systems, where the fundamental basis of the
separation process remains essentially the same technique, liquid chromatography.
with polar end-capping), and a Luna Pentafluoro-Phenyl (PFP). All column formats
were 150 4.6 mm, packed with 5 m Pd particles except for the Kinetex column,
which was 100 4.60 mm, packed with 2.6 m Pd particles.
2.3.4 Development of On-Line Post-Column DPPH Assay Technique
The aim was to set-up a primary technique for rapid antioxidants screening from
complex mixtures by combination of chromatographic separation and post-column
antioxidant assay in a single run. The study was conducted on Granny Smith apple
variety harvested in Australia in 2008 spring season and obtained from the local
market. Details of chemicals, sample and reagent preparation are given in Sections
2.2.1 and 2.2.2.
2.3.4.1 Instrumental Set-up
The schematic illustration of the instrumental set-up is depicted in Figure 2.1. The
initial set-up consisted merging a LC stream (without splitting) at 0.6 mL/min flow
rate with DPPH reagent flowing at 0.8 mL/min in a reaction coil (500 µL volume)
(refer to DPPH line only). The reaction coil was thermo stated at 60 ºC.
The chromatographic separations were completed on a Luna 100 Ǻ CN column (150
4.60 mm 5 M Pd). Solvent A was water, and solvent B was ACN. Initial
condition of 100% solvent A was followed by an increase to 100% solvent B over
100 min. After 60 min runtime no analytes were observed to elute thus for pictorial
presentation the chromatograms of this experiment are illustrated until 60 minutes
(Figure 2.2). The column was equilibrated for 10 min prior to injection. An injection
volume of 10 µL was used. Separated compounds were detected at 280 nm by
UV/Vis detector (sampling rate 5 Hz). The antiradical compounds were detected
simultaneously at 517 nm as a negative response by second UV/Vis detector (5 Hz
sampling rate).
39
DPPH line: CL line: Figure 2.1 Diagram of the flow system for antioxidants screening based on UV, DPPH and CL detection used in experiments. The arrows indicate flow direction.
2.3.4.2 Results and Discussion
UV and DPPH chromatograms show good separation and detection of antioxidant
compounds from apple flesh methanol extract (Figure 2.2).
0 10 20 30 40 50 60
-1000
-500
0
500
1000
DPPH'
UV
Inte
nsity (
mV
)
Retention Time (min)
Figure 2.2 Separation and identification of antioxidants in apple flesh (Granny Smith): UV/Vis (280 nm) and DPPH (517 nm) radical scavenging chromatograms.
CL detector
Analytical column UV/Vis 1 (280nm)
HPLC pump 2 (DPPH reagent only)
Peristaltic pump (CL reagent only)
UV/Vis 2 (517 nm)
Sample injection
Flow cell
PMT (photomultiplier tube)
T-piece eluents split 50-50
HPLC pump 1
T-piece
Reaction coil 60 ̊ C
•
•
40
The linearity of the DPPH detector response was calculated by a linear regression
analysis of absolute areas versus concentration of four well known antioxidant
standard compounds; catechol, caffeic acid, ascorbic acid, and gallic acid (Figure
2.3), over the range 0.001 to 1 mg/mL (n = 3). Standards were made up daily in
methanol. A linear relationship between the negative peak areas and injected
concentrations; R2 = between 0.9818 to 0.9906 was observed for all compounds
(Figure 2.4).
COOH
OH
OH
OH
OH
OH
COOH
OH
HO
OO
HO OH
HOH
HO
(a) (b) (c) (d)
Figure 2.3 Chemical structures of (a) caffeic acid, (b) gallic acid, (c) catechol, (d) ascorbic acid.
Figure 2.4 Linear dependence of the peak areas on the tested concentrations detected by DPPH assay.
Table 2.1 depicts the Limits of Detection and Limits of Quantification of tested
standard compounds, measured according to Harris [212]. Catechol and caffeic acid
had the lowest limits of detection, 0.056 and 0.068 µg/mL, respectively.
41
Table 2.1 Limits of Detection (LoD) and Limits of Quantification (LoQ) of standard antioxidants by DPPH detection.
Precision of the analytical method was confirmed based on the peak areas and
retention times. Intermediate precision was studied by comparing the Relative
Standard Deviations (RSD) of triplicate injections carried out within three days
interval, using the DPPH solution kept continuously for 3 days at 4 °C in a dark
(Table 2.2).
Table 2.2 Intermediate precision of the retention time and peak area of standard antioxidant compounds with a DPPH detector within 3 days interval (n = 3).
SD=Standard Deviation, RSD=Relative Standard Deviation in percentage (at 95% confidence level)
Tested compounds Radical Scavenging Detection at 517 nm
by DPPH Assay
LoD (µg/mL) LoQ (µg/mL)
Ascorbic acid 0.055 0.158
Caffeic acid 0.026 0.068
Catechol 0.021 0.056
Gallic acid 0.05 0.148
Tested compounds
Retention time (min)
Peak area
Mean ± S.D/R.S.D(%) Mean ± S.D/R.S.D(%) Ascorbic acid
Day 1 4.1 ± 0.06 1.5
0.14 ± 0.001 0.5
Day 2 4.1 ± 0.01 0.3 0.08 ± 0.005 6.2 Day 3 4.1 ± 0.05 1.2 0.06 ± 0.007 0.12
Caffeic acid Day 1
4.0 ± 0.36 9.0
0.12 ± 0.01 9.0
Day 2 4.0 ± 0.02 0.5 0.12 ± 0.003 2.4 Day 3 3.9 ± 0.04 1.0 0.10 ± 0.005 4.6
Catechol Day 1
4.3 ± 0.61 14.0
0.29 ± 0.001 0.3
Day 2 4.7 ± 0.37 3.8 0.27 ± 0.01 4.0 Day 3 4.6 ± 0.10 2.3 0.24 ± 0.007 2.8
Gallic acid Day 1
3.9 ± 0.14 3.6
0.21 ± 0.004 1.8
Day 2 4.0 ± 0.001 0.05 0.14 ± 0.001 0.5 Day 3 4.0 ± 0.05 1.2 0.12 ± 0.01 7.8
42
This study presents initial findings of the HPLC-DPPH assay that showed
satisfactory sensitivity. Some of the conditions of the presented HPLC-DPPH assay
were later improved for the antioxidant screening in green tea and red wine.
Optimum assay conditions were found to be: 5 10-5 M DPPH reagent prepared in a
degassed with nitrogen; reaction coil of 2 m 0.25 mm i.d. PEEK tubing; detection
at 521 nm; analysis at room temperature (results are in submission for publication)
[92].
Later to the HPLC-DPPH set-up a CL detector was added, by splitting the LC eluate
stream (50-50 ratio, controlled with a pressure regulator) at a T-piece between the
DPPH and CL detectors.
2.3.5 Chemiluminescence (CL) Detector
Chemiluminescence detection consisted of a transparent coil of tubing (~ 40 cm of
0.8 mm i.d), mounted against the window of a photomultiplier tube (Electron Tubes
Model 9828SB, ETP, Ermington, NSW, Australia), in a light-tight housing [213].
The total volume of the flow cell was approximately 200 μL. The reagent was
propelled to the T-piece using a Gilson Minipuls 3 peristaltic pump (John Morris
Scientific, Balwyn, Victoria, Australia) with bridged PVC tubing (DKSH,
Caboolture, Queensland, Australia). The reagent flow rate was 1.85 mL min-1.
2.4 Chromatographic Separation Methods
Experimental conditions for chromatographic separations and employed assays
varied between different studies. Details are provided in the appropriate chapters.
2.5 Data Analysis
The data plotting from one-dimensional HPLC and on-line antioxidant analyses has
been carried by using Microcal Origin (version 6.0) program (NSW, Australia).
Data obtained from a 2D HPLC analysis is a three dimensional data set, these
dimensions represent the first dimension retention time (or the cut time), the second
43
dimension retention time and the detector response. For graphical software packages
to display this data the entire two-dimensional separation must be merged into a
single dataset (in the three column format).
In this study to determine the location and number of separated components in a two-
dimensional separation domain a pick picking Wolfram Mathematica 7 (distributed
by Hearn Scientific Software, Melbourne, VIC, Australia) in-house written program
was utilised using optimised threshold conditions. Mathematica 7 peak picking
program was also employed to detail system performance derived from the geometric
approach to factor analysis (GAFA) [144]. This program is detailed in Appendix I.
44
CHAPTER 3
High performance liquid chromatography with two
simultaneous on-line antioxidant assays:
Evaluation and comparison of espresso coffees
45
3.1 Introduction
The use of multiple assays has been advocated to reconcile differences between
antioxidant data [44], including an attempt to derive „a complete and dynamic picture
of the ranking of food antioxidant capacity‟ [214].
The on-line antioxidant assays reported to date have almost exclusively been based
on DPPH or ABTS+ radical decolourisation, inhibition of luminol
chemiluminescence, or electrochemical techniques [7]. However, no single assay
provides definitive results, due to factors such as the multiple mechanisms of
antioxidant action, differences in the oxidant or free radical species used in each
assay, and interferences specific to particular assays or classes of assay [15, 48-53].
All studies focused on a single on-line assay, with the exception of that described by
Exarchou et al., [101], who used both DPPH and ABTS+ assays (after separate
chromatographic runs) to examine the antioxidant profiles of several plant extracts.
Recently, it has been proposed that the direct chemiluminescence reaction with
acidic potassium permanganate could be exploited as a rapid on-line assay to screen
for antioxidant compounds [9]. This reagent has previously been used for highly
sensitive quantitative detection of phenols and related compounds after
chromatographic separation [75], and to assess the total antioxidant status of wines,
teas, and fruit juices using flow injection analysis methodology [99].
Various off-line in vitro assays have been used to compare the total antioxidant
activity of coffees of different origin, variety and brewing processes [204, 215-218],
examine fractions/compounds isolated from coffee [200, 219, 220] and as part of
broader studies, comparing different plant extracts to identify rich sources of natural
antioxidants [221] or examining the contribution of different foods to the total
polyphenol/antioxidant consumption [222]. The application of on-line antioxidant
assays to examine coffee is limited to two recent studies on the effects of roasting
conditions, both of which combined reversed phase (C18) chromatographic
separation with the ABTS+ radical scavenging assay [223, 224]. The work described
in this chapter is the first use of an on-line DPPH assay to provide a detailed
antioxidant profile of coffee samples, which also serves as the first direct comparison
46
of on-line DPPH radical decolourisation and acidic potassium permanganate
chemiluminescence assays.
High separation efficiency is crucial for the analysis of complex natural products.
These types of samples contain multitudes of compounds, which often exceed the
peak capacity of the separation space. This problem is compounded when multiple,
sequential detectors are employed, or detection involves on-line chemical reactions,
which can lead to significant post-column diffusive band broadening and loss of
resolution. Therefore, to maximise detection of specific compounds, such as
antioxidants, the chromatographic separation efficiency and the time-scale and
degree of selectivity of each mode of detection must be considered.
Herein, a reversed-phase separation was combined with UV-absorbance detection
and two on-line chemical assays (DPPH decolourisation and acidic potassium
permanganate chemiluminescence). The majority of previously reported on-line
DPPH assays incorporated reaction coils constructed from 13-15 m of 0.25 mm i.d.
tubing [7], which provided reactor volumes of over 600 L, but significantly lower
volumes have also been successfully used [225]. To provide sufficient reaction with
minimal band broadening, a short reaction coil (100 L volume) was utilised and
heated to 60 °C. The chemiluminescence detector consisted of tightly coiled
transparent tubing (~ 40 cm of 0.8 mm i.d.), mounted against a photomultiplier tube.
Although the total volume of this flow cell was approximately 200 L, it should be
noted that the width of the peaks are also dependent on the rate of the transient
chemiluminescence response (i.e., the short-lived emission of light from a rapid
chemiluminescent reaction may be complete before the reacting mixture exits the
flow cell [226]). The aqueous-methanol gradient conditions selected for separation
(Section 3.2.4) are compatible with both the DPPH decolourisation [7] and
permanganate chemiluminescence [75] assays.
When combined with chromatographic separation, each of these three modes of
detection i.e., UV, DPPH and CL, provides a distinct perspective on the character of
these highly complex sample matrices. Almost any solute with a suitable
chromophore can be detected by absorption; 280 nm is most commonly used for the
47
quantitative post-column detection of phenolic antioxidants in foods [227], but it is
not specific to that functional group and provides no indication of reactivity. In
contrast, the responses for the DPPH decolourisation and permanganate
chemiluminescence assays are dependent on the reactivity of the compound toward
the respective reagent (as well as the concentration of the compound) [15, 228].
However, the mechanism of reaction and mode of detection are different [15, 95].
The DPPH reagent is consumed by radical scavenging compounds to produce
chromatograms comprising negative peaks from an ideally constant, high baseline
signal (517 nm) [7]. The acidic potassium permanganate reagent provides highly
sensitive detection of various phenols and other readily oxidisable compounds, based
on the emission of light from the manganese(II) product of the reaction [95]. Unlike
most other on-line assays used to assess the reactivity of antioxidant species [7],
permanganate chemiluminescence produces positive signals on a low, stable baseline.
These two assays are susceptible to very different interferences; examples include
colour pigments of natural products that absorb light of the same wavelength as that
used to measure DPPH, and the remarkable sensitivity of the permanganate reagent
towards certain phenolic alkaloids such as morphine and oripavine [229].
3.2 Experimental
3.2.1 Chemicals, Reagents and Samples
The chemicals and reagents used in this chapter are detailed in General Experimental
(Chapter 2). Three espresso coffees, Ristretto, Volluto, and Decaffeinato were
analysed. The manufacturer‟s description of these flavours is „subtle fruity full
bodied‟ (intensity of 10), „sweet and biscuity‟ (intensity of 4) and „aroma of red fruit‟
(intensity of 2), respectively.
3.2.2 Sample and Reagent Preparation (refer to Sections 2.2.1 and 2.2.2 for details)
3.2.3 Chromatographic Instrumentation and Columns
3.2.3.1 Chromatographic Instrumentation
The details of chromatographic instrumentation employed in this study are given in
General Experimental (Section 2.3.1).
48
3.2.3.2 Chemiluminescence (CL) Detector
The details of chemiluminescence detector are given in Section 2.3.5 (Chapter 2).
3.2.4 Chromatographic Separation and On-Line Antioxidant Assays
Separations were performed on either a Kinetex 90 Ǻ C18 (100 × 4.60 mm; 2.6 m
Pd) column or a SphereClone 100 Ǻ C18 (150 × 4.60 mm; 5 m Pd) column as
indicated in the appropriate text. Linear gradient conditions were employed on both
columns, starting from an initial mobile phase composition of 100% water and
running to a final mobile phase composition of 100% methanol, at a rate of 5% min-1.
The final mobile phase composition was held on for 4 min before re-equilibration
with the initial mobile phase. The flow rate was 1 mL/min and the injection volumes
were 10 µL. After UV-absorbance detection (280 nm), the eluate stream was split
(50-50 ratio, controlled with a pressure regulator) at a T-piece for the two
simultaneous on-line assays. The system had a gradient delay of ~ 4.5 min to the
head of the column.
3.2.4.1 On-Line DPPH Assay
One half of the eluate stream (0.5 mL min-1) was combined with the DPPH reagent
(0.66 mL min-1) at a T-piece. The combined stream entered a reaction coil (volume:
100 L), which was submersed in a water bath maintained at 60C. Radical
scavenging compounds were detected as a decrease in absorbance at 517 nm, using a
Waters 2487 series UV/Vis absorbance detector (Figure 2.1).
3.2.4.2 On-Line Chemiluminescence (CL) Assay
The other half of the eluate stream (0.5 mL min-1) (Figure 2.1) was merged with the
acidic potassium permanganate reagent (1.85 mL min-1) at a T-piece, immediately
prior to entering a flow-through chemiluminescence detection cell detailed in Section
2.3.5. For comparison purposes, the time axes of the respective chromatograms were
adjusted to account for the difference in volume between the column and the
detectors for the DPPH and CL assays.
49
3.3 Results and Discussion
Two critical aspects for high resolution screening are: (i) maximising separation
efficiency to isolate as many sample components as possible; and (ii) minimising the
time-scale of the assay (and thus the loss of resolution due to post-column band
broadening), whilst maintaining sufficient sensitivity [7]. To these ends, an efficient
reversed-phase separation has been coupled using a Kinetex C18 column with UV-
absorbance detection and two rapid, simultaneous on-line chemical assays: DPPH decolourisation and acidic potassium permanganate chemiluminescence. The
proposed hyphenated system was used to examine the antioxidant profile of three
espresso coffees.
3.3.1 Separation and Detection Conditions
The importance of separation efficiency and the ramifications it has on detection is
illustrated by the series of chromatograms in Figure 3.1, which show separations
achieved with a SphereClone C18 column (packed with conventional porous 5 m
particles) and a Kinetex C18 column (containing „core-shell‟ 2.6 m particles [222]),
for the same sample under identical conditions. In each case, the upper trace
represents UV-absorbance detection and the lower trace is the response for the
DPPH assay. The difference between the results obtained with the two columns was
substantial. For example, the details of the DPPH response in the first five minutes
of the analysis were lost in the complexity of the separation achieved on the
SphereClone column, whereas the use of the Kinetex column allowed the direct
association of many UV-absorbance peaks with DPPH detected bands. Of further
interest was the discrimination of peaks that absorb ultraviolet light, but did not
respond to the DPPH assay.
Three examples, labelled as A, B and C (of which C is caffeine) in Figure 3.1(b),
presented virtually no DPPH response. While two of these bands were also observed
in the separation on the SphereClone column, the third peak was less obvious. These
were by no means exclusive examples of these types of peaks. Caffeine standard (1
mg/mL) separated on the SphereClone C18 column did not show either any
scavenging response to DPPH assay (Figure 3.2). Greater separation could have
been obtained on the SphereClone column by decreasing the gradient rate, but at the
50
detriment of analysis time and band broadening. Better separation was achieved on
the shorter Kinetex column, using the same gradient rate and overall analysis time.
All subsequent data reported in this work were obtained from separations using the
Kinetex column.
(a)
0 5 10 15 20 25-1.0
-0.5
0.0
0.5
1.0
DPPH*
UV
CIn
tensity
Retention Time (min)
(b)
0 5 10 15 20 25-1.0
-0.5
0.0
0.5
1.0
DPPH*
UV
C
B
A
Inte
nsity
Retention Time (min)
Figure 3.1 Chromatograms for the Ristretto sample, separated on (a) SphereClone and (b) Kinetex columns. Response for UV-absorbance detection and DPPH assay shown.
51
0 5 10 15
0
1
2
3
4
Caffeine UV response
Caffeine DPPH responseIn
ten
sity (
mV
)
Retention Time (min)
Figure 3.2 Caffeine standard at 1 mg/mL on SphereClone C18 column with UV-absorbance and DPPH detection response shown. At 5% min-1 aqueous/methanol gradient going from 0 to 100% methanol.
3.3.2 Comparison of Espresso Coffees
The chromatograms obtained with UV-absorbance detection are shown in Figure 3.3.
Overall, the three coffees showed very similar fundamental chemical composition,
within the limitations of the information that can be derived from a unidimensional
separation of this highly complex matrix. In general, the sample complexity of the
Ristretto coffee was greater than those of both the Decaffeinato and the Volluto
espressos, which is consistent with the product description. The chromatographic
profiles of the Volluto and Decaffeinato espressos were almost perfectly overlaid;
thus it is tempting to suggest that the decaffeinated coffee was derived from similar
beans to those used in the preparation of the Volluto espresso.
52
0 5 10 15 20 25
0.0
0.5
1.0
1.5
2.0
Ristretto
Volluto
Decaffeinato
Inte
nsity
(mV
)
Retention Time (min)
Figure 3.3 Chromatograms for separation on Kinetex column and UV-absorbance detection, of Ristretto, Volluto and Decaffeinato café espresso samples.
The chromatograms obtained with the DPPH decolourisation and permanganate
chemiluminescence assays indicated that all three coffees contained a substantial
number of antioxidant-type compounds. The chromatograms in Figure 3.4, for
example, compare the results of the three methods of detection for the Ristretto
sample.
(a)
0 5 10 15 20 25-1.00
-0.75
-0.50
-0.25
0.00
0.25
0.50
0.75
1.00
C (Chemiluminescence)
B (DPPH')
A (UV)
Inte
nsity
(mV
)
Retention Time (min)
53
(b)
0 2 4 6 8 10-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
B (DPPH')
C (Chemiluminescence)
A (UV)
Inte
nsity
(mV
)
Retention Time (min)
(c)
10 12 14 16 18 20-1.00
-0.75
-0.50
-0.25
0.00
0.25
0.50
0.75
1.00
B (DPPH')
C (Chemiluminescence)
A (UV)
Inte
nsity
(mV
)
Retention Time (min)
Figure 3.4 (a) Chromatograms for separation of Ristretto coffee with (A) UV-absorbance detection, (B) DPPH decolourisation assay, and (C) acidic potassium permanganate assay. Figure 3.4(b) and Figure 3.4(c): as above, with close up view of 0-10 min and 10-20 min respectively.
Table 3.1 lists the most significant peaks in all three modes of detection, in order of
retention time. Detection was rated with a score of 0 to 3, with 0 indicating no peak
detected and 3 indicating an important peak. This relative score does not give any
information regarding the absolute nor even the relative concentration of each
component. If two modes of detection scored a 0 response while the third detector
scored a significant response, the rating was 3 by default. If two detectors were
equally sensitive and more so than the third, they both scored a value of 3. Of the 21
peaks listed in Table 3.1, 19 components were observed with UV-absorbance (280
54
nm) detection, with 70% yielding a strong response. The benchmark of choosing the
21 peaks was the UV detected separation of Ristretto. This data then was compared
with DPPH and CL sensitivity. Only 13 of the components responded to the DPPH assay, with 28% showing a strong response. In the chemiluminescence assay, 16
components were seen, with 52% showing a strong response, indicating that,
compared to DPPH, this reagent is sensitive towards a wider range of oxidisable
sample components. This clearly illustrates the advantage of employing multiple
modes of detection when searching for bioactive species in complex media. Each
mode was able to discriminate between sample components depending on certain
characteristics. In particular, it was interesting to examine the degree of
discrimination between chemiluminescence and DPPH assays, which revealed the
different behaviour of various oxidisable compounds contained in the coffee samples.
Hence, using multiple modes of detection may aid in not only identifying antioxidant
species, but also understanding their mode of action.
Table 3.1 Key peaks in the chromatograms for the Ristretto coffee sample obtained using UV-absorbance, DPPH and chemiluminescence modes of detection.
Comparison of the three coffee samples based on permanganate chemiluminescence
or DPPH decolourisation assays (Figure 3.5) shows a degree of similarity akin to
that observed with UV-absorbance detection.
(a)
0 5 10 15 20 25
0.0
0.5
1.0
1.5
2.0
(C) Decaffeinato
(B) Volluto
(A) Ristretto
Inte
nsi
ty (
mV
)
Retention Time (min)
(b)
0 5 10 15 20 25-2.0
-1.5
-1.0
-0.5
0.0(C) Decaffeinato
(B) Volluto
(A) Ristretto
Inte
nsity
(mV
)
Retention Time (min)
Figure 3.5 Chromatograms for Ristretto, Volluto and Decaffeinato samples: (a) acidic potassium permanganate assay and (b) DPPH decolourisation assay.
Interestingly, the overall intensity of the response for both assays were aligned with
the flavour „intensity‟ scale provided by the manufacturer (2, 4 and 10, for
Decaffeinato, Volluto and Ristretto, respectively). In agreement with previous
comparisons of the total antioxidant activities of coffees (with and without
decaffeination) using off-line in vitro assays [204], and examination of the
antioxidant profile of coffees using HPLC with an on-line ABTS+ assay [224],
caffeine did not exhibit antioxidant activity. However, caffeine has previously been
shown to be an effective inhibitor of lipid peroxidation (in vitro) induced by
hydroxyl (HO) and peroxyl (LOO) radicals and singlet oxygen (1O2) [230].
56
Moreover, Brezová et al., recently noted that although inert to ABTS+ and DPPH, caffeine is effective in scavenging HO radicals [216]. This demonstrates an
important limitation of off-line and on-line in vitro assays for antioxidant activity,
where some compounds that do not respond to particular assays may still exhibit
significant activity under other conditions. Nevertheless, the overall antioxidant
profile of the Decaffeinato coffee sample was similar to those of the Volluto and
Ristretto samples – rich in compounds that responded to both on-line assays – and it
is therefore likely that decaffeinated coffees have similar positive effects on human
health.
3.4 Conclusions
The antioxidant profiles of various espresso coffees were established using HPLC
with UV-absorbance detection and two rapid, simultaneous, on-line chemical assays
that enabled the relative reactivity of sample components to be screened. Results
from the two approaches based on (i) the colour change associated with reduction of
the 2,2-diphenyl-1-picrylhydrazyl radical (DPPH); and (ii) the emission of light
(chemiluminescence) upon reaction with acidic potassium permanganate, were
similar and reflected the complex array of antioxidant species present in the samples.
However, some differences in selectivity were observed. Chromatograms generated
with the chemiluminescence assay contained more peaks, which was ascribed to the
greater sensitivity of the reagent towards minor, readily oxidisable sample
components. The three coffee samples produced closely related profiles, signifying
their fundamentally similar chemical compositions and origin. Nevertheless, the
overall intensity and complexity of the samples in both UV absorption and
antioxidant assay chromatograms were aligned with the manufacturer‟s description
of flavour intensity and character. This approach provides much greater information
than total antioxidant batch-type measurements, but is often hindered by insufficient
resolution of chromatographic peaks due to the complexity of the samples.
57
CHAPTER 4
A Discussion on the Process of Defining Two-
Dimensional Separation Selectivity
58
4.1 Introduction
A prerequisite for two-dimensional separation is different retentive selectivity in
each of the dimensions. An understanding of the sample dimensionality is important
[105], because this will dictate the phase systems to be employed. However,
designing orthogonal separations to display selectivity towards ‘n’ dimensionality of
complex natural products samples is practically impossible, due to the physical
limitations of the number of system dimensions that may be coupled in the design of
real separation processes [111]. Nevertheless, there are examples in the literature of
two-dimensional separations whereby each separation dimension specifically
exploited only a single sample attribute, yielding nearly orthogonal systems, with the
result being a very high separation power [231] and a very ordered separation
displacement. An example of these high powered two-dimensional separations can
be found in the analysis of the diastereomers of low molecular weight polystyrenes
[231, 232]. Each dimension was reversed phase, one being a carbon adsorption
surface with a high degree of selectivity for diastereomers, while the other was a C18
phase, with almost no selectivity for diastereomers, instead separating only the
oligomer fractions. When mobile phases were employed that increased the selectivity
towards the other (or second) sample attribute, the theoretical peak capacity
increased [232], but the useable separation space decreased due to increased
correlation between each dimension. This resulted in a higher degree of crowding
within the separation space and made it more difficult to undertake the separation
[232].
The search for selectivity changes between different phase systems is simple in the
aforementioned example of the polystyrenes because, firstly the sample is easily
described, and secondly, it does not vary in its dimensionality as the sample
complexity increases. The increase in complexity comes about solely due to an
increase in molecular weight, which increases the number of diastereomers by the
order of 2(n-2), where n = the degree of polymerisation. Hence different molecular
weight fractions can be isolated and injected as standards into each dimension of the
system. The behaviour of the low molecular weight fractions also represents the
behaviour observed by the higher molecular weight fractions since the sample
dimensionality remains constant as a function of the molecular weight.
59
The search for differences in selectivity for samples of natural origin is, however,
often much more complex than that for the polystyrenes described above. Often very
little is known about the sample and hence defining the sample dimensionality may
be impossible. Whenever possible, the sample itself should be employed, as this will
result in the true measure of system dimensionality with respect to the separation
problem faced by the analyst.
An analysis of data, derived from a two-dimensional separation of Ristretto café
espresso was undertaken to illustrate how the measure of the degree of divergence
within the two-dimensional separation can change as a function of the sample set
employed to undertake the analysis, even though all sample components are directly
derived from the sample itself. The key outcome from this section is the importance
of correctly selecting a sample set that reflects the nature of the sample, or at least the
important aspects of the sample when undertaking optimisation studies of the
separation. For example, for complex samples, it may be feasible given limited
information regarding the nature of the sample itself, that the test compounds or
standards that are employed to measure selectivity differences between respective
dimensions may have similar chemistries, their behaviour in the two-dimensional
domain, may therefore not necessarily reflect the behaviour of the true sample.
Conversely, if the sample in its entirety is used to measure selectivity changes, then
all compounds to be separated are employed in the measure of selectivity, even if
their identity is unknown.
4.1.1 Statistical Metrics
The analysis presented here is undertaken using a geometric approach to factor
analysis (GAFA). This metric was chosen because it yields information that is
visually simple to interpret and calculation can be easily automated into peak picking
programs [Appendix I]. A geometric approach to factor analysis is commonly used to
examine variations within data sets, and in the context of two-dimensional HPLC,
Liu et al., [144] used a geometric approach to factor analysis to assess orthogonality
and estimation of peak capacity. The specific details of how a geometric approach to
factor analysis may be employed in two-dimensional HPLC have been assessed in
detail previously (Chapter 1, Section 1.7.1.1). The key metrics that the GAFA yields
60
are: the degree of correlation between each dimension, the peak spreading angle (the
measure of difference between the two separation vectors), theoretical and practical
peak capacities, and utilisation of the separation space.
In addition to the above metrics, a metric measuring space between detected peaks
was also used, which is called the normalised mean radius rn. This metric uses the
intuition that the average area around each point is an indication of the separation of
the detected peaks. The area around a peak is modelled by the area of the disk
centred at the peak, with radius given by the distance to the peak‟s nearest neighbour.
This area is then normalised by dividing by the area corresponding to the largest
possible distance between 2D peaks, namely the area of a disk whose radius is given
by the diagonal of the rectangular separation plane. If the distance from a given peak
to its nearest neighbour on the two dimensional plane is given by r, and if the
separation plane has dimensions a b, then the normalised mean radius is given by
Equation 4.1:
22
2
ba
rmeanrn (4.1)
where a and b represent the length and width dimensions of the separation plane (i.e.,
the distance between the maximum and minimum retention times in both
dimensions) and r is the distance to the nearest peak (calculated according to
Pythagoras‟ theorem).
4.2 Experimental
4.2.1 Chemicals and Samples
The chemicals and samples used in this chapter are detailed in Section 2.1.
Experimental work in this study was conducted on Ristretto café espresso (for
sample preparations refer to Section 2.2.1).
4.2.2 Chromatographic Instruments and Columns
All separations (one- or two-dimensional) were undertaken on a Waters HPLC
system (refer to Section 2.3.1 for details). A Luna Propyl Cyano column (150 mm
61
4.6 mm) packed with 5 m Pd particles and a SphereClone C18 column (150 mm
4.6 mm) packed with 5 m Pd particles were used.
4.2.3 Chromatographic Separation
Selection of the phase environments used in this chapter was based on an extensive
selectivity study undertaken that investigated a total of 17 stationary phase and
mobile phase combinations for Ristretto sample. The results derived from the CN
aqueous-methanol/C18 aqueous-methanol system is reported as this system yielded
the greatest degree of separation performance. Separations were undertaken using a
comprehensive (incremented) heart-cutting technique, in which the first dimension
was the cyano phase and the second dimension was the C18 phase. Both dimensions
were operated in the same mobile phase, which was a gradient run with the initial
conditions set at 100% water running to a final mobile phase of 100% methanol over
a 10 minute period. The final mobile phase composition was held on for 4 min before
re-equilibration with the initial mobile phase. Flow rates were 1 mL/min. All
injection volumes in the first dimension were 100 µL. The UV detection was set at
280 nm.
The comprehensive heart cutting process of analysis consisted of transferring a 200
µL heart-cut section from the first dimension to the second dimension. This process
was repeated for every second 200 µL aliquot of elution volume from the first
dimension across the entire first dimension separation. A total of 34 samples were
subsequently taken from the first dimension and sampled in the second dimension.
The total analysis time was over 20 hours. The entire process was automated and
continuous until completion.
Unidimensional separations were undertaken using the same mobile phases and flow
rates as for the two-dimensional analysis, but at gradient rates of 5% min-1. All other
conditions remained constant except there was no heart-cutting to a second
dimension.
62
4.2.4 Data Analysis
Data analysis was undertaken using a peak picking program written in Wolfram
Mathematica 7, which incorporated algorithms for the calculation of the key metrics
using the geometric approach to factor analysis. This software is described in
Appendix I.
4.3 Results
The chromatograms illustrated in Figure 4.1 depict the separation of the coffee
extract in each of the respective one-dimensional systems on (a) the cyano column
and (b) the C18 column. Quite clearly these separations illustrate significant
differences in the separation behaviour, albeit the complexity of the sample and the
fact that the peak capacity has been exceeded, negates the extraction of meaningful
information regarding the nature of these selectivity differences.
The two-dimensional surface plot representing the chromatographic separation of the
coffee brew in the two-dimensional system employed here is shown in Figure 4.2.
This representation of the chromatographic displacements in each mode of separation
depicts quite clearly the differences in selectivity of separation between each
dimension. Visual inspection of this information indicates that there are three key
regions of sample displacement. Intuitive assessment of this retention information
would suggest that these regions display varying degrees of retention correlation
between each dimension. Hence, describing separation „performance‟ would largely
depend on the sample set chosen for the analysis. Therefore the purpose of this
chapter is to illustrate how the selection of the sample set that is used to measure the
two-dimensional system performance can influence the process of separation
optimisation.
63
0 5 10 15
0
1
2
3
4(a)
Inte
nsity (
mV
)
Retention Time (min)
0 5 10 15
0
1
2
3
4(b)
Inte
nsity (
mV
)
Retention Time (min)
Figure 4.1 One-dimensional chromatograms Ristretto on (a) Cyano and (b) C18 stationary phases. Both columns 150 4.6 mm; 5 m Pd, mobile phase was aqueous/methanol going from 100% water to 100% methanol at a gradient rate of 5% min-1. Flow rate of 1 mL/min. Detection at 280 nm.
64
Figure 4.2 2D HPLC separation surface plot of Ristretto. 1st Dimension: Cyano column, 2nd Dimension: C18 column. Both dimension separations employed aqueous/methanol gradient elution going from 100% water to 100% methanol at a rate of 10% min-1. Flow rates in both dimensions were 1 mL/min. Injection volume in the first dimension was 100 µL, detection at 280 nm.
4.3.1 Sample Set Selection
Here the assumption is that either there is very limited information with respect to
the type of compounds present in the sample, or how the compounds that are known
may be displaced within the two-dimensional retention plane. Without detailed
knowledge of the sample it would be difficult to deduce whether separation
performance was limited by the system performance i.e., efficiency and selectivity,
or because the sample itself was of such a high complexity separation would always
be problematic, irrespective of the system design.
To illustrate the effect that the selection of the sample set may have on the measure
of two-dimensional separation performance a complex café espresso sample was
employed. Systematically the two-dimensional separation performance was
evaluated using subsections of the sample and the subsequent two-dimensional
retention times derived from practical separations. By selecting sample components
as a subset from the entire sample population various „potential‟ model systems
could be mimicked, which contain a variety of different compounds, for whatever
reason, may have been chosen by the analyst as representative samples of the entire
component population - in its entirety. These „hypothetical‟ sample sets thus may
65
represent sample sets chosen by the analyst for the purpose of measuring selectivity
changes.
4.3.2 System 1. 2D HPLC System Performance Measured Using the Entire Ristretto Espresso Sample
(a) Threshold 100%
In this test, the threshold was set to its most sensitive level i.e., 3 times the signal to
noise level, according to the separations that were obtained in the second dimension.
This value, for sake of simplicity, is denoted as threshold 100%, and represents the
sensitivity required to detect all components above the level of detection. In total,
142 components are recognised as peaks in the surface plot of the coffee sample
shown in Figure 4.2, and as a scatter plot in Figure 4.3(a). The measure of separation
quality achieved by the two-dimensional system is illustrated by the correlation
between dimensions, the peak spreading angle (, the percent usage of the
separation space and the mean separation between adjacent band centres (normalised
with respect to the separation plane available), given in Table 4.1. These results
indicate that despite both dimensions employing the same mobile phase for two
reversed phase dimensions, a moderate degree of separation divergence was
observed with a correlation of 0.74, and a spreading angle of 42. A total of 44% of
the theoretical peak capacity was lost due to correlation between each dimension i.e.,
56% of the space was utilised. The normalised mean radius to the nearest
neighbouring peak (rn) for the separation across the total two-dimensional plane in
Figure 4.2 was 0.036 10-2. The results for the measure of separation performance
on this „neat‟ sample, in its entirety, reflect the true measure of performance in the
2D HPLC separation of the café espresso, and this serves as the bench mark for all
other tests.
66
(a)
(b)
(c)
0 2 4 6 8 10 12 14
0
2
4
6
8
10
12
14Threshold 100%
2nd
Dim
ensi
on ret
entio
n tim
e (m
in)
1st Dimension retention time (min)
0 2 4 6 8 10 12 14
0
2
4
6
8
10
12
14Threshold 75%
2nd
Dim
ensi
on ret
entio
n tim
e (m
in)
1st Dimension retention time (min)
0 2 4 6 8 10 12 14
0
2
4
6
8
10
12
14Threshold 50%
2nd
Dim
ensi
on ret
entio
n tim
e (m
in)
1st Dimension retention time (min)
67
(d)
(e)
Figure 4.3 Scatter plots detailing the location of peak maxima across the two-dimensional separation plane. (a) Threshold 100%; (b) Threshold 75%; (c) Threshold 50%; (d) Threshold 25%; (e) Zones 1, 2 and 3.
0 2 4 6 8 10 12 14
0
2
4
6
8
10
12
14Threshold 25%
2nd
Dim
ensi
on ret
entio
n tim
e (m
in)
1st Dimension retention time (min)
0 2 4 6 8 10 12 14
0
2
4
6
8
10
12
14
Zone 3
Zone 2
Zone 1
2nd
Dim
ensi
on ret
entio
n tim
e (m
in)
1st Dimension retention time (min)
68
Table 4.1 Summary of the statistical measures of the peaks separated with the different thresholds and in the different zones.
(b) Thresholds - 75%, 50% and 25%
Subsequent to section (a) threshold 100%, the threshold for peak sensitivity was
reduced by 25%, 50% and 75%. As the threshold level decreased, the total number of
detected peaks decreased to 108, 72 and 37. The change in the distribution of peaks
located across the separation space, as a function of threshold is illustrated in the
scatter plots shown in Figure 4.3(a to d). Irrespective of the threshold level selected
there was, however, very little change in the measures of separation performance,
with the correlation being 0.77, 0.75 and 0.73, for each of the 75%, 50% and 25%
thresholds respectively. Consequently, for all thresholds the spreading angles were
between a low of 40 and a high of 44 for the 75% and 25% thresholds respectively.
This resulted in the percentage of peak capacity lost due to correlation being between
47% and 43% respectively. Consequently, it would appear that the selection of
sample components for the measure of system performance had little effect on the
overall measure of separation power. However, examination of the scatter plots in
Figures 4.3(a) to 4.3(d) show that the decrease in the threshold level, resulted
coincidently in an almost uniform loss of components across the separation plane,
System
Correlation
β
(º)
%Usage
rn
(x 10-2)
Zone 1 0.80 36.9 50 0.309
Zone 2 0.53 58.1 71 0.781
Zone 3 -0.26 75.0 87 0.936
Zone 1 and 2 0.85 31.0 44 0.099
Zone 1 and 3 0.99 10.0 16 0.026
Zone 1, 2 and 3 0.83 34.1 47 0.039
Zone 2 and 3 0.86 31.2 44 0.105
100% 0.74 42.3 56 0.036
75% 0.77 39.6 53 0.040
50% 0.75 41.9 55 0.058
25% 0.73 43.4 57 0.111
69
effectively therefore not changing the overall nature of the sample. This may not
necessarily be the case for other natural product samples, but is perhaps a fortuitous
aspect of this sample and separation conditions. A consequence, however, of
decreasing the sensitivity, was that the number of detected peaks decreased. This in
turn increased the mean radial distance between nearest neighbours, while at the
same time the theoretically available separation plane remained constant. The result
of this was an increase in the normalised mean radius between adjacent neighbours
(rn) indicating a decrease in the peak crowding across the separation plane. This is
consistent with there being fewer peaks representing the separation, and their spread
being larger.
4.3.3 System 2. 2D HPLC System Performance Measured Using Selected Regions
of the 2D Separation Space
If there is little prior information, with respect to how different compounds will
selectively interact with specific separation environments, it would be conceivable
that the selection of sample components could result in the displacement of bands
that are biased towards specific two-dimensional separation zones. This would be
particularly true when you consider that many systems would need to be evaluated
during the optimisation process and it would consequently be difficult to make a
model system that would provide representative displacement across numerous
systems. Under such situations, to what extent would the resulting two-dimensional
chromatographic behaviour reflect that of the sample in its entirety? To test this
effect, the separation shown in Figure 4.2 was divided into three separate regions,
each region largely reflecting the zones that show distinct chromatographic patterned
behaviour within the two-dimensional domain. These zones are denoted as zone 1,
zone 2 and zone 3, as indicated in Figure 4.3(e). The data points in Figure 4.3(e) are
exactly the same as those in Figure 4.3(a), except that all data points that were not
displaced in these three zones have been removed. The hypothesis set forth here is
that the analyst uses model compounds to represent the sample, and that these model
compounds elute in the regions described as zones 1, and/or 2 and/or 3. In effect this
would represent a scenario whereby the analyst had chosen compounds that were
weakly retained, and/or moderately retained and/or strongly retained in both
separation dimensions. Such a sample set selection should intuitively provide for a
reliable measure of orthogonality, however, selecting such compounds that would
70
elute reliable in these regions across numerous systems may be difficult. Following
the outcome for the measurement of the separation performance was examined based
on the conditions that: (a) Compounds elute in only zone 1, (b) Compounds elute
only in zone 2, (c) Compounds elute in only zone 3, (d) Compounds elute in only
zone 1 and zone 2, (e) Compounds elute in only zone 1 and zone 3, (f) Compounds
elute in only zone 2 and zone 3, and (g) Compounds eluting in zone 1, zone 2 and
zone 3.
(a) Compounds eluting only in zone 1
Zone 1, are strongly retained components in both dimensions, largely the
hydrophobic analytes. In this zone a total of 49 components were detected.
According to the distribution of these components across the two-dimensional
separation space, the correlation between the two dimensions was 0.80, marginally
more correlated than when the entire sample itself was employed (threshold 100%)
as the set of test compounds. Consequently, the spreading angle decreased to 37, with now 50% of the peak capacity being unavailable for separation purposes (see
Table 4.1). The peak density distribution gave a normalised mean radius from nearest
neighbours equal to 0.309 10-2, an order of magnitude larger than any of the
conditions associated with the threshold limits. This was a result of the decreased
region of the separation space employed, rather than a greater spread of peaks.
(b) Compounds eluting only in zone 2
The compounds that elute in zone 2 would be slightly more polar than those in zone
3. In this zone a total of 34 peaks were detected. In this region the correlation
between each dimension was measured to be 0.53, with the resulting spreading angle
being 58 and subsequently only 29% of the peak capacity was lost due to correlation
between each dimension i.e., 71% of the separation space was available, and the
density distribution of the peaks (rn) was 0.781 10-2, which represents the
combination of high space utilisation across a small region of separation space (see
Table 4.1).
71
(c) Compounds eluting only in zone 3
The compounds that elute in this zone are the polar analytes. They are in fact weakly
retained in both dimensions. It is in fact this region that provides the most interesting
measure of separation divergence between the dimensions. In this zone a total of 21
components eluted. The correlation between each dimension was 0.26, (but
negatively). The spreading angle was 75, and the percentage of space lost due to
correlation was only 13%. In this region the spread of the peaks was at its greatest in
any of the systems, with rn equal to 0.936 10-2, which reflects the high space
utilisation and low correlation between the dimensions across the relatively small
space in both dimensions. This result clearly illustrates the nature of the separation
behaviour on cyano phases, where these compounds of varying hydrophobicity had
litt le retention on the relatively polar cyano phase, and limited association to the non
resonance bonding on the carbon – nitrogen triple bond. The end result is that had
compounds that eluted in this region been chosen as the set of model analytes to
measure the separation performance of this system, then clearly the overall result
would not have been representative of the entire sample – compare to 100%
threshold results in Table 4.1. Figure 4.4 illustrates the extent of the almost
orthogonal, yet weak reverse correlation observed in zone 3, in comparison to that
for the higher and positively correlated elution components in zone 1.
(a) Zone 3
1.0 1.5 2.0 2.5 3.0
1.5
2.0
2.5
3.0
3.5
2nd
Dim
ensi
on ret
entio
n tim
e (m
in)
1st Dimension retention time (min)
72
(b) Zone 1
9 10 11 12 139
10
11
12
13
2nd
Dim
ensi
on ret
entio
n tim
e (m
in)
1st Dimension retention time (min)
Figure 4.4 Scatter plot illustrating the correlation in retention time data in (a) Zone 3 and (b) Zone 1.
(d) Compounds eluting in zones 1 and 2
By combining zones 1 and 2 a total of 83 components were observed to elute. The
correlation that was as a result of compounds eluting in these two regions was 0.85,
substantially higher than either zone 1 or 2, when assessed separately. The resulting
spreading angle decreased to 31 with a corresponding loss in peak capacity equal to
56%. The magnitude of rn was consistent with that observed across the entire
sample.
(e) Compounds eluting in zones 1 and 3
The combination of zones 1 and 3 resulted in the elution of 70 analytes. The
correlation measured as a result of the retention of the analytes in these zones was
now 0.99, with a spreading angle of 10 with a loss in the available peak capacity of
the two-dimensional system equivalent to 84%. Clearly, had components been
selected as a model set from the types of analytes that would be observed to elute in
zone 1 and 3, then the hypothetical separation performance of this system would
have negated its application in the analysis of the real sample, for which the actual
correlation was 0.74, with a utilisation of separation space being 56%. Figure 4.5
illustrates the distribution of the components across the two dimensional plane for
these components, showing the high degree of overall positive correlation.
73
Zones 1 and 3
0 2 4 6 8 10 12 140
2
4
6
8
10
12
14
2nd
Dim
ensi
on ret
entio
n tim
e (m
in)
1st Dimension retention time min)
Figure 4.5 Scatter plot of the retention times of the peaks contained in Zones 1 and 3.
(f) Compounds eluting in zones 2 and 3
Independently, GAFA for the compounds that eluted in zone 1 and zone 2 resulted in
the measure of the two highest degrees of separation divergence for this system, and
hence the greatest degree of space utilisation. When these two zones were combined
together the total number of components was 55. The correlation between the two
dimensions that was measured was in fact the second highest, despite individually
being the two least correlated regions. The correlation was 0.86, with a spreading
angle 31 and the resulting peak capacity unavailable for separation being 56% (see
Table 4.1). This clearly illustrates the importance of judicious selection of standard
compounds if models are to be employed in the assessment of separation
performance.
(g) Compounds eluting in zone 1, zone 2 and zone 3
The total number of components that eluted in these three zones was 104. Of the
seven systems tested (aside from the variation in threshold settings) this sample set
was the most representative of the sample in its entirety. The correlation between the
dimensions was 0.83, with a spreading angle of 34 and a loss in separation space
74
equivalent to 53%. Even despite the fact that the selection of components from these
three regions most represented the sample as a whole, in that this set contained
samples that were weakly retained, moderately retained and strongly retained in both
dimensions, the actual performance measure was well below that of the true sample
(see Table 4.1). Using these compounds as a system performance measure would
thus have resulted in an apparent loss in peak capacity equal to almost 10%.
4.4 Discussion
The performance of potential two-dimensional systems must be measured in order to
employ systems that have minimal correlation. How to select the most appropriate
set of model compounds in which to undertake the performance measure? The results
presented here show that depending upon what compounds would be used for the
assessment of selectivity, the degree of separation divergence will be highly variable.
In fact here, none of the selected compounds actually reflected the true behaviour of
the sample, and in some cases, there were variations by as much as a 70% change in
the effective use of the peak capacity. Under these circumstances it is evident that
using the sample itself provides the most accurate measure of separation performance,
and hence whenever possible the sample should be employed in the process of
optimisation.
These results also show quite clearly that for complex samples, it would be almost
impossible to couple two systems that are orthogonal for the separation at hand. The
vast variation in sample attributes present for complex samples in itself leads to
commonalities in the retention processes. Thus there may be no one „optimal‟ or best
performing system for the sample, rather, selection of the most appropriate two-
dimensional system may be based upon the desired outcome, with systems designed
accordingly to the key compounds being analysed or isolated. Hence the analyst
should not be dictated by the measure of orthogonality, but rather, the separation
should most suit the problem that is faced.
4.5 Conclusions
Designing efficient two-dimensional separation systems demands that at the very
least a measure of performance be made based on the degree of correlation between
75
each dimension. It is not enough to say that orthogonal dimensions are employed
simply based on the fact that there is a perceived difference in the mechanism in each
dimension. As the sample complexity increases so too does the likelihood that
various sample attributes will increase the correlation between the two supposed
orthogonal dimensions. Therefore, system correlation must be assessed. In that
process, absolute care must be paid to the appropriate selection of compounds that
best represent the sample, or rather, the objectives of the separation. Whenever
possible, the sample itself should be employed, as this will result in the true measure
of system dimensionality with respect to the separation problem faced by the analyst.
76
CHAPTER 5
The Assessment of Selective Stationary Phases
For Two-Dimensional HPLC Analysis of Foods:
Application to the Analysis of Coffee
77
5.1 Introduction
Selectivity studies with respect to both the stationary phase and also the mobile phase
must be undertaken to gauge an understanding with respect to the behaviour of the
sample solutes in each dimension within the 2D system. For such studies, whenever
possible, the sample itself should be employed, as this will result in the true measure
of system dimensionality with respect to the separation problem faced by the analyst
(Chapter 4). For complex samples utilisation of the sample itself during the design
phase of separation is far from straight forward, as there are usually multitudes of
components that co-elute, changes in selectivity are likely to go unnoticed, unless
hyphenated methods of detection are employed that can track specific sample
components as a function of the selectivity change.
Many detection methods such as UV (photodiode array) [233], mass spectrometry
(MS) [234, 235], infrared (IR) [236, 237] and nuclear magnetic resonance
spectroscopy (NMR) [238] can be used to assist in the overall selectivity analysis.
HPLC coupled with UV photodiode array detection (LC/UV) is one of the most
widespread techniques used for screening natural products extracts [233]. It provides
useful information on the type of compounds and in case of phenols also the
oxidation pattern [84]. Due to the high power of mass separation MS has been rightly
placed as an essential detection method in many laboratories completing natural
products analysis. Although MS offers high selectivity, the expense of MS, both the
initial purchase price and subsequent running costs and the basic operating
incompatibilities between HPLC and MS [239], detracts from its routine application
base in processes like selectivity screening. Infrared detection suffers from solvent
interference effects and relatively slow response time, limiting its application base.
NMR is expensive, relatively insensitive, but essentially absolute in its ability to
provide information that relates directly to the identity of a substance. In combination
with LC, these three hyphenated methods of detection yield a combined process of
analysis that is unsurpassed in its ability to provide qualitative and quantitative
sample information [240], however, the cost and upkeep of such instruments is the
limitation.
78
Carr and coworkers [132],],however, introduced the concept that a 2D HPLC system
could in fact be utilised as a separation process as the first dimension, and then the
second dimension serve as a selectivity detector. In that way, changes that are made
to the first dimension can be assessed in the retention distribution in the second
dimension. The relative change in selectivity of the different first dimensions can
therefore be gauged. Selectivity in RP chromatography has been extensively studied.
Cyano and phenyl phases showed little selectivity advantage to C18 columns in RP
separations when initially investigated [241]. Further studies proved that there is an
alternative selectivity for cyano and phenyl phases and theories on the interaction
mechanisms have been put forth [169, 242, 243].
More recent studies have investigated the different selectivity seen between a phenyl
phase using interaction and a cyano phase using non - resonance or a dipole-
dipole interaction [244]. Even the configuration of the phenyl phase has showed
changes in selectivity [245]. Fluoro-substituted columns have also shown alternative
selectivity to alkyl and phenyl phases [246]. In the majority of these selectivity
studies a finite number of test analytes are used to characterise the selectivity. In this
chapter, the selectivity is studied with respect to the behaviour of a complex sample
derived from espresso café containing a multitude of components. The focus is on
general selectivity differences rather than specific functional differences. To
illustrate this process the separation of Ristretto espresso on a number of - selective
stationary phases, has been assessed, employing 2D HPLC techniques with
selectivity detection, with the view of maximising the separation power for extended
studies on the analysis of coffee. In RPLC method development the solvent strength
(S) is an important optimisation factor [247] where the overall solvent strength is
adjusted to give a suitable retention value (k) from 2 to 10 (ideally 1 ≤ k ≥ 5) [248].
In this study, a combination of different stationary phases were used in conjunction
with MeOH, ACN and THF solvent systems to complete a detailed assessment of
selectivity with respect to the analysis of coffee in 2D HPLC.
79
5.2 Experimental
5.2.1 Chemicals and Samples
The chemicals and samples used in this Chapter are detailed in General Experimental
(Section 2.1 and 2.2). Experimental work in this study was conducted on Ristretto
café espresso (for sample preparations refer to Section 2.2.1).
5.2.2 Chromatographic Instrumentation and Columns
5.2.2.1 Chromatographic Instrumentation
The details of chromatographic instrumentation employed in this study are given in
the Section 2.3.1.
5.2.2.2 Chromatographic Columns
All chromatography columns were supplied by Phenomenex (Lane Cove, NSW,
Australia). Five different functionalities were tested: Luna 100 Ǻ Cyano (CN),
SphereClone ODS, Luna Phenyl-Hexyl (PH), Synergi Hydro-RP 80 Ǻ (C18 with
polar end-capping) and a Luna Pentafluoro-Phenyl (PFP). All column formats were
150 4.6 mm, packed with 5 m particles.
5.2.3 Chromatographic Separations
5.2.3.1 First Dimensional Separations
First dimensional separations were performed on either of the CN, C18 with polar
end-capping, PH or PFP columns. Selectivity studies were undertaken in aqueous
solvents of MeOH, ACN and THF. Since the second dimension was to serve as the
„detector‟, assessing the changes taking place in the first dimension, an aqueous/
methanol mobile phase in the second dimension was chosen, based on cost, and
environmental and laboratory impact. All separations, in both dimensions were
operated under the linear gradient conditions, starting with 100% aqueous mobile
phase and finishing in 100% methanol mobile phase at a gradient rate of 10% min-1.
The final mobile phase composition was held on for 4 min before re-equilibration
with the initial mobile phase. All flow rates were 1 mL/min and injection volumes
were 100 µL into the first dimension. Mobile phases were not buffered for all
experiments, despite the fact that coffee is known to contain a high number of
80
carboxylic acids. Initial experiments were undertaken using acidified mobile phases,
however, the separation performance was not improved (results not shown). This
also enhanced our ability to undertake mass spectral analysis in the negative ion
mode and reduced one further aspect of solvent mismatch between the respective
first and second dimensions i.e., pH shock in the second dimension, which would
result from a large bolus plug of solvent being heart cut to the second dimension,
operating at a different pH.
5.2.3.2 Second Dimensional Separations
The second dimension was conducted on the SphereClone C18 column, using
gradient elution with an initial mobile phase of 100% water, running to a final mobile
phase of 100% methanol at a gradient rate of 10% min-1. The final mobile phase
composition was held on for 4 min before re-equilibration with the initial mobile
phase. The flow rate was 1 mL/min. The transfer volume from the first dimension to
the second dimension was 200 µL. UV detection was set at 280 nm.
5.2.3.3 Operation
A comprehensive or more precisely, an incremental heart-cutting approach was used
to express the two-dimensional peak displacement, by which a 200 µL heart-cut
section was transferred to the second dimension, with subsequent second dimension
separation being undertaken. The first dimensional separation was repeated,
following which another 200 µL first dimension fraction was transferred to the
second dimension. This was repeated at every 0.4 mL across the entire first
dimension separation i.e., the first dimension separation was repeated a total of 34
times over a 20 hour period.
5.2.4 Mass Spectra Analysis
A 6210 MSDTOF mass spectrometer (Agilent Technologies, Forest Hill, VIC,
Australia) was used with the following conditions: drying gas, nitrogen (7 mL min-1,
Figure 5.1 One-dimensional separations of Ristretto on (a) Cyano, (b) Phenyl-Hexyl, (c) Pentafluoro-Phenyl, (d) Synergi-Hydro C18 and (e) C18 phases. Mobile phase was aqueous/methanol, going from 100% water to 100% methanol at a gradient rate of 10% min-1. All flow rates were 1 mL/min and injection volumes were 100 µL. All conditions were identical for all phase systems.
More information regarding the nature of the selectivity differences with respect to
the C18 phase can be seen in the two-dimensional surface plots illustrated in Figure
5.2(a to d). In each case the C18 phase was the second dimension, hence the change
in peak displacement reflects the nature of the selectivity change in the first
dimension. These surface plots clearly show that there were significant differences in
the retention behaviour of the solutes undergoing migration through the 2D system.
86
(a)
(b)
87
Figure 5.2 Two-dimensional separations of Ristretto. First dimension (a) Cyano, (b) Phenyl-Hexyl, (c) Pentafluoro-Phenyl and (d) Synergi Hydro-C18 and second dimension C18 phases. In both dimensions mobile phase was aqueous/methanol, going from 100% water to 100% methanol. All conditions identical for each phase system.
In order to assess qualitatively and quantitatively the separation power of the two-
dimensional systems, the number (N) and two-dimensional retention times of eluting
peaks were determined and then a geometric approach to factor analysis (GAFA)
was applied. The data in Table 5.2 depicts numerically the changes that have
occurred as a result of stationary phase selectivity. Each of these systems shows
(c)
(d)
88
relatively high correlation to that of the C18 phase, which is not unexpected since all
coupled systems were RP - RP. However, there are distinct regions whereby specific
systems would out-perform another system for certain sample components.
Therefore, in order to more specifically detail the selectivity differences that were
occurring, an extensive assessment of regional selectivity was undertaken. To do that
the two-dimensional separation plane was divided into quadrants, essentially
consistent with the bimodal nature of the unidimensional separations. That is, each
quadrant represents half the separation period from each dimension.
Then the number of components in each region was measured and the GAFA
assessment of each coupled region was undertaken. This is illustrated graphically in
Figure 5.3, which shows retention time scatter plots for the location of peak maxima
in each of the four coupled 2D systems. Thresholds of 2D peaks were adjusted to the
optimum for an illustration real separated and detected peaks. 2D Peaks in all
employed systems were detected under identical threshold condition. The results
from the GAFA for each system, and the four quadrants within each system are
detailed in Table 5.2.
Table 5.2 GAFA calculations for the 2D HPLC separations and in each of the quadrants.
5.3.2.1 Qualitative Assessment of the Selectivity Changes
Cyano Phase
Quadrant 1: The peaks eluting from the cyano column in quadrant 1 (Q1) do so with
very little retention on the stationary phase in the first dimension. However, these
components display substantial variability in retention across the C18 phase, as
shown by the separation on the C18 column for the heart cut fraction at 3.2 minutes
on the cyano column (Figure 5.4). Hence these compounds have a wide range in
polarity. This aspect of the two-dimensional retention behaviour indicates that these
compounds have limited interaction with the non resonance - electrons or the
dipole-dipole moment on the cyano phase, and separation in the second dimension is
based on their hydrophobicity/methylene selectivity. Mass spectral analysis of
compounds fractionated from this region of the chromatographic separation verified
that these compounds were highly hydroxylated and in some instances were low
molecular weight carboxylic or phenolic acids and monomeric flavan-3-ols (Figure
5.5). Within this quadrant retention of the acids and monomeric flavan-3-ols in the
second dimension increased with the aliphatic chain length (e.g., diCQA, di
procianidins). Alkaloids, such as, nicotinic acid, nicotinamide and trigonelline,
having cyclic amino rings were observed to elute in this quadrant (Figure 5.6).
90
(a)
0 2 4 6 8 10 12 140
2
4
6
8
10
12
14
Q4
Q3Q2
Q1
C18
Dim
ensi
on ret
entio
n tim
e (m
in)
Cyano Dimension retention time (min)
(b)
0 2 4 6 8 10 12 140
2
4
6
8
10
12
14
Q4
Q3Q2
Q1
C18
Dim
ensi
on ret
entio
n tim
e (m
in)
Phenyl Hexyl Dimension retention time (min)
(c)
0 2 4 6 8 10 12 14 160
2
4
6
8
10
12
14
Q4
Q3Q2
Q1
C18
Dim
ensi
on ret
entio
n tim
e (m
in)
Pentafluoro Phenyl Dimension retention time (min)
91
(d)
0 2 4 6 8 10 12 140
2
4
6
8
10
12
14
Q4
Q3Q2
Q1
C18
Dim
ensi
on ret
entio
n tim
e (m
in)
Synergi Hydro Dimension retention time (min)
Figure 5.3 Scatter plots for the 2D HPLC separations with (a) Cyano, (b) Phenyl-Hexyl, (c) Pentafluoro-Phenyl and (d) Synergi Hydro-C18 first dimension columns. The quadrants are defined by the red dashed lines.
Figure 5.4 Heart-cut segment separation of Ristretto on C18 phase at 3.2 min. Quadrant 2: The peaks in quadrant 2 (Q2) show a great deal of variability in their
retention across the cyano phase, however, limited variation on the C18 phase. They
are therefore compounds of similar hydrophobicity, with little change in the nature of
the carbon structure. Separation on the cyano phase is, however, obtained because of
their selective interaction with the non resonance - electrons, and this indicates
significant changes in the degree of functionalisation. Mass spectral analysis of the
components collected from this region of the separation indicates that some of these
compounds contain non-polar substituents, such as, the methoxy group in ferulic acid,
Retention Time (minutes)
AU
92
which contributes to greater retention on the C18 dimension, but offers little in the
way of increasing interactions with the cyano stationary phase in the first dimension
(Figure 5.6).
Malic acid Quinic acid Ferulic acid
Nicotinic acid Nicotinamide Trigonelline
Figure 5.5 Structures of some compounds identified in Ristretto.
Quadrant 3: The peaks in this quadrant (Q3) show discrimination in their retention on
both phases. These are thus compounds with variation in their carbon nature and in
their degree of functionality. Mass spectral analysis confirmed the presence of more
complex alkaloids, such as caffeine, which is the dominating compound in zone 3,
and polyphenols, such as rutin. The structures of these types of compounds are
consistent with the observation of increasing solute-stationary phase interactions on
these two phases. At this point in time nature of compounds eluting in this zone has
not been fully evaluated, as the complexity of the analysis is substantial. Having said
that, we suspect that these compounds may be melanoidins, higher molecular weight
polymeric species with a poly disperse structure containing nitrogen, carbohydrates,
amino acids and phenolics [249].
93
Figure 5.6 Components identified in the Cyano/C18 system: Note, for the purposes of illustration the location of the components on the 2D plot represents only the generalised location, and not the exact 2D retention time. a) caffeic acid b) malic acid c) quinic acid d) fumaric acid e) catechin f) a procyanidin dimer g) feruloylquinic acid h) ferulic acid i) 3-(4,5)-ο-caffeoylquinic acid j) 3,4-dicaffeoylquinic acid k) trigonelline l) nicotinic acid m) sucrose n) caffeine o) caffeoylquinic acid p) rutin q) acetylated hexose based oligosaccharide r) oligosaccharide containing anhydrohexose s) acetylformoin hexose based oligosaccharide t) caffeoylshikimic acid u) nicotinamide.
Quadrant 4: No UV absorbing compounds were observed to elute in this region
(although a line associate with a solvent peak is apparent at around 2 minutes in the
first dimension), indicating that there were likely to be few compounds that have
highly polar or -functional groups with a limited degree of carbon back-bone. (Two
acid Table 5.3 Mass Spectra data of protonated (a) and deprotonated (b) 21 compounds in Ristretto and their retention times on the first (CN, PFP, PHX) and second (C18) dimensions.
Schematically the HPLC-DPPH instrumental set-up is given in the Figure 2.1
(Chapter 2). Please note that the HPLC column eluent without splitting was sent to
the reaction coil to be combined with DPPH reagent at a T-piece (Figure 2.1). The
reaction coil had a volume of 100 L.
6.2.2.4 Chromatographic Columns
Chromatographic separations were performed on a Phenomenex Luna 100 Ǻ CN
(150 × 4.60 mm × 5 m Pd) in the first dimension, and SphereClone 100 Ǻ C18 (150
× 4.60 mm × 5 m Pd) column in the second dimension (Phenomenex, Lane Cove,
NSW, Australia). One-dimensional separations were performed on the same columns.
6.2.3 Chromatographic Separations
6.2.3.1 2D Chromatographic Separations and On-Line Chemiluminescence (CL)
Assay
The same linear gradient conditions were employed on both columns, starting from
an initial mobile phase composition of 100% water, running to a final mobile phase
composition of 100% methanol at a rate of 10% min-1. The final mobile phase
composition was held on for 4 min before re-equilibration with the initial mobile
phase. The flow rate was 1 mL/min and injection volumes in the first dimension were
100 µL. The chromatographic interface between the first and second dimensions
consisted of electronically controlled two-position six-port valves fitted with micro-
electric two-position valve actuators that allowed alternate sampling of the elute from
the first dimension into the second dimension, by which a 200 µL heart-cut section
108
was transferred to the second dimension, with subsequent second dimension
separation being undertaken. The first dimensional separation was repeated,
following which another 200 µL first dimension fraction was transferred to the
second dimension. This was repeated at every 0.4 mL across the entire first
dimension separation i.e., the first dimension separation was repeated a total of 34
times over a 20 hour period.
Following UV-absorbance detection (280 nm), the HPLC column eluent was merged
with the acidic potassium permanganate reagent at a T-piece, immediately prior to
entering a flow-through chemiluminescence detection cell (see Section 2.3.5). For
comparison purposes, the time axes of the respective chromatograms were adjusted
to account for the difference in volume between the column and the detector for the
CL assay.
6.2.4 Data Analysis and Plotting
The data plotting was carried by using Microcal Origin (version 6.0) program (NSW,
Australia). The data analysis for 2D HPLC and CL separations was done by using
Mathematica7 peak peaking program (Appendix I).
6.3 Results and Discussion
Figure 6.1 illustrates the two-dimensional surface plot (UV-absorbance detection) of
the chromatographic separation of (a) Ristretto, (b) Decaffeinato, and (c) Volluto
coffee brews. Each of these three samples produced very similar chromatographic
elution profiles across the two-dimensional domain, which was not unexpected given
the similarity of their unidimensional separations (Figure 3.3). For further discussion
the chromatographic separations were divided into three primary regions: Region A,
region B and region C (Figure 6.1). The compounds in region A were hardly
separated on the first dimension (cyano column), but had profound retention and
subsequent separation in the second dimension (C18), the retention correlation was
almost 0, or in other words, these two dimensions were orthogonal for the
compounds separated here (Table 5.2, Figure 5.3(a)). The compounds in region B
were selectively separated on the cyano phase, with less separation on the C18 phase,
and in this region correlation between both dimensions increased, yet the correlation
109
coefficient was only ~ 0.5, indicating significant differences were still apparent in the
retention mechanisms between the dimensions (Table 5.2).
Amongst these compounds in region B was caffeine, which dominated the separation
plane (Figure 6.1 (B region in (a) and (c) plots)), being present as the most abundant
species within the entire sample. Compounds in region C where strongly retained on
both phases, and hence correlation in this region of the separation increased to
around a correlation coefficient of 0.8 (Table 5.2).
A noteworthy difference between each sample of coffee was the intensity of the
elution profiles, which decreased in accord to the label claim associated with the
„strength‟ of the brew, that is, from Ristretto to Decaffeinato.
In comparison to the 2D separations depicted in Figure 6.1, limited information
about the sample could be obtained from either respective 1D separation, as shown
by the example of the separation of the Ristretto coffee (Figure 6.2) obtained on the
cyano column under the conditions employed for the 2D separation. This separation
shows a continuum of sample, partly due to the complexity of the sample and the fact
that the peak capacity has been exceeded, and partly because the column was
overloaded with sample. Nevertheless even at lower sample loads the sample
contains too many components to yield separation in a unidimensional sense.
A
B
C
(a) Ristretto
110
Figure 6.1 Two-dimensional separations of (a) Ristretto, (b) Decaffeinato and (c) Volluto café espresso. First dimension Cyano and second dimension C18 phases. In both dimensions mobile phase was aqueous/methanol, going from 100% water to 100% methanol, at 10% min-1 gradient. All conditions identical for each phase system.
C B
A
(c) Volluto
C
A
B
(b) Decaffeinato
111
0 5 10 15
0
1
2
3
4
Inte
nsity (
mV
)
Retention Time (min)
Figure 6.2 One-dimensional separation of Ristretto (undiluted) on Cyano column. Mobile phase was aqueous/methanol, going from 100% water to 100% methanol at a rate of 10% min-1. Flow rate at 1 mL/min. Detection at 280 nm.
Figure 6.3(curves a, b and c) illustrate the separation on the C18 column
(unidimensional 2nd dimension separations) of heart-cut fractions from the first
dimension of the weakly retained species on the cyano column for each of the three
coffees (a) Ristretto, (b) Decaffeinato and (c) Volluto. The heart-cut sections were
made at 3.2 minutes from the Cyano dimension. These separations highlight the
significant change in selectivity (practically orthogonal) between each dimension
(Chapter 5) and at the same time illustrate some of the differences between each of
these coffee brews.
In this particular heart-cut region all three coffees were different. In particular the
Decaffeinato coffee was the least complex, with a lower total intensity. Of note, is
the presence of two bands labelled as A and B that were present in the Ristretto and
Volluto coffees, respectively. Neither of these bands was present in the Decaffeinato
coffee. Thus demonstrating the fingerprinting potential of this technique, especially
for the description of sensory attributes of food samples.
While each of the three espressos showed similar bulk behaviour, there were,
however, subtle differences in their chemical composition, as well as the notable
difference in the lack of caffeine in the Decaffeinato sample (Figure 6.1(b) compared
to 6.1(a) and 6.1(c)). What is surprising is the specificity associated with the removal
of caffeine in the decaffeinated sample. Examination of the compounds that
neighbour the caffeine peak across both dimensions reveals that there was limited
112
interference to these compounds. Although, in the Decaffeinato sample, a new peak
is apparent that neighbours the caffeine band region. The new compound is not
present in either the Ristretto or Volluto samples, meanwhile both Ristretto and
Volluto have a single compound eluting at 10.8 minute (C) that is absent from the
Decaffeinato sample (Figure 6.4).
Figure 6.3 Heart-cut segment separation of (a) Ristretto, (b) Decaffeinato and (c) Volluto samples on C18 column at 3.2 min.
Figure 6.4(curves a, b and c) illustrate the unidimensional heart-cut slice separated
on the C18 column derived from the 7.6 minute region on the cyano dimension, for
the Ristretto, Decaffeinato and Volluto samples, respectively, illustrating the subtle
difference associated with this region of the sample. These differences could only be
visualised through high peak capacity separations.
(cB
(cA
113
Figure 6.4 Overlay of the separations of (a) Ristretto, (b) Deccaffeinato and (c) Volluto samples on C18 column heart-cut at 7.6 min. Peaks A, B and C marker peaks in this region.
Figures 6.5(a, b and c) illustrate the surface plots of two-dimensional separations for
each of the three coffee brews (Ristretto, Decaffeinato and Volluto, respectively)
using the acidic potassium permanganate chemiluminescence detection. In contrast
to UV-absorbance detection, an intense emission with this chemiluminescence
reagent is indicative of a reducing agent (antioxidant) with relatively high
concentration and/or reactivity. Substantial differences in the UV-absorbance and
chemiluminescence detection were apparent. Most notable was the complete absence
of the caffeine band in the chemiluminescence profile of both the caffeinated
samples. Furthermore, the new band labelled as A in Figure 6.4(curve b) was not
responsive to this chemiluminescence reagent.
More importantly, the chemiluminescence detection enabled visualisation of
compounds with „apparent high anti-oxidant activity‟ that were almost entirely
absent in the UV-absorbance detection mode, thus providing complementary
information about the sample matrix. The chemiluminescence detection primarily
exhibited sensitivity towards the range of compounds present in regions A and C (see
Figure 6.1) of the sample, and was not responsive to compounds in region B,
suggesting that most compounds in that region are not significant antioxidants. The
MS analysis revealed that compounds eluting in the A and C regions for example of
Ristretto‟s 2D separation plane are mainly phenolic acids and flavonoid glycosides
i.e., quinic acid, rutin etc. (Chapter 5), known for their antioxidant activity [199, 207].
In contrast, compounds highlighted as 1 in Figure 6.5(a, b, c) responded weakly in
114
terms of UV-absorbance, but exhibited intense chemiluminescence, indicating that
these compounds may possess strong antioxidant activity. The MS data show that
among the compounds detected by UV-absorbance in this region is the nonphenolic
trigonelline which is known for its high antioxidant activity [254]. Also of interest is
the elution of the band labelled as 2 in Figure 6.5(a, b, c) which was only visible in
the chemiluminescence mode of detection. Another region of distinct difference in
the detection responses is that of region 3 in Figure 6.5(a, b, c), where reasonably
strong chemiluminescence is observed, but only weak UV absorption.
Figure 6.5 Chemiluminescence detection plots of (a) Ristretto, (b) Decaffeinato and (c) Volluto samples.
(b) Decaffeinato
1
3
2
1
2
3
(c) Volluto
1
2
(a) Ristretto
3
115
These differences are more clearly shown in the series of UV-absorbance and
chemiluminescence detection responses separated on the C18 dimensions from heart-
cut sections at 3.2 minute on the cyano column (Figure 6.6).
There are numerous subtle differences between each of these three coffee brews,
depicted in both the UV-absorbance and chemiluminescence detection modes.
However, these differences are too complex to note individually; rather, it is easy to
simply state the difference in the number of components that were separated and
subsequently detected. In part these differences may arise because of the strength
profile in each coffee brew. The number of detected peaks for each sample (Table
6.1) was established using a peak picking algorithm [Appendix I].
Figure 6.6 UV-absorbance and chemiluminescence detection response of heart-cut fractions of (a) Ristretto, (b) Decaffeinato and (c) Volluto samples at 3.2 min. Blue and red lines represent UV-absorbance and chemiluminescence (CL) response, respectively.
The least number of peaks, in both modes of detection, was observed in the
Decaffeinato sample, which was described by the manufacturer as the „weakest‟ of
the coffee brews. The strongest of the coffee brews, Ristretto, had the most number
of peaks visible in the chromatograms obtained using UV-absorbance and CL
116
detection. Given the fact that the antioxidant compounds remained in the coffee even
after the food processing i.e., the roasting, we can say that café espresso is a good
source of dietary antioxidants.
Overall the benefit of such combinations of different modes of detections,
complementary to each other, is substantial and especially important in bio-assay
based screening analysis. Furthermore, the ability to detect the potential bio-active
compounds and the enhanced separation power of the 2D system present an
important pre-isolation tool enabling the rapid separation of already targeted
components from within this complex sample matrix.
Table 6.1 Number of peaks detected for each café espresso flavour for both UV-absorbance and chemiluminescence detection.
Systematic comparison of permanganate chemiluminescence (CL) with DPPH• post-
column antioxidant assay in terms of preparation, sensitivity, selectivity, resolution
are studied in a separate work [255]. Using flow injection analysis, experimental
parameters that afforded the most suitable permanganate chemiluminescence signal
for a range of known antioxidants were studied in a univariate approach. Optimum
conditions were found to be: 1 × 10-3 M potassium permanganate solution containing
1 % (w/v) sodium polyphosphates adjusted to pH 2 with sulphuric acid, delivered at
a flow rate of 2.5 mL min-1 per line. Further investigations showed some differences
in detection selectivity between HPLC with the optimised post-column permanganate
chemiluminescence detection and DPPH• and ABTS•+ assays towards antioxidant
standards. However, permanganate chemiluminescence detection was more sensitive.
Moreover, screening for antioxidants in green tea, cranberry juice and thyme using
Coffee flavour UV detected
peaks
CL detected
peaks
Ristretto 138 65
Volluto 88 56
Decaffeinato 68 44
117
potassium permanganate chemiluminescence offers several advantages over the
traditional DPPH• assay, such as: faster reagent preparation and superior stability;
simpler post-column reaction manifold; and greater compatibility with fast
chromatographic separations using monolithic columns.
6.4 Conclusions
The data obtained through the current combination of two-dimensional separations
with both UV-absorbance and acidic potassium permanganate chemiluminescence
detection offers relevant and comprehensive information for chemical matrix
characterisation and could serve as a fingerprint for the particular sample description.
Permanganate chemiluminescence detection is faster than other chemical assays for
antioxidant activity and therefore better suited for two-dimensional chromatography,
including systems involving very short second dimension separation times. The
technique can be used to target the isolation of key antioxidants from these complex
matrices with a relative degree of simplicity. Detailed information regarding the
complexity of the sample, and the variation in antioxidant profile between these three
coffees could be obtained using this multidimensional–hyphenated method of
analysis. These types of analyses also have the potential to generate simultaneously
valuable data on the bio-markers and become of special importance for designing
complete food „beneficial‟ compositional tables required for epidemiological
research.
118
CHAPTER 7
2D HPLC Fingerprinting Technique:
Applications To The Analysis of Coffee, Wine and
Apple Peel Samples
119
7.1 Introduction
Chemical fingerprinting is a means of providing a chemical profile or signature that
represents the components that are present in a sample and describes a range of
methods where the primary aim is to provide a unique graphical representation of the
sample by identifying the chemical elements within a matrix in comparison to similar
matrices. Chemical fingerprinting may provide the characterisation, quantification,
differentiation and the identification of complex mixtures based on their chemical
composition [256] and is particularly important in such fields as food, natural
products, pharmaceuticals, forensic and environmental sciences. Accordingly
techniques that imply to acquiring chemical fingerprinting information require high
levels of reproducibility and accuracy.
So far, for the separation and identification of antioxidant phytochemicals, high
performance liquid chromatography/mass spectrometry (HPLC/MS) [234], gas
chromatography/MS (GC/MS) [257, 258], high performance liquid
chromatography/diode-array detector (HPLC/DAD) [234] and more recently,
capillary electrophoresis/MS (CE/MS) [259] techniques together with high field
nuclear magnetic resonance (NMR) spectroscopy have been applied [238]. NMR has
the advantage of providing high structural information content from the experiment
and the NMR chemical shifts have relative stability, however, it is relatively less
sensitive in comparison to other above mentioned techniques [260]. GC is an
analytical technique of greater power for complex samples, but it usually requires
extensive sample preparation and perhaps analyte derivatisation [260] and due to the
lack of volatility of the majority of plant derived antioxidants its use in
phytochemistry is limited to oils and herbs [170]. CE offers several potential
advantages for the analysis of complex natural products, such as higher theoretical
separation efficiency, small sample injection volumes and rapid method development
[261]. Although, it is predominantly useful for highly polar/ionic compounds,
however CE is sometimes reported to lack the robustness required for analysing
biological samples [262], and CE is a separation process, not a means of
identification. HPLC/MS is very well suited for natural products profiling [263],
providing robust operation, coverage of various classes of plant metabolites, and no
need for prior sample derivatisation [264]. However, separation efficiencies in
120
conventional HPLC are limited, causing component co-elution in LC/MS given the
vast chemical heterogeneity of a natural products samples. This may hinder detection
and structure determination of unknown antioxidants at trace abundance. Two-
dimensional liquid chromatography has a strong potential for the analysis of complex
samples (Chapter 5) as it provides high theoretical peak capacity, extremely high
resolution, and can provide great sensitivity and selectivity ensuring optimum
phytochemicals chromatographic coverage [265]. To obtain a 90% probability of
separating a particular analyte as an isolated peak from a complex chemical matrix, a
chromatogram must be approximately 95% vacant [121]. In the expanded two-
dimensional separation space, the probability that two species will elute with exactly
the same retention time in both separation dimensions decreases compared to the
one-dimensional separation [112, 130], enhancing the fingerprinting capability of
two-dimensional separation. Furthermore, the separations in two-dimensions can be
tuned to specific targeted components, providing high resolution and timely
separations.
This chapter explores alternative methods for obtaining chemical fingerprints that is,
the two dimensional high performance liquid chromatography, applying the same
comprehensive (off-line) heart-cutting technique described through the Chapters 4 to
5. These studies were further followed by 2D HPLC-CL application as antioxidant
fingerprinting tool. Two complex, antioxidant-rich samples, that being, apple (peel)
and red wine were included in this study.
7.2 Experimental
7.2.1 Chemicals, Reagents and Samples
Chemicals, reagents and samples used in this chapter are detailed in the Section 2.1
(General Experimental). Ristretto, Capriccio, Volluto and Decaffeinato café
espressos were analysed. All coffees were made with the single coffee making
machine. The manufacturer‟s description of these flavours is „subtle fruity full
bodied‟ (intensity of 10), „a rich aroma‟ (intensity of 5), „sweet and biscuity‟
(intensity of 4) and „aroma of red fruit‟ (intensity of 2). Penfold‟s Rawson‟s Retreat
and Red Delicious apple (peel) samples were analysed. The wine samples of
Penfold‟s Rawson‟s Retreat were injected undiluted direct from the bottle. Sample
121
and reagent preparation details are given in the Sections 2.2.1 and 2.2.2. All samples
prior to injection into the HPLC system were filtered through 0.45 µm pore filter.
7.2.2 Chromatographic Instrumentation and Columns
7.2.2.1 Chromatographic Instrumentation
The details of chromatographic instrumentation employed in this study are given in
the Section 2.3.1.
7.2.2.2 Chemiluminescence (CL) Detector
The details of chemiluminescence detector are given in Section 2.3.5.
7.2.2.3 2D Chromatographic Columns
Chromatographic separations were performed on a Phenomenex Luna 100 Ǻ CN
(150 × 4.60 mm × 5 m Pd ) in the first dimension, and SphereClone ODS (150 ×
4.60 mm × 5 m Pd) in the second dimension (Phenomenex, Lane Cove, NSW,
Australia).
7.2.3 2D Chromatographic Separations
The same linear gradient conditions were employed on both columns, starting from
an initial mobile phase composition of 100% water, running to a final mobile phase
composition of 100% methanol at a rate of 10% min-1. The final mobile phase
composition was held on for 4 min before re-equilibration with the initial mobile
phase. The flow rate was 1 mL/min and injection volumes in the first dimension were
100 µL. UV detection was set at 280 nm. The chromatographic interface between the
first and second dimensions consisted of electronically controlled two-position six-
port valves fitted with micro-electric two-position valve actuators that allowed
alternate sampling of the elute from the first dimension into the second dimension.
The eluates from the first dimension across entire first dimension separation, was
comprehensively heart-cut at every 200 µL into the second dimension. This was
repeated at every 0.4 mL across the entire first dimension separation i.e., the first
dimension separation was repeated a total of 34 times over a 20 hour period.
122
2D separations of apple (peel) and red wine samples were conducted employing the
same method as described above but instead of methanol, tetrahydrofuran as an
organic modifier was used.
Mobile phases were not buffered to reduce the solvent mismatch between the two
operating dimensions i.e., pH shock in the second dimension.
7.2.4 2D HPLC-CL Analysis
Experimental set-up is detailed in Chapter 6 (Section 6.2.3.1). Following UV-
absorbance detection (280 nm), the HPLC column eluent was sent to a
chemiluminescence detector.
7.2.5 Data Analysis and Plotting
Data analysis was undertaken using a peak picking program (Appendix I) and
Excel2007.
7.3 Results and Discussion
Two-dimensional surface plots of each coffee sample are illustrated in Figures 7.1(a)
to 7.1(d). The 2D plots show elution locations and relative concentration of
compounds within the sample. Between the appropriate regions of 1st D 8-12 minutes
and 7.5-13 minutes on the 2nd D axis is a region of shading that is present in all the
flavours (quadrants in Figure 7.1(a-d)). The intensity of the hue varies between the
graphs (Figure 7.1(a-d)), which is in relation to the concentration of the compound
relevant to the shaded region within the samples. The concentration of certain
compounds generally found in coffee vary between the flavours, the more
concentrated the compound the darker the shaded region. Thus the intensities of the
shaded regions are directly related to the concentration of the compounds. Such
visual discrimination has been practically impossible to achieve by one-dimensional
separations (Chapters 3 and 5).
123
(a)
(c)
(b)
124
Figure 7.1 Two-dimensional separations of (a) Ristretto, (b) Capriccio, (c) Volluto and (d) Decaffeinato café espresso. First dimension Cyano and second dimension C18 phases. In both dimensions mobile phase was aqueous/methanol going from 100% water to 100% methanol. All conditions identical for each phase system.
This variability between coffee samples in accordance to the flavour description on
the label shows that potentially chemical profiles of tested coffees could be obtained.
As an approximate and preliminary measure of the reliability of these analyses the
retention time difference of the caffeine in the 2D retention plot, as the most well
known compound in the sample, was examined. Retention times in the second
dimension varied by 0.01 minutes (see Table 7.1) or 0.1% RSD. Such a level of
variability is consistent with the operation of one-dimensional HPLC.
Table 7.1 Caffeine second dimension retention times in three coffees and it‟s Mean and StDev at 95% confidence level.
Ristretto Capriccio Volluto Mean StDev RSD(%)
10.5800 10.5867 10.6133 10.5933 0.017 0.16
To test if this level of reproducibility is consistent across the entire 2D analysis, a test
in triplicate was performed for a segmented area covering between 3.6 to 6.6 min in
the first dimension of Ristretto’s 2D separation plane (Figure 7.2).
(d)
125
Figure 7.2 Overlay of second dimension retention times of the segmented area between 3.6 to 6.6 min in the first dimension (n = 3).
Standard deviation of the means of detected peaks in both first and second
dimensions in the same segmented region (Table 7.2) depicts that the data is
reproducible without significant difference between the retention times. The
repeatability for the entire 2D retention time was further tested by analysing
Ristretto’s two different samples independently by the same method and the overlay
of the retention times is shown in Figure 7.3.
Figure 7.3 Overlay of the second dimensional retention times of two independent runs of the Ristretto.
C18
Dim
ensi
on r
eten
tion
time
(min)
CN Dimension retention time (min)
CN Dimension retention time (min)
C18
Dim
ensi
on r
eten
tion
time
(min)
126
Table 7.2 Reproducibility of the first and second dimensional retention times in the segmented area between 3.6 to 6.6 min, represented as the Mean of three injections ± StDev.
Although an extensive assessment of the reliability of these measurements was not
undertaken and further research into the quality and reproducibility of the technique
would be required for establishing a definite chemical fingerprint, this initial study
indicates its potential applicability. The 2D HPLC retention time profile that has
been generated following a comprehensive (incremental) heart-cut approach could be
considered as a chemical fingerprint, that is, a profile of two-dimensional
reproducible retention times of each component that was eluted in the two-
dimensional retention space. This illustrates the importance and relative ease to
obtain the chemical fingerprint of complex samples, later explored on apple and red
The Penfold‟s Rawson‟s Retreat that was used was a Cabernet Sauvignon, the grapes
of which are sourced from vineyards throughout South Australia. The wine has
youthful and lively flavours of berry, chocolate, and mint characters complemented
by subtle oak nuances. Even given the youthfulness of such a wine, the character is
underlined by a vast chemical nature, the complexity of which is difficult to describe
and almost impossible to unravel. Red wine is a rich source of flavonoid antioxidants
and the determination of antioxidant capacity of wine by CL methods has been
reviewed in paper by Navas and Jimenez, 1999 [266]. Apple represents another
complex sample of natural origin, and numerous studies have supported the strong
total antioxidant capacity of apples [84, 211], with greater value in the peel rather in
the flesh [267].
Figure 7.4 illustrates representative sections of one dimensional separations of the
apple peel samples. Figure 7.4(a) is the first dimension separation of the sample on
the cyano (CN) column with an expanded section highlighted between the regions
12.9 and 14.7 minutes (i.e., between the vertical dashed lines) detailed in the inset.
Figure 7.4(b) is a separation in the second dimension of the cut fraction between 13.4
and 13.6 minutes from the first dimension (Figure 7.4(a)). These chromatograms
illustrate the complexity of this sample and the increased separation power gained by
incorporating a second separation dimension. For example, in the 200 µL cut from
the first dimension six additional peaks were resolved to near baseline resolution
with another two overlapping peaks with distinguished peak maxima in the second
dimension. Another peak is apparent on the shoulder of the fifth peak. Figure 7.4(c)
is a representation of eight consecutive first dimension cuts (0.2 minute slices from
Figure 7.4(a) between 12.9 and 14.7 minutes). While only eight of these cuts have
been displayed, the first dimension (Figure 7.4(a)) was in fact sampled into the
second dimension a total of 97 times. This illustration demonstrates the power of 2D
HPLC. The 1D chromatogram shown in Figure 7.4(a) understated the complexity of
the sample with the 2D separation able to identify many more peaks with each cut
possessing different components.
128
(a)
(b)
(c)
Figure 7.4 Figure 7.4(a) is the 1D separation of apple peel on a CN column using aqueous/THF mobile phase gradient. Inset is an expanded view of the retention between 12.9 and 14.7 minutes. Figure 7.4(b) is the second dimension separation (C18 column with aqueous/MeOH) of the cut at 13.4 minutes (between 17 and 20.2 minutes as the baseline is largely flat before this section). Inset is the expanded first dimension separation with the section that was cut to the second dimension outlined by vertical dashed lines. Figure 7.4(c) represents a stacking of the 8 cuts from the expanded first dimension separation (from 13.0 to 14.4 minutes in 0.2 minute increments (200 µL cut volumes)).
The surface plot of 2D HPLC separation of the apple peel sample using CN
(aqueous/THF) and C18 (aqueous/MeOH) phases, both with gradient elution, is
shown in Figure 7.5. A visual qualitative assessment of the 2D chromatographic
separations by evaluating how the components are scattered on the surface plot,
suggests that the chemical matrix of the apple peel extract is as complex as café
129
espresso‟s but the type of compounds representing these two samples are suggested
to be different. The chemical matrix of the latter sample was represented by at least
three major groups of chemical functionality i.e., hydrophobicity (retained on C18
2nd dimension), polarity (retained on CN 1st dimension) and compounds that were
retained on both dimensions (Figure 6.1). In apple peel there are two major clusters
of compounds (A and B) well retained in both dimension. Application of the peak
picking program (Appendix I) to Figure 7.5 further revealed that there were at least
187 separated peaks in apple peel extract. Such data, in contrast to one-dimensional
separations (Figure 7.4(a)), undoubtedly would be helpful to the analyst for later
hyphenated antioxidant screening experiments.
Figure 7.5 Two-dimensional separations of Red Delicious apple peel methanol extract. First dimension Cyano and second dimension C18 phases. In first dimension mobile phase was aqueous/THF going from 100% water to 100% THF at 10% min-1
gradient. In the second dimension mobile phase was aqueous/MeOH, going from 100% water to 100% methanol at 10% min-1 gradient. Detection at 280 nm.
The 2D HPLC method, previously described (Chapter 5), has been used for the
analysis of the red wine sample also. The surface plot of the 2D analysis of red wine
is depicted in Figure 7.6. In this separation the surface plot hints that the two
dimensions displayed similar characteristics to that of the coffee separations,
although statistically this was not tested. Furthermore, the sample dimensionality of
the wine when eluting under these conditions has been selectively reduced to almost
A B
130
two basic elements. The first, indicated by region A in Figure 7.6 represents
compounds hardly retained and not separated significantly on the cyano phase, but
showing a great deal of separation with a broad ranging selectivity on the C18 phase.
The second, indicated by region is almost the reverse – compounds hardly separated
on the C18 phase (although strongly retained), but on the cyano phase showing
widely separated components across a very large separation region. A third sample
dimensionality is apparent, but less ordered than the other two and this is identified
in Figure 7.6 as region C. In this region compounds were strongly retained in both
phase systems and the elution of these compounds utilised a greater spread in their
separation across both dimensions.
Figure 7.6 Two-dimensional separation of red wine using Cyano (1st dimension) and SphereClone C18 (2nd dimension) stationary phases, with UV absorbance detection. Mobile phase composition was aqueous/THF going from 100% water to 100% THF at 10% min-1 gradient. Detection at 280 nm.
The expanded separation space afforded by the 2D analysis has consequently
revealed that substantially more detail about the chemical nature of the wine could be
elucidated. At the most basic level, a qualitative assessment of the number of
components present could be determined. In order to do this a peak picking
algorithm [Appendix I] has been applied to the separations in Figure 7.6. From this
the tested wine sample contained at least 180 compounds. Some of these are likely to
be singlets, but due to the complexity of the sample there is still likely to be a
significant degree of co-elution, but to a greatly reduced level than in either
separation dimension employed one dimensionally (Figure 7.7(a and b)).
(a)
A
B
C
131
(a) (b)
0 5 10 15 20 25
0
1
Inte
nsity (
mV
)
Retention Time (min)
0 5 10 15 20 25
0
1
Inte
nsity (
mV
)
Retention Time (min)
Figure 7.7 1D separations of Penfold‟s Rawson‟s Retreat red wine on (a) Luna 100 Ǻ CN column (150 × 4.60 mm × 5 M Pd). Experimental conditions: A: water; B: THF at 5% min-1 linear gradient. Flow rate 1 mL/min, injection volume 100 µL, UV/Vis at 280 nm. (b) SphereClone ODS column (150 × 4.66 mm × 5 M Pd). Experimental conditions: A: water; B: MeOH at 5% min-1 gradient. Flow rate 1 mL/min, injection volume 100 µL, UV/Vis at 280 nm.
This information further coupled with CL detector i.e., 2D HPLC-CL (Chapter 6)
yield unprecedented information detailing the antioxidant matrix of red wine (Figure
7.8).
Figure 7.8 (a) Two-dimensional separation of Penfold‟s Rawson‟s Retreat Cabernet sauvignon using CN (1st dimension) and C18 (2nd dimension) stationary phases, with permanganate chemiluminescence detection. (b) Enlarged and re-scaled section containing peaks for dominant antioxidant compounds. However, unlike the coffee (Figure 6.5) the contour plot for chemiluminescence
detection of the separated wine sample showed a dominant cluster of antioxidant
species that were strongly retained on both stationary phases (Figure 7.8(a)). A closer
(b)
(a)
(a)
(b)
132
examination of this cluster revealed numerous distinct peak maxima (Figure 7.8(b)).
The contour plots for the wine and coffee samples illustrate not only the
characterisation of two complex matrices containing considerably different
antioxidant species, but also the vast potential for chemical fingerprinting based on
both the absorption of light and the chemical reactivity of sample components. The
combination of these two complimentary modes of detection further enhances the
information available in the expanded separation space afforded by two-dimensional
analysis.
7.4 Conclusions
The study conducted in this chapter was designed to show the potential application of
2D HPLC as a fingerprinting tool. Extensive studies were not undertaken, but future
work in this area is planned. Quite clearly the described 2D HPLC approach yields
extremely powerful separations, albeit, achieved in experiments that required up to
two days to complete. However, the information gained is enormous, and depending
upon the objectives set forth by the study design, it may well be worth the effort and
time for the analysis of these types of very complex samples.
A detailed profile that represents „most‟ compounds within the sample may provide
more information about the sample than is required by the analyst illustrating the
ramifications that this has on quality control procedures, forensic analysis and
traceability. Applications of this type of information, particularly if coupled with
chemometric analysis, may find use in, say, the determination of the source or region
of the grapes employed to make a specific wine, which would have been helpful in
the recent scandal associated with counterfeit wine on the French market.
133
CHAPTER 8
General Conclusion
134
The importance of antioxidants has been acknowledged by the food, medical and
cosmetic industries, especially for their potential health beneficial effects. As the
search for new sources of natural antioxidants still continues, so does the need for
more powerful separation and detection methods in this field. A technique that could
offer great separation selectivity, sensitivity and reproducibility to the antioxidant
screening studies of complex samples derived from natural origin would therefore
constitute a useful tool for their analysis.
A hyphenated technique combining chromatographic separation on-line with DPPH and CL antioxidant detecting assays based on (i) the colour change associated with
reduction of the 2,2-diphenyl-1-picrylhydrazyl radical (DPPH); and (ii) the
emission of light (chemiluminescence) upon reaction with acidic potassium
permanganate, has been developed (Chapter 3). Results from the two approaches
were similar and reflected the complex array of antioxidant species present in the
samples. However, some differences in selectivity were observed. Chromatograms
generated with the chemiluminescence assay contained more peaks, which was
ascribed to the greater sensitivity of the reagent towards minor, readily oxidisable
sample components. The three coffee samples produced closely related profiles,
signifying their fundamentally similar chemical compositions and origin. The study
suggests that regardless of the multi-component matrix of natural products, the
screening for their antioxidant composition should be based on multi-selective and
complementary detection processes. That is each mode should display selectivity
towards specific active sample components. Such application is a powerful tool for
rapid screening of antioxidant compounds without prior sample fractionation and
purification, and is therefore a great benefit for natural product discovery.
Because of the chemical complexity of natural products the information obtained
from either one-dimensional chromatographic separations or hyphenated on-line
antioxidant assays can be limited, triggering the need for high resolution 2D HPLC
analytical separations. As the sample complexity increases so too does the likelihood
that various sample attributes will increase the correlation between the two supposed
orthogonal dimensions. Therefore, system correlation must be assessed, in which
absolute care must be paid to the appropriate selection of compounds that best
represent the sample, or rather, the objectives of the separation (Chapter 4).
135
Whenever possible, such measurements should be undertaken on the sample itself, as
it will truly represent sample dimensionality dependent selectivity changes (Chapter
4). Understanding the nature of the selectivity may require the application of
hyphenated techniques, such as MS or NMR, but at the cost of more complex
analysis.
While LC × LC separations are not conventional with respect to hyphenated methods
of analysis, they do in fact serve that purpose. In this study selectivity screening data
was normalised by means of a 2D HPLC system where the second dimension of the
separation process remained constant, while the selectivity in the first dimension was
altered. In effect, the second dimension was a selective detector visualised through
the UV response in conventional unidimensional operation (Chapter 4). The two-
dimensional band displacement was then noted as a reflection of the selectivity
differences between systematic changes in the first dimension, not the second. In this
way, the second dimension served solely to reflect the changes taking place in the 2D
system. In some ways application of this type of process suggests that the answer
relating to the most effective separation conditions for a given sample composition is
known prior to undertaking the analysis. Certainly this would need to be true if a
comprehensive coupled 2D HPLC separation were to be employed, since the
constraints associated with sampling frequency and „wrap around effects‟ etc.
demands that a significant degree of sample-behaviour information is required prior
to operation. However, this is not the case if a comprehensive (incremental) heart-
cutting approach is employed. Application of this type of 2D analysis yields
potentially very high peak capacity separations that may yield chemical signature
information particularly useful in systems that are then to be employed for the
targeted detection of key antioxidant components from within the complex sample
matrix.
In order for the sample to be employed as a means of determining differences in 2D
selectivity changes to components that elute from the system must be assigned a
coordinate within the two-dimensional domain. To quantitatively locate 2D peaks in
a 2D separation plain while removing peaks associated with the solvent, multiple
sample components from successive heart-cut fractions have been transported into in
house written algorithm (Appendix I).
136
Differences between alkyl, dipole-dipole, hydrogen bonding, and selective
surfaces represented by non resonance and resonance - stationary phases with
mobile phase combinations were assessed for the separation of Ristretto café
espresso by employing 2D HPLC techniques with C18 phase selectivity detection. A
geometric approach to factor analysis (GAFA) was used as an assessment of
selectivity differences between regional quadrants of the two-dimensional separation
plane. The result suggests that it can be practically impossible to design fully
orthogonal two dimensional separations for complex samples (Chapter 5).
Depending on the elution zone within the entire separation space, where the
component distributions were assessed, the measure of separation quality varied
markedly. Thus there may be no one „optimal‟ or best performing system for the
sample, rather, selection of the most appropriate two-dimensional system may be
based upon the desired outcome. Hence the analyst should not be dictated by the
measure of orthogonality, but rather, the separation should most suit the problem that
is faced and upon the analytical goal to continue the method optimisation designed,
for example, according to the key compounds being analysed or isolated.
What was not covered in this thesis was the design of a 2D system in which the
second dimension was a column different to that of a C18 phase. This was
deliberately so, for the simple reason that the second dimension must be more
retentive than the first. If not sample cut from the first dimension will be in a large
bolus solvent plug of high elution strength for the second dimension. Hence the
driving force for solute interaction with the stationary phase will have to overcome
the strong solute-solvent interaction forces. This results in solute species eluting from
the second dimension earlier than theoretically estimated. This was in fact tested
during the selectivity studies in Chapter 5, where it was verified that even the phenyl
phases in the second dimension yielded 2D systems that displayed irregular
behaviour. These results were not presented and are not relevant to the outcomes of
the work described herein.
In Chapter 6 the conclusion was that combination of 2D HPLC high peak capacity
separations with both UV-absorbance and acidic potassium permanganate
chemiluminescence detection (2D HPLC-CL) offers the great advantages of high
sensitivity and specificity with respect to the antioxidant profiling of complex
137
samples. It is of major importance to find a bioassay or assays that are sensitive and
compatible with respect to the time-scale of the analysis and (bio) chemical
conditions within the 2D HPLC separations. The acidic potassium permanganate
reagent based chemiluminescence detection is fast and highly sensitive and therefore
better suited for two-dimensional chromatography than DPPH. Detailed information
regarding the complexity and the variation in antioxidant content between Ristretto,
Decaffeinatto and Volluto café espressos could be obtained using this
multidimensional–hyphenated method of analysis. The mass spectral analysis
revealed that the antioxidant profile of café espresso is dominated by mainly
phenolic acids and their adducts. Such in vitro systems could be a valuable tool in
antioxidants screening for potential in vivo studies, including targeting key
antioxidants from complex matrices with a relative degree of simplicity.
Chapter 7 explored the reliability of 2D HPLC technique as a fingerprinting tool. The
study was preliminary, and not quantitative but nevertheless proposed that 2D HPLC
could conceivably be used as fingerprinting tool for samples like coffee, fruits, and
wines. The results revealed that 2D HPLC offers reliable component identity of
separated compounds with sufficient reproducibility and repeatability. Specific
region of chemical clusters could be targeted by mass spectrometry and isolated with
high efficiency in 2D-preparative HPLC.
A series of future research should focus on detection and quantitative analysis in
statistical assessments in the differences between closely related samples i.e., source
of origin etc. Detection processes could be extended to include detectors that
measure enzyme activity by coupling 2D HPLC with enzyme immunoassays
The contents of the appendix are presented as it appears in the publication;
Stevenson P.G., Mnatsakanyan M., Shalliker R.A., Peak Picking from 2D-HPLC data.
Analyst 135 (2010); 1541.
154
I.1 Introduction
Complex samples (such as natural products) often contain hundreds of chemical
compounds that cannot be separated with sufficient resolution using one-dimensional
high performance liquid chromatography (1D HPLC) because the peak capacity is
exceeded. For these types of samples two-dimensional HPLC (2D HPLC) must instead
be employed as this technique usually provides a much greater peak capacity than does
1D HPLC. To maximise the two-dimensional information each separation dimension
should be orthogonal [I.1]. In order to assess the difference in selectivity between each
dimension it is usual that the retention behaviour of model compounds under a variety of
different separation conditions is studied in each of the potential systems to be coupled.
However, in the case of natural products the type of chemical components present may be
unknown and hence selecting a suitable set of model compounds that represent the
sample can be difficult. It is possible that selectivity changes in different 1D systems can
be followed by using mass spectrometry. However, this is time consuming, very labour
intensive and usually only appropriate for the more abundant species. Although HPLC-
MS may also provide information about the types of compounds that could be used to
construct an appropriate set of model compounds. In this paper we illustrate a simple
process that allows the sample itself to be used as the „model‟ set for the determination of
selectivity differences for two-dimensional separations.
Chemical fingerprinting is a means of providing a chemical profile or signature that
represents the components present in a sample. HPLC is often used in conjunction with
MS to generate chemical fingerprints for quality control and forensic analysis in
applications such as; traditional Chinese medicines [I.2-I.4], medicinal plants [1.5], fruits
and vegetables [I.6], proteins and peptides [I.7, I.8] and illicit drugs [I.9]. In many cases
MS is employed to identify each component peak in the one dimensional chromatograms.
2D HPLC is a cheaper option (as the acquisition and operation of MS is quite expensive),
and a large degree of information about the sample can be obtained relatively quickly.
Additionally the method can be used to generate chemical fingerprints as the retention
times in both dimensions have a higher probability of being unique, thereby minimising
the need for a MS if only comparisons between sample sets are required [1.10]. However,
155
for this method to be practical a rapid automated approach to 2D peak detection must be
employed. The algorithm in this paper can be applied to generate a two-dimensional
chemical fingerprint of complex samples and can be used to compare chemical profiles,
with the additional benefit of providing a measure of separation performance based on
the degree of divergence between each dimension, the estimation of the system peak
capacity, and the percentage space utilisation. These parameters are important as
conceivably a link could be established between the uniqueness in the two-dimensional
peak displacements and the system performance parameters.
In order for the sample to be employed as a means of determining differences in
selectivity changes, and to generate quantitative chemical fingerprints, components that
elute from the system must be assigned a coordinate within the two dimensional domain.
This paper outlines the approach employed, starting from the acquisition of the data,
plotting the two-dimensional separation as a contour, then extracting the peak retention
time information of the sample and finishing with the statistical analysis of the separation
performance.
I.2 Experimental
I.2.1 Chemicals and Samples
HPLC grade methanol (MeOH) and tetrahydrofuran (THF) were purchased from Merck
(Australia). Milli-Q water (18.2 MΩ, obtained in-house) was used throughout. Red
Delicious apples were obtained from fruit sourced from the local market, and used as
obtained following a washing in cold water. The apple flesh was extracted with 80% v/v
MeOH/water and stored at 4 ºC in the absence of light for the further experiments. All
samples prior to injection into the LC system were filtered through a 0.45 µm filter.
I.2.2 Chromatographic Instrumentation
All chromatographic experiments were conducted using a Waters 600E Multi Solvent
Delivery LC System equipped with Waters 717 plus auto injector, Waters 600E pumps,
Waters 2487 series UV/VIS detectors and Waters 600E system controller. Two columns
156
were chosen for this work, a Phenomenex Luna 100 Ǻ CN column (150 × 4.60 mm, 5 m
particle diameter), and a Phenomenex Synergi 80 Ǻ Hydro-RP column (C18-Hydro) (150
× 4.60 mm, 4 m particle size) (Phenomenex, Lane Cove, NSW, Australia). The
chromatographic interface between the first and second dimension columns consisted of
two electronically controlled VICI two-position six-port valves fitted with micro-electric
two-position valve actuators that allowed alternate sampling of the eluent from the first
dimension into the second dimension.
I.2.3 Chromatographic Separation
I.2.3.1 Two dimensional separation environments
A separation of an apple flesh extract was used to illustrate the process of analysis and
peak recognition in 2D HPLC. The apple extract separation was undertaken on a cyano
stationary phase in the first dimension, running an aqueous/THF mobile phase under
gradient conditions, where the initial mobile phase was 0% THF and the final mobile
phase was 100% THF over a period of 10 minutes. The second dimension employed a
C18-Hydro column with an aqueous/MeOH mobile phase operated under gradient
conditions, whereby the initial composition was 0% MeOH and the final composition
was 100% MeOH, over 10 minutes. All flow rates were 1 mL min-1 and the transfer
volume from the first dimension to the second dimension was 200 µL. UV detection was
set at 280 nm and the sampling rate was 5 Hz.
The two-dimensional separation illustrated in this work consisted of a series of
consecutive heart cut processes, whereby every second 200 µL fraction from the first
dimension was analysed in the second dimension in an on-line process. Each analysis was
undertaken in a separate run. A total of 96 analyses were undertaken over a period of
approximately 21 hours.
157
I.2.3.2 Data Processing
The peak detection and matching algorithms, all calculations and graphics were
constructed with Wolfram Mathematica 7 for Students (distributed by Hearn Scientific
Software, Melbourne, VIC, Australia).
I.3 Results
I.3.1 Collection of Chromatographic Data and Measurement of Peak Retention Time
For the chromatographer to qualitatively analyse the 2D HPLC experiments the data must
be arranged in a way that is visually easy to understand. A single 2D HPLC analysis will
usually produce output data in one of two forms. If the analysis was completed via a heart
cutting approach the output data will comprise a one-dimensional chromatogram, and a
corresponding second dimension chromatogram. If the heart cutting process is repeated
numerous times, then there will be the same number of second dimension chromatograms
as there were heart cutting processes undertaken (although the first dimension separation
would remain constant provided the same sample was tested each time). Alternatively, if
a comprehensive two-dimensional separation was employed the detector response of the
entire analysis would normally be a single data file that may contain rows that number in
the magnitude of hundreds of thousands. Depending on the instrument control software
the data will likely be output in a text format with either a single column that represents
the detector signal or in a two column format with the analysis time and detector response.
Regardless this data needs to be processed into an array so that the chromatographer can
derive useful information from the separation. To date, this aspect of 2D HPLC has been
only briefly studied, and there are very limited commercial software packages that have
the capabilities to import, display and perform analyses on 2D HPLC data.
I.3.1.1 Graphical representation of 2D HPLC
Effectively data obtained from a 2D HPLC analysis is a three dimensional data set, these
dimensions represent the first dimension retention time (or the cut time), the second
dimension retention time and the detector response. For the 2D HPLC data to be visually
represented all 2D HPLC data must be in this three column format. In the case of 2D
158
separations that were performed with a heart cutting approach the second dimension
separations will be contained in individual files. For graphical software packages to
display this data the entire two dimensional separation must be merged into a single
dataset (in the three column format).
Graphical representations are used for visualising the separation and obtaining an initial
indication of how well the 2D separation has performed. Often these graphical
illustrations are represented as 3D surface plots (Figure I.1(a)) or contour plots (Figure
I.1(b)). Figure I.1(b) is a good representation of a successful 2D separation. In particular
the peaks that have first dimension retention time 1tR = 13.4 minutes and 2tR = 18.7
minutes co-elute in the first and second dimensions respectively and appear as a single
component. However, these chromatograms are qualitative descriptions only and do not
tell the full story of the separation. For instance when comparing 2D contour plots it is
difficult to distinguish one forest of peaks from another and the strong absorbance of
some components makes it difficult to select a suitable detector response threshold (i.e.,
z-axis cut off). For example if the threshold is too great the components with a small
detector response (either due to low concentration or poor UV absorbance) will display
limited visibility. Alternatively if the threshold is too low peaks, with a strong response
that have similar elution times will not be identifiable. To resolve these issues the 2D
HPLC data needs to be analysed to determine the retention times for peaks in two-
dimensional space (i.e., first and second dimension retention times), thus quantifying the
separation with respect to displacement in the two-dimensional domain.
159
Figure I.1 (a) represents a 3D surface plot of apple flesh 2D comprehensive (off-line) heart-cut separation (0.2 minute increments, 200 µL cut volumes) using 1st D CN (aqueous-THF) and 2nd D C18 (aqueous-MeOH) gradient at 10% min-1 (segment from the first dimension between 13.0 to 14.4 minutes). The z-axis scale has been restricted so the less absorbing peaks can be observed. Figure I.1(b) is a contour plot of the same data. The dark regions represent 2D peaks with the darker regions having a greater detector response.
I.3.1.2 Automated peak detection
The quantification of 2D HPLC data requires the retention times of peaks be determined
in each one dimensional cut. Due to the large number of one-dimensional chromatograms
required for a comprehensive 2D HPLC analysis, a process should be enabled to
automate the detection of component retention times (tR) rather than manually examining
all components in many chromatograms. Often retention times of chromatographic peaks
are determined by least squares fitting peak functions to the chromatogram, however, this
technique begins to fail in cases where there are multiple overlapping or shouldering
peaks [I.11-I.13]. There are a vast number of peak models that can be fit to
chromatographic data that compensate for co-elution, an example of which is given in
reference I.14. However, in the absence of peak modelling automated peak detection can
also be achieved using the interpretation the first and second derivatives of each
chromatogram [I.15-I.19]. Vivó-Truyols and coworkers determined the derivatives of the
one-dimensional chromatographic response using Savitzky-Golay (SG) smoothing
methods [I.20, I.21]; the process of preparing the data and performing this analysis is
described below.
160
Initially the one dimensional chromatographic data is imported by a suitable
computational mathematics software package (Wolfram Mathematica 7 was used for this
work) to optimise the peak detection thresholds that are then applied to the entire 2D
chromatogram.
The chromatographic noise is calculated as the distance of point (pi) from the mean of the
neighbouring points (pi-1 and pi+1), for all data in the chromatogram. The median value is
the noise [I.19]. The noise value defines a threshold (e.g., multiplying the noise by 3),
thrh1, which must be exceeded in height in order for the detected response to be
determined as a peak. This removes spurious data points caused by lows levels of random
noise. Manual thresholds may also be set. The baseline drift is corrected using the
average detector response of the chromatogram across the void period of the separation.
The chromatogram is smoothed to compensate for the effects of noise amplification
during the calculation of first and second derivatives using SG smoothing procedures
according to the processes outlined by Savitzky and Golay [I.20], and Steinier et al.,
[I.21]. A second order polynomial with the smoothing window size selected
automatically with the Durbin-Watson test (DW) according to references 19 and 22 was
employed. The optimal smoothing window size was obtained when DW converges and
the correlation between the original and experimental data was minimised [I.19, I.23].
The second derivative defines the retention time, tR, of chromatographic peaks, while the
first derivative establishes the peak range, i.e., the peak start and finish. The minimum
values of the negative regions of the second derivative represent tR, and hence, it is
possible to determine the retention times of strongly overlapping and shouldering peaks
(artefacts from chromatogram noise are ignored by applying thresholds to the first and
second derivatives, εfd and εsd respectively). Figure I.2, for example, illustrates a
chromatogram that has undergone smoothing (a) and transformation to the first (b) and
second (c) derivatives. The first derivative of Figure I.2(b) is shown in Figure I.2(c) with
the εfd represented by a horizontal dashed line. Peak elution ranges are initially defined
when the smoothed chromatogram is greater than the threshold and is refined by
161
examining the first derivative. The peak elution range starts when the curve crosses the
first derivative at positive εfd, while having a positive gradient. The end point of the peak
elution range is where the curve again has a positive gradient though crosses negative εfd.
Figure I.2(d) is the second derivative of Figure I.2(b) with εsd represented by the dashed
line. In this example there are a total of three peaks (two overlapping peaks with clearly
visible peak maxima and two peaks with only one identifiable maximum) that the peak
detection algorithm has been able to identify and determine tR.
(a) (b)
(c) (d)
Figure I.2 An example of how derivatives of peaks can be used to determine retention times and peak regions from a chromatogram. Figure I.2(a) is the chromatogram to be analysed. Figure I.2(b) is an expanded section of the smoothed chromatogram; thrh2 is represented by the horizontal dashed line. The dots represent the retention time and peak height of the detected peaks. Figure I.2(c) is the first derivative of Figure I.2(b). The horizontal dashed lines represent εfd and the outer vertical lines represent the peak region define in Figure I.2(b). Figure I.2(d) is the second derivative of the chromatogram with εsd represented by the horizontal dashed line.
162
I.3.1.3 Peak Picking
Once the retention times have been determined for all peaks in the 2D HPLC
chromatogram quantitative data and selectivity analysis can be completed. However, due
to the nature of comprehensive analysis, where multiple cuts of a single peak are taken,
the same chemical component will be detected in adjacent chromatograms and will be
represented by several 2D retention times. Peters et al., [I.24] developed an algorithm
that compared peaks in adjacent chromatograms and then by applying a series of rules,
determined if the peaks were resulting from the same, or different, sources. This
algorithm was designed for GC × GC separations but is also valid for LC × LC
separations.
The software algorithm used in this work for 2D peak detection follows the same format
as outlined by Peters et. al., in reference I.24, though the results presented here were
acquired with an algorithm developed in house using Wolfram Mathematica 7. The
process of finding 2D peaks from 1D data involves examining the overlap of adjacent
peaks (calculated from first derivative). The peak maximum profile of grouped peaks is
then analysed and groups are divided at valleys. The peak maximum profile is enhanced
with 2D HPLC data. Detailed information about the peak matching algorithm can be
found in reference I.24.
At the completion of the peak picking algorithm the tR of the 2D peak is that with the
maximum detector response. This is illustrated in Figure I.3 where the peak detection
algorithm has detected 20 - 2D peaks in the small 1.4 × 3.2 minute window.
163
Figure I.3 The same separation displayed in Figure I.1 after the data has been applied to the peak detection algorithm. The blue points represent the 2D peak maxima and the red points are peaks that were detected in adjacent cuts that were deemed to be the same compound as those connected to it by a red line.
I.3.1.4 Enhancing Peak to Peak Resolution and Signal to Noise Response
In a separate publication [I.25] polynomial functions (in the form of xp, where x is the
data and p is a number between 1 and 4) were applied to the signal response for a series
of simulated and experimental one-dimensional chromatograms. SG smoothing
algorithms were also applied to the data. This process greatly improved the signal to
noise ratio of the peaks and decreased the peak variance while not changing Rt. This is
illustrated in Figure I.4 and Table I.1 whereby random noise was added to a model
Gaussian distribution and the power function was applied to the peak response. The SG
smoothing filter improved the signal to noise ratio tenfold, however, when the power
function was also applied to the peak intensity response, the signal to noise response
improved by a factor of 50000.
164
(a) (b)
(c)
Figure I.4 (a) Gaussian peak with simulated noise that is smoothed in (b) and applied to a polynomial function in (c). See Table I.1.
The polynomial smoothing function can be used to improve 2D HPLC data, and this
process can be incorporated into this peak picking program, for example, with the
addition of a few lines of code (or a simple loop) that transforms the data by the
polynomial function. Firstly, the data is normalised according to Equation I.1:
minmax
min, xx
xxx i
ni (I.1)
where xi,n is the normalised data point, xmin is the minimum signal response and xmax is the
maximum signal response. Once the data response is normalised the polynomial function
is applied to the signal response, Equation I.2:
pnipi xx ,, (I.2)
165
Table I.1 Simulated results of a single peak when applied to smoothing and polynomial functions. Rt and variance were calculated with the peak moments method.
To illustrate the gain in resolution and the improvement in the signal to noise ratio we
tested the process on a two-dimensional separation of an apple extract. Figure I.5(a) is the
two-dimensional surface plot of apple flesh extract, and Figure I.5(b) is an expanded
view of the region between 5.0 and 8.5 minutes on the first dimension and 11.0 and 14.5
minutes on the second. On these plots the dark regions represent the intensity of the
detector response, the darker the region the greater the detector absorbance. Figures I.5(c)
and I.5(d) show the detection response for the same region as in Figure I.5(b), however,
the signal responses were transformed using polynomial functions of x2 and x4. The signal
to noise response was greatly improved, as represented by the contrast between the
background and the peak, while the relative positions (two-dimensional retention times)
of the peaks did not change. A limitation of this application for the reduction in noise
response is that it yields a non-linear concentration response. However, the purpose of the
„enhancement‟ process is two-fold: (1) To increase baseline resolution between closely
eluting components in a crowded separation space, and (2) To minimise the noise
response, by improving the signal to noise ratio, which thus allows the analyst to
undertake a „fishing‟ expedition in the search for low concentration components that
potentially could serve as chemical signature identity markers. Both of these factors lead
to improvements in the surety of chemical signatures obtained using 2D HPLC.
166
(a) (b)
(c) (d)
Figure I.5 (a) an illustration of a 2D HPLC separation of an apple flesh extract. I.5(b) to (d) are expanded regions of this separation where the response has undergone different degrees of polynomial enhancement. I.5(b) emphasises the importance of selecting appropriate thresholds when analysing the data.
I.3.2 Evaluation of multidimensional separations
Determining separation correlation, and the practical peak capacity of the experimental
results presented in this work was conducted using the geometric approach to factor
analysis (GAFA) [I.27], a detailed explanation of which can be found in reference I.27,
however, a brief outline is described here. A GAFA allows the determination of the
correlation between dimensions, a measure of the practical peak capacity of the
separation space, a measure of the spreading angle between each dimension and
subsequently the percent usage of the separation space. The two-dimensional information
can be visualised by plots of the peak capacity in one dimension against the peak capacity
167
in the second dimension. A rectangular plot, the axes of which correspond to the peak
capacity in each dimension, illustrates the effective separation space utilized in the
separation process. The region bound between the spreading angle () is a measure of the
separation space utilisation. A truly orthogonal separation is achieved when 90, corresponding to complete utilisation of the theoretical separation space. As the
dimensions become more correlated the spreading angle decreases and the effectiveness
of the two-dimensional separation decreases. While this paper reports only the findings of
GAFA for the separation detailed, the Mathematica peak picking program can also detail
system performance measures derived from information theory [I.28] and the Bin
approach [I.25].
I.4 Discussion
Careful attention must be paid to the configuration of the peak detection section of the
algorithm in order to avoid, either the detection of false peaks due to the thresholds (thrh2,
εfd and εsd) being set too low or, information being lost if these thresholds are set too high.
Missing information is less important than the presence of false positives if the purpose
of the study is to determine selectivity differences between systems, because more than
likely there will still be enough components that can be used to validate a statistical
analysis. More so, since the components present in the higher concentrations are likely to
be the targeted components for maximising separation conditions. The presence of false
positives, however, could skew important statistical information derived from these plots.
It is more appropriate therefore to set higher thresholds and miss low concentration
components than to observe false peaks.
The peak detection and matching algorithm was applied here to the 2D HPLC separations
of apple flesh. The separations presented in this work illustrate the process of removing
false positive peaks, resulting from solvent effects following the transfer of mobile phase
from the first dimension to the second.
168
Figure I.5(a) is the two-dimensional contour plot of apple flesh extract, and Figure I.6 is
an expanded view of the region between 8.5 and 12.5 minutes on the first dimension and
13.0 and 17.0 minutes on the second. One dimensional peaks that are recognised by the
algorithm are represented by red points and 2D peaks are connected by red lines. The
retention times of the 2D peaks (i.e., 1D peak within the 2D peak that has the greatest
detector response) are represented by white points. This illustration shows that the
algorithm is able to successfully determine the 2D retention times of peaks automatically
from data obtained from 2D HPLC separations. In the expanded region shown in Figure
I.6 the algorithm detected 69 2D peaks.
Figure I.6 Expanded region of Figure I.5(a). White points represent peak maximum and red lines join 1D peaks deemed to belong to the same component.
Figure I.5(b) is an expanded section (between 5.0 and 8.5 minutes on the first dimension
and 11.0 and 14.5 minutes on the second) of the same separation of apple flesh shown in
Figure I.5(a). The intensity of this contour plot is amplified five-fold to that of Figure
I.5(a) so compounds that did not have strong detector responses could be visualised. In
this region the algorithm detected 30 2D peaks. Inspection of Figure I.5(b) shows there
are regions on the contour plot where peaks were not detected by the peak picking
algorithm. Decreasing the threshold values, however, increased the sensitivity of the peak
picking process, with the result being that 81 components were now recognised within
this region, as shown by the plot in Figure I.5(b). This illustrates the importance
associated with selecting the appropriate peak picking thresholds and also the difficulties
associated with visual representation of 2D HPLC data, as these peaks would be invisible
169
to the eye when the „z-axes‟ plot range was the same as Figure I.5(a). In total 187 peaks
were detected in the apple flesh extract illustrated in Figure I.5(a), but because of peaks
associated with solvent the actual number of peaks visually apparent is much larger.
Blank injections aid in the removal of these false peaks.
The importance and need to have access to an automated peak picking process is clearly
seen when numerical analysis that requires quantitative descriptions of peak maximum
profiles are attempted. With the 2D retention profile it is then relatively simple to design
a statistical analysis algorithm that assesses separation performance; an important aspect
associated with the optimisation of two-dimensional HPLC. To illustrate the statistical
analysis of separation performance GAFA was applied as a measure of separation quality
for the analysis of the apple extract.
The correlation between each dimension was 0.787. This indicates that despite both
separation processes being largely reversed-phase, there was considerable difference in
their retention processes. This is not surprising given the cyano phase essentially behaves
as a HILIC type stationary phase. The theoretical peak capacity of the apple separation
was 2550 (the peak capacity of the first dimension was 51 and the second dimension was
50). The spreading angle of this separation determined by the GAFA was 38.1º resulting
in a practical peak capacity of 1309. The results of the GAFA for the apple flesh
separation is illustrated in Figure I.7.
170
Figure I.7 Geometric approach to factor analysis when applied to the apple separation.
I.5 Conclusion
This paper has described the process of converting raw 2D HPLC data into information
that can then be used to quantitatively describe the selectivity of two-dimensional
retention data. This was illustrated by way of example on a 2D separation of the
components in an apple flesh extract using a comprehensive heart-cutting (off line
comprehensive) 2D HPLC approach. The algorithm quantitatively located 187 two-
dimensional peaks in the apple separation and was able to remove false peaks associated
with solvent, and multiple sample component transport from successive heart cut
fractions. As an illustration of the capabilities of the peak picking algorithm the peak
profiles were then subjected to a statistical analysis that measured separation performance
(i.e., a geometric approach to factor analysis). These types of analyses were relatively
simple to incorporate into the peak picking algorithm and resulted in a very powerful
means to analyse two-dimensional separation performance. This information can also be
applied to more sophisticated statistical algorithms for further examination of 2D HPLC
separations.
The 2D HPLC retention time profile that has been generated following this approach is a
chemical fingerprint. A detailed profile that detects all peaks may provide more
171
information about the sample than is required by the analyst. Compounds may only need
to be reported if they are of a certain concentration and the thresholds that are defined for
the peak detection can be adjusted to a suitable level. A series of future publications will
illustrate the importance and relative ease to obtain the chemical fingerprint of complex
samples and the ramifications that this has on quality control procedures, forensic
analysis and traceability.
I.6 References
I.1. M. Gilar, P. Olivova, A.E. Daly and J.C. Gebler, Anal Chem, 77 (2005), 6426-
6434.
I.2. S. Yan, W. Xin, G. Luo, Y. Wang, Y. Cheng, J. Chromatogr. A, 1090 (2005), 90-
97.
I.3. P. Xie, S.S. Hen, Y. Liang, X. Wang, R. Tian, R. Upton, J. Chromatogr. A, 1112
(2006), 171-180.
I.4. J.-L. Zhang, M. Cui, Y. He, H.-L. Yu, D.-A. Guo, Journal of Pharmaceutical and
Biomedical Analysis, 36 (2005), 1029-1035.
I.5. T.A. van Beek, J. Chromatogr. A, 967 (2002), 21-55.
I.6. C. Guo, G. Cao, E. Sofic, R.L. Prior, J. Agric. Food Chem., 45 (1997), 1787-1796.
I.7. M. Quadroni, W. Staudenmann, M. Kertesz, P. James, European Journal of
Biochemistry, 239 (1996), 773-781.
I.8. B. Thiede, A. Otto, U. Zimny-Arndt, E.-C. Müller, P. Jungblut, Electrophoresis,
17 (1996), 588-599.
I.9. T. Nagai, M. Sato, T. Nagai, S. Kamiyama, Y. Miura, Clinical Biochemistry, 22
(1989), 439-442.
I.10. E.P. Toups, M.J. Gray, G.R. Dennis, N. Reddy, M.A. Wilson, R.A. Shalliker, J.
Sep Sci., 29 (2006), , 481-491.
I.11. P. Nikitas, A. Pappa-Louisi and A. Papageorgiou, J. Chromatogr. A, 912 (2001),
13-29.
I.12. K. Lan and J.W. Jorgenson, J. Chromatogr. A, 915 (2001), 1-13.
I.13. T.L. Pap, Zs. Pápai, J. Chromotogr. A, 930 (2001), 53-60, J. Li, J. Chromatogr. A,
952 (2002), 63-70.
172
I.14. V.B. Di Marco, G.G. Bombi, J. Chromaogr. A, 931 (2001), 1-30.
I.15. G. Vivó-Truyols, J. R. Torres-Lapasió, A. M. van Nederkassel, Y. Vander Heyden
and D. L. Massart, J Chromatogr A, 1096 (2005), 133-145.
I.16. M.A. Korany, O.T. Fahmy, H. Mahgoub and H.M. Maher, Talanta, 66 (2005),
1073-1087.
I.17. B.R. Arora, B.N. Singh and R.K. Banerjee, J. Therm. Anal., 13 (1978), 499-508.
I.18. G. Espinosa and V.M. Castano, Appl. Radiat, Isot., 42 (1991), 377-381.
I.19. L. Antonov and S. Stoyanov, Anal. Chim. Acta, 324 (1996), 77-83.
I.20. A. Savitzy and M. J.E. Golay, Anal Chem, 36 (1964), 1627-1639.
I.21. J. Steinier, Y. Termonia and J. Deltour, Anal Chem, 44 (1972), 1906-1909.
I.22. J. Durbin and G.S. Watson, Biometrika, 37 (1950), 409-428.