:3:

 

 

 

 

 

“15818

97
(_Q (9(I.)L/

Lyn 3M9

This is to certify that the
dissertation entitled

IN SITU PROCESS MONITORING USING RAMAN
SPECTROSCOPY

presented by

Javier A. Falcon

has been accepted towards fulfillment
of the requirements for the

Chemistry and Chemical
Ph.D. degree in Engineering

 

 

%M@M

Major Professor’ 5 Signétfire
d QAWV
fl

Date

MSU is an Affirmative Action/Equal Opportunity Institution

 

LIBRARY '
Michigan State
University

 

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

JAN 1 0 2006

 

JAN 1 1 2006

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJClRC/DateDuopss-sz

 

IN SITU PROCESS MONITORING USING RAMAN SPECTROSCOPY
By

Javier A. Falcon

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Chemistry and
Department of Chemical Engineering & Material Science

2004

ABSTRACT

IN SITU PROCESS MONITORING USING RAMAN SPECTROSCOPY
By
Javier A. Falcon

The ability to monitor chemical processes in situ provides many advantages over more
conventional off-line analyses. Fundamentally, in Situ monitoring can aid in the
development of more in depth process understandings and allow for greater process
control. On a more practical level, in situ monitoring can reduce sample analysis time
and allow for increased analysis automation. These advantages ultimately translate to a
more robust and reproducible process. The studies presented here demonstrate the
superior capability Raman spectroscopy provides for in situ monitoring of various
processes.

Raman spectroscopy was used for in situ monitoring of antisolvent addition
crystallization of cortisone acetate. Raman spectral features of the solvent, the
antisolvent, and the solute were monitored to measure the relative concentration of the
components. Different rates of antisolvent addition were monitored and the resulting
solid-state form was characterized using Raman spectroscopy and differential seaming
calorimetry (DSC). Raman spectroscopy was used to monitor dehydration during storage
of the product crystals. This work demonstrates the advantages of using Raman
spectroscopy to monitor the solution phase during crystallization and simultaneously
monitoring the solid-state form.

In addition, Raman spectroscopy was used for in situ monitoring of anti-solvent

addition crystallization of progesterone. Raman spectral features of the solute and the

anti-solvent were subjected to principal components analysis (PCA) in order to attempt to
obtain information which is not readily apparent from the raw spectral data. For the
system utilized, PCA was able to distinguish Spectral features from the solute in solution,
the solute crystals and the anti-solvent. Furthermore, PCA was capable of detecting
subtle changes in the spectral data (as the addition of anti-solvent progressed) that could
be used as a warning for the onset of crystallization. The current study demonstrates the
advantages that can be gained by combining PCA with Raman spectroscopy for
monitoring crystallizations in situ.

A novel dispersive Raman spectroscopic procedure for the in sr’tu monitoring of fatty
acid cis to trans isomerization in edible oils is also described. The rapid determination of
trans isomer content in edible oils is of great importance in the characterization of fats
and oils in the food industry. Raman spectral changes observed during the high
temperature heating of oil were correlated with cis and trans isomer concentration
determined by gas chromatography (GC). The short acquisition times, low laser power
and ease of application of dispersive Raman spectroscopy using an NIR excitation source
combined with the correlation derived from the GC data provided the capability of
monitoring the isomerization in situ. The method developed here has potential
applications for increased process control of edible oil hydrogenation, deodorization and
quality control among other widely used processing steps. The in situ monitoring
capability provided by dispersive Raman spectroscopy offers considerable advantages

over the conventional off-line FT-Raman spectroscopy analysis.

ACKNOWLODGEMENTS

I would like to thank Prof. Kris A. Berglund for his guidance and support during my
time at MSU. His approach to managing our research lab has given me the opportunity
to grow as a scientist and as person.

I also have to thank my parents, my mom Joelle, my dad John and Kathy. You guys
have believed in me and been there for me every step of the way. Thanks to the best
three parents anyone could wish for.

Finally, I owe a BIG thank you (with a hug and a kiss) to Leslie. You managed to keep
me sane through the tough times and helped me celebrate the good times. Thank you for
everything you’ve done and for the sacrifices you’ve made for me. The best part is that

hopefiilly now I can begin to pay you back.

iv

TABLE OF CONTENTS

List of Tables ........................................................................................ ix

List of Figures ........................................................................................ x

Chapter 1: INTRODUCTION ...................................................................... 1

1-1 Theory of Raman Spectroscopy ................................................... 2

1-2 Background for Anti-Solvent Addition Crystallization Studies ............... 7

1-2.1 Crystallization .............................................................. 8

1-2. 1 . 1 Supersaturation ................................................... 8

1-2.1.2 Nucleation ......................................................... 9

1-2.1.3 Crystal Growth ................................................... 11

1-2.2 Tailor Made Additives ................................................... 13

1-2.3 Principal Components Analysis (PCA) ............................... 14

1-3 Background for Edible Oil Studies .............................................. 16

1-4 References ........................................................................... 18
Chapter 2: MONITORING OF ANT I-SOLVENT ADDITION CRYSTALLIZATIONS

WITH RAMAN SPECTROSCOPY ............................................... 20

2-1 Abstract ............................................................................. 20

2-2 Introduction ........................................................................ 20

2-3 Experimental Section ............................................................. 21

2-3.1 Materials and Instrumentation .......................................... 21

2-3.2 Calibration of Raman Spectra .......................................... 23

2-3.3 Anti-Solvent Addition Crystallization ................................. 23

2-3.4 Solid-State Analysis of Crystals ....................................... 23

2-4

2-5

2-6

2-7

2-3.5 Effect of Impurities ...................................................... 24

2-3.6 Principal Components Analysis ........................................ 24
Results and Discussion ........................................................... 25
2-4.1 Calibration of Raman Spectra .......................................... 25
2-4.2 Solubility Curve .......................................................... 25
2-4.3 Crystallizations ........................................................... 28
2-4.4 Solid-State Analysis ...................................................... 35
2-4.5 Effect of Impurities ...................................................... 40
2-4.6 Principal Components Analysis ........................................ 44
Conclusions ........................................................................ 47
Acknowledgements ............................................................... 47
References .......................................................................... 48

Chapter 3: IN SITU MONITORING OF ANTI-SOLVENT ADDITION

3-1

3-2

3-3

CRYSTALLIZATIONS WITH PRINCIPAL COMPONENTS

ANALYSIS OF RAMAN SPECTRA .............................................. 50
Abstract ............................................................................. 50
Introduction ........................................................................ 50
Experimental Section ............................................................. 52
3-3.1 Materials and Instrumentation .......................................... 52
3-3.2 Calibration of Raman Spectra .......................................... 53
3-3.3 Anti-Solvent Addition Crystallizations ................................ 55
3-3.4 Reproducibility of Crystallizations (a) ................................. 55
3-3.5 Blank Experiments (b) ................................................... 55
3-3.6 Varying Initial PG Concentration Crystallizations (c) ............... 55

vi

3-4

3-5

3-6

3-7

3-3.7 Principal Components Analysis (PCA) ............................... 56

Results and Discussion ........................................................... 56
3-4.1 Calibration of Raman Spectra .......................................... 56
3-4.2 Solubility Curve .......................................................... 5 7
3-4.3 Anti-Solvent Addition Crystallizations ................................ 60
3-4.4 Principal Components Analysis ........................................ 61
3-4.5 PCA of Reproducibility Crysallizations .............................. 64
3-4.6 PCA of Blank Experiments ............................................. 64

3-4.7 PCA of Varying Initial PG Concentration Crystallizations. . . . . ....69

Conclusions ........................................................................ 76
Acknowledgements ............................................................... 77
References .......................................................................... 78

Chapter 4: IN SITU MONITORING OF CHANGES IN SUNFLOWER OIL FATTY

4-1

4-2

4-3

4-4

4-5

ACID COMPOSITION DURING HIGH TEMPERATURE PROCESSING

USING A DISPERSIVE RAMAN SPECTROMETER .......................... 79
Abstract ............................................................................. 79
Introduction ........................................................................ 79
Experimental Section ............................................................. 81
4-3.1 Analytical Methods ...................................................... 83
Results and Discussion ........................................................... 84
Conclusions ........................................................................ 99
Acknowledgements .............................................................. 1 00
References ........................................................................ 101

vii

Chapter 5: CONTINUING INVESTIGATION: IN S] TU MONITORING OF OIL
OXIDATION DEGRADATION PRODUCTS USING A DISPERSIV E

RAMAN SPECTROMETER ...................................................... 102
5-1 Introduction ...................................................................... 102
5-1.1 Lipid Oxidation ......................................................... 103
5-1.2 Hydroperoxide Decomposition ....................................... 107
5-2 Experimental Section ............................................................ 107
5-2.1 Analytical Methods ..................................................... 109
5-3 Results and Discussion ......................................................... 111
5-4 Conclusions ....................................................................... 120
5-5 Acknowledgments ............................................................... 121
5-6 References ........................................................................ 122
Chapter 6: SUMMARY AND CONCLUSIONS ............................................. 124
6-1 References ........................................................................ 131
APPENDIX ......................................................................................... 132

viii

LIST OF TABLES

Table 4.1 Assignment of the major Raman scattered bands of sunflower
oil .................................................................................... 88
Table 4.2 Fatty acid composition of sunflower oil as a function of temperature
determined .......................................................................... 90

ix

Figure 1.1

Figure 1.2

Figure 1.3

Figure 2.1

Figure 2.2

Figure 2.3

Figure 2.4

Figure 2.5

LIST OF FIGURES

Vibration energy level diagram illustrating the energetic transitions
associated with Raman spectroscopy. h represents Planck’s constant, 0.,
indicates the excitation source frequency, v is the vibrational quantum

number and AD is the frequency difference between the ground state and
the first vibrational excited state .................................................. 4

Mechanisms ofnucleation.............................................................10

Crystal growth model. Dark gray (middle sphere) represents molecule of
interest. Light gray (surrounding spheres) represent solvent molecules.
A) transport of species from the bulk solution to a site at the crystal
surface, B) adsorption of the grth unit to a site at the crystal surface, C)
surface diffiision from the impingement site to a growth site D)
incorporation into the crystal lattice, E) initial and final solvent
losle

Schematic of the experimental setup. (1) Raman probe, (2) HoloLab 5000,
(3) PC, (4) water circulator, (5) jacketed reactor, (6) magnetic stir bar, (7)
water reservoir and pump, (8) digital thermometer ........................... 22

Molecular structure of cortisone acetate (CA) ................................. 26

Solubility of cortisone acetate in binary mixture of acetone/water at
224°C .............................................................................. 27

Real time change in the baseline peak area of the Raman spectrum
between 750—850 cm-1 for the first set of crystallizations as described in
the text. The symbols represent different water addition rates: (A) 0.25,
(D) 0.50, (O) 1.0 mL/min. Line A shows the change in concentration as
water is added to the solution, line B follows the sharp decrease in peak
area caused by turbidity in the solution which is caused by nucleation.
Line C follows the solution after the nucleation
event ................................................................................. 29

Real time change in the baseline peak area of the Raman spectrum of
cortisone acetate between 750-850 cm-1 for the second set of

Figure 2.6

Figure 2.7

Figure 2.8

Figure 2.9

Figure 2.10

Figure 2.11

Figure 2.12

crystallization experiments as described in the text. The symbols
represent different water addition rates: (A) 0.25, (D) 0.50, (O) 1.0
mL/min. Line A shows the change in concentration as water is added to
the solution, line B follows the sharp decrease in peak area caused by
turbidity in the solution which is caused by nucleation. Line C follows the
solution after the nucleation event .............................................. 31

Real time change in the baseline peak area for the Raman spectrum of
cortisone acetate between 1600-1640 cm-1 for the second set of
crystallization experiments. The different symbols represent different
water addition rates: (A) 0.25, ([1) 0.50, (O) 1.0 mL/min. Line A shows
the change in concentration as water is added to the solution, line B
follows the sharp decrease in peak area caused by turbidity in the solution
which is caused by nucleation. Line C follows the solution after the
nucleation event ................................................................... 32

Comparison of cortisone acetate Raman spectra for different environments
in the region between 1600-1680 cm". Spectrum (1): as received crystals.
Spectrum (2): solution of CA, acetone and water. Spectrum (3): CA
crystals in slurry. Spectrum (4): crystals after drying ........................ 34

Cortisone acetate crystals resulting from the three different rates of water
addition used. These photographs are at the same magnification and
encompass identical viewing areas. The pictures represent the different
water addition rates. (A) 0.25, (B) 0.50, (C) 1.00 mL/min .................. 36

Time resolved Raman spectra of cortisone acetate acquired daily to
monitor post crystallization phase transformation. The first spectrum in
was acquired the day of crystallization (day 0 after crystallization). The
final spectrum was acquired 8 days after crystallization ..................... 38

The DSC therrnograms of (A) CA hydrate and (B) same crystals after loss
of water. Apparent endotherms or exotherms at around 180 C are due to
change in the temperature program heating rate .............................. 39

Molecular structure of bisnoraldehyde ......................................... 41

Peak width of the Raman spectrum of cortisone acetate between 1585-
1635 cm'1 for three crystallizations conducted with a water addition rate of
0.50 mL/min. All three crystallizations required three days for

xi

Figure 2.13

Figure 2.14

Figure 3.1

Figure 3.2

Figure 3.3

Figure 3.4

Figure 3.5

Figure 3.6

dehydration to begin which is reflected by the increase of peak width.
Dehydration required 5.33 :h 1.53 days for completion ...................... 42

Peak width of the Raman spectrum of cortisone acetate between 1585-
1635 cm'1 for three crystallizations in the presence of bisnoraldehyde. The
crystallization at 0.04 weight percent BA in solution(A) shows no
statistically meaningful difference from the reference crystallizations. The
crystallization with 0.15 weight percent BA in solution (0) required seven
days for dehydration to begin and 12 days for dehydration to be
completed. Finally, the crystallization with 0.22 weight percent BA in
solution (D) had not begun dehydration after 12 days ........................ 43

The third factor obtained from PCA of the Raman spectra of cortisone
acetate in aqueous acetone solutions for the regions of 750-850 cm'1 and
1600-1630 cm'1 for the second set of crystallizations described in the text.
The different symbols represent different water addition rates: (0) 0.25,
(:1) 0.50, (A) 1.0 mL/min. The solution becomes saturated at the first
inflection point in time ............................................................ 46

Schematic of the experimental setup. (1) Raman probe, (2) HoloLab 5000,
(3) PC, (4) water circulator, (5) jacketed reactor, (6) magnetic stir bar, (7)
water reservoir and pump, (8) digital thermometer ........................... 54

Molecular structure of progesterone (PG) ..................................... 58

Solubility of progesterone in binary mixture of acetone/water at 22.4 °C
........................................................................................ 59

Rarnan spectra obtained for a PG crystallization experiment with an initial
PG concentration of 7.0wt %. The region enclosed by the dashed lines is
the region being investigated with PCA ........................................ 62

Principal component loadings of Raman spectra obtained from an initial
PG concentration of 7.0wt %. (PC 1) is the first principal component, (PC
2) is the second principal component, (PC 3) is the third principal
component .......................................................................... 63

Scores plots for the three reproducibility crystallization data sets. The
scores were calculated by constructing one PCA model using the three

xii

Figure 3.7

Figure 3.8

Figure 3.9

Figure 3.10

Figure 3.11

Figure 4.1

Figure 4.2

data sets. The similarity between the scores values indicates the
reproducibility of the experiment. Arrow indicates direction of increasing
water concentration ............................................................... 65

Raman spectra obtained for a "blank" experiment (no PG present). The
region enclosed by the dashed lines is the region being investigated with
PCA. Arrows highlight how the spectra change as water is
added ................................................................................ 66

Principal component loadings of Raman spectra obtained from a "blank"
experiment (no PG present). (PC 1) is the first principal component, (PC
2) is the second principal component. Numbers in italics indicate peak
position in cm' .................................................................... 68

PC 1 scores versus cumulative water volume (ml) plot from a "blank"
experiment (no PG present) ...................................................... 70

Principal component loadings of Raman spectra obtained for varying
initial PG concentrations. (A) 3.5wt%. (B) 7.0wt%, (C) 14.0wt%. (PC1)
first principal component, (PC2) second principal component, (PC3) third
principal component. Numbers in italics indicate peak position in
cm” .................................................................................. 71

Principal components scores plot of Raman spectra obtained for
crystallization with 7.0wt% initial PG concentration. The score plots were
calculated after a specific volume of water was added to the original
solution. a) 1.52 ml, b) 4.81 ml, 0) 8.14 ml and d) 12 ml. Beginning of
crystallization is circled in score pot c). Arrows indicate direction of
increasing water concentration. The circled point on plot (c) indicates the
beginning of crystallization ....................................................... 75

Schematic of the experimental setup. (1) Raman probe, (2) HoloLab 5000,
(3) PC, (4) hot-plate, (5) temperature controller, (6) thermocouple, (7)
stainless steel propeller stirrer .................................................... 82

Fluorescence emission spectra of sunflower oil obtained for oil at room
temperature. (A) 632.8 nm excitation, (O) 785 nm excitation .............. 85

xiii

Figure 4.3

Figure 4.4

Figure 4.5

Figure 4.6

Figure 4.7

Figure 5.1

Figure 5.2

Figure 5.3

Figure 5.4

Figure 5.5

Raman spectrum of sunflower oil acquired at room temperature using a
dispersive Raman spectrometer ................................................. 87

Average peak area and standard deviation for 1745 cm'1 peak plotted
against time for oil heated at 300 °C for 40 minutes ........................... 93

Peak area ratio 1265/1745 plotted against the total %cis concentration as
determined by GC. Linear fit displays reasonable correlation coefficient
of 0.95. (O) unheated sample, (X)200 °C, (A)250 °C,(O)300
°C .................................................................................... 94

Peak area ratio 1745/ 1265 plotted against the total %trans concentration as
determined by GC. Linear fit displays reasonable correlation coefficient
of 0.93. (O) unheated sample, (X) 200 °C, (A)250 °C, (0) 300
°C .................................................................................... 95

Peak area ratio 1745/1265 plotted against time for the three test
temperatures. (X) 200°C, (A)250°C, (0) 300°C ............................. 97

Molecular structure of linoleic and linolenic acids .......................... 105

A) Reaction mechanism of triplet oxygen with linolenic acid. B) reaction
mechanism of singlet oxygen with linoleic acid‘ ............................ 106

General decomposition pathway of hydroperoxide to produce volatile
compounds ........................................................................ 108

Schematic of the experimental setup. (1) Raman probe, (2) HoloLab 5000,
(3) PC, (4) hot-plate, (5) temperature controller, (6) thermocouple, (7)
stainless steel propeller stirrer, (8) dry air input ............................. 110

Difference Raman spectra obtained for the experiment conducted at 130
°C with 0.5 g of TBHQ added to the oil. Difference spectra calculated by
subtracting the first spectrum acquired once the oil reached 130 °C minus
A) the spectrum acquired after 180 minutes at 130 °C B) the spectrum
acquired after 5 minutes at 130 °C ............................................. 113

xiv

Figure 5.6

Figure 5.7

Figure 5.8

Difference Raman Spectra obtained for the experiment conducted at 130
°C with no TBHQ added to the oil. Difference spectra calculated by
subtracting the first spectrum acquired once the oil reached 130 °C minus
A) the spectrum acquired after 180 minutes at 130 °C B) the spectrum
acquired after 5 minutes at 130 °C ............................................. 114

First derivative spectrum calculated for differences spectrum A shown in
First derivative reveals a “hidden” peak at 1640 cm which Is not visible
in the raw spectral data due to overlap with a peak at 1656 cm

..................................................................................... 117

Difference Raman spectra obtained for the experiment conducted at 200
0C with no TBHQ added to the oil. Difference spectra calculated by
subtracting the first spectrum acquired once the oil reached 200 °C minus
A) the spectrum acquired after 180 minutes at 200 0C B) the spectrum
acquired after 5 minutes at 200 °C. . . . . . . . . .. ............................................ 118

XV

Chapter 1

INTRODUCTION

Historically, most industrial production has involved the manufacture of the finished
product, followed by laboratory testing and analysis to verify its quality, physical and
chemical properties. Recently, however, there has been an increase in interest by
industries and governmental regulatory agencies regarding in situ (on-line) process
monitoring and process control methods. The Process Analytical Technologiesl (PAT)
initiative is a good example of the increased interest. PAT is a joint initiative between
the pharmaceutical industry and the Food and Drug Administration (FDA) which aims to
utilize emerging process engineering and manufacturing technologies to achieve high
levels of quality through the use of in situ process monitoring and control. The main
reason for the formation of initiatives such as PAT is that the adoption of these in situ
monitoring and control methods provide companies with the opportunity to lower
manufacturing cycle times, increase production efficiency and diminish end-product
variability. This can in turn result in shorter time-to-market and a reduced likelihood of
product failures.

Raman spectroscopy provides many advantages over other analytical techniques for
monitoring processes in situ. This work consists of two major sections in which novel
methods and applicatiOns of Raman spectroscopy are described for the in situ monitoring
of two widely used industrial processes. The first section focuses on the in situ
monitoring of anti-solvent addition crystallizations of active pharmaceutical ingredients
(APIs) using Raman spectroscopy and principal components analysis (PCA). The second

section aims at using Raman spectroscopy to monitor in situ the cis-trans isomerization

of fatty acids and the formation of oxidation products in edible oils during high
temperature processing.
l-l Theory of Raman Spectroscopy

Raman spectroscopy can be used to obtain information about electronic and vibrational
transitions; this work will only deal with its use as a vibrational spectroscopy. When a
molecule is irradiated by monochromatic radiation of frequency v0, the oscillating electric
field of the light causes a short-lived distortion of the electron cloud of the bonds in the
molecule. The distortion of the electron cloud induces a dipole moment oscillating with
frequency v0, the magnitude of which is proportional to the “elasticity” of the electron
cloud, i.e., the electron polarizability. The oscillating dipole gives rise to “unshified”
scattered radiation also known as Rayleigh scattering—that is scattered radiation at the
same frequency as the excitation radiation.

Molecular vibrations cause the electron cloud belonging to the bonds in the molecule to
expand and compress. These vibrations may change the electron cloud size and/or shape
and this change translates to a change in the polarizability of the bonds involved with
those clouds. The polarizability of the molecule can therefore be modulated with
frequencies of vibrations v,. This means that some of the irradiating radiation will also be

modulated during scattering, resulting in “shifted” emission lines, the Raman lines, the
frequencies of which are V}; = v0 d: vs.

For the classic description of Raman spectroscopy discussed above”, the strength of
the induced dipole, p, scales with the second-order polarizability tensor, 0t, and the
incident electric field, E:

A: aE (1-1)

Each of the polarizability tensor components ajj depend on the normal coordinates of
the molecular vibration, Q. It is possible to expand each component using a Taylor series
with respect to the normal coordinates about the equilibrium position of the molecule.
Ignoring terms of second or higher order in the Taylor series leaves behind an
approximation that is referred to as the electrical harmonic approximation3

((11191: = (oar-)0 + (at-1):. Qk (1-2)
where k identifies the kth normal mode and or’u- is the derivative of or” with respect to Qk.
Assuming the normal coordinates execute simple harmonic motion (mechanical
harrnonicity), and writing the electric field in Sinusoidal form gives the induced
polarization3

P 2 [do + {a'kaaCOSWk + (150)] *EoCOS(Uot) (1-3)
where Q“, is the normal coordinate amplitude, 0k is the frequency of the kth normal
mode, (bk is a phase factor, 0., is the frequency of the excitation radiation, t is time and a’k
is the derivative of the polarizability tensor. Using the trigonometric identity
cos(A)*cos(B) = [cos(A+B) + cos(A-B)]/2 allows Eq. (1-3) to be expressed as3

P = aoEo[cos(Uot)] +

aOEoQko[cos {(00 _ Uk )t _ ¢k} + COS {(00 + Uk )t + ¢k}] (1-4)
2

 

Eq. (1-4) provides a mathematical representation of the Rayleigh and Raman scattering
process. It predicts that if light irradiates matter that is much smaller than its wavelength
(in our case molecules) the incident light will be scattered at three different frequencies.
The first term on the right accounts for Rayleigh scattering, which is at the same

frequency as the excitation source, and has a magnitude proportional to are. The second

Stokes Rayleigh Anti-Stokes

 

 

 

 

 

 

 

 

 

. Scattering Scattering Scattering
vrrtual_ _______________________
state “ i “ ‘ l

heo hoo
hoe h(\)o ' AU) hue h(Uo "' AU)
v = 1 "
v ‘7 AD
v = 0

Figure 1.1. Vibration energy level diagram illustrating the energetic transitions
associated with Raman spectroscopy. h represents Planck’s constant, 1)., indicates the
excitation source frequency, v is the vibrational quantum number and AD is the frequency
difference between the ground state and the first vibrational excited state.

and third terms account for Stoke and anti-Stokes Raman scattering respectively. Figure
1.1 shows the transitions associated with Stokes, Rayleigh and anti-Stokes scattering.
Although Eq. (1-4) is incomplete (because it was derived classically), it demonstrates
most of the important characteristics of Raman spectroscopy. First, polarization and
scattering (Rayleigh and Raman) intensities correlate linearly with laser intensity.
Second, only vibrations that change the polarizability (consequently a”, :4 0) yield Raman
scattering. This statement is the basis of the gross selection rule for Raman scattering.
Third, Eq. (1-4) demonstrates that anti-Stokes and Stokes scattering occur at higher and
lower frequencies (respectively) than the excitation fiequency. Fourth, two important
points that are not immediately apparent from Eq. (1-4), or’k may vary significantly for
any given molecule, this leads to wide variations in Raman scattering intensity strongly
depending on the molecule being probed. And, or’k is generally much smaller than do,
therefore Raman scattering is much weaker than Rayleigh scattering (approximately
1,000 times weaker).

Equation (1-4) does explain the variation of Raman intensity with the fourth power of
the excitation frequency for normal Raman scattering. The 1)4 factor (which can be seen
in Eq. (1-5)) is derived from the classical treatment of scattering from an oscillation-
induced dipole. For the same compound, the Raman spectrum will show the same
frequency shifi regardless of the excitation frequency, but the spectrum will be more
intense as the excitation frequency is increased.

Finally, Eq. (1-4) does not explicitly incorporate one of the most important
experimental parameters of Raman spectroscopy, temperature. Because Stoke and anti-

Stokes Raman scattering depend on the population of the ground and the first excited

vibrational state respectively, the ratio of the scattering intensity is related to temperature

by the Boltzmann distribution2

4
[MS _ (00 + Uk )4 exp[_huk]
) kT

 

(1-5)

IR.S — (00 -Uk

where IRAS is anti-Stokes scattering intensity, IRS is the Stokes scattering intensity, k is
Boltzmann’s constant, T is temperature and the other symbols have been defined above.
For a given molecular vibration, the anti-Stokes scattering intensity will increase while
the Stokes scattering intensity decreases. This effect becomes more pronounced as the
temperature is increased. In addition, molecular vibrations with higher frequency (higher
energy and lower Raman shift) will be affected more than lower fi'equency vibrations.
This means that at a given temperature vibrations with lower Raman shifts will have a
larger anti-Stokes to Stokes intensity ratio than vibrations with higher Raman shifts.
Apart from the effects of polarizability, laser wavelength, temperature conditions and
other parameters discussed above on the Raman scattering intensity, it is ofien the case in
analytical applications that the important parameter for Raman scattering intensity is the
empirically determined Raman cross section, Gj. Parameter Gj is proportional to the
probability of an incident photon being scattered as a Raman-shified photon with a
particular Raman shiftz. Equation (1-6) relates the Raman scattering intensity (in watts)
to the cross sectionz, with laser intensity ([0) in watts.
IR = I0 adez (1-6)
where D is the number density of scatterers (molecules per cubic centimeter) and dz is the
path length of the laser in the sample and 0,- tracks with O4 for reasons discussed above.

Equation (1-6) is now antiquated since modern Raman spectrometers almost exclusively

use photon counting detectors rather than measure watts. For this reason it is convenient
to rewrite Eq. (1-6) for photon counting systems2

PR = P00,- ’Ddz (1-7)
PR and P0 now have units of photon per second and aj’ has different frequency

dependence than that of Eq. (1-6). Because of this, when PR is measured in

photons per second, the Raman intensity scales with 00(00 - 01-)3 rather than (00 -
of.

1-2 Background for Antisolvent Addition Crystallization Studies

Crystallization is one of the most widely used chemical engineering operations in the
food, pharmaceutical and chemical industries. Of particular interest to the
pharmaceutical industry is the ability to monitor and control the solid-state form that
crystallizes. The solid-state form of a crystal, whether it is a polymorph, hydrate or a
solvate will have significant consequences for the physical properties of the final crystal
product. In order to control the physical and chemical properties of the crystal products it
is necessary to monitor the crystallization process itself. Therefore, the in situ monitoring
of crystallization is the first step to controlling the final product that crystallizes.

Historically, a variety of techniques have been used to monitor the crystallization
process in situ. Some of these include ATR-FTIR, turbidity and nephelometry. The
main disadvantage these techniques share is their inability to provide information with
regard to the solid-state form crystallizing. The development of these techniques
assumed that if a process involved the crystallization of one molecule, there was only one

kind of solid-state form that could crystallize. The increase in polymorphs being found in

the pharmaceutical industry has created a need for an analytical technique that can
simultaneously provide information about the solution and the solid state. We monitored
the crystallization process in situ using Raman spectroscopy, and upon crystallization
were immediately able to discern the solid state form that has crystallized. Furthermore,
we were able to utilize Raman spectroscopy to simply and efficiently monitor the solid-
state form of the product crystals during storage.
1-2.1.1 Cgstallization

Crystallization begins with the generation of supersaturation. Once enough
supersaturation is generated the solute molecules aggregate for form nuclei, this process
is referred to as nucleation. Further integration of molecules to the nuclei is defined as
crystal growth. Crystallization is a process governed by both the thermodynamics and
the kinetics of the interface between the crystalline phase and the solution phase. The
driving force behind the onset of crystallization is the disturbance of the thermodynamic
equilibrium between a supersaturated and a saturated solution. Nucleation and crystal
growth are governed more by the kinetics of mass transfer from the bulk solution phase
to the solid phase.
1-2.1.l Supersguration

Supersaturation is a thermodynamically metastable state in which nucleation and
crystal growth occur. Under supersaturated conditions, the chemical potential difference
provides the driving force for the phase transition from solution to solid crystals.

Supersaturation, S, is defined as S = a/a*, where a is the activity of the supersaturated
solution and a"' is the activity of the saturated solution. Typically the ratio of activity

coefficients is assumed to be unity; therefore, we can replace activity with concentration.

Under this assumption, supersaturation can be approximated as the ratio of the
concentration of the supersaturated solution to the concentration of the saturated solution
at a given temperature. While recent studies4 have shown that this assumption is
questionable, it is often used as a practical necessity.
1-2.1.2 Nucleation

Supersaturation alone is not enough cause for a system to begin to crystallize. Before
crystals can develop, tiny crystallites of the smallest size capable of independent
existence (termed nuclei) are formed in the supersaturated phase. This occurs when a
specific supersaturation, known as the metastable supersaturation (Acme!) is obtained in
the system. This metastable region comes about because crystallizations don’t always
begin when the solubility limit is exceeded. Therefore, at every temperature there is a
lower limit to supersaturation (the solubility limit) and an upper bound (the
supersaturation limit). Between these boundaries nucleation is possible but not a
certainty.
As shown in Figure 1.2, nucleation may be primary (which does not require preexisting
crystals of the substance that crystallizes) or secondary (in which nucleation is induced
by preexisting crystals of the substance). Primary nucleation may be homogeneous,
whereby the nuclei of the crystallizing substance arise spontaneously in the medium in
which crystallization occurs, or heterogeneous, whereby the nuclei incorporate foreign
solid matter such as dust particles or the walls of the container. It is widely accepted that

true homogeneous nucleation is not a common event.

 

 

 

 

 

 

 

 

 

 

Types:
1) Initial Seeding
2) Attrition Breeding

 

Secondary

 

, Homogeneous
I | Primary
, ' I Heterogeneous ]
LNucleation |.__1 .

 

 

Figure 1.2. Mechanisms of nucleation4.

10

There is no single nucleation theory that is accepted for predicting nucleation rate, but
several correlations based on the power law model have satisfactorily explained the

experimental data“.

B = kNAC”

me, (1-8)
Where B is the nucleation rate, kN is the nucleation rate constant and ACM is the
metastable zone width. The exponent n is the apparent order of nucleation. The
nucleation process eventually reaches equilibrium between the growth of nuclei and the
dissolution of nuclei. Some of the nuclei reach a critical size (r*), which is dependent on
the change of the free energy of the particle. Nuclei that are smaller than r* tend to
dissolve, whereas those larger than r* are true nuclei and tend to grow to form

macroscopic crystals“.

1-2.l.3 Crystal Growth

 

The final stage of the crystallization process is crystal growth. This stage is the time
when nuclei grow into macroscopic crystals. Crystal growth can be thought of as a
succession of events (Figure 1.3): a) transport of species from the bulk solution to a site
at the crystal surface (with initial solvent loss), b) adsorption of the growth unit to a site
at the crystal surface, c) surface diffusion from the impingement site to a growth site (I)
incorporation into the crystal lattice7 (with final solvent loss). The rate-limiting step may
be any of the above depending on the grth conditions such as supersaturation,

temperature, additives or solvent, and the hydrodynamics of the system.

11

 

 

 

 

 

 

Al
.6 a/o;
,. Pi /

Figure 1.3. Crystal grth model4. Black (middle sphere) represents molecule of
interest. White (surrounding spheres) represent solvent molecules. A) transport of
species from the bulk solution to a site at the crystal surface, B) adsorption of the growth
unit to a site at the crystal surface, C) surface diffusion from the impingement site to a
growth site D) incorporation into the crystal lattice, E) initial and final solvent loss.

 

 

 

 

 

 

 

 

 

 

 

 

12

The basic expression used to relate crystal grth to supersaturation is4
G = k gsg (1-9)

this equation uses a linear crystal growth velocity, correlating the growth rate (G) with
the supersaturation (S). kg is defined as the growth rate constant and the exponent g is the
order of crystal growth.

A number of models have been proposed to represent crystal growth. Many of these
models are mathematically complex, and only provide a tentative basis for the correlation
of experimental data4’ 8.

1-2.2 Tailor-Made Additives

There is an enormous amount of literatureg’l0 demonstrating the design principles for

tailor-made additives to control and inhibit the grth of molecular crystals.
The basic idea behind tailor-made impurities is best appreciated by realizing that a crystal
surface is a termination of the bulk crystal structure and that the different surfaces of a
crystal will expose different parts of the crystallizing molecule to the external
environment. This means that the stereochemistry and functionality of each surface will
be different.

To design additive molecules that capitalize on this feature of crystals, two rules must
be satisfiedg’lo’“. First, the additive must be able to enter the surface lattice site. To do
this, part of the additive molecule must resemble the crystallizing host. Second, once in
the crystal surface, the additive molecule must be capable of preventing crystallization by
discouraging other host molecules from joining the crystal. To fulfill this role, part of the

additive molecule must differ from the host either in its functionality or its

13

stereochemistry so that either electrostatic or steric effects can be used to prevent
crystallization.
It has long been recognized that the presence of impurities substantially affects the
kinetics of crystal nucleation, growth and dissolution). It is possible that by affecting
these processes, the impurity will promote the crystallization of a metastable solid-state
form.
1-2.3 Principal Components Analysis
Mathematically, PCA is based on an eigenvector decomposition of the covariance or
correlation matrix of the process variables. For a given data matrix X with m rows and n
columns, with each variable being a column and each sample a row, the covariance
matrix of X is defined as
cov(X) = xTx /m-1 (1-10)
with the condition that the columns of X have been “mean centere ” (by subtracting off
the original mean of each column to create columns that have a mean of zero). If the
columns of X have been “autoscaled” (by dividing each column by its standard deviation
to ensure columns with zero mean and unit variance), Eq. (1-10) gives the correlation
matrix of X. The data presented in this work was only mean centered. PCA decomposes
the data matrix X into a sum of the outer products of vectors ti and p.- plus a residual
matrix E:
X=tlpT1+t2pT2+...+tkak+E (1-11)
Here k must be less than or equal to the smaller dimension of X. The t,- vectors are
known as scores and contain information on how the samples relate to each other. Score

plots feature the scores of one principal component plotted versus the scores of another

14

principal component. These plots allow investigators to view the relevant variation of the
data set in a smaller number of dimensions. Samples that are close to each other on a
given score plot are similar with respect to the original measurements. This mathematical
proximity translates to chemical similarity if meaningful measurements have been
made”.

The pi vectors are known as loadings and contain information on how the variables
relate to each other. Principal component loadings are used to determine which variables
(Raman shift for Raman spectroscopy) influence a particular principal component. In
addition, principal component loading plots aid in the visualization of principal

components I 2 ’1 3 .

In the PCA decomposition, the p, vectors are eigenvectors of the covariance matrix,
therefore for each p.-

COV(X) P1 = 7% p1 (1-12)
where it, is the eigenvalue associated with the eigenvector pi. The ti form an orthogonal
set (ti tTj = 0 for i at j), while the p.- are orthonormal (pTi pj = 0 for i 73 j, pT, pj = 1 for i = j).
It should be noted that for X and any ti, p, pair

X P1 = ti (1-13)

This is due to the fact that the score vector t, is the linear combination of the original X
data defined by p,. The t,, pi pairs are arranged in descending order according to the
associated 7g. The it, are a measure of the amount of variance described by the t;, pi pair.
It is possible then to think of variance as information. Because the ti, p, pairs are in
descending order of M, the first pair captures the largest amount of information of any

pair in the decomposition. In fact, it can be shown that the tr , p1 pair captures the greatest

15

amount of variation in the data that is possible to capture with a linear factor. And, each
subsequent pair captures the greatest possible amount of variance remaining after
subtracting tipTi from X.

It is generally found that using PCA allows the data to be adequately described using
far fewer factors than original variables. Therefore, confusion created by data overload
can be solved by observing the data using fewer scores than original variables. It should
be also mentioned that many times PCA aids in finding combinations of variables that are
useful in describing certain phenomena. The combinations of variables are often more
reliable indicators of sample conditions than single variables because of the signal
averaging aspects of PCA.

There are a number of algorithms (the power method”, the Jacobi method”, singular
value decomposition [SVD]14, and NIPALSIS) used to calculate the loadings and scores
for PCA. SVD is the preferred algorithm because it is most stable under the widest range
of applications, and it is the algorithm used in our report. A more in depth discussion of
the algorithms used for PCA is outside the scope of this work. For detailed discussions
of the individual algorithms and of PCA in general the reader is directed to the references

provided12,l6,l7,18,l9

1-3 Background for Edible Oil Studies

trans Fatty acids are present in a variety of food products; some are derived from
natural sources, but most come from products that contain commercially hydrogenated
fats. The nutritional properties of trans fatty acids have been debated for many years,
particularly with respect to the amounts of low density and high density lipoprotein

(LDL, HDL) contained in serum. Some studies have shown that trans fatty acids elevate

l6

levels of serum LDL (“bad”) cholesterol and lower HDL (“good”) cholesterolzo’u’zz.
Results like these have prompted the Food and Drug Administration (FDA) to implement
new regulations23 which make labeling of trans fatty acids mandatory on food products.

Current standard methods for determining the cis and trans content of fats and oils
include gas chromatography and Fourier Transform infiared spectroscopy (FT-IR) for
samples captured off-line. In the past, Raman spectroscopy has also been used for this
purpose24'25’26’27’28’29, although its use has been almost exclusively limited to Fourier
Transform (FT) Raman spectroscopy (because of fluorescence from the oil samples).
The major drawback to using FT -Raman spectroscopy is that current technology only
allows for off-line analysis of samples.

Within the past few years there have been technological advances that allow dispersive
Raman spectrometers to use near infrared (NIR) lasers as excitation sources. The use of
an NIR laser for excitation eliminates the interfering fluorescence from oil and fat
samples. The present work takes advantage of the flexibility and the in situ analysis
capability offered by dispersive Raman spectrometers and combines it with an NIR laser
excitation source. This powerful combination allowed us to use dispersive Raman

spectroscopy to qualitatively and quantitatively monitor changes (in situ) of the fatty acid

composition of sunflower oil during high temperature processing.

17

1-4 References

1. Food and Drug Administration (FDA), FDA Process Analytical Technologies (PAT)
page. URL http://www.fdagov/cder/OPS/PAThtm

2. McCreery, R. L.; Raman Spectroscopy for Chemical Analysis. Wiley-Interscience:
New York, 2000.

3. Long, D. A.; Raman Spectroscopy, McGraw-Hill: New York, 1977.

4. Myerson, A. S., Handbook of Industrial Crystallization, Butterworth-Heinemann Ltd.:
Oxford, 1993.

5. Polymorphism in Pharmaceutical Solids. Ed. Brittain, H.G., Marcel Dekker: New
York, 1999.

6. Boistelle, R., Advances in Nephrology, 1986, 15, 173.
7. Bennema, P., J. Crystal Growth, 1967, 1, 128.
8. Mullin, J. W., Crystallization, 3rd Edition, Butterworth-Heinemann Ltd.: Oxford, 1993.

9. Mersmann, S. Crystallization Technology Handbook. Marcel Dekker: New York,
1995.

10. Weissbuch, I. et al, in The Lock and Key Principle, J. P. Behr (ed), Chap. 6.
Chichester: Wiley, 1994.

ll. Blagden, N.; Davey, R.; Chemistry in Britain. 1999, 44.

12. Beebe, K. R.; Pell, R. J.; Seasholtz, M. B. Chemometrics: A Practical Guide; John
Wiley & Sons: New York, 1998.

13. ThermoGalactic website. URL hm://www.galactic.com/algorithms/pcahtm
14. HoloReact Software Operations Manual. Kaiser Optical Systems, Inc.

15. Press, W. H.; F lanner, B. P.; Teukoslky, S.; Vetterling, W. T. Numerical Recipes:
The Art of Scientific Computing, Cambridge University Press: Cambridge, 1986.

16. Malinowski, E. R. Factor Analysis in Chemistry, 2nd Ed.; John Wiley & Sons: New
York 1991.

17. Beebe, K. R. Kowalski, B. R. Analytical Chemistry, 1987, 59, lOO7A-1017A.

18

l8.

19.

20.

21

22.

23.

24.

25.

26.

27.

28.

29.

Jackson, B. B. Multivariate Data Analysis: An Introduction, Richard D. Irwin, lnc.:
Homewood, Illinois 1983.

Jackson, J .E. A User’s Guide to Principal Components; John Wiley & Sons: New
York 1991.

Mensink, R.P.; Katan, M.B. N. Engl. J. Med, 1990, 323,439.

. Judd, J .T.; Clevidence, B.A.; Muesing, R.A.; Wittes, J .; Sunkin, M.E.; Podczasy, J.J.

Am. J. Clin. Nutr., 1994, 59, 861.

Almondingen, K.; Jordal, 0.; Kierulf, P.; Sandstad, B.; Pedersen, J .I. J. Lipid Res,
1995, 36, 1370.

FDA/DHHS, 68 Federal Register, 41,434 (July 11, 2003). Food Labeling: Trans
Fatty Acids in Nutrition Labeling, Nutrient Content Claims, and Health Claims (Final
Ruling). URL http:// www.cfsan.fda.qov/~Ird_[fr0371 1a.html

Bailey, G.F; Horvat, R]. J. Am. Oil Chem. Soc. 1972, 49 (8), 494-498.

Sadeghi-Jorabchi, H.; Hendra, P.J.; Wilson, R.H.; Belton, P.S. J. Am. Oil Chem. Soc.
1990, 67 (8), 483-486.

Sadeghi-Jorabchi, H.; Wilson, R.H.; Belton, P.S.; Edwards-Webb, J .D.; Coxon, D.T.
Spectrochimica Acta. 1991, 47A (9/10), 1449-1458.

Baeten, V.; Hourant, P.; Morales, M.T.; Aparacio, R. J. Agric. Food Chem. 1998, 46,
2638-2646.

Johnson, G.L.; Machado, R.M.; Freidl, K.G.; Achenbach, M.L.; Clark, P.J.; Reidy,
S.K. Organic Process Research & Development. 2002, 6 (5), 637-644.

Weng, Y.M.; Weng, R.H.; Tzeng, C.Y.; Chen, W. Applied Spectroscopy. 2003, 57
(4), 413-418.

19

Chapter 2

MONITORING OF AN TI-SOLVENT ADDITION CRYSTALLIZATIONS WITH
RAMAN SPECTROSCOPY *

*Crystal Growth and Design, 2003, 3(6), 947-952.

2-1 Abstract

Fiber optic Raman spectroscopy was used for in situ monitoring of antisolvent addition
crystallization of cortisone acetate. Raman spectral features of the solvent, the
antisolvent, and the solute were monitored to measure the relative concentration of the
components. Different rates of antisolvent addition were monitored and the resulting
solid-state form was characterized using Raman spectroscopy and differential scanning
calorimetry (DSC). Raman spectroscopy was used to monitor dehydration during storage
of the product crystals. The current study demonstrates the advantages of using Raman
spectroscopy to monitor the solution phase during crystallization and simultaneously

monitoring the solid-state form.

2-2 Introduction

The solid-state form of a compound (polymorphs, hydrates, or solvates) must be
controlled to satisfy regulatory authorities as to the bioavailability of formulations of new
chemical entities]. One of the major contributory factors to bioavailability is the
solubility of a compound. Solubility can be largely dependent on the solid-state forrrr, as
different solid-state forms have different energies and therefore different solubilities.2’3
To control the solid-state form that will crystallize, it is necessary to be able to monitor

the solid-state form in situ during crystallization, and to be able to monitor the different

stages of the crystallization process itself. In the present work, Raman spectroscopy is

20

used to monitor the crystallization process in situ. In addition, in situ analysis of the
spectra obtained during the crystallization gives insight into the solid-state form that
crystallizes as it crystallizes. Finally, we present the use of Raman spectroscopy to

monitor the dehydration of crystal hydrates over time.

2-3 Experimental Section
2-3.1 Materials and Initrumentation

Solutions used in the crystallization experiments were prepared with deionized water
and analytical grade acetone (J. T. Baker). Cortisone acetate (CA) and bisnoraldehyde
(BA) were provided by Pharmacia Corporation (Kalamazoo, MI) and were used without
further purification. Recrystallizations were conducted in a 60-mL glass water-jacketed
reactor. The temperature of the water jacket was maintained at 22.4 i 0.6 0C using a
Brinkmann water circulator. Water addition rates and total water added volumes were
controlled with a 665 Dosimat (Metrohm Ltd.) water pump. A schematic diagram of the
experimental setup is shown in Figure 2.1. All crystallizations were performed using the
same size stir bar with the same stir rate. Raman spectra were collected using a HoloLab
Series 5000 Kaiser Optical Systems instrument. Details of the HoloLab 5000 have been
described previously."’5 Briefly, a 100 mW external cavity stabilized diode laser at 785
nm is used for sample illumination. A CCD camera, a spectrograph, and a fiber optic
probe comprise the rest of the HoloLab 5000 system. Remote sampling was
accomplished by employing a fiber optic probe attached to an immersion optic to focus
the incident beam into the solutions. The positioning of the immersion optic within the

crystallizer was maintained as constant as possible through out all the experiments. This

21

 

 

 

 

 

 

2.1. Schematic of the experimental setup. (1) Raman probe, (2) HoloLab 5000, (3) PC,
(4) water circulator, (5) jacketed reactor, (6) magnetic stir bar, (7) water reservoir and
pump, (8) digital thermometer.

22

ensured that any effects the immersion optic may have had on the crystallization process
impacted all the experiments in the same way. All spectra were a sum of five
accumulations collected over 5 seconds each, for a total spectra acquisition time of 25
seconds.
2-3.2 _C_alibration of Rgman Spectra

The Raman spectra of standards were used to construct a partial least-squares (PLS)
regression model using Quanth, a PLS regression analysis software package by ASI-
Mettler Toledo. The model consists of 30 standard solutions ranging in concentration
from 0.04 to 2.56 wt % of CA, from 40.00 to 99.00 wt % acetone and from 60.00 to 0.00
wt % water.
2-3.3 Anti-Solvent Addition Crystallizations

Binary solutions of acetone and CA were prepared (1 wt % CA in acetone) for the
starting point of the first set of crystallizations. For the starting point for the second set of
crystallizations, ternary solutions of acetone, water, and CA were prepared (87.00, 10.50,
and 2.50 wt %, respectively). The solutions were stirred in the glass-j acketed reactor for 2
hours to ensure dissolution of CA. Once this was achieved, water was added continuously
to the solutions. For both sets of crystallization experiments, three different water
addition rates were used, 0.25, 0.50, and 1.00 mL/min. During the continuous addition of
water, the crystallization was monitored by acquiring Raman spectra of the solution in 1-
min intervals.
2-3.4 Solid-State Analysis of Crvsta_l§

The resulting crystals were stored in scintillation vials in a dark cabinet at room

temperature (23 °C) and humidity. Solid-state Raman spectra were recorded in the

23

scintillation vials immediately after crystallization and each day thereafter. Images of the
crystals were obtained using an Olympus IMT-2 light microscope equipped with a 40X
magnification objective. The thermal properties of the different solid-state forms were
characterized with a differential scanning calorimeter, TA instrument DSC Q 100. The
temperature program used was heating at 10 °C/min to 180 C, heating from 180 to 230
°C at 5 °C/min, and heating from 230 to 250 °C at 2.5 °C/min.

2-3.5 Effect of Impurities

Another set of three crystallizations was performed. These crystallizations were all
conducted using a 0.50 mL/min water addition rate. BA was chosen as the impurity for
study due to its structural similarity to CA. BA was added during the solution preparation
step, prior to the beginning of the crystallization experiment. These solutions were
initially prepared with the same concentrations of acetone, CA, and water as the second
set of crystallizations. The amounts of BA added to each crystallization were as follows:
0.04, 0.15, 0.22 wt %.

2-3.6 Principal Components Analysis (PCA)

The Raman spectra obtained during the second set of crystallizations were subjected to
PCA using Matlab based HoloReact software provided by Kaiser Optical Systems. In
HoloReact, the PCA mode uses singular value decomposition (SVD) for the calculations.
The regions selected for analysis were 750-850 and 1600-1630 cm], and PCA was
applied to both regions simultaneously. Examination of the third factor obtained from

PCA was used to monitor the saturation point of the solutions.

24

2-4 Results and Discussion
24.] Calibration of Raman Spectra

Raman spectra of standard solutions were collected and subjected to a partial least
squares (PLS) calibration analysis. The peak region selected to correlate spectra and
solution concentrations was 1610-1680 cm”. This is the only region in the spectra of the
CA, acetone, water solution that contains accessible CA vibrational bands. The standard
error determined by a leave-one-out cross validation was 0.04 CA weight percent. The
two vibrational bands found in the 1610-1680 cm'1 region are explained by conjugation
theory.6 Figure 2.2 shows the molecular structure of CA. The conjugation between the
C=O at position 3 and the C=C at position 4 results in the delocalization of the 7:
electrons and reduces the double-bond character of both sets of double bonds. This causes
the Raman scattering to occur at lower wavenumbers than those of unconjugated systems.
Using this interpretation, the peak found between 1600 and 1625 cm'1 in the spectra is
assigned to C=C vibrations. The peak found between 1660 and 1680 cm-1 in the spectra
is assigned to C=O vibrations.
2-4.2 Solubility Curve

CA solubility in acetone/water solutions of varying compositions at 22.4 i 0.6 °C was
determined using the PLS model described above. Figure 2.3 shows the solubility curve
of CA determined via this procedure. Each point on the curve is the average of at least
three measurements. The error bars represent the standard deviation of three
measurements. Solid CA in excess of solubility was added to solutions initially

containing 40/60, 50/50, 60/40, 70/30, 80/20, 90/ 10 wt % acetone/water. Each solution

25

 

Cortisone Acetate
CZ3H3006
it
CH3 0
CH3 0
OH
H3C
O 4

 

 

 

Figure 2.2. Molecular structure of cortisone acetate (CA).

26

 

CA Weight Percent

 

 

O . r i r

35 50 65 80 95
Acetone Weight Percent

 

Figure 2.3. Solubility of cortisone acetate in binary mixtures of acetone/water at 22.4 °C.

27

was stirred for 24 h to ensure equilibrium was achieved. Raman spectra from the liquid
phase were collected, and the CA concentration was determined using the calibration.
The solubility of CA was found to be higher (2-3 times higher) in acetone/water solutions
of composition from 90/10 to 60/40 acetone/water compared to the solubility of CA in
acetone. The increase in solubility was an unexpected result since the literature value
reported for the solubility of CA in water is 2.20 x 10'3 wt %7 The reported solubility of
CA in acetone is 1.71 wt %;7 therefore, the expectation was that the addition of water to
acetone would decrease the solubility of CA. Although unexpected, an enhanced
solubility or maximum solubility effect8 has been observed for a variety of solid systems
dissolved in a binary mixed solvent.
2-4.3 Crystallizations

The CA and acetone concentrations decrease as water is added into the solution. As
more water is added into the solution, the solubility of CA is also decreased. Eventually,
the decrease in solubility due to water addition drives the CA concentration beyond
solubility. At this point, the solution can no longer sustain the level of supersaturation and
CA nucleates, resulting in a rapid decrease of CA concentration in solution. The last
period reflects the final slow growth of the crystals in a solution with a near-depleted
supersaturation. Figure 2.4 shows the solution phase peak area plotted versus time for the
first set of crystallizations, which depicts the decrease in concentration and the
crystallization events described. The low concentration of CA precluded the use of the
peak area of the solute to monitor the crystallization. It was necessary to use a solvent

band to follow the crystallization. The 787 cm'1 acetone band assigned to the C-C(=O)~C

28

 

    
 

_J\
I.

Peak Area 750-850 cm‘1

 

 

 

Time (minutes)

Figure 2.4. Real time change in the baseline peak area of the Raman spectrum between
750-850 cm-1 for the first set of crystallizations as described in the text. The symbols
represent different water addition rates: (A) 0.25, (1:1) 0.50, (O) 1.0 mL/min. Line A
shows the change in concentration as water is added to the solution, line B follows the
sharp decrease in peak area caused by turbidity in the solution which is caused by
nucleation. Line C follows the solution after the nucleation event.

29

symmetric stretch9 was used to monitor the crystallization event. It is important to note
that the rapid decrease in signal intensity (evident in Figure 2.4) that occurs immediately
after crystallization begins is not solely due to a decrease in concentration of the species
being monitored. The growing crystals will reflect and refract the excitation and Raman
scatter; this consequently decreases the amount of Raman scatter that reaches the
collection optics and therefore the detector. In effect, the Raman spectrometer is acting as
a turbidity meter. This is particularly useful when (as above) monitoring the
crystallization of a compound present at low concentrations. It is possible to monitor a
solvent band with the knowledge that the crystal growth will be evident even though the
compound being crystallized is not being directly monitored. As expected, the slope of
the line marked A (when normalized to 1) in Figure 4 matches the slope of the line of the
calculated concentration change for acetone for all three curves. To monitor the peak
intensity of CA accurately, a second set of crystallizations was performed at a higher
starting CA concentration. This second set of experiments was made possible because of
the a priori knowledge from the solubility curve that CA displays increased solubility in a
range of binary mixtures of acetone/water.

Figure 2.5 shows the peak area of the 787 cm'1 acetone band plotted versus time for the
second set of crystallization experiments. Again as expected, the slope of the line marked
A (when normalized to 1) in Figure 2.5 matches the slope of the line of the calculated
concentration change for acetone for all three curves.

Due to the increased concentration of CA in the second set of experiments, it was
possible to use a solute band to monitor the crystallization event. The CA 1600-1625 cm'1

band assigned to C=C vibrations was used. Figure 2.6 shows the peak area of the band

30

 

 

 

 

Peak Area 750-850 cm'1

 

 

 

Time (minutes)

Figure 2.5. Real time change in the baseline peak area of the Raman spectrum of
cortisone acetate between 750-850 cm-1 for the second set of crystallization experiments
as described in the text. The symbols represent different water addition rates: (A) 0.25,
(1:1) 0.50, (O) 1.0 mL/min. Line A Shows the change in concentration as water is added to
the solution, line B follows the sharp decrease in peak area caused by turbidity in the
solution which is caused by nucleation. Line C follows the solution after the nucleation
event.

31

 

Area 1600-1640 cm-1

 

 

 

 

O 1 5 30 45 60
Time (minutes)

Figure 2.6. Real time change in the baseline peak area for the Raman spectrum of
cortisone acetate between 1600-1640 cm-1 for the second set of crystallization
experiments. The different symbols represent different water addition rates: (A) 0.25, (:1)
0.50, (O) 1.0 mL/min. Line A shows the change in concentration as water is added to the
solution, line B follows the sharp decrease in peak area caused by turbidity in the solution
which is caused by nucleation. Line C follows the solution after the nucleation event.

32

versus time for the second set of experiments. The data for the CA concentration are in
qualitative agreement with those obtained for acetone.

It is important to note that no calibration model has been applied to the data presented
in Figures 2.4-2.6. The peak area profiles come directly from the raw data. With a
reliable instrument that exhibits negligible signal drift from day to day (or more likely by
avoiding signal drift through peak area ratio methods or daily calibrations), the idea of
using the raw peak area data vastly simplifies the process of monitoring the concentration
changes in situ. The simplification comes fi'om the fact that it is possible to monitor the
relative concentrations of components during the crystallization without having to
perform the crystallization and then subsequently subject the data to the calibration
model. Certain methods10 have claimed to monitor crystallizations in situ, even though
the data must first be collected and then run through a calibration model. In that scenario,
there is no possibility for changing experimental conditions in response to a perturbation
to the crystallization.

Monitoring the crystallization process has been previously demonstrated using ATR-
FTIRll and dynamic light scattering12 among other techniques. The drawback to these
techniques is that it is not possible to obtain information regarding the structure of the
crystals during the crystallization. With Raman spectroscopy, the spectra of the solution
and then of the slurry can be monitored for changes in the solid-state form that
crystallizes.

Figure 2.7, trace 1, shows the Raman spectrum of the as received CA crystals used to

prepare the solutions. Trace 2 shows the spectrum of CA in solution, trace 3 shows the

33

 

 

1615 1671
Q 2
g 1669
*5 1616

 

1650

  

Raman Shift (cm")

Figure 2.7. Comparison of cortisone acetate Raman spectra for different environments in
the region between 1600-1680 cm". Spectrum (1): as received crystals. Spectrum (2):
solution of CA, acetone and water. Spectrum (3): CA crystals in slurry. Spectrum (4):
crystals after drying.

34

spectrum of the CA in slurry after crystallization, and trace 4 shows the dried product
crystals after crystallization. The 1600-1680 cm'1 region is useful to track the changes the
crystals underwent from the solution state through to the product crystals. From Figure
2.7, it can be seen that the original crystals are not the same as the product crystals.
Perhaps more importantly, the spectrum of the crystals in the slurry shows that the solid-
state form that has crystallized is not the same as the original crystals. Even before the
crystals are vacuum filtered and dried, it is possible to know which solid-state form has
crystallized. The product crystals turn are a hydrate of CA (see solid state analysis below
for explanation). Both sets of crystallizations, with different concentrations of CA,
yielded the same CA hydrate crystals.
2-4.4 Solid-State Analysis

Generally, as the rate of supersaturation increases (increasing rate of water addition),
the induction time decreases. Accordingly, higher rates of supersaturation exhibit larger
numbers of small crystals or the formation of precipitate. Figure 2.8 shows
photomicrographs of the product crystals resulting from the three different water addition
rates of the second set of crystallizations (the results were the same for the first set). It is
evident from the different sizes of the crystals that, as expected, the highest rate of water
addition yielded the smallest crystals. The Raman spectra change with the passage of
time indicating a transformation occurs. Again, the region of focus is between 1600 and
1680 cm”. Immediately after crystallization and for some time after, there are two large
peaks that appear at 1606 and 1650 cm”1 in the spectrum of the product crystals. As time
passes, the 1606 peak develops a shoulder. The peak at 1606 cm'1 then becomes a doublet

at 1605 and 1613 cm”. The 1650 cm'1 peak also develops a shoulder and in

35

 

 

 

 

 

 

 

Figure 2.8. Cortisone acetate crystals resulting from the three different rates of water
addition used. These photographs are at the same magnification and encompass identical
viewing areas. The pictures represent the different water addition rates. (A) 0.25, (B)
0.50, (C) 1.00 mL/min.

36

addition shifts toward higher wavenumbers. The 1650 cm“1 peak then becomes a peak at
1651 cm'1 with a small shoulder at 1669 cm”. Figure 2.9 shows the Raman spectra that
illustrate the transformation described above. To further investigate this phenomenon, a
DSC experiment was conducted on crystals that had yet to undergo the transformation.
Figure 2.10A shows the themrodiagram of these crystals. It is evident from the
endotherm centered at 100 °C that there is water present in the crystals. The endotherm
along with the transformation observed in the Raman spectrum indicates that the product
crystals are a hydrate of CA, and as time goes on the water is lost from the crystals due to
phase transformation.

To conclusively prove the crystals are CA hydrates, DSC experiment was also
conducted on the same crystals after the transformation was observed in the Raman
spectrum. Figure 2.10B shows that there is no endotherm near 100 °C after the
transformation was observed in the Raman spectrum. This proves that the product
crystals are hydrates, and as time passes the water is slowly lost from the crystals. The
formation of CA hydrates seen here agrees with past studies,13 where it has been reported
that CA will form a hydrate when crystallized from acetone if the solvent contains more
than 10 wt % water. This previous study did not analyze the crystals over time to
investigate the dehydration of the crystals. The Raman approach of the present study
uniquely allows tracking of the crystal phase during the various steps of the

crystallization process.

37

.......

...... . . Days After

 

 

 

.é‘ Crvstallization
3
s
w I.
>
'3
2
a .......
n:
l
1550 1700
Raman Shift (cm'l)

Figure 2.9. Time resolved Raman spectra of cortisone acetate acquired daily to monitor
post crystallization phase transformation. The first spectrum in was acquired the day of
crystallization (day 0 after crystallization). The final spectrum was acquired 8 days after
crystallization.

38

 

 

 

 

 

 

 

 

A ,flflv /‘:N‘ f
. x m ,1
.. N :
.2 l
'. I
‘l
E, 1 96 c =1
“- ll
*1 1
as
a r...
3 MW 1
...... r" * i
,MWM-‘J ‘! i
"" 1‘ I:
'1 l
l Endotherm if
1 00 1 75 250

Temperature (°C)
Figure 2.10. The DSC thermograrns of (A) CA hydrate and (B) same crystals after loss

of water. Apparent endotherms or exotherrns at around 180 C are due to change in the
temperature program heating rate.

39

2-4.5 Effect of Impurity

Impurities can serve as tailor-made additives that alter the solid-state form produced
from particular crystallization conditions.” The impurity chosen for the current study was
BA. Figure 2.11 shows the molecular structure of BA. This molecule meets the criteria,
which have been discussed extensively”16 for tailor-made additives. That is, BA is
structurally similar enough to CA to be incorporated into the CA crystal lattice, but BA
also has differences in its structure that should disrupt the growth of the CA crystal
lattice.

The Raman spectra obtained for the crystallizations where BA was present were
identical to the CA hydrate spectra that were obtained for all the other crystallizations
without BA. The addition of BA appeared to have no effect on the solid-state form that
crystallized (under the conditions used). Even though BA failed to provide a different
solid-state form, the dried crystals were monitored daily. During the monitoring of the
crystals, it became evident that the crystals with BA added to them took considerably
longer to dehydrate. Figure 2.12 shows the peak width in the region between 1600 and
1640 cm'1 versus the number of days after crystallization for three crystallizations
performed at 0.50 mL/min without BA.

The crystallization was performed in triplicate to ensure that the dehydration time used
for comparison was statistically meaningful. From Figure 2.12, it can be seen that all
three crystallizations required 3 days for the dehydration to begin. To complete the
dehydration, the crystallizations required 5.33 (1.53 days. Figure 2.13 shows the peak
width versus the number of days after crystallization for the solutions with BA. The three

crystallizations were conducted at a water addition rate of 0.50 mL/min with increasing

40

 

 

Bisnoraldehyde
C22H3202

 

 

 

Figure 2.11. Molecular structure of bisnoraldehyde

41

 

 

 

 

17*‘ xkgamageeuwo
e :1
O
[o X
8
‘714 o
ID
no
3
5
E
311 a
x.
111
o
Q
5119!!
8' ' ‘ ' 1
0 4 8 12 16

Time (days)

Figure 2.12. Peak width of the Raman spectrum of cortisone acetate between 1585-1635
cm'1 for three crystallizations conducted with a water addition rate of 0.50 mL/min. All
three crystallizations required three days for dehydration to begin which is reflected by
the increase of peak width. Dehydration required 5.33 i 1.53 days for completion. .

42

 

 

 

 

17~ Aééoooo
A
é o .
0
‘0 i
”14,. E
:2 A l
u':
66
$3
'5
211“
g o E]
1:690 13:10
iiEEBOO D
8 ' 1 1
0 3 6 9 12 15

Time (days)

Figure 2.13. Peak width of the Raman spectrum of cortisone acetate between 1585-1635
cm'1 for three crystallizations in the presence of bisnoraldehyde. The crystallization at
0.04 weight percent BA in solution(A) shows no statistically meaningful difference from
the reference crystallizations. The crystallization with 0.15 weight percent BA in
solution (0) required seven days for dehydration to begin and 12 days for dehydration to
be completed. Finally, the crystallization with 0.22 weight percent BA in solution (El)
had not begun dehydration after 12 days.

43

amounts of BA added. The crystallization with the smallest amount of BA shows no
statistically meaningfirl difference from the crystallization with no BA added. The two
other samples with higher concentrations of BA demonstrated a significant increase in the
number of days required for dehydration. The sample from second highest BA containing
solution required 6 days for dehydration to begin and 9 days for the dehydration to be
completed. The sample from the solution with the most BA began to dehydrate 12 days
after crystallization occurred, which is approximately a five to six times increase in the
number of days required for dehydration when compared to the “reference” experiments.
The morphology of the crystals did not change and the Raman spectrum looks identical to
the Raman spectrum of a sample without BA.

2-4.6 Principgfi Components Analysis QCA)

The goal of this analysis was to investigate the possibility of identifying and isolating
subtle features that are part of the Raman spectra that are susceptible to changes in
solution structure, for utilization in measurement of precrystallization phenomena. The
arrangement or rearrangement of solute molecules in solution (caused by supersaturation
or nucleation) should be manifested in Raman spectral features such as shifts in peak
positions and changes in peak areas and or peak widths. These variations in spectral
features may not be readily apparent in the raw spectra. Techniques such as spectral

deconvolutionn’18

and derivative spectroscopy” have been used in the past to investigate
the occurrence of these difficult to track changes in the spectra of crystallization
experiments. PCA is a powerful mathematical technique that has not been used in the

past to investigate crystallization experiments. PCA of the spectral data for the second set

of experiments using the HoloReact software yielded 25 different factors. Using a percent

44

variance plot, it was determined that only the first three factors would have to be
examined closely since they described 99.95% of the variance in the data. The first and
second factor provided information that was readily apparent from the raw spectra. The
third factor, however, yielded an interesting result. Figure 2.14 shows the third factor
plotted versus time for the second set of crystallization experiments. The plots Show an
increase up to an apex and then a steady decrease. Using the mass balance from the
crystallization experiments and the solubility curve presented above, it was possible to
determine that the apex in all three plots corresponds with the point at which the solution
becomes saturated. Therefore, using this technique, it is possible to determine the onset of
supersaturation in the crystallization experiment. A full explanation of this phenomenon
would involve a thorough discussion of PCA itself. That is beyond the scope of this
work, so the interested reader is referred to a number of texts available.”’20 A simplistic
molecular level explanation of the trend in the data displayed from the third factor is that
this factor is correlated to the interactions between solvent and solute during the water
addition. As the solution becomes saturated and subsequently supersaturated, the changes
in this interaction that take place are correlated to the variance displayed in the raw
spectra. The purpose of this work is not to conclude that this effect is general to
antisolvent crystallization or even to corticosteroid antisolvent crystallization. It is only
intended to show that PCA is a powerfirl technique that may provide useful information
about the crystallization phenomena beyond that which is obvious from the overall

spectra.

45

 

 

.33
re

 

c':
or
n l [‘1

 

 

Time (minutes)

Figure 2.14. The third factor obtained fiom PCA of the Raman spectra of cortisone
acetate in aqueous acetone solutions for the regions of 750-850 cm'1 and 1600-1630 cm'1
for the second set of crystallizations described in the text. The different symbols
represent different water addition rates: (0) 0.25, (1:1) 0.50, (A) 1.0 mL/min. The solution
becomes saturated at the first inflection point in time

46

2-5 Conclusions

The use of Raman spectroscopy was demonstrated for simultaneous monitoring of the
crystallization process and of the solid-state form crystallizing during antisolvent addition
crystallization. Raman spectroscopy is able to provide accurate information regarding not
only the crystallization process such as ATR-FTIR and other methods, but it is also
capable of providing information on the solid-state form during crystallization and
subsequent unit operations such as drying. Finally, Raman spectroscopy was also used to
monitor the dehydration of crystals during an extended storage time including the effect
of an impurity on the transformation rate. These results provide further evidence that
Raman spectroscopy can be very useful in rapid screening of different solid-state forms

during processing or storage.

2-6 Acknowledgments
Kaiser Optical Systems, Inc. provided the Raman spectrometer and HoloReact software
and Pharmacia Corporation provided the steroids used in this study. Their contributions

are appreciated. Partial support of J .A.F . was received fi'om a GAANN fellowship.

47

2-7 References
1. Threlfall, T. Analyst 1995, 120, 2435-2460.

2. Polymorphism in Pharmaceutical Solids; Brittain, H. G., Ed.; Marcel Dekker: New
York, 1999.

3. Bym, D. E.; Pfieffer, R. R.; Stowell, J. G. Solid-State Chemistry of Drugs, 2nd ed.;
SCCI, Inc.: West Lafayette, 1999.

b

. Battey, D. E.; Slater, J. B.; Wludyka, R. 8.; Owen, H.; Pallister, D. M.; Morris, M. D.
Appl. Spectrosc. 1993, 47, 1913.

5. Owen, H.; Battey, D. E.; Pelletier, M. J .; Slarter, J. B. Proc. SPIE 1995, 24, 260.

6. Silverstein, R. M.; Bassler, C. G.; Morril, T. C. Spectrometric Identification of
Organic Compounds, 5th ed.; John Wiley & Sons: New York, 1991.

7. The Merck Index, 8th ed.; Stecher, P. G., Windholz, M., Leahy, D. 8., Bolton, D. M.,
Eaton, L. G., Eds.; Merck & Co., Inc: New Jersey, 1968.

8. Prausnitz, J. M.; Lichtenthaler, R. N.; Azevedo, R. G. Molecular Thermodynamics of
F laid-Phase Equilibria, 3"d ed.; Prentice Hall PTR: New Jersey, 1999.

9. Dollish, F. R.; Fateley, W. G.; Bentley, F. F. Characteristic Raman Frequencies of
Organic Compounds; Wiley-lnterscience: New York, 1974.

10. Schwartz, A. M.; Berglund, K. A. J. Cryst. Growth 1999, 203, 599.
11. Feng, L.; Berglund, K. A.; Cryst. Growth Des. 2002, 2, 449.

12. Ansari, R. R.; Suh, K. 1.; Arabshahi, A.; Wilson, W. W.; Bray, T. L.; DeLucas, L. J.
J. Cryst. Growth 1996, 168, 216-226.

13. Kuhnert-Brandstatter, M.; Gasser, P. Microchem. J. 1971, 16, 590.

14. Weissbuch, I.; Leiserowitz, L.; Lahav, M. In Advances in Crystal Growth Research;
Sato, K., Nakajima, Y., Furukawa, Eds.; Elsevier Science B.V.: Amsterdam, 2001;
Chapter 23.

15. Kuzmanko, 1.; Rapaport, H.; Kjaer, K.; AIS-Nielsen, J .; Weissbuch, 1.; Lahav, M.;
Leiserowitz, L. Chem. Rev. 2001, 114, 1659-1696.

16. Weissbuch, 1.; Lahav, M.; Leiserowitz, L. Adv. Cryst. Growth Res. 2001, 381-400.

17. Dunuwila, D. D.; Berglund, K. A. Trans. 1. ChemE 1995.

48

18. Cerreta, M. K.; Berglund, K. A. J. Cryst. Growth 1987, 84, 577-588.

19. Malinowski, E. M. Factor Analysis in Chemistry; John Wiley & Sons: New York,
1991.

20. Beebe, K. R.; Pell, R. J.; Seasholtz, M. B. Chemometrics: APractical Guide; John
Wiley & Sons: New York, 1998.

49

Chapter 3
IN SI T U MONITORING OF ANTI-SOLVENT ADDITION
CRYSTALLIZATIONS WITH PRINCIPAL COMPONENTS ANALYSIS OF
RAMAN SPECTRA *
*Crystal Growth and Design, 2004, 4(3), 457-463.

3-1 Abstract

Fiber optic Raman spectroscopy was used for in situ monitoring of anti-solvent addition
crystallization of progesterone. Raman spectral features of the solute and the anti-solvent
were subjected to principal components analysis (PCA) in order to attempt to obtain
information which is not readily apparent from the raw spectral data. For the system
utilized, PCA was able to distinguish spectral features from the solute in solution, the
solute crystals and the anti-solvent. Furthermore, PCA was capable of detecting subtle
changes in the spectral data (as the addition of anti-solvent progressed) that could be used
as a warning for the onset of crystallization. The current study demonstrates the

advantages that can be gained by combining PCA with Raman spectroscopy for

monitoring crystallizations in situ.

3-2 Introduction

Monitoring and control of the crystallization of active pharmaceutical ingredients (API)
are important to the pharmaceutical industry. This importance is due in large part to the
pharmaceutical industry’s need to satisfy regulatory authorities as to the morphology,
polymorphism, crystal size and other properties of new chemical entities'.
Pharmaceutical companies have also recognized that on-line or in situ monitoring of

crystallizations will greatly aid in the reproducibility of results and in detecting problems

50

that may arise during crystallization. Consequently, in situ monitoring of crystallizations
will reduce money and time spent in reworking “failed” crystallizations.

The use of Raman spectroscopy to monitor crystallization processes in situ offers many
advantages over using turbidity or nephelometry measurementsz. Even with the
advantages afforded, Raman spectroscopy has yet to meet the expectations placed on it
regarding crystallization monitoring3. The formation of solution structures (caused by
supersaturation or nucleation) should be noticeable in the Raman spectrum. Peaks arising
from the solution structures should provide information regarding the intermolecular
interactions (e.g. hydrogen bonding) that are present in the solid-state phase3 and could
help to predict the solid-state form that ultimately crystallizes. The appearance of bands
related to these kinds of interactions could also be used as a signal of the irnminence of
crystallization.

The foregoing discussion is based on a solid theoretical foundation; however, to the
authors’ knowledge there have been few results reported in this area. Some studies have
reported the detection of solution structures” in supersaturated solutions, but the solution
structures were found to be unrelated to the crystal structure. A possible explanation for
the lack of results in this area could be the difficulty in interpreting the spectral data
obtained. The kinds of solution structure interactions described above are expected to
have varying effects on the Raman spectrum of a sample. For example, the interactions
could cause: new peaks to appear, changes in the intensity of existing peaks, peak shifts,
changes in peak width, variations in the baseline of the spectrum and probably a
combination of the listed changes. It is also likely that any bands related to the solution

structure interactions would overlap with existing solvent or solution bands, thus masking

51

their appearance. Another factor that makes data interpretation difficult is the
concentration of the crystallizing species in solution. Many APIs are not readily soluble
in organic solvents and are often present in relatively low concentrations compared to the
solvent. The low concentration of the API in solution combined with the strong Raman
scattering properties of many organic solvents often leads to small, broad bands from the
crystallizing species. A final consideration for the lack of results is that crystallizations
are dynamic experiments with at least one condition varying at all times. Two of the
most common crystallization techniques are cooling and anti-solvent addition.
Temperature changes used in cooling crystallizations and the introduction of another
compound to a solution (anti-solvent addition crystallization) are both known to impact
the Raman spectrum (strongly dependent on the sample). With these variations factored
in, it becomes even more difficult to single out any changes in the Raman spectrum that
are due solely to any solution structure phenomena.

For these reasons, multivariate analysis, and in particular principal components analysis
(PCA) seems well suited for processing spectral data for anti-solvent addition
crystallization experiments. PCA is purely a mathematical technique whose goal it is to

reduce the dimensionality of a data matrix.

3-3 Experimental Section
3-3.1 Materials and Instrumentation

Solutions used in the crystallization experiments were prepared with de-ionized water
and analytical grade acetone (J .T.Baker, Phillipsburg, NJ). Progesterone (PG) was
provided by Pharmacia Corporation (Kalamazoo, MI) and was used without firrther

purification. Crystallizations were conducted in a 60 mL glass water-jacketed reactor.

52

The temperature of the water jacket was maintained at 22.4 i 2.00C using an RTE-140
bath/circulator with microprocessor controller (NESLAB, Postrnouth, NH). Water
addition rates and total water added volumes were controlled with a 665 Dosimat
(Metrohm Ltd., Switzerland) water pump. A schematic diagram of the experimental
setup is shown in Figure 3.1. All crystallizations were performed using the same size stir
bar with the same stir rate.

Raman spectra were collected using a HoloLab Series 5000 instrument provided by
Kaiser Optical Systems, Inc. Details of the HoloLab 5000 have been described
previously”. Briefly, a 100 mW external cavity stabilized diode laser at 785 nm is used
for sample illumination. A CCD camera, a spectrograph, and a fiber optic probe
complete the HoloLab 5000 system. Remote sampling was accomplished by employing
a fiber optic probe attached to an immersion optic to focus the incident beam into the
solutions. The positioning of the immersion optic within the crystallizer was maintained
constant through out all the experiments. This positioning ensured that any effects the
immersion optic may have had on the crystallization impacted all the experiments
equally.

3-3.2 glibration of Raman Spectra

The Raman spectra of standards were used to construct a multiple linear regression
(MLR) model using the MATLAB (The MathWorks, Inc.) based HoloReactTM software
provided by Kaiser Optical Systems, Inc. The model consists of eleven standard
solutions ranging in concentration from 1.01 to 8.22 weight percent of PG, from 56.96 to

88.45 weight percent acetone and from 10.03 to 39.99 weight percent water.

53

 

 

 

 

 

Figure 3.1. Schematic of the experimental setup. ( 1) Raman probe, (2) HoloLab 5000,
(3) PC, (4) water circulator, (5) jacketed reactor, (6) magnetic stir bar, (7) water reservoir
and pump, (8) digital thermometer.

54

3-3.3 Anti-Solvent Addition Crystallizations

For all crystallization experiments, binary solutions of acetone and PG were prepared
as the starting point. The solutions were stirred in the glass-j acketed reactor for 2 hours
to ensure dissolution of PG. Once this was achieved, water was added continuously to
the solutions. For all sets of experiments three different water addition rates were used,
0.25, 0.50 and 1.00 mL/min (except for the set of crystallizations used for reproducibility
[see (a) below], where the water addition rate was 0.5 mL/min). For all experiments,
10.3 mL of acetone comprised the initial solution and 12 mL was the total volume of
water added. During the continuous addition of water the crystallization was monitored
by continuously acquiring Raman spectra of the solution. The spectra were acquired
using 10 second acquisition time and 10 second intervals between spectra.

3-3.4 Reproducibility Crystallizations (a)

A replicate set of three crystallizations initially contained 7.0wt% PG. These
crystallizations were performed in order to demonstrate the reproducibility of the
experimental results.

3-3.5 m Experiments (b)

For this set of three experiments water was added to neat acetone. The results were
used as “blanks” for comparison purposes.

3-3.6 V_arying Initial PG Concentration Crystallizations (c)

Three sets of three crystallizations were performed. The first set of three contained an
initial PG concentration of 3.5wt %, the second set contained 7.0wt % PG, and the third

set contained 14.0wt % PG.

55

3-3.7 Principal Components Analysis (PCA)

The Raman spectra obtained were subjected to PCA using two different software
packages, HoloReactTM and PLS_toolbox (Eigenvector Research, Inc.). HoloreactTM and
PLS_toolbox are MATLAB based software and both use singular value decomposition
(SVD) for the PCA calculations. In order to perform the SVD calculations the software
packages use the svd command found in MATLAB. MATLAB’s svd command8 uses a
version of the LINPACK9 routine ZSVDC for the SVD calculations. HoloReactTM was
used to collect the spectral data and analyze it in situ; PLS_toolbox was used for all post
experiment data analysis.

All the spectral data were mean centered before being subjected to PCA, this was the
only data pre-processing performed. The region selected for analysis was 1594-1688 cm’
1. Examination of the scores and loadings for the most important principal components

(PCs), as determined from percent variance plots, were used to investigate changes in the

spectral features of the Raman data.

3-4 Results and Discussion
3-4.1 Calibration of Rm Spectra

Raman spectra of standard solutions were collected and subjected to the MLR
calibration analysis. The peak regions selected to correlate spectra and solution
concentrations were 750-850 cm'1 and 1600-1630 cm”. The standard error of prediction
(SEP) determined by a leave-one-out cross validation was 0.11 PG weight percent and
the standard error of calibration (SEC) determined by a leave-one-out cross validation

was 0.06 PG weight percent.

56

The 750-850 cm'1 region was chosen because it contains the 787 cm’1 acetone band
assigned to the C-C(=O)-C symmetric stretch”). This band provides the model with the
spectral changes caused by the various concentrations of acetone and water present in the
standards. The vibrational band found in the 1600-1630 cm'1 region provides the model
with the spectral changes caused by the different concentrations of PG used in the
standards. The presence of this band in the spectrum can be explained by conjugation
theory“. Figure 3.2 shows the molecular structure of PG. The conjugation between the
C=O at position 3 and the C=C at position 4 results in the delocalization of the it
electrons and reduces the double-bond character of both sets of double bonds. This
causes the Raman scattering to occur at lower wavenumbers than those of un-conjugated
systems. Using this interpretation, the peak found between 1600-1630 cm"1 in the spectra
is assigned to C=C vibrations.

3-4.2 Solubility Curve

PG solubility in acetone/water solutions of varying compositions at 22.4 i 0.5 0C was
determined using the MLR model described above. Figure 3.3 shows the solubility curve
of PG determined via this procedure. Each point on the curve is the average of at least
three measurements. The error bars represent the standard deviation of the three
measurements.

Solid PG in excess of solubility was added to solutions initially containing 4985/5015,
5875/4125, 7346/2654, 87.81/ 13.19, 89.99/ 10.00 weight percent acetone/water. Each
solution was stirred for twenty-four hours to ensure equilibrium was achieved. Raman

spectra from the liquid phase were collected and the PG concentration was determined

57

 

 

Progesterone
Cz1H3002

 

 

 

Figure 3.2. Molecular structure of progesterone (PG).

58

18-

 

.3
N

PG Weight Percent
05

 

 

 

40 60 80 1 00
Acetone Weight Percent

Figure 3.3. Solubility of progesterone in binary mixtures of acetone/water at 22.4 °C.

59

using the MLR calibration. The solubility of PG was found to be within the expected
limits found in the literaturelz’13 for PG dissolved in neat water and neat acetone. The
solubility of PG was used to determine the concentrations that could be used in the anti-
solvent crystallization experiments. Also, the solubility data were used to monitor the
progress of the crystallizations.
3-4.3 Anti-Solvent Addition Crystallizatiog

The PG and acetone concentrations decrease as water is added to the solution. As more
water is added into the solutions the solubility of PG decreases. Eventually, the water
addition drives the PG concentration beyond solubility. Continued water addition beyond
this point ensures that eventually the solution will no longer be able sustain the level of
supersaturation and PG nucleates. The last phase is the growth of the crystals in a
solution with near-depleted supersaturation. As described previouslyz, it is possible to
monitor this process in situ using Raman spectroscopy. Monitoring the crystallization
process in this way provides invaluable information. However, the raw Raman spectral
data without being subjected to a quantitative calibration does not provide a key piece of
information: How close is the solution to the nucleation and/or crystallization point?
Advanced warning regarding the impending point of nucleation and crystal growth would
be useful in a variety of fields. A simple example of the usefirlness of this would be in
monitoring the reproducibility of the crystallizations and reproducibility of the crystal
physical properties. Another example would be the possibility of altering the crystal size
distribution by stopping the addition of anti-solvent at a certain time before crystallization

occurs.

60

3-4.4 Principal Components Analysis (PCA)

The goal of this analysis was to investigate the possibility of identifying and isolating
subtle features of the Raman spectrum that are susceptible to changes in solution
structure for utilization in measurement of pre-crystallization phenomena. The
arrangement or rearrangement of solute and solvent molecules (caused by supersaturation
or nucleation) should be manifested in Raman spectral features such as shifts in peak
positions and changes in peak areas and/or peak widths. These variations may not be
readily apparent in the raw spectra. The system chosen for this work represents a good
example, not only for subtle changes in the spectrum, but also for the difficulty in
interpreting changes that occur to the spectra during crystallization. Figure 3.4 shows the
Raman spectral data for a typical experiment. It can be seen from Figure 3.4 that there
are at least three different changes occurring as water is added. The signal intensity, the
peak position and the baseline of the spectra are changing as water is added to the
solution. In the past, techniques such as spectral deconvolution4 and derivative
spectroscopy5 have been used to investigate the occurrence of these difficult to track
changes in the spectra.

PCA is a powerful mathematical technique that has not been used in the past to
investigate crystallization experiments. Two matrices are obtained from the analysis.
The loadings matrix shows the relationship of the variables to each other and to the
principal components. The scores show the similarities and differences between the

samples based on the principal components.

61

 

No water ded

lntensi

 

 

 

 

1640 1530

Raman Shift (cm-1)

Figure 3.4. Raman spectra obtained for a PG crystallization experiment with an initial
PG concentration of 7.0wt %. The region enclosed by the dashed lines is the region
being investigated with PCA.

62

015 PC1 (82.08%)

A...—-\/\

1590 1620 1650 1680

 

 

 

Loading

 

 

 

       
 

 

 

-0.1
PC2 (15.65%)

0.15

D,

.E

'U ' 1

3 15 1620 1650
_l

 

 

PC3 (2.06%)

 

 

    

 

 

 

-0.1u *
Raman Shift 1cm")

Figure 3.5. Principal component loadings of Raman spectra obtained from an initial PG
concentration of 7.0wt %. (PC 1) is the first principal component, (PC 2) is the second
principal component, (PC 3) is the third principal component.

63

3-4.5 PCA of Reproducibility Crystallizations

Three crystallizations were performed in order to demonstrate the reproducibility of the
results. The data shown in this section are presented for comparative purposes only and
are not discussed in depth with relation to the interpretation of changes relating to the
crystallization experiments. Figure 3.5 shows three major principal component (PC)
Figure 3.5 is presented in order to illustrate how the changes in the Raman spectra
translate into loading plots of the PCA.

The three data sets obtained for the reproducibility analysis were used to construct one
PCA model. Figure 3.6 displays the graph of PCI scores versus PC2 scores for this
model. Without performing the standard deviation calculations it is evident that the
scores of the values are similar for the three data sets. This similarity between the scores
values for the three data sets indicates that experimental results are reproducible when
comparing crystallizations with the same initial concentrations and the same rate of water
addition. It is important to note here that the present work does not use the values of the
scores in order to monitor the progress of the crystallizations. Only the “trend” that the
scores plots display (as seen in Figure 3.6) is used to monitor the crystallization process.
3-4.6 PCA of Blank Experiments

“Blank” experiments were performed in order to simplify data analysis. Inspection of
the data from the blank experiments was used in order to investigate the effects of
continuous water addition on the PCA calculations and results. These experiments also
ensured that acetone-water interactions would not be mistaken for crystallization
phenomena. Figure 3.7 shows the Raman spectra collected for one blank experiment

with a water addition rate of 0.5 mL/min. Similar to Figure 3.4, Figure 3.7 displays the

i D
A o
. .1
"‘ “— A o 1" r
.20 E+04 0 .
N
2 l 31.2,;
a--- __—1.oe+04 ...% z,___A -1“ ,_ . 2p
E C
- 6
l ‘i
.20 BM 1
P‘c1

 

1.0 E+O4

 

 

 

 

 

 

Figure 3.6. Scores plots for the three reproducibility crystallization data sets. The scores
were calculated by constructing one PCA model using the three data sets. The similarity

between the scores values indicates the reproducibility of the experiment.

indicates direction of increasing water concentration.

65

Arrow

 

Intensity

 

 

 

 

1560 1600 1640 1680
Raman Shift (cm'l)

Figure 3.7. Raman spectra obtained for a "blank" experiment (no PG present). The

region enclosed by the dashed lines is the region being investigated with PCA. Arrows
highlight how the spectra change as water is added.

66

changes that occur to the region of interest in the Raman spectra as water is added to the
initial solution. Figure 3.8 shows the first two PC loadings for the data shown in Figure
3.7, the two loadings account for 99.78% of the variance in the data. The loading plot for
PC 1 in Figure 3.8 appears to have two major features. The first feature is the broad band
that begins at 1594 cm’1 and ends at approximately 1665 cm”. In the region between
1594 and 1665 cm"1 Figure 3.7 shows that there is an increase in signal intensity as water
is added. The increase in intensity can be explained by considering that even though
water is known to be a weak Raman scatterer, it still has a Raman spectrum. One of the
features of the water Raman spectrum is a broad band near 1645 cm'1 that arises from the
H-O-H bending modes of the molecules”. As water is added, almost every region in the
spectrum should display decreases in signal intensity (since most of the visible bands are
due to acetone). However, the regions of the spectrum that coincide with the Raman
water bands should increase in signal intensity.

The second major feature seen in the Figure 3.8 PC 1 loading plot is the large increase
along the PC 1 axis in the region between 1670 and 1688 cm”. Figure 3.7 shows a peak
shifting into this region. The peak that is shifting is due to C=O stretching of acetone
molecules”). With no water added the peak is centered on 1706 cm], after 12 ml of water
are added the peak is centered on 1695 cm". As the peak shifts the signal intensity
changes for the Raman bands in the area of interest, this is detected by PCA and
displayed in PC 1 on Figure 3.8. Using the preceding discussion, it can be concluded that
PC 1 is affected mostly by the changes in signal intensity caused by water addition.

The PC 2 loading plot in Figure 3.8 displays similar features to those seen in the PC 1

loading plot. The PC 2 loading plot demonstrates that PC 2 is influenced by the change

67

 

 

 

 

 

O 3PC1 (98.81%) 02IPCZ (0.97%)
0.2 166
3' 0.1

0.1 E

1655 1:
.3 got - ,
‘3? 0 1590 1640 690
.J 1590 1640 1690 l
01' I -0-1

 

 

Raman Shift (cm'l) Raman Shift (cm'l)

Figure 3.8. Principal component loadings of Raman spectra obtained from a "blank"
experiment (no PG present). (PC 1) is the first principal component, (PC 2) is the second
principal component. Numbers in italics indicate peak position in cm".

68

in signal intensity at Raman shifts below 1640 cm". There is a significant feature
beginning at 1640 cm'1 and also a rapid decrease in values along the PC 2 axis from 1670
to 1688 cm”. This indicates that PC 2 is being influenced by the appearance of the water
band and the peak shift that also affect PC 1. PC 2 is describing subtle additional
information that is not described by PC 1. PC 1 describes 98.84 % of the variance while
PC 2 describes only 0.81 % of the variance. For the blank experiments PC 1 alone is
capable of describing most of the variance in the data. Figure 3.9 is a plot of cumulative
water voltune versus the scores for PC 1. This plot demonstrates how PC 1 is capable of
describing the behavior of the samples in the data set (this figure is identical to a graph of
signal intensity versus cumulative water volume for the raw spectral data [data not
shown]). As water is added to the solution at a constant rate, the signal intensity of water
should increase and the signal intensity of acetone should decrease. The increase or
decrease of signal intensity is not expected to be linear (as might be expected with a
constant water addition rate) since Raman spectroscopy is affected by concentration and
not volume.
3-4.7 PCA of Varfl'nglnitial PG Concentration Crystallizations

As described in the experimental section above, three sets of three crystallizations were
performed. Each set of three crystallizations was conducted with a different initial PG
concentration. Figure 3.10 shows the loading plots of the first three PCs for
crystallization experiments with initial PG concentrations of 3.5, 7.0 and 14.0 wt%.
Separate PCA was performed for each set of three replicates at the different initial

concentrations.

69

 

2.0E+04

1.0E+04 *6 ~- We

 

 

 

 

I.
0 0.0E+00 44.!” _--- ,___1___,,
.. 9 .-
B o 4 8 12 16
8 ,6
-1.0E+04 -—4.—--——- __ — —--—— —— ——-———4
.0
O
0
-2.0E+04 ._:__._, -L 4w- ‘— H- ——
.0
-3.0E+O4

 

 

 

Cumulative Water Volume (ml)

Figure 3.9. PC 1 scores versus cumulative water volume (ml) plot from a "blank"
experiment (no PG present).

70

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

O 16 3.5% PG 7.0% PG 14% PG
5 1672 1671
a; 1609 1671 1608 A 1609 A
5,008
5 M W
1:
(U
o
_l
o . _ .
1590 1640 1690 1590 1640 1690 1590 1640 1690
0.15
1650 F 1645 1645
Q 1606
0 1606 1606 /,I/\
(3.005 ,\
2‘ A l A j
:3 1590 \l164o \169'0 1590 \l1640 \1690 1590 /\/1640\ 1690'
8005 V I V I
"I V 1615 V
1616 1616
1672 1671 1672
-015
0'15 1659 1659 1658
6‘ 1611 1611 1612
0 A
$0.05
5 I \ I \ 11 I I1 I1
=5 1590 \1640 \1690 1590 1640 \1690 1590 \1640' 7690
30.05 I K
.1
-o.15 -
- -1
Raman Shift (cm )

Figure 3.10. Principal component loadings of Raman spectra obtained for varying initial

PG concentrations.

(A) 3.5wt%. (B) 7.0wt%, (C) 14.0wt%. (PC1) first principal

component, (PC2) second principal component, (PC3) third principal component.
Numbers in italics indicate peak position in cm' .

In Figure 3.10 the loading plots of PC 1 for the three crystallization experiments
demonstrate that 1609 and 1672 cm'1 are the Raman bands that have the most influence
over PC 1. These features can be matched to the PG solution peaks for the C=C
vibrations at 1610 cm'1 and the C=O vibrations at 1670 cm”. From that it can be
concluded that PC 1 is most affected by decrease in intensity of the PG solution peaks
observed as water is added.

In Figure 3.10 the PC 2 loading plots for all three crystallizations Show that there are
four main Raman bands that influence PC 2. The first two Raman bands can be
addressed together, with features at 1606 and 1616 cm”. These two features in the PC 2
loading plots demonstrate the shift that the first solution PG peak (originally at 1610 cm'
1) undergoes as water is added. This peak shift can be seen in Figure 3.4, where the first
PG solution peak begins at 1617 cm’] and then shifts towards lower Raman wavenumbers
as water is added to the solution (this is prior to crystallization, once crystallization
occurs the peak shifts back to higher wavenumbers). The third feature of the PC 2
loading plots covers a broad range of waventunbers (from 1620 to 1660 cm”). This
feature is approximately centered on 1640 cm"1 and has already been interpreted as
arising from water molecules in a previous section of this work. The fourth and final
feature seen in the PC 2 loadings plots of Figure 10 is centered on 1670 cm". This
feature is due to the shifting (to lower wavenumbers) of the second PG solution peak as
water is added. In addition, this feature also tails towards higher PC 2 values at 1688 cm'
1. This strong tailing is due to the peak shift (to lower Raman shifts) of the C=O band of

acetone. Both of the peak shifts described above can be seen in Figure 3.4. Using the

72

above discussion it is possible to conclude that PC 2 is mostly influenced by changes in
the Raman band positions caused by the addition of water.

Finally, it is possible to analyze the PC 3 loading plots for the three crystallizations
shown on Figure 3.10. The PC 3 loading plots exhibit two main features. The first
feature is centered on 1612 cm‘1 and the second is centered on 1658 cm". These features
are a good match with the expected peak positions of crystalline PG. From the
assignments above it can be concluded that PC 3 is mostly influenced by the formation of
crystals in solution. This conclusion is corroborated by the fact that PC 3 loading plots
display only random noise before crystallization occurs (data not shown). It is only after
crystallization occurs that the PC 3 loading plots begin to be influenced by the crystalline
PG Raman bands.

Examining the three PC 2 and PC 3 loading plots shown in Figure 3.10, it becomes
evident that many of the variables have negative PC 2 and PC 3 loadings. It should be
noted that only the absolute magnitude of the loading is discussed here. In some
applications, the Sign of the loadings have been utilized to make physical
interpretations15 ; in this work the use is mainly in establishing the relative contributions
of the variables to the PCs.

For that reason careful examination of the data must be undertaken when
spectroscopic bands are correlated to PC loading plots. In the present work the
correlations for all three PCs are simplified by the well-defined loading plots and the high
resolution of the spectral data. In addition, the positioning of the loadings for each PC

matches well with the raw Raman data.

73

In the discussion above regarding the influences of Raman shifts on the PC loading
plots, all of the raw data from the crystallization experiment were used as input. When
monitoring crystallizations in situ, this would not be the case. The data would be input
sequentially as it was collected. The PCA results would not reflect the appearance of the
solid until crystallization occurred. For this reason, PC 3 could not be used to monitor
the crystallization in situ. Because PC 1 and PC 2 are influenced by the signal intensity
from the PG, acetone and water in solution they are largely unaffected by the removal of
the post crystallization data. This allows for in situ monitoring of the crystallization
process using PC 1 and PC 2.

Figure 3.11 displays selected score plots for a crystallization conducted with a 7wt%
initial PG concentration and a 0.5 ml/min water addition rate ([6] having the lowest and
[(1] having the highest water concentrations). The score plots were calculated in situ, as
water was added to the initial solution spectra were acquired and immediately placed into
the PCA data set (arrows indicate the direction of increasing water concentration).
Beginning with score plot [a] it can be seen that PC 1 is capable of differentiating
between the spectra according to water concentration. Score plots [b] and [c]
demonstrate that as water is added PC 2 can also aid in differentiating the spectra with a
noticeable change in slope for the data. Finally, score plot [(1] shows how PC 1 and PC 2
can differentiate between PG in solution and crystalline PG suspended in a liquid.

The scores plots do not provide any information regarding the passing of the solubility
limit for the solutions (the solubility limits do not correspond to any particular feature in
the score plots), but they do seem to provide a fair amount of warning as to the

impending crystallization. The change in SIOpe along the PC 2 axis can be used as a

74

 

 

;-2.0 E+04 l

 

 

 

 

 

 

 

 

 

 

 

§ 2.0 E+04
I
g-1.0 E+04 ’
i I
E-2.0 E+04
PC1
1.0 E+04
c) .
1 » _‘._—_ 7a_r
-2.0 E+04 j ’9 2.0 E+04
. e j .
O.
-r-1.o E+04 r — —" * '
1
l
.. -20 E+04 -
IPC1

 

 

 

-2.0 E+04

 

O

S
2.0 E+04

,_ ——<

 

 

 

l
92 I
-1.0 E+04 7'
i
-2.0 E+O4 I
“' PC1
' 1.0 E+04
d)
O l
._ t v-,.__
N
o
n.

1
r
l

 

 

    

l
l
> -10 E+04 ”1— -—-. ‘
‘ O
* e

 

 

 

Figure 3.11. Principal components scores plot of Raman spectra obtained for
crystallization with 7.0wt% initial PG concentration. The score plots were
calculated after a specific volume of water was added to the original solution.
a) 1.52 ml, b) 4.81 ml, c) 8.14 ml and d) 12 ml. Beginning of crystallization is

circled in score pot c).

concentration.
crystallization.

75

Arrows indicate direction of increasing water
The circled point on plot (c) indicates the beginning of

warning of the onset of crystallization, and was reproducible in all the other
crystallization experiments performed (data not shown). These results are promising for
crystallization monitoring. PCA is able detect subtle changes in signal intensity that are
not visible from the raw Raman spectral data. The detection of these slight changes
provides a signal for the irnminence of the approaching crystallization. In addition, PCA
differentiates the spectral features of the three major "components" present in the region
analyzed. That is, the spectral features of PG in solution mostly influence PC 1, water
spectral features mostly influence PC 2 and the PG slurry spectral features mostly
influence PC 3. This helps reveal the presence of a water band that is not readily
apparent from the raw Raman data.
3-5 Conclusions

The use of PCA was demonstrated for aiding in the monitoring of anti-solvent addition
crystallizations. For the region analyzed, PCA was able to distinguish spectral features
from the solute in solution, the solute crystals and the anti-solvent. Furthermore, PCA
was capable of detecting subtle changes in the spectral data (as the addition of anti-
solvent progressed) that could be used as a warning for the onset of crystallization. It
must be emphasized that the methodology introduced in this work must be considered
system specific. It is expected that using a different system will yield vastly different
results. For that reason, this work is only intended to show that PCA is a powerful
technique that has not been used previously to study crystallizations. In this case PCA
was able to provide useful information on crystallization phenomena beyond that, which

was obvious from the raw spectral data.

76

3-6 Acknowledgments

Dr. David Burgard provided invaluable assistance in applying and understanding the
results obtained from the PCA of the spectral data. Kaiser Optical Systems, Inc. provided
the Raman spectrometer and HoloReact software and Pharmacia Corporation provided
the steroids used in this study. Their contributions are appreciated. Partial support of J.

A. F. was received from GAANN and Alfred P. Sloan Foundation fellowships.

77

3-7 References

1.

2.

10.

ll.

12.

13.

14.

15.

Threlfall, T. Analyst, 1995, 120, 2435-2460.

Falcon, J. A.; Berglund, K. A. Journal of Crystal Growth and Design, 2003, 3(6),
947-952.

. Polymorphism in Pharmaceutical Solids; Brittain, H.G., Editor; Marcel Dekker: New

York, 1999.

Dunuwila, D. D.; Berglund, K. A. Trans. 1. ChemE, 1995.

. Cerreta, M. K.; Berglund, K.A. Journal of Crystal Growt,. 1987, 84, 577-588.

Battey, D.E.; Slater, J.B.; Wludyka, R.S.; Owen, H.; Pallister, D.M.; Morris, M.D.
Applied Spertroscopy, 1993, 47, 1913.
Owen, H.; Battey, D.E.; Pelletier, M.J.; Slarter, J .B. Proc. SPIE, 1995, 24(6), 260.

MATLAB Function Reference, version 5. The Math Works Inc.
Dongarra, J. J .; Bunch, J. R.; Moler, C. B.; Stewart, G. W. LINPACK Users’ Guide,
Chapter 11; Society for Industrial and Applied Mathematics: Philadelphia, 1979.

Dollish, F.R.; Fateley, W.G., Bentley, F.F. Characteristic Raman Frequencies of
Organic Compounds; Wiley-Interscience: New York, 1974.

Silverstein, R. M.; Bassler, C. G.; Morril, T. C. Spectrometric Identification of
Organic Compounds, 5th Edition; John Wiley & Sons: New York, 1991.

Brotherton, J. Sex Hormone Pharmacology; Academic Press: New York, 1976.
Walker, J. Private Communications.

Eisenberg, D.; Kauzmann, W. The Structure and Properties of Water; Oxford
University Press: Oxford, 1969.

Martans, H; Naes, T. Multivariate Calibration; John Wiley & Sons: New York,
1989.

78

Chapter 4

IN SI T U MONITORING OF CHANGES IN SUNFLOWER OIL FATTY ACID
COMPOSITION DURING HIGH TEMPERATURE PROCESSING USING A
DISPERSIVE RAMAN SPECTROMETER"

*Submitted to Organic Process Research & Development, April, 2004.

4-1 Abstract

The rapid determination of trans isomer content in edible oils is of great importance in
the characterization of fats and oils in the food industry. A novel dispersive Raman
spectroscopic procedure for the in situ monitoring of fatty acid cis to trans isomerization
in edible oils is described. Raman spectral changes observed during the high temperature
heating of oil were correlated with cis and trans isomer concentration determined by gas
chromatography (GC). The short acquisition times, low laser power and ease of
application of dispersive Raman spectroscopy using an NIR excitation source combined
with the correlation derived from the GC data provided the capability of monitoring the
isomerization in situ. The method developed here has potential applications for increased
process control of edible oil hydrogenation, deodorization and quality control among
other widely used processing steps. The in situ monitoring capability provided by
dispersive Raman spectroscopy offers considerable advantages over the conventional off-
line FT-Rarnan spectroscopy analysis.
4-2 Introduction

The accurate determination of the concentration of various types of unsaturation
(especially cis and trans isomers) of fats and oils has become a subject of great
importance to the food industry. This is due in part to new regulations1 set forth by the

Food and Drug Adminisu'ation (FDA) regarding food and nutritional labeling. The FDA

79

has been urged to set forth these new regulations by consumer groups with increasing
health consciousness and the mounting evidence for cardiovascular health risks from the
consumption of high levels of saturated and trans fatty acids.

Current standard methods for determining the cis and trans content of fats and oils
include gas chromatography and Fourier transform infi'ared spectroscopy (FT-IR) of
samples captured off-line. In the past Raman spectroscopy has also been used for this
purposez’3’4’5’6’7, although its use has been ahnost entirely limited to Fourier Transform
(FT) Raman spectroscopy. The one exception is the seminal work in this area, in which
Bailey and Horvat2 described the use of a dispersive Raman spectrometer, but used it to
analyze only pure lipid samples. Although it has been proven that FT-Raman
spectroscopy is capable of providing good quality fluorescence-free spectra (due to the
use of 1064 nm excitation wavelength) the current technology only allows for off-line
analysis of samples.

Past publications3’4’5’6’7 have correctly argued that dispersive Raman spectroscopy using
visible light excitation sources could not be used to analyze commercial oils and fats.
This was due to fluorescence (fluorescence in oils and fats is most often caused by
natural antioxidants, carotenes, nutritional supplements and/or coloring ingredients) fiom
the samples which severely degraded the quality of the spectrum. However, within the
past few years there have been technological advances that now allow dispersive Raman
spectrometers to use near infrared (NIR) lasers as excitation sources. The use of an NIR
laser for excitation eliminates the interfering fluorescence from oil and fat samples. The
present work takes advantage of the flexibility and the in situ analysis capability offered

by dispersive Raman spectrometers and combines it with an NIR laser excitation source

80

in order to demonstrate the feasibility of using dispersive Raman spectroscopy to
qualitatively and quantitatively monitor changes (in situ) of the fatty acid composition of
sunflower oil during high temperature processing.
4-3 Experimental Section

Three 32 F1 oz. 100% Sunflower oil bottles (Meijer, Grand Rapids, MI) were
purchased from a local retail supplier. “Fresh” 250 mL batches of oil were heated to one
of the three following temperatures: 200, 250 or 300 °C and held at the particular
temperature for 40 minutes. Experiments were performed in triplicate, always using
fresh oil at the start. Oil was heated in a 1.19 L stainless steel beaker (Polar Ware,
Sheboygan, WI) using a stirrer/hotplate model PC-420 (Corning Incorporated Life
Sciences, Acton, MA). Hotplate temperature was maintained within 5 °C of the desired
temperature using a CN7600 temperature controller (Omega Engineering, Inc., Stamford,
CT). The oil was continuously stirred during heating using a stainless steel four-blade
propeller attached to a StedFast model SL 1200 mechanical stirrer (Fisher Scientific
International, Inc., Hampton, NH). All experiments began by obtaining the Raman
spectrum of the oil at room temperature (20 °C), the oil was then heated at approximately
10 °C/minute until reaching the desired temperature. A schematic diagram of the
experimental setup is shown on Figure 4.]. Once the 40 minute heating time period
ended, the oil was immediately placed into a water-ice bath. This ensured that the oil was
maintained at a constant temperature for exactly 40 minutes. Once cooled, aliquots of the
oil were placed in a fi'eezer (0 0C) for storage until they were analyzed by GC or firrther

analyzed by Raman spectroscopy.

81

 

 

 

 

Figure 4.1. Schematic of the experimental setup. (1) Raman probe, (2) HoloLab 5000,
(3) PC, (4) hot-plate, (5) temperature controller, (6) thermocouple, (7) stainless steel
propeller stirrer.

82

4-3.l Analytical Methods

The gas chromatography (GC) operational settings used in this work are similar to
AOCS official method Ce 1C-898. This method specifies GC analysis for methyl ester
derivatives of fatty acids. Methyl ester derivatives were obtained by reacting sunflower
oil with methanol/BF; as specified by AOCS official method Ce 2-668. Analysis of the
fatty acids composition of the oil was carried out using a Shimadzu GC-17A (Shimadzu
Corp., Kyoto, Japan) GC fitted with a Supelcowax-lO (Supelco, Inc., Bellefonte, PA)
capillary column (60 m X 0.32 mm id, 0.5 pm fihn thickness). The carrier gas used was
helium at a flow rate of approximately 1mL/min. The sample injected was 1 11L and
injection was by a split injection system (100:1 split ratio). The injector temperature was
210 °C, while the flame ionization detector was maintained at 230 °C. The column oven
was temperature programmed from 150 to 200 °C at 1.3 °C/min and held at that
temperature for 30 minutes. Identification of peaks was by comparison of relative
retention times with those of authentic standards.

Fluorescence spectra were acquired using a Spex F1uorolog-2 Spectrofluorometer
(Spex Industries, Inc., Edison, NJ). Sunflower oil was tested using 632.8 and 785 nm
excitation wavelengths. The collection ranges were 635 to 800 nm for 632.8 nm
excitation and 790 to 900 nm for 785 nm excitation wavelength. Acquisition time was 1
seconds per collection wavelength.

Raman spectra were collected using a HoloLab Series 5000 instrument (Kaiser Optical
Systems, Inc., Ann Arbor, MI). During heating Raman spectra were acquired in one
minute intervals. Once the desired temperature was reached, Raman spectra were

acquired using two minute intervals. All spectra were obtained using two acquisitions

83

with two second exposure leading to a four second total acquisition time per spectrum.
Details of the HoloLab 5000 have been described previouslyg’lo. Briefly, a 100 mW
external cavity stabilized diode laser at 785 nm is used for sample illumination. A CCD
camera, a spectrograph, and a fiber optic probe complete the HoloLab 5000 system.
Remote sampling was accomplished by employing a fiber optic probe attached to an
immersion optic to focus the incident beam into the solutions. The positioning of the
immersion optic was maintained constant through out all the experiments. This ensured
that any effects the immersion optic may have had on the experiments remained constant
throughout.
4-4 Results and Discussion

As discussed in the introduction section, many previous FT-Raman studies of edible
oils have argued that only FT-Rarnan and not dispersive Rarnan spectroscopy can be used
to analyze edible oils. The reason given was that dispersive Raman spectrometers were
limited to using excitation lasers in the visible wavelength range. Using visible
wavelength range excitation sources caused fluorescence from impurities or additives in
the oil. The fluorescence would overwhelm the Raman scattering signal and therefore
obtaining a spectrum with acceptable sigrral-to-noise ratio was impossible. This argument
was correct, until approximately five years ago when a dispersive Raman spectrometer
using NIR excitation laser became available. However, some FT-Raman studiess’7
analyzing edible oils continue to use the outdated argument. It can be demonstrated that
in order to avoid the fluorescence interference when analyzing edible oils it is not

necessary to use an excitation source as far into the IR region as FT-Raman does using

84

A;

 

 

 

 

Arbitrary Intensity

l 3.
L _L-- afi— EAL

630 720 810 900
Wavelength (nm)

    

 

Figure 4.2. Fluorescence emission spectra of sunflower oil obtained for oil at room
temperature. (A) 632.8 nm excitation, (O) 785 nm excitation.

85

the 1064 nm wavelength. Figure 4.2 shows the fluorescence emission spectra of
sunflower oil using excitation at 632.8 and 785 nm. It is evident from these spectra that
using 785 nm excitation sources reduce the fluorescence intensity by approximately half.
These results indicate that it may be possible to use a dispersive Raman spectrometer to
analyze edible oils. Using a lower wavelength excitation source provides one main
advantage, improved sensitivity. Because the efficiency of Rarnan scattering is inversely
proportional to wavelength to the fourth power, a lower wavelength excitation source
allows the use of lower laser power and shorter acquisition times. The shorter acquisition
times allow for in situ monitoring of processes.

Figure 4.3 displays the Raman spectrum of sunflower oil at room temperature. This
figure demonstrates that by using an NIR laser as an excitation source it is possible to
obtain high signal to noise ratio, fluorescence fiee Raman spectra of edible oils using
dispersive Raman spectrometers. Table 4.1 displays the assignments of the major Raman
bands of the sunflower oil spectrum. There are numerous references in the literature that
have utilized various ratios of the peaks presented in Table 4.1 in order to quantitatively
analyze the fatty acid composition of edible oils. As mentioned in the introduction
section all of these references have been limited to analyzing samples off-line. This is
due to all but one reference using FT-Raman spectroscopy for the analyses. By using a
dispersive Raman instrument it was possible to implement similar peak area ratio
methodologies used by the previous studies in order to monitor (in situ) changes in the
fatty acid composition of edible oils during high temperature processing. As described in

the experimental section batches of fresh sunflower oil were heated to 200, 250 or 300 °C

86

 

2850 2900
\ /

Intensity

3010
/

 

 

 

 

 

 

0 i 1000 I 2000 I 3000
Raman Shift (cm‘I)

Figure 4.3. Raman spectrum of sunflower oil acquired at room temperature using a
dispersive Raman spectrometer.

87

 

Molecule 4 Vibration

 

3010 RCH=RCH =C-H Asymmetric

 

 

 

 

 

 

 

 

stretching
Symmetric
2900 -CH3 C-H stretching
2850 -CH2 C-H Synm‘tmc
stretching
1 745 RC=OOR C=O Stretching
trans only _ .
1665 RCH=CHR C—C Stretching
cis only _ .
1655 RCH=CHR C—C Stretching
1440 -CH2 C-H Deformation
1300 -CH2 ‘ C-H Deformation
cis only __ .
1265 RCH=CHR —C-H Deformation

 

 

 

 

 

 

 

1070 -(CH2)n— C-C Stretching

 

Table 4.1. Assignment of the major Raman scattered bands of sunflower oil.

88

and held at the corresponding temperature for 40 minutes. GC analyses of unheated
samples of sunflower oil revealed the following fatty acid profile: 6.01 i 0.02 % pahnitic
acid, 4.15 i 0.33 % stearic acid, 24.53 i 3.04 % cis oleic acid, 0.59 i 0.03 % trans oleic
acid, 64.41 i 2.24 % cis,cis linoleic acid, 0.26 i 0.03 % cis,trans linoleic acid. These
values are in good agreement with previously reported11 sunflower oil fatty acid
concentrations. It should be noted that because it was necessary to purchase three bottles
of oil in order to perform the experiments, the standard deviation values calculated for all
the data presented in this work are influenced by the differences in the fatty acid
composition of natural oils. Table 4.2 lists the results obtained from GC analysis for the
total %cis and total %trans of the samples heated at the three temperatures plus a control
sample which was not heated. The calculation of %cis and %trans is not straight
forward, since sunflower oil contains high amounts of linoleic acid. Linoleic acid with its
two double bonds can be present in the samples as one of four isomers: cis,cis / cis,trans/
trans, cis / trans, trans. The recent FDA regulation1 that requires trans fat concentration
to be listed in the nutritional label of foods provides the solution. If the double bonds are
not conjugated, cis, trans and trans, cis isomers should be considered trans.

An examination of Table 4.2 reveals that there are significant changes in the total %cis
and the total %trans concentration (determined by GC) of fatty acids as a function of
temperature. The total %cis concentration decreases by 8.5% while the total %trans
concentration increases by 7.5%. A more in-depth analysis of the GC data reveals that
there is no change in the concentrations of cis or trans oleic acid and only linoleic acid is

. . . . . . . . 12
undergorng rsomerrzatron. These results are in good agreement wrth prevrous findings

89

Table 4.2.
determined

 

 

 

 

 

 

 

 

1 Temperature Total cis I ' Total trans

* (°C) (%) (%) .
25 88.69 i 3.78 0.94 i 0.11
200 87.18 3: 5.66 1.68 :t 0.64
250 87.67 :t 4.47 1.22 d: 0.17
300 80.12i3.18 8.48:t 1.27

 

Fatty acid composition of sunflower oil as a function of temperature

90

 

which indicate that polyunsaturated fatty acids are susceptible to heat induced
isomerization.

Table 4.1 was utilized to find peak areas and peak area ratio that could be correlated to
the isomerization taking place. Previous studiesz’3’7 have focused on the 1655 cm'1 peak
as an indicator of cis isomer concentration. Our attempts to correlate this peak area with
the total %cis concentration determined by GC resulted in a poor linear model. The poor
linearity can be explained by the fact that during isomerization the peak area of the 1665
cm"1 peak increases. This peak has been assigned to the trans C=C stretching, as the
isomerization progresses and the 1665 cm'1 peak area increases it overlaps with the 1655
cm'1 peak and thus interferes with the calculation of the 1655 cm"1 peak area. Because of
this interference it was necessary to find another peak that could be used in order to
monitor the concentration of cis isomer. The 1265 cm”l peak, which has been assigned to
in—plane =C-H deformation in an unconjugated cis double bond3, was the other available
peak. Although the 1265 cm'1 peak also suffers from some overlap with another peak
(the 1300 cm'1 peak assigned to in-plane CH2 twisting) the overlap is not as severe, and
the correlation between this peak area and the total %cis concentration provides a
reasonable linear correlation. In order to improve the reliability of the correlation
between the spectroscopic data and the GC data a peak area ratio was used instead of the
absolute peak area of the 1265 cm'1 peak.

The 1745 cm’1 peak was chosen to ratio with the 1265 cm'1 peak, this peak ratio has
successfully been used previously4 to monitor the cis and trans isomer concentration in
off-line oil samples. The 1745 cm'1 peak has been assigned to the carbonyl stretching of

fatty acids. The carbonyl moiety should be unaffected by the isomerization of the C=C

91

bond, which indicates that the 1745 cm'1 peak should also be unaffected. Although the
discussion above regarding the use of the 1745 cm'1 peak is valid, it fails to take into
account the many degradation products that are known to form during the hearting of oil
at high temperatures”. More specifically, previous work14 has demonstrated that
sunflower oil heated at 300 0C for 40 minutes degrades the oil into a variety of oxidation
products such as ketones and aldehydes (lower temperatures produced insignificant
changes). In that work, an increase in concentration of those carbonyl containing
compounds due to the heating of sunflower oil was demonstrated to increase the peak
area and the peak width of the 1745 cm'1 peak during off-line studies using FT-IR for
analysis. However, even using FT-IR, which is known to be far more sensitive to
carbonyl vibrations than Raman spectroscopy, the changes to the spectrum of sunflower
oil due to the formation of the oxidation degradation products was quite small. Figure
4.4 shows the average (of three experiments) peak area of the 1745 cm'1 peak plotted
versus time for sunflower oil being heated at 300 °C for 40 minutes. It can be seen that
there is no meaningful change in the peak area. Without more in-depth data analysis
(such as derivative calculations or multivariate analysis) the use of baseline integration
calculations of unprocessed Raman spectra is incapable of detecting the changes to the
1745 cm'1 peak that should accompany the increase in carbonyl containing compounds.
Because the 1745 cm’1 peak area remains constant through the heating process at a
particular temperature, it was a good candidate to “normalize” the 1265 cm'1 peak area.
The plot of the 1265/1745 peak area ratio versus the total %cis determined by GC for
sunflower oil unheated and after heating to the three test temperatures is shown in Figure

4.5. This plot gives rise to a linear fit of with a reasonable correlation coefficient of 0.95.

92

 

 

 

1745 cm'1 Peak Area

 

 

 

0 9 18 27 36 45
Time (minutes)

Figure 4.4. Average peak area and standard deviation for 1745 cm"1 peak plotted against
time for oil heated at 300 °C for 40 minutes.

93

 

1 . 90 TCLVCMLL. a-.- “IDEALS.” M-,-” --,. - _.-.......t._._.mr_-- -- - - _... W-, - _ - . - .. ._, Hwy}

1.73 :

 

 

1.55

 

1.38 1*

 

1265" 745 Peak Area Ratlc

 

 

1.20

 

78 82 86 90
Total cis Concentration (%)

Figure 4.5. Peak area ratio 1265/1745 plotted against the total %cis concentration as
determined by GC. Linear fit displays reasonable correlation coefficient of 0.95. (O)

unheated sample, (X) 200 °C, (A)250 °C, (0) 300 0C.

94

090 ..._____...__. .. _ _.-.,, . “Jaw--- _. ”a,“ . , ... .. ”N---“ Dunucwny- . . -,-,L,,-H _ , - - --_ ..

0.80 1-. ——— *7 -~ -9. --.-_.__¥, -- A-

0.70 *-

1745I1265 Peak Area Ratic
.0
8

 

 

 

0.50

Total trans Concentration (%)

Figure 4.6. Peak area ratio 1745/1265 plotted against the total %trans concentration as
determined by GC. Linear fit displays reasonable correlation coefficient of 0.93. (O)

unheated sample, (X) 200 °C, (A)250 °C, (0) 300 °C.

95

If the peak area ratio of 1265/1745 can be correlated to the concentration of cis
isomers, it should be possible to correlate the inverse of the peak area ratio to the trans
isomer concentration. The plot of 1745/1265 peak area ratio versus total %trans is
presented as Figure 4.6. The plot on Figure 4.6 provides a linear fit with a reasonable
correlation coefficient of 0.93. For clarity, it should be stated that the spectral data used
for Figures 4.5 and 4.6 were obtained at room temperature. That is, the samples were
heated and cooled as described in the experimental section and then Raman spectra were
acquired. Figures 4.5 and 4.6 demonstrate that at a constant temperature, a linear
correlation can be made between these peak area ratios and the GC data which can be
used to monitor changes in the total %cis and total %trans concentrations of edible oils.
Figure 4.7 displays the 1745/1265 peak area ratio versus time for sunflower oil during
heating at the three test temperatures. From Figure 4.7 it can be seen that during the 40
minutes for which oil was held at 200 °C the 1745/1265 peak area ratio did not change
from its initial value. The ratio can be seen to increase slightly for the oil held at 250 °C.
Finally, for the oil held at 300 °C the ratio undergoes a relatively large increase from its
initial value. Utilizing the linear correlation developed in Figure 4.6 it is possible to state
that the changes in the peak area ratio seen in Figure 4.7 are directly related to the change
in concentration of the trans isomer in the oil samples. Figure 4.7 demonstrates that the
rate of thermally induced cis to trans isomerization increases with increasing temperature
under the circumstances examined in this work. The increased rate therefore leads to a
higher total %trans concentration in oils heated at 300 0C as compared to oils heated at
200 or 250 °C for the same period of time. The results from Figure 4.7 are in good

agreement with the GC data presented in Table 4.2. Only at 300 °C is there a marked

96

 

 

 

g l/I' ; I
£0.90 i ————-_- ’E ’I’ . f]. l l l
2 ' {Jlrj 1
13060 i 11‘; lTIf%,
n. 5 ..
§ “ 5...“ ‘ A‘I “II—5% .. If 1
E070 1 ~ : X a;
= l
Mdc, -- __a__1

30 37.5 45

O
N
01
A
01
N
.N “1
0|

Time (minutes)

Figure 4.7. Peak area ratio 1745/1265 plotted against time for the three test temperatures.
(X) 200 °C, (A)250 °C, (0) 300 °C,

97

increased in the concentration of trans isomers. The results displayed in Figure 4.7 also
agree well with previous work15 which demonstrated that the thermally induced cis to
trans isomerization of linoleic acid was strongly dependent on temperature.

It would now seem possible that the linear correlation obtained from Figure 4.6 could
be used to roughly quantify the data from Figure 4.7 by converting the 1745/1265 peak
area ratio to trans isomer concentration. This quantification would result in trans isomer
concentrations much higher than the actual values. The error can be (apart from the fact
that the calibration curve being used would be comprised of only four points) explained
by the dependence of the ratio of anti-Stokes to Stokes Raman scattering on Boltzmann’s
distribution. At lower temperatures (up to approximately 225 0C) the anti-Stokes to
Stokes peak area intensity for the 1265 cm'1 peak is smaller than the same ratio for the
1745 cm'1 peak. While at higher temperature (approximately 250 °C and above) the
opposite is true. This inversion in peak intensities means that the 1745/1265 peak area
ratio will not be constant when compared between different temperatures. Because of the
dependence of the peak areas on temperature, it would be necessary to create a
calibration curve for every temperature tested.

In addition to monitoring cis to trans isomerization, it should also be possible to
monitor degradation of fatty acids and the resulting degradation products. Initially, the
monitoring of fatty acid degradation appears to be a straightforward task, analyze the in
situ monitoring spectral data for the appearance of new peaks and the decrease in area of
existing peaks. However, as alluded to earlier in this section, the heating of oils at high
temperatures (whether food is present in the oil or not) produces a multitude of

degradation products13 . Some of the degradation products that have been detected are:

98

monoglycerols, oxidized triacylglycerols, tryacylglycerol polymers (also dimers and
trimers), cyclic fatty acids, fi'ee fatty acids, ketones, aldehydes, alcohols and esters. With
the appearance of such a large number of degradation products it would be difficult to
single out the effect of one particular molecule to the changes observable in the Raman
spectral data. Another difficulty with the monitoring of degradation products will be the
concentrations levels at which they are formed. Depending on the experimental
conditions many of the degradation products may be formed at concentrations below
those which Raman spectroscopy can detect. For these reasons great care should be
taken with the interpretation of Raman spectral data obtained from the in situ monitoring
of high temperature heating of edible oils.
4-5 Conclusions

The results of this study demonstrate the possibility and advantages of using dispersive
Raman spectroscopy with an NIR excitation source for analysis of edible oils on or off-
line. It can no longer be argued that FT-Raman spectroscopy is superior to dispersive
Raman spectroscopy for the analysis of edible oils. Using GC concentration
measurements in combination with Raman spectroscopy measurements it was possible to
create correlations between changes in the Raman spectra of oils and cis and trans isomer
concentrations. The methodology developed in this work for in situ monitoring of cis to
trans isomerization of edible oils has a number of immediate potential applications for
the oil production industry. It also provides significant advantages with regard to sample
handling/storage, sample analysis time and process monitoring/process control
capabilities over conventional techniques (GC, off-line FT-Raman, off-line FT-IR) for

the analysis of natural oils hydrogenation process. Other clear applications would be for

99

monitoring the deodorization of natural oils and also for on-line quality analysis of oils
during cooking. These advantages should translate to improvements in the process

productivity, production efficiency and superior quality control capabilities.

4-6 Acknowledgments

Dr. Dale LeCaptain of Central Michigan University provided one of the Raman
spectrometers. Kaiser Optical Systems, Inc. provided a second Raman spectrometers and
HoloReact software. Their contributions are appreciated. Partial support of J. A. F. was

received from GAANN and Alfred P. Sloan Foundation fellowships.

100

4-7 References

10.

ll.

12.

13.

14.

15.

. FDA/DHHS, 68 Federal Register, 41,434 (July 11, 2003). Food Labeling: Trans

Fatty Acids in Nutrition Labeling, Nutrient Content Claims, and Health Claims (Final
Ruling). (Internet address: http:// www.cfsan.fda.gov/~qu_/fr03711a.html)

Bailey, G.F; Horvat, R.J. J. Am. Oil Chem. Soc. 1972, 49 (8), 494-498.

Sadeghi-Jorabchi, H.; Hendra, P.J.; Wilson, R.H.; Belton, P.S. J. Am. Oil Chem. Soc.
1990, 67 (8), 483-486.

Sadeghi-Jorabchi, H.; Wilson, R.H.; Belton, P.S.; Edwards-Webb, J.D.; Coxon, D.T.
Spectrochimica Acta. 1991, 47A (9/10), 1449-1458.

Baeten, V.; Hourant, P.; Morales, M.T.; Aparacio, R. J. Agric. Food Chem. 1998, 46,
2638-2646.

Johnson, G.L.; Machado, R.M.; Freidl, K.G.; Achenbach, M.L.; Clark, P.J.; Reidy,
S.K. Organic Process Research & Development. 2002, 6 (5), 637-644.

Weng, Y.M.; Weng, R.H.; Tzeng, C.Y.; Chen, W. Applied Spectroscopy. 2003, 57
(4), 413-418.

Oflicial Methods and Recommended Practices of the AOCS, 5th ed.; Firestone, D.,
Ed.; American Oil Chemist’s Society: Champaign, Illinois, 1998.

Battey, D.E.; Slater, J.B.; Wludyka, R.S.; Owen, H.; Pallister, D.M.; Morris, M.D.
Applied Spectroscopy. 1993, 47, 1913.

Owen, H.; Battey, D.E.; Pelletier, M.J.; Slater, J.B. Proc. SPIE. 1995, 24 (6), 260.

Romero, A.; Sanchez-Muniz, F.J.; Cuesta, C. J. Sci. Food Agric. 2000, 80, 2135-
2141.

Jorge, N.; Guaraldo Goncalves, L.Ap.; Dobarganes, M.C. Grasas y Aceites. 1997, 48
(1), 17-24.

Food Lipids, Chemistry, Nutrition, and Biotechnology, Akoh, C.C., Min, D.B., Ed.;
Marcel Dekker: New York, NY, 1997.

Moya Moreno, M.C.M.; Mendoza Olivares, D.; Amezquita Lopez, F.J.; Peris
Martinez, V.; Bosch Reig, F. Journal of Molecular Structure. 1999, 482-483, 557-
561.

Leon-Carnacho, M.; Ruiz-Mendez, M.V.; Graciani-Constante, M.; Graciani-
Constante, E. Eur. J. Lipid Sci. T echnol. 2001, 103, 85-92.

101

Chapter 5

CONTINUING INVESTIGATION: IN SI TU MONITORING OF OIL
OXIDATION DEGRADATION PRODUCTS USING A DISPERSIVE RAMAN
SPECTROMETER

5—1 Introduction

The oxidation of edible oils containing a large amount of polyunsaturated fatty acids
(PUFAs) is a major concem for the food industry because it is directly related to
economic, flavor, and storage problems. Although relatively well understood in general
terms, lipid oxidation is a quite complex and variable process which depends on oil type
and conditions of oxidation”. A wide range of end products are associated with the
oxidative deterioration of oils at ambient and elevated temperatures”, the most important
being hydroperoxides, alcohols and aldehydes. Hydrocarbons, free fatty acids (FFAs),
esters, ketones, lactones, firrans, triacylglycerol polymers and other minor products may
also be produced. In addition, there is significant cis to trans isomerization and
conjugation of double bonds in the hydroperoxides formed as an oil oxidizes. The rate of
oxidation and the distribution of accumulated products depend strongly on the oil source,
fatty acid composition, degree of unsaturation, presence of metal ions and antioxidants,
time and thermal stress. This wide array of variables affecting the oxidation process
makes it essential to have a reliable means of assessing the oxidative status and oxidative
stability of edible oils.

Standard methods for determining oxidation degradation product content of fats and
oils include titrations (which are largely empirical in nature), capillary GC-MS and size
exclusion HPLC of samples captured off-line”. Within the last 5 to 10 years there has

been an increased interest in utilizing Fourier Transform infrared spectroscopy (FT-IR) to

102

analyze oxidation degradation productss’6’7’8. Surprisingly there has been little or no work
done on investigating the possibility of using Raman spectroscopy. There are two likely
explanations for the lack of interest in Raman spectroscopy. First, up until our recent
study9 the analysis of fats and oils with Rarnan spectroscopy was confined to the use of
FT-Raman spectroscopylo’n’n’”. The current FT-Raman technology does not provide in
situ analysis capabilities, which makes the technique inadequate for monitoring such a
dynamic process as oil oxidation. Second, is the fact that Raman spectroscopy is known
to be relatively insensitive (when compared to FT-IR) to functional groups often
correlated with oxidation products such as carbonyls, aldehydes and alcohols. Therefore,
it is likely that past researchers have assumed that Raman spectroscopy would be unable
to detect changes in the oil due to oxidation degradation. We propose that even though
Raman spectroscopy may be insensitive to the specific functional groups related to
oxidation degradation, the degradation products will have other functional groups which
Raman spectroscopy is particularly sensitive to such as peroxides and conjugated double
bonds”.

The present work attempts to take advantage of the flexibility and the in situ analysis
capability offered by dispersive Raman spectrometers in order to demonstrate the
feasibility of using Raman spectroscopy to qualitatively monitor changes (in situ) in the
composition of sunflower oil during oxidative degradation.

5-1.1 Lipid Oxidation

Long chain fatty acids are abundant in nature, most commonly in combination with

glycerol as either as tryacylglycerides or as the arnphiphilic membrane phospholipids.

From most sources, tryacylglycerides will be found to comprise a mixture of compounds

103

with differing proportions of saturated, mono-unsaturated, and poly-unsaturated fatty acid
residues. Particularly common are fatty acid residues with eighteen carbon atoms,
consequently we will focus our attention on derivatives of linoleic and linolenic acids
(structures shown on Figure 5.1). In the context of lipid oxidation and free radical
formation it is important to mention that both linoleic and linolenic acid incorporate
“skipped” diene units in their molecular structure. It is known that allylic C-H bonds are
relatively weakz, more specifically the C-H bonds of the central methylene group in
skipped dienes (like linoleic and linolenic acdis) are “doubly” allylic and are especially
weak (ca. 75 kcal / mol). These hydrogen atoms are therefore unusually susceptible to
radical attack leading to the formation of resonance stabilized dienyl radicals (see Figure
5.2 A).

Lipid oxidation is believed to be due to a combination of triplet and singlet oxygen
reactions with fatty acids. Triplet oxygen lipid oxidation has been extensively studied to
improve the oxidative stability of foods]. Triplet oxygen, a di-radical compound, does
not react with the singlet state linoleic acid instead linoleic acid must be in a free radical
form to react as shown in Figure 5.2A. This reaction is known to produce only
conjugated cis and trans hydroperoxides. Although triplet oxygen oxidation has been
widely studied it does not fully explain the initiation step of lipid oxidation. For this
reason researchers‘ have suggested that singlet oxygen is involved in the initiation of
lipid oxidation. Singlet oxygen is an electrophilic, non-radical compound which reacts
with electron-rich, non-radical compounds (like linoleic acid) as shown on Figure 5.2B.
Figure 5.2B demonstrates that the reaction between singlet oxygen and linoleic acid leads

to the formation of conjugated and non-conjugated cis and trans hydroperoxides.

104

OH

 

I Linoleic Acid

 

 

 

I Linolenic Acid

 

Figure 5.1. Molecular structure of linoleic and linolenic acids.

105

 

 

 

 

A 0011
.— ' 302 m...— f-‘1
H Rearrangement Conjugated
H00
3
O
13 \ ____3_._ \
Rearrangement Conjugated

 

 

 

 

 

_

Hmmr‘vfim

Conjugated

 

 

wmmflh

0'11 Non-conjugated

 

Figure 5.2. A) Reaction mechanism of triplet oxygen with linolenic acid. B) reaction
mechanism of singlet oxygen with linoleic acid].

106

Because singlet oxygen can react with the non-radical compounds, its reaction rate with
linoleic acid is known to be at least 1000 times higher than that of triplet oxygen‘.
Despite the large difference in reaction rates, free radical triplet oxygen oxidation is
considered the primary mechanism for the formation of volatile flavor components in
edible oils. This is due to triplet oxygen being more abundant than singlet oxygen and
also to the relative ease with which fi'ee radicals are formed in linoleic and linolenic acid
(due to the doubly allylic central group mentioned above) which allow the triplet oxygen
oxidation reaction to occur.
5-1.2 Hydroperoxide Decomposition

A general decomposition pathway of hydroperoxide to produce volatile compounds is
shown in Figure 5.3. The most likely first step in the decomposition pathway is the
cleavage between the oxygen and oxygen of the R-O-O-H, that is, R-O-O-H -) R0* +
*O-H. The unsaturated alkoxy radical can be cleaved by B-scission in two mechanisms
of cleavage (A and B of Figure 5 .3). It is clear that the decomposition pathway will lead
to a large number of degradation products. An in-depth discussion of all possible
degradation products is beyond the scope of this work. For a more comprehensive
analysis readers are directed towards the extensive body of literature available on the

subj ect1 '3 .
5-2 Experimental Section

Two 32 F1 oz. 100% Sunflower oil bottles (Meijer, Grand Rapids, M1) were

purchased from a local retail supplier. “Fresh” 150 mL batches of oil were heated to

107

 

R’ - CH=CH - CH — R”
OOH
-OH
B A
R’ — CH=CH 1 CH {- R”
. (I), .
Cleavage B Cleavage A
R’— CH=CH- + OHC - “R R’ - CH=CH - CHO + R”
OH . RH OH . RH
R - R .
B1 82 A1 A2
R’ - CH=CH - OH R’ — CH=CH2 R"— OH R” — H
R’ — CH2 - CHO

 

 

 

Figure 5.3. General decomposition pathway of hydroperoxide to produce volatile
compoundsl.

108

either 130 or 200 0C and held at that particular temperature for 180 minutes. Each
experiment was performed twice at each temperature. For the first experiment 0.5 g of
tert-Butylhydroquinone (TBHQ) was added to the oil once the desired temperature had
been reached. The second experiment was performed with neat oil. All experiments
were performed using fresh oil at the start. Oil was heated in a 1.19 L stainless steel
beaker (Polar Ware, Sheboygan, WI) using a stirrer/hotplate model PC-420 (Coming
Incorporated Life Sciences, Acton, MA). Hotplate temperature was maintained within 3
°C of the desired temperature using a CN7600 temperature controller (Omega
Engineering, Inc., Stamford, CT). The oil was continuously stirred during heating using
a stainless steel four-blade propeller attached to a StedFast model SL 1200 mechanical
stirrer (Fisher Scientific International, Inc., Hampton, NH). In order to facilitate
oxidation dry air (AGA Specialty Gas, Cleveland, OH) was bubbled continuously
through the oil at approximately 15 mL/min. The oil was heated at approximately 10
°C/minute until reaching the desired temperature. A schematic diagram of the
experimental setup is shown on Figure 5.4. Once the 180 minute heating time period
ended, the oil was immediately placed into a water-ice bath. This ensured that the oil was
maintained at a constant temperature for exactly180 minutes.

5-2.1 Analytical Methods

 

Raman spectra were collected using a HoloLab Series 5000 instrument (Kaiser Optical
Systems, Inc., Ann Arbor, MI). Once the desired temperature was reached, Raman
spectra were acquired using five minute intervals. All spectra were obtained using two

acquisitions with two second exposure leading to a four second total acquisition time per

109

 

 

 

Figure 5.4. Schematic of the experimental setup. (1) Raman probe, (2) HoloLab 5000,
(3) PC, (4) hot-plate, (5) temperature controller, (6) thermocouple, (7) stainless steel
propeller stirrer, (8) dry air input.

110

spectrum. Details of the HoloLab 5000 have been described previously15 ’16. Briefly, a
100 mW external cavity stabilized diode laser at 785 nm is used for sample illumination.
A CCD camera, a spectrograph, and a fiber optic probe complete the HoloLab 5000
system. Remote sampling was accomplished by employing a fiber optic probe attached
to an immersion optic to focus the incident beam into the solutions. The positioning of
the immersion optic was maintained constant through out all the experiments. This
ensured that any effects the immersion optic may have had on the experiments remained
constant throughout.

5-3 Results and Discussion

As described in the experimental section two different experiments were used to
investigate the possibility of using Raman spectroscopy to monitor the oxidation of edible
oils. The first experiment involved heating the oil to a specified temperature, then adding
the commercial (Food & Drug Administration [FDA] approved) antioxidant tert-
butylhydroquinone (TBHQ) and finally holding the oil at the specified temperature for
180 minutes. The second experiment is identical to the first, except that no TBHQ is
added. It should be mentioned that for this work TBHQ was added at a higher weight
percent (based on the “fat” weight of the product) than is allowed by the FDA (current
work used 0.4 wt%, FDA allows a maximum of 0.02 wt%)l7.

With this simple experimental design it was possible to analyze the Raman spectra
collected for the two experiments and determine if Raman spectroscopy was capable of
detecting any changes in the oil due to oxidation. Spectral subtraction was used in order
to elucidate any changes to the spectrum of sunflower oil that occurred during the

experiments. Unless otherwise specified the spectral subtraction involved subtracting the

111

first spectrum acquired once the oil had reached the desired temperature (spectrum at
time = 0) from the spectrum of interest. Figure 5.5 shows the results of spectral
subtraction for the experiment conducted at 130 °C with TBHQ added. In Figure 5.5, A
is the spectrum obtained after 5 minutes and B is the spectrum obtained after 180 minutes
of holding at 130 °C (spectra have been offset for clarity). Except for the change in
baseline linearity, spectra A and B do not display any significant differences. The change
in baseline linearity will be discussed below.

Figure 5.6 shows the results of spectral subtraction for the experiment conducted at 130
°C without TBHQ. Spectrum A was obtained after 5 minutes and spectrum B was
obtained after 180 minutes of holding at 130 °C (spectra have been offset for clarity). In
this case it is clear that there are substantial differences between the two spectra. One of
the major differences between the spectra shown in Figure 5.6 is the change in baseline
linearity between spectra A and B. A similar effect is seen in the spectra shown on
Figure 5.5. The difference in baseline is due to the degradation of naturally present
antioxidants in the oil. Work performed in our lab9 has demonstrated that these naturally
present antioxidants fluoresce weakly when exposed to 785 nm light. As the oil is
maintained at 130 0C the antioxidant concentration will decrease as they react with free
radicals formed in the oil. The first spectrum obtained in a particular data set will have a
relatively high fluorescence signal incorporated into its baseline, while spectra obtained
later in the experiment will have a lower concentration of antioxidants and thus lower

fluorescence. The difference in the fluorescence intensity due to the

112

 

Intensity

 

 

 

 

500 1000 1500 2000 2500
Roman Shift (crrr‘)

Figure 5.5. Difference Raman spectra obtained for the experiment conducted at 130 °C
with 0.5 g of TBHQ added to the oil. Difference spectra calculated by subtracting the
first spectrum acquired once the oil reached 130 °C from A) the spectrum acquired after
180 minutes at 130 °C B) the spectrum acquired after 5 minutes at 130 °C.

113

 

1660

 

 

 

 

 

 

 

 

 

870
E A
a 1
i
B
l I L L I l
0 500 1000 1500 2000 2500 3000

Raman Shlft (cm-I)

Figure 5.6. Difference Raman spectra obtained for the experiment conducted at 130 °C
with no TBHQ added to the oil. Difference spectra calculated by subtracting the first
Spectrum acquired once the oil reached 130 0C from A) the spectrum acquired after 180
minutes at 130 °C B) the spectrum acquired after 5 minutes at 130 0C.

114

decrease in antioxidant concentration leads to the changes in baseline observed in Figures
5.5 and 5.6. This theory is further supported by the fact that spectra A in Figure 5.5
displays a smaller change in baseline (when compared to its corresponding spectra B)
than spectra A in Figure 5.6. Since spectra A in Figure 5.5 contained a large amount of
synthetic antioxidants, it is expected that only a small fraction of its naturally present
antioxidant would have reacted with the free radicals formed in the oil. This would lead
to a sample with a stronger fluorescence signal at the end of the experiment and therefore
a smaller difference in baseline between itself and the first spectrum acquired.

Figure 5.6 A also displays significant features or peaks which are not seen in Figure 5.6
B or in Figure 5.5. It is therefore possible to conclude that the peaks seen in Figure 5.6 A
are due to the formation oxidation products during the experiment. The obvious next step
is the identification of the compound(s) responsible for the appearance of the peaks. The
first visible peak is located at 870 cm”, it is possible to assign this peak to the O-O
stretching vibration of organic peroxides. Organic peroxides are known to strongly
Raman scatter in the 870 cm'1 region”. In addition, we have observed that an addition of
5 wt% of tert-Butylhydroperoxide to sunflower oil produces a peak in the Raman
spectrum at approximately 875 cm”. The 875 cm'1 peak area is positive since the
concentration of peroxide compounds is higher at the end of the experiment. The next
visible peak at 1260 cm'1 can be assigned to the in-plane =C-H deformation in an
unconjugated cis double bond”), in fact previous work in our lab has correlated changes
in the area of this peak to the concentration of cis double bondsg. As expected the 1260
cm'1 peak area is negative, at the temperature examined this is most likely due to the cis

to trans isomerization occurring during oxidation. The third and most pronounced peak

115

in Figure 5.6 A is located at 1660 cm”. Closer examination of the spectrum reveals that
the 1660 cm'1 peak has a small shoulder near 1640 cm”. In order to better resolve the
two peaks, the first derivative of Figure 5.6 A was calculated (shown on Figure 5.7).
Figure 5.7 clearly demonstrates that the 1660 cm'1 peak in Figure 5.6 A is two overlapped
peaks at 1640 and 1660 cm]. The 1660 cm'1 peak has been previously assigned to trans
C=C bondsn’lg, while the 1640 cm'1 peak has been assigned to conjugated C=C
bondsm’lg. The appearance of both of these peaks during our experiments can be linked
to the oxidation of fatty acids (see Figure 5.2) since both of the functional groups
responsible for the peaks are known to form as oxidation products.

The final peak seen in Figure 5.6 A is located in the 2990-3020 cm”1 region. This peak
has been assigned to olefinic C-H stretching”, more specifically, Sadeghi-Jorabchi et
al.11 have demonstrated that the exact position of this peak can be correlated with to the
composition of C=C cis and trans isomers. In our particular case the peak appears at
3010 cm], the previous work assigned that peak position to cis,cis dienes. Figure 5.6 A
indicates that the concentration of cis,cis dienes decreases during our experiments. This
result is in accord with the fatty acid oxidation theory described previously which
predicts the transformation of cis C=C bonds to conjugated and unconjugated trans C=C
bonds.

Figure 5.8 shows the results of spectral subtraction for the experiment conducted at 200
°C without TBHQ. Spectrum A was obtained after 5 minutes and spectrum B was
obtained after 180 minutes of holding at 200 0C (spectra have been offset for clarity).

The data for the experiment conducted at 200 °C with TBHQ added will not be presented

ll6

 

1640

1656

/

Intensity

 

 

 

 

L L I l J I l
1600 1620 1640 1660 1680 1700 1720
Ramon Shift (cnr‘)

Figure 5.7. First derivative spectrum calculated for difference spectrum A shown in
Figure 5.6. First derivative reveals a “hidden” peak at 1640 cm'1 which is not visible in

the raw spectral data due to overlap with a peak at 1656 cm'l.

117

 

 

 

 

 

 

 

 

1660
1694
I‘ 1260
i?
O
i
" 3010
B
0 500 1000 1500 2000 2500 3000
Raman Shit! (cm‘)

Figure 5.8. Difference Raman spectra obtained for the experiment conducted at 200 °C
with no TBHQ added to the oil. Difference spectra calculated by subtracting the first
spectrum acquired once the oil reached 200 °C from A) the spectrum acquired after 180
minutes at 200 °C B) the spectrum acquired after 5 minutes at 200 0C.

118

because it did not show significant differences when compared to the data displayed on
Figure 5.8. Essentially, the addition of TBHQ did not affect the oxidation reactions for
the experiment conducted at 200 °C. Since the melting range of TBHQ is 126-128 0C it
is likely that upon its addition to the 200 °C oil, TBHQ immediately evaporated or
thermally decomposed and thus was incapable of performing its role as an antioxidant.
Most of the peaks visible in Figure 5.8 were also visible in Figure 5.6, their
assignments remain the same. However, there are some significant differences between
the two Figures that are worth discussing. One of the biggest differences between
Figures 5.6 and 5.8 is the intensity of the change in baseline linearity. Figure 5.6 A
demonstrates a much larger change than Figure 5.8 A. The explanation for this
difference is that by the time the oil had reached 200 °C most of the antioxidants had
already decomposed or degraded. This would mean that the change in concentration of
antioxidants between the first spectrum obtained at 200 °C versus the last spectrum
obtained after 180 minutes at 200 °C would be low when compared to the same change
for the experiment performed at 130 °C. Another difference between Figures 5.6 A and
5.8 A is that Figure 5.8 A does not show a peak in the 870 cm'1 region. There are two
likely explanations for this; the first is that the melting range of many organic
hydroperoxides is known to be in the 140-160 0C range. It could then be expected that at
200 °C the hydroperoxides that were formed would evaporate almost immediately. The
second likely explanation would be an increased in the reaction rate of the
hydroperoxides at the higher temperatures. Both of these explanations rely on the fact
that the concentration of hydroperoxides in the oil would be kept at very low levels by

being heated to 200 °C, making their detection with Raman spectroscopy difficult.

119

The last major difference between Figures 5.6 A and 5.8 A is the appearance of a peak in
Figure 5.8 A in the 1690 cm’1 region which is not visible in Figure 5.6 A. Previous
works’6 using ATR-FTIR observed the appearance of a band in the 1690-1715 cm”1
region under very similar experimental conditions. The authors assigned this band to the
keto functional group present in free fatty acids and also to the appearance of conjugated

keto groups, both of which are formed as products of oil oxidation.

5-4 Conclusions

It is obvious that, as the title of this chapter states, this is a continuing investigation.
Much work remains to be done before any definitive conclusions can be made about the
data presented here. As a starting point, it would be beneficial to perform the
experiments described above at least in triplicate in order to gain a better understanding
of the reproducibility (or lack there of) of the oxidation process. In addition, different
oxidation conditions (temperature, air-flow rate, humidity) should be examined along
with a variety of synthetic and natural antioxidants with the aim of defining the
concentration range for which Raman spectroscopy can be used to detect the various
oxidation products. Finally, the theories developed using Raman spectroscopy alone
must be investigated further by analyzing the oil with chemical methods (e.g. peroxide
value (PV), anisidine value (AV), etc.) and other analytical techniques such as GC, FT-IR
and perhaps HPLC. Despite the above mentioned shortcomings, the overall objective of
the present work, which was to lay the foundation for the development of Raman
methods for assessing oil quality in relation to lipid oxidation and thermal stress, was

accomplished successfully. We are aware that because of the large number of

120

compounds created during oil oxidation (many with very similar molecular structures) it
is unlikely that Raman spectroscopy (and for that matter vibrational spectroscpies
altogether) will ever be able to fully characterize and identify every individual oxidation
product. However, if we can continue to monitor spectral changes that occur as oxidation
proceeds and assign wavelengths to the more common molecular species produced, it
may be possible to create a generalized, rapid approach to monitoring and quantifying
key products associated with oxidation. This may in turn lead to the development of a
dependable, quantitative, in situ, general-purpose Raman spectroscopy based quality
control method for the food industry. The results of this study demonstrate the promising
future of using dispersive Raman spectroscopy with an NIR excitation source for the

analysis of edible oil oxidation on or off-line.

5-5 Acknowledgments
Dr. Dale LeCaptain of Central Michigan University provided the Raman spectrometer.
His contribution is appreciated. Partial support of J. A. F. was received from GAANN

and Alfred P. Sloan Foundation fellowships.

121

5-6 References

10.

ll.

12.

13.

14.

15.

16.

. Food Lipids, Chemistry, Nutrition, and Biotechnology, Akoh, C.C.; Min, D.B., Eds;

Marcel Dekker: New York, NY, 1997.
Perkins, M. J .; Radical Chemisty, Ellis Horwood: London, England, 1994.

O’Brien, R. D.; Fats and Oils, Formulating and Processing for Applications,
TECHNOMIC publications: Lancaster, PA, 1998.

Official Methods and Recommended Practices of the AOCS, 5th ed.; Firestone, D.,
Ed.; American Oil Chemist’s Society: Champaign, Illinois, 1998.

Hamilton, R. J .; Cast, J., Eds.; Spectral Properties of Lipids, Sheffield Academic
Press: Sheffield, England, 1999.

van de Voort, F .R.; Ismail, A.A.; Sedman, J .; Emo, G. J. Am. Oil Chem. Soc. 1994,
71 (3), 243-253.

Moya Moreno, M.C.M.; Mendoza Olivares, D.; Amezquita Lopez, F .J .; Peris
Martinez, V.; Bosch Reig, F. Journal of Molecular Structure. 1999, 482-483, 557-
561.

Guillen, M.D.; Cabo, N. Journal of the Science of Food and Agriculture. 2000, 80,
2028-2036.

Falcon, J .A.; Berglund, K.A. Submitted for publication: Organic Process Research &
Development, April 2004.

Sadeghi-Jorabchi, H.; Hendra, P.J.; Wilson, R.H.; Belton, P.S. J. Am. Oil Chem. Soc.
1990, 67 (8), 483-486.

Sadeghi-Jorabchi, H.; Wilson, R.H.; Belton, P.S.; Edwards-Webb, J .D.; Coxon, D.T.
Spectrochimica Acta. 1991, 47A (9/10), 1449-1458.

Johnson, G.L.; Machado, R.M.; Freidl, K.G.; Achenbach, M.L.; Clark, P.J.; Reidy,
S.K. Organic Process Research & Development. 2002, 6 (5), 637-644.

Weng, Y.M.; Weng, R.H.; Tzeng, C.Y.; Chen, W. Applied Spectroscopy. 2003, 57
(4), 413-418.

F .R.; Fateley, W.G.; Bentley, F .F.; Characteristic Raman Frequencies of Organic
Compounds, John Wiley & Sons, Inc.: New York, NY, 1974.

Battey, D.E.; Slater, J.B.; Wludyka, R.S.; Owen, H.; Pallister, D.M.; Morris, M.D.
Applied Spectroscopy. 1993, 47, 1913.

Owen, H.; Battey, D.E.; Pelletier, M.J.; Slater, J .B. Proc. SPIE. 1995, 24 (6), 260.

122

17. Food and Drug Administration, 21CFR172.185
18. Bailey, G.F; Horvat, R.J. J. Am. Oil Chem. Soc. 1972, 49 (8), 494-498.

19. Baeten, V.; Hourant, P.; Morales, M.T.; Aparacio, R. J. Agric. Food Chem. 1998, 46,
2638-2646.

123

Chapter 6

CONCLUSIONS & NEXT STEPS

This work consists of two major sections in which novel methods and applications of
Raman spectroscopy are described for the in situ monitoring of two widely used
industrial processes. The first section focused on the in situ monitoring of anti-solvent
addition crystallizations of active pharmaceutical ingredients (APIS) using Raman
spectroscopy and principal components analysis (PCA) in order to gain a better
understanding of the crystallization process. The second section aimed at using Raman
spectroscopy to monitor in situ the cis-trans isomerization of fatty acids in edible oils
during high temperature processing and to lay the foundation for the development of
Raman methods for assessing oil quality in relation to lipid oxidation and thermal stress.

In Chapter 2 Raman spectrosc0py was used for in situ monitoring of antisolvent
addition crystallization of cortisone acetate. Raman spectral features of the solvent, the
antisolvent, and the solute were monitored to measure the relative concentration of the
components. It was found that Raman spectroscopy also provides useful information
about the on-set of crystallization. Furthermore, Raman spectroscopy is probably the
only analytical technique available that allows for the in situ monitoring of the solid-state
form that crystallizes. Therefore, the present work provides a method for the in situ
monitoring of the entire crystallization process, from solution to possible polymorph
transformation after crystallization. This study demonstrated the advantages of using
Raman spectroscopy is able to provide accurate information regarding not only the
crystallization process such as ATR-FTIR and other methods, but it is also capable of

providing information on the solid-state form during crystallization and subsequent unit

124

operations such as drying. Finally, Raman spectroscopy in tandem with differential
scanning calorimetry (DSC) was used to monitor the dehydration of crystals during an
extended storage time including the effect of an impurity on the transformation rate.
These results provide further evidence that Raman spectroscopy can be very useful in
rapid screening of different solid-state forms during processing or storage.

The work presented in Chapter 2 demonstrates the usefulness (and advantages
compared with other analytical techniques such as DSC or X-ray powder diffraction) of
using Raman spectroscopy to monitor solid-state forms after crystallization in a general
sense. More in-depth understanding of the specific changes observed for cortisone
acetate during storage will require supplementation of the results presented with
additional experimental data. A logical first experiment would be the use of
therrnogravimetric analysis (TGA) or Karl-Fischer titrations in order to accurately
determine the amount of water present in the crystal lattice beginning at day zero through
the end of the storage period. These experiments would provide valuable information
regarding the evaporation rate of water from the crystals which should aid in the
development of a conclusive dehydration mechanism. Replacing water with deuterated
water as the antisolvent should provide a more solid foundation for peak assignment of
the peaks observed to appear during the storage time period (see Figure 2.9).

Further investigations into the project presented in Chapter 2 should also include
monitoring the crystallization of compounds other than cortisone acetate. These studies
would prove that the in situ monitoring technique introduced does not depend on a
particular chemical species. Regardless of the solvent, antisolvent and solute used,

Raman spectroscopy will be able to provide qualitative and quantitative information

125

about the components in solution (assuming there is no significant overlap of peaks from
the three components) before crystallization occurs and will also provide qualitative
information about the crystals once they have formed. Another benefit of investigating
different crystallizing species would be the opportunity to study compounds with vastly
different metastable regions. In particular it would be interesting to monitor the
crystallization of compounds which can sustain very high supersaturation and thus should
reach the nucleation stage at a much slower pace than cortisone acetate. High
supersaturation conditions have been proven to be conducive to the formation of pre-
crystallization solution structuresl. In situ monitoring of these pro-crystallization solution
structures could provide some insight into the crystal structure that ultimately forms. It
should be noted that the relationship between solution structures formed under high
supersaturation conditions and the final crystal structures formed has been investigated in
the past1 and these studies did not find a strong correlation between the structures.
However, those studies only focused only on dihydrogen orthophosphate compounds, and
it can be argued that different classes of compounds should behave quite differently.
Being able to correlate any pre-crystallization solution structure with the final crystal
structure would be an important breakthrough particularly for compounds that crystallize
in various polymorphic forms.

Another interesting avenue to follow would be the modification of the current approach
in order to monitor in situ other often used crystallization methods. Temperature change
crystallization would be an obvious candidate due to its wide application in the
pharmaceutical industry. Assuming the use of a cooling profile to bring about

crystallization, the overall Raman scattering intensity would increase as the temperature

126

is decreased (see Eq. (1-5) in Chapter 1). Obviously this would cause a different peak
area profile than those displayed in Figures 2.4-2.6. Despite the differences, the large
drop in Raman scattering signal that comes about due to the formation of crystals should
still be evident. Furthermore, a cooling crystallization would not require the addition of a
third component (assuming the simplest case where there is only one solvent and one
solute) which should simplify the analysis of the spectral data.

Using the method developed in Chapter 2 for in situ monitoring of anti-solvent
crystallization of cortisone acetate, the Raman spectral features of the solute and the anti-
solvent were subjected to principal components analysis (PCA) in Chapter 3. The
purpose of this analysis was to obtain information from the PCA model which was not
readily apparent from the raw spectral data. For the system utilized, PCA was able to
distinguish spectral features from the solute in solution, the solute crystals and the anti-
solvent. Furthermore, PCA was capable of detecting subtle changes in the spectral data
(as the addition of anti-solvent progressed) that could be used as a warning for the onset
of crystallization. To our knowledge this is the first reported study that combines PCA
with Raman spectroscopy for monitoring crystallizations in situ.

In Chapter 3, PCA was used solely as a qualitative tool; only the pattern of the scores
was used to monitor the progress and onset of crystallization. With the method used, the
model was recalculated with the incorporation of every new spectrum acquired. Because
the model was constantly recalculated the scores and loadings were also continually
changing. The constant change is important because the interpretation of the principal
components changes with time as the data set changes (see Figure 3.11). The first

principal component always represents the largest variation in the data set; however this

127

variation will change over time. For example, at the beginning of the experiment (before
crystallization) the larges variation in the data set is due to the dilution of the solution by
the addition of water, therefore PC 1 will likely be due to those changes. Following
crystallization the largest variation will be due to the drastic drop in signal intensity, and
PC1 will then reflect that variation. Under these circumstances, PCA is not being used to
create a model. Instead, it is essentially another way to deconvolute spectra.

A solution to this problem would be to use a training set to create a PCA model which
could then be used to “predict” real time scores as the newly acquired spectra were
introduced into the model. With this approach, the “predicted” scores values could be
used quantitatively to monitor the crystallization process by creating a feedback control
loop. The control loop would use the scores values as indicators to vary the water
addition rate. As with most models that are created the benefits that may be gained by
the construction of the predictive model might be undone by the inflexible nature of the
model. Unexpected changes such as an increase or decrease in water addition rate or a
variation in the temperature would likely cause the model to fail.

Chapters 4 and 5 concentrate on in situ monitoring of edible oil degradation and
oxidation under simulated processing conditions. In Chapter 4, a novel dispersive Raman
spectroscopic procedure for the in situ monitoring of fatty acid cis to trans isomerization
in edible oils is described. Raman spectral changes observed during the high temperature
heating of oil were correlated with cis and trans isomer concentration determined by gas
chromatography (GC). The short acquisition times, low laser power and ease of
application of dispersive Raman spectroscopy using an NIR excitation source combined

with the correlation derived from the GC data provided the capability of monitoring the

128

isomerization in situ. The method developed has potential applications for increased
process control of edible oil hydrogenation, deodorization and quality control among
other widely used processing steps.

Chapter 5 describes preliminary studies which aim at utilizing Raman spectroscopy to
in situ monitor the oxidation of thermally stressed edible oil. Spectral subtraction was
used to enhance spectral changes observed under the accelerated oxidation conditions.
The difference spectra displayed peaks which could be assigned to various functional
groups, such as hydroperoxides, conjugated C=C, trans C=C and ketones, which are
known to form as products of oil oxidation. It was also possible to qualitatively monitor
the disappearance or degradation of natural antioxidants during oxidation. Although
further work must be done in order to complete this study, the results obtained thus far
indicate that it will be possible to develop a dependable, quantitative, in situ, general-
purpose Raman spectroscopy based method to monitor and quantify key products
associated with oil oxidation.

The degradation of edible oils subjected to high temperature in open air conditions is
known to be a complicated processz’3 dominated by the oxidation of fatty acids.
The formation of many of the decomposition products is largely dependent on the exact
stress conditions applied to the oilz. For these reasons, future work on the projects
presented in Chapters 4 and 5 must include high performance liquid chromatography
(HPLC) analysis. Because the decomposition of the oils is so complex and involves the
production of such a wide variety of compoundsz’4 (fi'om polymers to volatile
compounds) it should be expected that Raman spectroscopy alone will not be able to

differentiate many of the compounds created. HPLC analysis would provide more

129

definitive information regarding the degradation products formed. In addition, it may
then be possible to correlate the HPLC and Raman spectral data. This correlation could
lead to the ability of using Raman spectroscopy to quickly analyze stressed oils for
particular decomposition products.

Once an HPLC method was in place, it would be possible to monitor oil oxidation in
situ under “normal” cooking conditions. Researcherss”6 have defined cooking conditions
of interest in order to analyze the used oil for chemical composition changes. To our
knowledge there are no reports of monitoring cooking oil in situ. One obvious obstacle
to this experiment will be the added chemical compounds that food can introduce to the
oil during the cooking process. It would be expected that fresh food, for example fiesh
potatoes, would introduce a large amount of water into the oil. The water will quickly
turn to steam in the hot oil, and it is reasonable to argue that the steam generated would
increase the rate of oxidation of the oil. Raman spectroscopy would be well suited to

monitoring in situ the rapidly changing chemical environment during frying.

130

6-1 References

1.

Cerreta, M.K.; Ph.D. Dissertation, Michigan State University, East Lansing, Michigan,
1988.

. Food Lipids, Chemistry, Nutrition, and Biotechnology, Akoh, C.C., Min, D.B., Ed.;

Marcel Dekker: New York, NY, 1997.

. O’Brien, R. D.; Fats and Oils, Formulating and Processing for Applications,

TECHNOMIC publications: Lancaster, PA, 1998.

. Hamilton, R. J .; Cast, J ., Eds.; Spectral Properties of Lipids, Sheffield Academic Press:

Sheffield, England, 1999.

. Houhoula, D.P.; Oreopoulou, V.; Tzia, C. J. Am. Oil Chem. Soc. 2002, 79 (2), 49133-

137.

. Sebedio, J-L.; Bonpunt, A.; Grandgirard, A.; Prevost, J. J. Agric. Food Chem. 1990,

38, 1862-1867.

131

APPENDIX

132

Table A1.

 

 

 

 

 

 

 

 

 

 

 

Data for Figure 2.3
Water Acetone 02:21:29 Standard
0 0

Weight A Weight /o W 9' ght % Deviation

60.08 39.74 0.18 0.04

51.11 48.40 0.49 0.05

42.22 56.93 0.85 0.03

29.72 68.31 1.97 0.07

16.43 80.53 3.04 0.06

8.04 88.49 3.47 0.10

 

133

 

 

Table A2.

Data for Figure 2.4 (water addition rate: 0.25 mL/min).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

(13:19) Peak Area 1
' 750-850 cm'
0.0 1 .67E+07
1 .0 1 .65E+07
1 .9 1 .62E'I'07
3.0 1 .58E+07
3.9 1 .55E+07
5.0 1.51 E+07
5.9 1 .48E+07
6.9 1 .45E+07
8.0 1 .43E+07
9.0 1 .40E+07
9.9 1 .38E'I'07
10.9 1.35E+07
12.0 1.3ZE+07
12.9 1.30E+07
14.0 1.28E'I'07
15.0 1 .25E+07
15.9 1.23E+07
17.0 1.22E+07
18.0 1.2OE+07
18.9 1.18E+07
20.0 1.16E+07
21 .0 1 .14E+07
21.9 1.12E+07
23.0 1.10E+07
24.0 1 .08E+07
25.0 1 .07E+07
26.0 1 .05E'I'07
27.0 1 .04E+07
28.0 1 .025+07
29.0 1.01 E+07
30.0 9.95E+06
30.9 9.81 E+06
32.0 9.69E+06

 

 

 

Time (cont) Peak Area

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

33.0 9.56E+06
33.9 9.43E+06
35.0 9.32E+06
36.0 9.20E+06
36.9 9.10E+06
38.0 9.02E+06
39.0 8.92E+06
39.9 8.81 E+O6
41 .0 8.71 E+06
42.0 8.59E+06
42.9 8.50E+06
44.0 8.40E+06
45.0 8.33E+06
45.9 8.24E+06
47.0 8.15E+06
48.0 8.06E+06
48.9 7.77E+06
50.0 6.92E+06
51 .0 5.89E+06
51 .9 5.24E+06
53.0 5.19E+06
54.0 4.96E+06
54.9 4.80E+06
56.0 4.55E+06
57.0 4.43E+06
57.9 4.39E+06
59.0 4.34E+06
60.0 4.29E+06
60.9 4.1 1 E+06
62.0 4.03E+06
63.0 4.02E+06
63.9 4.05E+06
65.0 4.13E+06
66.0 4.05E+06
66.9 4.00E+06

 

 

134

Time (cont) Peak Area

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

68.0 3.95E+06
69.0 3.83E+06
69.9 3.78E+06
71 .0 3.76E+06
72.0 3.78E+06
72.9 3.75E+06
74.0 3.69E+06
75.0 3.70E+06
75.9 3.62E+06
77.0 3.59E+06
78.0 3.61 E+06
78.9 3.61 E+06
80.0 3.58E+06
81 .0 3.54E+06
81 .9 3.47E+06
83.0 3.48E+06
84.0 3.44E+06
84.9 3.40E+06
86.0 3.43E+06
87.0 3.39E+06
87.9 3.39E+06
89.0 3.34E+06
90.0 3.30E+06
90.9 3.28E+06
92.0 3.31 E+06
93.0 3.31E+06
93.9 3.21 E+06
95.0 3.12E+06
96.0 3.12E+06
96.9 3.1 1 E+06
98.0 3.02E+06
99.0 3.06E+06

 

 

Table A3.

Data for Figure 2.4 (water addition rate: 0.50 mL/min).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Time (min.) Peak Area
750-850 cm"
1 .0 1.07E+07
2.0 1.01E+07
3.0 9.90E+06
4.0 9.76E+06
5.0 9.67E+06
6.0 9.31 E+06
7.0 9.01 E+06
8.0 8.74E+06
9.0 8.44E+06
10.0 8.24E+06
11.0 7.98E+06
12.0 7.71 E+06
13.0 7.46E+06
14.0 7.28E+06
15.0 7.12E+06
16.0 6.94E+06
17.0 6.79E+06
18.0 6.65E+06
19.0 6.50E+06
20.0 6.37E+06
21 .0 6.32E+06
22.0 6.15E+06
23.0 6.04E+06
24.0 5.93E+06
25.0 5.81 E+06

 

 

 

135

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Time (cont) Peak Area
26.0 5.69E+06
27.0 5.58E+06
28.0 4.48E+06
29.0 3.25E+06
30.0 2.99E+06
31.0 2.93E+06
32.0 2.87E+06
33.0 2.79E+06
34.0 2.75E+06
35.0 2.74E+06
36.0 2.72E+06
37.0 2.70E+06
38.0 2.69E+06
39.0 2.68E+06
40.0 2.64E+06
41.0 2.67E+06
42.0 2.63E+06
43.0 2.63E+06
44.0 2.60E+06
45.0 2.6ZE+06
46.0 2.61 E+06
47.0 2.59E+06
48.0 2.58E+06
49.0 2.59E+06
50.0 2.56E+06
51.0 2.54E+06

 

 

 

 

 

Table A4. Data for Figure 2.4 (water addition rate: 1.00 mUmin).

 

Time (min.) Peak Area

 

 

 

 

 

 

 

 

 

 

750-850 cm"
0.0 1 .18E+07
10 1.06907
2.0 9.22906
3.0 8.95E+06
4.0 6.64906
5.0 8.39906
6.0 7.97906
7.0 7.57906
8.0 7.24906
9.0 6.92906

 

10.0 6.60E+06
11 .0 6.50E+06
12.0 6.23E+06
13.0 6.01E+06
14.0 5.16E+06
15.0 3.45E+06
16.0 2.98E+06
17.0 2.83E+06
18.0 2.78E+06
19.0 2.71 E+06
20.0 2.68E+06
21 .0 2.72E+06
22.0 2.73E+06
23.0 2.69E+06
24.0 2.71E+06
25.0 2.67E+06
26.0 2.66E+06

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

136

Table A5.

Data for Figure 2.5 (water addition rate: 0.25 mL/min).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Time (min.) Peak Ar 93.1

750-850 cm

0.0 1 .70E+07
1 .0 1 .69E+07
2.0 1 .6654'07
3.0 1 .61 E+07
4.0 1 .58E+07
5.0 1 .57E+07
6.0 1 .53E+07
7.0 1 .50E+07
8.0 1.46E+07
9.0 1 .44E+07
10.0 1 .42E+07
1 1 .0 1 .41 E+07
12.0 1 .39E+07
13.0 1 345107
14.0 1.30E+07
15.0 1 .29E+07
16.0 1.27E+07
17.0 1 .24E+07
18.0 8.09E+06

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Time (cont) Peak Area
19.0 4.49E+06
20.0 3.98E+06
21.0 3.77E+06
22.0 3.72E+06
23.0 3.59E+06
24.0 3.51 E+06
25.0 3.37E+06
26.0 3.33E+06
27.0 3.24E+06
28.0 3.13E+06
29.0 3.12E+06
30.0 3.05E+06
31.0 3.01 E+06
32.0 2.95E+06
33.0 2.90E+06
34.0 2.88E+06
35.0 2.87E+06
36.0 2.88E+06
37.0 2.81 E+06
38.0 2.79E+06
39.0 2.78E+06
40.0 2.75E+06

 

 

 

 

137

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Time (cont) Peak Area
41.0 2.72E+06
42.0 2.70E+06
43.0 2.70E+06
44.0 2.69E+06
45.0 2.65E+06
46.0 2.65E+06
47.0 2.62E+06
48.0 2.65E+06
49.0 2.63E+06
50.0 2.57E+06
51.0 2.60E+06
52.0 2.55E+06
53.0 2.53E+06
54.0 2.58E+06
55.0 2.59E+06
56.0 2.56E+06
57.0 2.54E+06
58.0 2.54E+06
59.0 2.53E+06
60.0 2.53E+06
61.0 2.52E+06

 

 

 

 

 

Table A6.

Data for Figure 2.5 (water addition rate: 0.50 mUmin).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Time (min.) P9“ Ar “.1

750-850 cm

0.0 1.63E+07
1.1 1 .63E+07
2.0 1 .60E+07
3.0 1.55E'l'07
4.0 1.50E+07
5.0 1 .45E+07
6.0 1.4OE+07
7.0 1.35E+07
8.0 1.31 E+07
9.0 1 .27E+O7
10.0 1.21 E+07
1 1.0 7.09E'l’06
12.0 3.77E+06
13.0 3.17E+06
14.0 2.965'1'06
15.0 2.78E'l'06
16.0 2.66E+06
17.0 2.54E+06
18.0 2.48E+06
19.0 2.4ZE+06
20.0 2.40E+06
21 .0 2.35E+06
22.0 2.345106
23.0 2.32E+06
24.0 2.345406
25.0 2.29E+06
26.0 2.27E'l'06
27.0 2.25E'l'06
28.0 2.24E+06
29.0 2.24E+06
30.0 2.21 E+06
31 .0 2.22E+06
32.0 2.21 E+06

 

138

 

 

Table A7. Data for Figure 2.5 (water addition rate: 1.00 mL/min).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Time (min.) P9“ N 93.1

750-850 cm
0.0 1.44E+07
1.0 1 .31E+07
2.0 1.205107
3.0 1 .1GE+07
4.0 1.1 1 E+07
5.0 1 .045+07
6.0 2.31 E+06
7.0 1 .71E+06
8.0 1 .59E+06
9.0 1.51 E+06
10.0 1.49E+06
1 1 .0 1.49E+06
12.0 1.48E+06
13.0 1.4GE+06
14.0 1 .4GE+06
15.0 1.47E+06
16.0 1.48E+06

 

139

 

Table A8. Data for Figure 2.6 (water addition rate: 0.25 mL/min).

Time (cont) Peak Area

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

30.0 2.24904
31.0 2.01904
Time (min.) Peak Area.1 32.0 2.27E+04
1 '0 75193535“ 33.0 1 .98904
20 9.68904 34.0 2.33904
30 9.65904 35.0 2.21904
40 9.10904 36.0 2.25904
5.0 9 1 1904 37.0 2.24904
- 38.0 1.97904
60 8365104 39.0 2.22904
7.0 821904 40.0 2.29904
8.0 8-41 5104 41.0 2.12904
9.0 8-1 1E+04 42.0 2.05904
10.0 7.91904 430 2.26904
11.0 770904 44.0 2.31904
12.0 7.32904 45.0 214904
13.0 721904 46.0 2.11904
14.0 7.01904 47.0 2.11904
15.0 6.86E+04 43.0 2_05E+04
16.0 694904 49.0 2.24904
17.0 6.58904 500 1.97904
18.0 4.33E+04 510 2545104
19.0 2.5QE+04 520 2105904
20.0 2.45E+04 53.0 231 154-04
21.0 2.38904 54.0 219904
22.0 233904 55.0 2.1 1 904
23.0 231904 56.0 2.14904
24.0 2.17904 57.0 200304
25.0 2.58E+04 5&0 1 ,95E-1-04
26.0 2.37E+04 59.0 2,135+04
27.0 2.10E+04 60.0 23354-04
28.0 2.14E+04 61.0 225904
29.0 221904

 

 

 

 

 

140

Table A9. Data for Figure 2.6 (water addition rate: 0.50 mL/min).

 

Time (min.) Peak Area

 

 

 

 

 

 

 

 

 

 

 

750-850 cm"
1.1 8.95904
2.0 8.41E+04
3.0 7.84904
4.0 7.19904
5.0 6.87E+04
6.0 6.76904
7.0 6.10904
8.0 6.01E+04
9.0 3.72904
10.0 1.96904
11.0 1.91904

 

12.0 1.54E+04
13.0 1.43E+04
14.0 1.69E+04
15.0 1.87E+04
16.0 1.32E+04
17.0 1.58E+04
18.0 1.45E+04
19.0 1.60E+04
20.0 1.56E+04
21 .0 1 .68E+04
22.0 1 .74E+04
23.0 1 .82E+04
24.0 1 .80E+04
25.0 1 .78E+04
26.0 1 .66E+04
27.0 1 .74E+04
28.0 2.04E+04
29.0 1 .95E+04
30.0 1 .76E+04

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

141

 

 

Table A10. Data for Figure 2.6 (water addition rate:1.00 mUmin).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

(2%: Peak Area.1

750-850 cm
1.0 7.24E+04
2.0 6.69E+04
3.0 5.87E+04
4.0 5.98E+04
5.0 5.03E+04
6.0 1.28E+04
7.0 1 .01E+04
8.0 9.26E+03
9.0 1.11E+04
10.0 1.06E+04
11.0 1.25E+04
12.0 1.32E+04
13.0 1.33E+04
14.0 1.36E+04
15.0 1.37E+04
16.0 1.11E+04

 

 

142

Table A11.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Data for Figure 2.12
Days After (A) Peak Width (B) Peak Width (C) Peak Width
Crystallization 1585-1635 cm" 1585-1635 cm" 1585-1635 cm"

0 8.67 8.88 8.67

1 8.87 8.88 8.72

2 8.88 8.88 8.73

3 11.00 11.00 15.03

4 14.00 15.75 16.91

5 ----- - 17.06 16.86

6 16.49 ------ 16.86

7 17.15 16.85 16.86

8 -------- 17.03 16.86

9 17.09 16.89 16.86

10 16.89 17.18 ------

11 17.04 16.91 -----

12 17.06 16.98 -----

13 ------ 17.03 -------

14 16.92 16.88 ------

15 16.98 ------ -----

16 16.83 ------------- -

 

 

 

 

 

143

 

Table A12. Data for Figure 2.13

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Days After Peak Width 1585- Peak Width 1585-1635 Peak Width 1585-
Crystallization 1635 cm (0.04 wt% cm (0.15 wt% 1635 cm (0.22 wt%
Bisnoraldehyde) Bisnoraldehyde) Bisnoraldehyde)
0 8.61 8.72 9.05
1 8.64 8.75 8.87
2 8.66 8.75 8.97
3 8.70 8.85 9.15
4 8.78 8.76 9.17
5 9.44 8.85 9.14
6 13.49 8.93 9.27
7 16.44 10.35 9.00
8 16.95 15.09 9.06
9 17.07 16.92 9.10
10 17.03 16.86 9.30
11 -------- 16.86 10.40
12 -------- 16.86 --------
13 --------- 16.86 --------
14 ------- 16.86 -------

 

144

 

Table A13. Data for Figure 3.3

 

 

 

 

 

 

Water Acetone 023329 Standard
Weight % Weight % Weight % Deviation
49.72 49.40 0.88 0.03
40.16 57.13 2.71 0.09
24.65 68.26 7.09 0.17
16.20 72.90 10.9 0.04
8.43 76.47 15.1 0.14

 

 

 

 

 

 

145

Table A14. Data for Figure 3.9

Water Added (cont.) PC1 Scores

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

9.53 9.03E+03
“1112:1333" PC1 s°°'°s 9.80 9.06E+03
0,00 43354.04 10.08 9.30E-1-03
0.32 -2.28E+04 10.37 9.52E+03
050 -2.10E+O4 10.64 9.56E+03
0,88 -1.93E+04 10.93 1.00E+04
1.15 -1 1554.04 11.21 9.84E+O3
1.43 -1 5151.04 11.48 1.00E+04
1.71 -1.45904 "-77 9-83E+03
1.98 -1.28E+04 12.05 1.00E+04
2.27 -1.14E+04 12.33 9885-1-03
2,54 33054.03 12.61 9.89E+03
2.82 -8.52E+03
3.10 -6.75E+03
3.38 -5.70E+03
3.65 -4.25E+03
3.93 -3.64E+03
4.21 -2.25E+03
4.49 -1 .38E+03
4.77 5.13E+02
5.05 9.89E+02
5.33 1.18E+03
5.61 1 .65E+03
5.88 2.40E+03
6.17 3.06E+O3
6.44 3.81 E+03
6.73 4.73E+03
7.01 5.85E+03
7.28 6.02E+03
7.57 6.67E-1-03
7.84 7.09E+03
8.13 7.74E-1-03
8.40 7.86E+03
8.68 8.00E+03
8.96 8.54E-1-03
9.24 8.90E-1-03

 

 

 

146

 

Table A15. Data for Figure 4.4

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

-1
(m... 21:19" 33.23.13.
0 6.96E+04 1 .32E+03
2.5 7.01E+04 8.58E+02
5 7.17E+04 1.80E+03
7.5 7.09E+04 1 .19E+03
10 7.03E+04 1.94E+03
12.5 6.99E+04 1 .42E+03
15 7.09E+04 1.10E+03
17.5 7.13E+04 1.21E+03
20 7.12E+04 6.62E+02
22.5 7.12E+04 6.63E+02
25 7.15E+04 1.69E+02
27.5 6.95E+04 1 .09E+03
30 6.92 E+04 2 .88E+03
32.5 7.11E+04 1.58E+03
35 7.00E+04 9.79E+02
37.5 7.18E+04 9.53E+02
40 7.01E+04 8.54E+02

 

147

 

Table A16. Data for Figure 4.5 and 4.6

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Heating Cis Isomer 1265l1745 Average Standard
Temperature (°C) Concentration (wt%) Peak Area Ratio Deviation
No heating 88.59 1.71 0.06

200 88.24 1 .66 0.06
250 87.67 1 .59 0.09
300 80.12 1.35 0.03
Heating Trans Isomer 1745" 265 Average Standard
Temperature (°C) Concentration (wt%) Peak Area Ratio Deviation
No heating 0.94 0.59 0.02
200 1 .68 0.60 0.02
250 1 .22 0.63 0.03
300 8.48 0.74 0.02

 

 

148

 

l-;

Table A17. Data for Figure 4.7

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1745/1265 1745/1265 1745/1265
Time Average Peak Standard Average Peak Standard Average Peak Standard
(min.) Area Ratio at Deviation Area Ratio at Deviation Area Ratio at Deviation
200 °C 250 °C 300 °C
0 0.69 0.01 0.75 0.02 0.77 0.01
2.5 0.69 0.00 0.74 0.01 0.77 0.00
5 0.70 0.02 0.74 0.03 0.80 0.02
7.5 0.70 0.00 0.75 0.00 0.81 0.01
10 0.71 0.02 0.74 0.03 0.81 0.03
12.5 0.70 0.02 0.75 0.02 0.82 0.04
15 0.70 0.01 0.75 0.04 0.83 0.03
17.5 0.71 0.02 0.74 0.01 0.84 0.04
20 0.69 0.01 0.75 0.03 0.85 0.04
22.5 0.70 0.01 0.76 0.05 0.86 0.03
25 0.70 0.01 0.77 0.06 0.88 0.03
27.5 0.70 0.01 0.78 0.03 0.87 0.05
30 0.70 0.02 0.77 0.05 0.87 0.06
32.5 0.70 0.01 0.78 0.05 0.92 0.06
35 0.69 0.00 0.78 0.04 0.92 0.04
37.5 0.69 0.01 0.79 0.04 0.94 0.03
40 0.71 0.02 0.79 0.03 0.93 0.05

 

 

 

 

 

 

 

149

 

   

u111111111111131113111611