Professional Documents
Culture Documents
by
Sang-Hee Shim
Doctor of Philosophy
(Chemistry)
at the
U&IVERSITY OF WISCO&SI&-MADISO&
2008
i
Sang-Hee Shim
Two dimensional infrared (2D IR) spectroscopy has proven itself as a unique tool for probing the
structures and dynamics of chemical and biological systems. Now, 2D IR spectroscopy has
progressed to the point of being a ready-to-operate method for general researchers to address
novel scientific questions. For that purpose, this thesis includes three practices: inventing a
spectra; applying it to delineate the pathway of amyloid formation. First, we developed a means
to shape femtosecond pulses in the mid-infrared using a germanium acousto-optic modulator (Ge
AOM). The Ge AOM directly modulates the amplitude and phase of mid-IR light between 2~18
µm, producing intense shaped pulses with high resolution and good phase stability. Second, we
combined the pulse shaper with a pump-probe geometry to develop a new method of 2D IR
spectroscopy. The pump-probe geometry simplifies the optical setup and produces properly
phased absorptive lineshapes without additional data processing. The pulse shaper not only
reproduces most conventional methods, but also allows new pulse shapes and phase
accelerates data collection so that an entire 2D IR spectrum can be obtained in <1 sec. Such
ease of use, versatility and speed of our pulse shaping 2D IR method enables more sophisticated
ii
and reliable experiments. Finally, we tracked amyloid formation of human Islet amyloid
can follow the hIAPP fibril formation indefinitely by collecting data on-the-fly without re-
initiating the aggregation for each data point. We used isotope-labeling to probe the structural
evolution of six different residues along the 37-residue hIAPP peptide. By separately monitoring
the six residues, we found that the peptides nucleate near the turn of the hairpin, followed by a
propagation of the two parallel β-sheets with the N-terminal β-sheet forming twice as fast as the
C-terminal sheet. The experimental approach provides a detailed view of the assembly pathway
of hIAPP fibril as well as a general methodology for studying other amyloid forming peptides.
iii
Table of Contents
List of Figures v
Acknowledgements vii
Chapter 1 Introduction 1
1.1 References 11
List of Figures
Chapter 1
Figure 1.1 1D and 2D IR spectra of hIAPP before and after amylodosis. 8
Figure 1.2 1D and 2D IR spectra of hIAPP labeled at Ala 25 before and after amyloid
formation. 9
Chapter 2
Figure 2.1 Layout of the ultrafast laser system. 16
Figure 2.2 Layout of the home-built Kapteyn-Murnane oscillator. 21
Figure 2.3 Layout of the Spitfire-HPR Ti:Sapphaire amplifier. 25
Figure 2.4 Layout of the home-built optical parametric amplifier. 31
Figure 2.5 Dependence of mid-IR bandwith on OPA optical configuration. 32
Chapter 3
Figure 3.1 Schematic of a femtosecond pulse shaper. 39
Figure 3.2 Experimental setup of a mid-IR pulse shaper using a Ge AOM. 43
Figure 3.3 Triggering and timing scheme of the laser and the shaper. 46
Figure 3.4 Experimental setup for characterizing the shaped pulse. 48
Figure 3.5 Measurement of the shaper resolution. 51
Figure 3.6 Amplitude and position dependence on deflection efficiency. 53
Figure 3.7 Phase stability measurements. 56
Figure 3.8 Compressing the shaped pulse. 58
Figure 3.9 Pulse pairs with various spacing and relative phase. 61
Chapter 4
Figure 4.1 Schematic layouts of 2D IR experiment. 72
Figure 4.2 Experimental setup of the automated 2D IR spectroscopy 79
Figure 4.3 2D IR spectra of W(CO)6 obtained with various shapes of pump pulses. 84
Figure 4.4 Shifting signal frequency of the amide I frequency of NMA in D2O by
incrementing the relative phase. 91
Figure 4.5 Removing transient absorption background from 2D IR specta of NMA in D2O
using phase cycling. 94
Figure 4.6 Removing scatter from 2D IR spectra of hIAPP fiber. 97
Figure 4.7 2D visible spectra of atomic Ru vapour. 104
Chapter 5
Figure 5.1 Morphologies of hIAPP fibers. 118
Figure 5.2 Kinetic studies of hIAPP aggregation using FTIR and ThT fluorescence. 119
Figure 5.3 Concurrent ThT fluorescence and TEM snapshots during hIAPP amyloid
formation. 120
Figure 5.4 2D IR spectra and kinetics curves of unlabeled hIAPP during amylod
formation. 124
Figure 5.5 Structure of hIAPP fibrils obtained from solid state NMR. 126
Figure 5.6 2D IR spectra of hIAPP labeled at Ala25 during aggregation. 129
vi
Figure 5.7 Kinetics curves of hIAPP labeled at Ala25 during aggregation. 130
Figure 5.8 Reproducibility of four trials on hIAPP labeled at Ala 25. 135
Figure 5.9 Difference 2D IR spectra of mature fibers of the six labeled peptides. 139
Figure 5.10 Difference 2D IR spectra at various aggregation stages of the six labeled
peptides. 140
Figure 5.11 Kinetics data and fits of 6 labeled peptides. 142
Figure 5.12 Comparison of kinetics of 6 labeled peptides. 143
Figure 5.13 Effects of running average. 144
Figure 5.14 Diagonal slices of 2D IR spectra taken for fibers with and without diluting the
label. 146
Figure 5.15 hIAPP aggregation pathway that is consistent with our data. 151
vii
Acknowledgements
The past five years in graduate school have been the most challenging as well as the most fruitful
time of my life. Not only is Madison distant from my homeland, Korea by half way of the earth
circumference, but also learning and practicing the art of science have been a rare experience that
combines education and amusement. In each step of the period, I have been blessed to have
First of all, I can not thank enough my advisor, Professor Martin Zanni. He has been
extremely supportive, inspiring, and fun to be with. Whenever I encounter hurdles, personal or
professional, he was always with me, never running out of support and advice. He is enthusiastic
and intuitive in science, but never forgets to enjoy the fun of science. When I decided to study
abroad, my goal was to find a true role model, and I have found him.
In the Zanni group, I have been accompanied by smart, hard-working, and fun individuals.
Eric Fulmer was the very first person whom I worked with. We constructed the home-built
oscillator together and he was always patient and kind to the clueless first-year student. Amber
Krummel and Prabuddha Mukherjee were always supportive and helpful whenever I bother them
with my lack of understanding. Although we never worked on a specific project together, Terry
Ding has been a pleasant and smart character during my days in the laboratory. I worked with
David Strasfeld the most and we constructed the new laser table, pulse shaper and 2D IR
spectrometer together. Dave is not only an uplifting and humorous friend, but also an inspiring
colleague, full of scientific insights. The hIAPP folding project could have never begun without
Yun Ling’s hard work in tackling the finicky sample preparation. Once Wei Xiong joined the
group, he has often been my companion in the laser lab. His intelligence and hard-work will
viii
definitely benefit him well. Ann Woys and Emily Blanco, smart and sweet ladies of the group,
and I had great girl talks lighting up the darkness of the laboratory. Sudipta Mukherjee and Dr.
Chris Middleton were the last people with whom I worked; although short, I had great time with
them.
I was blessed to have many great collaborators. I acknowledge them in the following
chapters that we collaborated, but I thank them here as well for giving me variety of chances to
explore. I thank Professor David Blank and Dr. David Underwood at the University of
Minnesota who gave us their design of the Ti:Sapphire oscillator and helped us through the
construction. I also thank Professor Niels Damrauer and Eric Grumstrup at the University of
Colorado at Boulders for the collaboration in 2D visible spectroscopy. I never met them in
person, but through numerous emails and phone calls, we could accomplish the experiment in an
efficient way. Finally, I thank Professor Daniel Raleigh and Ruchi Gupta at the State University
of New York at Stony Brook for the collaboration in the labeled hIAPP folding experiment.
They were generous in providing the labeled peptides as well as in educating me with their vast
knowledge in amyloids.
Last, but not least, I sincerely appreciate my family. My parents have always allowed me,
a whimsical character, to find my way by myself, supporting me with all their hearts. Father, I
can no longer see you, but I do know that you are blessing me in heaven. Mother, without your
support, I am sure that I would not have finished graduate school. Brother, you are a great role
model as a scientist as well as a parent. I also thank my new family, my parents in law, who
supported me like their own daughter. Of course, I thank them for giving birth to my husband,
my other half, my companion. I so look forward to sharing the future with you.
1
Chapter 1
Introduction
The structures and dynamics of proteins have been the focus of study for many researchers to
understand the mechanism of their function. In order for a protein to function, a linear chain of
amino acids must fold into a functional three-dimensional structure, which is uniquely
determined by the sequence of its amino acids through their chemical characters. When a protein
fails to fold into the native form, it misfolds (i.e. incorrectly folds) into an inactive or even toxic
form. Some misfolded proteins form aggregates, known as amyloid, that damage cells and cause
many degenerative diseases (1, 2). Even after folding into their native state, many proteins
undergo conformational changes in response to their environments or upon binding with other
molecules to perform their biological function. For instance, an ion channel controls the flow of
ions by opening and closing its pore in response to a membrane potential or upon binding of a
ligand (3). Many experiments aim to monitor the structural evolution of proteins at sufficient
spatial and temporal resolution to capture the details of their mechanisms. A molecular movie of
amino acids interacting and assembling during protein folding would allow us to answer the
question of how the amino acid sequence translates into the native structure. Following this line
or reasoning, one way to gain insights into the misfolding and toxic activity of amyloids is to
follow the structural changes of a misfolding protein. Understanding the mechanism of ion
gating requires a means to track the structure of the pore at each step of opening and closing of
an ion channel. Therefore, studies associating structure, dynamics and function of proteins need
a method to capture structural details at a sufficient speed so that the individual steps can be
resolved.
2
One promising method for capturing protein dynamics is two-dimensional infrared (2D
IR) spectroscopy. 2D IR spectroscopy has proven itself as a useful tool to study structures and
dynamics of chemical and biological systems in condensed phases (4-6). The structural
sensitivity comes from vibrational couplings bridging multiple bonds to yield a collective
vibration. The delocalized mode reflects the spatial arrangement of the coupled bonds. In
proteins, backbones arranged in α-helix or β-sheet vibrate in one or two distinct cooperative
motions so that IR absorption from α-helix or β-sheet can be distinguished from one another (7).
To obtain more precise structural information, a local mode (i.e. one residue) can be separated by
labeling the bond with isotopes such that its frequency shifts away from the rest (8, 9). 2D IR
spectroscopy cannot obtain atomic level information like X-ray crystallography or nuclear
while NMR spectroscopy uses radio-frequency pulse sequences to study protein motions taking
obtains dynamical contents from lineshapes that reveal femtosecond dynamics, and from the
evolution of lineshapes that confers picosecond dynamics. The time scale can extend much
an evolving system. In fact, snapshots can be taken by other techniques, but each 2D IR
snapshot reflects its unique ultrafast dynamics at the very moment when the snapshot is captured.
Moreover, 2D IR spectra can be collected from virtually any kind of sample, as demonstrated by
linear IR spectroscopy, including membranes and fibers that other structure-resolving tools have
difficulty probing. These capabilities of 2D IR spectroscopy for protein dynamics have been
3
membrane peptide structure assessment (13), proton channel gating (14) and amyloid formation
(15).
Since Hamm, Lim and Hochstrasser collected the first 2D IR spectrum in 1998 (16), the
field of 2D IR spectroscopy has focused in advancing the technique and demonstrating its
capabilities. Now the field faces a new phase calling for a turnkey equipment that allows many
far, its complicated implementation has restricted the number of researchers to a few dedicated
insights on variety of topics including protein dynamics. To initiate the new phase of 2D IR
pulse shaper and applied the method to delineate the pathway of amyloid formation.
Among many possible applications, amyloid formation is of particular interest, not only
because it is well suited for demonstrating the unique capabilities of our method, but because it
will also propose potential drug targets for many human diseases including diabetes,
Altzheimer’s and Parkinson’s disease. More than 20 diseases are known to associate with
proteins misfolding into amyloid fibers (1, 2). Abnormal accumulation of amyloid fibers is
commonly observed in diseased tissues. Recently, evidence pointing to the partially folded
intermediates as cytotoxic entities has accumulated, which is steering the focus of amyloid
research from mature fibers to intermediates (17-19). Identifying the structure and defining the
evolution of such cytotoxic species is vital to design inhibitors that intervene the peptide’s
damaging action. Studies containing claims of critical intermediates have provided little
structural information, mostly because it is extremely difficult to obtain both structural and
4
NMR spectroscopy do not have sufficient time resolution to track the structural changes, not to
mention that they can seldom work with aggregating systems. Circular dichroism (CD), Fourier-
transform infrared (FT IR) and fluorescence spectroscopy have sufficient time-resolution, but
only provide a low resolution in structure (20, 21). Moreover, studies of amyloid with CD or FT
IR spectroscopy often encounter difficulties on finding a unique fit when deconvoluting the
congested spectrum into features for β-sheet, random coil and α-helix. Resolving the structural
evolution of the cytotoxic intermediates requires a technique with sufficient time and structural
modes of secondary structures in 1D IR spectra are spread over two dimensions so that each
secondary structure can be assigned to a unique spectral component. For instance, shown in Fig.
1.1 are 1D and 2D IR spectra of hIAPP (human Islet amyloid polypeptide), associated to type II
diabetes, before and after amyloid formation (15, 22). The amide I band of a peptide at
1600~1700 cm-1 can be approximated as a sum of carbonyl stretches in the backbone and is
sensitive to the secondary structure. When freshly dissolved in buffer, the peptides begin as
monomers in random coil (Fig. 1.1(a)). When carbonyls in a peptide are randomly distributed,
thus randomly coupled, the frequencies of the carbonyl stretches are broadly distributed.
Therefore, the FT IR spectrum in Fig. 1.1(c) gives a broad peak centered at 1644 cm-1; the 2D IR
signature in Fig 1.1(e) is an out-of-phase doublet elongated along the diagonal. In 2D IR spectra,
each vibrational mode produces a pair of peaks that probe two different sequences of interactions
between the mode and the light pulses; one involves only with the fundamental band; another
5
includes the overtone, thus shifts the peak along ωprobe. After the amyloid fibril mature, the
fibril from hIAPP comprises stacks of hairpins with a turn connecting two β-sheets, one of which
tails to a random coil (Fig. 1.1(b)) (23). In a β-sheet where the carbonyls form a pleated sheet in
two dimensions, two delocalized modes are allowed, antisymmetric and symmetric stretches.
The most obvious feature in the FT IR spectrum of fibers is the antisymmetric mode at 1618 cm-1
(Fig. 1.1(d)) that gives the dominant doublet in the 2D IR spectrum (Fig. 1.1(f)). Note that the
major feature at 1618 cm-1 is more prominent in the 2D spectrum than in the 1D spectrum
because 2D IR intensities are more dependent on the transition dipole strengths than 1D IR
intensities are. Moreover, the 2D IR spectrum gives a unique feature of the vertical crosspeak
along ωpump from 1618 to 1673 cm-1 that connects the two overtone peaks of β-sheet
antisymmetric and symmetric mode. The crosspeak indicates strong coupling between the two
modes, which is expected from the fact that the two modes stem from the same carbonyls
arranged in β-sheet. As seen above, it is common that spectra in one dimension suffer from
congested features because all the secondary structures absorb similar wavelengths. 2D IR
spectra give better distinction between the spectral features from different secondary structures
by highlighting features from ordered structures and by providing unique features of cross peaks
(15).
separated from the rest of the protein with 12C16O (9). Shown in Fig. 1.2 are the 2D IR spectra
taken from hIAPP labeled at Ala 25 before and after aggregation (24). The spectral features
from the unlabeled portion of the peptide (dotted box in Fig. 1.2(c-d)) are similar to the ones
6
described before (see Fig. 1.1(e-f)). New features located at ωpump=ωprobe=1570~1580 cm-1
belong to the 13C18O-label at Ala25 (solid box in Fig. 1.2(c-d)). The spectral feature of the label
before aggregation is very broad and weak (Fig. 1.2(c)) because the disordered monomers
distribute the label randomly in space (black circles in Fig. 1.2(a)). In the mature fibril
composed of parallel stacks of peptides, the isotope-labels at Ala25 form a column along the
fibril long axis (black circles in Fig. 1.2(b)). The linear alignment strongly couples the labeled
carbonyls so that the label feature narrows and intensifies in the 2D IR spectrum (solid box in
Fig. 1.2(d)). In fact, the fibril spectrum in Fig. 1.2(d) is a difference spectrum from the first and
the last spectrum of a single aggregation experiment. The subtraction highlighted the spectral
changes so that the weak feature from a single residue became noticeable. Even a crosspeak was
detected between the label at Ala 25 and the β-sheet antisymmetric stretch from the rest of the
peptide (follow the arrow in Fig. 1.2(d)), indicating the proximity of the label to the β-sheets.
Therefore, a single residue can be tracked during amyloid formation with isotope-edited 2D IR
spectroscopy.
amyloid studies. Those methods spend tens to hundreds of minutes to collect a single spectrum,
which is too long considering that amyloid formation is often complete within hours. Transient
2D IR spectroscopy can be collected with time resolution of picoseconds (12), but it requires re-
initiating the reaction hundreds of times to average a signal at each point of a 2D IR spectrum
(25). However, the transient method will not work for amyloid aggregation, which is infamous
for its poor reproducibility. This is because the kinetics of amyloid formation is extremely
sensitive to the composition and purity of the sample (26, 27), and to the conditions of initiation
and incubation (28). To track a single aggregating sample with 2D IR spectroscopy, we need a
7
rapid means of scanning time delays between multiple pulses that generates and samples the 2D
splitting one IR light into multiple beam paths. Their time delays are scanned by controlling the
path lengths with translational stages that spend a deadtime of 50-100 ms for each translation
(29). Therefore, a new means of generating pulse sequences without spending time on moving
parts will speed up the data collection so that 2D IR spectra can be collected on-the-fly while the
Recently, we developed a mid-infrared pulse shaper that can directly shape mid-IR light (30, 31).
Our shaper spends only 10 microseconds to load a new pulse shape, which is even faster than the
repetition rate of most lasers used in 2D IR spectroscopy. With this powerful tool in hand, we
developed an automated version of 2D IR spectroscopy combining the mid-IR pulse shaper with
a pump-probe geometry (22). While the pump-probe geometry simplifies the setup into two
beams, the pulse shaper generates pulse pairs with accurate phase and time delay in a speed rapid
enough to use every laser shot. When using 1 kHz laser systems, an entire 2D IR spectrum can
be collected <1 sec. Thus, the speed of the method would encourage many researchers to use 2D
IR spectroscopy in their kinetics studies. In fact, the rapid data collection allowed us to track the
amyloid formation from a single hIAPP sample with a sufficient time resolution to follow the
kinetics (15).
Chapter 3 discusses the technology we invented, mid-infrared pulse shaping. The mid-IR
pulse shaper uses germanium acousto-optic modulator (Ge AOM), a programmable diffractive
optic whose efficiencies and phases across the crystal aperture are modulated by the amplitude
8
and phase of the acoustic wave passing through the crystal (30, 31). The Ge AOM that shapes
mid-IR at 2~18 µm has efficiency and resolution on par with visible modulators. The optical
alignment and electronics configurations are discussed in detail for future reconstruction. The
Chapter 4 will describe. The method combines the mid-IR pulse shaper with a pump-probe
geometry (22). The two-beam geometry simplifies the optical setup; the programmable
generation of pulse sequence makes data collection straightforward and instantaneous. Chapter
4 details the optical and electronic configuration for the rapid-scanning method that continuously
cycles a set of waveforms assigning time delays and synchronously detects the 2D IR signal.
Various pulse shapes and phase combinations are demonstrated along with rapid data collection
to further enhance 2D IR spectroscopy in ways that only pulse shaping can accomplish.
defining the pathway of amyloid formation of hIAPP (15, 24). Site-specific isotope-labeling was
used to selectively probe six residues evenly spaced along the 37-residue peptide. Since the
fibril consists of parallel β-sheets crossing the fiber axis, isotope-labeled carbonyls form a linear
column along the fiber long axis. Thus, the kinetics curve of each residue reflects the timing of
the column formation during aggregation. By acquiring kinetics curves of the six residues
separately, the order of amyloid assembly was delineated; the aggregation nucleates around the
turn, and then propagates along the final hairpin. The experimental approach combining the
automated 2D IR spectroscopy and isotope-labeling provides not only a detailed view of the
aggregation pathway of hIAPP, but also a general methodology for studying self-assembling
proteins to develop inhibitors for disease-causing amyloids and to better design biomaterials.
9
Figure 1.1 Amyloid formation of hIAPP. The monomers in random coil (a) aggregate into
fibrils rich in β-sheets (b). 1D IR and 2D IR spectra of hIAPP are presented before (c and e) and
after aggregation (d and f). The left column shows structure and spectra of the monomers before
aggregation and the right column contains those for mature fibrils. The top row shows a
schematic of structures exemplified with four peptides. The center row and the bottom row
Figure 1.2 Amyloid formation of hIAPP labeled at Ala 25. The monomers in random coil
(a) aggregate into fibrils rich in β-sheets (b). 2D IR spectra of hIAPP labeled at Ala 25 are
presented before (c) and after aggregation (d). The left column shows structure and spectra of
the monomers before aggregation and the right column contains those for mature fibrils. The top
row shows a schematic of structures exemplified with four peptides labeled with black circles
representing the isotope-label at Ala 25. The bottom row presents the 2D IR spectra with solid
black boxes highlighting the label feature, dotted boxes enclosing the unlabeled features, and a
black arrow pointing the crosspeak between the label and the β-sheet feature from the remainder
of the peptide.
11
1.1 Reference
2. Chiti F & Dobson CM (2006) Protein Misfolding, Functional Amyloid, and Human
Disease. Annu. Rev. Biochem. 75:333-366.
3. Alberts B, Johnson A, Lewis J, Raff M, Roberts K, & Walter P (2007) Molecular Biology
of the Cell (Garland Science, New York) 5th ed. Ed.
7. Barth A & Zscherp C (2002) What vibrations tell us about proteins. Quart. Rev. Biophys.
35(4):369-430.
9. Torres J, Adams PD, & Arkin IT (2000) Use of a new label C-13=O-18 in the
determination of a structural model of phospholamban in a lipid bilayer. Spatial restraints
resolve the ambiguity arising from interpretations of mutagenesis data. J. Mol. Biol.
300(4):677-685.
10. Wuthrich K (1986.) MR of proteins and nucleic acids (John Wiley & Sons, New York).
11. Chung HS, Khalil M, Smith AW, Ganim Z, & Tokmakoff A (2005) Conformational
changes during the nanosecond-to-millisecond unfolding of ubiquitin. Proc. atl. Acad.
Sci. U.S.A. 102(3):612-617.
12. Kolano C, Helbing J, Kozinski M, Sander W, & Hamm P (2006) Watching hydrogen-
bond dynamics in a beta-turn by transient two-dimensional infrared spectroscopy. ature
444(7118):469-472.
13. Mukherjee P, Kass I, Arkin I, & Zanni MT (2006) Picosecond dynamics of a membrane
protein revealed by 2D IR. Proc. atl. Acad. Sci. U.S.A. 103(10):3528-3533.
14. Manor J, Mukherjee P, Leonov H, Zanni MT, & Arkin I Gating mechanism of the
Influenza A M2 channel revealed by 1 and 2D IR spectroscopies. submitted.
12
15. Strasfeld DB, Ling YL, Shim S-H, & Zanni MT (2008) Tracking fibril formation in
human Islet amyloid polypeptide with automated 2D-IR spectroscopy. J. Am. Chem. Soc.
130(21):6698-6699.
16. Hamm P, Lim MH, & Hochstrasser RM (1998) Structure of the amide I band of peptides
measured by femtosecond nonlinear-infrared spectroscopy. J. Phys. Chem. B
102(31):6123-6138.
17. Kayed R, Head E, Thompson JL, McIntire TM, Milton, S. C. C, C. W,, & G. GC (2003)
Common structure of soluble amyloid oligomers implies common mechanism of
pathogenesis. Science 300:486-489.
18. Kagan B (2005) Oligomers and cellular toxicity in amyloid proteins: The beta sheet
conformation and disease. (WILEY-VCH Verlag GmbH & Co., Weinheim).
19. Silveira JR, Raymond GJ, Hughson AG, Race RE, Sim VL, Hayes SF, & Caughey B
(2005) The most infectious prion protein particles. ature 437:257-261.
20. Jayasinghe SA & Langen R (2005) Lipid membranes modulate the structure of islet
amyloid polypeptide. Biochemistry 44(36):12113-12119.
21. Knight JD, Hebda JA, & Miranker AD (2006) Conserved and cooperative assembly of
membrane-bound alpha-helical states of islet amyloid polypeptide. Biochemistry
45(31):9496-9508.
22. Shim S-H, Strasfeld DB, Ling YL, & Zanni MT (2007) Automated 2D IR spectroscopy
using a mid-IR pulse shaper and application of this technology to the human islet amyloid
polypeptide. Proc. atl. Acad. Sci. U.S.A. 104(36):14197-14202.
23. Luca S, Yau WM, Leapman R, & Tycko R (2007) Peptide conformation and
supramolecular organization in amylin fibrils: Constraints from solid-state NMR.
Biochemistry 46:13505-13522.
24. Shim S-H, Gupta R, Ling YL, Strasfeld DB, Raleigh DP, & Zanni MT 2D IR
spectroscopy and isotope labeling defines the pathway of amyloid formation with residue
specific resolution. Submitted.
25. Bredenbeck J, Helbing J, Kolano C, & Hamm P (2007) Ultrafast 2D-IR Spectroscopy of
transient species. Chemphyschem 8(12):1747-1756.
26. Padrick SB & Miranker AD (2002) Islet amyloid: Phase partitioning and secondary
nucleation are central to the mechanism of fibrillogenesis. Biochemistry 41(14):4694-
4703.
28. Petkova AT, Leapman RD, Guo ZH, Yau WM, Mattson MP, & Tycko R (2005) Self-
propagating, molecular-level polymorphism in Alzheimer's beta-amyloid fibrils. Science
307(5707):262-265.
29. Nee MJ, McCanne R, Kubarych KJ, & Joffre M (2007) Two-dimensional infrared
spectroscopy detected by chirped pulse upconversion. Opt. Lett. 32(6):713-715.
30. Shim S-H, Strasfeld DB, Fulmer EC, & Zanni MT (2006) Femtosecond pulse shaping
directly in the mid-IR using acousto-optic modulation. Opt. Lett. 31(6):838-840.
31. Shim S-H, Strasfeld DB, & Zanni MT (2006) Generation and characterization of phase
and amplitude shaped femtosecond mid-IR pulses. Opt. Express 14(26):13120-13130.
14
Chapter 2
Experimental Basics
Since the early 1990s, ultrafast lasers have created revolutions in physics and chemistry by
opening new fields, including nonlinear optics and time-resolved spectroscopy (1). The term
“ultrafast” comes from their output pulse duration of femtoseconds or picoseconds (2).
Scientists have applied ultrafast lasers to monitor dynamics occurring on time scales of
reactions. For chemists, the dynamics of molecules in liquids is of particular interest because
most chemistry performed in the laboratory occurs in the liquid phase. In liquids, vibrations of
chemical groups and molecules occur in femtoseconds and relax in picoseconds. Thus, a laser
system that can investigate time scales sufficiently short to observe femto/picosecond dynamics
has been the tool of choice for many researchers in physical chemistry.
crystals gave birth to modern ultrafast lasers. Further technical advances in amplification (3) and
nonlinear crystals (4, 5) have expanded the spectrum of ultrafast lasers from soft X-rays to
Terahertz. This chapter aims for describing our way of generating femtosecond pulses in the
femtosecond lasers directly emit mid-infrared light, 800-nm light from a mod-locking laser is
15
The femtosecond pulses at 800 nm should have sufficiently high energy in 100 µJ to mJs to
pump an OPA generating mid-IR with microjoules of energy. Most mode-locking lasers cannot
produce such a high power, so a regenerative amplifier is used such that femtosecond pulses
from a mode-locking laser gain more power from a pulsed green laser with high power (tens of
watts). In our laboratory, a commercially available ultrafast laser system from Spectra-Physics
generates sub-45fs pulses at 800 nm with 2-mJ of energy. Shown in Fig. 2.1 is the ultrafast laser
system including a home-built Ti:Sapphire oscillator (88 MHz) pumped with a Millenia
amplifier pumped with a Evolution-30 Nd:YLF pulsed laser (1 kHz). The oscillator generates
femtosecond pulse trains by Kerr-lens mode-locking, and then seeds the regenerative amplifier to
gain high power via chirped pulse amplification. Since the oscillator pulses spontaneously, all
monitoring the oscillator. For instance, the Evolution-30 is triggered by a 1 kHz TTL pulse from
Figure 2.1 Layout of our ultrafast laser system. The Millenia Nd:YVO4 continuous wave
laser pump the home-built Ti:Sapphire oscillator to generate 800 nm <25fs laser pulse. The
pulse train from the oscillator is seeded to the Spitfire-HPR Ti:Sapphire regenerative amplifier,
which is pumped by the Evolution-30 Nd:YLF ns-pulsed laser. Note that the Evolution is
synchronized to the oscillator by an external trigger generated from a divider circuit that
produces a 1 kHz TTL pulse synchronized to the 88 MHz pulse train from a photodiode
monitoring the oscillator. The output of the laser system is <45 fs pulses with 2 mJ energy per
Millenia home-built
Nd:YVO4 Ti:Sapphire
CW laser CW Oscillator
532nm
2.4W
88MHz
divider
circuit
800nm, <25fs,
1kHz ~2 nJ/pulse
TTL
In most ultrafast laser systems, ultrashort pulses are spontaneously generated from a Ti:Sapphire
laser that is pumped with green continuous wave light. The Ti:Sapphire crystal exhibits the
optical Kerr effect where an intense optical field causes a variation in the refractive index of the
n = n0 + n2 I (2.1)
where n is the refractive index of the material, n0 and n2 are the zeroth and second-order term,
and I is the intensity of the incident electric field. As a pulse propagates through the crystal, the
intensity dependence of the refractive index induces the “self-focusing” of the femtosecond
pulsed mode; i.e. the peak portion of the pulse is focused more than the tail portion with lower
intensity. As a result, the peak intensity of self-focusing pulse keeps increasing as the pulse pass
through the crystal. This divergence of the beam profiles leads the pump laser to associate
preferentially to the femtosecond mode, which in turn gains higher power than the continuous
wave mode. Thus, once the laser mode-locks, the mode-lock mode remains and dominates,
when the laser is aligned properly. In practice, the laser does not start mode-locking by itself and
lases in the continuous wave mode when turned on. A “seed” noise spike, commonly produced
We built an oscillator (Fig. 2.2) based on the design of Kapteyn and Murnane which is
capable of producing pulses as short as 11 fs (2). Our design uses a short Ti:Sapphire crystal
(Saint-Gobain, 0.25% doping, 3 mm diameter) that has a path length of 4 mm. The crystal is
pumped by a Nd:YVO4 laser (Spectra-Physics Millenia), focused by a lens to focus the beam
into the crystal. The crystal sits in a copper, water cooled mount between a pair of dichroic
18
concave mirrors. The crystal, curved mirrors, and lens are mounted on translational stages
which are mounted to a baseplate predrilled at correct positions. A pair of prisms in conjunction
of a retroreflector compensates dispersion from the crystal and the cavity optics. Following is a
(1) Align the green pump beam. First, construct a periscope to rotate the polarization of the
pump beam in vertical polarization. Second, align the turning mirror for the pump beam
through two irises mounted to the baseplate. Third, place the pump lens and adjust it to
keep the pump beam on the two irises. Fourth, place the curved mirror close to the lens
and adjust it so that the pump beam passes through the center of the curved mirror.
(2) Place the Ti:Sapphire crystal. First, mount the crystal into the custom-machined, water-
cooled, copper crystal mount after greasing the crystal with thermal grease. Second,
rotate the mount so that the crystal reflects the green pump at minimum intensity (i.e. at
Brewster angle).
(3) Assemble the curved mirror away from the pump lens similarly to another curved mirror.
(4) Build a small laser cavity by placing the end mirror and output coupler as close as
possible to the curved mirrors. While pumping the crystal, align the end mirror and
output coupler for lasing by overlapping the fluorescence spots. Optimize power and
mode (round) by alternately tweaking horizontal and vertical knobs on both end mirrors,
(5) Extend the cavity gradually. Do this by moving the cavity mirrors, one at a time, while
maintaining lasing. Stop every few inches and optimize the laser power by moving the
curved mirrors and crystal, and by adjusting the end mirror. Usually the laser mode-
locks best when the extension is done asymmetrically, with the arm containing prisms
19
much longer than the other arm. The distances between optics are denoted in the Fig.
(6) Assemble the first prism. Once the arm containing the end mirror extends out of the
potential position of the first prism, replace the end mirror with a gold-coated plane
mirror, while keeping lasing. Insert the first prism by grazing a bit of the laser beam off
the apex of the prism. Rotate the prism for minimum deviation of this light. Retro-
reflect the diffracted beam from the prism using the end mirror. Once the beam is
retracting itself, translate the prism until the laser is barely lasing. Optimize the lasing
power from a cavity containing the first prism and the end mirror. Extend the cavity
further as in Step 5
(7) Assemble the second prism in the same way in Step 6. Then, extend the cavity as in
Step 5 until the cavity length becomes as designed. Once the two prisms are placed,
place a card in front of the gold mirrors paired with each prism to prevent additional
lasing. These gold mirrors are used only when the laser stop lasing and need to shorten
(8) Pick up the continuous wave laser power. Adjust the end mirror, the output coupler, the
lens and the curved mirrors to get round continuous wave mode and maximum power.
(9) Now the laser is ready to mode-lock. To detect the mode-locking, we use a spectrometer
and a fast photodiode. When the cavity lases in continuous wave, a very narrow
photodiode. When the laser mode-locks, the spectrum broadens and the photo diode
detects pulse trains. First, align the laser for maximum continuous wave power and for a
20
round TEM00 mode. Translate the prisms so that the beam is within 1-2mm from
edge. Then move the curved mirror away from the pump lens towards the crystal until
the output beam looks oval, vertically. When the cavity is ready for mode-locking, the
spectrum will get “jumpy” and the photodiode signal will become noisy. Pushing the
translational stage mounting the curved mirror away from the pump lens should cause
(10) Once the laser mode-locks, translate the curved mirror until the laser mode-locks easily.
Adjust the translation of the two prisms (varying the amount of glass) for a desired
center frequency and bandwidth. Then, tweak up the mode-locked power with the end
mirrors. When aligned properly, the mode-locked laser has perfectly round mode and
more power than in continuous wave mode (up to 50% more), while the continuous
(11) Often, there is some continuous wave light co-lasing with the mode-locked light, which
produces a sharp and tall spike in the spectrum. The best way to get rid of this is to turn
down the pump power. You can suppress the continuous wave by using an aperture
medium, a continuous wave pump source, curved mirrors to pass the pump light (532 nm) and to
reflect the emitted light (800 nm), a high-reflective end mirror, an output coupler with 12%
TM
Periscope Nd:YVO4
Curved TM Pump Laser
beam Mirror
block Curved
Mirror Pump 2.4W
Ti:Sapphire Lens
CW
800 nm
15.3o
<25 fs
58.8 cm 15.3o TM
88MHz Prism
Output 30 cm Au
Coupler
58.3 cm
Prism
Au 12.7 cm
End
Mirror
Optics
End Mirror plane, HR (0º, >99.9%) coated for 710-880 nm (Layertec #101241)
Ti:Sapphire oscillators produce a few nJ per pulse, which is not enough to pump an OPA to
amplify the pulses using Spitfire-HPR from Spectra-Physics that includes a stretcher, a
regenerative amplifier, and a compressor in conjunction with a high power pump laser at 532 nm
running at 1 kHz as well as fast electronics for accurate timing the amplification (3). The
Spitfire-HPR uses Ti:Sapphire crystal as an amplifying medium due to its high saturation
fluences and a large gain-bandwidth. Since intense beams can cause the crystal to self-focus the
beam destructively or to get damaged, it is necessary to limit the intensity present in amplifiers.
The Spitfire-HPR overcomes this obstacle by using chirped pulse amplification. Briefly, a
femtosecond pulse from an oscillator is stretched to reduce its peak power significantly. This
stretched pulse with low peak power is then amplified, so that damage of the laser rod is much
less likely to happen. Following amplification, the pulse is recompressed to near its original
duration. The Spitfire-HPR (shown in Fig. 2.3) stretches the femtosecond seed pulses by as
much as 10,000 times and amplifies the pulse energy up to 106 times.
Pulse stretching and compression use diffraction grating (or prism as in the oscillator) to
disperse the incident beam so that different frequencies travel through different path lengths.
The grating and retroreflectors can be configured such that the bluer frequency components have
to travel longer distance through stretcher than the redder ones. Thus, the redder the frequency,
the earlier it exits the stretcher, and vice versa. Therefore, the pulse is stretched in time with
redder frequencies traveling in the front and bluer frequencies retarded in the back. The
stretcher in Spitfire-HPR (Fig. 2.3) uses a single grating and a multi-passed beam configuration
that the incident beam passes the grating four times to reconstruct the stretched beam spatially.
23
Pulse compressor, in essence, reverses what pulse stretcher does. The gratings and mirrors are
arranged so that the bluer frequencies travel the shorter path to catch up with the redder
The stretched seed is sent to the regenerative amplifier cavity to gain high energy up to
106 times in Spitfire-HPR. In our system, 2 nJ of seed is amplified to 2 mJ. In principle, the
regenerative amplifier controls polarization to confine a single pulse out of seed pulse train, to
amplify the pulse to a high energy level, and then to dump the output from the cavity. The
amplification take place in the laser rod pumped by a high power ns-pulse from an Nd:YLF laser
(23 W, for us). A single pass through the rod only gives a factor of 3-4 amplification; however,
the regenerative amplifier allows the pulse pass through the crystal multiple times to gain much
higher amplification. The Spitfire-HPR use a pair of Pockels cells, which are programmable λ/4
waveplates, to control the multi-pass. When both of the Pockels cells (PC#1 and #2 in Fig. 2.3)
are off, the seed oscillates only once and gets little amplification. The amplification process
begins as the PC#1 is turned on, following the activation of the Q-switch on the pump laser. A
λ/4 waveplate placed near PC#1 orthogonally rotates the polarization of the double-passing beam
(i.e. λ/2 rotation). Thus, when the PC#1 is off, the beam transmits the resonator optics.
However, when the PC#1 is activated to serve as additional λ/4 waveplate, it negates the effect
of the neighboring λ/4 waveplate, trapping the pulse in the resonator. After the pulse makes a
number of round trips (usually ~20), the PC#2 is activated so that the pulse polarization is
orthogonally rotated after double-passing the Pockels cell. Then, a thin-film polarizer placed
after PC#2 expels the multi-passed pulse from the resonator. During the trapping period in the
resonator, the seed pulse multi-passes the rod with a gain of 106.
24
Correct timing of switching the Pockel cells is essential for the regenerative amplifier
guarantee admission of a single pulse to the resonator, the Pockels cell must be switched
synchronously to the mode-locked pulse train. For this aim, a photodiode monitors the pulse
train out of the mode-locking oscillator to generate a reference signal. In addition, the switching
phase should be fine-tuned to prevent additional output pulse. The Spitfire-HPR provides the
“Synchronization and Delay Generator” (SDG) module containing the synchronizing and phase
adjusting electronics. The SDG module is triggered by a TTL pulse that is provided from the
pump laser, Evolution-30, which is synchronized to the oscillator by a 1 kHz reference from a
divider circuit that divides the 88 MHz reference by 88,000 times. The SDG produces trigger
pulses to control two Pockel cells with adjustable delays. To eject the amplified output after
sufficient round trips, the second Pockel cell is activated approximately 200 ns after the first
Figure 2.3 Layout of the Spitfire-HPR Ti:Sapphaire amplifier with pulse stretcher and
compressor.
Seed
Compressor
Output
Stretcher
PC#2
λ/4 PC#1
Pump
Regenerative Amplifier
Optics
PC#_
Pockels cell
26
An optical parametric amplifier (OPA) transfers the Spitfire output at 800 nm to the mid-infrared.
The frequency conversion is accomplished in two steps: optical parametric generation using β-
barium borate (BBO), followed by difference frequency mixing with AgGaS2 crystal, both of
The χ(1) is the first-order susceptibility describing linear optics; The χ(2), χ(3) … are the second
and third-order susceptibilities that account for nonlinear optical effects (4). With a sufficiently
weak illumination, the material linearly responds to the electric field. The nonlinearity is
observed when the electric field is sufficiently intense to break the linear assumption, mostly
when ultrashort pulses are illuminated. Note that the optical Kerr effect described in Eq. 2.1 is a
nonlinear response. Here, the second-order susceptibility, χ(2), accounts for optical parametric
Materials with large χ(2) are used to generate new frequencies via parametric effects
involving three photons. The three photons obey energy conservation by linearly combining the
three frequencies as well as momentum conservation by linearly combining the three wave
27
vectors. Let us assume that the three photons have frequencies of ω1, ω2 and ω3 with an order
of ω1< ω2< ω3 and have wavevectors k1, k2, k3, respectively. Then,
ω3 = ω1 + ω 2 (2.3),
k 3 = k1 + k 2 (2.4),
Nonlinear crystals are cut and arranged to satisfy phase-matching conditions for a desired
scheme of frequency conversion (5). Most of such nonlinear crystals are birefringent material,
whose indices of refraction depend on the polarization and direction of the light passing through.
Among three axes of the birefringent crystal, one axis has different refractive index, thus is
called as “extraordinary” axis; the other two axes are called “ordinary” axis. The polarization of
the lights and the crystal orientation are designed to obey the phase-matching condition. Among
several schemes of choosing polarizations, type I and type II are most common. In type I phase-
matching, the lights at frequencies of ω1 and ω2 (i.e. two lower frequencies) have the same
polarized. Once a crystal is chosen, the angle between the incident lights and the axes of the
crystal determines the frequencies of outputs from nonlinear crystals by fulfilling phase matching
conditions. Thus, the frequency conversion via nonlinear crystals are tunable by rotating the
crystal.
28
Of many parametric effects, our OPA utilizes optical parametric generation and difference
frequency mixing. The BBO crystal generates a signal and an idler light in near-infrared from
the pump pulse at 800 nm via optical parametric generation obeying the phase matching
conditions of
From the signal and idler lights are generated, the following AgGaS2 crystal produces mid-
Note that the wave vector depends on the refractive index of the crystal, i.e.
n(ω )ω
k= . (2.9)
c
Then, a nonlinear crystal can be designed from the mathematical representation of the refractive
indices for extraordinary and ordinary axes of the crystal regarding the frequencies and the
polarizations of the incident and outgoing beams, as well as a desired angle of the crystal.
We chose the cutting angles and types of BBO and AgGaS2 from the frequencies and
polarizations of the given pump at 800 nm and the desired mid-IR at 5 or 6 µm. Also, the crystal
orientation was set to be normal to the table so that the path length of the incident beam through
the crystal, thus material dispersion, kept minimal. Once the OPA built, the mid-IR frequency is
29
tuned by rotating AgGaS2 crystal. Also, the signal and idler frequencies mixed at AgGaS2 are
The layout of our optical parametric amplifier (OPA) is shown in Fig. 2.4. The design of the
OPA is similar to others (8, 9). The 2 mJ 45 fs transform limited pulses from the Spitfire is
equally divided by two with a 50% beamsplitter to pump two OPAs. Two OPAs have similar
layout and each are pumped with 1 mJ of 800 nm pulses. Downconversion of the 800 nm into
signal and idler pulses (140 µJ) takes place in a type II BBO (θ=25.9°) crystal in two stages with
collinear alignment. Mid-IR pulses are generated by difference frequency mixing the signal and
idler beams into a type II AgGaS2, cut at θ=50.4°. The signal and idler beams are 2 mm dia.
before focusing with f=30 cm spherical mirrors. The generated mid-IR beam is collimated with
a spherical gold mirror. The residual signal and idler are removed by a 1-mm-thick Ge filter,
which is also used to overlap a collinear HeNe alignment beam on top of the mid-IR beam path.
amounts of dispersion occurs when pulses of <50 fs duration pass through even a 1-mm thick
material. Therefore, the thicknesses of the non-linear crystals must be optimized for bandwidth
and power. Fig. 2.5 shows how the mid-infrared spectrum depends upon the components
constituting our OPA. Starting with a 4 mm BBO and a 1.5 mm AgGaS2 crystal, which are
typical thicknesses in mid-IR OPAs (9), the OPA produced mid-IR pulses with a bandwidth of
200 nm and ~4 µJ of energy (Fig. 2.5, solid thick). Since the cube polarizer (14 mm thick) that
controls the white light generation stretches the 800 nm pulses from 50 to 77 fs, we began by
removing the cube polarizer. This change increased the bandwidth to ~300 nm (Fig. 2.5, dashed)
30
without altering the energy of the mid-IR (the white light continuum was instead optimized by
translating the sapphire crystal along the beam focus). Turning to the crystals, we changed the
1.5 mm to a 1.0 mm thick AgGaS2 crystal, which brought the bandwidth to 400 nm, again
without power reduction (Fig. 2.5, solid thin). Replacing a 4 mm-thick BBO with a 2 mm BBO
(Fig. 2.5, dotted) did not alter the mid-IR spectrum, but the shot-to-shot stability of the mid-IR
improved and the power increased (~6 µJ). The shot-to-shot stability improved because the
signal and idler were no longer spontaneously generated without the white light seed present.
Finally, a 0.5 mm AgGaS2 dramatically increased the bandwidth to 850 nm (Fig. 2.5, dash-dot-
dashed), with only a slight sacrifice of energy (~5 µJ). Thus, the largest improvements are made
by proper choice of AgGaS2 crystals, which is consistent with the phase-matching bandwidths of
AgGaS2 increasing inversely to crystal thickness (570, 850 and 1700 nm for crystals with
thicknesses of 1.5, 1.0, and 0.5 mm, respectively). In the end, with proper selection of crystals
and optics, the bandwidth could be nearly quadrupled without significant loss of intensity,
Figure 2.4 Layout of the home-built optical parametric amplifier based on β-barium borate
Regenerative SM AgGaS 2
MD type II
Amplifier f=30cm
SM 0.5mm
800 nm 1450 Au f=15cm
f=30cm θ=50.4 o
45 fs LWP
SM
1kHz Au
1mJ Ge filter
HR HR
BS Au mid-IR
R=90% Au Au f=25cm L4 HR
MD
HR MD
Cube
λ/2 Pol HR MD
BS L5
R=2% Au MD MD
f=-5cm
MD
Optics
MD manual delay
32
Figure 2.5 Dependence of mid-IR bandwith on OPA optical configuration. (solid thick) 4
mm BBO and 1.5 mm AgGaS2; (dashed) cube polarizer is removed; (solid thin) 1.0 mm
AgGaS2; (dotted) 2 mm BBO; (dash-dot-dashed) 0.5 mm AgGaS2. The abrupt drop in intensity
4000 4200 4400 4600 4800 5000 5200 5400 5600 5800
wavelength (nm)
33
2.4 Acknowledgments
We thank Professor David Blank and Dr. David Underwood for providing the design of the
2.5 References
2. Asaki MT, Huang CP, Garvey D, Zhou JP, Kapteyn HC, & Murnane MM (1993)
Generation of 11-fs pulses from a self-mode-locked Ti-Sapphire laser. Opt. Lett.
18(12):977-979.
3. Backus S, Durfee CG, Murnane MM, & Kapteyn HC (1998) High power ultrafast lasers.
Rev. Sci. Instrum. 69(3):1207-1223.
4. Boyd RW (2003) onlinear Optics (Academic Press, New York) 2nd ed. Ed.
5. Dmitriev VG, Gurzadyan GG, & Nikogosyan DN (1999) Handbook of onlinear Optical
Crystals (Springer, New York) 3rd ed. Ed.
6. Siegman AE (1986) Lasers (University Science Books, Sausalito, CA) 1st ed. Ed.
8. Kaindl RA, Wurm M, Reimann K, Hamm P, Weiner AM, & Woerner M (2000)
Generation, shaping, and characterization of intense femtosecond pulses tunable from 3
to 20 mu m. J. Opt. Soc. Am. B 17(12):2086-2094.
9. Hamm P, Kaindl RA, & Stenger J (2000) Noise suppression in femtosecond mid-infrared
light sources. Opt. Lett. 25(24):1798-1800.
34
Chapter 3
Femtosecond Pulse Shaping Directly in the Mid-Infrared
Using Acousto-Optic Modulation
3.1 Introduction
Two-dimensional infrared (2D IR) spectroscopy is becoming an increasingly important tool for
studying the structures, environments and dynamics of molecules (1, 2). Collecting 2D or higher
dimensional spectra requires sequences of femtosecond and/or picosecond mid-IR pulses whose
time delays and phases can be specified. Currently, these pulse trains are generated using
beamsplitters where the delays are controlled with translation stages (3-5). This conventional
sophisticated control over the pulse frequencies and phases. For example, recent simulations
have shown that the 2D IR cross peaks can be enhanced using pulse trains composed of pulses
with sophisticated chirps (7). Another example is work done in the near-IR, where 2D electronic
spectra were collected by cycling the phases of the pulses rather than using phase matching, in
analogy to the way that 2D NMR spectra are often recorded (8). Thus, mid-IR pulse shaping
In the visible and near-IR spectral regions it is very common to use pulse shaping to
create trains of pulses or pulses with complicated chirps. These wavelength regions access
excited electronic states and can be used to optimize processes such as fluorescence, ionization
and photodissociation (9-11). Most experiments have utilized these wavelength regions the most
35
because there exist a number of commercially available devices for shaping visible and near-
IR light, including liquid crystal modulators (12, 13), TeO2 acoustooptic modulators (14), and
deformable mirrors (15). However, these devices do not work in the mid-IR. Liquid crystal
modulators and TeO2 absorb wavelengths longer than 1.5 microns while deformable mirrors
have insufficient translation for a π-phase shift beyond λ=900 nm. Shaped mid-IR pulses have
been created indirectly, by shaping in the near-IR using a commercial shaper and transferring
that shape to the mid-IR via difference frequency mixing (16-19). However, the nonlinearity of
difference frequency mixing results in poor resolution and mid-IR intensities that strongly
depend on the pulse shape and phase. As a result, indirect shaping methods have limited utility
in 2D IR spectroscopy. A better way is to shape directly in the mid-IR. In this manner, intense
mid-IR pulses can first be generated from transform limited near-IR pulses and then shaped with
a resolution and efficiency dictated only by the design of the pulse shaper. Based on
groundbreaking work in TeO2 acousto-optic pulse shaping by Warren and co-workers (14), our
shaper uses germanium acousto-optic modulator (Ge AOM) directly modulates the phase and
amplitude of femtosecond mid-IR pulses, providing superior resolution and efficiency (20).
In this chapter, we report a pulse shaper built from a Ge AOM that works directly in the
mid-IR with a resolution and efficiency on par with visible pulse shapers. The high resolution
and efficiency enable our mid-IR shaper to be used for coherent control over ground states (21).
Unlike mid-IR generation via DFM, the mid-IR shaped by our Ge AOM is phase stable with
respect to the input mid-IR. As a result, these shapers can be used in experiments requiring high
Femtosecond pulse shapers modulate the amplitude, the phase, or the polarization of a light pulse
in femtosecond duration. The electric field of a light pulse propagates at position r in time t in a
generalized form of
where ŷ is the unit vector defining the direction of E; E0 is the amplitude; ω0 is the center
frequency; φ is the phase; k is the wave vector and orthogonal to ŷ . A femtosecond pulse
shapers tailors E0, φ, ŷ or a combination of the three. Polarization shapers, shaping are
rather new and most commercial pulse shapers control the amplitude and/or the phase.
In practice, most of electronic devices cannot shape light pulses in femtoseconds directly
in the time domain. Therefore, the incident electric field Ein (t ) is Fourier-transformed to give
~
Ein (ω ) , then modulated by a mask M (ω ) in frequency domain (22). This results in an outgoing
~
shaped spectral electric field Eout (ω ) :
~ ~
Eout (ω ) = Ein (ω )M (ω ) . (3.2)
~
The mask may modulate the spectral amplitude A(ω ) and the phase φ (ω ) , i.e.,
~
M (ω ) = A(ω ) exp[iφ (ω )] . (3.3)
~
Finally, Eout (ω ) is inversely Fourier-transformed to yield Eout (t ) in the time domain.
37
Fig. 3.1(a) shows a basic pulse shaping apparatus, which consists of a spatial modulator that
serves as a mask and an optical configuration that Fourier-transforms from the time domain into
the frequency domain and vice versa (22). The spatial modulator is placed at the “Fourier plane”
inside a “4-f geometry” composed of two pairs of a grating and a lens spaced by f, the focal
length of the lens. The first pair of a grating and a lens disperses the incident pulse into
frequency, thus into space. Each frequency component of the incident beam is focused at the
focal plane of the lens (Fourier plane) and spatially separated along one dimension. The spatial
modulator located at this Fourier plain manipulates the amplitude and/or phase of the spatially
dispersed frequency elements at a computer’s command. Then, the second pair of a lens and a
grating − a mirror image of the first pair − transforms the pulses back to the time domain.
For instance, let us consider a case of generating a pulse pair: one pulse is a replica of the
incident pulse and another pulse is a copy of the incident pulse shifted forward in time by τ. Any
multiplied with exp(iωτ), then inverse Fourier-transformed. Thus, a mask function for a pulse
1 iωτ
M (ω ) = (e + 1) (3.4)
2
where 1 generates another at t = 0 and exp(iωτ) generates the pulse at t = τ. Eq. 3.4 can be
rearranged to
generate a pulse pair composed of a fixed pulse and a scanning pulse with a time delay of τ. Fig.
3.1(b) displays how a pulse shaper converts one pulse into two pulses. The first pair of a grating
and a lens Fourier-transforms a short femtosecond pulse (gray dotted) into frequency domain. At
the focal plane, the spectrum of the incident pulse (gray solid) is multiplied by a sigmoidal
amplitude mask (thick, solid) and a linear phase mask (thick, dashed). After the second pair of a
lens and a grating transfers the resultant spectrum with fringes into time domain, the shaped
beam becomes a pair of pulses whose time delay is set by the periods of fringes in the spectrum.
39
Figure 3.1 (a) A schematic of a femtosecond pulse shaper. (b) An example how a pulse
(a)
f f f f
t ω ω ω t
40
3.2 Experimental
Our pulse shaper is built from a germanium acousto-optic modulator (Isomet) that works directly
in the mid-IR between 2 and 18 µm (23, 24). Our design is based largely on work by Warren and
co-workers who pioneered AOM-based pulse shaping in the visible (14, 25). An AOM is a
programmable transmission grating whose efficiency and phase are modulated by an acoustic
wave pass through the AOM crystal. Fig. 3.2 details the setup of our pulse shaper. It follows the
general layout shown in Fig. 3.1 with three variations: First, we used cylindrical mirrors (f = 125
mm) instead of lenses to minimize material dispersion and aberration and to focus in one
dimension along the aperture of the modulator; Second, the Ge AOM is oriented at the Bragg
angle of ~2º to allow the diffracted beam to go through the remainder of the 4-f setup; Third, the
beam paths are separated in vertical dimension instead of the horizontal dimension. Gratings are
placed in quasi-Littrow configuration, where the incoming and the outgoing beams are the same
To align the shaper, we utilized a collinear HeNe beam overlapped on top of the mid-IR
beam path. ~55 fs mid-IR pulses (1.2x transform limited) are generated by DFM of the signal
and idler beams of a BBO optical parametric amplifier (OPA) in a AgGaS2 crystal (26). The
residual signal and idler are removed by a 1-mm-thick Ge filter, which is also used to overlap a
collinear HeNe alignment beam on top of the mid-IR beam path. The quasi-Littrow
configuration allows easy adaptability to multiple light sources such as a HeNe and mid-IR (27).
(1) Before inserting the Ge AOM, the 4-f geometry of the cylindrical mirrors and gratings
is initially set using three parallel propagating HeNe beams diffracted from the first
grating deflects the incoming beam downward and a folding mirror is placed in a lower
height to allow the incoming beam to travel above the folding mirror. Build the first 2-f
geometry for The 7, 8 and 9th orders to have collimated beam path and horizontally
parallel spot shape by fine-tuning the tilt of the grating, the tilt of the cylindrical mirror
and the distance between the grating and the cylindrical mirror. Then, the second 2-f
geometry is constructed as a mirror image of the first 2-f. The second cylindrical mirror
reflects upward for the beam to pass above the second folding mirror, and then the
grating parallelizes the downward beam direction. Again, the second 2-f geometry is
completed with the tilt and the translation of the second pair of the grating and the
cylindrical mirror so that the outgoing beam from the shaper is collimated and parallel to
the ground.
(2) Let the mid-IR pass through the shaper, check the resultant spectrum and compare it with
the incoming spectrum. Block a small portion of each optic, check if any portion of the
mid-IR is clipped off by the optic, and then attune the grating angles and the vertical
alignment accordingly.
(3) Immediately before placing the AOM, the HeNe beams are horizontally tilted against the
expected deflection of 2º by the first folding mirror. Place the AOM at the Fourier plane,
and then tilt the AOM to maximize the diffracted mid-IR power outside the shaper. The
chopping the acoustic wave propagating the AOM, so that the repetition rate of the
42
shaped beam is 0.5 kHz. The AOM is designed to operate with an acoustic wave at a
75 MHz center frequency and a 50 MHz bandwidth. The acoustic wave is generated by a
propagates along the 5.5-cm length of the AOM at 5.5 mm/µs, giving a time aperture of
10 µm. The trigger and timing of the acoustic wave will be discussed later in Sec. 3.3.2.
(4) The output power of the shaper is further tuned up by tweaking the folding mirrors before
and after the AOM as well as putting a waveplate in the incoming beam and rotating for
the optimal efficiency of the gratings and the Ge AOM. The Ge AOM has a deflection
conjunction with two gratings of 80% efficiency. We achieved 30% efficiency because
(5) The 2º deflection by the Ge AOM adds a linear chirp to the pulse. The chirp is
compensated by translating the second grating while maximizing the second harmonic
signal (SHG) of the shaped beam through a 0.5-mm thick doubling AgGaS2 cyrstal
(θ=33º). The leftover linear chirp and higher-order chirps are further compensated by
adding chirps to the phase of the acoustic wave that maximizes the SHG.
43
Figure 3.2 The experimental setup of a mid-IR pulse shaper using a Ge acousto-optic
modulator. The angles of the incident and the outgoing beams for the grating are almost
identical (quasi-Littrow configuration), but the beam paths are separated in height: The incoming
beam for the shaper travels above the folding mirror (dashed line); The grating deflects the beam
downward (dotted line); The cylindrical mirror compensate the vertical tilt to parallelize the
beam to the optical table (solid); After Ge AOM, the optics are configured to a mirror image of
Cylindrical Cylindrical
Mirror Mirror
Ge
AOM
Folding Folding
Mirror Mirror
Arbitrary
Grating Waveform Grating
Generator
Optics
Since the acoustic wave appears static on the timescale of the ultrafast pulse traversing the Ge
AOM, the acoustic wave acts as a modulated grating, deflecting desired frequencies with
amplitude and phase specified by the acoustic wave. As a result, the phase of the shaped mid-IR
pulses is set by the phase of the acoustic wave. Therefore, for pulses with reproducible phase
from one laser shot to the next, it is imperative that the arbitrary waveform generator is
Fig. 3.3 details the synchronizing scheme. A photodiode monitors the repetition rate of
the Ti:sapphire oscillator as a reference for all the electronics afterwards. A custom-made
divider circuit uses the 88 MHz reference to generate a 1 kHz trigger pulse for both the arbitrary
waveform generator and the amplifying Nd:YLF pump laser. A 300 MHz clock signal is also
generated from the 88 MHz reference wave using a phase-locked loop circuit (also custom-made)
that serves as an external clock for the arbitrary waveform generator. This electronic
configuration produces clock and trigger pulses synchronized to within 0.6 ns, which is
substantially smaller than the periods of the 75 MHz acoustic wave (13.3 ns) and the 300 MHz
clock (3.3 ns). A 0.6 ns electronic jitter results in the theoretical phase stability of λ/50. Note
that the phase fluctuation cannot be reduced less than λ/4 when using the 300MHz “internal”
clock of the arbitrary waveform generator, which is not synchronized to the laser.
To time the bursts of acoustic wave to the incoming mid-IR pulses at the Fourier plane, a
delay generator defers the RF waveform right before the piezoelectric transducer. In our setup,
the electronic controller for the regenerative amplifier spends less time in response to the trigger
than the arbitrary waveform generator does. As a result, the acoustic wave arrives to the AOM
after the light pulse pass through the AOM. So we set the delay slightly less than 1 ms so that
45
the acoustic wave burst encounters the next laser shot. The delay is further attuned by using a
short waveform burst centered within the 10-µs time aperture. While checking the deflected
spectrum with a monochromator, the delay is adjusted to place the narrow spectrum at the center
Figure 3.3 Triggering and timing scheme of the laser and the shaper. The basis of the
timing sequence is the oscillator. A photodiode monitors the pulse train from a Ti:Sapphire
oscillator running at 88 MHz. The 88 MHz reference is sent to a divider and a phase-locked loop
that provides 1 kHz trigger and 300 MHz clock, respectively, for the arbitrary waveform
electronically chopped to run 0.5 kHz. The RF waveform is delayed to locate the 10-µs time
aperture on top of the mid-IR pulse at the AOM. The delayed RF waveform is transformed into
11.4 ns
Ti:Sapphire
88 MHz Oscillator
1 ms 1 ms
1 ms
Divider
Circuit 1 kHz
Pulse AOM
Shaper
Trigger 75 MHz
13.3 ns
Arbitrary
3.3 ns Waveform
Generator Delay Piezoelectric
Phase-locked 2 ms Generator Transducer
Loop Circuit Clock 0.5 kHz
300 MHz
47
Fig. 3.3 shows optical devices for characterizing the shape of the pulse. A monochromator
equipped with a MCT (mercury cadmium telluride) array measures the spectrum in frequency
domain, while an auto- and a cross-correlator retrieve the amplitude and phase of the pulse in
time domain. The monochromator utilizes a focal length of 150 mm, giving a resolution of 3.6
nm. When characterizing the spectrum, we usually use a single pixel out of 64 pixels in the
MCD array detector to avoid the pixel-to-pixel variation. The auto- and cross-correlators are
collinear interferometers whose time delays are controlled with computer-controlled translational
stages outfitted with ZnSe wedges that afford a 0.02 fs time resolution and a maximum delay of
shaped beam focused into a 0.5-mm thick doubling AgGaS2 crystal (θ=33º) by an off-axis
parabolic mirror (f=50.8mm). A short-pass filter with a cut-off wavelength of 3.5 µm removes
the un-doubled mid-IR at 5~6 µm, while allowing the doubled mid-IR at 2.5~3 µm to pass
through. In the crosscorrelator, the shaped pulse is heterodyned by a small portion (2~3%) of the
Figure 3.4 The experimental setup for characterizing the shaped pulse. A monochromator
measures the spectrum, while the autocorrelator or the crosscorrelator retrieve the amplitude and
pulse shaper
mid-IR G
CaF2 CM
AOM AWG
CM
G
FM
FM
DET
FM CD
MCT
BS array
DET
crosscorrelator monochromator
MD
CD
BS
SF
DET
SHG autocorrelator
Optics
2D IR spectroscopy or coherent control experiments necessitates intense and phase stable pulses
with accurate pulse shape. In this section, we demonstrate these capabilities by resolving
3.3.1 Resolution
Our shaper is based on Ge AOM, which serves as a programmable diffractive grating whose
phase and efficiency is controlled by shaping the phase and the amplitude of the acoustic wave
passing through the Ge crystal. A 75-MHz acoustic wave with a bandwidth of ∆f = 50 MHz
travels the 5.5-cm window of the Ge crystal at a speed of 5.5 mm/µs, giving the time aperture, ∆t
= 10 µs. Thus, for our design, ∆t · ∆f = 500, which indicates a maximum of 500 equivalent
narrowest width that the arbitrary waveform generator can afford. Shown Fig. 3.5(b) is a
spectrum collected for a shaped pulse generated by a series of 3.3 ns acoustic waves separated by
665 ns, shown in Fig. 3.5(a). Each wave deflects a particular frequency in the mid-IR pulse
spectrum. The resulting peaks each have a FWHM of 5 nm and are separated by 63 nm,
demonstrating the high resolution and contrast ratio of the pulse shaper. From the 5 nm FWHM,
we estimate the actual resolution to be 190 equivalent pixels, after taking into account
monochromator resolution (3.6 nm) and timing jitter (1 ns jitter creates 0.1 nm broadening). The
limiting factor for resolution is currently the ~250 µm spot size, which is more than twice as
50
large as the optimal resolvable element (110 µm). Thus, expanding the beam prior to the
We use the comb-shaped spectrum to correlate the spatial position on the AOM and the
spectral position in the mid-IR spectrum for accurate shaping. The arbitrary waveform generator
digitizes the 10-µs time aperture of the RF waveform with an array of 3008 samples. A comb-
shaped RF waveform with a series of peaks with a width of one sample at every 100 samples was
used to generate the calibration curve shown in Fig. 3.5(c). A single-peak spectrum is also
measured to center the RF waveform in the mid-IR spectrum. The peak frequencies in the
resultant spectrum are measured in THz and the calibration curve correlating sample indices to
mid-IR frequencies is fitted by a second-order polynomial. The fitting parameters are used when
digitizing a designed mask in optical frequencies to a discrete set of samples for the arbitrary
waveform generator.
51
Figure 3.5 Resolution measurements. (a) The RF signal used to modulate the spectrum. (b)
The spectrum resulting from the modulated RF waveform of (a). (c) The calibration curve
correlating times at the RF waveform to frequencies of the mid-IR. The measured peak
frequencies in the comb-spectrum (circles) are fitted by a second-order polynomial (solid line).
(a)
amplitude
1 3ns 665ns
0
0 2 4 6 8 10
time (µs)
(b)
intensity (a.u.)
63nm
5nm
60
59
58
57
56
55
54
53
600 800 1000 1200 1400 1600 1800 2000
time (sample index)
52
3.3.2 Efficiency
use in 2D IR and coherent control experiments. The Ge AOM has a deflection efficiency of
~60%, giving a theoretical throughput efficiency of 40% when used in conjuction with two
gratings of 80% efficiency. We usually achieve 30% efficiency because of suboptimal optics.
Most likely the folding mirrors is the limiting factor since the aluminum coating has ~90%
reflection efficiency. The total power of our shaped pulses is typically ~1.5 µJ when starting
from an OPA output of 5 µJ at 5 micron, and ~0.8 µJ when starting from an OPA output of 3.5
µJ at 6 micron.
To accurately program the pulse shape, the efficiency needs to be correlated for the
amplitude and the position of the acoustic wave. Shown in Fig. 3.6(a) is the deflection
efficiency of AOM as a function of acoustic wave amplitude. The deflection efficiency scales
nearly linearly with acoustic wave amplitude up to 0.5 of the maximum voltage of the arbitrary
waveform generator. This linear correlation holds across the 55-mm aperture, exampled at 10, 20
and 30 mm. Shown in Fig. 3.6(b) is the AOM deflection efficiency as a function of position
along the aperture for an acoustic wave amplitude of 0.5. The maximum deflection efficiency is
~70% and drops to ~50% at 15-mm from the end of the aperture due to divergence of the
acoustic wave. When shaping the pulses, we include the aperture position and acoustic wave
amplitude dependent deflection efficiencies in the software to design the shaped pulses.
53
Figure 3.6 Amplitude and position dependence on deflection efficiency. (a) Deflected
intensity at three positions across the aperture vs. acoustic wave amplitude. (b) Deflection
efficiency across the aperture for tan acoustic wave amplitude of 0.5.
(a) (b)
1.0
70
0.8
normalized
65
0.6
60
0.4 at 10 mm 55
at 20 mm
0.2 at 30 mm 50
0.0 45
0 0.2 0.4 0.6 0.8 1 5 10 15 20 25 30 35 40
normalized position at the aperture (mm)
acoustic wave intensity
54
3.3.3 Stability
In order for shaped pulses to be generally incorporated into a coherent 2D IR pulse sequence,
they must be phase stable since the 2D IR signal is collected interferometrically. Interferometric
stability is degraded by changes in pathlength that are usually caused by variations in mirror
pointing stability. These fluctuations can be actively compensated with stabilized beampaths or
passive correction. Besides pathlength drift, phase instability can be caused by the Ge AOM
electronics. Since the phase of the acoustic wave sets the phase of the mid-IR pulse, the acoustic
wave must be reliably triggered to within a fraction of the 75 MHz center frequency for each
laser pulse. A timing jitter of 6.7 ns would lead to randomly phased pulses. The arbitrary
waveform generator locates samples separated by 3.3 ns according to a 300 MHz clock (internal
or external). When using the internal clock unsynchronized to the laser, the arbitrary waveform
generator produces rf waveforms with a phase jitter of λ/4 due to the timing jitter of 3.3 ns. For
an external clock, a phase-lock circuit generates a 300 MHz clock signal synchronized to the
Shown in Fig. 3.7 are plots of the phase of the shaped pulse relative to an unshaped
reference during data collection of 100 sec, quantified by with our crosscorrelator. The phase
was measured by setting the delay stage to a position where the crosscorrelation signal amplitude
is close to zero so that intensities are linearly relative to the phase. Each point in the plot of Fig.
3.7 is averaged over 5 laser shot. Notice that the phase fluctuation with the internal clock (Fig.
3.7(a)) is ~4 times larger that the phase jitter with the external clock (Fig. 3.7(b)). The standard
deviation over the 100 second experiment gives a phase stability of λ/27 when the external clock
used. When normalized for shot-to-shot noise, this is about half the phase stability of a standard
2D IR optical setup without a shaper. Since the period of the acoustic wave corresponds to 13.3
55
ns, with 0.6 ns electronic jitter, we do not expect better than λ/50 phase stability. Electronic
jitter might be further reduced by locking to a higher harmonic of the Ti:Sapphire oscillator.
Figure 3.7 Phase stability measurements using (b) the internal clock and (c) the external
clock for the arbitrary waveform generator. 5 laser shots were averaged.
π/2
π/4
(a)
0
phase (radian)
−π/4
−π/2
π/4 (b)
0
−π/4
π/2
0 10 20 30 40 50 60 70 80 90 100
time (sec)
57
Mid-IR pulses generated from the OPA are not necessarily transform-limited. Bragg deflection at
the AOM also adds a significant amount of linear chirp to the pulse and the Ge AOM itself
causes material dispersion. Since the 4-f configuration is a grating-pair compressor (22), the
linear chirp can be reduced by translating the second grating in the shaper to maximize the SHG
of the shaped beam using a doubling AgGaS2 crystal. Following grating optimization, we use
the AOM to refine the linear chirp compensation as well as compensate for higher order phase
distortions in the pulses. In fact, in order to program pulses with well-defined shapes, it is
essential to first remove the phase distortions of the input pulses. The phase of the electric field
1 1
φ (ω ) = φ0 + φ1 (ω − ω 0 ) + φ2 (ω − ω 0 ) 2 + φ3 (ω − ω 0 ) 3 + ⋅ ⋅ ⋅ (3.6)
2 6
where φn are the dispersion coefficients, each of which can be independently varied.
To create transform limited pulses, we varied the 2nd and 3rd-order dispersion
coefficients (φ2 and φ3) while recording the resultant SHG signal, shown in Fig. 3.8. The
optimum values resulting in the maximum SHG were φ2 = 0.06 ps2 and φ3= 0.06 ps3, which
corresponds to the shortest pulse duration. With these optimum coefficients, the duration
measured from the autocorrelation in Fig. 3.8(b) was reduced to 67 fs from 80 fs. While this
process produces transform limited pulses, the pulse duration is not as short as might be expected
from the bandwidth of the pulses emitted directly from the OPA (Fig. 2) because the deflection
efficiency of the shaper effectively reduces the pulse bandwidth (Section 3.3.2, above).
58
Figure 3.8 Compressing the shaped. (a) Scanned SHG intensity as a function of phase
terms φ2 and φ3. Autocorrelations with (b) unoptimized and (c) optimized phase parameters.
0.16
(a) (b) (c)
0.12
φ3 (ps3)
0.08
0.04
0.00
-0.04
-0.05 0.00 0.05 1.00 1.50 2.00 -200 -100 0 100 200 -200 -100 0 100 200
φ2 (ps2) time (fs) time (fs)
59
phases and time delays. These pulse trains are conventionally created using beam splitters.
Pulse shaping has the potential of generating arbitrary numbers of pulses with control over the
amplitude and phase of each pulse, which is impossible for conventional splitting configurations.
Here, we create a pulse pair with variable time delay and phase. The mathematical
representation of generating a pulse pair in Eq. 3.4 can be extended to specify a phase for each
1 iωτ iφ1
M (ω ) = (e e + eiφ 2 ) (3.7)
2
where φ1 is the phase of the pulse at t = τ where as φ2 is the phase of the pulse at t = 0. The mask
ωτ + φ1 − φ2 ωτ + φ1 + φ2
M (ω ) = cos exp i . (3.8)
2 2
(ωτ + φ1 + φ2 ) / 2 generates two pulses: one at t = τ with a phase of φ1; another at t = 0 with a
phase of φ2.
Fig. 3.9(a) includes the autocorrelation of a pulse pair created by sinusoidally modulating
the intensity of an RF waveform. The relative delay is determined by the period of the sine wave
while the phase of the sinusoidal modulation dictates the relative phase between the two pulses.
Also shown in Fig. 3.9(b) are enlarged sections of 1 ps autocorrelations with the relative phase
set to ∆φ = 0 and π, revealing a corresponding phase difference. In Fig. 3.9(c) the relative phase
60
is incremented in steps of 0.1π rad, and the measured deviation from linearity gives a relative
phase control of the order of 0.008π rad. Accurate relative phase control is critical in phase
cycling 2D IR experiments.
61
Figure 3.9 Pulse pairs with various spacing and relative phase. (a) The autocorrelation of
pulse pairs with time separations of 1, 2, and 3 ps. (b) Enlarged section of the 1 ps
autocorrelation from (a) with relative phases between pulses of 0 rad (solid) and π rad (dashed).
(c) Theoretical and experimental relative phases of pulse pairs with 1 ps time separation, for
(a)
intensity (a.u.)
-3 -2 -1 0 1 2 3
time (ps)
(b) π (c)
experimental ∆φ
intensity (a.u.)
0 π/2
-1
0
960 1000 1040 1080 0 π/2 π
time (fs) theoretical ∆φ
62
3.5. Conclusion
Mid-IR pulse shaping has been previously limited to simple linear chirps using material or
gratings and to indirect shaping methods with low resolution and power (16-19). Although
simply shaped, applications of these pulses demonstrated vibrational ladder climbing and
optimized pulse compression (28, 29). We demonstrated pulse shaping directly in the mid-IR
using a Ge AOM with high efficiency and resolution. We generated a series of phase and
amplitude pulses, the accuracy of which is tested with simulations. Phase-stable mid-IR shaping
energy transfer, and study solute-solvent interactions. In fact, simulation studies suggest that
well-designed pulses can populate certain high vibrational modes (30), control intra- and inter-
molecular proton transfer (31-33), control electron transfer through vibrational excitation (34),
and direct isomerization (35). Considering that 2D IR experiments are usually carried out with
<1 µJ of power (5) and that vibrational populations have recently been inverted with only 2.2 µJ
(36), it stands to reason that this new shaper promises to open a new class of experiments geared
3.6. References
3. Cowan ML, Bruner BD, Huse N, Dwyer JR, Chugh B, Nibbering ETJ, Elsaesser T, &
Miller RJD (2005) Ultrafast memory loss and energy redistribution in the hydrogen bond
network of liquid H2O. ature 434(7030):199-202.
8. Tian PF, Keusters D, Suzaki Y, & Warren WS (2003) Femtosecond phase-coherent two-
dimensional spectroscopy. Science 300(5625):1553-1555.
9. Gordon RJ & Rice SA (1997) Active control of the dynamics of atoms and molecules.
Annu. Rev. Phys. Chem. 48:601-641.
10. Assion A, Baumert T, Bergt M, Brixner T, Kiefer B, Seyfried V, Strehle M, & Gerber G
(1998) Control of chemical reactions by feedback-optimized phase-shaped femtosecond
laser pulses. Science 282(5390):919-922.
11. Rabitz H, de Vivie-Riedle R, Motzkus M, & Kompa K (2000) Chemistry - Whither the
future of controlling quantum phenomena. Science 288(5467):824-828.
12. Weiner AM, Leaird DE, Patel JS, & Wullert JR (1990) Programmable Femtosecond
Pulse Shaping by Use of a Multielement Liquid-Crystal Phase Modulator. Opt. Lett.
15(6):326-328.
13. Wefers MM & Nelson KA (1993) Programmable Phase and Amplitude Femtosecond
Pulse Shaping. Opt. Lett. 18(23):2032-2034.
14. Dugan MA, Tull JX, & Warren WS (1997) High-resolution acousto-optic shaping of
unamplified and amplified femtosecond laser pulses. J. Opt. Soc. Am. B 14(9):2348-2358.
64
16. Eickemeyer F, Kaindl RA, Woerner M, Elsaesser T, & Weiner AM (2000) Controlled
shaping of ultrafast electric field transients in the mid-infrared spectral range. Opt. Lett.
25(19):1472-1474.
17. Belabas N, Likforman JP, Canioni L, Bousquet B, & Joffre M (2001) Coherent
broadband pulse shaping in the mid infrared. Opt. Lett. 26(10):743-745.
18. Witte T, Zeidler D, Proch D, Kompa KL, & Motzkus M (2002) Programmable
amplitude- and phase-modulated femtosecond laser pulses in the mid-infrared. Opt. Lett.
27(2):131-133.
19. Tan HS & Warren WS (2003) Mid infrared pulse shaping by optical parametric
amplification and its application to optical free induction decay measurement. Opt.
Express 11(9):1021-1028.
21. Strasfeld DB, Shim S-H, & Zanni MT (2007) Controlling vibrational excitation with
shaped mid-IR pulses. Phys. Rev. Lett. 99(3).
22. Weiner AM (2000) Femtosecond pulse shaping using spatial light modulators. Rev. Sci.
Instrum. 71(5):1929-1960.
23. Shim S-H, Strasfeld DB, Fulmer EC, & Zanni MT (2006) Femtosecond pulse shaping
directly in the mid-IR using acousto-optic modulation. Opt. Lett. 31(6):838-840.
24. Shim S-H, Strasfeld DB, & Zanni MT (2006) Generation and characterization of phase
and amplitude shaped femtosecond mid-IR pulses. Opt. Express 14(26):13120-13130.
25. Hillegas CW, Tull JX, Goswami D, Strickland D, & Warren WS (1994) Femtosecond
Laser-Pulse Shaping by Use of Microsecond Radiofrequency Pulses. Opt. Lett.
19(10):737-739.
26. Kaindl RA, Wurm M, Reimann K, Hamm P, Weiner AM, & Woerner M (2000)
Generation, shaping, and characterization of intense femtosecond pulses tunable from 3
to 20 mu m. J. Opt. Soc. Am. B 17(12):2086-2094.
28. Kleiman VD, Arrivo SM, Melinger JS, & Heilweil EJ (1998) Controlling condensed-
phase vibrational excitation with tailored infrared pulses. Chem. Phys. 233(2-3):207-216.
65
30. Beyvers S, Ohtsuki Y, & Saalfrank P (2006) Optimal control in a dissipative system:
Vibrational excitation of CO/Cu(100) by IR pulses. J. Chem. Phys. 124(23):234706.
32. Ohta Y, Bando T, Yoshimoto T, Nishi K, Nagao H, & Nishikawa K (2001) Control of
intramolecular proton transfer by a laser field. J. Phys. Chem. A 105(34):8031-8037.
33. Petkovic M & Kuhn O (2004) Ultrafast wave packet dynamics of an intramolecular
hydrogen transfer system: from vibrational motion to reaction control. Chem. Phys.
304(1-2):91-102.
34. Skourtis SS, Waldeck DH, & Beratan DN (2004) Inelastic electron tunneling erases
coupling-pathway interferences. J. Phys. Chem. B 108(40):15511-15518.
35. Artamonov M, Ho TS, & Rabitz H (2006) Quantum optimal control of molecular
isomerization in the presence of a competing dissociation channel. J. Chem. Phys. 124(6).
36. Ventalon C, Fraser JM, Vos MH, Alexandrou A, Martin JL, & Joffre M (2004) Coherent
vibrational climbing in carboxyhemoglobin. Proc. atl. Acad. Sci. U.S.A. 101(36):13216-
13220.
66
Chapter 4
Automating 2D IR Spectroscopy Using Mid-Infrared Pulse
Shaping
4.1 Introduction
Two-dimensional infrared (2D IR) spectroscopy is becoming a very useful tool for probing the
fast structural dynamics of chemical and biological systems (1, 2). Analogous in many ways to
frequencies, couplings, and transition dipole angles (3-7). Environmental dynamics are probed
through lineshape analysis (8-10). The combination of structural sensitivity and fast time-
resolution (fs/ps) makes this technique and its variants (11-14) especially adept at monitoring
2D IR spectrometer, each pulse has its own optical path composed of several mirrors to route the
beam to the sample and a mechanical delay stage that controls the time delay by changing the
physical length of the optical path. All of the pulses have the same frequencies and shapes
unless additional optics are added such as an etalon to narrow a pulse bandwidth (3, 19) or a
second mixing crystal to generate pulse sequences with two center frequencies (20, 21). Thus,
implementing even the simplest pulse sequence is a tremendous amount of work. This is in
sharp contrast to 2D NMR spectroscopy, where NMR pulse sequences are easily programmed
straightforward in NMR to change the pulse sequences or add additional pulses in order to create
67
the best spectrum for measuring the desired information. In this chapter, we take a step closer
towards making 2D IR more like 2D NMR spectroscopy. Using a mid-IR pulse shaper recently
developed by our group (see Chapter 3), we can now programmably generate 2D IR pulse
sequences with controlled phases, delays and shapes (22, 23). Using traditional 2D IR methods
as a starting point, we improve upon their accuracy, speed and ease of implementation with
several alternative pulse sequences optimized to maximize the structural content of the spectra
(24). We also introduce phase control and rapid-scan data collection to tackle practical issues in
and his group at the University of Colorado-Boulder, we also recently extended our pulse
shaping approach to 2D visible (2D Vis) spectroscopies (25) which are being used to explore the
semiconductors. While 2D IR and 2D Vis spectroscopies probe very different processes, their
implementation is virtually identical. The only important difference is that the wavelength is
much shorter in the visible, which puts a stringent requirement on the phase stability of 2D Vis
spectrometers. However, this tricky problem is automatically overcome with pulse shaping. In
this chapter, we focus on 2D IR spectroscopy, but because the infrared and visible techniques are
so similar, the contents for 2D IR can also be used to understand pulse shaping methods for
The most obvious way of collecting 2D IR spectra would be stacking pump-probe spectra from
narrow-band pump pulses whose center frequencies are scanned. This so-called “hole-burning”
method was indeed the first method of collecting 2D IR spectra introduced by Hamm, Lim and
Hochstrasser (3) using etalons. An etalon consists of two partial reflectors aligned parallel so
that a pulse entering on one end bounces back and force between the two mirrors, leaking a
portion of the pulse with each bounce. The leaked portions interfere with one another, creating a
narrow-band whose center frequency is determined by the distance between the two mirrors.
The output pulse shape in time contains a long exponential tail as in Fig. 4.1(a), which limits the
time delay between the pump and the probe, thus limits time-resolution of the measurement to >1
ps. Moreover, the lineshape along the pump axis broadens due to the convolution of the intrinsic
spectroscopy that is analogous in many ways to modern pulsed 2D NMR spectroscopy (Fig. 1b)
(8, 11). Hinted from Fourier-transform (FT) spectroscopy useful for improving time and
frequency resolution, both the pump and the probe beams are split into two transform-limited
pulses whose time delays are scanned and Fourier-transformed to generate two frequency axes.
In practice, the first three pulses serves as excitation pulses aligned in a boxcar geometry (Fig.
4.1(b)). The excitation beams are separated in space by beamsplitters and separated in time by
translational stages. A signal comes off at a direction determined by a vector sum of the three
beams. A local oscillator pulse is collinearly combined to the signal for sampling the amplitude
and the phase of the emitted signal. Best time resolution is achieved by using transform-limited
pump pulses; Best frequency resolution is obtained by using proper time increments and number
69
of data. Moreover, this method provides superior selectivity of choosing a specific signal by
spreading combinations of excitations in space according to their phase matching conditions and
a specific order of angles. However, it is challenging to separate and re-overlap the four pulses
using many numbers of optics and translational stages while keeping the phase stability of all
four beams within a small fraction of a micrometer. Moreover, the method does not
automatically generate absorptive spectra or spectra that are properly phased, unlike its hole-
burning counterpart. Instead, two separate 2D IR spectra must be collected with different pulse
sequences and added together to generate absorptive features and optimal spectral resolution (26).
Performing this addition is tricky and requires an appropriate pump-probe spectrum (27), which
does not exist for all pulse sequences. In addition, data collection is usually slow because each
time delay accompanies hundreds of millisecond of a wait time for a computer to communicate
with translational stages. Thus, any means that is more robust and rapid to generate and control
pulse sequence would greatly improve the methods for collecting 2D IR spectra.
A mid-IR pulse shaper is a convenient tool to generate pulse sequences with accurate and
stable phase (22, 23). Femtosecond pulse shaping allows arbitrary control over amplitude and/or
phase and is often capable to generate very complicated shapes that are almost impossible for
other optical means to reproduce (28). Generating a sequence of pulses is one of the simplest
shapers are ready to carry out 2D optical spectroscopy including our mid-IR pulse shaper. Our
shaper is based on Ge AOM, which serves as a programmable diffractive grating whose phase
and efficiency is controlled by shaping the phase and the amplitude of the acoustic wave passing
through the Ge crystal. Our design affords a maximum of 500 equivalent resolvable elements
70
across the aperture. The actual resolution is 190 equivalent pixels, which enables us to
generate a pulse pair with a time delay as large as 13 ps (22). Our shaper is also demonstrated to
generate pulse pairs with relative phases within an accuracy of 0.008π rad. Also the relative
phases are perfectly stable during hours of data collection because the two pump pulses pass
through exactly the same optical setup. Even their absolute phases were demonstrated to be
stable within λ/27 over 100 seconds for 5 micron light because there is no moving part during
Regardless of choice of method, generating 2D IR spectra rely on the same principles. Three
interactions with a light pulse are required for the sample to generate a 3rd-order nonlinear signal.
When a pulse interact with the sample just once, at least three pulses are necessary. In some
cases like hole-burning, the same pulse interacts with sample twice and generate 2D IR signal.
When a pump-probe geometry is used, the probe serves as a one of the three interactions as well
as a local oscillator that heterodynes the emitted field. The measured 2D IR signal can be written
as convolution of the response functions with the electric fields of the input pulses;
where E1, E2, and Eprobe are the three interacting electric fields and each of them can be written as
Ei = Ai exp[i (ω0 t + φi − k i ⋅ r )] where A, ω0, φ, and k stands for the amplitude, the center
frequency, the phase and the wavevector, respectively, for i=1,2 and probe; Rn (τ , T , t ) are the
vibrational and orientational responses of the system; and ⊗ s represent convolutions of the
71
electric fields with the response function. The electric field of a light, Ei, is convoluted with or
without complex conjugation depending on which response of the system interacts with the
electric field. Therefore, phase cycling and phase matching conditions are interchangeable with
each other.
When a hole-burning scheme is used, the shaper carves a narrow band that serves as both
E1 and E2, and scans the center frequency of the narrow band. A 2D spectrum is generated by
stacking pump-probe spectra along the center frequency of the narrow-band pump. When a pulse
pair is used instead of a narrow band, the time delay, τ, between the pulses is scanned and the
− iω pumpτ
S (ωpump , T , ωprobe ) = ∫ S (τ , T , ωprobe )e dτ . (4.2)
Consequently, the lineshapes of the 2D IR spectra depend upon the pulse shape used because the
linewidth along the pump axis is convolution of the intrinsic linewidth of the sample with the
pump shape.
72
Figure 4.1 Schematic layouts of 2D IR experiments: (a) hole-burning method, (b) pulsed
(a) T
k1
k2 kS
k3
(b) t T τ
k1
kS
k2
k3
kLO
(c) T t
k1
k2 kS
k3
73
4.2 Experimental
Shown Fig. 4.2(a) is the optical layout of the experimental setup for the automated 2D IR
spectroscopy. A mid-IR pulse at 2~18 micron and ~50 fs duration is generated by difference
frequency mixing the signal and the idler from a BBO-based optical parametric amplifier that is
pumped by a 45-fs Ti:sapphire regenerative amplifier running at 1 kHz. A small fraction of the
IR beam is served as a probe, and the remainder is sent to a mid-IR pulse shaper described in
Chapter 3 and serves as a pump. The pulse shape of the pump beam is tailored by a Ge AOM, a
modulated grating, whose phase and efficiency across the aperture is controlled by a 75-MHz
waveform at 300 Msample/s, and then the resultant RF waveform is amplified and transferred to
the acoustic wave. A pair of parabolic mirrors focuses and collimates both the pump and the
probe at the sample. The pump and the probe are overlapped temporally at the sample using a
mechanical time delay placed at the pump pass while keeping the probe path length constant.
The probe beam is spectrally dispersed with a monochromator and its spectrum is measured with
a MCT (mercury cadmium telluride) array detector with 64 pixels. Changes of the probe
absorption are measured with and without presence of the pump pulses that are tailored to spread
the probe absorption along the pump axis. Since the shaper allows arbitrary pump shapes, the
setup can carry out both the hole-burning and the pulsed photon echo experiments. For the hole-
burning, the shaper slices a narrow band out of the input spectrum at various center frequencies
and the resultant probe absorption spectra are stacked along the frequency of the narrow-band
pump to give ωpump. For the pulsed photon echo, pulse pairs are generated with their time delay
74
varied from 0 to a time delay long enough to allow vibrations relaxes completely. Then the
time scans at each ωprobe are Fourier-transformed to generate the second frequency domain of
ωpump.
Shown in Fig 4.2(b) is the electronics layout to cycle a new shape for every laser shot and collect
data without deadtime. Data collection in this method is two to three orders of magnitude faster
than conventional methods because the setup uses no moving mechanics. In conventional
methods, one or two automatic stages are translated to a new position to scan the center
frequencies of the etalon output or the time delays between pulses. At each translation, the
computer spends hundreds of millisecond for communicating with the automatic stages. In
contrast, it only takes 10 µs for the Ge AOM to upload a new mask. Therefore, every laser shot
can bear a new shape since the repetition rate of the light source is usually 1~20 kHz.
To reduce the communication time between the computer and the arbitrary waveform
generator (CompuGen 4300, GaGe Applied Technologies), usually 426 waveforms are loaded to
the 1 M on-board memory of the arbitrary waveform generator at once prior to the experiment (it
takes ~40 seconds to upload 426 waveforms). A time aperture of 10 µs at Ge crystal is filled up
with 3008 samples when 300 MHz clock used. Then, 1 M memory can hold 340 waveforms
consist of 3008 samples. Usually, since the deflection efficiency diminishes as the acoustic wave
propagates on the crystal from the piezoelectric transducer, we use only 2400 samples close to
the transducer and load 426 waveforms to the 1M memory of the arbitrary waveform generator.
75
The arbitrary waveform generator has four output channels and four 1M memory for each
channel.
Once the 426 waveforms are loaded, the arbitrary waveform generator brings out each
waveform from its own memory and cycles through the loaded waveforms, one at a time, at each
trigger of 1 kHz from the regenerative amplifier. Thus, at 1 kHz, an entire 2D IR data set is
collected in only 0.426 sec. This process is continuously repeated until the necessary signal to
noise is averaged. Each waveform is synchronized to the laser by a trigger and a clock signal
generated from a 88 MHz pulse train from a photodiode monitoring the Ti:Sapphire oscillator.
The 1 kHz trigger is generated from a circuit that divides the 88 MHz of the Ti:Sapphire
oscillator by 88,000 to produce a 1 kHz TTL pulse. The 300 MHz external clock is produced
from a phase-locked loop circuit that relates the clock to the frequency and phase of the 88 MHz
reference. With the trigger and clock synchronized to the 88 MHz reference, the RF waveform is
As the 426 RF waveforms cycles continuously, the resultant 2D IR signal also cycles
accordingly. To catch the initiation point of the 426 waveform set, a “start” reference signal in
TTL format is also generated from the arbitrary waveform generator. In addition, every two
waveforms within 426 waveforms are paired to generate one pump-probe signal, thus a
“chopper” reference is necessary to process the data correctly. These reference signals can be
loaded to the arbitrary waveform generator along with the 426 pump shapes since the arbitrary
waveform generator affords four output channels sharing the same trigger and clock. Four
separate sets of 426 waveforms can be uploaded to the four channels all at once and cycled
synchronously with each other. Channel 1~4 of the arbitrary waveform generator are used as
follows.
76
(1) Channel 1 produces a “start” reference for the array integrator. It contains a square
pulse at the first element of the waveform set and a zero voltage elsewhere. Thus, for a
set containing 426 waveforms, the reference repeats every 426 ms, thus the repetition rate
of the reference becomes 2.35 Hz. This reference is sent to the integrator electronics box
for the 64-pixel MCT array detector and used to detect the starting point of the 426
signals. All the channels in the integrator shares one gate pulse that is set for the probe
beam impinging the array detector located meters away from the Ge AOM. Thus, the
signal from the arbitrary waveform generator that is timed for the Ge AOM arrives to the
integrator a few microseconds earlier than the gate pulse. To retard the reference closer
to the gate without wasting the memory of the arbitrary waveform generator, we use a
separate delay generator for the reference before sent to the array integrator.
(2) Channel 2 creates a chopper reference signal for the array integrator. In practice, to
reduce number of delay generators, Channel 2 is not utilized for data collection and the
“start” reference is used instead for chopping the signal at Channel 4 because the start
reference indirectly indicates the chopping order. However, in case of collecting the data
less than 426 ms (for instance, checking the pump-probe signal from a reference pump
shaped with the RF waveform from Channel 3), Channel 2 will be ideal to chop the
(3) Channel 3 generates a reference pump shape for the Ge AOM to generate a single
transform-limited pulse. The reference pump is turned on/off every other laser shot
the optical alignment. Since either Channel 3 or Channel 4 should be sent to the Ge
77
AOM, a two-way switch is used to allow the user to switch between aligning the setup
(4) Channel 4 is loaded with the 426 waveforms containing 426 different pump shapes
varying time delays of pulse pairs or center frequencies of narrow bands. The output of
Channel 4 is sent to the Ge AOM to scan time delays or center frequencies for a 2D IR
spectrum. Since pump-probe signals are calculated from pairs of probe intensities with
and without present of the pump, 426 RF waveforms correspond to 213 time delays or
center frequencies. Thus, our setup collects a single 2D IR spectrum with 213 scanning
The array integrator (IR-0128, Infrared Systems Development Corp.) collects data from 62
channels measuring spectrally resolved intensities of the probe beam as well as data from the
external input monitoring the reference indicating the start point of the waveform set. The array
integrator affords total 128 channels for inputs: Channel 1~62 for one 64-pixel array; Channel
65~126 for another 64-pixel array; Channel 63~64 for two external input channels for signals
with ±1V maximum from single detectors to be integrated; Channel 127~128 for two external
input channels of DC-10 kHz level sensor for integrated signals and TTL pulses in ±10V
maximum. The integrator also includes one sync input to generate a gate pulse synchronized to
the trigger read from the sync input. During data collection, Channel 65~126 collects the probe
spectrum from one of the 64-pixel arrays and Channel 127 detects the start reference. After
78
integrating channel 1~126 and detecting channel 127~128, the data from all 128 channels are
sent to the computer. When 426 waveforms used, the computer collects 426 data from the
integrator. Since the computer randomly initiate data collection, the 426 data are not necessarily
ordered in a specific sequence of time delays or center frequencies. Thus, the computer use the
start reference to correlate each data to the shape of the pump pulse, i.e. the waveform sent to the
Ge AOM. The start reference is used to assign the initiation point (time zero when time delays
are scanned) of the 426 data as well as to pair the 426 data to calculate changes of optical density,
∆OD, for 213 time delays or center frequencies. Each pump-probe signal, ∆OD, is calculated as
I
∆OD = − log on (4.3)
I off
where Is are the spectrally resolved intensities of the probe beam when the pump is turned on/off.
Note that ∆OD is equivalent to the 2D IR signal, S(τ, T, ωprobe) in Eq. 4.1. A single 2D IR
spectrum is collected as 2D array of 62x213 ∆OD signals from 213 scanning parameters (time
delay or center frequency) and 62 pixels (probe frequency). ωprobe is optically resolved by the
monochromator to the 62 pixels, whereas ωpump is numerically given after the data collection is
completed. When hole-burning method in frequency domain is used, 213 pump-probe spectra
are stacked along the center frequency of the narrow-band pump to give ωpump. When pulsed
photon echo method in time domain is used, time scans composed of 213 time delays are
Figure 4.2 Experimental setup with (a) optical and (b) electronical layouts.
(a) optics
unshaped probe
mid-IR light
5~6 µm,
~4 µJ/pulse Ge AOM
~55 fs duration shaped pump
sample
λ/2 P
mono- MCT
chromator array
(b) electronics
Ti:Sapphire
Oscillator
1 ms
Divider
88 MHz Circuit 1 kHz AWG
ch4
TRIG
switch
ch3 Sync IN
Array IN
Phase-locked ch2
CLK
OUT
Computer Integrator
80
spectrum (3, 29). To compare the automated 2D IR with conventional 2D IR methods, we used a
mid-IR pulse shaper as a tool to generate a pump pulse with a narrow bandwidth (22, 23). To
demonstrate the utility of collecting 2D IR spectra with our shaper, in what follows we show a
series of 2D IR spectra that have been collected with differently shaped pulse sequences (30).
We start by mimicking the conventional hole-burning approach and then sequentially optimize
the spectra with improved pulse sequences. Shown in Fig. 4 are four series of 2D IR spectra
collected for W(CO)6 solvated in n-hexane, which is a model compound that has a single IR
active anti-symmetric stretch near 5 microns. Each series is generated using a different pump
shape.
First choice among many potential shapes was an etalon-like shape that is used in the
conventional hole-burning method. An etalon consists of two partial reflectors aligned parallel so
that a pulse entering on one end bounces back and force between the two mirrors, leaking a
portion of the pulse with each bounce. The leaked portions interfere with one another, creating a
narrow-band whose center frequency is determined by the distance between the two mirrors, d,
according to
1− R
M (ω ) = (4.4)
1 − R ⋅ exp[+ i (2d cosθ / c)ω ]
where R is the reflection coefficient, θ is the angle the light travels through the etalon and c is the
speed of light (31). To generate the 2D IR spectra in the first column of Fig. 4.3, we set the
bandwidth to 4.6 cm-1 (which is slightly larger than the homogeneous linewidth of 4.4 cm-1),
81
scanned d, and plotted the change in absorption of the probe with and without the pump. 2D
IR spectra were collected at two different pump-probe delay times, T = 2.0 and 0.2 ps. At T =
2.0 ps, a negative and a positive peak pair is observed (blue and red in Fig. 4.3, respectively),
which is exactly as expected according to standard hole burning approaches on metal carbonyl
systems. The negative peak appears along the diagonal with ωpump = ωprobe = 1983 cm-1, which is
the fundamental frequency of W(CO)6. The positive peak is shifted by ∆ωprobe = 13.3 cm-1,
which is the anharmonicity of the W(CO)6 antisymmetric stretch (32). The spectra contain all of
the desirable characteristics about hole-burning that we discussed above: it is properly phased
without any post-data corrections and the peaks have absorptive lineshapes. Thus, pulse shaping
A significant drawback of the hole burning approach is that the picosecond pump pulse
can distort the signal. An etalon generates a picosecond pulse with a sharp front and a long
exponential tail (Fig. 4.3(a)). As a result, the 2D IR spectra collected with an etalon-type pump,
when separated 2 ps from the probe, generates peaks elongated along ωpump because the natural
linewidth is convoluted with a Lorentzian spectrum of a 4.6-cm-1 bandwidth (Fig. 4.3(b)). When
the exponential tail of the etalon-shaped pump is close enough to overlap with the probe
significantly (T = 0.2 ps), the 2D IR peak shapes were severely distorted (Fig. 4.3(c)). This
probably was because the portion of the pump pulse coming after the probe served as the second
and the third pulse among the three excitations generating the non-rephasing 3rd-order signal and
interfered with the desired 2D IR signal. For this reason, an etalon-type shape can be used only
when T > 1 ps. This could be a significant drawback for biological applications. For instance,
the amide I band of a peptide or protein, for which the vibration lifetime is ~1.4 ps, a 3×
enhancement of the signal strength would occur for T = 0.2 vs. 1.8 ps. Of course, the tail can be
82
minimized by using a shorter duration for the pump pulse, but this leads to a larger bandwidth
Pulse shaping provides alternative approaches for data collection using hole burning.
With pulse shaping, the time and frequency profiles of the pump pulse are not limited to just
what can be created by an etalon, but can be programmed to optimize the spectral features. For
instance, we devised a new kind of the etalon-like shape with the same frequency distribution,
but flipped the direction of the exponential tail in time so that it decays away from the probe
pulse (Fig.4.3(d)). This may be almost impossible to accomplish with a conventional optical
setup, but a pulse shaper can easily reverse the pulse shape by reversing the sign of the phase
term in Eq.4.4. This new shape of “time-reversed” etalon produced similar 2D IR peak shapes
for both T = 0.2 and 2.0 ps without peak distortion (Fig. 4.3(e) and (f)). Since the pump pulse
tails away from the probe, the pump and probe pulses do not overlap until much shorter T,
resulting in smaller peak distortions. Furthermore, the signal strength is also stronger for a time-
reversed etalon pulse shape, since the population has had less time to relax with shorter T.
convolution of the pump spectrum with the natural linewidth of the molecular vibration (Eq. 4.1).
Thus, the etalon lineshapes in Lorentzian have a very broad baseline, and as a result, the 2D
peaks have broad tails, which degrades the frequency resolution. An improvement would be to
An alternative shape of a Gaussian with a 7.0-cm-1 FWHM generated peak shapes with a
shorter tail along ωpump (Fig. 4.3(g)). Note that the tails along pump axis are now much shorter,
which is because the Gaussians decay much more rapidly than Lorentzians and thus have
83
improved resolution even for pump pulses with the same full-width-at-half-maximum (Fig.
4.3(h) and (i)). Although we have explored Lorentzians and Gaussians here, any number of
used, the linewidth measured with such a pump suffers from elongation or distortion. If one tried
a pump with a narrower band to reduce the elongation effect, the pulse duration in time would
extend even further and make workable T even farther out. Therefore, using transform-limited
pulses would be an ideal solution. The peak shape distortion could be negligible when the
response function is convoluted with a pulse far shorter than the lifetime of the response function
so that the light pulse can be approximated as a delta function, an infinitely sharp pulse. A time-
domain method utilizing transform-limited pulses will be discussed in the next section.
84
Figure 4.3 2D IR spectra of W(CO)6 dissolved in hexane, obtained with pump pulse
shapes: (a-c) a traditional etalon; (d-f) a time-reversed etalon; (g-i) a Gaussian; (j-l) a pulse pair.
The left column shows the pulse shapes in time domain, and the center and right columns are 2D
T = 2.0 ps T = 0.2 ps
(a) traditional 2000 (b) (c)
etalon 1990
k1 k3
1980
1970
T t
1960
(d) time-reversed 2000 (e) (f)
etalon 1990
k 1 k3
1980
ωpump (cm-1)
1970
T t
1960
(g) Gaussian 2000 (h) (i)
k1 k3 1990
1980
T t 1970
1960
2000
(j) pulse pair (k) (l)
k1 k1 k3 1990
1980
τ T t 1970
1960
1950 1960 1970 1980 1990 2000 1960 1970 1980 1990 2000
ωprobe (cm-1)
85
Hinted from Fourier-transform spectroscopy, a narrow-band pump can be replaced with a pair of
transform-limited pulses whose time delay is scanned to generate ωpump via Fourier-
transformation. The method presented here more closely resembles the pulsed four wave mixing
method (8, 11, 33) except that the experiment is performed in a pump-probe beam geometry. In
this method, the pulse shaper is used to create two transform limited pulses. Rather than scan the
frequencies like was done above, the 2D IR spectrum is instead generated by collecting the
signal as a function of the time delay between the two pump pulses, τ, and Fourier transforming
A mask function for a pulse pair is sum of “1” for one pulse at t = 0 and exp(iωτ) for
another pulse at t = τ (any function gets shifted in time by τ when its counterpart in frequency is
1 iωτ
M (ω ) = (e + 1) (4.5)
2
Thus, combining an amplitude mask of cos(ωτ/2) and a phase mask of ωτ/2 generate a pulse pair
composed of a fixed pulse and a scanning pulse with a time delay of τ. The mask in Eq. 4.5 sets
the phases of the two pump pulses inside their respective envelopes zero (φ1 = φ2 = 0), mimicking
the way that pulse delays are typically generated using translational stages.
and shown in the fourth row of Fig. 4.3 for τ = 0 to 10,000 fs in 14 fs steps. Like the frequency-
86
domain methods in the Sec. 4.3, the spectra are automatically phased and have absorptive
lineshapes, but now the narrowest W(CO)6 pump linewidth is measured and there is no
discernable distortion in the peaks even at T = 200 fs (Fig. 4.3(l)). Moreover, the peak intensity
was ~75 times larger than the intensities measured with the narrow-band shapes when T = 2000
fs. This is because a pulse pair has more intensity as well as shorter duration than a narrow-band
pulse. Due to this, the 2D IR spectrum in Fig. 5d showed a new peak at 1955 cm-1 from the
Note that the time-domain method generated pure absorptive spectra with neither an
and + k1 − k 2 + k 3 , respectively, and show line shapes with wings tilted along the diagonal and
the anti-diagonal axes, respectively. When adding equally-weighted and well-phased rephasing
and non-rephasing spectra, the wings cancel out to yield a purely absorptive line shape that gives
the optimal spectral resolution (26). Therefore, in traditional four wave mixing experiments,
absorptive spectra ask for two 2D IR scans for rephasing and nonrephasing signals as well as a
transient absorption experiment to set time zeros (27). Once the spectra have been collected,
they must be phased and added. In a pump-probe geometry, since k 1 = k 2 , both the rephasing
and the non-rephasing signal are emitted collinearly and detected simultaneously while properly
weighted and phased automatically. Since the time-zeros are precisely zero, the desired
measured. We also point out that the peak shapes are symmetrical about ωpump and that there are
no spurious ghost images, indicating that the shaper time-resolution is sufficiently accurate.
87
Asymmetric shapes and ghost images are common problems that occur when 2D IR spectra
are collected using translational stages to increment time delays (34, 35).
We usually use the time-domain method for most of applications. The time-domain
method is indeed the best way of collecting 2D IR spectra providing the best possible features:
intrinsic lineshapes, maximal signal strength, maximal time and frequency resolution, and
absorptive line shapes. Most uniquely, pulse shaping allows many new tricks of cycling and
incrementing phase that further enhance the pulse-pair method. So far, the phases of the pulses
in pulse pairs were held constant ( φ1 = φ2 = 0) to mimic the way that pulse delays are typically
arbitrarily set the phases and using this feature would make the automated 2D IR method truly
Compared to the three frequency-domain methods in Sec 4.3, the time-domain method in Sec.
4.4 has the optimal time and frequency resolution in a straightforward manner. Furthermore, the
shaper allows arbitrary control over the phase so that 2D IR spectroscopy can borrow tricks that
have been extensively demonstrated in NMR spectroscopy including phase incrementing and
phase cycling. The pulse shaping mask in Eq. 4.5 for a pulse pair can be extended with
1
M (ω ) = [exp(iωτ ) exp(iφ1 ) + exp(iφ2 )] (4.7)
2
where φ1 and φ2 are the phases of the pulse at t = τ and 0, respectively. Any functional form in
frequency can be given for φ1 and φ2 . In a pump-probe geometry, two phase matching
directions of –k1+ k2+ k3 and +k1− k2+ k3 are detected, so the phase of the signal, φsig , can be
written as
Thus, the 2D IR signal depends on the relative phase between the pulses in a pulse pair, ∆φ .
Then, in Eq. 4.1, a phase term exp(i∆φ ) is convoluted with the response function and the 2D IR
signal changes its phase or frequency accordingly. Using this fact, we routinely implement
phase incrementing or phase cycling to shift the signal to a lower frequency, remove the
unwanted transient absorption background, and subtract scatter from the signal. For these
reasons, the pulsed method is the one most often used in our research group. In this section, we
outline each of these capabilities and describe the pulse sequence most often used in our group.
89
In NMR spectroscopy, the signal frequency is shifted by incrementing the phase of radio-
frequency pulses proportional to the time delay so that fewer data points need to be collected
(36). This scheme is called as time-proportional phase increments. It can be easily implemented
to optical spectroscopy with a pulse shaper capable of phase control. In phase incrementing, the
phase of a femtosecond pulse is linearly dependent on the time delay, φ = ωiτ where ωi is the
frequency incrementing the phase. When ∆φ = 0 , the 2D IR signal oscillates in τ with ω0 , the
vibrational frequency of the system. When ∆φ = ωiτ , the 2D IR signal is convoluted with an
Shown in Fig. 6(a) are three time scans of n-methyl amide (NMA) in D2O at ωprobe=1625
cm-1 using three different ωi to increment ∆φ. In this particular case, we used φ1 = ωiτ and
φ2 = 0 . The ω0 of the amide I mode of NMA in D2O is 1625 cm-1, so when ωi = 0, the signal
oscillates with a period of 20 fs. When ωi = 1625 cm-1, then ω0 − ωi = 0 and the signal no longer
oscillates with τ. When 0 < ωi < ω0 , the signal oscillates with a period larger than 20 fs. In Fig.
6(a), ωi = 1200, 1400 and 1625 cm-1, so the period of the time scans increased accordingly (78 fs,
148 fs and infinity). These time scans were Fourier-transformed and their center frequencies
were measured in Fig. 6b. The ωi and the resultant signal frequencies were linearly correlated
with a slope of −0.99 and a root-mean-square of 2.0 cm-1, indicating that our shaper is
sequence, a function of discrete time. An analogue signal that has been digitized can be perfectly
90
reconstructed if the sampling increment in time, ∆t, is less than the half of the period of the
sinusoidal signal, ts (i.e. ts <∆t/2) (36). When this “Nyquist sampling criterion” is not satisfied,
the frequencies are overlap; that is, frequencies above 1/2ts will be reconstructed as and appear as
frequencies below 1/2ts. The resulting distortion is called aliasing. Aliasing not only confuses
the frequency, but also distorts the phase and the amplitude of the signal. Especially, the
sampled amplitude reduces when ts >> ∆t/2 because the “under”-sampling misses the top and the
bottom of the sinusoid. However, the undersampling is often used when the vibrational
relaxation takes too much time to use ts < ∆t/2 that enables “regular” sampling for data collection.
When the relative phase increments used to shift the observed frequency, the same ts that has
been used for undersampling can sample the signal without breaking the Nyquist criterion and
reducing the measured amplitude. This is particularly useful when a good signal-to-noise
spectrum should be recorded in a faster speed. For instance, to follow kinetics of amyloid
formation completing in a couple of hours (will be discussed in Chapter 5), a sampling scheme
using ωi of 1225 cm-1 and ∆t of 22 fs collects a 2D IR spectrum of the amide I band of the fibril
at 1618 cm-1 that relaxes in ~3 ps while spending only 0.4 sec for a single spectrum and keeping
Figure 4.4 Sampling the amide I mode of NMA in D2O by incrementing the relative phase
in proportion to τ. (a) Time scans at ωprobe = 1625 cm-1 by using ωi = 1200, 1400 and 1625 cm-1.
(b) The measured frequency after Fourier-transforming the scans in (a) plotted with ωi used for
sampling.
(a)
30
∆OD at ωprobe=1625 cm-1
25 ωi =1200 cm-1
20
15 ωi =1400 cm-1
10
5 ωi =1625 cm-1
τ (fs)
(b)
400
measured frequency (cm-1)
300
200
100
0
1200 1300 1400 1500 1600
ωi (cm-1)
92
According to Eq. 4.1, the 3rd-order signal is generated by two interactions with the pump pulse
(E1 and E2) and one interaction with the probe (E3). However, a 3rd-order signal is also generated
when E1 or E2 interacts twice with the sample, which gives rise to two transient absorption
spectra, e.g.,
( )
S B (τ , T , ω probe ) = ∫ dt ∑ Rn (τ , T , t ) ⊗ Ei* ⊗ Ei ⊗ Eprobe + Eprobe e
−iωprobet
(4.9)
n
where i=1 or 2 and ⊗ s represent convolutions of the electric fields with the response function.
Fig. 4.5 (as well as Fig. 4.4) outlines the three signals that arise from the two transient absorption
spectra and the desired 3rd-order spectrum: an exponential decay from the moving pulse (E1)
alone; a constant offset from the stationary pulse (E2) alone; an oscillation from E1 and E2. The
oscillatory part is the desired signal, but the exponential decay and the constant offset are
backgrounds of no interest. These background signals are easy to discriminate against in the
frequency domain, because neither oscillates when τ is incremented. However, in some cases
they can be a problem, such as when the signal frequency is shifted closer to ωpump=0 (when
The two transient absorption signals can be removed by programming all three pulse
sequences (E1 only; E2 only; E1 and E2) so that the two background signals can be subtracted
from the total signal. However, this effectively cuts the repetition rate of the laser system by 2/3.
A more efficient way is to use phase cycling. The phase of desired signal depends on the phase
difference between E1 and E2, e.g. ∆φ = φ1 − φ2, whereas the background transient absorption are
insensitive to the phase difference. Thus, for each time delay, we collect probe spectra with ∆φ =
I (∆φ = 0 )
∆OD = − log , (4.10)
I ( ∆φ = π )
instead of measuring a signal by tuning on/off the pump as in Eq. 4.3. Thus, background
The subtraction procedure is demonstrated in Fig. 4.5, where we show the data for each
of the pulse sequences separately. Shown in Fig. 4.5(a) are two time scans of NMA in D2O at
∆φ = π appeared out of phase to the oscillation when ∆φ = 0, while the backgrounds remained the
same. Therefore, when the two scans with ∆φ = 0 and π were subtracted, the backgrounds were
removed; only oscillatory part remained and doubled in intensity (thick). Fig. 4.5(b) displays the
resultant spectra after FT giving a peak with the opposite sign for ∆φ = 0 (solid) and π (dotted)
containing an oscillatory noise along the entire ωpump and a peak doubled in intensity and free
from the oscillatory noise for the subtracted signal (thick). The oscillatory noise comes from the
constant offset because a step function in time generates oscillations along the entire frequency
after FT. Numerically subtracting the background in time domain after the experiment may
remove the oscillatory noise in frequency, but subtracting the two out-of-phase signals is
Figure 4.5 Removing backgrounds from transient absorption using a phase cycling scheme
for the amide I mode of NMA in D2O. (a) Time scans at ωprobe = 1625 cm-1 when ∆φ = 0 (solid)
and ∆φ = π (dotted) and the subtracted product of the two scans (thick). (b) Fourier transforms
(a) 15
10
∆OD (x103)
0
∆OD(∆φ=0)
-5 ∆OD(∆φ=π)
∆OD(∆φ=0) − ∆OD(∆φ=π)
-10
0 500 1,000 1,500
τ (fs)
200
(b)
∆OD(∆φ=0)
150
∆OD(∆φ=π)
2D IR intensity (a.u)
∆OD(∆φ=0) − ∆OD(∆φ=π)
100
50
-50
-100
1400 1450 1500 1550 1600 1650 1700 1750 1800
ωpump (cm-1)
95
Phase incrementing or cycling allows scatter to be shifted or subtracted from the signal. Laser
light from the pulses can be scattered off spatially inhomogeneous samples such as crystals,
aggregates or any sample with length scales comparable to the wavelength of light. This is an
especially grievous problem with amyloid peptides, which we study in our laboratory, because
the peptides form fibrils spanning lengtzh scales of 30 nm to several microns (37). Fortunately,
much of this scatter can be removed by phase incrementing or cycling. The desired 3rd-order
signal arises from each pulse interacting with the sample once, while the scatter occurs from each
Thus, the phase of a background from the scattered pump depends on φ1 or φ2, whereas the signal
phase depends on ∆φ . If φ1 = φ2 = ωiτ , then ∆φ = 0 , then the signal remains the same in
frequency. In the meanwhile, the scatter phase is incremented by ωiτ and the scatter frequency
Fig. 4.6 presents 2D IR spectra of amyloid fibrils from human Islet amyloid polypeptide
(hIAPP) in D2O. hIAPP forms fibrils ~100 nm in diameter and up to several microns in length
(37). As a result, the sample scatters the mid-IR pump, which interferes with the desired
spectrum. Scatter from the stationary pulse in the pulse-pair pump can be subtracted or filtered
because it appears at the ωpump = 0 (Fig. 4.6(a)). However, numerical subtraction is almost
impossible for the scatter from the moving pulse in the pulse pair because the scatter appear at
ωpump = ω0 as a broad line elongated along the diagonal axis that overlaps with the desired
spectrum (Fig. 4.6(b)). Rather than post-correcting our data, we shift the scatter frequency to
96
separate the scatter from the desired spectrum during the experiment. When ωi = 1800 cm-1
was used to increment both φ1 and φ2, the scatter from the stationary pulse showed up at ωpump =
1800 cm-1; the scatter from the moving pulse appeared at ωpump = ~220 cm-1, elongated along the
anti-diagonal (Fig. 4.6(c)). The scatter lineshape is flipped because ω0 − ωi = −220 cm-1 and a
cosine function appears as mirror images at both the positive and the negative ωpump after FT.
Then, the desired 2D spectrum at ωpump = 1580~1680 cm-1 was completely separated from the
Instead of shifting scatter frequency with phase increments, phase cycling can be used to
subtract scatter from the signal. We collect four probe spectrum: I(φ1=0, φ2=0), I (φ1=0, φ2=π), I
(φ1=π, φ2=π), and I (φ1=π, φ2=0). We then add them to give ∆OD for each time delay:
I (φ = 0,φ2 = 0 ) I (φ1 = π , φ2 = π )
∆OD = − log 1 − log . (4.12)
I (φ1 = 0, φ2 = π ) I (φ1 = π ,φ2 = 0 )
The result is that the scatter subtracts and the signal adds. This sequence also subtracts off the
unwanted transient absorption spectra discussed in Sec. 4.5.2. As a result, this is the series of
pulse sequence that we most commonly use when collecting 2D IR spectra. Since each pulse
sequence in the series is collected from one laser shot to another, it only takes 4 ms at 1 kHz
repetition rate to measure all four probe spectra. On this timescale, the scatter is essentially
constant, even for aggregating amyloids, so that the scatter subtraction can be done even when
Figure 4.6 2D IR spectra of hIAPP fiber in D2O when using (a and b) no phase increments
and (c and d) ωi = 1800 cm-1. The left panel shows the entire ωpump from 0 to 2034 cm-1 and the
2000
(a) 1680 (b)
1600
1660
ωpump (cm-1)
ωpump (cm-1)
1200
1640
800 1620
400 1600
2000
(c) 1680 (d)
1600
1660
ωpump (cm-1)
ωpump (cm-1)
1200
1640
800 1620
400 1600
0
1580 1600 1620 1640 1660 1680 1580 1600 1620 1640 1660 1680
ωprobe (cm-1) ωprobe (cm-1)
98
One of the most noticeable advantages of the automated method presented here is the far more
prompt data collection than traditional methods for colleting 2D IR spectra. In hole-burning
experiments, the center frequencies are adjusted by translating one of the two high reflectors in
an etalon. Because displacements of ~1 nm are needed to scan the center frequency by 1 cm-1,
piezo crystal actuators are typically used. Piezo crystal actuators are slow and not reproducible.
Thus, at each point during scanning, the pump pulse frequency is set using feedback from a
spectrometer, which is time consuming. In a traditional four wave mixing setup, the delay times
are adjusted by either translating a retroreflector to change the pathlength of the pulses to the
sample or translating a wedged optic to change the amount of material the pulses pass through.
For each translation, the computer spends 50-100 ms to communicate with the automatic stage.
To collect a typical 2D IR data set, each data point might be averaged for 50 or 100 laser shots,
which means that for a 1 kHz laser system, data collection is only half of as efficient as it could
be. The efficiency is even worse for higher repetition rate laser systems, where a larger fraction
In contrast, a pulse shaper can create pulse sequences without moving any optic on a
shot-to-shot basis, enhancing the efficiency of data collection drastically. For our germanium
acousto-optic modulator, it takes 10 microseconds for the sound wave to travel the length of the
repetition rate of 100 kHz. Thus, at 1 kHz, the time delay is easily incremented at every laser
shot, making data collection twice as fast with a pulse shaper. Furthermore, high repetition rate
has a much higher impact with pulse shaping 2D IR, because data collection rate decreases
linearly with repetition rate, unlike traditional methods in which the deadtime cannot be reduced.
99
are uploaded into the memory of the arbitrary waveform generator prior to the experiment and
cycled continuously during the experiment. At each trigger from the laser, the waveform
generator cycles through the loaded waveforms, one at a time. Thus, at 1 kHz, an entire 2D IR
data set is collected in only 0.4 sec. This process is continuously repeated until the signal to
noise reaches a reasonable number. Averaging 100 shots (like in a traditional set), thus takes 40
s. The faster data collection is also expected to improve the averaging. Laser sources vary
noise of the light source by averaging out slow random fluctuations, which can contribute
significantly to the noise when scanned slowly (ref). Besides, most importantly, such a rapid
data collection allows monitoring dynamic events happening in seconds and minutes with 2D IR
spectroscopy on-the-fly without initiating the event for every frequency or time delay.
In practice, the memory of the arbitrary waveform generator limits the number of time
delays to collect a 2D IR spectrum. The number of time delays cannot exceed the half (or the
quarter when four laser shots are used for one ∆OD) of the number of loaded waveforms for
continuous cycling. When using 426 waveforms, the number time delays are 213 for regular
pump-probe signals (Eqs. 4.3 and 4.10) and 106 for the four-shot scheme in Eq. 4.12. For
tracking amyloid formation that scatters light extensively, we use four laser shots for each data.
Regarding the relaxation of amyloid fibrils (~2.5 ps), the increment of time delays should be
longer than 22 fs, which is larger than 20 fs, the period of a 6-micron mid-IR light. Rather than
undersampling, the signal frequency can be shifted using the phase increments in Sec. 4.5.1 so
that a 28-fs increment can sample the signal without violating Nyquist sampling criterion.
100
Notice that the method can be tailored for each application by simply changing the software.
This is the most unique difference of the pulse shaping 2D IR comparing to traditional ways
which necessitates an extensive optical reconfiguration for any effort to enhance the method.
Our method of collecting 2D spectra is not limited to the infrared. Our method of pulse shaping
in a partly collinear beam geometry is also applicable for generating 2D visible (2D Vis) spectra.
Pulse shaping has been used previously to collect 2D Vis spectra by generating fully collinear
pulse sequences (38) and by both spatially and temporally shaping for added control over
using the protocols described here with a standard liquid crystal modulator and a transient
absorption optical layout (25). Regardless of which method is used, one of the biggest
advantages of using pulse shaping to collect 2D Vis spectra is that it overcomes one of the most
To demonstrate the method, a 64-pixel liquid crystal modulator controlling both phase
and amplitude was used to shape a ~50 fs pulse at 800 nm running at a 1-kHz repetition rate.
The pulse pair is followed by an ultrashort probe pulse that is spectrally resolved. The delay
between the collinear pulses is incremented using phase and amplitude shaping and a 2D Vis
spectrum is generated by a following Fourier transformation. Shown in Fig. 4.7 are 2D Vis
spectra of atomic rubidium vapour. Rb vapor has atomic transitions at 794.76 (5P1/2 ← 5S1/2)
101
and 780.03 nm (5P3/2 ← 5S1/2) and has been used to test other 2D E methodologies since it
has very narrow linewidths. The real part of the 2D E spectrum is shown in Fig. 4.7 and exhibits
diagonal peaks and cross peaks as expected. The diagonal peaks appear at λpump= λprobe= 780 and
795 nm along with crosspeaks between the two transitions. Shown in Fig. 10b is an expanded
plot of the real part for the diagonal peak at λpump= λprobe = 780 nm. Well-shaped lineshapes are
observed, indicating that the shaper can very accurately step the delay time. The linewidths
differ along λpump and λprobe, which is to be expected under our experimental conditions where the
linewidth along λprobe is determined by the natural linewidth convoluted with the spectrometer
resolution (~0.65 nm), while the linewidth along λpump is given by the natural linewidth
convoluted by the spectral resolution of the Fourier transform, 1/τmax, where τ max is the largest τ
and 2 ps here. We also measured the phase stability of the pulse pair generated by the shaper and
got λ/67 over the course of 35 hours. This number compares to phase drift of λ/100 typical of
diffractive optics and λ/250 with HeNe interferometry measurements made over the course of ~5
hours (40-42).
In most 2D spectrometers, each of the four pulses traverse an independent delay line
whose length must not vary more than a fraction of the wavelength during the experiment.
Achieving adequate phase stability with visible laser pulses can be difficult in conventional four
wave mixing methods. One approach for stabilizing the pathlengths is to use diffractive optics
and phase-compensating mirror arrangements (40). Another is to actively adjust the optical
pathlengths using piezocrystals and two sets of HeNe interferometers for feedback (42). If the
drift is slow enough, then it can be measured and the data corrected (43). All of these
approaches result in high quality multidimensional electronic spectra, but the extensive
infrastructure and effort required to implement these methods has discouraged all but a few
102
dedicated groups from using this extremely powerful research tool. On the other hand, it is
straightforward to generate phase-locked pulse pairs with a pulse shaper capable of modulating
the spectrum of femtosecond pulses, like we have described above. Warren and coworkers
collected 2D Vis spectra by using an entirely collinear train of 4 pulses in analogy to the way that
2D NMR spectra are generated using phase cycling (38). However, the fully collinear geometry
necessitates the signal be measured by some observable other than absorption (e.g. fluorescence),
since the intense excitation pulses will saturate the detector in a linear absorption arrangement.
In contrast, Nelson and coworkers used a two-dimensional pulse shaper to collect 2D Vis spectra
in a non-collinear boxcar geometry where all 4 pulses in the pulse train impinge on the sample
from different directions (39). The approach by Nelson and coworkers takes advantage of phase
matching that provides background free signal and eliminates the need for phase cycling.
However, the two dimensional pulse shaper has very low throughput efficiency owing to the fact
that the input laser beam is spatially filtered to obtain the non-collinear geometry and few pixels,
Our method that significantly eases the hurdles associated with collecting 2D electronic
spectra. 2D Vis spectra can be measured using commonly available pulse shapers and a partly
collinear pump-probe beam geometry. With a pulse shaper it is a trivial matter to make phase
stable pulse trains, thus eliminating difficulties associated with working at short (visible)
wavelengths. Furthermore, the beam geometry used in our experiments eliminates the need for
complicated phasing procedures. Our method eliminates many of the technical hurdles to
implementing 2D Vis spectroscopy, making it possible for many more research groups to exploit
Figure 4.7 2D electronic spectra of atomic Ru vapour: (a) real part over λpump=
λprobe=775~800 nm (b) expanded plots of real part for the diagonal peaks at λpump= λprobe=780 nm.
795 782
781
λpump (nm)
λpump (nm)
790
780
785
779
780
778
775 777
775 780 785 790 795 800 779 780 781
λprobe (nm) λprobe (nm)
104
4.8 Conclusion
Femtosecond pulse shaping greatly enhances 2D IR spectroscopy with its ease of use, versatility
and speed. Programmable control of pulse shapes makes designing and implementing new pulse
sequences just a matter of computer programming. On a daily basis, this method is much easier
to maintain than a four-wave mixing setup, since the pulse shaper and the pump/probe geometry
requires minimal alignment. In many ways, the ease of alignment and accuracy of the collected
data is the greatest attribute of this method. In our experience, much more time can be spent on
the application of the technique to scientific questions rather than on technical maintenance of
shaper that operates in the mid-IR. Further advances will be made possible by straightforward
four pulses selectively allows a combination of transitions at a specific order of angles (33).
Recently, we demonstrated a polarization scheme that assigns different polarization for the two
signal strength and eliminating unwanted background signals (44). It is now straightforward to
construct polarization pulse shapers in the visible/near-IR (45, 46), and it should be possible to
extend these approaches into the mid-IR. Coherent control methodologies can now also be
mid-IR pulse shaping can be used to selectively enhance vibrational excitation of desired
quantum levels (47), which, if incorporated into a 2D IR pulse sequence, could be used to
enhance or eliminate spectral features (48). It should also now be straightforward to implement
105
5th and higher-order signals to measure 3D spectra (49), for example, since addition of laser
pulses no longer requires complicated optical alignments but just additional programming. The
speed of rapid-scanning 2D IR spectroscopy can be further accelerated with high repetition rate
lasers. With current laser technology, it should be straightforward to increase data acquisition by
a factor of 30 or more, resulting in higher signal-to-noise spectra and faster kinetics. It is clear
that the utility of 2D spectroscopies for studying structures and dynamics will only improve in
4.9 Acknowledgment
We appreciate Prof. Niels Damruer and his group at University of Colorado at Boulders for the
4.10 Reference
3. Hamm P, Lim MH, & Hochstrasser RM (1998) Structure of the amide I band of peptides
measured by femtosecond nonlinear-infrared spectroscopy. J. Phys. Chem. B
102(31):6123-6138.
7. Krummel AT & Zanni MT (2006) DNA vibrational coupling revealed with two-
dimensional infrared spectroscopy: Insight into why vibrational spectroscopy is sensitive
to DNA structure. J. Phys. Chem. B 110(28):13991-14000.
10. Loparo JJ, Roberts ST, & Tokmakoff A (2006) Multidimensional infrared spectroscopy
of water. I. Vibrational dynamics in two-dimensional IR line shapes. J. Chem. Phys.
125(19):194521.
13. Fulmer EC, Mukherjee P, Krummel AT, & Zanni MT (2004) A pulse sequence for
directly measuring the anharmonicities of coupled vibrations: Two-quantum two-
dimensional infrared spectroscopy. J. Chem. Phys. 120(17):8067-8078.
14. Fulmer EC, Ding F, Mukherjee P, & Zanni MT (2005) Vibrational dynamics of ions in
glass from fifth-order two-dimensional infrared spectroscopy. Phys. Rev. Lett.
94(6):067402.
15. Chung HS, Khalil M, Smith AW, Ganim Z, & Tokmakoff A (2005) Conformational
changes during the nanosecond-to-millisecond unfolding of ubiquitin. Proc. atl. Acad.
Sci. U.S.A. 102(3):612-617.
17. Zheng JR, Kwak K, Asbury J, Chen X, Piletic IR, & Fayer MD (2005) Ultrafast
dynamics of solute-solvent complexation observed at thermal equilibrium in real time.
Science 309(5739):1338-1343.
18. Kolano C, Helbing J, Kozinski M, Sander W, & Hamm P (2006) Watching hydrogen-
bond dynamics in a beta-turn by transient two-dimensional infrared spectroscopy. ature
444(7118):469-472.
19. Cervetto V, Helbing J, Bredenbeck J, & Hamm P (2004) Double-resonance versus pulsed
Fourier transform two-dimensional infrared spectroscopy: An experimental and
theoretical comparison. J. Chem. Phys. 121(12):5935-5942.
20. Rubtsov IV, Wang JP, & Hochstrasser RM (2003) Vibrational coupling between amide-I
and amide-A modes revealed by femtosecond two color infrared spectroseopy. J. Phys.
Chem. A 107(18):3384-3396.
21. Heyne K, Huse N, Nibbering ETJ, & Elsaesser T (2003) Ultrafast coherent nuclear
motions of hydrogen bonded carboxylic acid dimers. Chem. Phys. Lett. 369(5-6):591-596.
22. Shim S-H, Strasfeld DB, Fulmer EC, & Zanni MT (2006) Femtosecond pulse shaping
directly in the mid-IR using acousto-optic modulation. Opt. Lett. 31(6):838-840.
23. Shim S-H, Strasfeld DB, & Zanni MT (2006) Generation and characterization of phase
and amplitude shaped femtosecond mid-IR pulses. Opt. Express 14(26):13120-13130.
24. Shim SH, Strasfeld DB, Ling YL, & Zanni MT (2007) Automated 2D IR spectroscopy
using a mid-IR pulse shaper and application of this technology to the human islet amyloid
polypeptide. Proc. atl. Acad. Sci. U.S.A. 104(36):14197-14202.
108
25. Grumstrup EM, Shim S-H, Montgomery MA, Damrauer NH, & Zanni MT (2007)
Facile collection of two-dimensional electronic spectra using femtosecond pulse-shaping
technology. Opt. Express 15(25):16681-16689.
26. Khalil M, Demirdoven N, & Tokmakoff A (2003) Obtaining absorptive line shapes in
two-dimensional infrared vibrational correlation spectra. Phys. Rev. Lett. 90(4):047401.
27. Asbury JB, Steinel T, & Fayer MD (2004) Vibrational echo correlation spectroscopy
probes of hydrogen bond dynamics in water and methanol. J. Lumin. 107(1-4):271-286.
28. Dugan MA, Tull JX, & Warren WS (1997) High-resolution acousto-optic shaping of
unamplified and amplified femtosecond laser pulses. J. Opt. Soc. Am. B 14(9):2348-2358.
30. Shim S-H, Strasfeld DB, Ling YL, & Zanni MT (2007) Automated 2D IR spectroscopy
using a mid-IR pulse shaper and application of this technology to the human islet amyloid
polypeptide. Proc. atl. Acad. Sci. U.S.A. 104(36):14197-14202.
32. Witte T, Yeston JS, Motzkus M, Heilweil EJ, & Kompa K-L (2004) Femtosecond
infrared coherent excitation of liquid phase vibrational population distributions (v > 5).
Chem. Phys. Lett. 392:156-161.
33. Zanni MT, Ge NH, Kim YS, & Hochstrasser RM (2001) Two-dimensional IR
spectroscopy can be designed to eliminate the diagonal peaks and expose only the
crosspeaks needed for structure determination. Proc. atl. Acad. Sci. U.S.A.
98(20):11265-11270.
34. Volkov V, Schanz R, & Hamm P (2005) Active phase stabilization in Fourier-transform
two-dimensional infrared spectroscopy. Opt. Lett. 30(15):2010-2012.
35. Yan LM, Tatarek-Nossol M, Velkova A, Kazantzis A, & Kapurniotu (2006) A. Design of
a mimic of nonamyloidogenic and bioactive human islet amyloid polypeptide (IAPP) as a
nanomolar affinity inhibitor of IAPP cytotoxic fibrillogenesis. Proc. atl. Acad. Sci.
U.S.A. 103:2046-2051.
36. Ernst RR, Bodenhausen G, & Wokaun A (1987) Principles of uclear Magnetic
Resonance in One and Two Dimensions (Oxford University Press,, Oxford).
37. Strasfeld DB, Ling YL, Shim S-H, & Zanni MT (2008) Tracking fibril formation in
human Islet amyloid polypeptide with automated 2D-IR spectroscopy. J. Am. Chem. Soc.
130(21):6698-6699.
109
38. Tian PF, Keusters D, Suzaki Y, & Warren WS (2003) Femtosecond phase-coherent
two-dimensional spectroscopy. Science 300(5625):1553-1555.
39. Vaughan JC, Hornung T, Stone KW, & Nelson KA (2007) Coherently controlled
ultrafast four-wave mixing spectroscopy. J. Phys. Chem. A 111:4873-4883.
40. Goodno GD, Dadusc G, & Miller RJD (1998) Ultrafast heterodyne-detected transient-
grating spectroscopy using diffractive optics. J. Opt. Soc. Am. B 15:1791-1794.
41. Brixner T, Mancal T, Stiopkin IV, & Fleming GR (2004) Phase-stabilized two-
dimensional electronic spectroscopy. J. Chem. Phys. 121(9):4221-4236.
42. Zhang TH, Borca CN, Li XQ, & Cundiff ST (2005) Optical two-dimensional Fourier
transform spectroscopy with active interferometric stabilization. Opt. Express 13:7432-
7441.
43. Ding F, Mukherjee P, & Zanni M (2006) Passively correcting phase drift in 2D IR
spectroscopy. Opt. Lett. 31(19):2918-2920.
44. Xiong W & Zanni MT (2008) Signal enhancement and background cancellation in
collinear two-dimensional spectroscopies. Opt. Lett. 33(12):1371-1373
45. Brixner T & Gerber G (2001) Femtosecond polarization pulse shaping. Opt. Lett.
26(8):557-559.
46. Polachek L, Oron D, & Silberberg Y (2006) Full control of the spectral polarization of
ultrashort pulses. Opt. Lett. 31(5):631-633.
47. Strasfeld DB, Shim S-H, & Zanni MT (2007) Controlling vibrational excitation with
shaped mid-IR pulses. Phys. Rev. Lett. 99(3).
49. Ding F & Zanni MT (2007) Heterodyned 3D IR spectroscopy. Chem. Phys. 341(1-3):95-
105.
110
Chapter 5
Defining the Pathway of Amyloid Formation of human Islet
Amyloid Polypeptide with Residue-Specific Resolution
5.1 Introduction
More than twenty different diseases are associated with proteins that form insoluble amyloid
fibers (1-3). In large quantities, organ function is disrupted by the formation of amyloid deposits,
but for several amyloid diseases there is evidence that the toxic entities are actually pre-fibril
intermediates (4-6). However, details about these critical intermediates have been elusive,
mostly because it is extraordinarily difficult to obtain structural and kinetic information for
amyloid aggregation. The difficulty arises because high-resolution techniques do not have the
time-resolution required to track the structural changes nor can they be easily applied to
aggregating systems. Optical techniques that do have sufficient time-resolution, such as circular
inhibitors that subvert the pathway of amyloid formation. Inhibitors which intervene early in the
process are particularly desired because they have the potential to prevent both fibril formation
and toxic prefibril intermediates (4-6). What is needed is a technique with sufficient time-
A technique that satisfies these criteria is two dimensional infrared (2D IR) spectroscopy
when used with site-specific isotope labeling (7). In chapter 4, we demonstrated a new
111
studying amyloid formation (8). Our method uses a mid-IR pulse shaper to automate data
collection so that spectra can be collected quickly enough that fibril kinetics can be monitored
on-the-fly. In this chapter, we combine this automated version of 2D IR spectroscopy with site-
13 18
specific C O labeled peptides. Isotope labeling permits the kinetics of individual residues to
resulting kinetics and secondary structure information provides a detailed pathway for the
the pathway of amyloid formation in the 37-residue human islet amyloid polypeptide (hIAPP), a
peptide associated with type II diabetes. The deposition of amyloid fibrils in the islets of
Langerhans of the pancreas is a common pathological feature of type 2 diabetes and hIAPP is the
major protein component of these amyloid deposits (9-14). The polypeptide hormone is stored
with insulin in the secretory granules of the pancreatic β-islet cell and is secreted to the
extracellular space in response to the same factors that cause the release of insulin. Synthetic
amyloid aggregates are toxic to insulin-producing β-cells, arguing that hIAPP fibril formation
could play a role in the loss of β-cell mass in type 2 diabetes. The extent of amyloid deposition
appears to correlate with the severity of the disease, further highlighting a relationship between
amyloid deposition in the pancreas and the progression of the latter stages of type 2 diabetes (11-
14).
112
5.2 Experimental
Peptide synthesis and purification. Peptides were synthesized on a 0.25 mmol scale using an
form an amidated C-terminal. Standard Fmoc reaction cycles were used. The first residue
attached to the resin, β-branched residues, residues directly following β-branched residues and
pseudo-prolines were double coupled. Crude peptides were oxidized by dimethyl sulfoxide
(DMSO) for 24 hours at room temperature (16). The peptides were purified by reverse-phase
HPLC using a Vydac C18 preparative column. Analytical HPLC were used to check the purity of
the peptides before each experiment. The identity of the pure peptides was conformed by mass
spectrometry using a Bruker MALDI-TOF MS. 13C18O labeled Fmoc protected amino acids were
prepared as described (17). The level of 18O enrichment was at least 91% or greater and the 13C
enrichment was 98%. The high level of enrichment is critical to maximize signal to noise and to
avoid complication with 13C16O signals. The peptide was lyophilized in D2O and stored as a dry
powder.
at a concentration of ~0.8 mM. An aliquot of the d-HFIP stock solution was evaporated under a
stream of N2 gas to generate a film. Amyloid formation was initiated by re-dissolving the film in
mM.
113
Transmission Electron Microscopy (TEM). TEM images were taken of the 3 µL aliquot
of fibril solutions generated from three peptides that are unlabeled, Ala13, and Val32 labeled.
Fibril samples were prepared in the same way for the 2D IR spectra after incubating at least 3
days to complete aggregation. The aliquot was loaded onto a polyvinyl butyral coated copper
grid and stained with methylamine tungstate (Nanoprobes, Inc.). Excess stain was blotted off and
the sample was allowed to air dry. Images were acquired using a Philips CM 120 microscope
during aggregation of the peptide labeled at Ala13 using a Hitachi F-4500 fluorescence
spectrometer. The fluorescence shift due to binding of Thioflavin T to β-sheets was measured
with excitation wavelength at 450 nm and emission at 482 nm. Concurrent infrared absorption
data was taken using Bruker Optics Tensor 27 Fourier-transform infrared (FTIR) spectrometer.
The fluorescence data and growth of the β-sheet peak at 1618 cm-1 in the FTIR spectra are
compared in Fig. 5.2. Both the fluorescence and the FTIR data were measured with an average
time window of 100 seconds. The excellent agreement in the two different spectroscopic
measurements indicates that the unlabeled β-sheet feature at 1618 cm-1 is still an accurate
Automated 2D IR spectroscopy. The aggregation process, which took place between two CaF2
windows separated by a 100 µm Teflon spacer, was observed by the rapid scanning of four
phase-cycled pulse sequences. We create these 2D-IR pulse sequences using a Ge acousto-optic
modulator (AOM) based pulse shaper, as has been reported previously (18, 19). Briefly,
114
femtosecond mid-IR pulses were initially generated by difference frequency mixing (DFM)
the two femtosecond near-IR outputs of a BBO-based optical parametric amplifier (OPA). A
ruled grating (150 g/mm) was then used to disperse the mid-IR pulses into the frequency domain.
The Ge AOM was used to modulate the phase and intensity of the frequency profile so as to
create the desired, phase cycled pulse pair. Rather than implementing a pump-probe experiment
improve signal to noise and to get rid of the scatter from pump pulses, we took the difference in
transmission generated from four pump pulse trains with absolute phase shifted by π so that
∆OD = −ln(transmissionpump on, φ1 =0, φ2 =0/transmissionpump on, φ1 =π, φ2 =0) − ln(transmissionpump on,
φ1 =π, φ2 =π/transmissionpump on, φ1 =0, φ2 =π). The waveform generator for the pulse shaper can
continuously cycle 426 pulse shapes at maximum, thus we can scan 106 time delays when using
four phase cycles. To scan over >2 ps of the first coherence time with the 106 time delays, the
relative phase of the pump pulse pair was cycled with a rotating frame frequency of 1225 cm-1,
which enables regular sampling with a 22 fs step. It took 0.43 seconds to measure a single 2D IR
spectrum at a 1 kHz repetition rate. A running average was then performed in the same way for
all experiments until adequate signal to noise was achieved. The 2D IR spectra shown in Figs.
5.4, 5.6, 5.7, and 5.8 are averaged over a 9 minute window while the kinetic traces are averaged
over a 18 minute window. There is ~2 minute dead-time between initiation of the aggregation
Isotope dilution. Samples of mature fibers, whose isotope-label diluted to 25%, were prepared
by mixing the d-HFIP stock solutions of the labeled and the unlabeled peptides, evaporating d-
two CaF2 windows, and incubating in room temperature for 2 hrs at least. The concentration
of the labeled peptides was 1 mM for both the 25% and the 100% label samples. Three sets of
samples, along with a 100% unlabeled peptide, were prepared in the same way for Ala13, Ala25
and Leu27. Each set composed of a 100% and a 25% labeled peptide. >5,000 of 2D IR spectra
for the seven samples were collected, averaged and sliced through the diagonal to give Fig. 5.10.
The diagonal intensities of each sample were normalized with the intensity of the unlabeled β-
sheet feature at 1618 cm-1, then subtracted from the diagonal slice from the 100% unlabeled
sample.
the wild-type hIAPP in the full length (37 residues). Many of previous studies rely on small
peptide fragments, which do not necessarily translate into the context of the full protein. For
example, studies using fragments of hIAPP proposed that residues 20 through 29 form the core
of the cross b-structure of the fiber (20), but recent solid-state NMR (ssNMR) studies of full-
length hIAPP reveal that this is not the case (21). Furthermore, the kinetics of amyloid formation
by short fragments of hIAPP differs dramatically from that observed for the full length hormone.
The same difficulties arise from studies of fragments of the Alzheimer’s Aβ1-40 and Aβ1-42
peptide. For these reasons, we have chosen the full 37-residue peptide.
The polypeptide hormone is stored with insulin in the secretory granules of the pancreatic
β-islet cell in a concentration of 0.5~2 mM. Such a high concentration is also necessary for a
116
single carbonyl (i.e. a single residue isotope-labeled) to produce reasonable signal strength at
the amide I mode in FTIR and 2D IR spectra. To begin with unstructured state in such a high
concentration, amyloid formation was initiated by the addition of buffer to dried films of
unstructured hIAPP. According to transmission electron microscopy (TEM) images (Figs. 5.1A
to C), the fibrils produced using this protocol have the same structure as those formed using
other methods (22). Figs. 5.1A to C show a common morphology of long fibrils with a mean
diameter 10 nm for all the three fibril samples prepared from three differently labeled peptides
(unlabeled, labeled at Ala13 and labeled at Leu27). Moreover, kinetic curves generated from
Thioflavin T (ThT) fluorescence and from IR absorbance of the β-sheet are sigmoidal, as is
typical of amyloid kinetics (Figs. 5.2B). The fluorescence shifts due to binding of ThT to β-
sheets, thus increasing trend of the emission measured at 482 nm reflects the growing
populations in fibril. The amide I band of the FTIR spectrum right after initiation gave a broad
peak centered at 1644 cm-1, indicating that the dominant structure was random coil at the
beginning. As aggregation progressed, the random coil feature at 1644 cm-1 decreased and the
antisymmetric stretch of β-sheets located at 1618 cm-1 increased (Fig. 5.2A). Thus, FTIR spectra
indicate the structural progression of the peptide from the soluble state in random coil to the
Concurrent to the fluorescence experiment, TEM images were taken from four aliquots
removed from the sample, diluted and flash-frozen at four times during aggregation to monitor
morphological changes (Figs. 5.3). The TEM image taken immediately after initiation of the
fibers (Fig. 5.3B). At t = 20 min, the kinetics curve began to leap and we observed large
spherical oligomers in diameters of 20~35 nm as well as short fat fibers (~20 nm wide) (Fig.
117
5.3C). After the kinetics curve slightly passed the time-to-half-maximum (t50), the TEM
snapshot showed aggregates of short fat fibers and networks of long thin fibers (final fiber
structure) concentrated mostly on the left and right of Fig. 5.3D, respectively, free from spherical
oligomers. After ~4 hrs, the amyloid formation was completed and only the final fibers in a ~10
nm diameter were found in Fig. 5.3E similar to the fiber structures shown in Fig. 5.1. From the
oligomers whose their diameter increase to reach a critical point that interconnect the granules
into short fat fibrils, which transform into the final structure of long and thin fibrils. Therefore,
not only the secondary structure evolves, but also the macroscopic morphologies progress during
amyloid formation.
118
synthesized peptides with no labeling(A) and with isotope-labeling at Ala13(B) and Leu27(C).
The three TEM images share common motif of fibrils in a diameter of ~10 nm and lengths that
A B C
100 nm
119
500 µM in 20 mM phosphate buffer (pH 7.0). (A) FTIR spectra of the amide I band at t = 4 min
(solid), 60 min (dotted), 120 min (dashed) and 300 min (thick solid). (B) Kinetics curves from
intensities of the anti-symmetric β-sheet feature at 1618 cm-1 and fluorescence intensities of
Thioflavin T.
0.03
A
absorption (a.u.)
0.02
4 min
60 min
0.01
120 min
300 min
0.00
1580 1600 1620 1640 1660 1680 1700
frequency (cm-1)
1.0
B
Normalized Intensity
0.8
0.6
0.4
0.2
ThT fluororescence
IR absorbance of the β-sheet
0.0
Figure 5.3 Concurrent ThT fluorescence and TEM snapshots during amyloid formation
of hIAPP isotope-labeled at Leu 27. (A) Kinetics curve generated from fluorescence. Circles
indicate the time points of the four aliquots removed from the sample for TEM snapshots. (B to
E) TEM snapshots taken from the four samples removed at t = 3, 20, 57 mins and ~4 hrs after
initiation of amyloid formation. All four images are scaled in the same way.
A
fluorescence intensity
1.0
E
0.8
D
normalized
0.6
0.4
0.2 TEM
B C
0.0
ThT Fluorescence
0 20 40 60 80 100 120 140 160
time after mixing (min)
B C
100 nm
D E
121
unlabeled hIAPP
The FTIR spectra collected from hIAPP transforming into amyloid fiber contain all secondary
structural information during the process. However, only the β-sheet feature could produce a
kinetics curve without deconvolution, which is prone to error. The random coil feature at 1644
cm-1 is hard to separate from the final IR spectrum of the fiber (Fig. 5.2A), which contains
unignorable amount of absorbance around 1644 cm-1. This is a common problem of one-
kinetics since the spectral features of interest all absorb in the same wavelength range. If the
overlapped spectral features are spread over two dimensions, they could be separately monitored
without erroneous deconvolution. 2D IR spectroscopy has the potential to improve our structural
knowledge of amyloid formation due to its increased structural resolution and its femtosecond
time resolution.
spectroscopy to amyloid formation. First, amyloid fibers as well as intermediates scatter light,
which creates a large background signal. Second, the aggregation kinetics of amyloids is poor in
reproducibility, so that it is not possible to initiate the reaction over hundreds times to collect 2D
IR spectra at various stages of amylodosis. We overcome these two obstacles by using the
automated method for 2D IR spectroscopy using a mid-IR pulse shaper. Our method removes
scatter by incrementing or cycling phase described in Sec. 4.5.3 and collects a single 2D IR
spectrum in <1 sec as explained in Sec. 4.6. Thus, we can directly track the amyloid formation
of hIAPP on-the-fly.
122
Shown in Figs. 5.4A to C are three representative 2D IR spectra generated from many
thousands of spectra collected during fiber formation. Each peak in a FTIR spectrum produces a
pair of peaks in the opposite sign in a 2D IR spectrum because both fundamental and overtone
bands are probed. At the beginning of the amyloid formation, the most noticeable feature is an
out-of-phase doublets with the negative peak at ωpump = ωprobe = 1644 cm-1. The peak position
and shape (broad, diagonally elongated) indicate that the peptides are predominantly random coil
with a wide distribution of structures. As aggregation proceeds, the random coil doublet gives
way to a doublet at ωpump = ωprobe = 1618 cm-1 with a narrow linewidth of 15 cm-1 that is created
by the antisymmetric stretch of the β-sheet amyloids. At ωpump = ωprobe = 1673 cm-1, the β-sheet
symmetric stretch gives a weak doublet. IR absorbance between 1618 and 1673 cm-1 come from
Populations of random coil and β-sheet were monitored from the two features in the 2D
IR spectra without deconvolution. Kinetics curves in Fig. 5.4D were generated from intensities
of two features in 2D IR spectra: fundamental sequence peak of β-sheet at ωpump = ωprobe = 1618
cm-1; overtone sequence peak of random coil at ωpump = 1644 cm-1 and ωprobe = 1624 cm-1. The
two features are marked with a solid and a dashed arrow on Fig. 5.4B for the β-sheet and random
coil features, respectively, chosen for the kinetics curves. At the position for the random coil, β-
sheets contribute negligibly, as can be seen at t = 120 min, when the fibers mature and the
overtone intensity peak has gone away. The kinetics curves of the two features (Fig. 5.4C) are
sigmoidal, typical of amyloid kinetics, as in kinetics curves from ThT fluorescence and FT IR
spectra (Fig. 5.2B). However, the kinetics curves for β-sheet and random coil populations differ
in the t50, the rate of rise and the length of lag phase. The difference between the two kinetics is
more noticeable when normalized and smoothed as in Fig. 5.4E. In our previous study of hIAPP
123
amylodosis in pure D2O at pH ~6, the kinetics of random coil and β-sheet were also
significantly different, although both of them had no lag phase. The random coil kinetics fit to a
The different kinetics of random coil and β-sheet indicate that the amyloid formation
cannot be explained with a simple two-state model. As shown in Fig. 5.4E, the lag phase of
random coil is almost half of the β-sheet lag phase, indicating that some intermediates exist
during the intermission. The intermediates, neither in random coil nor in the final β-sheets, may
consume random coils, then remain for ~20 minutes, and then transform into mature fibrils. In
Sec. 5.3, TEM images revealed intermediate morphologies including spherical oligomers and
short fat fibrils that disappear once the fibril fully mature. These macroscopic intermediates may
have secondary structures emitting unnoticeable amount of signal or overlapping with the
dominant features in the 2D IR spectra, but contributing to the lag phase and the rise time of the
kinetics curves of random coil and β-sheet. In Aβ40 amylodosis, a ssNMR study suggested that
random coils aggregate into small spherical oligomers, then transform into large spherical
oligomers (15-35nm) composed of β-sheets prior to the final fibril state (23). To identify
intermediate structures of hIAPP amylodosis, we need to further resolve the amide I band of the
13
peptide. Isotope-labeling with C=18O separates a single residue or a number of residues in
frequency by ~50 cm-1 away from the rest of peptide backbone composed of 12
C=16O. In the
Figure 5.4 Representative 2D IR spectra and kinetics curves of hIAPP during aggregation.
(A to C) The 2D IR spectrum at (A) t = 5 min, (B) 40 min, (C) 120 min. (D) Kinetics of the
diagonal peaks of the β-sheet at ωpump=ωprobe=1617 cm-1 and the random coil at ωpump=1645 cm-1
and ωprobe=1624 cm-1. 500 spectra were averaged. (E) Normalized and smoothed kinetics curves
from the kinetics curves of the β-sheet and random coil features in D. 2,000 spectra were
1680 A B C
1660
ω1 (cm-1)
1640
1620
1600
1580 1620 1660 1700 1620 1660 1700 1620 1660 1700
ω3 (cm-1) ω3 (cm-1) ω3 (cm-1)
50
D 100 E
40
normalized intensity
peak intensity (a.u.)
80
30 60
β-sheet β-sheet
20 40
random coil random coil
20
10
0
0
0 20 40 60 80 100 120 140 160 180 0 20 40 60 80 100 120 140 160 180
time (min) time (min)
125
of isotope-labeled hIAPP
To monitor the structures and kinetics of individual residues in hIAPP, we use 13C18O labeling of
the backbone carbonyl groups to resolve single amide I stretches (24). Isotope labeling and IR
spectroscopy has been used previously to study amyloids, which have provided insight into
residue (21). The positions of the six labeled residues are shown in Fig. 5.5 superimposed upon
the recent model of the hIAPP protofibril deduced from solid state NMR (21). hIAPP forms a
cross β-strand structure that is composed of two twisted columns of peptides. According to the
ssNMR studies, each column is created from a stack of peptides that are folded into a structure
containing two β-strands with an ordered loop between residues 18 and 27. Hydrogen bonding
occurs along the stacks so that there are four β-strands running in parallel along the lengths of
the fibrils. The labeled residues are spaced along the peptide (Fig. 5.5) and the positions were
chosen to provide probes throughout key regions of the fibril. Note that residues 1-7 do not form
part of the ordered core of the fibril, in part because there is a disulfide bond between Cys2 and
Cys7 that is not compatible with β-sheet structure. Since the peptides are stacked, each isotope
label forms two columns, illustrated in Fig. 5.5C for Ala25. The kinetics we report below
Figure 5.5 Structure of hIAPP fibrils according to solid state NMR. (A) Sequence of
hIAPP with the six isotope-labeled residues highlighted: Ala8 (yellow), Ala13 (blue), Val17
(purple), Ala25 (red), Leu27 (cyan), and Val32 (green). (B) Cross-section through a fibril,
illustrating that the fibrils are composed of two columns of peptides to create four β-strands (that
run in and out of the page). The labeled residues are highlighted. (C) Side-view of the
protofibril illustrating how a peptide with a single labeled residue creates two columns of isotope
A
1 6 11 16 21 26 31 36
KCNTATCATQRLANFLVHSSNNFGAILSSTNVGSNTY
B C
127
Shown in Fig. 5.6 are four 2D IR spectra collected at successively longer times during amyloid
formation for a sample of hIAPP labeled at Ala25. The four spectra are generated from many
thousands of spectra that were collected. The first 2D IR spectrum in Fig. 5.6 is at t = 5 min,
which is the first averaged 2D IR spectrum collected after aggregation is initiated. The most
prominent features in these spectra are between 1617 and 1670 cm-1 that are due to the unlabeled
residues (inside the dashed black squares) as shown in Sec. 5.4. The unlabeled peaks give
information on the overall assembly kinetics of the amyloid (8). At t = 5 min., the unlabeled
features consists of two out-of-phase peaks at ωpump = 1645 cm-1, typical of random coil peptide
structures with large structural distributions. The isotope labeled features appear near 1580 cm-1
(inside the solid black squares), which provides information specific to the folding kinetics of
Ala25. At t = 5 min., the isotope labeled absorption is very broad and weak, indicating that
Ala25 is conformationally disordered, which is consistent with the nature of the random coil
state.
To highlight the changes which occur in the 2D IR spectra during the folding processes,
the remaining 2D IR spectra in Fig. 5.6 are plotted as difference spectra, calculated by
subtracting the first 2D IR spectrum in Fig. 5.6A from the others. As time progresses, the
random coil doublet disappears while a doublet grows in at ωpump = 1618 cm-1, the signature of β-
sheet amyloid. Concurrent with the growth of the β-sheet, two isotope labeled features appear at
1574 and 1585 cm-1. The lower frequency peak eventually becomes more intense (Figs. 5.6D
and 5.7A). Cross-peaks also grow in between the isotope labels and the unlabeled β-sheet peak
at 1618 cm-1, indicating that Ala25 is strongly coupled to the β-sheets. Thus, the cross peaks
measure the kinetics of A25 folding into the β-sheet configuration of the amyloid fibril. Figure
128
5.7B plots the intensity of the peaks for the unlabeled β-sheet, Ala25, and the cross peak as a
function of time. All three kinetic curves are sigmoidal, as is typical of amyloid kinetics, and all
three have a time-to-half-maximum (t50) that is virtually identical, indicating that when Ala25
becomes part of the ordered fibril, it assembles directly into a β-sheet structure.
129
31 min, (C) 66 min and (D) 187 min, calculated by subtracting the t = 5 min. spectrum. Insets
highlight the diagonal peaks from 13C18O Ala25. Contours and color scales are set in the same
way for B to D. Black boxes surround spectral features of the unlabeled portion of the peptide
while red boxes enclose the diagonal peaks of the isotope labeled Ala25. Blue and green arrows
highlight the two labeled features, while the red arrows points to the cross peak between the
13 18
C O Ala25 and the unlabeled β-sheet.
10 5 0 −5 −10
1680 A
ωpump (cm-1)
1640
1600
1560
1560 1600 1640 1680
ωprobe (cm-1)
4 3 2 1 0 −1 −2 −3 −4
1680
B C D
ωpump (cm-1)
1640
1600
1560
1560 1600 1640 1680 1560 1600 1640 1680 1560 1600 1640 1680
ωprobe (cm-1) ωprobe (cm-1) ωprobe (cm-1)
1600 1600 2 1600
1
1580 1580 0 1580
−1
1560 1560 −2 1560
1560 1580 1600 1560 1580 1600 1560 1580 1600
130
Figure 5.7 Kinetics curves of hIAPP labeled at Ala25 generated from 2D IR spectra during
aggregation. (A) Kinetics of the diagonal peaks of the unlabeled β-sheet at 1617 cm-1 and the
two label features (black and blue arrows). (B) Kinetics of the cross peak are compared to the
diagonal peaks. The left y-axis is for the β-sheet feature at 1617 cm-1 and the right y-axis is for
other features.
200
16
A B
β-sheet Intensity
160
label Intensity
12
120
8
80
β-sheet β-sheet 4
40 label (1583 cm -1) label (1583 cm -1)
label (1576 cm -1) crosspeak
0 0
0 20 40 60 80 100 0 20 40 60 80 100
t (min) t (min)
131
The reproducibility of our experiments are particularly important issues with regards to amyloid
formation and this is one of the first detailed studies of amyloid kinetics using 2D IR
spectroscopy. It is well-known that amyloid kinetics are not perfectly reproducible from one
experiment to the next because they are extremely sensitive to conditions which are difficult to
control, such as how rapidly the solution is mixed (27). Procedures for comparing separate data
sets have been established using fluorescence and CD spectroscopy. Our procedure is similar to
these well-established methods, but accounts for issues specific to infrared spectroscopy. We
with the same peptide labeled at Ala25 prepared in the same condition. Shown in Fig. 5.8A are
the kinetic traces of the unlabeled β-strand feature from four separate 2D IR experiments on the
same sample of hIAPP labeled at Ala25. The time-to-half-maximum (t50) times for these 4
experiments fall between 30 and 60 min. This large range of times prohibits the averaging of
fluorescence studies that kinetic scans on wide ranging timescales can be compared by scaling
the time axes with their respective t50, so long as the nucleation mechanism is unchanged.(28)
Following this procedure, we first established the value of t50 for each scan by fitting the kinetic
traces associated with the spectral features due to the unlabeled 12C16O oscillators to a sigmoidal
function
1 1
f (t ) = [r1 − m1 exp(−t / τ )] + (r1 + m1t )1 −
1 + exp(−(t − t 50 ) / σ ) 1 + exp(−(t − t 50 ) / σ )
(5.1)
132
where the exponential decaying function of r1− m1exp(−t/τ) defines the upper baseline and
the linear equation r1+ m1t defines the lower baseline. Flat baselines were used for the kinetics
curves with low signal-to-noise ratio. Fits were performed using a nonlinear algorithm in
MatLab. For fitting, we used 2D IR data averaged over a 1 minute window. (In Fig. 5.8, we
smoothed the data (dots) and the fitted curves (solid lines) with a 18 mins window for
visualization purpose after fitting.) We then used the t50 determined by fitting to scale the time
axes.
On this scaled axis, the sigmoidal curves are quite similar (Fig. 5.8B). The similarity of
the slopes indicate that all 4 experiments are monitoring the same aggregation mechanism, as is
expected from identical samples. Having determined the scaling parameters from the unlabeled
β-strand features, we then scale the time course of the spectral features associated with the
labeled site using the same parameters. For example, Fig. 5.8C shows the time course of the
Ala25 signal at 1585 cm-1 for the same 4 experiments after scaling with the t50 times used above,
along with the sigmoidal fits. The signal-to-noise of the label kinetics is lower than the
unlabeled β-strand features because the label intensities are about 13 times weaker.
To account for signal-to-noise, we calculate the error in the ratio of t50s between the label
and the β-strand features. To estimate the error, the vector containing standard errors of the
fitted parameters (σ) were calculated from the sum of (residuals)2 and Jacobian matrix (J) using
where “residuals” are the difference between the original data and the fitted data and J is the
matrix of all the first-order partial derivatives of the fitting function with all the fitted parameters.
The scaled data is calculated by dividing by the t50 of the unlabeled β-sheet feature
133
tl
tn = (5.3)
tb
where tn is the normalized t50, tl is the t50 of the labeled kinetics and tb is the t50 of the unlabeled
β-sheet kinetics. The standard error in the scaled t50 times (σn) accounts for error in both the
2
dt n dt n
2
2
σb
2
2 σl
σ = σ l + σ b = tn +
2
n (5.4)
dtl dtb tl tb
The unlabeled β-sheet kinetic traces usually have a slow rise after the fast transition,
which is commonly observed in amyloid studies using ThT fluorescence (28). The variation of
the slow rise between experiments is illustrated in Fig. 5.8D for the four Ala25 trials. Our fitting
function accounts for this rise with a simple exponential function and the fits predict that it has a
minor contribution to the t50 of the sigmoidal transitions because it occurs on a much longer
timescale than the fast sigmoidal kinetics. The difference in time scale is readily apparent in the
Ala25 data for which both isotope label peaks exhibit nearly identical fast kinetics but very
different slower kinetics. Even so, fits give both Ala25 peaks nearly identical t50 times of 49.24
and 49.51 mins for 1585 cm-1 and 1574 cm-1 peaks, respectively, for the experiment shown in
Figs. 5.6 and 5.7. Of course when the data is plotted in the figures the kinetic traces must be
normalized. We chose the relative time t/t50 =2 (e.g. Fig. 5.7, 5.8 and 5.12). The choice of
normalization time has no bearing on the reported t50 data which comes directly from the fits
themselves. The choice of normalization does slightly alter the slopes of the kinetic traces at t50,
but we know that the effect is small by comparing the sigmoids from the fits with the additional
134
exponential function, shown in Fig. 5.8D for the 4 Ala25 trials. Thus, our fitting routine is
robust with regards to the t50 and the slopes visualized in the manuscript figures are accurate.
The resulting ratios of t50 and their standard error bars (σn) are plotted in Fig. 5.8E. All 4
measurements give a very similar ratio, with much smaller error bars associated with the higher
quality data sets, as expected. Taken together, these 4 measurements give a combined t50(label)/
Figure 5.8 Summary of experiments using Ala25 labeled peptides to test reproducibility
and data analysis method. (A) Kinetics of unlabeled β-strand features at 1617 cm-1 for 4
independent experiments (dots), along with sigmoidal fits (solid line) using Eq. 5.1. Data and
fits are smoothed with a 18-min. window after performing fits on data averaged with a 1 minute
window. (B) Curves from (A) scaled by their respective t50 times. (C) Kinetics of Ala25
feature at 1585 cm-1, using the same t50 times along with sigmoidal fits. (D) Curves in (B)
shown in longer time scale. The intensities are normalized to the fitted value at the relative time
t/t50 =2. The experimental data in dots are shown along with its fits in solid line with the same
color for each experiment. (E) t50 times and standard errors extracted from fits of isotope
kinetics shown in (C). Colors are assigned in the same way for each of the 4 experiments.
A B C
normalized intensity (a.u.)
1.2
1.0
0.8
0.6
0.4 exp1 exp1 exp1
exp2 exp2 exp2
0.2
exp3 exp3 exp3
0.0 exp4 exp4 exp4
-0.2
0 20 40 60 80 100 0.2 0.6 1.0 1.4 1.8 0.2 0.6 1.0 1.4 1.8
t/t50 t (min) t/t50
1.2
normalized intensity (a.u.)
D E
t50 (label) / t50 (unlabeled)
1.0
1.1
0.8
0.6
1.0
0.4
exp1
0.2 exp2 0.9
0.0
exp3
exp4
-0.2 0.8
0 1 2 3 4 1 2 3 4 Avg
t/t50 repeats of experiments
136
The kinetics of the other five labeled sites were also measured. They have similar spectral
features to Ala25 (Figs. 5.9 and 5.10), but strikingly different kinetics; the kinetic transition
measured for some labeled sites is more rapid than the transition monitored by the unlabeled β-
sheet peak, while others fold much slower. To better compare the relative folding times, we
have fitted the kinetics curves with sigmoids as in the previous section (Data and fits shown in
Fig. 5.11. Function given by Eq. 5.1), then scaled the kinetic curves for the unlabeled β-sheet
peaks from all 7 samples (6 labeled and 1 unlabeled peptides), so that their t50 times are equal
(Fig. 5.12A). Scaling is necessary because amyloid kinetics are not perfectly reproducible from
one experiment to the next, which is why our rapid-scan version of 2D IR spectroscopy is
necessary to accomplish these experiments. However, the kinetics from different experiments
can be compared by dividing the time-axes by the respective t50 measured for the unlabeled
peaks (28) as we demonstrated for multiple experiments of the same peptide labeled at Ala 25 in
the previous section. We find that when the axes are scaled, the unlabeled β-sheet peaks for all
six labeled peptides have virtually the same folding curve and that this folding curve matches
that observed for an unlabeled peptide without any isotope labels (Fig. 5.12A). The slopes show
slightly more variation (mostly because of Ala8 and Ala13) than for multiple runs of the same
sample (Fig. 5.8B), which may be caused by small differences in HPLC purification of the
peptides, but is not large enough to indicate a significantly different aggregation mechanism (28,
29). Thus, we conclude that the kinetics from individual 2D IR measurements can be compared
by following the standard analysis methods of fluorescence experiments by using the unlabeled
β-sheet feature as a reference. The β-sheet feature provides an internal standard which allows
one to deconvolve the normal run-to-run variations from the more interesting differences
137
associated with the kinetics of assembly. The presence of such a built in internal standard is
Having established the validity of our 2D IR approach, we now turn to comparing the
kinetics of all 6 labeled residues, which is the primary motivation for this study. The scaling
parameters in Fig. 5.12A are then used to compare the relative folding rates of the isotope labels,
which are shown in Fig. 5.12B. Simple visual inspection of this data reveals that these 6 residues
exhibit strikingly different kinetics; some residues transition slower than the unlabeled β-strand
feature, while others are faster. The data shown in Fig. 5.12B are solely presented to visualize the
differences in kinetics. Like Ala25, each labeled peptide was measured multiple times and the
t50 were calculated from the best 2 to 4 data sets (Fig. 5.12C). The averaged t50 times are plotted
in Fig. 5.12D. Fig. 5.12D exhibits a trend in which the first residues to transition are near the
middle of hIAPP sequence, followed by aggregation at the ends. Val17 is the first residue to
transition and has a t50 that is 10% sooner than the t50 of the unlabeled residues (in all
experiments performed, the Val17 isotope label rises before the unlabeled β-sheet features),
while Val32 is the last to transition with a t50 time that is 30% slower. On an absolute timescale,
this 40% difference in folding rates translates into a 20 min. difference between the folding of
From the scaled t50s of 6 residues, we estimated the order in which each residue
assembles into the final structure. The first residue to fold is Val17, on the N-terminal end of
the ordered loop, followed by Ala25 on the C-terminal end of the loop. The folding then
proceeds down the β-strands, although at unequal rates towards the two termini. hIAPP clearly
has a diverse and interesting folding landscape. The landscape must contain multiple folding
barriers to cause the different dynamics. Our previous study of unlabeled hIAPP (Sec. 5.4)
138
suggested a much simpler landscape, since the data could be largely explained with only an
initial random coil state and a final folded state (8). In principle, the shape and width of the
unlabeled β-sheet transition should reflect the different dynamics since it contains the kinetics of
all residues. However, since the unlabeled feature is a convolution of 37 amino acids, the
kinetics of 1 or a few residues with unique transition times will be obscured by the others. These
experiments demonstrate that when individual residues are monitored using isotope labeling,
details of the folding and assembly process are resealed that are otherwise masked by
As mentioned above, the quantitative determination of all t50 times was done using 2D
data sets smoothed with a 1 min. running average. For visualization purposes, the reported 2D
IR spectra (Figs. 5.6, 5.9 and 5.10) are smoothed with 9 min. windows and the kinetic traces use
an 18 min. window. Nonetheless, to understand the effects of smoothing on our results, we have
also calculated the averages and errors in t50 following the above procedure using various
smoothing windows. Shown in Fig. 5.13 is the labeled t50 data reported in Figs. 5.8B and 5.12D
that was calculated with 1 min. windows compared to the same data calculated from data
smoothed with 18 min. windows. Smoothing with 18 min. windows reduces the error bars and
tends to decrease the variation in the labeled t50 times (as would be expected since the time-
resolution is poorer), but scarcely alter the relative t50 times of the 6 labels. Therefore, we
conclude that reasonable amounts of smoothing do not alter our fundamental conclusions and we
have chosen 1 min. windows for data analysis, even though they give larger error bars, because it
Figure 5.9 Difference 2D IR spectra at the last stage of aggregation of the six peptides
labeled at Ala8(A), Ala13(B), Val17(C), Ala25(D), Leu27(E), and Val32(F). Each difference
2D IR spectrum was generated by subtracting the spectrum collected at the end of the reaction
from the first 2D IR spectrum of each experiment. Fig S1D is identical to Fig. 2D in the text.
Except for Ala 25, all the other five labeled peptides exhibit a single label feature (black arrows)
around 1580 cm-1 with their frequencies and lineshapes varying as their location in the final fibril
structure. Cross-peaks between the isotope labels and the unlabeled β-sheet peak at 1618 cm-1
are seen in all six labeled peptides (red arrows), indicating strong coupling of the label to the β-
sheets.
1700
1680 A B C
A8 A13 V17
ωpump (c m-1)
1660
1640
1620
1600
1580
1560
1680 D E F
A25 L27 V32
ωpump (c m-1)
1660
1640
1620
1600
1580
1560
1560 1600 1640 1680 1560 1600 1640 1680 1560 1600 1640 1680
ωprobe -1
(c m ) ωprobe -1
(c m ) ωprobe -1
(c m )
140
the six peptides labeled at Ala8(A), Ala13(B), Val17(C), Ala25(D), Leu27(E), and Val32(F).
The center column displays spectra at t = ~t50 of the label. The left column shows spectra at t <
t50 of the label, where as the right column was chosen at t > t50 of the label when the label
intensities begin to plateau. At the top-left corner of each spectrum, corresponding scaled time is
denoted by multiples of t50 of the unlabeled β-sheet. Each difference 2D IR spectrum was
generated by subtracting the spectrum collected at certain time from the first 2D IR spectrum of
each experiment.
141
1640
A8
1600
1560
0.6t50 1.1t50 1.6t50
B 1680
ω pump (c m -1)
1640
A13
1600
1560
0.7t50 0.9t50 1.1t50
C 1680
ω pump (c m -1)
1640
V17
1600
1560
0.7t50 0.9t50 1.5t50
D 1680
ω pump (c m -1)
1640
A25
1600
1560
0.6t50 1.1t50 1.6t50
E 1680
ωpump (c m-1)
1640
L27
1600
1560
0.9t50 1.8t50 4.2t50
F 1680
ωpump (c m -1)
1640
V32
1600
1560
1560 1600 1640 1680 1560 1600 1640 1680 1560 1600 1640 1680
ωprobe (c m -1) ωprobe (c m -1) ωprobe (c m-1)
142
Figure 5.11 Time course of the spectral features (black line) and their fits (red line) from
the best data of six labeled peptides at Ala8(A), Ala13(B), Val17(C), Ala25(D), Leu27(E), and
Val32(F). For each residue, the feature associated with unlabeled b-strand (left column) and the
13 18
C O label (right column) are shown. For visualization purpose, the data and fitted curves are
intensity (a.u.)
intensity (a.u.)
intensity (a.u.)
A8 120 6 A25 120
8
80 4 80
4
40 data 2 data 40 data data
fit fit fit fit
0 0 0
0
40 80 120 160 200 40 80 120 160 200 50 100 150 200 250 50 100 150 200 250
time (min) time (min) time (min) time (min)
B 300
8
E 400
intensity (a.u.)
intensity (a.u.)
intensity (a.u.)
intensity (a.u.)
A13 6
L27 300
30
200
4 20
200
100 2
data data 100 data 10 data
fit fit fit fit
0
0 0
50 100 150 200 50 100 150 200 40 80 120 160 200 40 80 120 160 200
time (min) time (min) time (min) time (min)
C 200 40 F 160 40
intensity (a.u.)
intensity (a.u.)
intensity (a.u.)
intensity (a.u.)
V17 150 30 V32 120 30
100 20 80 20
Figure 5.12 Summary of the kinetics for all 6 labeled peptides. (A) Scaled kinetic curves
for the unlabeled β-sheet feature at 1617 cm-1 for all 6 labeled samples plus an unlabeled peptide
sample obtained from the best experiment among 2 to 4 sets for each peptide. (B) Using the
same scaling and data set, comparison of kinetic curves for the isotope labeled peaks. (C) Ratio
of t50 of the label to t50 of the unlabeled β-sheet for the six residues measured with standard error
bars from the best 2 to 4 data sets. Blue circles represent the best experiment, used for in (A) and
(B), for each labeled peptide. Red, magenta and green asterisks are from other separate
experiments. (D) Averaged ratio of t50 of the label to t50 of the unlabeled β-sheet for the six
residues. All the data in (C) was used to produce the means and error bars.
normalized intensity
100
A B
80
60 A8
A13 A8
40 V17 A13
A25 V17
20 L27 A25
V32 L27
0 UL V32
0.2 0.6 1.0 1.4 1.8 0.2 0.6 1.0 1.4 1.8
normalized time (t/t50) normalized time (t/t50)
t50(label) / t50(unlabeled)
1.6
C D
1.4
1.2
1.0
0.8
5 10 15 20 25 30 35 10 15 20 25 30 35
residue number residue number
144
13
Figure 5.13 Effects of running average in the ratio of t50 between C18O label and the
unlabeled 12C16O for (A) four separate experiments on the same peptide labeled at Ala 25 and (B)
six separate experiments on six peptides labeled at six different residues. β-strand and label
kinetics from data generated in two different ways were separately fitted with Eq. S1 to generate
the ratio. Blue circles are from data averaged with 1 min window and red asterisks are from data
1.2 1.6
t50 of label / t50 of unlabeled
1.0 1.2
0.9 1.0
0.8 0.8
1 2 3 4 5 10 15 20 25 30 35
Repeats of the experiment residue number
145
The above experiments reports on local structure through the changes in intensity and
frequency of the isotope labeled amide I band. There are two factors that affect the frequency
and intensity the bands in the 2D IR spectra: their environment and their coupling. The
electrostatic environment around the atoms of the amide I mode (the backbone C, O, N, and H
atoms) influence its frequency. Electrostatics is largely influenced by solvation and hydrogen
bonding, which is why the amide I band has a lower frequency in water than in membrane bound
systems. As for coupling, the isotope labeled band is vibrationally coupled to nearby residues,
which can also shift its frequency. These couplings create the cross peak to the unlabeled amide
13 18
I mode. However, the coupling to other C O labels is more important for the observed
frequency, because the degeneracy of the modes leads to larger frequency shifts. To ascertain
the effects of electrostatics versus coupling, we collected 2D IR spectra of Ala13, Ala25 and
Leu27 where only 25% of the peptides contain an isotope label, thereby largely eliminating the
effects of coupling between labeled residues. (The spectra are reported in Fig. 5.10) We found
that upon isotope dilution, the frequencies of these three residues lie within 5 cm-1 of each other
(1589 to 1594 cm-1), indicating that their local electrostatics are comparable. Moreover, the
isotope frequencies are higher when diluted, indicating that the coupling between isotope labels
causes a negative frequency shifts, which is consistent with the negative coupling constants
predicted for β-sheets in amyloid fibril (26). These results emphasize that the isotope labeled
features that are the focus of this study arise from delocalized vibrational modes that involve the
cooperative motions of several labeled amide I groups. Thus, the kinetics experiments are not
monitoring the folding of a single peptide, but are instead measuring β-sheet formation through
Figure 5.14 Diagonal slices of 2D IR spectra taken for matured fibers of three peptides
labeled at Ala13(A), Ala25(B), and Leu27(C) in a 100% (solid) and 25% (dashed) concentration
of the isotope.
normalized intensity (a.u.)
20 A B C
A13 A25 L27
15 A13/UL A25/UL L27/UL
10
0
1570 1580 1590 1570 1580 1590 1570 1580 1590 1600
frequency (cm-1) frequency (cm-1) frequency (cm-1)
147
5.6 Discussion
Our data is consistent with the mechanism shown in Fig. 5.15 whereby a series of peptides
nucleate into a β-sheet conformation at Val17 on the N-terminal end of the ordered loop,
followed by the formation of the ordered loop to nucleate a second β-sheet at Ala25 on the C-
terminal end of the loop. The N-terminal β-strand then folds about twice as quickly as the C-
terminal strand. Some portion of the ensemble may be nucleating at Ala25 before Val17, since
the kinetic trace of Ala25 follows closely enough behind Val17 that the two kinetic curves
partially overlap (Fig. 5.13B). Also broadly consistent with the data would be nucleation at the
ordered loop, followed by folding down the N- and C-terminal β-strands. We do not yet have a
spectroscopic signature for when loop formation occurs or independent measures for each of the
two columns, but we suspect that the two columns of peptides grow at different rates. Different
growth rates would explain why the C-terminal sheet, which resides at the interface of the two
columns (see Fig. 5.5B), has slower kinetics than the outside N-terminal sheet e.g. the region
containing Val32 does not form β-sheet until it is stabilized by the second column of peptides.
In principle, more sophisticated isotope labeling schemes, like using pairs of labels that span
across the β-strands, could be used to test this hypothesis. Regardless, it is clear that nucleation
occurs at or near Val17, followed by elongation of the two parallel β-strands, with the N-
terminal β-strand forming more rapidly. The folding mechanism is somewhat reminiscent of the
zipper mechanism for β-hairpin folding. In the zipper mechanism, the β-turn forms first,
followed by elongation of the hairpin towards the N- and C-termini by rapidly forming a series
of hydrogen bonds (30, 31). However, we emphasize that the hydrogen bonding in amyloid is
not the same; instead of hydrogen bonds linking the two halves of each peptide individually,
148
hydrogen bonds in amyloids run along the fiber long axis to connect stacked hairpins. Thus,
fundamentally different forces are likely to be responsible for hairpin formation in amyloid.
Another very important point is that our data rules out the presence of large β-sheet structures
during the lag phase. We estimate that not more than 5% of the peptide ensemble can reside in
β-sheets with more than 4 strands, based on the background intensity at 1618 cm-1 where large β-
sheets absorb. β-sheet aggregates of 3 strands or less could be present during the lag phase,
because small β-sheets absorb near the random coil at 1645 cm-1 (32).
Of the six residues measured, Ala25 is particularly interesting because it is the only
substitution that exhibits two isotope labeled bands (See Figs. 5.6B to D for Ala25 and SI Figs.
5.9 and 5.10 for other residues). Isotope dilution of Ala25 causes both peaks to collapse into a
single feature (Fig. 5.14). This finding indicates that Ala25 contains two coupling constants of
3.5 and 15.5 cm-1. Two coupling constants indicate that there is a bimodal distribution of
distances between adjacent Ala25 residues. A bimodal distribution of distances could arise from
two different populations of fibril structures or from the helix twist which breaks the symmetry
between the two columns of peptides. The last possibility seems unlikely, since the twist is small
over the delocalization length of a vibrational exciton (4-6 strands), and thus should be negligible.
It seems more likely that there are two subpopulations of β-sheet structures and that the relative
ratio of the 1574 and 1585 cm-1 features gives their relative population. If true, then the fact that
the two features have similar intensities up to 1.5t50 indicates that the two subpopulations
aggregate in tandem, but that the subpopulation with the lower frequency isotope label (and, thus,
shorter Ala25 spacing) is eventually preferred (Fig. 5.7A). Considering that Ala25 is the only
residue that exhibits two peaks and the only residue we measured that was assigned to random
149
coil in the ssNMR structure, it appears that the ordered loop is the most sensitive region of
At this time, we do not know why nucleation occurs near Val17. However, some insight
may be gained by considering our results in the context of the well known correlation between
variations in the primarily sequence of IAPP and the prevalence of islet amyloid in different
species (20). For example, rodents do not form islet amyloid even though they produce IAPP.
The sequence of rat IAPP differs from that of hIAPP at six positions; residues 18, 23, 25, 26, 28
and 29. Four of these residues flank the ordered loop and are located close to the amino acids
that we postulate are the nucleation sites for β-sheet formation. It is also worth noting that
human IAPP as does a single mutation from Ile26 to proline (33, 34). Interestingly substitutions
outside of the 20-29 region but within the region we have identified as being involved with
nucleation also profoundly affected amyloid formation. A triple mutant in which residue 17, 19
and 30 were replaced by proline had a dramatically reduced tendency to aggregate (35). Thus
studies all highlight the putative importance of the region(s) we have identified as being critical
for nucleation. Our results have important implications for inhibitor development since they
provide the information required to design inhibitors which target the critical nucleation site.
Along these lines insulin and its isolated B-chain are known to be among the most potent
inhibitors of amyloid formation by IAPP and they are thought to interact predominately within
residues 7 to 19 of IAPP (36). Thus, rat and mouse IAPP might have evolved to prohibit
amyloid formation by mutating the first two sites in the aggregation pathway: Val17 and Ala25.
In principle, much of the information reported here could have been obtained using FTIR
spectroscopy. However, we have not succeeded at resolving the kinetics of the isotope labels in
150
these samples using FTIR spectroscopy. One advantage of 2D IR spectroscopy over FTIR is
that background signal from weak absorbers is suppressed because the intensity of 2D IR spectra
scales as |µ|4, where µ is the transition dipole, whereas FTIR scales as |µ|2. Thus, the non-linear
technology for collecting 2D IR spectra is that we can phase-cycle the pulses to help eliminate
scatter from these heterogeneous samples. But the real advantages of 2D IR spectroscopy will be
utilized in the future in experiments that monitor 2D lineshapes and quantify couplings which
have the potential to time-resolve solvent expulsion during fiber formation and identify the
Figure 5.15 hIAPP aggregation pathway that is consistent with our data. Labeled residues
in register at each assembly stage are denoted with their residue numbers and colored circles
(purple: Val17, red: Ala25, cyan: Leu27, blue: Ala13, yellow: Ala8, green: Val32). Thick black
arrows are β-strands, while the portion of the peptide in random coil or loop is shown in gray.
152
5.7 Conclusion
We have utilized isotope labeling and new technological advances in 2D IR spectroscopy to gain
residue specific kinetic information about the development of the aggregation pathway of hIAPP.
We found that the amyloid fibrils associated with type 2 diabetes follow a multistep pathway
involving a series of intermediate, largely β-sheet structures. The methodology presented here is
not limited to IAPP nor is it limited to amyloid formation in homogenous solution. Membrane
peptide interactions have been postulated to play an important role in amyloid formation (37) and
our methodology is well suited to investigate such systems. It is also applicable to a wide range
of other proteins now that advances in peptide synthesis have made it possible to synthesize large
polypeptides with unnatural or labeled amino acids (38). Furthermore, 2D IR spectra can be
simulated from X-ray, NMR or molecular dynamics structures. Thus, our methodology has the
potential to test and link disparate studies that together hold the promise of providing a
5.8 Acknowledgments
We thank Professor Daniel Raleigh and Dr. Ruchi Gupta in State University of New
York at Stony Brook for the collaboration and for providing the synthesized peptides.
153
5.9 Reference
2. Sparrer HE, Santoso A, Szoka FC, & Weissman JS (2000) Evidence for the prion
hypothesis: Induction of the yeast PSI+ factor by in vitro-converted Sup35 protein.
Science 289(5479):595-599.
3. Aguzzi A & Haass C (2003) Games played by rogue proteins in prion disorders and
Alzheimer's disease. Science 302:814-818.
4. Kayed R, Head E, Thompson JL, McIntire TM, Milton SC, Cotman CW, & G. GC (2003)
Common structure of soluble amyloid oligomers implies common mechanism of
pathogenesis. Science 300:486-489.
5. Kagan B (2005) Oligomers and cellular toxicity in amyloid proteins: The beta sheet
conformation and disease. (WILEY-VCH Verlag GmbH & Co., Weinheim).
6. Silveira JR, Raymond GJ, Hughson AG, Race RE, Sim VL, Hayes SF, & Caughey B
(2005) The most infectious prion protein particles. ature 437:257-261.
8. Strasfeld DB, Ling YL, Shim S-H, & Zanni MT (2008) Tracking fibril formation in
human Islet amyloid polypeptide with automated 2D-IR spectroscopy. J. Am. Chem. Soc.
130(21):6698-6699.
10. Cooper GJS, Willis AC, Clark A, Turner RC, Sim RB, & Reid KBM (1987) Purification
and characterization of a peptide from amyloid-rich pancreases of type 2 diabetic patients.
Proc. atl. Acad. Sci. U.S.A. 84:8628-8632.
11. Lorenzo A, Razzaboni B, Weir GC, & Yankner BA (1994) Pancreatic islet cell toxicity
of Amylin associated with type II diabetes mellitus. ature 368:756-760.
12. Kahn SE, Andrikopoulos S, & Verchere CB (1999) Islet amyloid: A long recognized but
underappreciated pathological feature of type II diabetes. Diabetes 48:241-246.
13. Ritzel RA, Meier JJ, Lin C-Y, Veldhuis JD, & Butler PC (2007) Human islet amyloid
polypeptide oligomers disrupt cell coupling, induce apoptosis, and impair insulin
secretion in isolated human islets. Diabetes 56:65-71.
154
14. Hayden MR, Karuparthi RR, Manrique CM, Lastra G, Habibi J, & Sowers JR (2007)
Longitudinal ultrastructure study of islet amyloid in the HIP rat model of type 2 diabetes
mellitus. Exper. Bio. Med, 232:772-779.
15. Abedini A & Raleigh DP (2005) Incorporation of Pseudoproline Derivatives Allows the
Facile Synthesis of Human IAPP, A Highly Amyloidogenic and Aggregation-Prone
Polypeptide. Org. Lett. 7:693-696.
16. Abedini A, Singh D, & Raleigh DP (2006) Recovery and Purification of Highly
Aggregation-Prone Disulfide-Containing Peptides: Application to Islet Amyloid
Polypeptide. Anal. Biochem. 351:181-186.
17. Marecek J, Song B, Brewer S, Belyea J, Dyer RB, & Raleigh DP (2007) A Simple and
Economical Method for the Production of 13C 18O Labeled Fmoc-Aminoacids with
High Levels of Enrichment: Applications to Isotope Edited IR Studies of Proteins. Org.
Lett. 9:4935-4937.
18. Shim S-H, Strasfeld DB, Fulmer EC, & Zanni MT (2006) Femtosecond pulse shaping
directly in the mid-IR using acousto-optic modulation. Opt. Lett. 31(6):838-840.
19. Shim S-H, Strasfeld DB, Ling YL, & Zanni MT (2007) Automated 2D IR spectroscopy
using a mid-IR pulse shaper and application of this technology to the human islet amyloid
polypeptide. Proc. atl. Acad. Sci. U.S.A. 104(36):14197-14202.
20. Westermark P, Engstrom U, Johnson KH, Westermark GT, & Betsholtz C (1990) Islet
amyloid polypeptide: Pinpointing amino acid residues linked to amyloid fibril formation.
Proc. atl. Acad. Sci. U.S.A. 87:5036-5040.
21. Luca S, Yau WM, Leapman R, & Tycko R (2007) Peptide conformation and
supramolecular organization in amylin fibrils: Constraints from solid-state NMR.
Biochemistry 46:13505-13522.
22. Jaikaran E, Higham CE, Serpell LC, Zurdo J, Gross M, Clark A, & Fraser PE (2001)
Identification of a novel human islet amyloid polypeptide beta-sheet domain and factors
influencing fibrillogenesis. J. Mol. Biol. 308(3):515-525.
23. Chimon S, Shaibat MA, Jones CR, Calero DC, Aizezi B, & Ishii Y (2007) Evidence of
fibril-like beta-sheet structures in a neurotoxic amyloid intermediate of Alzheimer's beta-
amyloid. at. Struct. Mol. Biol. 14(12):1157-1164.
24. Torres J, Briggs JAG, & Arkin IT (2002) Multiple site-specific infrared dichroism of
CD3-zeta, a transmembrane helix bundle. J. Mol. Biol. 316(2):365-374.
26. Kim YS, Liu L, Axelsen PH, & Hochstrasser RM (2008) Two-dimensional infrared
spectra of isotopically diluted amyloid fibrils from Ab40. Proc. atl. Acad. Sci. U.S.A.
105(22):7720-7725.
27. Petkova AT, Leapman RD, Guo ZH, Yau WM, Mattson MP, & Tycko R (2005) Self-
propagating, molecular-level polymorphism in Alzheimer's beta-amyloid fibrils. Science
307(5707):262-265.
28. Padrick SB & Miranker AD (2002) Islet amyloid: Phase partitioning and secondary
nucleation are central to the mechanism of fibrillogenesis. Biochemistry 41(14):4694-
4703.
29. Konarkowska B, Aitken JF, Kistler J, Zhang S, & Cooper GJS (2006) The aggregation
potential of human amylin determines its cytotoxicity towards islet b-cells. in FEBS
Journal), pp 3614-3624.
30. Munoz V, Henry ER, Hofrichter J, & Eaton WA (1998) A statistical mechanical model
for beta-hairpin kinetics. Proc. atl. Acad. Sci. U.S.A. 95(11):5872-5879.
31. Munoz V, Ghirlando R, Blanco FJ, Jas GS, Hofrichter J, & Eaton WA (2006) Folding
and aggregation kinetics of a beta-hairpin. Biochemistry 45(23):7023-7035.
32. Hahn S, Kim SS, Lee C, & Cho M (2005) Characteristic two-dimensional IR
spectroscopic features of antiparallel and parallel beta-sheet polypeptides: Simulation
studies. J. Chem. Phys. 123(8).
33. Yan LM, Tatarek-Nossol M, Velkova A, Kazantzis A, & Kapurniotu (2006) A. Design of
a mimic of nonamyloidogenic and bioactive human islet amyloid polypeptide (IAPP) as a
nanomolar affinity inhibitor of IAPP cytotoxic fibrillogenesis. Proc. atl. Acad. Sci.
U.S.A. 103:2046-2051.
34. Abedini A, Meng F, & Raleigh DP (2007) A single-point mutation converts the highly
amyloidogenic human islet amyloid polypeptide into a potent fibrillization inhibitor. J.
Am. Chem. Soc. 129:11300-11301.
35. Abedini A & Raleigh DP (2006) Destabilization of human IAPP amyloid fibrils by
proline mutations outside of the putative amyloidogenic domain: is there a critical
amyloidogenic domain in human IAPP. J. Mol. Biol. 355:274-281.
36. Gilead S, Wolfenson H, & Gazit E (2006) Molecular mapping of the recognition
interface between the islet amyloid polypeptide and insulin. Angew. Chem., Int. Ed.
45:6476-6480.
37. Engel MFM, Khemtemourian L, Kleijer CC, Meeldijk HJD, Jacobs J, Verkleij AJ, de
Kruijff B, Killian JA, & Hoppener JWM (2008) Membrane damage by human islet
amyloid polypeptide through fibril growth at the membrane. Proc. atl. Acad. Sci. U.S.A.
105(16):6033-6038.
156
38. Dawson PE & Kent SBH (2000) Synthesis of native proteins by chemical ligation.
Annu. Rev. Biochem. 69:923-960.
157
Appendix I
List of Publication
11. Ling YL, Strasfeld DB, Shim S-H, Raleigh DP, and Zanni MT, “2D IR provides evidence of
an on-pathway intermediate in the membrane-catalyzed assembly of diabetic amyloid”,
Submitted.
10. Shim S-H and Zanni MT, “How to turn your pump-probe instrument into a multidimensional
spectrometer: 2D IR and Vis spectroscopies via pulse shaping.” Invited article for Phys.
Chem. Chem. Phys. Accepted.
9. Shim S-H, Gupta R, Ling YL, Strasfeld DB, Raleigh DP and Zanni MT, “2D IR
spectroscopy defines the pathway of amyloid formation with residue specific resolution.”
Submitted.
8. Xiong W, Strasfeld DB, Shim S-H, and Zanni MT (2008) “Automated 2D IR spectrometer
mitigates the influence of high optical densities.” Vibrational Spectroscopy, Accepted.
7. Strasfeld DB, Ling YL, Shim S-H, and Zanni MT (2008) “Tracking fibril formation in
human Islet amyloid polypeptide with automated 2D-IR spectroscopy.” J. Am. Chem. Soc.
130:6698-6699.
6. Grumstrup EM,* Shim S-H,* Montgomery MA,* Damrauer NH, & Zanni MT (2007)
“Facile collection of two-dimensional electronic spectra using femtosecond pulse-shaping
technology.” Optics Express 15(25):16681-16689. *Equal authors.
5. Shim S-H, Strasfeld DB, Ling YL, & Zanni MT (2007) “Automated 2D IR spectroscopy
using a mid-IR pulse shaper and application of this technology to the human islet amyloid
polypeptide.” Proc. atl. Acad. Sci. U.S.A. 104:14197-14202.
4. Strasfeld DB, Shim S-H and Zanni MT, “New advances in mid-IR pulse shaping and its
applications to 2D IR spectroscopy and ground state coherent control.” Invited article for
Advances in Chemical Physics. In press.
3. Strasfeld DB, Shim S-H, & Zanni MT (2007) “Controlling vibrational excitation with shaped
mid-IR pulses.” Phys. Rev. Lett. 99:038102.
2. Shim S-H, Strasfeld DB, & Zanni MT (2006) “Generation and characterization of phase and
amplitude shaped femtosecond mid-IR pulses.” Opt. Express 14:13120-13130.
1. Shim S-H,* Strasfeld DB,* Fulmer EC, & Zanni MT (2006) “Femtosecond pulse shaping
directly in the mid-IR using acousto-optic modulation.” Opt. Lett. 31(6):838-840. *Equal
authors.