Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

2. Basic Principles

2.1 Theoretical aspects

This section presents some basic theoretical aspects of NMR which are relevant for a technical
discussion of the principles and experimental procedures used when studying biological macro-
molecules in solution by NMR. A rigorous discussion of the theoretical foundation of NMR can
be found in many textbooks [e.g. 16, 19, 24, 26, 29]. Only a very limited theoretical foundation is
necessary for the technically oriented discussion of NMR methods in the following sections. The
concept of energy levels, the Bloch equations and the product operator formalism are sufficient in
most cases. NMR is intimately related with frequencies and to obtain a clear distinction between
angular frequencies with units rad/s and technical frequencies in Hertz (Hz) the symbols ω or Ω
are used for the former and ν for the latter.

The basis of all NMR experiments is the nuclear spin which can be interpreted as a magnetic
moment. A spin 1--2- nucleus in this view forms a small dipole. This dipole orients either parallel (α
state) or antiparallel (β state) to a magnetic field leading to a small energy difference ∆E between
the two states
hγ
∆E = ------ Bo = h γBo (2.1)
2π
where Bo is a large externally applied homogeneous magnetic field, h is Planck’s constant
( h = h ⁄ ( 2π ) ), and γ the gyromagnetic ratio which is a property of the nucleus and can have a

Table 1
Properties of selected nuclei
natural relative
Nucleus Spin γ † [107rad/(Ts)]
abundance [%] sensitivity§
1H 1---
2 26.75196 99.985 1.00
2H 1 4.106625 0.015 9.65 10–3
3H 1---
2 28.53495 - 1.21
13C 1---
2 6.72828 1.108 1.59.10–2
14N 1 1.93378 99.634 1.01.10–3
15N 1---
2 –2.71262 0.366 1.04.10–3
17O (-----
5
2 –3.6281 0.037 2.92.10–2
19F 1---
2 25.18147 100.00 0.83
31P 1---
2 10.8394 100.00 6.64.10–2
† γ: gyromagnetic ratio; γ =γ/(2π)
§ For an equal number of nuclei relative to protons

-7-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

positive or a negative value. Table 1 lists the gyromagnetic ratio and some other properties of
nuclei important in NMR of biological macromolecules. From the two states of a dipole the α
state is energetically slightly more favourable and thus possesses a higher population than the β
state. Transitions between adjacent energy levels can be induced by small additional magnetic
fields perpendicular to Bo which oscillate with a frequency νo fulfilling the resonance condition
νo = ∆E/h . The frequency νo typically lies in the radio-frequency range and is often referred to as
Larmor frequency. In the equilibrium state the Boltzmann distribution favours the lower energy
states. Thus, the sum of all contributing nuclear magnetic moments of the individual nuclei leads
to a resulting macroscopic magnetization M along the homogeneous external field Bo. In the
framework of classical physics the behaviour of this magnetization under the action of time
dependent magnetic fields can be described by the Bloch equations [30]. Because the spin is a
quantum mechanical phenomenon this description has a very limited scope, but it proves very
useful for the description of single resonance lines under the action of radio-frequency (rf) pulses
and thus for the characterization of the effect of rf pulses.

2.1.1 Magnetization, precession and Bloch equations

In a classical description the macroscopic magnetization M created by the spins is described by a

vector M parallel to the magnetic field vector Bo. M is forced to move away from the direction of
Bo by an additional linearly polarized magnetic field B1 perpendicular to Bo. B1 must fulfil the
resonance condition and oscillate with the resonance frequency νo. The magnetization vector M
precesses about the resulting magnetic field B = Bo+ B with an angular velocity vector ω point-
ing in the opposite direction and the components as described in Eq. (2.2)

ω = –γB = –γ(2Bx cos(2πνot + φ), 2By sin(2πνot + φ), Bo) (2.2)

where ω = (ωx, ωy, ωo), Bo is chosen along the z axis and φ describes the angle between the x axis
and B1. The magnetic field B1 is often applied only for short time periods as radio-frequency (rf)
pulses. The discussion of the motion of the magnetization vector M in space due to rf pulses is
usually based on a rotating frame of reference which has the same z axis along the static magnetic
field Bo as the laboratory frame but rotates around the z axis with a frequency which is often cho-
sen equal to the resonance frequency νo. In this rotating frame of reference the relevant compo-
nent of the applied oscillating field B1 appears static making the discussion and visualization
much easier. To fulfil the physical requirement that the magnetization vector M returns to its
equilibrium position in a finite period of time after a disturbance, a longitudinal relaxation time
T1 (spin-lattice relaxation) is introduced. The loss of coherent precession is described by a trans-
verse relaxation time T2 (spin-spin relaxation). The motion of the magnetization vector M under
the action of the magnetic field B and hence under ω can be described by the Bloch equations
[30]. These equations are presented in the Appendix for further reference.

In the rotating frame where B1 becomes static the main magnetic field Bo vanishes for nuclei
with resonance frequency νo. Hence, Eq. (2.2) in this rotating frame contains only a transverse
component ω = –γ(B1, 0, 0) with B1 chosen along the x axis. Consequently M precesses by an
angle β around the magnetic field B1 applied as a pulse for the short duration τ

β = –γB1τ (2.3)

where β is called the flip angle of the pulse. The flip angle is often indicated in degrees, for exam-

-8-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

ple during a 90o pulse M can precess from the z axis to the x axis. The angle β depends on γ and
is negative for positive γ values such as for protons (Table 1). For positive γ values a magnetic
field B1 pointing along the positive x axis turns M towards the negative y axis. A B1 field along
the +y axis turns M towards the +x axis [16, 31]. When applying a rf pulse with a frequency dif-
fering from νo the action of the rf pulse becomes more complex as described in the Appendix. A
situation often referred to as non-ideal behaviour of the rf pulses or as off-resonance effects.

The oscillating magnetic field 2B1 cos(ωrft) used for excitation is linearly polarized in the labora-
tory frame. The transformation into the rotating frame can best be followed when this is thought
of as a superposition of two counter-rotating, circular polarized fields with an amplitude B1.
When transforming into the rotating frame one component matches the Larmor frequency
whereas the other oscillates at twice the Larmor frequency and does not fulfil the resonance con-
dition. Bloch and Siegert [32] calculated the effect of the non-resonant field and found that it
slightly shifts the frequency of the observed resonance lines away from the disturbing field by the
small amount νΒ= ( γ B1)2/∆ν where γ = γ/(2π) and ∆ν stands for twice the resonance frequency.
The Bloch-Siegert shift is small and amounts for example to 0.5 Hz for a frequency of 600 MHz
during a rf pulse with duration τ of 10 µs and a flip angle of π--2- (Eq. (2.3) or 90o. The shift disap-
pears as soon as B1 is switch off. An effect similar to the Bloch-Siegert shift occurs whenever a rf
field is applied with a frequency ∆ν off-resonance for the nuclear spins. First described by Ram-
sey [33] it is still very often referred to as Bloch-Siegert effect. To better distinguish it from the
effect due to the counter rotating field the term "non-resonant effect" was introduced [34]. Since
this additional field may be rather strong and close to the resonance frequency this effect can
become quite large and should be compensated by adequate means [35] (Section 2.2.1).

2.1.2 Operators, coherence, and product operator formalism

The description of NMR experiments by the Bloch equations and by magnetization vectors in the
rotating frame has rather significant limitations particularly for the description of multipulse
experiments. On the other hand, a full quantum mechanical treatment which describes the state of
the system by calculation of the time evolution of the density operator under the action of the
appropriate Hamiltonian can be rather cumbersome. In a quantum mechanical description a rf
pulse applied to the equilibrium state creates a coherent superposition of eigenstates which differ
in their magnetic quantum number by one, often simply referred to as a coherence. In more com-
plex experiments the magnetic quantum numbers between states may differ by a value q different
from one, leading to a q quantum coherence with at least q spins involved. However only in-
phase single quantum coherences (Table 2) are observable and correspond to the classical mag-
netization detected during the acquisition of an NMR experiment. Multiple quantum coherences
cannot directly be observed but they influence the spin state and this information can be trans-
ferred to observable magnetization.

In a step towards a full quantum mechanical treatment the product operator formalism for spin 1--2-
nuclei was introduced [28]. In this approach it is assumed that there is no relaxation and that the
difference in the resonance frequency ∆ν of two nuclei is much larger than their mutual scalar
coupling J; this situation is often referred to as weak spin-spin coupling whereas strong spin-spin
coupling specifies the case where ∆ν is close to or even smaller than J. With these assumptions
simple rules can be calculated which describe the evolution of spin operators under the action of
chemical shift, J coupling and rf pulses. The formalism combines the exact quantum mechanical
treatment with an illustrative classical interpretation and is the basis for the development of many

-9-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

NMR experiments. However, some parts of experimental schemes, for example TOCSY
sequences (Section 4.2.1), can only be described with a full quantum mechanical treatment. Cal-
culations with the formalism are not difficult, but many terms may have to be treated and imple-
mentations of the formalism within computer programs are very helpful in such situations [36,
37]. Although most of the experiments applied in biomolecular NMR correlate three or more
spins, the majority of interactions can still be understood based on an analysis of two spins. For
two spins I and S the operator basis for the formalism contains 16 elements. Two sets of basis
operator, cartesian and shift operators, have proven very useful for the description of experimen-
tal schemes and are used in parallel. The two basis sets and the nomenclature used to characterize
individual states are summarized in Table 2 [28].

Table 2
Product operator basis for a two spin system
cartesian operator basis nomenclature shift operator basis
Iz , Sz longitudinal magnetization Iz , Sz
Ix , Iy , Sx , Sy in-phase transverse magnetization I , I–, S+, S–
+

2IxSz , 2IySz anti-phase I spin magnetization 2I+Sz , 2I–Sz

2SxIz , 2SyIz anti-phase S spin magnetization 2S+Iz , 2S–Iz
2IxSx , 2IySx , 2IxSy , 2IySy two spin coherence 2I+S+, 2I+S–, 2I–S+, 2I–S–
2IzSz longitudinal two spin order 2IzSz
E unity operator E

The operators Iz and Sz are identical in the two basis sets and a simple relationship exists between
the two other cartesian and shift operators:

Ix = (I+ + I–) /2 I+ = Ix + iIy (2.4)

Iy = –i(I+ – I–) /2 I– = Ix – iIy

Three operators, which represent the action of the Hamiltonians for chemical shift, scalar cou-
pling and rf pulse, act on these basis operators and may transform them into other operators
within the basis set. In this way the spin states created during a NMR experiment can be
described and the observable magnetization calculated. The operator formalism can be summa-
rized by simple rules [28] which are listed for both basis systems in the Appendix for further ref-
erence. Operators transform individually under these rules even in products of operators except
for anti-phase terms which have to be considered as a unit and transformed accordingly, however,
they can be treated consecutively when different couplings to the same nucleus exist.

The cartesian operators transform more easily under pulses and their single operators have a
direct classical interpretation as magnetization vectors. The shift basis provides a useful alterna-
tive for the description of the evolution due to chemical shift and/or the influence of magnetic
field gradients (Section 2.3) and is better suited for the description of coherence orders and coher-
ence pathways (Fig. 1). Their single operators describe a transition from the α to the β state, I+,

-10-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

or from the β to the α state, I–. Hence shift product operators are uniquely associated with one
coherence order, for example I+S+ describes only double quantum coherence (DQC), whereas
the cartesian product operator may be associated with several coherence orders, for example
2IxSx describes a linear combination of both DQC and zero quantum coherence (ZQC). The car-
tesian x and y components of a multiple quantum coherence are given by linear combinations of
the shift or cartesian operators. For example, the ZQCs and DQCs of a two spin system can be
described as follows

(ZQC) x = (2IxSx + 2IySy) = (I+S– + I–S+)

(DQC) x = (2IxSx – 2IySy) = (I+S+ + I–S–) (2.5)
(ZQC) y = (2IxSy – 2IySx) = i(I+S– – I–S+)
(DQC) y = (2IxSy + 2IySx) = –i(I+S+ – I–S–)

On the basis of the operator formalism the selection of particular states using phase cycling of rf
pulses in a NMR experiment can be rationalized. Coherences present in a experiment can be clas-
sified into their different orders or coherence levels which can be represented in a pictorial way
(Fig. 1) to visualize the coherence transfer pathways in a experimental scheme [38, 39]. The

p rf1 rf2 rf3

+2
+1
0
–1
–2

Fig. 1. Coherence level diagram. The coherence levels p are formally represented by products of the shift
operators I+ and I– which are conserved during periods of free evolution. The application of radio-fre-
quency pulses may transfer coherences from one order (level) to another. The positions of three pulses are
indicated in the figure by vertical arrows labelled rf1, rf2 and rf3. Thick lines represent the coherence path-
way starting at the equilibrium state (p = 0) passing through single quantum coherences after the first pulse
(|p| = 1), reaching double quantum coherences after the second pulse (|p| = 2) and ending as observable
magnetization (p = –1) after the third pulse. Only single quantum coherences (|p| = 1) can be observed. The
spectrometer detects only one of these two coherence levels which is usually assumed to be p = –1 [38] and
hence all other coherence orders after the third pulse cannot be detected and therefore are not drawn. Thin
lines indicate alternative pathways which have to be suppressed if only the pathway indicated by thick
lines should contribute to the signal measured at the end of the sequence.

order of coherence p corresponds to the change q in the magnetic quantum number between the
two connected states [28]. Hence for n coupled spins the maximal coherence level that can be
reached is n. Free precession conserves the coherence order whereas pulses may cause coher-
ences to be transferred from one order to another (Fig. 1). The sensitivity of a coherence to the
phase of an rf pulse is proportional to its order, a q quantum coherence will experience a phase
shift ∆φ of an rf pulse as q∆φ. If the pulse results in a change in the coherence order of ∆p, the
corresponding phase shift experienced by the affected coherence will be –∆φ∆p. Proper phase
cycling of consecutive rf pulses allows for selection of a specific succession of coherence levels
that define a coherence pathway (Fig. 1). The concept of coherence transfer pathways clarifies the

-11-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

role of phase cycling in NMR experiments and describes their action with a simple set of rules
[21, 24, 38, 40]. A particular rf pulse can be designed to select a certain difference in coherence
order ∆p ± nN (n = 0, 1, 2, ..,) with a phase cycle comprising N phase steps ∆φ of the same size
equal to 360o/N. The N signals obtained must be summed together with the proper receiver phase
–k∆φ∆p to compensate for the phase change experienced by the coherence, k takes on values {0,
1, 2, 3, ...}. An example for the design of a phase cycle using this recipe is given in Section 4.2.1
with the discussion of the double quantum filter. A detailed discussion of phase cycling can be
found in most textbooks on NMR [e.g. 16, 21, 24, 26].

2.1.3 Descriptive representations of experimental schemes

An NMR experiment can be graphically described to a limited extent based on a classical physi-
cal model using populations and magnetization vectors in the rotating frame or based on quantum
mechanical principles using the product operator formalism. Both descriptions find widespread
applications for the discussion and development of NMR experiments. The different representa-
tions are discussed on the basis of the scheme shown in Fig. 2. The application of this experiment

y x

I
1/(2J) y
S
t
ab c de

Fig. 2. Sketch of the experimental scheme used for the discussion of different representations shown in Fig
3. The black narrow bars indicate 90o rf pulses, five time points on the time axis t are denoted by the letters
a, b, c, d and e. The rf pulses applied on-resonance to the two species of nuclei I and S are indicated on the
lines marked with the corresponding letters, a particular pulse acts only on one nuclear species. The scalar
coupling between the two spins is J. The two pulses applied on spin I are separated by the time period
(2J)–1. The phases of the rf pulses are indicated with x or y at the top of the pulses, where x or y stand for
the application of the B1 field in the rotating frame along the positive x or y axis, respectively.

to a system of two scalar coupled spins I and S is described with four different representations in
Fig. 3 for each of the five time points a to e. Fig. 3 represents energy level diagrams (E), the
observable spectra (S), magnetization vector diagrams (V) and the notation in the product opera-
tor formalism (O) showing the cartesian and the shift operator basis. In Fig. 3 spins I and S are
assumed to be proton and carbon nuclei, respectively, with the size of representative vectors pro-
portional to the corresponding populations. But qualitatively the figure applies to all nuclei with
spin 1--2- and positive gyromagnetic ratio.

The experiment in Fig. 2 starts at time point a in thermal equilibrium (Fig. 3) where the popula-
tions on the upper Pu and the lower energy level Pl across a transition fulfil the Boltzmann distri-
bution (Eq. (2.6)). Because ∆E in Eq. (2.1) is typically much smaller than kT we can approximate
this exponential distribution by the first term in a Taylor expansion

Pu/Pl = exp(– ∆E ------- = 1 – γ-----------

------- ) ≅ 1 – ∆E hBo- (2.6)
kT kT kT

-12-
 a b c d e

Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......
ββ
MS– 1
1-2.5δ
1
1-0.5δ
1
1-0.5δ –3MS– 1
1+1.5δ 1
1+0δ
2 2 2 2 2
1-1.5δ 1+0.5δ 1+0.5δ 1-1.5δ 1+0δ
E: βα
x
MI– x y –MI–

MI+ x y MI+
3
1+1.5δ
3
1-0.5δ
3
1-0.5δ 3
1-2.5δ x 3
1+0δ
4 4
MS αβ
4 4 4
1+2.5δ + 1+0.5δ 1+0.5δ 1+2.5δ 5MS+ 1+0δ
αα

13 12
S: 0 ωI 13 24 0 ωI 24 0 ωI 0 ω 34 0 ωS
Fig. 3
-13-

z z z z z
5MS+
MI– MI+ MI+
3MS–
MI +
V: MI – MI–
y y y y y
x x MI+ x x – x
3MS 5MS+
MI–

Iz + Sz Ix + Sz 2IySz + Sz 2IzSz + Sz 2SxIz + Sx

O: Iz + S z (I– + I+)/2 + Sz –i(I– + I+)Sz + Sz 2IzSz + Sz S–Iz + S+Iz + S+/2 + S–/2
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

Fig. 3. Different representations used for the description of NMR experiments illustrated for the experi-
mental scheme shown in Fig. 2 using a two spin system consisting of a proton (I) and a carbon (S) nucleus
with a scalar coupling J>0. The representations are qualitatively valid for all spins with a positive gyro-
magnetic value (Eqs. (2.2) and (2.6)). The labels a, b, c, d and e refer to the different time points indicated
in Fig. 2. Four different representations are indicated. E: energy level diagrams, S: sketch of the observable
spectrum, V: vector diagram and O: product operator formalism in the cartesian and in the shift operator
basis. In E the spin states α and β associated with each energy level are indicated with the first character
representing the proton spin and the second the carbon spin. The polarizations are indicated by M with the
corresponding nucleus indicated as a subscript, the superscript plus or minus signs refer to the sign of the
corresponding frequency of the components after excitation in the rotating frame (compare V). The popu-
lations of individual energy levels are normalized to one and deviations are given in units of δ with
δ = γc h Bo/(kT) (Eq. (2.6)) where γc is the gyromagnetic ratio for carbons. The ratio of 4 between the γ
values of protons and carbons was used to represent populations. The coherences evolving after excitation
have a "direction" (arrows) [16] and are labelled x and y for x and y phase, respectively. In the schematic
stick spectra S the transitions are labelled according to the numbers of the connected energy levels. The
spectra in S and the rotating frame in V are at the proton frequency at the time points a, b and c, at the pro-
ton and carbon frequency at time point d and at the carbon frequency at time point e. The vectors M in V
are labelled with the same conventions as the polarizations in E. In the product operator representation the
natural S spin magnetization is indicated by the operator S.

where k is Boltzmann’s constant and T the absolute temperature. With Eq. (2.1) the energies E1,
E2, E3, and E4 of the four different energy levels in the system can be calculated

E1 = h(–νI – νS + J/2)/2
E2 = h(–νI + νS– J/2)/2 (2.7)
E3 = h( νI – νS – J/2)/2
E4 = h( νI + νS + J/2)/2

where νI and νS stand for the resonance frequencies of the I and S nuclei, respectively. The reso-
nance frequencies of nuclei with positive gyromagnetic ratio γ such as protons and carbons are
negative (Eq. (2.2)) and, hence, E1 becomes the highest and E4 the lowest energy (Fig. 3). For
nuclei with a positive γ value the α state (spin 1--2-) has lower energy than the β state (spin – 1--2-). The
polarizations MI+ and MI– are proportional to the energy differences (E4–E2) and (E3–E1),
respectively, and they determine the intensity of the corresponding transitions 2<-->4 (24) and
1<-->3 (13). The consistent use of signs and transformation properties as presented in Fig. 3 may
seem not to be of great importance and, indeed, has very often no direct experimental conse-
quences. But there are situations where inconsistencies occur and the interpretation of data
becomes confusing or wrong [31, 41].

Apart from the consistent illustration of different representations for the description of a NMR
experiment, Fig. 3 demonstrates that the scheme shown in Fig. 2 transfers polarization from pro-
ton to carbon spins. At time point d the proton polarization MI– is inverted. As a consequence the
populations across the carbon transitions 12 and 34 acquire a larger difference than at thermal
equilibrium at time point a and, hence, the experimental scheme allows measurement of carbon
spectra with higher sensitivity. Such polarization transfer experiments are extremely important in
heteronuclear NMR experiments and are further discussed in Section 4.2.3.

-14-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

2.1.4 Relaxation

Relaxation processes re-establish an equilibrium distribution of spin properties after a perturba-

tion. After a disturbance, the non-equilibrium state decays in the simplest case exponentially
characterized by the spin-lattice relaxation time T1. Re-establishing thermal equilibrium requires
changes in the population distribution of the spin states and lowers the energy of the spin system.
Thus, it involves energy transfer from the spin system to its surroundings which is usually
referred to as the lattice. Microscopically, relaxation is caused by fluctuating magnetic fields.
Dynamical processes such as atomic or molecular motion produce fluctuating dipolar interac-
tions and facilitate spin-lattice relaxation. The efficiency of the relaxation process depends on the
extent of the overlap between the frequency spectrum of the motional process and the relevant
resonance frequencies. This overlap is described by the spectral density function J(ω). Since J(ω)
is the Fourier transform of the time correlation function describing the motion, its functional
form depends on the mechanism of motion. An exponential correlation function with correlation
time τc results in the spectral density function [16, 24, 29]
τc
J(ω) = 2--- ⋅ -----------------
- (2.8)
5 1+ ω2 τc2
Fig. 4 shows a plot of J(ω) as a function of the frequency ω for the three correlation times τc of
5 ns, 10 ns and 20 ns. These correlation times represent the motion of small, medium and large
globular proteins in solution. In addition Fig. 4 illustrates that low frequency motions are espe-

J(ω)
[10-9s]
τc = 20ns
8

τc = 10ns
4
τc = 5ns

0
0.001 0.01 0.1 0.2 0.5 1 2 5 ω [109 rad/s]
0.16 1.6 16 32 80 160 320 800 ν [MHz]

Fig. 4. Plot of the spectral density function J(ω) (Eq. (2.8) versus the frequency ω on a logarithmic scale.
Three correlation times 5 ns, 10 ns and 20 ns are indicated which represent small, medium and large pro-
teins. The frequency scale is given in units of rad/s and in MHz.

cially effective in NMR relaxation processes for proteins. Using the concept of spectral density
functions the different behaviour of longitudinal relaxation and relaxation of transverse magneti-
zation can be rationalized. When considering only relaxation due to fluctuating dipolar interac-
tions caused by stochastic motion, the relaxation rate T1–1 is proportional to J(ωo) since only

-15-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

stochastic magnetic fields in the transverse plane at the resonance frequency ωo are able to inter-
act with the transverse magnetization components bringing them back to the z axis. For frequen-
cies larger than 25 MHz the values J(ωo) decrease for the three increasing values of τc
represented in Fig. 4. The longitudinal relaxation times, therefore, increase for increasing molec-
ular weight of the protein. For very small molecules with very small τc the values J(ωo) get
smaller again leading to an increase of T1 compared to the value for a τc of 5 ns (Eq. (2.8). The
minimum T1 value is obtained when ωoτc = 1, e.g. at 600 MHz for a τc of 0.26 ns. Transverse
relaxation shows a different dependence on the molecular weight of the molecule. T2 relaxation
not only depends on J(ωo) but also on J(0) since the z components of stochastic magnetic fields
(zero frequency) reduce the phase coherence of transverse magnetization components which con-
sequently sum up to a smaller macroscopic magnetization. Since J(0) monotonously increases
with increasing correlation times (Eq. (2.8), Fig. 4) T2 decreases monotonously with increasing
molecular weight. Short T2 values reduce the performance of NMR experiments with large mole-
cules. However, relaxation depends not only on the size of a molecule but also on its internal
motions. Two molecules with the same molecular weight may show quite different relaxation
behaviour depending on their particular internal motions.

2.1.5 Through-bond correlations

The magnetic dipole-dipole interaction describes the effect of the local magnetic fields associated
with the magnetic moments of surrounding nuclei. Two mechanism contribute to this effect: the
"direct" (through-space) coupling and the "indirect" spin-spin coupling or J coupling transmitted
via polarization of bonding electrons. The complete analysis of protein spectra is based on inter-
actions between different spins, either mediated by electrons in through-bond correlations or by
direct interactions through space. Through-bond correlations group individual spins into spin
systems [15] which are characteristic for individual amino acids. In proteins couplings over more
than three chemical bonds are usually not observed. Consequently only spin systems for amino
acid types can be obtained for unlabelled or 15N labelled proteins but the sequential arrangement
of these spin systems relies on through-space correlations [5, 6] which may be ambiguous for
larger proteins. The efficiency of through-bond correlations depends on the size of the coupling
constants involved. Heteronuclear coupling constants often are much larger than proton-proton
couplings (Fig. 5). The use of heteronuclear coupling constants requires the protein to be labelled
with 15N and/or 13C isotopes. With 13C, 15N doubly labelled proteins spin systems of individual
amino acid residues can be connected via J couplings across the peptide bond. Based on hetero-
nuclear couplings a complete assignment can be obtained from through-bond correlations alone.
Fig. 5 summarizes some typical coupling constants found for nuclei in proteins. A wide range of
experiments for the determination of homo- and heteronuclear scalar coupling constants in pro-
teins exist. An excellent survey of these methods can be found in a recent review [42].

Two principally different mechanisms for the through-bond correlation of spins are used. Either
individual spin pairs are correlated or all spins in a spin systems interact simultaneously. The first
case is often referred to as a COSY-type and the second as TOCSY-type correlation. TOCSY
stands for total correlation spectroscopy [43] also known under the acronym HOHAHA for
homonuclear Hartmann-Hahn transfer [44]. Both types of correlation can transfer magnetization
between two nuclei. Thus the sensitivity of a nucleus can be enhanced when the experiment starts
with the polarization of a nucleus of a different species with a higher gyromagnetic ratio (Fig. 2
and Fig. 3). Polarization transfer based on a COSY-type sequence was given the acronym INEPT
[45] which stands for insensitive nuclei enhanced by polarization transfer. Polarization transfer

-16-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

H H O H H O H H O H H O

140
1-5 0-5 93 0-4 1-7
2-11 0-2
14
N C C N C 53
C N 7-11 C C N C C
0-4

37
2 6-8 0-4
132 0.5-2.5
H C H C

40
3
56
C
160
H C C H

58
7

H C C H
C
H

Fig. 5. Typical absolute values for coupling constants and their range in Hertz. If the variation is less than
10% of the maximal value single average values are given otherwise the range is indicated by the maximal
and the minimal value. One bond coupling constants are written along the bond, for multiple bond cou-
pling constants a line drawn along the chemical bonds connects the two coupled nuclei.

based on a TOCSY-type sequence usually is referred to as HEHAHA (heteronuclear Hartmann-

Hahn) experiment (Section 4.2.3).

The COSY-type correlation can easily be rationalized using the product operator formalism for
two scalar coupled spins I and S (Eq. (A.2.2) in the Appendix). Transverse Iy magnetization will
evolve into anti-phase magnetization of the form 2IxSz due to scalar coupling. A 90o rf pulse with
phase y on both spins transforms this operator product into 2IzSx which can evolve into the
observable operator Sy. The crucial element in a COSY-type transfer is the 90o rf pulse acting on
an anti-phase state. If the two nuclei belong to two different nuclear species the polarization
transfer from spin I to spin S (Fig. 3) will change the sensitivity of the spin S spin by the ratio
γ I ⁄ γ S of the gyromagnetic ratios.

In a homonuclear TOCSY-type transfer a strong rf field is applied to one nuclear species. Viewed
in the rotating frame this field locks the spins along the axis it is applied. In this spin-locked state
individual precession around Bo is suppressed and replaced by a collective precession with the
frequency of the applied rf field. Without their characteristic precession frequencies the spins lose
their individuality and can no longer be distinguished and behave as part of a strongly coupled
spin system. The product operator formalism cannot describe such a state and an analysis is only
possible using a quantum mechanical treatment. A homonuclear two spin system with scalar cou-
pling J evolves under spin-locking for a period τm from the state Ix as follows [43]:

Ix ----> Ix [1 + cos(2πJτm)]/2 + Sx [1 – cos(2πJτm)]/2 + [IySz – IzSy] sin(2πJτm) (2.9)

For τm = (2JIS)–1 complete in-phase magnetization transfer from Ix to Sx will occur. For more
than two coupled spins different coupling constants will govern the transfer and complete trans-
fer from one spin to another is usually not possible. The theoretical evaluation leading to Eq.
(2.9) does not consider offset effects of rf pulses. When the effective fields for two nuclei are not

-17-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

aligned (Eqs. (A.1.3) and (A.1.4) in the Appendix) the effective J coupling during the mixing
sequence is reduced resulting in a slower transfer. In addition to magnetization transfer through
bonds TOCSY mixing sequences transfer magnetization through space as discussed in the next
section. This pathway requires special attention only for very sensitive nuclei such as protons and
can safely be neglected for all other nuclei. Detailed descriptions of the foundations of the
TOCSY experiment can be found in literature with both experimental [46] and theoretical treat-
ments [43, 47].

In the more general case of the heteronuclear TOCSY, magnetization is transferred between
nuclei of different species. In this situation two rf fields B1I and B1S at two different nuclear reso-
nance frequencies have to be applied, with B1I = γIωI and B1S = γSωS. As in the HOHAHA exper-
iments the spins must experience the same magnetic field strength to loose their individuality and
to form a strongly coupled system. Setting B1I and B1S equal one obtains the well known Hart-
mann-Hahn condition [16, 19, 48] which must be fulfilled to obtain a heteronuclear TOCSY
transfer:

γIωI = γSωS (2.10)

If for example I stands for a proton and S for a 15N nucleus the locking rf field applied to 15N has
to be almost ten times larger than the one applied to protons (Table 1). Based on a quantum
mechanical treatment the transformation properties of the operator Ix of a heteronuclear two-spin
system submitted to a HEHAHA sequence can be formulated in analytical form [16, 24]

Ix --->Ix [1 + cos(πJτ)]/2 + Sx [1 – cos(πJτ)]/2 + [IySz – IzSy] sin(πJτ) (2.11)

where J stands for the heteronuclear coupling constant between the spins I and S. For a full trans-
fer of the magnetization in a HEHAHA experiment the mixing time τm must be 1/JIS which cor-
responds to double the duration compared to the homonuclear transfer (Eq. (2.9)).

2.1.6 Through-space correlations

A nucleus with a spin different from zero generates a magnetic dipolar field proportional to its
magnetic moment. As the molecule tumbles in solution, this field fluctuates and constitutes a
mechanism of relaxation for nearby spins. Since the dipole-dipole interaction involves a pair of
spins, four states can occur for a system with two spins 1--2-. Due to the double and zero quantum
transitions which are possible in such a system, the longitudinal relaxation process for the two
spins I and S are coupled [49] and the expectation values <Iz> and <Sz> describing the z magnet-
ization fulfil the equation

d/dt(<Iz> – Io) = –ρN (<Iz> – Io) – σN(<Sz> – So) (2.12)

d/dt(<Sz> – So) = –ρN (<Sz> – So) – σN(<Iz> – Io)

where ρN stands for the longitudinal relaxation rate constant of the two spins I and S of the same
nuclear species, σN for the cross relaxation rate constant and Io or So are the equilibrium magnet-
izations. The coupling of the relaxation of the two nuclei will alter the magnetization of one spin
when the other spin is not in its equilibrium state. For example, when the equilibrium value So is
selectively disturbed leading to a deviation ∆S from So then the I magnetization will change from
Io with a initial rate proportional to σN∆S (Eq. (2.12). The following expression can be derived

-18-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

for σN and ρN using two spins of the same species without scalar coupling and with an internu-
clear distance r [50]:
K- (6J(2ω ) – J(0))
σN = --- (2.13)
r6 o

K- (J(0) + 3J(ω ) + 6J(2ω ))

ρN = --- (2.14)
r6 o o
µ 2
h γ 2 4
K =  -----o- ⋅ -----------2 (2.15)
4π 16π
These expressions are valid in the case of isotropic tumbling and dipolar relaxation. σN can take
on positive and negative values depending on the value of the spectral density function in Eq.
(2.8) (Fig. 4) and has a zero crossing (Fig. 6). For globular proteins measured at high magnetic
fields σN is negative. For protons K equals 1.424.105 nm6/s2. For example in a globular protein
with a rotational correlation time of 10 ns, σN becomes approximately –10 s–1 for two protons at
a distance r of 0.2 nm and σN = –0.04 s–1 for r = 0.5 nm. Nuclear Overhauser enhancement
(NOE) experiments make use of the cross relaxation between spins and allow detection of nuclei
which are close in space, in practice at a maximal distance of about 0.5 nm for protons in a glob-
ular protein. In a system with more than two spins consecutive cross relaxation can occur leading
to so called spin diffusion [51] where two distant spins exhibit a larger apparent cross relaxation
due to the contributions of the intervening spins. This effect complicates the derivation of dis-
tances from NOE measurements. Since the initial NOE is proportional to σN spin diffusion
becomes more important for larger proteins which exhibit a longer rotational correlation time τc
(Fig. 6). Practical implementations exploiting cross relaxation use the NOESY segment dis-
cussed in Section 4.2. 2.

Based on the observation of cross relaxation between Iz and Sz states one may expect a similar
effect with transverse magnetization. In general, a net magnetization transfer between transverse
magnetization components does not occur because the spins precess with different frequencies
and the continuously changing phase relationship between the corresponding magnetization vec-
tors prevents the accumulation of a net transfer of magnetization. The situation changes when
different spins are forced to precess at the same frequency by applying a strong rf field. With the
individual precession frequencies removed a net transfer can be established which couples the
transverse relaxation process for two spins I and S. In analogy to the calculation of σN a cross
relaxation rate constant, σR, and a relaxation rate constant, ρR, can be obtained [50, 52]:
K- (2J(0) + 3J(ω ))
σR= --- (2.16)
r6 o

K- (5J(0) + 9J(ω ) + 6J(2ω ))

ρR= ------- (2.17)
2r 6 o o

The NOE between spin-locked transverse magnetization components is usually referred to as

NOE in the rotating frame (ROE). Originally the ROE experiment was called CAMELSPIN [53],
but was later renamed ROESY for the two-dimensional ROE experiment [54]. Fig. 6 illustrates
that the cross relaxation rate σR does not have a zero crossing and that for molecules with a cor-
relation time τc larger than 0.05 ns σR is larger than σN. A detailed theoretical derivation of
homonuclear cross relaxation in the rotating frame can be found in an excellent review [52] and a
number of textbooks [e.g. 16, 24, 26, 50].

Since ROE and TOCSY mixing sequences, both use a spin-lock field to suppress the individual
precession frequencies of transverse magnetization components special care is required in the

-19-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

σ [s–1]
2 40
σR σR
1 20

0 0

–1
σN –20
σN
–2
0.01 0.02 0.05 0.1 0.2 0.5 1 2 5 10 20 τc [ns]
Fig. 6. Plot of cross relaxation rates σ versus the rotational correlation time τc for a two spin system. σ is
given for the NOE in the laboratory frame denoted with σN and in the rotating frame, σR, for two spec-
trometer frequencies: 800 MHz indicated with solid lines and 500 MHz drawn with dotted lines. For these
curves, the scale for σ on the left hand side of the figure applies. The dashed lines show σN and σR on a 20-
fold smaller vertical scale shown on the right hand side. On this scale the curves for the two field strength
are indistinguishable. For the calculation the equations (2.8), (2.13), (2.15) and (2.16) were used assuming
two protons with a constant internuclear distance of 0.2 nm.

implementation to separate the two effects (Section 4.2.1). Although there are experimental
implementations which minimize the simultaneous occurrence of both effects, a strict separation
is not possible and both processes can contribute to correlations between scalar coupled nuclei.
Nevertheless ROE and TOCSY type mixing sequences find widespread applications because the
residual interference between them can in many practical cases be distinguished since the two
effects generate signals with different signs.

Both the NOE and ROE enhancements for short mixing times are proportional to the cross relax-
ation rate which for globular proteins is dominated by the spectral density function at zero fre-
quency, J(0), and therefore from Eqs. (2.13) and (2.16) σR ≅ –2σN. This relation indicates that
the ROE effect builds up signal with increasing mixing time twice as fast and in the opposite
direction from NOE spectra. When results from ROE and NOE spectra are to be directly com-
pared the mixing time for the ROE experiment is often chosen half as long as for the NOE exper-
iment. Both ROE and NOE experiments not only detect cross relaxation but also magnetization
transfer by chemical or conformational exchange. Whereas the sign for the NOE and the ROE
effect may be different this is not the case for exchange contributions which always have opposite
sign compared to the ROE effect. This allows a separation of exchange contributions from NOE
effects.

-20-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

2.2 Radio frequency pulses

2.2.1 Rectangular pulses

A NMR experiment consists of a series of rf pulses and delays, the pulse sequence, followed by
the measurement of the voltage induced by the resulting magnetization in a rf coil (Section
3.5.1). A delay specifies a time period during which no external rf field is applied and the nuclear
spin states evolve due to their intrinsic properties (chemical shift, scalar coupling and relaxation).
A pulse represents a time period during which rf is delivered to the coil in the probe. Four param-
eters describe a rf pulse: frequency, phase, duration and strength. The frequency of the pulse is
often called the carrier frequency since, generally, it is identical to the frequency used to demod-
ulate the detected NMR signal. For a rectangular pulse the strength stays constant during its
application. The magnetic field strength applied during such a pulse is often given in frequency
units γ B1 which can be obtained with Eq. (2.3) by setting the pulse flip angle to 2π for a 360o
pulse with the corresponding pulse length τ360:
γ B1 = γ B = -------
1
--------
2π 1 τ360- (2.18)

For example, a 90o pulse with a duration of 12.5 µs is produced by a field strength γ B1 of
20 kHz.

Ideally a rf pulse applied to a given nuclear species will rotate all magnetization components irre-
spective of their individual resonance frequencies by the same flip angle β about the axis in the
rotating frame defined by the phase of the pulse. However, the performance of a real rf pulse
degrades with increasing offset of the nuclear precession frequencies from the applied rf fre-
quency. The precession axis introduced by a rf pulse applied off-resonance deviates from the
direction of the B1 field of the pulse (see Appendix). Whereas for a 90o excitation pulse satisfac-
tory performance can be obtained over a wide bandwidth, the efficiency of a 180o pulse degrades
rather rapidly. Fig. 7 shows excitation profiles for 180o and 90o pulses in dependence of the offset
frequency. Best use of these excitation profiles can be made when the carrier frequency sits in the
middle of the spectral range of interest. A 90o pulse applied to z magnetization brings most mag-
netization into the transverse plane when the offset frequency νoff fulfils the condition νoff < γ B1.
However, the magnetization acquires an offset dependent angle ε to the direction it would reach
after an ideal 90o pulse with duration τ90. This angle ε can be calculated to be 4τ90νoff in units of
radians [16, 24, 26].

A further consequence of the non-ideal behaviour of rf pulses are specific offset frequencies at
which the pulse does not perturb the resonances. This feature can be used to excite one group of
resonances while selectively avoiding excitation of another group. A technique that finds frequent
applications in heteronuclear experiments involving carbon nuclei where carbonyl carbons and
aliphatic carbons are excited separately. Using the Bloch equations (see Appendix) the k-th null
in the excitation profiles can be calculated for a 90o pulse with duration τ90 to be at a frequency
ν90 and for a 180o pulse with duration τ180 at a frequency ν180 from the carrier frequency:

ν90 = ± k – -----
2
16
1- / τ90 (2.19)

ν180 = ± k – --14- / τ180

2
(2.20)

For a 90o pulse the first null is at ν90 = ±0.97/τ90 and for a 180o pulse at ν180 = ±0.87/τ180.

-21-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

Mi
1
C
D

A
0

B
B1
–1
–1 –0.5 0 0.5 1 [ γ B1]

Fig. 7. Excitation profiles of rf pulses represented by the normalized magnetization Mi plotted against the
offset given in units of γ B 1 where B1 is the applied field strength. (A) Mz after a 90o excitation pulse and
(B) Mz after a 180o inversion pulse applied to z magnetization. (C) Mx after a 180o refocusing pulse with
phase x applied to x magnetization. (D) Same as (C) but My is shown.

Typically several 180o pulses occur in a pulse sequence and the signals created by their non-ideal
behaviour may limit the spectral quality. With a phase cycling scheme, EXORCYCLE [55], mag-
netization components not inverted by the 180o pulse and those which underwent a coherence
transfer due to off resonance effects can be removed from the detected signal. The EXORCYCLE
consists of a 4-step phase cycle where the phase of the 180o pulse changes according to the
scheme x, y, –x, –y together with the receiver phase cycling through x, –x, x –x.

Many applications require the inversion of z magnetization. In this case simple composite pulses
exhibit much broader inversion profiles than a single 180o pulse. A composite pulse based on a
180o pulse with phase y embraced by 90o pulses with phase x, in short notation 900 18090 900,
shows more than 80% inversion over a bandwidth of ± γ B1 [56]. The composite pulse
900 22590 900 where the 180o pulse is replaced by a 225o pulse results in even better inversion of
more than 98% however at the cost of a smaller bandwidth of ±0.7 γ B1. Analogous simple com-
posite pulses for improved refocusing of transverse magnetization do not exist [57].

When pulses or more complex pulse trains with low power are applied selectively to only a small
frequency range in a spectrum the signals at an offset ∆ν far from the irradiation frequency can
still be significantly affected. These off-resonance or non-resonant effects can lead to a phase
shift φnr and a rotation ρ of the evolving magnetization and depend on the strength ν2(t) = γ B2 ( t )
of the applied magnetic field B2(t) [34, 35]. The rotation ρ is towards the positive z axis around an
axis perpendicular to the axis along which the rf pulse is applied. Both φnr and ρ depend on the
length τp of the pulse train applied and on the average of the square of the field strength, <ν22(t)>,
in addition ρ is proportional to the average of the field strength:

-22-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

φnr = π <ν22(t)> τp/∆ν (2.21)

ρ = π <ν22(t)><ν2(t)> τp/(∆ν)2 = φnr<ν2(t)>/∆ν (2.22)

In these equations φnr and ρ are given in units of radians. Eq. (2.21) and (2.22) exhibit clear dif-
ferences between ρ and φnr. The rotation ρ is always smaller than the phase shift φnr and when the
average strength of the applied field <ν2(t)> is zero then the rotation ρ, but not φnr, vanishes com-
pletely, a situation encountered with composite pulse decoupling where <ν2(t)> = 0 due to the
applied phase cycling schemes. A typical numerical example relates to the decoupling of alpha
carbon resonances from carbonyl carbons which are separated by ∆ν = 18000 Hz on a 600 MHz
NMR spectrometer. The application of an 8 ms long, low power WALTZ decoupling sequence to
carbonyl carbons using a 1 kHz decoupling field strength results in φnr = 1.4 rad (80o) and ρ = 0.
Using instead of WALTZ a rectangular 180o pulse at the carbonyl frequency with its sixth null at
18000 Hz (Eq. (2.20)) to refocus the effect of the carbonyl couplings, the values change to
φnr = 0.13 rad (7.5o) and ρ = 0.6o.

In the course of a multidimensional NMR experiment selective decoupling may be applied dur-
ing an evolution time which is incremented. In this situation τp linearly increases and considering
Eq. (2.21) it becomes clear that the time dependent phase φnr will manifest itself as a frequency
shift νnr:

νnr = <ν22(t)>/2∆ν (2.23)

With the numerical example given above using WALTZ decoupling, the frequency shift νnr
becomes 27.8 Hz. The numerical examples show that the non-resonant effects described by φnr
and ρ may cause severe signal loss and care has to be taken to correct for their influence. Four
different procedures can be envisaged to compensate for non-resonant effects [35]:

1. Adjusting the phase and flip angle of the pulse directly following or preceding the occurrence
of non-resonant effects.
2. A phase error occurring during an evolution time can be corrected by applying a phase correc-
tion after Fourier transformation.
3. Compensation by modulating the amplitude of the pulse train with a cosine function which
results in the application of the disturbing field at +∆ν and –∆ν and thereby cancelling the
effects at the frequency of interest.
4. Applying the disturbing field twice, once before and once after a non-selective 180o pulse and
thereby refocussing the adverse effects.

2.2.2 Amplitude modulated pulses

Modulating the amplitude of a pulse permits the design of specific excitation profiles. Since the
shape of a pulse and its excitation profile are not related by a Fourier transformation [16] more
elaborate procedures must be used to find the optimal pulse shape based on a desired excitation
profile. A large selection of pulse shapes have been developed and characterized [58]. From a
given pulse shape the excitation profile can be calculated by integrating the Bloch equations
given in the Appendix. Software packages on commercial spectrometers include corresponding
routines (Bloch simulator) which help to choose the appropriate shape for a specific experiment

-23-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

and to determine its parameters. In general a good amplitude modulated pulse should have an
adequate frequency selectivity, uniform excitation, uniform phase behaviour and a short duration.
Some of these desired properties contradict each other. Improving the selectivity, for example,
tends to increase the pulse duration which should be kept as short as possible to counteract relax-
ation losses. During the application of a selective pulse, the magnetization components of interest
accumulate an offset-dependent phase error. If the phase error is approximately linear across the
excitation bandwidth it can be refocused by a non-selective 180o pulse after a suitable delay
within or after the selective pulse [59]. Far from its irradiation frequency a selective pulse may
introduce non-resonant phase and amplitude errors as described by Eqs. (2.21) and (2.22). These
deficiencies can be corrected using the same methods as described for rectangular pulses in Sec-
tion 2.2.1.

The performance of an amplitude modulated pulse depends on the initial state of the magnetiza-
tion. A selective 180o pulse that provides good inversion properties for longitudinal magnetiza-
tion in general does not perform well as a 180o refocusing pulse for transverse magnetization. For
this reason, some shapes of pulses, for example the BURP pulses [60], are grouped into families
with a member for excitation, inversion and refocusing. The most frequently used shaped pulses
are Gaussian, sinc with no or one pair of side lobes, Gaussian cascades [61] which are based on
individual Gaussian shaped pulses, and pulses of the BURP family [60].

The application of amplitude modulated pulses in a pulse sequence requires all stages of the
transmitter pathway of the spectrometer to be linear, otherwise the shape and hence the excitation
profile deviate from the one selected. The direct determination of the pulse length of an ampli-
tude modulated pulse can be rather tedious. In this situation the use of a Bloch simulator program
which integrates the Bloch equations (Eq. (A.1.1)) seems more efficient for the determination of
the parameters of a shaped pulse by using the known parameters of a rectangular 90o pulse to
determine the field strength of a given power setting. A prerequisite for such calculations is the
linearity of the transmitter channel. In addition amplitude modulated pulses can be rather sensi-
tive to rf inhomogeneity of the coil in the probe and therefore require good rf homogeneity for
best performance.

2.2.3 Amplitude and phase modulated pulses

A pulse excites a spectral range around its irradiation frequency, however some experimental
techniques require the pulse to excite at a frequency different from this position. For example in
the middle of a period of free precession a selective 180o pulse may need to be applied to the
amide protons to decouple them from the alpha protons. If the carrier is to remain on the water
resonance this requires a homonuclear off-resonance selective pulse. Off-resonance pulses have
the advantage over switching the carrier frequency in that the latter method generates a phase
shift which must be taken into account and complicates the phase setting of subsequent pulses
applied to the precessing magnetization. The center of excitation of a pulse cannot only be
changed by changing its frequency but also by changing its phase during the application. Two
types of phase modulated pulses can be distinguished: pulses shifted off-resonance by a fixed fre-
quency and pulses where the frequency is swept during their application. A pulse with a fixed off-
resonance frequency shift νoff during the application can be obtained by linearly increasing the
pulse phase Φ(t) with time while keeping the carrier frequency fixed at νo:

cos(2πνot + Φ(t)) = cos(2πνot + (Φo + 2πνoff t)) (2.24)

-24-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

The increment per unit time depends on the required offset frequency νoff from the carrier fre-
quency νo. A positive phase increment will shift the frequency to higher values, decrementing the
phase lowers the effective frequency of the pulse. For example, amide protons or carbonyl car-
bons resonate at a higher absolute frequency than methyl resonances of protons or carbons,
respectively. The time point t = 0 sets the reference phase Φo for the off-resonance pulse. Any
further pulse on transverse magnetization of the same nuclear species during the pulse sequence
at a later time point T will have the phase Φ(T) which in general deviates from Φo. The perform-
ance of pulses which invert z magnetization will usually be independent of Φ(T). However, off-
resonance pulses that create or act on transverse magnetization require the phase Φ(T) to be
adjusted properly otherwise signal may be lost. Φ(T) can be calculated on the basis of Eq. (2.24)
and for example the proper time chosen where Φ(T) = Φo.

When the phase Φ(t) in Eq. (2.24) depends non-linearly on the time t the effective frequency
changes during the pulse. An important group of pulses using frequency sweeps during their
application are the adiabatic pulses. These pulses excite, invert or refocus magnetization over a
very wide frequency range at the cost of a longer pulse duration and a phase dispersion across the
excitation bandwidth. In applications where such pulses excite or invert magnetization they are
robust to rf inhomogeneities but not when applied for refocusing [62]. An additional feature
makes adiabatic pulses very attractive: doubling the rf field strength of the pulse quadruples the
bandwidth covered [62]. A distinct advantage over conventional pulses which excite a bandwidth
proportional to the strength of the pulse.

The concept of adiabatic pulses can be understood with the help of a description in the rotating
frame. With the applied radio frequency field B1 far above resonance the effective field Beff corre-
sponds to the residual magnetic field Bz (Eq. (A.1.3) and Fig. A1 in the Appendix) and the mag-
netization M stays aligned along the positive z axis. Sweeping the frequency of B1 towards
resonance will tip away Beff from the z axis towards the transverse plane. For a sufficiently slow
sweep the magnetization M stays aligned with the changing direction of the effective field. At
resonance Beff and M lie in the transverse plane. As the frequency of the exciting field B1 passes
through resonance and subsequently is swept to frequencies much lower than resonance, the
magnetization M moves on towards the negative z axis (Eqs. (A.1.3) and (A.1.4)). For M to fol-
low Beff the adiabatic passage condition must be fulfilled [29]:

|dΘ/dt| << γBeff with Beff = (B12(t) + Bz2(t))1/2 (2.25)

where Bz represents the offset from the resonance frequency and Θ describes the angle between
the effective field and the z axis (Eq. (A.1.4)). With Eq. (2.25) and Eq. (A.1.4) one finds that the
critical stage of any adiabatic sweep is the point where B1 sweeps through resonance for a given
magnetization component resulting in the relation

dBz(t)/dt << γB12(t) (2.26)

In addition to the condition described in Eq. (2.26) adiabatic passage requires that no relaxation
occurs during the sweep period. In practical implementations the field cannot be swept starting at
an infinitely large offset, however when choosing a finite starting value the effective field and the
magnetization M are not aligned. Consequently M precesses around Beff which degrades the per-
formance of adiabatic pulses. To reduce this initial precession the starting angle Θο must be as
small as possible. For small values of Θ one finds with Eq. (A.1.4)

-25-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

Θ = B1/Bz (2.27)

Based on Eq. (2.27) small values of Θ require large initial offsets for the start of the sweep. For
an angle Θ corresponding to 1o (0.0175 rad) and with γ B1 = 2 kHz an offset of at least 115 kHz
is necessary. Eq. (2.27) suggests an alternative approach to reduce the initial offset substantially.
If the sweep starts with an amplitude of B1 at zero and is smoothly increased to its nominal value,
then the offset where the sweep must start, can be reduced making adiabatic sweeps more effi-
cient [62]. The same consideration for the end of the adiabatic sweep leads to the conclusion that
the amplitude of B1 should be smoothly reduced to zero.

As long as Eq. (2.26) stays fulfilled, the time dependent phase Φ(t) of the pulse can have any
functional form. For simplicity very often linear sweeps are applied (Eq. (2.28)). With the total
sweep range F and the total duration of the sweep τs the change of Bz per unit time (dBz/dt)
becomes 2πF/γτs. Linear sweeps result in a quadratic time dependence for the phase Φ(t) of the
constant frequency νo during the application of an adiabatic pulse:

cos(2πνot + Φ(t)) = cos(2πνot + (Φo + (2πF/γτs) t2)) (2.28)

Most applications of adiabatic pulses use single inversion pulses or trains of inversion pulses in
decoupling sequences. When used to refocus transverse magnetization, their inherent long execu-
tion time may cause problems with fast relaxing magnetization as well as with evolution due to
scalar couplings which may modulate the signals. On the other hand, depending on the chemical
shift range of scalar coupled nuclei, partial decoupling may be achieved during the adiabatic
pulse [63, 64]. Applications for refocusing are not yet very common but their applicability has
been demonstrated [63, 65, 66].

2.3 Magnetic field gradients

Pulsed magnetic field gradients (PFGs) have been used in NMR spectroscopy for more than 30
years to study the diffusion of molecules in solution [67, 68]. However, PFGs became a routine
tool in high resolution NMR in solution [69–73] only since the introduction of actively shielded
gradient coils which offer short recovery times (Section 3.5.2) for the re-establishment of the
very homogeneous magnetic field after the application of a PFG. The resonance frequency of
nuclei placed in a magnetic field gradient becomes dependent on their spatial location. A linear
PFG with the strength G per unit length along the z direction produces the following dependence
of the resonance frequency ω(z) on the z coordinate

ω(z) = ωo+ γGz (2.29)

where ωo is the resonance frequency without the application of a PFG. The evolution of the oper-
ators used to describe an NMR experiment (Section 2.1.2) become dependent on the vertical
position in the sample. Using the shift operator basis the following transformations are obtained
due to the action of a gradient pulse applied for a duration τ:

I– ---> I– eiγGzτ
I+ ---> I+ e–iγGzτ (2.30)
Iz ---> Iz

-26-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

In these transformation rules the evolution due to chemical shift and J coupling are not included.
The phase factor γGz describing the spatial dispersion of the resonance frequencies depends on
the gyromagnetic ratio γ which makes the action of a gradient less efficient for nuclei with a
small γ. For example, applying an identical gradient on transverse proton and nitrogen magneti-
zation will result in a spread of resonance frequencies across the sample which is ten times larger
for proton than for nitrogen nuclei. Eq. (2.30) describes how the evolution of magnetization com-
ponents depends on the z coordinate. However, the magnitude of the macroscopic magnetization
M will be the integral over the sample length L which results in the following relationship
sin ( γGτL ⁄ 2 )
M = ---------------------------- = sinc(γGτL/2) (2.31)
γGτL ⁄ 2
The sinc function in Eq. (2.31) has a damped oscillatory behaviour and is zero only at specific
values. In between the zero crossings the detectable signal can reach appreciable values and can
interfere with the performance of the experiment requiring the value for γGτL to be optimized.
For example, a moderate gradient with a duration of 1 ms and a strength of 0.1 T/m applied to
proton magnetization using a typical commercial probe will show a maximum signal recovery
between the first and second zero crossing of about 0.5% of the original intensity which, for
example, results in a substantial residual signal for the solvent resonance.

Pulsed magnetic field gradients find three main applications: (i) spatial encoding of coherences,
(ii) elimination of unwanted coherences and (iii) selection of coherences. Whereas point (i) is
used in diffusion measurements already for a long time, points (ii) and (iii) have become increas-
ingly important in high resolution NMR spectroscopy only in recent years replacing or supple-
menting phase cycling schemes for the selection of a specific coherence transfer pathway
(Fig. 1). Eq. (2.30) illustrates the basic principle used. Every application of a PFG introduces a
phase factor of the form γGzτ. For the desired coherence pathway the sum of all these phase fac-
tors must be zero. All other pathways do not result in detectable signal if sufficiently strong gra-
dients are used (Eq.2.31). For multiple quantum coherence every operator in the product will
evolve according to Eq. (2.30). Consequently the sensitivity of the precession frequency of a
coherence to magnetic field gradients, will be proportional to its order. Thus homonuclear double
quantum coherence will be two times as sensitive to magnetic field gradients as single quantum
coherence, while homonuclear zero quantum coherence will be unaffected. For heteronuclear
multiple quantum coherences the different gyromagnetic ratios have to be taken into account (Eq.
(2.30)). The application of gradients has the potential to select the desired signal in one scan in
contrast to phase cycling which requires the repetition of the experiment and the subtraction of
unwanted signals. Unfortunately, only half the signal defocused by a gradient can be refocused if
a 90o rf pulse is applied between the two gradients. This drawback can be verified using Eqs.
(A.2.6) and (2.30). Such a signal loss constitutes a common feature when gradients are used for
pathway selection. As discussed in Section 4.5.3 experimental techniques exist for some cases
that prevent this signal loss. Further signal losses can occur due to diffusion losses. In between
the defocusing and refocusing gradients the molecules with the nuclei diffuse to a different loca-
tion preventing a complete refocussing (Eq. 4.3). Diffusion losses are largest for small molecules
particularly the solvent magnetization.

Magnetic field gradients can also be applied in the form of spatially inhomogeneous radio-fre-
quency pulses [74, 75]. Either using the inherent rf inhomogeneity of the transmitter coil or using
a separate coil designed to deliver inhomogeneous B1 fields [76]. The first method is frequently
used in the form of spin-lock purge pulses which destroy magnetization components that are not

-27-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

aligned along the axis defined by the phase of the rf pulse [77]. The second more efficient method
requires special hardware and has not yet found many applications. Radio-frequency gradients
possess the inherent advantage over static pulsed magnetic field gradients that they can be applied
frequency selective [78]. On the other hand limitations arise due to rf heating effects and the
rather modest maximal strength for rf gradients which typically is limited to 0.1T/m.

2.4 Data acquisition

2.4.1 Digitizing the signal

At the end of a pulse sequence the free induction decay (FID) of the magnetization of one nuclear
species is measured. The sensitivity of the detection is proportional to γ3/2 (Eq. (4.2)), and when-
ever feasible the magnetization should be transferred to and detected on the nuclear species with
the largest gyromagnetic ratio. This procedure requires an efficient coupling between the differ-
ent species of nuclei to retain the sensitivity advantage despite of the inevitable signal losses dur-
ing the transfer. When working with macromolecules in solution the proton constitutes the
preferred nucleus for detection. However, irrespective of the nuclear species detected, the reso-
nance frequency is dozens or hundreds of MHz. Whereas the difference of resonance frequencies
of one particular nuclear species in different environments, the chemical shift range, is very
small, often only a few kHz. Technically speaking a high frequency, the carrier frequency, is
modulated by low frequency signals which represent the spectrum of interest. Subtracting the
carrier frequency from the signal one obtains a modified signal that contains only frequencies
between zero and a few kHz. Spectrometers use frequency mixing schemes to transform the high
resonance frequencies to a lower frequency range since the necessary high dynamic range digitiz-
ers exist only up to frequencies of a few hundred kHz. In addition a quadrature detection scheme
(Section 3.3.1) delivers two signals which are 90o out of phase which allows discrimination
between positive and negative frequencies with respect to the carrier frequency. The carrier fre-
quency and the frequency of the rf pulses on the observe channel are usually identical and conse-
quently quadrature detection enables the user to place the carrier in the middle of the spectrum
resulting in a more efficient excitation of the resonances (Fig. 7). The low frequency analog sig-
nal thus obtained is digitized at equidistant time points and stored on a computer.

The sampling theorem [79] specifies that for the representation of the analog signal in digital
form the sampling rate of the digitizer must be at least twice the highest frequency in the signal.
With a quadrature detection system the highest frequency in the spectrum is half the spectral
range covered by the signals. In other words, the time increment, the dwell time ∆, between two
digitized points must be equal to or smaller than the inverse of the sweep width 2νN which covers
the frequency range from –νN to +νN with the carrier frequency being at zero frequency. Slower
sampling results in folding of the signals with absolute frequencies larger than νN into the spec-
tral range of interest. The folded signals will be represented by a wrong frequency and in general
will have a phase that differs from the one of unfolded signals. One dimensional 1H NMR spectra
of proteins contain a very large number of resonances and folding of additional signals into the
spectral range is not desired. This is in contrast to multidimensional experiments where folding
quite frequently helps to reduce the spectral range in additional dimensions.

The data can be digitized with two different sampling techniques. The first method digitizes the
two quadrature channels simultaneously and delivers data points that can be regarded as complex

-28-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

numbers. This simultaneous digitization method simply measures the orthogonal components of
the precessing magnetization which correspond to the two quadrature channels. The minimal
sampling rate 1/∆ becomes equal to the sweep width 2νN and a complex Fourier transform of the
FID produces the spectrum. The second method digitizes the two quadrature channels sequen-
tially with a sampling rate of 4νN. Consequently two time shifted signals are measured and cor-
rect processing requires that every second data point pair must be inverted in sign before the
signal is submitted to a real Fourier transformation resulting in the spectrum [80]. The sequential
digitization can be understood using the concept of the rotating frame of reference at the carrier
frequency (Fig. A1 in the Appendix). Including the effect of the sign inversion the sampling
method corresponds to a sequential sampling of the magnetization components along the x, y, –x,
–y, x, y, –x, ... axis in the rotating frame. From one digitized point to the next the axis along which
the magnetization is sampled rotates by 90o about the z axis instead of being static as with simul-
taneous digitization. In other words the reference frame for detection rotates within two dwell
times by 360o with respect to the carrier frequency which corresponds to a frequency shift of νN.
Consequently, the zero frequency shifts from the middle to the edge of the spectrum and all reso-
nance lines have a frequency larger than zero that lies between 0 and 2νN. This detection scheme
is based on a linear phase incrementation with time and is referred to as TPPI method which
stands for time-proportional-phase-incrementation [80, 81].

The requirement for a constant dwell time and the sampling theorem are consequences of the
Fourier transformation method used to obtain the spectrum from time domain data. Other trans-
formation methods permit varying dwell time lengths. In general, time periods in the FID which
have better signal-to-noise ratio (S/N) should be sampled more frequently than those with less
S/N. Different sampling schemes were proposed in combination with maximum entropy recon-
struction for the transformation into the frequency domain [82–84]. Considering the modest ben-
efits obtained, these procedures do not seem to be justified in most applications due to the
increased complexity involved in the data collection, transformation, and sensitivity to parameter
settings. Currently, equidistant sampling is by far the most frequently used scheme.

2.4.2 Handling the water resonance

Many NMR measurements with biological macromolecules rely on the presence of exchangeable
protons in the molecules and must therefore be performed in H2O solutions which contain only a
small amount of D2O for the field-frequency lock (Section 3.3.3). Measuring the spectrum of a
protein dissolved in H2O at a typical concentration of 1 mM requires a dynamic range of the
spectrometer which cannot be obtained without the reduction of the signal intensity of the water.
Numerous techniques for the reduction of the intense solvent line have been developed and a
number of excellent reviews on these techniques exist [e.g. 21, 85]. Here only a few selected
techniques can be mentioned. When 2D methods were first developed it was a real challenge to
obtain spectra from H2O solutions and initial success was achieved by presaturation of the water
resonance [86]. Later spin-lock pulses were introduced which destroy the water magnetization
due to the spatial rf field inhomogeneity during the pulse [77]. Unlike solvent presaturation
which happens during the relaxation delay spin-lock pulses occur later in the sequence allowing
the water to recover during the relaxation delay reducing the degree of saturation of the water
magnetization. This enhances the signal intensities of exchangeable protons of the protein, for
example for the amide protons which play an important role for many experimental schemes.
Magnetic field gradients act in a very similar way to spin-lock pulses, however, often better sup-
pression can be obtained [69, 70, 87]. Work at higher pH demands for even better conservation of

-29-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

water polarization due to the faster exchange rates of labile protons and methods were developed
that attempt to conserve the equilibrium water magnetization during the pulse sequence and flip
the water back to the positive z axis before the start of the acquisition [88].

The huge magnetization of water exhibits special behaviour after the excitation by a rf pulse. The
magnetization of water rotates back to the positive z axis much faster than expected from relaxa-
tion. The large signal induced in the receiving coil during precession of the water magnetization
creates a magnetic field that acts on the water resonance like a shaped rf pulse and rotates it back
to the z axis. This effect, known as radiation damping, may interfere with the performance of
experiments where the water resonance cannot be destroyed, for example when studying interac-
tions between water and protein protons (Section 5.3). When using water flip-back methods [88,
89] the H2O resonance should never be aligned with the negative z axis just prior to acquisition
because radiation damping can efficiently rotate the water magnetization into the transverse plane
which can overload the receiver system during acquisition. The strength of radiation damping is
proportional to the size of the water magnetization and the quality factor of the probe. Conse-
quently it becomes a more serious problem with high magnetic field strengths and higher sensi-
tivity probes. Different techniques have been developed to suppress radiation damping during a
pulse sequence without destroying the water resonance. Using pulse field gradients the water can
be defocused during periods of free precession and refocused just before the next pulse which is
different from 180o [90]. This procedure conserves the water magnetization except for diffusion
losses which must be minimized. Another method temporarily reduces the quality factor Q of the
receiving coil and therefore decreases the current induced in the coil and hence the radiation
damping effect [91]. This technique requires special circuitry in the probe known as Q switch
which allow rapid changes in the Q value. A third method uses an active feed back loop driven by
the detected water magnetization. Such a device sends very weak rf pulses to the coil counteract-
ing the radiation damping effect. In this way the state of the water magnetization can actively be
controlled and suppression as well as an enhancement of the radiation damping effect can be
obtained [92, 93].

The very large polarization of water not only interferes with the water resonance itself but
slightly alters the main static magnetic field which changes the resonance frequencies of all
nuclei. This demagnetizing field effect amounts to a rather small change of the main magnetic
field of about 1.5 Hz for protons at 750 MHz when the water magnetization is switched from the
positive z axis to the negative z axis. None the less difference experiments such as hydration
experiments may exhibit subtraction artifacts and suffer losses of performance due to these small
field changes [94, 95] (Section 5.4).

2.4.3 Decoupling during acquisition

During acquisition of the signal, scalar coupling between nuclei splits the resonance line into
different components. Often it would be preferable to decouple the individual nuclei from each
other to obtain the best sensitivity and the minimum number of lines. For different species of
nuclei decoupling is usually straight forward. For example protons attached to 15N nuclei in
labelled proteins can be decoupled from the 15N nuclei by continuous inversion of the magnetiza-
tion of all the 15N nuclei during acquisition of the proton signal without adverse effects on the
measured spectrum. Practical consequences and limitations of heteronuclear decoupling will be
discussed in Section 4.3.

-30-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

Decoupling nuclei of the same species from each other requires more elaborate schemes and, of
course, the decoupled nuclei cannot be directly detected. Homonuclear decoupling requires rf
irradiation at the receiving frequency necessitating that the preamplifier (Section 3.3.1) is
switched off during the period of irradiation. The total dwell time, DWt, is split into a period
where the preamplifier is turned off and a period where the signal is received, DWr. This time
sharing reduces the measured signal since the receiver can integrate the signal only during the
time DWr. Switching off the preamplifier reduces not only the signal by a factor DWr/DWt but
simultaneously the noise proportional to (DWr/DWt)1/2. For the signal-to-noise (S/N)hd with
homonuclear decoupling one obtains the reduction

(S/N)hd = (DWr/DWt)1/2 (S/N) (2.32)

where S/N is the signal-to-noise ratio without decoupling. For example using half the dwell time
for homo-decoupling including the necessary switching delays of the hardware reduces (S/N)hd
by 2 compared to that obtainable without decoupling. Instrumental limitations may reduce the
obtainable (S/N)hd further. Not only decoupling but any disturbance which interferes with the
integration of the signal reduces the sensitivity, for example, when applying defocusing and refo-
cusing gradients during the acquisition of the signal [96].

2.4.4 Oversampling and digital filtering

Before the signal reaches the digitizer the spectral bandwidth must be limited by a suitable ana-
log filter to the frequency range (sweep width) selected. Otherwise higher frequency noise folds
into the spectral region of interest and reduces the obtainable S/N. Such analog filters introduce a
frequency dependent retardation of the signals and show transient oscillations after a sudden
change of the voltage which typically happens at the start of the FID. This characteristic behav-
iour can result in baseline distortions in the spectrum and the necessity of first order phase correc-
tion which itself introduces baseline curvature [97–99]. The distortions in the spectrum due to the
filters can be minimized by judicious choice of the time at which the receiver opens and the first
data point is sampled [100]. Spectrometers do start accumulation according to this scheme but
the appropriate delays usually require fine adjustment for best performance. Use of analog filters
present somewhat of a dilemma; on the one hand analog filters introduce distortions which
increase with the steepness of the frequency cut off of the filters, on the other hand the filter
should be as steep as possible for efficient suppression of signals and noise outside the chosen
sweep width. This dilemma can be overcome with oversampling when the digitized frequency
range is much larger than the one finally desired. Oversampling allows the use of analog filters
with a much broader transition frequency range between full passage and full attenuation. Such
filters have short filter delays and negligible transient oscillations which reduces their detrimental
effects on the baseline of the acquired spectrum [101]. The analog narrowband filters with steep
cutoffs for the adjustment of the final spectral range can then be replaced by digital filters. Spe-
cial digital signal processors are programmed to apply very steep filters without introducing base-
line distortions. The digital signal processors are inserted between the digitizer and the computer
memory which stores the data so that the storage requirements of oversampled data does not
exceed the space used by conventionally acquired data. Oversampling provides a further advan-
tage by increasing the dynamic range of the digitizer [102]. The extra bits n obtained with over-
sampling can be calculated from the following expression

n = log2(SWov/SW) / 2 (2.33)

-31-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

where SWov is the oversampled frequency range, SW the desired spectral range and the logarithm
is base 2. For example, with SW = 10 kHz and SWov = 160 kHz the effective dynamic range of a
16 bit digitizer corresponds to 18 bits.

The quadrature detection scheme used in spectrometers (Section 3.3.1) requires a phase differ-
ence of exactly 90o and identical amplification in the two signal paths which in practice cannot be
achieved. As a consequence positive and negative frequencies with respect to the carrier cannot
be completely separated, leaving typically less than 1% of the wrong signal as a quadrature
image on the wrong side of the carrier frequency. Still, these signals may have a detrimental
effect in multidimensional NMR spectra which contain resonances with widely different signal
intensities. The images can usually be suppressed beyond detection by exchanging the signal
pathways in subsequent scans which are added [97]. This cancellation method requires simulta-
neous changing of the phases of all rf pulses and the receiver by 90o between scans. When this
procedure is extended to a four-step phase cycle using the phases 0o, 90o, 180o and 270o it is
called CYCLOPS [103]. In addition to the quadrature images CYCLOPS suppresses the spike at
zero frequency which is caused by different dc offsets in the two receiver channels. When the car-
rier is set on-resonance with H2O the original two step phase cycle suffices since the water reso-
nance usually causes a distortion at zero frequency which is much larger than the zero frequency
spike. Modern spectrometers use a different scheme sometimes referred to as digital quadrature
detection. In this scheme the signal is mixed to a low frequency range which does not contain
zero frequency. A narrow bandpass filter programmed into the digital signal processor used to
reduce oversampled data then selects the desired spectral range which does not contain quadra-
ture images. The technique produces spectra which cannot contain quadrature images and hence
renders CYCLOPS unnecessary which reduces the phase cycling schemes, a definite advantage
in multidimensional NMR spectroscopy where phase cycles may be longer than necessary to
obtain sufficient signal-to-noise causing unnecessary long experiment times.

2.5 Multidimensional NMR

Multidimensional NMR was first proposed in 1971 [3] and first applications were published a
few years later [4]. A multidimensional data set depends on different time variables which can be
transformed into multiple frequency domains. For the introduction of an additional time variable
a suitable delay during the pulse sequence, the evolution time, is incremented during the experi-
ment [16, 40, 104]. Only during the last time period the signal is physically acquired which is
thus often referred to as the direct dimension in contrast to all other dimensions referred to as
indirect dimensions. The magnetization measured depends on the evolution times during the
pulse sequence and this dependence is reflected by a change in amplitude and/or phase of the
acquired signal. A typical n-dimensional NMR experiment follows the scheme

excitation - (evolution - mixing)n–1 - detection (2.34)

where the bracket repeats (n–1) times. In a simple example the excitation just consists of a 90o
pulse and the single mixing could be another 90o pulse which mixes the non-equilibrium states
obtained after the first pulse and the evolution time. This sequence is usually referred to as COSY
experiment [3, 4]. The sampling of frequencies in evolution periods must fulfil the same require-
ments with respect to folding and the sampling theorem as the measurement of the FID in the
detection period. A subsequent n-dimensional Fourier transform provides a n-dimensional spec-

-32-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

trum that depends on n frequency variables. If two nuclei suitably interact with each other in a
mixing time between two evolution periods or between the last evolution time and the detection
period this interaction will be manifested by a resonance in the spectrum, a cross peak, at a posi-
tion characterized by the precession frequencies of the interacting nuclei. Nuclei that did not have
any interaction in the mixing time will show the same frequency in the two successive evolution
times. The later resonances, for example, form the diagonal resonances observed in homonuclear
2D spectra. Magnetization components that relax during the scheme will show zero frequency for
all evolution periods that occurred before the magnetization relaxed. Such signals are called axial
peaks and are usually not of interest.

The number of dimensions measured should be kept minimal since increasing their number
reduces the sensitivity due to the additional quadrature sampling step (see below), larger signal
losses due to relaxation and additional rf pulses. Theoretical considerations show that nD spectra
do not offer fundamentally new information but are just the mathematical products of the corre-
sponding basic 2D experiments [105]. More than two dimensions should mainly be used to
reduce spectral overlap. Most multidimensional experiments are run with two or three dimen-
sions. Only for few experiments a four-dimensional (4D) scheme may be beneficial. One exam-
ple for a rather frequently used 4D experiment is a homonuclear [1H, 1H]-NOESY which is
resolved into a 13C and a 15N dimension [106].

Based on the scheme given in Eq. (2.34) every evolution time is represented by a separate dimen-
sion in the spectrum. But experiments can be designed which combine the information from dif-
ferent evolution periods into one dimension. The basic principle of these experiments is very
simple. Whenever two evolution times are incremented simultaneously the corresponding fre-
quencies occur in one spectral dimension reducing the number of dimensions by one but keeping
the same information [107, 108]. In such experiments one frequency is encoded conventionally
on the chemical shift scale. This signal is modulated by the offset frequency of the second
nucleus leading to a splitting similar to a J coupling but representing the second chemical shift.
Hence, the carrier frequency for the second nucleus must be set to one edge of the spectrum since
positive and negative frequencies with regard to the carrier result in the same splitting. The chem-
ical shifts of the two nuclei are obtained from the center frequency of the two lines and from their
splitting [107, 108].

In multidimensional spectra, resonances should have absorptive lineshapes in all dimensions for
the best spectral resolution and quality of a nD spectrum. To prevent the undesirable mixture of
absorptive and dispersive signals the two orthogonal transverse magnetization components repre-
sented by the operators Ix and Iy must be obtained separately from the final signal. In the frame-
work of coherence pathways both +k and –k coherence levels of a k quantum coherence during an
evolution time must be maintained [16, 40] which then allows to separate the operators Ix and Iy
in the final signal. For practical applications this requires that sine and cosine components of the
signal during the evolution time are not mixed when accumulating the data. In other words these
components should not be summed into one FID otherwise only n- or p-type peaks are obtained
which do not have a pure phase; p-type peaks result from a coherence pathway whose coherence
levels have the same sign during the evolution and final detection period, opposite signs lead to n-
type peaks [16] (Fig. 1).

In analogy to the direct acquisition of the signal a quadrature scheme must be applied in the indi-
rect dimension. For the indirect dimensions the sine and cosine components of the signal cannot
be measured simultaneously. The two quadrature components are typically sampled consecu-

-33-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

tively changing the phase of the 90o rf pulse at the start of the evolution time by 90o according to
one of the schemes given in Table 3. This procedure reduces the sensitivity for every additional
dimension by a factor 2 . With respect to sensitivity it is important to note that all noise in the
spectrum originates from the measurement in the direct dimension. During the evolution periods
the signal may be reduced due to relaxation but no noise is added. As there is no noise to be
excluded from the bandwidth of interest there is no need for filters in the indirect dimensions.
None the less the baseline in indirect dimensions may be distorted. This baseline distortion origi-
nates from linear phase corrections applied in the indirect dimensions. Therefore linear phase
corrections must be avoided by adjusting the initial delay of the evolution time properly. In prin-
ciple the minimal evolution time should be zero. Often this delay cannot be zero. For example,
the finite duration of the rf pulses at the beginning and at the end of the evolution period result in
evolution of the magnetization requiring a linear phase correction for an absorptive spectrum. It
is customary to sample the first point at half the dwell time. For data sampled as complex data
points this requires a linear phase correction of 180o across the spectrum which does not intro-
duce baseline curvature [97, 99]. As an additional advantage folded peaks have opposite sign and
are easily detected if they do not overlap with an unfolded resonance line. This feature proves
useful in some instances to reduce the spectral range together with the number of increments
required, as for example in the indirect carbon dimension of heteronuclear multidimensional
spectra [109]. Hence folding can be used to extend the phase cycling and/or to reduce disk stor-
age requirements.

For quadrature detection in indirect dimensions several techniques are available [110] (Table 3).
The individual schemes sample the data at different time points and with a different sign which
requires suitable Fourier transform (FT) algorithms to be used. The particular method affects
both the frequency position of the axial peaks and the folding behaviour for resonances outside

Table 3
Properties of different quadrature sampling schemes

quadrature sampling of x and y components‡ FT mode axial folded

scheme n n+1 n+2 n+3 peaksƒ peaks†
Redfield x(t) y(t+∆/2) x(t+∆) y(t+3∆/2) real§ center mirrored

TPPI x(t) y(t+∆/2) –x(t+∆) –y(t+3∆/2) real edge mirrored

Complex x(t) y(t) x(t+∆) y(t+∆) complex center cyclic

Complex-TPPI x(t) y(t) –x(t+∆) –y(t+∆) complex§ edge cyclic

‡ The four columns indicate consecutive time points at which four magnetization components are
sampled which evolve during the evolution time, the time period ∆ stands for the dwell time which is
equal to the inverse of the frequency range sampled; x(t) and y(t) refer to the x- and y-component at
time point t, respectively; the sign indicates the use of the positive or negative component.
ƒ Axial peaks appear either in the center or at the edge of the spectral range selected.
† Folded peaks are either mirrored at the closer edge of the spectral range or cyclically shifted into the
spectrum by subtraction of the spectral range.
§ Every second pair of data points in the indirect dimension has to be inverted in sign before Fourier
transformation (FT), the same goal can be achieved by changing the phase setting of the receiver by
180o.

-34-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

the selected spectral range (Table 3). The best overall performance can be expected using a com-
bination of complex sampling [111] with TPPI phase incrementation [81]. This scheme allows to
adjust the initial delay to obtain optimal baseline behaviour and at the same time places axial
peaks at the edge of the spectrum [110]. In principle, axial peaks are suppressed using a suitable
phase cycling scheme, for example the phase cycle x, –x applied to the excitation pulse and the
receiver. None the less residual intensity should still be moved to the edge of the spectrum. TPPI
[81] and Redfield [80] quadrature detection are nowadays only used when their specific folding
properties are required.

2.6 Data processing

2.6.1 Transforming the time domain data into a spectrum

A detailed description of data processing methods used in biomolecular NMR is outside the
scope of this article. There are excellent monographs dealing with all aspects of data processing
in detail [112–114]. Here only a few basic aspects shall be mentioned. In NMR time dependent
oscillations of the magnetization are measured which can be represented in an analytical form as
a complex signal S(t).

S(t) = Σk exp(2iπνkt) exp(–t/T2k) (2.35)

where νk are the individual resonance frequencies and T2k the corresponding transverse relaxa-
tion times. The frequency components νk encoded in the modulation are typically analysed by a
complex Fourier transformation resulting in a spectrum with a real part containing absorptive
lines Ak(νk–ν) of the form

Ak(νk–ν) = T2k/(1+4π2T 2k
2
(νk–ν)2)) (2.36)

and an imaginary part containing dispersive lines Dk(νk–ν)

2
Dk(νk–ν) = 2πT 2k (νk–ν)/(1+4π2T 2k
2
(νk–ν)2)) (2.37)

In practice the signals in the real part of the spectrum may not be completely absorptive due to a
phase factor in the modulation of the signal. By a suitable mixing of the real and imaginary parts
using a phase correction routine, absorptive lines can be obtained in the real part and dispersive
lines in the imaginary part of the spectrum. Absorptive lineshapes are preferred since they have
narrower lineshapes (Eq. (2.36) and (2.37)) and hence offer a better resolved spectrum. In multi-
dimensional NMR experiments, proper acquisition schemes have to be used to avoid summation
of absorptive and dispersive lines in the real part of the spectrum (Table 3). A sum of absorptive
and dispersive resonances cannot be corrected by a mathematical procedure and severely
degrades the spectral quality. Alternative analysis methods to Fourier transformation are availa-
ble but are not frequently used because they often require extensive calculation times, are non-
linear algorithms which may distort the signal intensities and the quality of the results is sensitive
to parameter settings [113, 114].

Because the causality principle requires the measured FID to be zero for times smaller than zero
Fourier transformation of the FID leads to a relation (Kramers-Kronig relation) between the real

-35-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

and imaginary part of the spectrum [16]. The two parts can be calculated from each other by a
Hilbert transformation. Since Fourier transformation results in a real and an imaginary part with
identical information content the number of data points N in the FID must be doubled by adding
N zeroes at the end of the FID (zero filling). In this way N independent data points in the FID are
represented by N independent data points in the spectrum. More zero filling does smooth the rep-
resentation of the data by interpolation but no additional information can be obtained [7, 115]. To
avoid truncation artifacts which result in oscillations around individual signals zerofilled FIDs
must be brought smoothly to zero by multiplication with a weighting function which becomes
zero at the end of the FID.

The first data point in the FID is often measured as close as possible to the time point zero in
order to avoid a first order phase correction in the spectrum. With this procedure the intensity of
the first time domain data point should be averaged with the last data point in the FID to result in
a properly periodic signal [116], in practice this is often equivalent to a division of the first point
by two. The first data point represents the integral of the spectrum and overemphasizing its inten-
sity adds a constant offset to the spectrum which can be very disturbing in multidimensional
spectra since the offset usually varies from trace to trace. NMR software provides the possibility
to scale the first point to remove the baseline offset. In indirect dimensions where the measure-
ment starts after half a dwell time the situation is different. In this case the first data point repre-
sents the correct integral and no additional scaling is necessary but a linear phase correction of
180o has to applied across the spectrum.

In multidimensional experiments the maximal evolution time in indirect dimensions must often
be chosen to be rather short to achieve the best sensitivity and therefore often truncated FIDs are
obtained for indirect dimensions. Applying weighting functions directly to truncated FIDs dis-
cards some of the measured signal. In this situation linear prediction provides a possibility to
extend the experimental data in the time domain before applying weighting functions [113].
However, typically used algorithms determine four parameters for every signal: frequency, inten-
sity, linewidth, and phase. Enough data points must be available to prevent overfitting the data
and the number of predicted points should as a rule of thumb not exceed the number of measured
points. Linear prediction works best when the time-domain data to be extended contains a small
number of signals. In order to achieve this, all of the other dimensions of the spectrum should be
Fourier transformed before linear prediction in a given dimension. If more than one dimension is
to be linear predicted inverse Fourier transformations will be necessary. It should be noted that in
the first round of Fourier transformations window functions are not applied because this would
throw away information present in the FID. Even with linear prediction the resulting spectrum in
an indirect dimension of a 3D and especially a 4D experiment may still have a resolution of only
40 Hz/point or even less. Applying resolution enhancement filters to FIDs with such a low resolu-
tion is of little use and often simple cosine functions are applied. There are a variety of commer-
cial and non commercial programs which allow interactive evaluation of NMR data using a
graphics interface. However for time consuming calculations, such as linear prediction or sophis-
ticated baseline correction algorithms, it may be advantageous to use a program with no graphic
interface to enable processing in batch mode on high performance computers [117, 118].

-36-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

2.6.2 Referencing the chemical shift

The nuclear shielding constant σi describes the effect of magnetic interactions between electrons
and nuclei which affect the local magnetic field Bi experienced by the nucleus i with Bi = (1–
σi)Bo, and hence influences its resonance frequency. This chemical shift has long been known to
be extremely sensitive to a multitude of structural, electronic, magnetic and dynamical variables
and in principle contains a wealth of information on the state of the system under investigation.
This potential is largely offset by the difficulties of a detailed interpretation which attributes the
measured chemical shift to the various contributions. However, the large amount of data repre-
sented by the 1H, 13C and 15N shifts in known 3D NMR structures of proteins forms the basis for
a multitude of empirical and semi-empirical correlations with structural parameters. These corre-
lations can be used during the assignment process as an additional source of information [15,
119–122]. It is also possible to refine protein structures based on chemical shifts and even with
insufficient information for a structure determination chemical shifts can be used to distinguish
between alpha helix and beta sheet via the chemical shift index [120, 123]. This index provides a
simple tool for identifying protein secondary structure through analysis of backbone 13C chemi-
cal shifts. Further information on many aspects of the use of chemical shifts in studies of proteins
can be found in a recent review article [124].

The evaluation of chemical shift data requires an accurate referencing of NMR spectra both for
reproducibility and for the correlation of the chemical shift with structural properties. A wide
variety of reference compounds exist for various solvents and many nuclei. Referencing each
nuclear species individually is cumbersome and prone to errors. Reference substances may inter-
act with the protein or not be soluble in water. In addition the reference chemical shift may
depend on parameters like temperature, pH or ionic strength. External referencing circumvents
most of these problems. However, separating the solution and the standard into two compart-
ments introduces susceptibility effects which depend on the sample geometry and disappear only
for spherical inserts. With the proton omnipresent in biological samples indirect referencing of
the heteronuclei to a proton standard becomes attractive. Recently it was proposed to exclusively
use dimethylsilapentane sulfonic acid (DSS) as standard for referencing of proton chemical shifts
due to its favourable properties as a reference compound and to relate standards for heteronuclei
to DSS by a scaling factor Ξ which is specific for a selected heteronuclear standard [125]. The
absolute methyl proton frequency in DSS in the sample at a given field is simply multiplied by Ξ
to obtain the zero frequency for the heteronuclear chemical shift scale. Values of Ξ for 13C and
15
N [125] as well as for 19F and 31P [126] have been published for the standards 10mM DSS in
water (DSS), liquid NH3 (NH3), trifluoroacetic acid (TFA), and 85% ortho-phosphoric acid
(H3PO4), respectively, and are reproduced in Eqs. [2.38]–[2.42]

Ξ (1H, DSS) = 1.000000000 (2.38)

Ξ (13C, DSS) = 0.251449530 (2.39)
Ξ (15N, NH3) = 0.101329118 (2.40)
Ξ (19F, TFA) = 0.940866982 (2.41)
Ξ (31P, H3PO4) = 0.404807356 (2.42)

In practice the frequency difference between the methyl proton resonance of DSS and the carrier
(usually at the water resonance) must be subtracted from the absolute frequency of the carrier to
obtain the zero ppm reference for protons. Chemical shifts referenced to a variety of different ref-
erence compounds can be readjusted to conform to the indirect referencing discussed above using
published conversion data [125].

-37-
 Gerhard Wider: Technical aspects of NMR spectroscopy with biological macromolecules .......

The low digital resolution typically obtained in 3D and 4D NMR spectra makes calibration criti-
cally dependent on the spectral data point that represents the position of the carrier frequency. In
general the number of points N contained in a NMR spectrum is a power of 2. Assuming a num-
bering of the points from 1 to N the question arises if the carrier frequency is represented by point
N/2 – 0.5 or by point N/2 + 0.5. Using the fact that the frequency domain spectrum is a periodic
function point 1 and point N + 1 have identical values and hence point N + 0.5 represents the
position of the carrier frequency. However, the implementation in the software used may be dif-
ferent. When details of the transformation software are not known, a simple test allows determi-
nation of the position of the carrier frequency in the spectrum. Adding a constant to the FID
before Fourier transformation will create a spike in the spectrum which appears at the carrier fre-
quency.

-38-