Dependence of nuclear spin singlet lifetimes on RF spin-locking power
Stephen J. DeViencea*, Ronald L. Walsworthb,c, Matthew S. Rosenc,d,e
aDepartment of Chemistry and Chemical Biology, Harvard University, 12 Oxford St., Cambridge, MA 02138
bHarvard-Smithsonian Center for Astrophysics, 60 Garden St., Cambridge, MA 02138
cDepartment of Physics, Harvard University, 17 Oxford St., Cambridge, MA 02138
dHarvard Medical School, 25 Shattuck Street, Boston, MA 02115
eMartinos Center for Biomedical Imaging, 149 Thirteenth St., Charlestown, MA 02129
Harvard-Smithsonian Center for Astrophysics,
MS 59, 60 Garden St., Cambridge, MA 02138
We measure the lifetime of long-lived nuclear spin singlet states as a function of the strength of the RF spin-locking
field and present a simple theoretical model that agrees well with our measurements, including the low-RF-power
regime. We also measure the lifetime of a long-lived coherence between singlet and triplet states that does not require
a spin-locking field for preservation. Our results indicate that for many molecules, singlet states can be created using
weak RF spin-locking fields: more than two orders of magnitude lower RF power than in previous studies. Our
findings suggest that in many biomolecules, singlets and related states with enhanced lifetimes might be achievable
in vivo with safe levels of RF power.
The spin-lattice relaxation time T1is a limiting factor for a broad class of NMR experiments in which nuclear spin
polarization or order needs to be preserved or transported [1–4]. However, long-lived nuclear spin singlet states with
lifetimes up to 37 T1 have recently been measured in thermally-polarized samples [5–10]. These long-lived states
have been used to study slow processes such as diffusion, chemical exchange, and conformational dynamics in vitro
Nuclear spin singlet states are typically created from pairs of coupled spin-1/2 nuclei with equal or near-equal
resonance frequencies. The most well-known example is the H2molecule, which exists in two forms: para-H2, with
the singlet eigenstate |S0? = (| ↑↓? − | ↓↑?)/√2; and ortho-H2, with triplet eigenstates |T−? = | ↑↑?, |T0? = (| ↑↓
? + | ↓↑?)/√2, and |T+? = | ↓↓? . In our notation ↑ represents a spin aligned with the applied magnetic field, B0,
while ↓ represents a spin that is anti-aligned. In the singlet state, the net nuclear spin is zero, and there is no net
magnetic dipole moment. Hence interactions with the environment are weak and the rate of interconversion between
singlet and triplet states is very slow, often much slower than the spin-lattice relaxation rate 1/T1. On the other
hand, the triplet states have non-zero magnetic moments and couple strongly with the environment. Relaxation
among the triplet states occurs on the timescale T1.
Recently, Levitt and colleagues demonstrated a general technique for the creation of singlet states for pairs of
magnetically inequivalent nuclei [7, 15]. Such inequivalent nuclear spins cannot form ideal, long-lived singlet states
as outlined above: the different local environments of the nuclear spins leads to rapid conversion to the triplet
arXiv:1201.1482v1 [physics.chem-ph] 6 Jan 2012
state and thus coupling to the environment. Nonetheless, as Levitt et al. showed, a properly designed RF pulse
sequence can prepare a singlet state, which is then preserved from triplet interconversion by the application of a
continuous resonant RF field. This “spin-locking” field forces the average Hamiltonian of the two nuclear spins to
be effectively equivalent. While this technique is applicable for a large variety of molecules, the large continuous
RF power employed in spin-locking experiments to date implies an RF specific absorption rate (SAR) that is likely
prohibitive for animal and human studies [16, 17].
In this paper, we report measurements of singlet state lifetime for a variety of organic molecules and as a function
of RF spin-locking field strength. We find that the measured RF power required to preserve a singlet state agrees
well with the predictions of a simple theoretical model with inputs from the molecule’s NMR spectrum. We also
present measurements of a singlet-triplet coherence with an extended lifetime that does not require the use of RF spin
locking for preservation. Moreover, our findings demonstrate that for many molecules of interest, singlet lifetimes
many times longer than T1can be achieved with much weaker RF spin-locking fields than have been used to date –
more than an order of magnitude smaller than in previous studies – leading to both an RF power and an SAR more
than 100 times lower. This result suggests that in vivo application of long-lived singlet NMR might be possible in
biomolecules with the appropriate properties, despite limitations imposed by RF SAR.
2.1. Singlet Relaxation Mechanisms
Many relaxation mechanisms are forbidden by symmetry from converting the singlet state to triplet states. For
example, magnetic dipole-dipole interactions between the singlet’s two spins cannot couple the antisymmetric singlet
state to the symmetric triplet states. Since this intra-pair interaction is often the dominant driver of relaxation,
the typical result is a singlet population with a lifetime TSmany times longer than the spin-lattice relaxation time
T1. Dipole-dipole interactions between the singlet spin pair and more distant spins can also lead to relaxation, but
the singlet is protected from dipolar fluctuations common to both singlet spins: i.e., in the far-field the net dipole
moment of the singlet is zero. Thus singlet-state relaxation must instead occur through differential interactions on
each spin of the singlet; these include chemical shift anisotropy (CSA), spin rotation (SR) due to collisions between
molecules, and magnetic quadrupole interactions with a third spin [18–21].
Since these three singlet relaxation mechanisms respond differently to the applied static magnetic field and temper-
ature, the dominant effect can be determined experimentally. CSA has a strong magnetic-field dependence with a
lifetime scaling as TCSA∝ 1/B2
as TSR∝ exp(E/kBT) ∝
scaling linearly with temperature in the extreme-narrowing regime (when molecular rotation rates are much greater
than the Larmor frequency). As demonstrated below, for the molecules used in the present study, the quadrupolar
mechanism dominates singlet-state relaxation, i.e., TS≈ TQ.
0. Spin rotation collisions result in a lifetime scaling non-linearly with temperature
kBT. Magnetic quadrupole interactions, on the other hand, produce a singlet lifetime
Magnetic quadrupole relaxation results from the two spins of the singlet interacting differently with a third spin.
This relaxation mechanism was modeled at high magnetic field by Tayler et al. , who derived an expression for
the enhancement of the singlet lifetime∗:
2j− b1jb2j(3cos2φ1j2− 1))
∗Ref.21 contains a typesetting error in which the summation has been taken over the whole expression rather than only the
b Strong scalar coupling
c Strong resonant
a Two weakly-coupled spins
Figure 1: A system containing two spins with resonant transition frequencies ν1and ν2can be represented by four spin-pair eigenstates.
(A) For weak spin coupling, relaxation occurs via magnetic dipole-dipole interactions due to zero-, single-, and double-quantum transitions
(with rates W0, W1, and W2). (B) If the two spins are coupled strongly by scalar coupling (J >> ∆ν, the chemical shift splitting), the
spin-pair eigenstates are mixed into singlet and triplet states. The triplet states still interact via dipole-dipole interactions (solid lines),
but singlet-triplet transitions are forbidden (dashed lines). (C) Mixing of the spin-pair eigenstates can also be induced by a strong RF
spin-locking field (νn > 5∆ν), which produces a singlet state and three mixtures of triplet states. Singlet-triplet transitions are again
forbidden. The new triplet states exhibit new transition rates W?
Here spins 1 and 2 compose the singlet while j represents another nearby spin; bjk = γ2/r3
dipolar coupling strength between spins; and φ1j2is the angle between the vectors connecting 1 with j and 2 with
j. In principle, there is no limit to the singlet lifetime enhancement given the proper molecular geometry. However,
in practice other relaxation mechanisms gain importance if magnetic quadrupole relaxation is highly suppressed. In
a previous study, Equation 1 was found to agree well with measurements of singlet-state lifetimes using high RF
spin-locking power .
jkis a measure of the
2.2. RF-Power Dependence
A detailed theoretical analysis of the singlet’s lifetime during RF irradiation has been given by Pileio and Levitt,
who performed exact numerical calculations for the relationship between singlet lifetime and RF power . Here,
we develop an approximate model that leads to a simple calculation of the measured singlet lifetime at a given
spin-locking frequency. Our model can easily be fit to measurements of singlet lifetime at a number of RF field
strengths so that the maximum singlet lifetime can be extracted.
A pair of spin-1/2 nuclei creates a system with four spin-pair eigenstates as shown in Fig. 1A: | ↑↑?,| ↑↓?,| ↓↑?, and
| ↓↓?. In general, there is mixing between the two eigenstates with no net z-component of spin (| ↑↓? and | ↓↑?) if the
spin states are degenerate, or if scalar coupling between the spins is strong compared with any difference between
the individual spin transition frequencies (ν1and ν2). As a result of such mixing, the spin-pair system is described
by one singlet state, |S0? = (| ↑↓? − | ↓↑?)/√2, and three triplet states |T−? = | ↑↑?, |T0? = (| ↑↓? + | ↓↑?)/√2, and
|T+? = | ↓↓?, with the energy levels shown in Fig. 1B. In this case, spin polarization cannot be transfered to the
singlet state from the triplet states via an RF pulse sequence because HRF|S0? = 0.
For many molecules of interest, however, chemical shifts induce a difference between spin transition frequencies
(∆ν = |ν1− ν2|) that is much larger than the scalar coupling, and hence there is little mixing of the bare spin-
pair eigenstates. In this case unitary transformations, via the RF pulse sequence shown in Fig. 2A, can transfer
initial thermal spin polarization to the singlet state with at most 50% efficiency by creating the singlet-enhanced
90x 180x 90-y RF 90x Acquire
τ1 τ2 τ5
90x 90-y RF 90x Acquire
τ1 τ2 τ3 τ5
Figure 2: (A) The singlet-enhanced superposition state ρST is initialized via a 3-pulse preparation sequence. (B) A long-lived coherence
(LLC) between the singlet and triplet states ρLLC, as well as mixtures of the singlet and LLC states, are initialized via a 2-pulse
preparation sequence. Both preparation sequences are followed by application of the RF spin-locking field and a signal acquisition pulse.
superposition state 
ρST=|T0??T0| − |S0??S0|
=| ↑↓??↓↑ | + | ↓↑??↑↓ |.
Similarly, the RF pulse sequence shown in Fig. 2B can transfer initial thermal spin polarization into a long-lived
coherence between the singlet and triplet states with density matrix
ρLLC=|S0??T0| + |T0??S0|
=| ↑↓??↑↓ | − | ↓↑??↓↑ |.
A similar long-lived coherence has been previously studied as a way to extend T2[22, 23].
For both ρST and ρLLC, population in the singlet state will be rapidly interconverted with the triplet states on a
timescale ∼ 1/∆ν, providing strong coupling to the environment and thus rapid relaxation to the thermal state.
However, a strong on-resonance RF field has been shown to be effective for maintaining the singlet-enhanced super-
position state ρST [7, 15]. When the spin-locking RF field is set to the average resonant transition frequency of the
two spins, the Hamiltonian in the bare spin-pair basis becomes
Here νnis the effective spin nutation frequency due to the RF field, which drives single-quantum spin transitions.
Also, we assume scalar coupling is weak and neglect its contributions to the Hamiltonian.
H = h
Diagonalizing this Hamiltonian yields four spin-locked eigenstates given by
2sinθ(| ↑↑? + | ↓↓?) +1
2(| ↑↓? + | ↓↑?)
√2cosθ(| ↑↑? + | ↓↓?)
√2sinθ(| ↑↓? − | ↓↑?)
|φ−? = −1
2(| ↑↓? + | ↓↑?).
2cosθ(| ↑↓? − | ↓↑?)
√2(| ↑↑? − | ↓↓?) (8)
|φS? = −
2sinθ(| ↑↑? + | ↓↓?) −1
2cosθ(| ↑↓? − | ↓↑?)
The mixing angle θ is controlled by the ratio of the spin nutation frequency to the chemical shift splitting:
θ = arctan2νn
At very large nutation rates (νn>> ∆ν), i.e., high RF spin-locking power, the spin-locked eigenstates simplify to
2(| ↑↓? + | ↓↑? + | ↑↑? + | ↓↓?)
2(|T−? + |T+?)
2(|T−? + |T+?).
√2(| ↑↑? − | ↓↓?) =
√2(|T−? − |T+?) (13)
√2(| ↑↓? − | ↓↑?) = |S0?
2(| ↑↓? + | ↓↑? − | ↑↑? − | ↓↓?)
Note that the spin-locked singlet state |φS? corresponds to |S0? in this limit of large spin nutation (i.e., large RF
spin-locking field), whereas the three spin-locked triplet states are each mixtures of eigenstates |T0?,|T+? , and |T−?.
In this case, the initial state ρST is described well by Equation 2. The singlet, |S0?, is well-protected by the RF
spin-locking field, and after a short initial period during which the triplet states equilibrate, the remaining |S0?
component relaxes exponentially with the characteristic time predicted by Equation 1.
In the high-RF-power regime, the long-lived coherence ρLLCis a sum of coherences containing |φ+?, |φ−?, and |φS?,
which experience decoherence due to both dipole-dipole interactions and inhomogeneities in the RF spin-locking
If instead very small RF spin-locking power is applied (νn<< ∆ν), the singlet component of ρSTrapidly interconverts
with the central triplet state, |T0?. When no RF power is applied, ρSTis a zero-quantum coherence that precesses in
the xy-plane, with a lifetime up to 3.25 T1if inter-pair dipole-dipole interactions are the sole relaxation mechanism
. The addition of a small amount of RF power quickly decreases the lifetime of the ρST coherence because the
RF field efficiently drives single-quantum transitions but creates very little long-lived singlet component.
In the low-RF-power regime, the long-lived coherence ρLLCis well-described by Equation 5 as a population difference
between the two central bare spin-pair eigenstates. The conventional two-spin dipole-dipole relaxation model of
Solomon [25, 26] predicts that in most cases TLLC= 3T1(see supplement S1).
For intermediate RF spin-locking power (νn≈ ∆ν), a more complex analysis is required. For an arbitrary RF power,
the initial state ρST can be represented as
ρST=| ↑↓??↓↑ | + | ↓↑??↑↓ |
2(|φ+??φ+| + |φ+??φ−| + |φ−??φ+| + |φ−??φ−|)
(|φ+??φ−| − |φ+??φ+| + |φ−??φ+| − |φ−??φ−|)
(|φ−??φS| − |φ+??φS| + |φS??φ−| − |φS??φ+|)
where the four spin-locked eigenstates are given by Eq. (7)-(10).
At moderate RF powers (νn > ∆ν), ρST is still mainly composed of the population |φS??φS| and mixed triplet
states. However, the eigenstate |φS? no longer consists solely of the singlet |S0?. It also contains a triplet component
cosθ(|T−? + |T+?)/√2, which interacts with |φ0? via a double-quantum transition, with relaxation rate scaling as
cos2θ. The triplet component also interacts with |φ+? and |φ−? via single-quantum transitions, with relaxation rate
scaling as cos2θ; and via double quantum transitions, with relaxation rate scaling as cos2θsin2θ.
The above scaling of the relaxation of ρST suggests a model for the measured singlet lifetime as a function of RF
1 + (2νn/∆ν)2+
where Tx is the lifetime at low RF power and TS is the maximum singlet lifetime, typically achieved at high RF
power. Significantly, this model predicts that for typical maximum singlet lifetimes, the measured singlet lifetime
reaches 95% of its maximum value when the nutation rate νnis approximately 5 ∆ν.
Relaxation of the long-lived coherence ρLLC can be modeled using a similar analysis. In terms of the spin-locked
eigenstates, we have:
ρLLC=| ↑↓??↑↓ | − | ↓↑??↓↑ |
=cosθ(|φ+??φ+| − |φ−??φ−|)
√2(|φS??φ+| + |φ+??φS| + |φS??φ−| + |φ−??φS|).(19)
At low RF spin-locking powers, the long-lived coherence is mainly composed of |φ+??φ+| − |φ−??φ−|, and these two
eigenstates interact with one another via a zero-quantum transition. However, as the RF power is increased, these
states begin to mix with |T+? and |T−?, which opens up double-quantum transitions with relaxation rates scaling
as sin4θ. A double-quantum transition with |φ0? also becomes available, with relaxation rate scaling as sin2θ. The
latter relaxation rate increases more quickly with RF power and dominates at small θ.
This above scaling suggests a simple model for the ρLLCrelaxation rate:
1 + (2νn/∆ν)2+
where 1/TLLCis the relaxation rate at zero RF power and 1/Tyis the additional relaxation rate due to the applied
We find that our model for the measured singlet lifetime agrees well with the detailed treatment of Pileio and
Levitt (see eq. 43 in ), which contains terms up to eighth power in cosθ. Our model includes only lowest-order
terms, but satisfactorily describes the measured relationship between singlet lifetime and RF power, as described
below. The two models deviate most at low RF powers (νn< 0.5∆ν), where higher-order terms in cosθ make larger
contributions. See supplement S2 for a comparison.
3. Experimental Results
We performed NMR studies at 4.7 T of proton pair singlet states in a number of small organic molecules using a wide
range of RF spin-locking powers. We chose citric acid and p-hydroxybenzoic acid, as Pileio et al. had previously
studied these using high RF power . Aditionally, we studied aspartic acid, trans-1,4-cyclohexanediol, and glycerol
formal as examples of molecules with a range of structures.
Measurements: Coherence Singlet Mixture Model: Coherence Singlet
Figure 3: Measurements of the lifetimes of the singlet state, long-lived coherence (LLC), and a mixture of the two as a function of the
effective RF spin-locking field B1 for proton pairs in small organic molecules. Also shown are fits to models for the singlet and LLC
lifetimes, as described in the main text. (A) Citric acid, Tx= 500 ms and Ty = 600 ms; (B) p-hydroxybenzoic acid Tx= 2 s and Ty = 7 s;
(C) glycerol formal, Tx= 250 ms and Ty = 1 s; RF power is quantified by the induced nutation frequency about the B1field. Molecular
structures are shown, protons of the singlet(s) are circled, and values for chemical shifts (∆ν) and spin-lattice relaxation times (T1) are
Our experimental protocol (Fig. 2) initialized proton pairs into one of three different states: the singlet-enhanced
superposition state ρST, which contained predominantly singlet population after rapid initial decay of the triplet
component; a long-lived coherence between singlet and triplet, ρLLC; or a mixture of the two. In all molecules we
measured the singlet (ρST) lifetime to increase with the applied RF power, reaching a plateau at the maximum singlet
lifetime, TS, when νn> 5∆ν. In contrast, we found that the lifetime of the long-lived coherence decreases from its
maximum value of TLLCwith the application of RF power. Both of these cases are well-modeled by equations 18 and
21 above. When we created a mixture of ρST and ρLLC, the measured lifetime was that of the state with the longest
lifetime at a given RF power, although the measured amplitude was lower as the contribution from the faster-relaxing
state was quickly lost. We individually fit the two regimes of the mixed-state lifetime-vs.-RF-power measurements
with the corresponding models for ρST and ρLLC, which provided a good characterization of the system’s behavior,
as shown in Fig. 3A-C. Note that for each molecule studied, we found that the shortest mixed-state lifetime occurs
near νn≈ 0.5∆ν. Results for maximum singlet and LLC lifetimes (TSand TLLC) are summarized in Table 1.
Table 1: Measured values of spin-lattice, singlet, and long-lived coherence (LLC) relaxation times.
0.58 ± 0.03
p-hydroxybenzoic acid D2O2.9 ± 0.1
p-hydroxybenzoic acid H2O2.3 ± 0.1
aspartic acid0.83 ± 0.03
trans-1,4-cyclohexanediol1.34 ± 0.02
glycerol formal0.68 ± 0.01
1.5 ± 0.1
7.3 ± 0.7
3.9 ± 0.1
2.3 ± 0.4
2.90 ± 0.01
1.37 ± 0.02
2.6 ± 0.2
2.5 ± 0.3
1.7 ± 0.1
2.8 ± 0.5
2.16 ± 0.03
2.01 ± 0.04
4.5 ± 0.3
16 ± 2
5.8 ± 0.2
7.48 ± 0.3
3.9 ± 0.9
1.91 ± 0.03
7.8 ± 0.7
6.2 ± 0.8
2.5 ± 0.1
9.0 ± 0.5
2.9 ± 0.7
2.81 ± 0.06
We also investigated possible mechanisms for proton-pair singlet relaxation. First, we compared our measurements
with those at other magnetic fields to probe the importance of chemical-shift anisotropy (CSA) relaxation (see Table
1). Our result of TS= 4.5 s and 7.8-fold lifetime enhancement over T1for citric acid at 4.7 T are consistent with a
previous measurement of 4.81 s and 7.6 T1at 9.4 Tesla . Due to hardware limitations, there was insufficient RF
power to reach the maximum singlet lifetime for p-hydroxybenzoic acid. Nevertheless, a fit to data at finite RF power
gave TS = 16 s and 5.5 T1, which is consistent with previous measurements . Since these singlet-state lifetimes
do not significantly depend on magnetic field strength, we conclude that CSA is not a primary mechanism of singlet
relaxation. To distinguish between the spin-rotation and magnetic quadrupole relaxation mechanisms, we performed
singlet lifetime measurements at several sample temperatures. We found that the singlet lifetime increases linearly
with temperature, which identifies the magnetic quadrupole relaxation mechanism as dominant in such proton-pair
singlet state molecules (see supplement S3).
Note that most previous proton-pair singlet measurements were conducted using deuterated solvents, which should
result in weaker singlet-solvent interactions and larger enhancements of singlet state lifetime. To test whether such
lifetime enhancement changed in a normally protonated solvent, we studied p-hydroxybenzoic acid in both D2O and
H2O. We found that the enhancement of both the singlet and LLC lifetimes were significantly lower in H2O (see
Table 1). The enhancement is likely higher in D2O due to the substitution of deuterium for the phenolic proton as
well as reduced dipolar interactions with nearby solvent protons.
The above experimental results and associated modeling establish an operational spin-locking condition νopt
to realize maximum singlet lifetime with minimal RF power. In the context of this operational condition, we can
reassess the past work by Levitt and colleagues using high-power RF spin-locking fields [5, 7, 15]. As shown in Table
2, most of the previous experiments employed νn>> 5∆ν: i.e., they used much higher RF power than was needed to
achieve a long singlet-state lifetime. For example, for citric acid Pileio et al.  used νn= 3.5 kHz, whereas ∆ν = 72
Hz at 9.4 T, which is an order-of-magnitude higher spin-locking field than necessary.
Furthermore, we note that similarly low RF powers will be required for practical in vivo singlet-state creation in
a wide variety of molecules using clinical MRI scanners, where the static magnetic field is commonly between 1.2
and 7 T; see example values for νopt
at 1.5 T given in Table 2. For example, at 4.7 T glycerol formal’s protons
have a frequency difference ∆ν ≈ 16 Hz; hence νopt
enhancement. For common biomolecules such as citric acid and aspartic acid, νopt
within the spin-locking regime commonly used in clinical MRI [27, 28]. The spin-locking times and strengths used
in [27, 28] imply that a 60 Hz spin-lock could be safely applied for 3.5 s, and a 20 Hz spin-lock for 30 s. These
timescales are of the same order as the singlet lifetimes we measured for typical small molecules, and thus should be
≈ 80 Hz is sufficient to achieve significant singlet-state lifetime
< 100 Hz at 1.5 T, which is well
sufficient to conduct a variety of in vivo measurements using singlets. Alternatively, the long-lived coherence can be
utilized without the need for any spin-locking if only moderate lifetime enhancements are required.
In summary, our measurements and theoretical description show that for many molecules long-lived nuclear-spin
singlet states and singlet/triplet coherences can be created using RF spin-locking powers that are more than two
orders of magnitude lower than in previous studies; and that the effectiveness of the spin-locking can be accurately
predicted from spectral parameters. These insights will be useful in the development of new applications for singlet
states in vivo, where the RF specific-absorption rate (SAR) must be minimized.
Table 2: Comparison of the chemical shift; optimal spin nutation frequency for RF spin-locking, νopt
previous experiments. Also listed are values for νopt
citric acid 72 Hz (9.4 T)
p-hydroxybenzoic acid 445 Hz (11.75 T)
aspartic acid 100 Hz (11.75 T)
≈ 5∆ν; and values of νn used in
at 1.5 T
for a clinical MRI scanner.
νnused in previous experiments
5. Materials and Methods
Solutions of citric acid, aspartic acid, p-hydroxybenzoic acid, and 1,4-cyclohexanediol were made in D2O, with the
addition of sodium hydroxide where necessary for dissolution. Glycerol formal was analyzed neat. Concentrations
and conditions can be found in Table 3. All reagents were purchased from Sigma-Aldrich . All samples were prepared
in 10 mm diameter NMR sample tubes and bubbled with nitrogen gas for three minutes. Spectra were acquired on
a 200 MHz Bruker spectrometer without spinning.
Experiments shown in Fig. 2 were run on each compound using varying lengths for τ4. Pulse sequence parameters can
be found in Table 4. To remove any remaining triplet polarization, phase cycling was used in which the experiment
was repeated with both the first and last 90◦pulses along -x rather than x. Between 8 and 32 averages were used
to provide sufficient signal-to-noise. The intensity of each peak was then measured and plotted against τ4. The
resulting data was fit with a single exponential time decay. Further details of the experimental pulse sequences are
discussed in Supplement S4. Multiple datasets were collected using different RF power levels for spin-locking. The
RF power was characterized by measuring the nutation rate induced by the RF B1field, which was calibrated using
a sequence of single-pulse experiments performed with increasing pulse length. T1relaxation rates were measured
through conventional inversion-recovery experiments.
For variable temperature experiments, the temperature was controlled by supplying hot air to the probehead. Blown
air was heated with a Hotwatt cartridge heater controlled by an Omron temperature control box, and the temperature
of the sample was monitored with an RTD in the probehead.
Table 3: Sample preparations for the study of long-lived states.
1,4-cyclohexanediol (cis/trans mixture)
Table 4: Delays, in ms, for pulse sequences used in the experiments: τ1− τ2− τ3− τ4− τ5
p-hydroxybenzoic acid 29.3-31.5-1.14-τ4-1.14
1,4-cyclohexanediol (cis/trans mixture)
 K. Golman, R. in’t Zandt, M. Lerche, R. Pehrson, J. H. Ardenkjaer-Larsen, Cancer Research 66 (2006) 10855–
 K. Golman, J. S. Petersson, Academic Radiology 13 (2006) 932–942.
 H. Yu, Proc Natl Acad Sci USA 96 (1999) 332–334.
 P. T. Callaghan, Principles of Nuclear Magnetic Resonance Microscopy, Oxford University Press, Oxford, Eng.,
 G. Pileio, M. Concistr´ e, M. Carravetta, M. H. Levitt, Journal of Magnetic Resonance 182 (2006) 353–357.
 G. Pileio, M. H. Levitt, Journal of Magnetic Resonance 187 (2007) 141–145.
 M. Carravetta, M. H. Levitt, Journal of the American Chemical Society 126 (2004) 6228–6229.
 P. R. Vasos, A. Comment, R. Sarkara, P. Ahuja, S. Jannin, J.-P. Ansermet, J. A. Konter, P. Hautle, B. van den
Brandt, G. Bodenhausen, Proceedings of the National Academy of Sciences USA 106 (2009) 18469–18473.
 P. Ahuja, R. Sarkar, P. R. Vasos, G. Bodenhausen, ChemPhysChem 10 (2009).
 W. S. Warren, E. Jenista, R. T. Branca, X. Chen, Science 323 (2009) 1711–1714.
 R. Sarkar, P. Ahuja, P. R. Vasos, G. Bodenhausen, ChemPhysChem 9 (2008) 2414–2419.
 S. Cavadini, J. Dittmer, S. Antonijevic, G. Bodenhausen, Journal of the American Chemical Society 127 (2005)
 R. Sarkar, P. R. Vasos, G. Bodenhausen, Journal of the American Chemical Society 129 (2006) 328–334.
 A. Farkas, Orthohydrogen, Parahydrogen, and Heavy Hydrogen, Cambridge University Press, Cambridge, Eng.,
 M. H. Levitt, Encyclopedia of Magnetic Resonance (2010).
 P. A. Bottomley, R. W. Redington, W. A. Edelstein, J. F. Schenck, Magnetic Resonance in Medicine 2 (1985)
 International standard, medical equipment–part 2: particular requirements for the safety of magnetic resonance
equipment for medical diagnosis, 2nd revision, Geneva: International Electrotechnical Commission, 2002.
 G. Pileio, M. H. Levitt, Journal of Chemical Physics 130 (2009) 214501.
 E. Vinogradov, A. K. Grant, Journal of Magnetic Resonance 194 (2008) 46–57.
 E. Vinogradov, A. K. Grant, Journal of Magnetic Resonance 188 (2007) 176–182.
 M. C. D. Tayler, S. Marie, A. Ganesan, M. H. Levitt, Journal of the American Chemical Society 132 (2010)
 R. Sarkar, P. Ahuja, P. R. Vasos, G. Bodenhausen, Physical Review Letters 104 (2010) 053001.
 A. Bornet, R. Sarkar, G. Bodenhausen, Journal of Magnetic Resonance 206 (2010) 154–156.
 R. R. Ernst, G. Bodenhausen, A. Wokaun, Principles of Nuclear Magnetic Resonance in One and Two Dimen-
sions, Oxford University Press, Oxford, Eng., 1987.
 I. Solomon, Physical Review 99 (1955) 559–565.
 C. P. Slichter, Principles of Magnetic Resonance, Springer-Verlag, Berlin, 1990.
 A. J. Wheaton, A. Borthakur, J. B. Kneeland, R. R. Regatte, S. V. S. Akella, R. Reddy, Magnetic Resonance
in Medicine 52 (2004) 1453–1458.
 S. Sammet, M. Bock, H.-P. Schlemmer, P. Bachert, Proceedings of the International Society for Magnetic
Resonance in Medicine 8 (2000) 1097.