Hyperpolarization Methods for MRS

This article covers the fundamental principles and practice of NMR hyperpolarization techniques, which are proving useful for in vivo magnetic resonance spectroscopy (MRS) studies of metabolism in animal models, and clinical trials with hyper-enhanced sensitivity. Fundamentally, hyperpolarization methods enhance nuclear spin polarization by orders-of-magnitude, resulting in concomitant improvement in NMR detection sensitivity. The hyperpolarization methods described here – dynamic nuclear polarization (DNP), para-hydrogen induced polarization (PHIP), signal ampliﬁcation by reversible exchange (SABRE), and spin-exchange optical pumping (SEOP) – are capable of achieving nuclear spin polarization approaching the theoretical maximum of unity on nuclear spin sites of molecular or atomic agents suitable for in vivo administration. Importantly, hyperpolarization is inherently nonequilibrium in nature: the duration of the hyperpolarization is frequently short-lived, often being limited by the in vivo spin–lattice relaxation times (T 1 ) that are on the order of seconds to a minute. Nevertheless, sufﬁcient amounts of nuclear spin polarization can survive the process of preparation, administration, and in vivo circulation to provide extraordinary enhancement of the hyperpolarized agent. The chemical shift dispersion of these agents at the molecular location of interest reports on functional, metabolic, and other processes at the molecular level, enabling true molecular MRS imaging.


Introduction
The sensitivity and signal-to-noise ratio (SNR) of the NMR experiment is directly proportional to the nuclear spin polarization factor P, which represents the fraction of the alignment of the nuclear spin ensemble with an applied magnetic field B 0 . In conventional NMR, P ≈ γ B 0 /2kT under thermal-equilibrium conditions governed by the Boltzmann distribution (with γ = the gyro magnetic ratio, = Planck's constant/2π, k = Boltzmann's constant, and T = absolute temperature). In vivo conditions imply temperatures in excess of 300 K. As a result, even if a high-field magnet -e.g., a 3-T clinical magnetic resonance (MR) imaging (MRI) scanner -is employed for MR spectroscopy (MRS), the value of P of proton ( 1 H) and carbon-13 ( 13 C) nuclear spins is only 1 × 10 −5 and 2.5 × 10 −6 , respectively. Therefore, the vast majority of the nuclear spin ensemble does not in fact contribute to the NMR signal. For this reason, NMR in general and MRS, in particular, are frequently considered insensitive probes.
However, P can be artificially increased significantly relative to the thermal-equilibrium level, via a process called NMR hyperpolarization. Several hyperpolarization techniques realize enhancements of P by several orders-of-magnitude over the be confronted during HCA production. First, HCAs are nearly always prepared exogenously in a hyperpolarization apparatus, frequently requiring sophisticated and expensive hardware. Second, HCAs must retain a HP state sufficiently long for quality assurance, transport, administration, in vivo delivery, and NMR/MRS measurements. These requirements limit the number of molecular sites (i.e., moieties of 13 C, 15 N, 129 Xe) suitable for HP MRS, to those whose spin-lattice relaxation times (T 1 ) are long enough to accomplish all this. Finally, the decay of the HP state to thermal equilibrium is irrecoverable. Therefore, signal acquisition must be fast and efficient to account for polarization losses via T 1 relaxation, RF pulsing (for detection), and the effects of in vivo metabolism and dilution. 3 This article describes four hyperpolarization techniques: (1) dynamic nuclear polarization (DNP), (2) para-hydrogen induced polarization (PHIP), (3) signal amplification by reversible exchange (SABRE), and (4) spin-exchange optical pumping (SEOP) of 129 Xe. These techniques have already proven effective for preparing HCAs that can be used for tracking metabolic and functional processes in vivo (see Hyperpolarized Carbon-13 MRI and MRS Studies 3 ), because they are capable of achieving P values approaching the theoretical maximum of unity on nuclear spin sites of molecular or atomic HCAs. Moreover, the HCAs for MRS fundamentally offer a wide dynamic range of chemical shift dispersions either through metabolism (e.g., 13 C metabolites) or by their sensitivity to the local environment (e.g., 129 Xe). Many HCAs have been validated in animal models, 2 while some HCAs (e.g., HP 13 C-pyruvate and 129 Xe) are being tested in clinical trials. 4 The primary focus in this article is on the fundamentals of these hyperpolarization techniques (i.e., how they are generated and their physical and chemical properties), their existing or emerging MRS relevance, and a description of the hyperpolarization equipment (frequently referred to as hyperpolarizers) required to perform these techniques. Accordingly, the article describes and cites only selected works in these areas, and the reader may additionally benefit from some recent comprehensive reviews. 2,5 -9

Fundamentals
Despite the demonstration by Overhauser in the early 1950s of the underlying basis of DNP -that P could be enhanced by cross-relaxation with nearby electrons owing to differences in their gyromagnetic ratios (γ e /γ I ∼660) -only recently has interest in its potential applications for biomedical MRI and MRS emerged. 10 While various methods fall under the DNP umbrella, the 'dissolution DNP' (d-DNP) method is the most commonly used method for in vivo applications. 1,9 In biological d-DNP MRS experiments, the general preparation procedure involves (i) mixing the MR-active species of interest with a paramagnetic source of free electrons; (ii) placing the mixture in a high magnetic field at low cryogenic temperatures; (iii) subsequent irradiation with a microwave source resonant on the electron spin resonance (ESR) frequency to mediate the transfer of electron polarization to nearby nuclei 11 followed by (iv) rapid sample thawing using a hot solvent to warm the frozen polarized sample and transferring it to the MR scanner for in vivo injection as a liquid. These conditions generate electron spin polarizations that can approach unity as shown in Figure 1, with a significant fraction of the prepared P largely surviving the rapid dissolution in step (iv).

Dissolution DNP (d-DNP) Instrumentation
Both commercial and home-built d-DNP devices have been extensively used for in vivo MRS. While each polarizer design may have unique characteristics, they are all comprised of the same basic components: a superconducting magnet that provides a B 0 of 3-7 T, a liquid helium cryostat for sample cooling to 1-4 K, a microwave source to transfer polarization from the electrons to nuclei (∼100 mW), and a mechanism that rapidly dissolves the frozen sample pellet and delivers the solution for collection. 9 Step (i) of the typical d-DNP experiment proceeds as follows: the species of interest is mixed with a radical source and a glassing agent in a typically flat sample container (up to a few hundred μL) and lowered into the cryostat, as depicted in Figure 2. There in step (ii), the sample is frozen into an amorphous solid at low temperatures either by submersion into a liquid helium bath (where vacuum-pumping reduces the vapor pressure, and hence bath temperature) or by exposure to a stream of continuously flowing liquid helium. The microwave source, whose frequency is set to the ESR line of the chosen radical species at the B 0 of the polarizer (typically within ∼80-140 GHz, with a ∼0.5 GHz sweep range), irradiates the sample for ∼1 h in step (iii), using an antenna that takes advantage of the large flat surface area of the sample container. An on-board NMR system can check the HP signal while sweeping the microwave source to allow for optimization of the microwave frequency and excitation power ( Figure 1c) and for monitoring the buildup of hyperpolarization over time ( Figure 1d).
Step (iv) of the procedure requires a superheated solvent with high heat capacity, which is typically satisfied by pH-buffered saline. This is injected into the sample cup inside the cryostat. The lifetime of the HP signal is determined by the T 1 of the HCA, so to mitigate P losses, the rapid dissolution is performed at high B 0 to minimize the sudden depolarizing effects of the paramagnetic centers as the temperature rises above ∼4 K. Dissolution also reduces the relative concentration of radicals and lowers their contribution to the nuclear spin-relaxation rate 9 as the sample volume expands by an order of magnitude (this process also dilutes the HP agent). The room-temperature (RT) T 1 values of most HCAs used for biomedical MRS are ≤1 min but B 0 changes between the polarizer and MR scanner may further reduce T 1 . Rapid shuttling of the sample from the polarizer to the scanner is key and is typically performed pneumatically using helium gas through small diameter tubing. There, the polarized solution is quickly filtered to remove persistent radicals using anion-exchange or polyethylene filters 9 and injected into the subject (Figure 2).
A commercially available d-DNP device for clinical use has recently been developed (SPINlab; GE Healthcare) that operates at 5 T and 1.1 K. 12 The system uses an on-board helium recycling system that dramatically lowers the consumption of liquid helium, which is expensive. While many other features of its design and operation are similar to that described above, additional quality-control steps are incorporated to ensure that the injectable solution meets clinical criteria for pH, temperature, P value, sterility, purity, and residual radical species concentration. In general, the larger sample volumes required for human use necessitate longer dissolution, transport, quality control, and injection periods, such that significant time (e.g., >65 s) may elapse between the initial dissolution and final injection. Nevertheless typical 13 C P values are ∼15-20% after quality control, which is certainly sufficient to perform HP MRI/MRS in human patients with great advantage vs conventional MRS with P ∼10 −3 %. 4 Common HP Agents Produced with d-DNP 13 C-labeled Small Biomolecules. The overall P, T 1 , and biological relevance of the agent are the main criteria for choosing a HP species for MRS applications. Further considerations include the ability to form a glassy matrix in a sufficiently high concentration to mitigate dilution effects from the dissolution step, sufficient chemical shift range to differentiate the primary polarized species from any downstream products, the cost of isotopic enrichment, the ability to lessen the depolarization effects of nearby protons via deuteration, 13 and toxicity. While a number of MR-active nuclei are eligible for DNP, the most commonly pursued are 13 C-labeled carboxyl sites of metabolites. Of the latter, 13 C-pyruvate has garnered the most attention because of its relatively high P values (>50 %), long T 1 (∼60 s), and ability to probe the tricarboxylic acid (TCA) cycle and glycolysis within a T 1 -relevant timescale. 14 Figure 2. Overview of the d-DNP process. Within the high magnetic field of the DNP polarizer, high-power microwaves excite free electrons in free radical species mixed with the HP substrate and frozen into an amorphous solid. The electrons undergo spin-exchange with nearby 13 C nuclei and over time, the entire sample becomes polarized via direct dipole-dipole interactions as well as nuclear spin diffusion involving the 1 H nuclei present in the HP substrate or glassing agent. After polarization, the HCA sample is rapidly dissolved in hot solvent and shuttled into the MR scanner room for injection into the animal or human subject gradient strengths for spatial localization with typical scanner hardware. Other species that can be HP by DNP and may have potential for future in vivo applications include 1 H, 6 Li, 29 Si, 31 P, 89 Y, 107 Ag, and 129 Xe.
Secondary and Co-polarization. The enhanced signal from isotopically labeled HP small molecules may be translated into their chemical reaction products. This process, referred to as secondary polarization, allows the study of biologically relevant molecules that may not polarize well on their own, but have metabolic predecessors that do. This process can either be enzymatically regulated, such as with [1,1-13 C 2 ]acetic anhydride breaking into different [1-13 C]N-acetylated amino acids, 13 or nonenzymatically regulated -for instance, the reaction of 1,2-13 C-pyruvate with H 2 O 2 to form 13 CO 2 , 13 C-acetate, and 13 C-bicarbonate. 15 Co-polarization of a single 'batch' of different small molecules using d-DNP during agent production allows the simultaneous study of different biophysical parameters and processes such as pH, metabolism, necrosis, and perfusion. 13 While the commingled molecules' solid-state polarization P and T 1 values are often similar to their individual values, drawbacks include lower concentrations of, and potential interactions among, the small molecules involved.

Radicals and Glassing Agents
Choice of Radical Species. The buildup time and hyperpolarization level for all species are highly dependent on the choice of radical used to mediate the DNP. Typical radical concentrations range from 15 to 60 mM and may persist post-thawing, requiring filtering before administration. The most widely used sources of free electrons for 13 C DNP have been the trityl and 1,3-bisdiphenylene-2-phenylallyl (BDPA) radicals. Because their ESR lines are narrower than the 1 H resonance frequency dispersion, the probability of cross-polarization with 1 H is lowered, which increases the chances of polarizing nearby 13 C nuclei. Further improvements in the P level and reductions in buildup time can be gained by the addition of a Gd 3+ complex to the trityl radical solutions. The highly efficient electron-1 H cross-polarization of nitroxyl radicals (e.g., TEMPO; 2,2,6,6tetramethylpiperidine-1-oxyl) can also be used in conjunction with a 1 H-13 C cross-polarization scheme to decrease 13 C polarization time, albeit with lower overall 13 C polarization compared to trityl or BDPA radicals. Liquid-state P can be further increased using a deuterated solvent for the nitroxyl radicals. 9 Additional gains have been reported using biradicals comprised of two TEMPO radicals tethered by a polyethylene chain. 16 Choice of Glassing Agent. The DNP process is most efficient when the HP molecule is homogenously mixed with the radical species and formed into a glass, as already noted. This amorphous state allows the paramagnetic centers to achieve close contact with nearby nuclei, which is not the case in a crystalline lattice. Typical glassing agents include glycerol and dimethyl sulfoxide (DMSO), which act to prevent crystallization. Some HP molecules such as neat 13 C-pyruvate are capable of self-glassing, 17 which allows for a higher concentration of 13 C nuclei that speeds 13 C spin diffusion resulting in a more homogenously polarized sample. 18

DNP Advantages, Limitations, and Future Prospects
As a hyperpolarization method, DNP's versatility lies in its ability to hyperpolarize a variety of biologically relevant molecules not easily polarized by other methods. Target molecules can achieve high P with sufficiently long T 1 values for probing metabolic pathways, and clinical trials are already underway for HP 13 C-pyruvate. The main limitations thus far are the high cost of purchase, maintenance, and infrastructure required for the polarizers. Improvements in cryogen recovery and solid-state cooling can lower operational costs, and further optimization of the physical and chemical environment of the HP species should improve P and T 1 values, as would the ability to store the enhanced polarization in a singlet state until it is ready for MRS applications. Finally, the demonstration of a clinically important application of HP MRI -such as using 13 C-pyruvate to study prostate cancer or heart dysfunction -will greatly hasten the development of all d-DNP aspects. 9

para-Hydrogen (para-H 2 ) Production
In addition to its more recent applications in HP MRS, the properties of para-hydrogen (para-H 2 ) have proved important for hydrogen's use as a rocket fuel by minimizing boil-off due to state conversion, and even earlier, para-H 2 served as a prototypical system during the development of quantum theory.
The dihydrogen molecule possesses two nuclear spin isomers, denoted as 'ortho-' and 'para-' states, comprised of a nuclear triplet state and a singlet state, respectively. Importantly, while representing the simplest nuclear singlet state, para-H 2 , per se, has net spin I = 0 and is therefore NMR invisible. As will be seen in the following sections, its use in NMR/MRS is as an extremely powerful vector for transferring HP to HCAs of interest. For further discussion of the physics of para-hydrogen, we direct the interested reader to the books by Farkas on hydrogen 19 and Linus Pauling on quantum mechanics, 20 and the review by Green et al. 5 Inter-conversion of ortho-hydrogen (ortho-H 2 ) to para-H 2 (with a heat of conversion of 670 J g −1 ) is forbidden according to the selection rules of quantum mechanics. It thus occurs very slowly. Spontaneous emission of radiation was calculated by Wigner to result in a conversion rate of ∼10 −10 s −1 , or approximately one transition per 1 gram every 300 years. 19 In 1930, Hall and Oppenheimer calculated that molecular collisions at atmospheric pressure result in a rate of one per ∼10 8 s/1 g for the ortho-para transition half lifetime, or about 3 years. 19 While such a slow inter-conversion rate is nominally unfavorable to para-H 2 production, Bonhoeffer and Harteck discovered in 1929 that paramagnetic catalysts, including, for example, activated charcoal, nickel, and hydrated iron-III oxide, greatly accelerate the establishment of thermodynamic equilibrium for the ratio of ortho-H 2 /para-H 2 .
The instrumentation used to generate high-purity para-state dihydrogen is commonly called a 'para-hydrogen generator', shown in Figure 3. The essential purpose of the generator is to pass bulk H 2 at RT (i.e., ∼293 K) over a paramagnetic catalyst at a cryogenic temperature of ∼20 K to rapidly equilibrate the 75% ortho-H 2 fraction to the lower-temperature Boltzmann thermal equilibrium, thereby creating para-H 2 with nearly 100% para-fraction. Typically, the gas flows through a catalyst chamber attached to a two-stage, Joule-Thomson effect, helium cryocooler cold-head. The flow is regulated with either (i) flow restrictors such as mass flow controllers, snubbers, or needle valves 21,22 or (ii) by releasing small batches of compressed H 2 between automatically sequenced solenoid valves. 23 After separation of the para-H 2 from the catalyst and subsequent storage at RT, quantum-mechanical symmetry selection rules forbid reestablishment of the RT equilibrium. Provided the para-H 2 storage vessel contains no paramagnetic impurities (as is the case for those glass-lined or of aluminum, although the absence of O 2 is also important), the relaxation rate of the para-hydrogen back to the RT equilibrium can be on the order of months. 21 -23 Thus, para-H 2 experiments are possible long after production.

Fundamentals
While the para-H 2 molecule is itself NMR invisible, Bowers and Weitekamp 24 have demonstrated that the para-H 2 singlet can be unlocked via the chemical reaction of hydrogenation provided that the para-H 2 is added in a pairwise manner (best exemplified in the molecular mechanism of H 2 addition) and that the nascent protons in the resulting hydrogenation product  are no longer magnetically equivalent. This phenomenon of symmetry breaking the para-H 2 nuclear singlet state via hydrogenative processes is termed PHIP. 25 While the para-H 2 singlet can be preserved for weeks and months, once the symmetry is broken and the para-H 2 pair is incorporated into the molecular framework of a product molecule (Figure 4), the relaxation decay processes (via T 1 and the spin-spin-relaxation time, T 2 ) are significantly more efficient (on a timescale of seconds) -leading to rapid depolarization. For the majority of metabolically relevant injectable contrast agents that can be envisioned, fast T 1 relaxation limits any meaningful preparation of HCA for biomedical use.
This initial fundamental PHIP challenge was overcome in the early 2000s, when an adjacent 13 C carboxyl (and potentially 15 N or others) with significantly longer T 1 was introduced by Golman et al. 26 and Goldman and Johannesson. 27 In addition, deuteration of the PHIP molecular precursor can extend the para-H 2 state symmetry breaking through chemical reaction with pairwise addition followed by polarization transfer to 13 C produces a HP contrast agent comprised of a 13 C HP carboxyl moiety with a long T 1 lifetime T 1 lifetime of the nascent protons, which has proven especially useful in PHIP of gases. 28 In the PHIP process, polarization from nascent para-H 2 protons is efficiently transferred to the 13 C nucleus via the network of J-couplings using either magnetic field cycling or RF pulse-based methods. Regardless of the approach used, the chemical reaction between para-H 2 and the unsaturated substrate molecule must be performed on a relatively fast timescale (seconds), i.e., significantly shorter than the effective relaxation times of the nascent para-H 2 protons, in order to minimize polarization losses during the reaction. Post-reaction, the polarization is typically transferred to the 13 C nucleus within a second. Moreover, even in the case of RF-based polarization transfer, which has proven to be more efficient as gauged by the %P delivered to 13 C, to date, the PHIP method is relatively low cost because the PHIP process can be performed at very low magnetic fields (a few mT) using relatively simple NMR hardware.

Molecular Precursors and Contrast Agents
The design of PHIP precursor molecules for biomedical applications and MRS is a challenging process that must meet critical requirements. In particular, the molecule must possess an unsaturated (typically C=C) bond adjacent to 13 C to provide a J-coupling contact between the nascent para-H 2 protons and the 13 C to enable efficient polarization transfer. Furthermore, the efficiency of this polarization transfer in RF pulse-based polarization transfer methods can greatly benefit from deuteration of the precursor molecule 29 as a means to keep the nuclear spin system effectively constrained to just the three involved spins: the two nascent para-hydrogens and the 13 C nucleus. 27 Lastly, because the hydrogenation reaction may be incomplete, both the precursor and product molecules should have relatively low in vivo toxicity. Meeting all these requirements represents the main disadvantage of PHIP, and only a limited number of potentially amenable precursor molecules and HCAs have qualified for in vivo use, so far. Those that have been successfully HP and used for in vivo MRS include HP 13 C-succinate, which highlights metabolites of the TCA cycle 30 ; 13 C-tetrafluoropropyl propionate, for sensing lipids in atherosclerotic plaques; 31 and 13 C-phospholactate, 29 for preparation of HP 13 C-lactate, which has been shown to be useful for in vivo metabolic imaging of elevated glycolysis in tumors.

Homo-and Heterogeneous PHIP Catalysis
Homogeneous Catalysis in Organic Solvents. PHIP 24,25 has attracted the attention of those working in catalysis owing to its promise to increase the NMR sensitivity of key lowconcentration species. The PHIP process of pairwise para-H 2 addition requires an efficient catalyst to complete the process within seconds (i.e., on the timescale of relaxation processes). Initially, PHIP catalytic systems were based on Wilkinson's catalyst RhCl(PPh 3 ) 3 , 24 Vaska's complex Ir(CO)Cl(PPh 3 ) 2 , and Crabtree's catalyst [(COD)Ir(PCy 3 )(Py)] + PF 6 − , (COD = 1,4-cyclooctadiene, PCy 3 = tricyclohexylphosphine, Py = pyridine). These transition metal-based systems catalyze the molecular hydrogenation reaction, where para-H 2 is added to an unsaturated precursor via a molecular mechanism. Undergoing a molecular rather than a radical mechanism is essential for preserving the para-H 2 spin order, which allows the observation of a PHIP effect. 5,24 A number of other transition metals and chelating ligands have been evaluated. 5 In particular, implementation of bidentate ligands such as Ph 2 P(CH 2 ) 4 PPh 2 using Rh(I) as a catalyst has proven to be particularly effective for the ultrafast molecular addition of para-H 2 to double and triple bonds, resulting in polarization enhancements of ∼10 4 upon transfer to adjacent 13 C nuclei. 26 Despite the apparent synthetic success of this methodology, the approach has been limited to organic solvents, thus limiting biological applications where aqueous catalysts and substrates are desired.
Homogeneous Aqueous Catalysis. Chemical modification of phosphine-based ligands has permitted preparation of efficient water-soluble Rh(I)-based catalytic systems. By far, the most common variation utilizes 4-bis[(phenyl-3-propanesulfonate) phosphine] butane di-sodium or its close variant, 22 where Rh(I) is chelated by norbornadiene and two phosphines connected by a four-or three-carbon bridge. 5 While aqueous homogeneous catalysts have opened the door to the first generation of mammalian in vivo imaging aided by PHIP, 30,31 the presence of Rh(I) catalysts in the HCA solution is presently an obstacle to clinical applications. Replacement of homogeneous PHIP by heterogeneous PHIP (HET-PHIP) processes will likely be the key to PHIP clinical applications. HET-PHIP catalysis was extensively reviewed a few years ago, 32 and is briefly discussed here in the context of recent advances and applications for aqueous media.

Heterogeneous (Solid/Liquid) Catalysis.
A number of immobilized Rh complexes, based on either Wilkinson's catalyst or closely related systems based on Rh(I) derived from [Rh(COD) 2 ] + [BF 4 ] − , have been evaluated for PHIP, typically using diphenylphosphinoethyl-modified silica as a stationary phase. While some PHIP signal was often observed, the enhancement values were generally low due to slow hydrogenation rates, loss of para-H 2 polarization on the stationary phase surfaces, and chemical degradation of the catalyst. Alternatively, the application of transition metal nanoparticles is well established for gas-solid PHIP (see below) and has been demonstrated for capped platinum nanoparticles 33 in liquid-solid PHIP, where HP liquids are generated over solid-phase catalysts. Finally, Rh/TiO 2 nanoparticles have also been used for PHIP processes, and PHIP hyperpolarization of acrylamide and allyl methyl in D 2 O solutions was demonstrated using these Rh/TiO 2 nanoparticles. This approach is promising for in vivo applications in spite of the low P levels demonstrated to date.

Heterogeneous (Solid/Gas) Catalysis. A silica-embedded
Wilkinson's catalyst was also the first catalytic system enabling PHIP of gases at polarization levels sufficient for in vitro imaging. 34 Further progress in this area has been driven by the development of more robust nanoparticle catalysts such as Pt, Pd, or Rh on TiO 2 , SiO 2 , Al 2 O 3 , or ZrO 2 solid substrates. In general, better polarization transfer results were obtained with smaller nanoparticles: Rh and Pt provide superior results to Pd, while Al 2 O 3 and especially TiO 2 were superior performing substrates. 32

Hyperpolarization Instrumentation
The controlled reaction of para-H 2 with an unsaturated chemical precursor to produce a HCA takes place within an instrument referred to as a 'PHIP hyperpolarizer'. The features and underlying polarization transfer technique differentiate the hyperpolarizers constructed to date. 26,35,36 In general, inert gases such as N 2 shuttle the precursor through a manifold during injection into a heated reactor wherein it is rapidly and thoroughly mixed with a para-H 2 atmosphere, as shown in Figure 5. Before injection, the precursor is typically preheated to further speed the chemical reaction of pairwise addition. Two approaches to precursor/para-H 2 mixing have involved: (i) jet entrainment mixing 22 and, more commonly, (ii) spray mixing ( Figure 5). 26,35,36 Other differences in hyperpolarizer designs concern the method used to transfer the spin order of para-H 2 after chemical synthesis to the heteronucleus, which include (i) field cycling 26 or (ii) more efficiently, 26,35,36 the application of RF pulses. 27 In the latter method, RF pulses may be synchronized with the operation of the manifold by a LabView interface to a National Instruments data acquisition card 26,35,36 or by a low-field NMR spectrometer. 36 In order to facilitate calibration of the RF pulse sequence parameters, in situ NMR detection capability for the hyperpolarizer is desirable. 36 The latter can result in a stand-alone PHIP hyperpolarizer with built-in quality assurance for the P produced.

Signal Amplification by Reversible Exchange (SABRE)
Another route to achieve hyperpolarization in solution is SABRE, introduced by Duckett and coworkers in 2009. 37 Similar to traditional PHIP, SABRE also takes advantage of the transfer of the spin order of para-H 2 to a target substrate  (v) the solution containing the 13 C HP compound is ejected from the reactor through an optional purification stage for in vivo administration. The system can be purged with N 2 gas (typically ∼1 min) to remove any residual solution from the PTFE lines to prepare for the next PHIP hyperpolarization production cycle brought together by an organometallic catalyst. The key difference is that with SABRE, the transfer of spin order is not dependent on the hydrogenation of an unsaturated bond of the substrate. This difference brings about two advantages (i) the expansion of the range of applications of PHIP 24,25 with respect to the types of molecules that can serve as substrates and (ii) that destruction or alteration of the original structure of those substrates is not required -one does not need to develop a precursor structure that will yield a desired agent upon hydrogenation. SABRE is the newest of the hyperpolarization techniques described in this article, and consequently, its development for MRS/MRI is less advanced. It has some intrinsic advantages including simplicity, low cost, scalability, and the potential for continuous agent production, which are promising for biomedical applications. Indeed, it has already achieved substantial polarization enhancements in solution, with polarizations reaching ∼10%. 38 -40 In order for SABRE to work, the catalyst must transiently bind both the target substrate and the atoms of para-H 2 with residence times that are long enough to allow the transfer of spin order or polarization to occur, but short enough to allow rapid turnover and subsequent accumulation of HP substrate in the solution. The process is illustrated in Figure 6. Although multiple organometallic complexes have been tested for SABRE, all to date use an iridium atom at the center -including the most successful SABRE catalyst, IrCl(COD)(IMes), where  IMes = 1,3-bis(2,4,6-trimethyl-phenyl)imidazol-2-ylidene and COD = cyclooctadiene. While the latter is not commercially available, its preparation is straightforward for those experienced with synthesis in inert-atmosphere environments. 41 As with many catalysts used in traditional PHIP, the basic iridium-based SABRE catalyst must first be activated before use. This typically involves dissolving the catalyst in an organic solvent along with excess substrate such as pyridine or a structural derivative: attempting to activate the catalyst without excess substrate can lead to irreversible deactivation of the catalyst. 42 The solution is charged with para-H 2 , either by loading a few atm of para-H 2 over the sample within a pressureresistant vessel or by continuously bubbling the para-H 2 within a specially designed apparatus/tube, such as that depicted in Figure 7. Stepwise hydrogenation of the molecular COD-ring and binding of the H 2 to the Ir leads to loss of the eight-carbon ring, and the formation of a hexacoordinate site complex, with ligands comprising the IMes group, two hydrides, and three substrates, two of which are exchangeable. The progress of activation may be followed by observing changes in the characteristics of hydride resonances of transient species and the final activated structure. 42 SABRE experiments can be divided into two types: ex situ and in situ, depending on the location of the sample vessel during polarization transfer with respect to the MR magnet. The vast majority of SABRE experiments are performed ex situ: spin order is transferred via scalar couplings between the involved nuclear spins during the lifetime of the catalyst complex. The spin flips must be nearly energy conserving: thus, the magnetic field must be sufficiently low to make the frequency difference between the hydride spins and the substrate spins similar to the scalar coupling between them. This corresponds to ∼6 mT for 1 H and ∼1 μT for 15 N. 40 Thus, the sample must be outside the MR magnet or within a fringe field for polarization transfer, and the sample rapidly moved to the magnet for detection. However, SABRE enhancements can also be observed in situ at high field. 41 In this case, the mechanism appears to be mediated by dipolar cross-relaxation between the HP hydride spins and the substrate spins. 41,42 While much less efficient, in situ SABRE does not require sample shuttling. In addition, the application of RF pulse sequences, for example, to allow mixing of spin states by driving the spins into level anti-crossing regimes, 43 may dramatically improve the efficiency of SABRE in situ.
To date, only a limited number of HCAs have been HP via SABRE. Besides pyridine (with 1 H and 15 N polarizations approaching ∼10% 38,40 ), SABRE enhancements of amino acids and small peptides have been reported at low field. 44 Most attention has been directed at biologically relevant pyridine derivatives. For example, 1 H enhancements of ∼230 and ∼1400 vs conventional MRS were reported for the tuberculosis drugs isoniazid and pyrazinamide in deuterated methanol at 700 MHz, respectively, corresponding to 1 H P values (P H ) of nearly ∼1.3 and ∼8%. 39 Nicotinamide was used to demonstrate SABRE in pure water, albeit with a more modest enhancement of ∼33-fold at 9.4 T, 42 by first activating the catalyst in ethanol, drying it, and reconstituting it in an aqueous solution.
Finally, in addition to the limitations imposed by the catalysts and substrates that are presently available and the need to perform SABRE in biologically compatible media, the catalysts themselves present an obstacle to the application of SABRE in the clinic. The catalysts are expensive and potentially toxic, making their recovery desirable after agent hyperpolarization. In this regard, following efforts of heterogeneous PHIP (HET-PHIP) by Koptyug and coworkers, 32 'HET-SABRE' enhancement of pyridine using SABRE catalysts that were covalently immobilized onto solid substrates of microscale polymer beads has been achieved. 45 Although the enhancements were small (approximately fivefold), the feasibility of creating separable catalysts for the creation of pure SABRE HP agents was demonstrated.

Spin-exchange Optical Pumping (SEOP)
Hyperpolarization of noble gases via SEOP was first demonstrated by Bouchiat et al. in 1960, following earlier pioneering work involving OP of alkali metal vapors by Kastler and others. 46 Over time, the technique has been refined with an aim to increase efficiencies of the OP and SE steps, as well as to reduce spin-relaxation or 'spin-destruction' rates, to enable the production of HP gases with sufficiently high P in large enough quantities for NMR/MRI applications. Major technical advances in these areas came with the implementation of high-power laser diode arrays (LDAs) 47 and the introduction of buffer gases and cell coatings. 46 These improvements have contributed to increases in noble gas polarization, including values as high as ∼90% for 129 Xe (e.g., Ref. 48), upon which our discussion will focus.
Although the complexity of SEOP setups may vary, one minimally needs (i) an OP 'cell' that contains the noble gas of interest ( 129 Xe) and a small amount of alkali metal (Rb in the following discussion); (ii) a source of circularly polarized light resonant with the Rb electronic transition; (iii) a heater for the cell (to vaporize the Rb); and (iv) a magnetic field source. Circularly polarized laser light is used to optically pump Rb electrons into a selected spin state, as shown in Figure 8. A usually weak (∼30 Gauss) magnetic field provides hyperfine Zeeman splitting of the Rb electronic energy states. Angular momentum is then transferred from the Rb electrons to the 129 Xe nuclear spins (Figure 8b) via gas-phase collisions. 46 Provided the OP/SE rates, respectively, exceed the electron and nuclear spin-relaxation rates, 129 Xe polarization can be increased far above thermal equilibrium according to 49 where P Xe and P Rb represent the polarizations of xenon nuclei and rubidium electrons, respectively, γ se is the SE rate, γ w the rate of 129 Xe spin destruction, and t p the polarization time. Also present is N 2 as a buffer gas, which at a pressure of ∼100-200 Torr is sufficient to collisionally quench alkali metal fluorescence, thus avoiding re-emission of photons with random polarization that would otherwise decrease P Rb . 46,50 One challenge for hyperpolarizing 129 Xe is that most Rb/Xe collisions tend to depolarize Rb atoms without conserving spin angular momentum, making the Rb spin-destruction rate effectively proportional to the xenon nuclear density [Xe] -an issue that can be mitigated through the use of higher resonant laser fluxes (albeit with greater demands on thermal regulation 50 ) and/or lower Xe partial pressures. Conducting SEOP for 129 Xe at lower total cell pressures also enables more efficient spin-exchange through three-body van der Waals collisions, in addition to the binary spin-exchange that is dominant at higher gas pressures. 51 On the other hand, higher cell pressures can broaden the Rb spectral lines resulting in more efficient absorption of the light. 49 Finally, silicon coatings are often used to reduce Xe nuclear spin destruction, and all the components of the SEOP setup must be chosen to minimize exposure of the HP gas to paramagnetic centers, strong field gradients, and zero crossings of the magnetic field. Many different polarizers have been developed to maximize the achievable polarization, P Xe , and/or the rate of production of HP Xe. Their designs are typically grouped into two categories. First, in the continuous-flow (CF) method illustrated in Figure 9, 49,51 a xenon gas mixture is polarized as it passes continuously through the OP cell and is subsequently collected using a cryogenic 'cold finger', 49 where it is stored at high field (>500 G) before it is carefully warmed 52 to allow transfer to a secondary vessel with minimal losses of P Xe . For example, Ruset et al. 51 (Figure 9a) employed a number of features -including a Rb presaturator upstream of a ∼1.8 m optical cell, a high laser power (∼90 W), a carefully designed optical path, and a relatively low Xe density and total pressure (allowing it to exploit the more efficient three-body van der Waals SE interaction) -to attain a peak 129 Xe polarization of ∼67% and a high output rate of 6 L Xe/h at %P Xe = 22%.
Second, the stopped-flow (SF) method (Figure 9b 48,53 ) takes place in a closed system where gas delivery, SEOP, and collection are performed sequentially. The cell is loaded with the desired gas mixture, and the gas is illuminated by the laser  for ∼10-30 min until P Xe nears a steady state. Before collection, the SEOP cell is cooled to condense the alkali metal vapor. The HP Xe can then be cryo-collected or simply expanded into another container (e.g., a Tedlar ® bag).
The second approach is simpler and easier to automate, and the reduction in magnetization density that results from HP gas dilution with buffer gas can be mitigated using high [Xe] mixes. 48 Nikolaou et al. recently presented an 'open-source' SF polarizer design (Figure 9b-d 54 ) utilizing a 200-W narrowed (∼0.27 nm FWHM) LDA capable of attaining polarizations of ∼90%, ∼57%, ∼50%, ∼30%, for Xe partial pressures of ∼300, ∼500, ∼760, and ∼1570 Torr backfilled with N 2 to 2000 Torr, with a throughput of ∼1 L h −1 . 48 Standard diagnostic techniques used in SEOP experiments include in situ low-field NMR spectroscopy to determine and control for P Xe , and near-infrared optical absorption spectroscopy to monitor pump laser wavelength, absorption profile, and inferred electronic P Rb . While not strictly necessary for HP Xe preparation, additional diagnostic techniques such as Faraday rotation, 55 ESR 46 to measure alkali metal density and spin polarization, and Raman spectroscopy to map elevated gas temperatures within the OP cell have also been implemented to better understand the complex processes underlying mass, energy, and polarization transport within the SEOP apparatus.
MRS applications of HP Xe typically exploit the extraordinary sensitivity of xenon's chemical shift to its local molecular environment (Figure 8c). 7 Of the two NMR-active xenon isotopes -129 Xe (I = 1/2) and 131 Xe (I = 3/2) -only 129 Xe has sufficiently long relaxation times in condensed phases to allow MRS. Xenon's highly polarizable electron cloud also allows it to participate in relatively substantial van der Waals interactions with proteins, lipid membranes, and living tissues. Xenon is not naturally found in the body in appreciable concentrations, which is both an advantage and a limitation: HP 129 Xe spectra will be background-free, but achieving efficient HP Xe delivery to the body is key. Delivery options are respiration and dissolution into the blood from the lungs; direct or intravenous injection of Xe-saturated solutions 7 ; or ex vivo gas dissolution into the blood or other solutions using off-the-shelf gas-exchange modules. 56 When Xe is dissolved in bodily tissues, a number of resonances corresponding to different cellular and tissue compartments may be observed generally within a range ∼185-215 ppm downfield from the gas reference at ∼0 ppm. 7 For example, the shift of 129 Xe is so sensitive that its resonance in red blood cells responds to changes in blood oxygenation. 57 Functional studies of the brain are thus possible. The large 129 Xe shift between xenon in the lung space and that dissolved in tissues may be exploited for a variety of pulmonary functional MRS/MRI studies, including studies of gas-exchange efficiency in healthy versus pathological tissues. Nevertheless, without further intervention xenon's concentration in the body is generally too low -and its interactions too weak and nonspecific -for 129 Xe MRS to provide much more in the way of specific biomolecular information.
To overcome this low sensitivity and improve its specificity, HP 129 Xe can be augmented with a host molecule, which can be functionalized to act as a biosensor as exemplified in Figure 10. 58 The approach exploits the strong interaction between xenon and cryptophanes, which are molecules that can bind Xe atoms within their internal hydrophobic cavities. Indeed, cryptophanes are an excellent choice for the superstructure of xenon biosensors, not only because of their relatively high xenon affinity (association constant, K a ∼ 10 3 -10 4 M −1 ) but also because the exchange and relaxation timescales are compatible with MRS. Moreover, they can be chemically functionalized to facilitate both aqueous solubility and covalent linkage to a chosen functional group to target specific bioanalytes. In principle, almost any ligand or antibody can be used as a functional group for the purposes of biosensing, 8 and multiple cages/biosensors could be simultaneously employed to allow multiplexed bioanalyte detection. The key feature empowering the biosensor approach is that the chemical shift of Xe within a biosensor cage that has bound a bioanalyte is measurably different from that of Xe residing within an unbound cage -thereby allowing the binding event to be detected via HP 129 Xe MRS. It is expected that for in vivo applications, rather than delivering Xe-loaded biosensors to the body in a single bolus, one would first administer the biosensors to the subject and allow them time to bind to their biomolecular targets. Then later, HP Xe may be delivered (using an above method) so that it would have the opportunity to locate the biosensor cages immediately before MRS.
This type of direct-detection experiment would yield three types of 129 Xe peaks: one for unbound Xe in the bulk environment, one for Xe trapped in an unbound host biosensor molecule, and one for Xe residing in biosensor molecules bound to targeted analytes. However, if the targeted bioanalyte is in low concentration, the bound Xe signal can be buried in noise. One way around this problem is to employ HP chemical exchange saturation transfer, or Hyper-CEST. 59 Hyper-CEST requires that the exchange of HP nuclei between sites is slow enough to give rise to unique, addressable resonances -yet fast enough to allow significant fractions of the populations to undergo exchange during the experiment. 8 Thus, if a saturating RF pulse is applied at the frequency for a bound Xe resonance, it will reduce the magnetization not only for the bound Xe pool but also for the (typically much larger) bulk Xe pool: The chemical structure of a xenon biosensor molecule comprises a cryptophane cage, a linker group, and a molecule (e.g., biotin) to bind to a targeted bioanalyte (e.g., avidin) with high affinity and specificity. 58 Below the structure are 129 Xe spectra monitoring the binding of biotin-functionalized xenon to the prototypical bioanalyte avidin. The spectra show only the cryptophane-bound peaks; the (much larger) peak corresponding to free xenon in water is at 193 ppm. The spectra differ by the absence (top) and presence (bottom) of 80 nmol avidin in solution, manifesting in the bottom spectrum by the appearance of a new peak at ∼72.5 ppm. Adapted from figures courtesy of Prof. Alex Pines off-resonance saturation provides a control/reference signal. The greater sensitivity afforded by indirectly measuring the presence of activated biosensors via the more sensitive bulk Xe signal allows detection of concentrations as low as ∼100 nM, with further optimization not inconceivably improving the detection limit to the ∼700 fM range. 8

Conclusions
In summary, four hyperpolarization techniques have been described here in detail that either successfully enable in vivo MRS studies (d-DNP, PHIP, and SEOP/Xe) or have a strong potential for in vivo MRS in the near future (SABRE). Key features of these hyperpolarization methods from the perspective of their biomedical use are compared in Table 1. Some of these techniques have evolved rapidly from proof-ofprinciple studies using custom-built instrumentation to robust commercial equipment. This in turn has fueled reliable in vivo experimentation -including the first clinical trials in human volunteers. The ongoing innovations in instrumentation and the fundamental technology of hyperpolarization will likely continue to significantly and positively impact access and applications for the broader biomedical community, expanding the utility of HP methods as a new tool for probing fundamental biomedical questions.