Pathway to Cryogen Free Production of Hyperpolarized Krypton-83 and Xenon-129

Hyperpolarized (hp) 129Xe and hp 83Kr for magnetic resonance imaging (MRI) are typically obtained through spin-exchange optical pumping (SEOP) in gas mixtures with dilute concentrations of the respective noble gas. The usage of dilute noble gases mixtures requires cryogenic gas separation after SEOP, a step that makes clinical and preclinical applications of hp 129Xe MRI cumbersome. For hp 83Kr MRI, cryogenic concentration is not practical due to depolarization that is caused by quadrupolar relaxation in the condensed phase. In this work, the concept of stopped flow SEOP with concentrated noble gas mixtures at low pressures was explored using a laser with 23.3 W of output power and 0.25 nm linewidth. For 129Xe SEOP without cryogenic separation, the highest obtained MR signal intensity from the hp xenon-nitrogen gas mixture was equivalent to that arising from 15.5±1.9% spin polarized 129Xe in pure xenon gas. The production rate of the hp gas mixture, measured at 298 K, was 1.8 cm3/min. For hp 83Kr, the equivalent of 4.4±0.5% spin polarization in pure krypton at a production rate of 2 cm3/min was produced. The general dependency of spin polarization upon gas pressure obtained in stopped flow SEOP is reported for various noble gas concentrations. Aspects of SEOP specific to the two noble gas isotopes are discussed and compared with current theoretical opinions. A non-linear pressure broadening of the Rb D1 transition was observed and taken into account for the qualitative description of the SEOP process.


Introduction
Nuclear magnetic resonance imaging (MRI) of the respiratory system using hyperpolarized (hp) 129 Xe is increasingly attracting attention for clinical [1,2,3,4,5,6] and preclinical research [7,8] despite the associated lower signal intensities compared to the more established hp 3 He MRI [9,10]. Hp 129 Xe provides additional information due to its chemical shift and tissue solubility [11] and its attractiveness is further augmented by the limited availability of the 3 He isotope [12,13]. The isotope 83 Kr possesses a nuclear electric quadrupole moment (eQ) that may enable hp 83 Kr to be used as a surface sensitive contrast agent and biomarker [14,15].
Both noble gas isotopes, 129 Xe (nuclear spin I = 1/2) and 83 Kr (I = 9/2), can be hyperpolarized through spin exchange optical pumping (SEOP) with alkali metal vapor [16,17,18,19]. Alternatively, dynamic nuclear polarization (DNP) at 1.2 K temperature was reported recently that allows for at least 7% hp 129 Xe production [20]. For SEOP, the noble gases are typically diluted in helium -nitrogen mixtures and, in the case of 129 Xe, the hp xenon is subsequently separated from the other gasses by a freezethawing cycle using a cold trap at 77 K [5,21,22,23]. This process is not viable for hp 83 Kr because of its rapid quadrupolar relaxation in the frozen state [24,25]. Although cryogenic separation of hp 129 Xe is straightforward in a physics or chemistry laboratory with acceptable losses [23,26], it would be desirable to eliminate cryogen usage to facilitate hp 129 Xe MRI applications in typical clinical and pre-clinical settings.
A high noble gas concentration in the SEOP gas mixtures would reduce the need for gas separation and could open up the pathway for cryogen free hp noble gas MRI. Unfortunately, a high noble gas density, [NG], adversely affects the obtained noble gas spin polarization, P NG , because it reduces the alkali metal electron spin polarization in the SEOP process. The adverse effect of [NG] on P NG is further exacerbated by the diminishing effect of [NG] upon the spin exchange rate, c SE [21,27,28,29,30]. If cryogenic separation is omitted, a trade off between noble gas concentration and obtained spin polarization exists. For example, a spin polarization of approximately 1% was generated in a previously reported 83 Kr SEOP experiments using a mixture of 95% krypton with 5% N 2 . Reducing the noble gas concentration to 25% krypton led to four fold higher spin polarization but the MR signal did not improve because polarization increase was offset by the noble gas dilution [31].
A potential solution for the conundrum to generate high P NG at high noble gas concentrations is to reduce [NG] through decreasing the total pressure of the gas mixture containing a high percentage of the respective noble gas. Optical pumping far below ambient pressure had been the method of choice in many of the pioneering SEOP studies [16,17,32,33,34], but low pressure SEOP was largely abandoned with the advent of high power solid state lasers that provide better polarization at elevated gas pressures due to pressure broadening of the rubidium D 1 transition. However, line narrowed high power diode array lasers have become available [28,34,35] that make pressure broadening less beneficial. Even non-narrowed (typically 2 nm linewidth) solid state lasers benefit from 129 Xe SEOP at a gas pressure below ambient, as previously demonstrated by Imai et al. [36]. Unfortunately high spin polarization .12% was obtained (at 15 kPa pressure) only for mixtures with low xenon concentration leaving cryogenic separation as a remaining desirable step. However, the work by Imai et al. also demonstrated that recompression of hp 129 Xe to ambient pressure after SEOP is feasible without significant losses in spin polarization. Recompression of the hp noble gas to ambient pressure would be a crucial step for intended low pressure SEOP usage for in vivo MRI applications.
In contrast to 'continuous flow' SEOP [5,21,22,23,37,38,39,40] that is technically more demanding [22,23,40], 'stopped flow' SEOP is applied to a stagnant gas mixture until the steady state polarization has been reached. The hp noble gas is then shuttled through pressure equalization into a pre-evacuated chamber for high field MR detection without repressurization. The advantage of 'stopped flow' 129 Xe SEOP was noted previously [41] and remarkably high 129 Xe spin polarization were reported recently [28]. With the noticeable exception of the work by Fujiwara and coworkers [42,43], pulmonary MRI typically uses hp gas in batched volumes. Therefore stopped flow SEOP may be of interest for pulmonary hp 129 Xe MRI applications, in particular if it provides some advantages beyond current continuous flow methods.
To date, stopped flow SEOP is the only viable technique for hyperpolarizing noble gases with nuclear electric quadrupolar moment such as 83 Kr [44,45]. In this publication, stopped-flow SEOP was studied with mixtures containing 5-78% of either krypton or xenon at total gas pressures ranging from 5 kPa to 200 kPa and above. Current theory was applied to attempt a qualitative interpretation of the experimental data.

Stopped Flow SEOP
The experimental setup is sketched in Fig. 1. Mixtures containing various concentrations of 129 Xe and 83 Kr were hyperpolarized in borosilicate glass SEOP cells (length = 120 mm, inner diameter = 28 mm) containing ,1 g Rb (99.75%; Alfa Aesar, Heysham, England, UK). The SEOP cell was housed in an aluminum oven with quartz windows and temperature controlled using heated air. The fringe field of a 9.4 T superconducting magnet provided the magnetic field of B 0 &0:05T for the SEOP process. Unless otherwise specified, a line narrowed diode-array laser system (30 W, 0.25 nm linewidth Comet Module, Spectral Physics, Santa Clara, CA, USA) tuned to the D 1 transition of Rb (794.7 nm) was used to irradiate the SEOP cell with collimated, circularly polarized light of 23.3 W power (incident at SEOP cell).
Steady state, nuclear spin polarization was reached after 6 minutes for 129 Xe SEOP and after approximately 18 minutes for 83 Kr SEOP. However, due to time restraints 83 Kr SEOP times of only 8 minutes were used resulting to 80% completion of the built up, as verified by measurements at both high and low SEOP pressure. During SEOP, the gas mixture was contained within the SEOP cell with valve 2 closed (see Fig. 1A). Valve 1 was kept open initially to allow for pressure monitoring but was closed approximately 2 minutes before delivery. The borosilicate detection cell and PFA transfer tubing were evacuated (valve 3 open) during the SEOP duration. After SEOP completion, valve 3 was closed and valve 2 opened. Pressure equalization caused rapid hp gas transfer via 1.5 mm (inner diameter) PFA tubing into the 15 mm borosilicate detection cell. The detection cell located within a 9.4 T superconducting magnet and a Magritek Kea 2 spectrometer (Wellington, NZ) with custom-built probes tuned to the resonance frequencies of 129 Xe (110.5 MHz) and 83 Kr (15.4 MHz) where used for detection.

Laser Power Adjustment and Optical Measurements
In addition to the line narrowed Comet laser, two broadband 30 W Coherent (Santa Clara, CA, USA) fiber array packaged (FAP) lasers were also used as a non-narrowed laser system (2 nm linewidth) for SEOP efficiency comparison with the line narrowed Comet laser. Due to the experimental setting only 15.6 W of FAP laser power was used to irradiate the SEOP cell. To have a proper comparison between the narrowed and broadband laser systems the laser power of the narrowed laser was reduced to approximately match the power of the broadband system. Fig. 1B displays the optical elements used to reduce the power of the Comet laser. The first beam splitter in the path of the laser light was present in all experiments in this work and ensured that only a single plane of linearly polarized light would continue toward the SEOP cell. It was found that B 2 = 19 B 1 for the highly linear polarized Comet system and B 2 = B 1 for the FAP system (i.e. no linear polarization remaining due to passage through the long fibre optic cable of the FAP system). Laser power control was obtained through a l/2 wave plate followed by a second beam splitter. By rotating the l/2 wave plate the laser rejection (B 3 ) was controlled, thus enabling the power control for the laser irradiation (B 4 ) of the SEOP cell without changes in the irradiation profile (i.e. wavelength and spatial distribution). The incident laser power was measured at the SEOP cell using a Coherent PM150-50C water-cooled power meter. The same power adjustment procedure was also used for the power dependent measurements described in section 4.9.
The rubidium absorption linewidth in the presence of pure krypton, xenon, N 2 , and a Xe -N 2 mixture was measured through absorption experiments similar to those by Driehuys and coworkers [46]. An incandescent light source with a consistent emission over the observed wavelengths irradiated the SEOP cell in place of the laser. A fibre optic cable leading to the optical spectrometer, HR2000+ Ocean Optics (Dunedin, Fl, USA) with a spectral resolution of 0.04 nm was placed at the rear of the SEOP cell to measure the D 1 absorption line width at 794.72-795.15 nm.

Temperature Control
The temperature of the SEOP cell inside the oven was maintained by an inflow of heated air near the back of the cell. Two thermocouples attached to the SEOP cell were used to measure the cell temperature. The first thermocouple was placed at the frontal region of the cell (i.e. in approximately 10 mm distance from the laser illuminated window) where it was carefully shielded from IR radiation, while the second thermocouple was positioned near the back region of the cell. The data from the two thermocouples were fed into a temperature controller. With this setup, the temperature controlled incoming air provided sufficiently stable temperature conditions, although the actual temperature inside the cell could not be determined. The temperature was measured on the surface of the SEOP cell at the thermocouple locations during ramping and steady-state processes. Typical temperature difference across the cell was less than 10 K after the steady-state conditions were reached.

Gas Mixtures
Research grade Xe (99.995% natural abundance, 26.4% 129 Xe; Airgas, Rednor, PA, USA), Kr (99.995% natural abundance, 11.5% 83 Kr; Airgas, Rednor, PA, USA), and N 2 (99.999% pure, Air Liquide, Coleshill, UK) were used to prepare the gas mixtures used in this study. The mixtures with varying noble gas contents were prepared prior to the SEOP experiments using a custom built gas mixing system. The 'standard mixture' described in section 2.6 required the use of research grade He (99.999% pure, Air Liquide, Coleshill, UK) in addition to other gases.

Determination of Obtained Polarization Values
For the determination of the actual polarization value, the integrated signal intensities of the hp noble gases were compared to the integrated signal intensity of a thermally polarized sample of the respective gas. For the thermal 83 Kr NMR measurement, a 15 mm borosilicate sample tube was pressurized to 560 kPa of natural abundance Kr gas leading to T 1 &65s at 298 K [47]. Data were averaged from 360 acquisitions with a 360 s recycle delay time between pulses. Similarly, for the 129 Xe thermal measurement, a sample tube was pressurized to 500 kPa containing 4 amagat of natural abundance Xe gas and approximately 1 amagat of O 2 in order to reduce the longitudinal relaxation time to T 1 v5s (T 1 &2:6s at 4.7 T [48]). Data were averaged from 120 acquisitions with 120 s recycle delay time between pulses. Taking into account the differences in concentration, pressure, and number of scans the integrated intensities from the thermal samples were compared with the integrated intensity of the hp samples to obtain the polarization enhancement over the thermal spin polarization.
For nuclei with arbitrary spin I the spin polarization P in a thermal equilibrium is given [45]: with c as the gyromagnetic ratio, k B as the Boltzmann constant, and B~h=2p as the Planck constant. Eq. 1 assumes Boltzmann population distribution at high temperatures where TwwDcDBB 0 =k B , a condition that is fulfilled for the thermally polarized samples described above. Note that the thermal samples and the 'standard mixture' (described in section 2.6) where rerecorded with another NMR system (Bruker, Avance III at 9.4 T) in order to confirm the obtained hyperpolarization values with the Kea 2 spectrometer.

Accuracy of Polarization Measurements
The SEOP generated polarization can be measured with high precision through high field NMR spectroscopy. However, the polarization values will scatter due to fluctuations in the SEOP cell. For example, the cell surface will 'cure' after reloading with rubidium metal, probably due to redistribution of surface condensed Rb, and the obtained hyperpolarization will increase initially for up to a few hours for cells newly loaded with rubidium. Further, contamination with oxygen, CO 2 , or H 2 O will lead to a slow decrease in the obtainable hyperpolarization. Some of the cells that appear to be nearly identical lead to slightly different hyperpolarization values. Because of the many factors that may influence these measurements data sampling was randomized during parts of the experiment. To characterize experimental variation in cell performance over time a polarization value was obtained for a standard mixture (5% Xe, 5% N 2 , 90% He at 230620 kPa and 373 K). This polarization value, averaged over a few experiments, was measured to be 44.065.4% and was further used for the 'quality control' test of a given SEOP cell. Three different SEOP cells that consistently achieved polarization values in this range were used during the course of the experiments. If the achievable polarization of a cell fell outside this range, it was cleaned and refilled with rubidium. Errors reported for the polarization measurements are based on the 65.4% error of the standard mixture and scaled accordingly.

Data Analysis
Data analysis was performed using Igor Pro Version 6.2 from Wavemetrics (Lake Oswego, OR, USA). Fitting parameters for spin-exchange optical pumping were extracted using built-in nonlinear least squares fitting algorithms.

Background to the 83 Kr and 129 Xe SEOP Experiments
The unit 'amagat' for the number density [M i ] of gas phase atoms or molecules is often used for convenience. In this work an amagat is defined as the density of an ideal gas at standard pressure and temperature of 101.325 kPa and 273.15 K and therefore 1amagat~2:6868|10 25 m {3 . Note that the amagat was historically defined as the density of the specific gas at standard pressure and temperature resulting to the slightly different value for instance for xenon with 2:7048|10 25 m {3 [49]. The small difference of less than 1% between the two definitions indicates almost ideal gas behavior for xenon at this condition.

Expected Pressure Dependence
The increase of the noble gas spin polarization as a function of the total pressure decrease is expected from [21,50]: where c op is the optical pumping rate caused by laser irradiation of the alkali metal atoms (i.e. by irradiation of rubidium (Rb) atoms with circular polarized light at the D 1 transition at 794.7 nm for all experiments described in this work). In principle, the rate c op (z) is a function of position within the pump cell due to the weakening of the laser in the optically thick medium [39,51], however for the purpose of this work an averaged value c op is assumed for simplicity, noting also the presence of significant gas convection in the SEOP cell [52]. The rate constant c SE describes the spin exchange rate and C is the longitudinal relaxation rate of the noble gas atoms. The polarization, P NG , increases with increasing SEOP time, t p , until the contribution from the exponential term in Eq. 2 becomes negligible and the steady state value of polarization P NG has been reached. The Rb electron spin polarization is limited by spin depolarizing collisions with inert gas atoms described by the gas (M i ) specific rate constants k i sd multiplied by the number density of the corresponding gas, M i ½ . A further limitation is through radiation trapping described by the rate constant c trap [30] that is further discussed below (see section 3.3) and by the rate constant c vdW that is caused by spin rotation interactions (i.e. interaction of the Rb 5s electron spin with Rb-M i molecular rotation -see section 3.4). A major contribution to the Rb depolarization in the gas phase at SEOP pressures p tot w20{50kPa is caused by binary atomic collision. The rate constants caused by these interactions are directly dependent on the density of the respective atoms [33]. The rate constant of xenon is k Xe sd~5 :2|10 {21 m 3 s {1 and is about 500 times larger than that of molecular nitrogen and more than 3 orders of magnitude larger than that of helium (see Table 1). Similarly, the rate constant of krypton, k Kr sd~1 :1|10 {21 m 3 s {1 , is a factor of 100 higher than that of molecular nitrogen. Therefore, even in the 95% nitrogen and 5% krypton gas mixture the contribution of molecular nitrogen to the overall Rb electron spin relaxation is only about 14% of the total gas phase relaxation caused by binary collisions. Moreover, in all other mixtures used in this work the nitrogen contribution to rubidium 5s electron spin depolarization through binary collisions is assumed to be below 4%.

Contribution of Rb-Rb Collisions
Unlike typical experiments at high SEOP pressure, depolarization of the rubidium electron spin due to rubidium-rubidium atom collisions may contribute significantly to Rb depolarization in the gas phase at low SEOP gas densities. The fairly large corresponding rate constant k Rb{Rb sd &8:1|10 {19 m 3 s {1 indicates that electron magnetic dipole -dipole interactions are responsible for the relaxation mechanism [51]. Depolarization due to Rb-Rb collisions depends on the rubidium number density [Rb] and is therefore a function of the SEOP cell temperature. An empirical equation (replacing an older, similar equation by Killian [53]) for [Rb] in m 23 as a function of temperature T in Kelvin is [54,55]: Using Eq. 3 one obtains that Rb ½ 373K~6 :0|10 18 m {3 at 373 K. However Eq. 3 should be used with caution for Rb concentration calculations as uncertainties arise from the difficulty of proper temperature monitoring inside the SEOP cell during onresonance irradiation with a high-powered laser as explained further in the text (see section 4.3 for discussion of the correction factor, c Rb , to [Rb]).
The potential uncertainty in temperature is quite inconsequential for the rubidium depolarization in 129 Xe SEOP since the rubidium density at a temperature of 373 K leads to a relaxation rate of ½Rb 373K : k Rb{Rb sd~4 :8s {1 that contributes less than 2% to the Rb gas phase relaxation at the lowest pressure (5 kPa) and the lowest xenon concentration (5.0%) used. The significance of Rb-Rb collisions to the Rb depolarization decreases further as the total gas pressure and the xenon concentration increase. However, the situation is quite different in 83 Kr SEOP. Firstly, the rate constant k Kr sd is about 5 times smaller than k Xe sd , thus increasing the relative importance of k Rb{Rb sd for the rubidium depolarization. Secondly, 83 Kr SEOP produces the highest nuclear spin polarization at 433 K and, according to Eq. 3, Rb ½ 433K~1 :6|10 20 m {3 . This translates into 27 fold increase in Rb concentration as compared to Rb ½ 373K and Rb-Rb collisions contribute therefore significantly to the rubidium depolarization, in particular at low SEOP pressures. For example, at 30 kPa total gas pressure the contribution of Rb ½ 433K : k Rb{Rb sd to the Rb gas phase depolarization ranges from approximately 2% (for the 74% krypton mixture) to 5% (for the 25% krypton mixture) to about 20% for the leanest (5%) krypton mixture. Therefore uncertainties in SEOP temperature (and therefore [Rb]) can affect the second term in Eq. 2 for 83 Kr SEOP.

Radiation Trapping
Molecular nitrogen is an important component of an SEOP gas mixture because it can, unlike mono-atomic noble gasses, dissipate energy from excited rubidium electronic states into vibrational modes [32,56]. This non-radiative relaxation pathway reduces rubidium fluorescence, depending on the N 2 number density [30]. In SEOP mixtures with high rubidium density [Rb], fluorescence may be detrimental to the Rb spin polarization because it can lead to radiation trapping where a single incident circularly polarized photon gives rise to multiple scattered photons that are arbitrarily polarized. Wagshul and Chupp [56] have reported a formula based on earlier experimental work [57] that quantifies the extent of quenching through N 2 . A slight modification of this formula, i.e. multiplication with the c trap N2 ½ ~0 term from SEOP in the absence N 2 , leads to an expression similar to the one reported by Brunner and co-workers [52]:

Rb Depolarization Caused by Spin-rotation Interactions
At lower pressures with correspondingly longer lifetimes of the Rb-Xe van der Waals complexes, a significant Rb polarization loss is induced by spin rotation interaction [58]. In Eq. 2 this effect is represented by the rate c vdW . The functional dependence of c vdW on SEOP gas pressure and composition is difficult to quantify. For an SEOP gas mixture with fixed concentration in the long-lifetime pressure regime (i.e. at very low pressures), the relaxation rate c vdW will increase with the pressure increase due to the intensified complex formation. At sufficiently high pressure the short molecular lifetime regime is reached and the further increase of complex formation with increasing pressure will be offset by higher breakup rates, thus resulting in pressure independent c vdW . In this regime, the Rb nuclear-electron hyperfine interaction limits the influence of spin-rotation relaxation. At further pressure increase however, the very short lifetime regime is reached with a diminished hyperfine interaction and therefore, c vdW starts to increase again with increasing pressure until the hyperfine interaction has become completely negligible. For a 1% Xe, 1% N 2 , and 98% He SEOP mixture, a rate of c vdW &2|10 3 s {1 at 423 K (and an approximately 60% higher value at 353 K) has been reported for the short lifetime limit [58]. This value is comparable to that of k Xe sd Xe ½ &2|10 3 s {1 caused by binary collisions in 129 Xe SEOP at 40 kPa and 373 K in the 5% Xe -95% N 2 mixture. The relaxation rate c vdW is however mixture dependent. For instance completely replacing helium by nitrogen should considerably reduce c vdW [59] as N 2 facilitates the break-up of the Rb-NG van der Waals dimer better than helium. Unfortunately literature data of c vdW for the mixtures used in this work are not available. SEOP conditions in the current work are likely to create short to very short lived Rb-NG van der Waals complexes. Therefore, to a first approximation and within the scope of this work, c vdW will be considered as pressure independent because of its general pressure independence in the short lifetime limit and because of its relatively small pressure dependence compared to binary relaxation, k NG sd NG ½ in the very short lifetime limit. In the lower pressure regime, where c vdW actually dominates Rb depolarization rate this crude approximation is destined to fail, therefore experimental data fitting with Eq. 2 (or modifications thereof) was not attempted in this pressure limit.

The Spin Exchange Rate
The spin exchange rate c SE results from the added contributions of (1) spin exchange in rubidium -noble gas van der Waals complexes that is characterized by the rate constant, c RbNG and (2) from spin exchange caused by binary collisions quantified by the velocity averaged binary spin-exchange cross section SsvT. Literature values of c RbNG and SsvT for 83 Kr and 129 Xe are listed in Table 1 [18,27,60,61], while Eq. 5 shows the contribution of both rates to c SE [27]: The rates c RbNG : Kr ½ {1 and c RbNG : Xe ½ {1 are comparable to their corresponding SsvT rates at a densities of 0.25 amagat and 0.4 amagat respectively (in the absence of nitrogen). In this density range, van der Waals dimers (mediated through three-body collisions) and binary collisions contribute about equally to the spin exchange. However, binary collisions will eventually dominate in the spin exchange process as the contributions from van der Waals complexes is expected to decline with the increase of the noble gas concentration and therefore its density [NG].
The N 2 molecules in the SEOP mixture also contribute to the Rb-NG dimer break up. This contribution is quantified by the characteristic pressure ratio b~p 0 NG ð Þ=p 0 N 2 ð Þ listed in Table 1 with the specific values for xenon and krypton [18,27,62]. The parameter r in Eq. 5 is the partial pressure ratio p N 2 ð Þ=p NG ð Þ (or N 2 ½ = NG ½ density ratio) in a mixture. The ratio b shows that a dilution of xenon in nitrogen can be beneficial to c SE . However, a dilution of krypton in nitrogen can be detrimental to c SE because the break up of van der Waals complexes is facilitated by nitrogen more than by krypton. Note however, that nitrogen is still beneficial for 83 Kr SEOP because of its radiation quenching effect (section 3.3) and because k Kr sd &100 : k N2 sd (section 3.1).

Noble Gas Polarization as a Function of SEOP Gas Pressure
Steady state, or near steady state spin polarization was obtained for the 129 Xe mixtures after about 6 min of SEOP at 373 K and a near steady state (approximately 80%) was reached after 8 min of SEOP for 83 Kr mixtures at 433 K. The steady state polarization P is shown as a function of the total SEOP pressure p tot in Figs. 2 and 3 for hp 83 Kr and hp 129 Xe respectively. The noble gas polarization P of both isotopes in all mixtures increased as the total gas pressure was decreased from 350 kPa to below ambient in all studied mixtures. The maximum steady state polarization P max for hp 83 Kr was obtained at a total gas pressure p tot , in the range of 35-50 kPa, depending on the krypton concentration used. Similarly, a polarization maximum was observed for hp 129 Xe, however at a lower total pressure range of p tot~2 0{30kPa. Reducing the pressure below these values resulted to a rapid drop in the steady state polarization of the noble gases. In order to facilitate the following discussions, the SEOP pressure that resulted to the highest observed steady state polarization P max , will be labeled as p P max . Table 2 lists P max for various mixtures, the corresponding total SEOP pressure p P max tot , and the corresponding SEOP partial pressure p P max NG . As can be seen from Table 2, the maximum 83 Kr polarization of P max~2 6:5% was reached for the 5% krypton -95% nitrogen mixture at an SEOP pressure of 54 kPa. This is a remarkably high spin polarization for a quadrupolar spin system observed at ambient temperature. 129 Xe SEOP at a pressure of 46 kPa using a 5% xenon mixture resulted to P max &65%spin polarization. Both results were obtained with a 23.3 W laser irradiation that resulted in a power density of 2.6 W/cm 2 at the SEOP cell front window.
Since hp noble gasses remain diluted without cryogenic separation process, the obtained polarization does not enable easy comparison with experiments that utilize cryogenic separation. It is therefore useful to define an apparent polarization, P app , scaled to the polarization, P, in the pure hp noble gas that would result to the same MRI signal.
The apparent polarization, P app , provides a measure of the 'usable' spin polarization in MR experiments if the hp noble gas is not separated from the nitrogen after SEOP. Table 2 also lists the apparent maximum steady state polarization P max app . The highest P max app was obtained for krypton with the 25% and 50% krypton  Note the maximum polarizations listed in Table 2 were generated every 6 minutes for hp 129 Xe and every 8 minutes for hp 83 Kr (and with slightly increasing values for P Kr at SEOP times up to 18 min). The ideal pumping time for MRI applications however may be shorter than these values if polarization can be compromised in favor for faster experimental repetition.

SEOP Temperature
The three-body spin exchange rate c RbNG and the binary cross section SsvT are both more than two orders of magnitude smaller for the Rb-83 Kr system than for the Rb-129 Xe system. The resulting small c SE rate has two adverse consequences for 83 Kr SEOP as predicted by Eq. 2. Firstly, a smaller c SE in the presence of a higher relaxation rate C leads to a reduced steady state polarization P for 83 Kr compared to that for 129 Xe under otherwise identical SEOP conditions. Secondly, smaller c SE values further result in slower 83 Kr SEOP polarization build up as compared to 129 Xe SEOP, thus increasing the repetition time in MRI applications. In order to, at least partially, offset this effect [Rb] needs to be raised through elevated 83 Kr SEOP temperatures. In addition to the increased [Rb], a further advantage of elevated 83 Kr SEOP temperatures comes from reduced quadrupolar relaxation of 83 Kr on the cell surface, as discussed in Appendix 2 in Supporting Information S1. It was found that up to a temperature of 433 K the benefit from the increased spin exchange rate c SE for 83 Kr SEOP outweighs other detrimental effects arising from elevated temperatures. In contrast, a temperature of 373 K was found to produce the highest 129 Xe spin polarization in this work. Examples of adverse effects at higher temperatures are increased Rb-Rb collision rates, as discussed in the section 3.2, and increased laser absorption in the rising optical density of the rubidium vapor phase.

Results from Inversion Recovery 83 Kr SEOP Experiments
The noble gas self-relaxation rate C is difficult to obtain from published data as it is specific to some SEOP conditions, for example SEOP cell dimensions and its surface temperature. However, the combined rate constants B~c SE zC can be extracted from the time dependence of the polarization obtained in SEOP experiments according to Eq. S1 in Appendix 1 in Supporting Information S1 (i.e. utilizing the time dependence of Eq. 2). In principal, build up curves can be measured directly inside the SEOP cell [28,61,63]. However, in this work the SEOP time dependence is determined through remotely detected NMR experiments (i.e. after hp gas transfer into the high field magnet) as no further experimental modification was required for the existing instrumentation. The drawback of this procedure was that the measurement of the build up curves required time-consuming point-by-point experiments. The data from inversion recovery 83 Kr SEOP experiments (see Appendix 1 in Supporting Information S1) are shown in Fig. 4A and the rate constants, B~c SE zC, obtained from fitting with Eq. S1 are listed in Table 3.
The spin exchange rates c calc SE listed in Table 3 were calculated using Eq. 5 with the relevant literature values reported in Table 1. However, the experimental value B&3:7|10 {3 s {1 obtained from the inversion recovery experiments for 83 Kr SEOP below 200 kPa presents a problem when combined with the calculated spin exchange rate values c calc SE in order to determine the first fraction in Eq. 2, c SE =(c SE zC) . Using c SE calc =B , Eq. 2 predicts an upper limit for the 83 Kr polarization of P max &11{14%. In reality, any experimentally measured value for P Kr would be further reduced because of P Rb ,1 and due to incomplete (approximately 80%) build up at t p~8 min in SEOP. In remarkable disagreement, the experimental data show polarization values of up to P max~2 6:5% and P max~1 7:7% for the 5%  Table 2. A. Solid lines represent data analysis with Eq. 8. Extrapolation of these theoretical curves to pressure ranges outside the fitted region are shown by dotted lines. B. Same experimental data as in (A) but the solid lines represent now the data analysis using Eq. 8 with the pressure dependence of the Rb D 1 absorption taken into account through Eq. 9. Extrapolation to pressure ranges outside the fitted region are shown by dotted lines. Fitting parameters for (A) and (B) are reported in Table 5A and 5B, respectively. doi:10.1371/journal.pone.0049927.g003 krypton and 25% krypton mixtures, respectively (see Fig. 2 and Table 2). The discrepancy between predicted maximum possible polarization and observed polarization may be due to incorrect literature data in Table 1 used for determining c calc SE . Note that the literature data was obtained at temperature conditions different from the ones used in this work. Another potential culprit is a wrong value of [Rb] obtained from Eq. 3 based on temperature measurements outside the cell. The temperature inside the cell under high power laser irradiation in the presence of the liquid rubidium metal is unknown. Wagshul and Chupp [51] noted a discrepancy of a factor of two or more in [Rb] under 129 Xe SEOP conditions from the prediction by the equilibrium vapor equation. Further doubt about [Rb] determination through external temperature measurements arises from Raman spectroscopical experiments by Happer and co-workers that provide access to the in situ temperature distribution within the SEOP cell by measuring the rotational -vibrational N 2 temperature [64]. The internal temperatures were found to substantially exceed those measured externally at the cell outside surface. Finally, a numerical simulation study [52] also draws a very complex picture about a non-uniform temperature distribution within a static SEOP cell with significantly elevated internal temperatures. The same, perhaps amplified problem may occur for 83 Kr SEOP experiments that are run at the cell outside temperature of 433 K. A correction factor c Rb for the rubidium concentration from Eq. 3 is therefore introduced for this work. It follows from the discrepancy between observed and calculated P max described above, that c Rb .2. An upper limit for the correction factor c Rb ,8 is obtained from the fact that C cannot be negative. Further, the upper limit can be reduced to c Rb ,6 if one assumes that relaxation rate C of 83 Kr is not significantly lower than typical rates found for 129 Xe under SEOP conditions. Further determination of c Rb for 83 Kr SEOP was not possible from the data in this work, however the qualitative outcome of the fittings in Fig. 2 is not strongly affected within the range 2, c Rb ,6. The correction factor was set to c Rb~4 for further data analysis in Fig. 2.
The similarity in the c calc SE values in Table 3 for 83 Kr SEOP is caused by the [Kr] independent rate constant SsvT that dominates over the c RbNG : Kr ½ {1 term even at the low pressures of p P max tot for all krypton mixtures. As pressure p tot wp P max tot , the van der Waals contributions will be even further marginalized. As a consequence, the inversion recovery 83 Kr SEOP curves in Fig. 4A   Xe polarization after SEOP time, t p , for two xenon-nitrogen gas mixtures at different SEOP pressures. The inversion recovery data from both (A) and (B) were analyzed using Eq. S1. Polarization data were normalized to their values at t p~2 040s for 83 Kr and t p~1 200s for 129 Xe to visually compare the rate differences of the mixtures and pressures. The obtained rate constants from fitting of both (A) and (B) are reported in Table 3. doi:10.1371/journal.pone.0049927.g004 all display similar time dependence at SEOP pressures below 200 kPa. At 310 kPa, the combined rate constant is increased due to the increased relaxation rate constant C. The functional form of the pressure dependence of C is explored in Appendix 2 in Supporting Information S1. Rewriting Eq. S4 as a function of the krypton number density and using r~N 2 ½ = Kr ½ leads to:   Table 4. At a first glance, the fitting result in Fig. 2 [30]. These values are quite high but an increase in c N 2 ½ ~0 trap with increasing rubidium density is expected. The c op rates listed in Table 4 are low and indicate low pumping rates as it would be expected for an optically thick medium with high [Rb]. The 2.8 fold decrease of c op with increasing krypton concentration is further discussed in section 4.8.

Results from Inversion Recovery 129 Xe SEOP Experiments
In contrast to 83 Kr SEOP, the time behavior of the 129 Xe SEOP polarization shown in Fig. 4B depends strongly on total pressure and gas composition (see Table 3). This observation is in agreement with previous work [28] and was expected since c RbNG , i.e. the van der Waals contribution to the spin exchange rate caused by three-body collisions, plays a more dominant role for 129 Xe SEOP than for 83 Kr SEOP. An increased c RbNG relative to the rate SsvT caused by two body collisions will result in a stronger noble gas density dependency for c SE in Eq. 5. Furthermore, the time scale of the inversion recovery is accelerated at low xenon density compared to that of 83 Kr (Fig. 4A) The combined rate constants B~c SE zC and the rates c calc SE for 129 Xe, as listed in Table 3, imply that the correction factor for [Rb], if needed at all, must be c Rb v1:6 because of the requirement C §0. Once again, c Rb cannot be further determined and the average c Rb~1 :3 of the range is taken. Furthermore, the assumption is made that Cis caused mainly by interactions with the surface and is therefore pressure and gas composition independent. This seems to be indeed the case with the exception of the data taken at 50 kPa that scatter widely. However, for 129 Xe SEOP at this pressure the values for C are relatively small compared to B and a significant error is not unlikely. Excluding 50 kPa data and averaging the 180 kPa and 300 kPa data one obtains C~9|10 {4 s {1 using c Rb~1 :3. Note, for c Rb~1 it follows that C~1:9|10 {3 s {1 in better agreement with data by Goodson et al. [28] who previously determined C~1:7|10 {3 s {1 in a coated SEOP cell. However, as will be discussed in the following section, the exact value is not very important for the description of 129 Xe SEOP in this work. 4.6. P Xe vs. SEOP Pressure Dependence above p max tot A qualitative analysis of the data shown in Fig. 3A was attempted with Eq. 8 derived from Eq. 2 with the inclusion of the correction factor for the rubidium density, c Rb . During the fitting procedure the rates c op and c vdW were used as the fitting parameter with the correction factor set to c Rb~1 :3 and the nuclear relaxation term to C~9|10 {4 s {1 . Unlike for 83 Kr SEOP that is run at a temperature of 433 K, the radiation trapping term for 129 Xe SEOP could be taken from literature data Table 4. Values for c op and c trap from fitting experimental data of 83 Kr spin polarization as a function of SEOP cell pressure in Fig. 2  with c trap~3 3000s {1 [30]. Furthermore, the SEOP duration was long enough to reach the steady state polarization value and therefore one could set f = 1. The rest of the constants used in the fitting procedure were taken from Table 1, in the case of the multiple choices of the literature data the constants from reference [27] were used. The resulting fits over the pressure range from 45 to 240 kPa are displayed (solid lines) in Fig. 3A (see also Table 5A for the relevant fitting parameters). The theoretical curves were further extended over the entire pressure range using the values for c op and c vdW obtained from fitting (dotted lines). Although, fitting curves using Eq. 8 seem to qualitatively describe the experimental behavior in Fig. 3A, the results listed in Table 5A  Note that the general appearance of the overall shape of the fitting curves is not dramatically affected by c Rb (at least within the range 1ƒc Rb v1:6), nor do the resulting values for the fitting parameters change significantly. Generally, the larger c SE =C ratio makes the first term in Eq. 8 less important for 129 Xe SEOP compared to 83 Kr SEOP. However, the unsatisfactory results of the data fitting with Eq. 8 will need some further considerations. The rubidium D 1 absorption linewidth may hold important information for the second term in Eq. 8 and may provide a better understanding of the experimental data. The effect of the D 1 linewidth is discussed in the following section. Fig. 5A shows IR absorption spectra of rubidium within the SEOP cell when illuminated by an incandescent light source. Spectra were acquired at 433 K with pure krypton for three pressures: 9 kPa, 68 kPa and 434 kPa. Only the D 1 transition (i.e. the 1 S 1=2 ? 1 P 1=2 transition at 794.7 nm) and its linewidth are relevant for the SEOP studied in the present work. The pressure behavior of the D 1 linewidth is depicted in Fig. 5B. Further theoretical analysis suggests that a [Xe] 1/3 , [Kr] 1/3 , and [N 2 ] 1/3 functional form provides a reasonably good description of the absorption linewidth behavior over the studied pressure range. The non-linear Rb D 1 line dependence on gas density dependence is in contrast to the linear gas density dependence usually found for alkali metal D 1 or D 2 transitions (see for instance [46,65]). The cause for this unexpected behavior was not further investigated and the exact functional description would benefit from refinement in future research. Fig. 5B shows that the linewidth in the presence of either krypton or N 2 at 433 K is much broader than that in the presence of xenon at 373 K. The Rb absorption linewidth with N 2 at 373 K was too close the resolution limit of the optical spectrometer used (i.e. 0.04 nm). The data demonstrates that all krypton-nitrogen mixtures at 433 K should lead to a D 1 broadening that is much larger than the laser linewidth (0.25 nm -dashed line in Fig. 5B) at all pressures above p max tot . However, a different situation occurs for xenon at 373 K, in particular in mixtures with N 2 . In these cases the laser linewidth may exceed the D 1 linewidth and thus not all of the laser power will be absorbed. The effect of the linewidth is difficult to quantify, in particular since exact on-resonance irradiation can be disadvantageous as explored in detail by Wagshul and Chupp [51] and recently observed for high power irradiation by Wild and co-workers [66] and by Goodson and co-workers [67]. However, for this work the simple assumption is made that laser irradiation with a wider linewidth than the D 1 linewidth will lead to a pressure dependent pumping rate that follows the same dependence as the D 1 linewidth itself:

Non-linear Pressure Broadening of the Rb D 1 Absorption Linewidth
with c op Ã as the optical pumping rate at 1 amagat total gas density.
The density dependent rate constant c op r ð Þ as defined in Eq. 9 replaces c op in Eq. 8. Using c op Ã and c vdW as fitting parameters with all other parameters kept identical to the ones used in section 4.6, fitting with Eq. 8 leads to the solid lines depicted in Fig. 3B with the values for rate constants listed in Table 5B. Once again, the theoretical curves were further extended over the entire pressure range using the values for c op Ã and c vdW obtained from fitting (dotted lines). The results for c op Ã listed in Table 5B Table 5. Values for c op , c Ã op , and c vdW rates obtained from the fitting of experimental data of 129 Xe spin polarization as a function of SEOP cell pressure (Fig. 3) using Eq. 8. A .

Mixture
A. Data fitting using Eq. 8 (Fig. 3A) B. Data fitting using Eqs. 8 and 9 (Fig. 3B) pressure dependence of the Rb D 1 (Eq. 9) in Eq. 8 appears to result to more realistic values for c op Ã and c vdW . While there is little effect on the qualitative appearance between the fitted curves in Figs. 3A and 3B, the extended curve (dotted line) in Fig. 3B provides a better description of the observed data compared to the one in Fig. 3A.
It should be noted again that Eq. 9 should be handled with care since it is based on a number of simplifying assumptions. Firstly, neither the line shape of the pressure broadened Rb D 1 transition nor the emission line shape of the frequency narrowed diode-array laser are Lorentzian or otherwise straightforwardly defined. Further, at high xenon concentration and pressure, the adsorption linewidth starts to exceed the laser linewidth causing the validity of the underlying concept in Eq. 9 to end. This may be the case in particular at high SEOP pressures for the mixture containing 78.2% xenon. Another factor, not considered here, is the pressure dependent shift of the D 1 transition. For 129 Xe SEOP at 373 K this shift is small with 0.13 nm over the used pressure range for pure xenon. Although the shift is larger at 433 K with 0.43 nm over the used pressure range for krypton (see Fig 5A) it is still small compared to the D 1 line broadening. Despite the limitation of Eq. 9, requiring more refinement in future research, the current work suggests that the effect of pressure broadening needs to be considered for a correct description of variable pressure 129 Xe SEOP with narrowed lasers.

Thermal Properties of SEOP Gases
The c op values for 83 Kr SEOP listed in Table 4 of change by a factor of approximately 2.8 between the gas mixtures used. The c op Ã rates found in 129 Xe SEOP summarized in Table 5B are less affected by [Xe] except for the mixture containing 78.2% xenon where the rate drops significantly. However, nothing in the general theory outlined in section 3 gives rise to the expectation that c op is affected by the noble gas-nitrogen ratio of the various mixtures. Nevertheless, at the same time it has been noted that the temperature gradient between the front and the back of the SEOP cell changed when SEOP mixture was altered.
The mixture dependent changes in the temperature gradient across the SEOP cell may have been induced by the different thermal conductivity of the used gas mixtures. Under the experimental SEOP conditions, N 2 has an approximate 2.5 times larger thermal conductivity than krypton (and 4.5 times larger than xenon) [68]. Therefore, as the krypton or xenon concentration in the SEOP cell is increased, the decreasing thermal conductivity allows for higher temperature difference between the laser-illuminated front of the SEOP cell and its back. The consequences of this temperature gradient are unknown but changes in local rubidium concentration, thermal convection, and laser penetration are likely to lead to different convection patterns within the cell [52,69]. Note also, that the heat capacity, C V , of N 2 is more than 5/3 larger than that of a mono-atomic noble gas. Therefore, the corresponding changes between the gas mixtures may potentially have a profound impact on quantitative SEOP measurements and comparison of data between different noble gas mixtures needs to be handled with great caution. Due to the higher temperature, 83 Kr SEOP may be stronger affected than 129 Xe SEOP.
Thermal conductivity and heat capacity effects may explain the mixture dependent c op values but would of course also require mixture dependent c Rb values. Unfortunately, the limited data in this work does not make the usage of a further fitting parameter reasonable in particular since the differences between the c op values are not too excessive.
However, a serious concern for the fitting of the experimental data would be SEOP gas pressure of the temperature, c op , and c Rb . Fortunately, no effect on the pump cell temperature gradient with pressure changes has been noted. Moreover, the well-known equation for the thermal conductance, k, of an ideal gas is where c c is the mean average velocity of the gas molecules, l is the mean free path, C V ,m is the molar heat capacity at constant volume, [M] the density of the gas, and N A is Avogadro's number. The thermal conductivity of an ideal gas is pressure independent because the gas density is directly proportional to the pressure, whereas l!p {1 and c c is also pressure independent.

Effect of Laser Power and Laser Linewidth
The effects of laser power on the polarization curves are shown in Fig. 6. The power of the laser irradiation was adjusted in the linear polarized part of the laser beam rotating the l=2 plate positioned in front of a beam splitter (see experimental section or Fig. 1B). This procedure allowed for the control of the laser irradiation power (incident at the SEOP cell) without changing the linewidth, the line shape, and irradiation pattern (i.e. beam shape). Fitting of the data was performed using Eq. 8 in the same fashion as in section 4.7 using c op r ð Þ as defined in Eq. 9. The parameter trap~3 3000s {1 at 23.3 W power was taken from literature [30] and was scaled linearly with the relative decrease of laser power.
Measurements at 23.3 W power were performed redundantly under the same pumping conditions as the ones used for 5% and 50% Xe mixtures displayed Fig. 3. The resulting rates, c Ã op , are listed in Table 6 for the two mixtures at various laser power levels.
The increase in c Ã op as the laser power is raised from 5.7 W to 23.3 W is 3.0 fold for the 50% mixture and is 2.6 fold for the 5% xenon mixture. However, the dependence of P max Xe on laser power (see Fig. 6) is more pronounced for the 50% mixture (approximately 2.0 fold increase in the polarization P max Xe between 5.7 W to 23.3 W) compared to the 5% xenon gas mixture (1.3 fold increase). The increasing importance of laser power for SEOP with higher noble gas concentration is due to the second fraction in Eq. 8 that makes the c Ã op (or c op ) values more relevant for the obtained polarization, P max Xe , if the destructive rates k NG sd NG ½ are high. Therefore higher laser power is particularly beneficial for higher noble gas concentration SEOP. This is an important observation for the concept of cryogen-free SEOP. Fig. 7 depicts a comparison of SEOP results obtained with a line narrowed (0.25 nm) Comet laser module using reduced laser power (17.3 W) and with a similar power (15.6 W) but using much larger linewidth (Coherent FAP, approximately 2 nm line width). Data were analyzed with Eq. 8 in identical fashion as above and the resulting c op Ã for broadband laser 129 Xe SEOP are listed in Table 6. Clearly, laser line narrowing is beneficial for SEOP as it leads to a 9.3 fold increase of c op for the 50% xenon mixture and to the 6.4 fold increase for the 5% xenon mixture. Similar to the laser power trend, the resulting improvement of P max Xe through line narrowing is particularly strong for SEOP with high xenon concentration. A 4.7 fold increase of P max Xe is observed in Fig. 7 for the 50% xenon mixture as compared to the 2.7 fold increase for the 5% xenon mixture. 4.10. Rapid Decrease of P NG with Decreasing Pressure below p max tot When the SEOP pressure was reduced below p P max tot (i.e. p P max tot~2 0{35kPa for 129 Kr SEOP and p P max tot~3 0{50kPa for 83 Kr SEOP) a sharp decrease in polarization was observed. Note, that data fitting was limited to pressures above p max tot , however simple extrapolation of the (high-pressure) fitting curves into the lower pressure region are shown as dotted lines in Figs. 2 and 3. These extensions seem to provide a remarkably good description of the low-pressure behavior. This result should however not be over-interpreted, in particular since the assumption of a constant c vdW will fail in the low-pressure regions (see section 3.4). The rate c vdW , caused by spin-rotation interaction, will lead to significant depolarization at lower pressure but its effect is overestimated in this work because its absolute value will decrease with decreasing pressure. Figure 6. 129 Xe polarization, P, dependence on laser power. 129 Xe spin polarization as a function of SEOP cell pressure for two different gas mixtures at four different SEOP laser power levels. Please refer to the figure legend for symbol explanation. The laser power was measured in the front of the SEOP cell. Data were analyzed using Eq. 8 (utilizing Eq. 9) within the fitting region (solid lines). Extrapolations to pressure ranges outside the fitted region are shown by dotted lines. The fitting procedure is discussed in section 4.9 and the results of the data analysis are listed in Table 6. doi:10.1371/journal.pone.0049927.g006 There are further effects that contribute to the rapid polarization drop below p max tot . Radiation trapping, discussed in section 3.3, reduces the rubidium electron spin polarization. Radiation trapping will increase with lower p tot values, in particular in mixtures with high noble gas concentration (i.e. low N 2 concentration) as described by Eq. 4.
A contribution to the polarization drop at pressures below p max tot , that is not accounted for in Eq. 8, may be caused by an optically dense boundary layer of rubidium at the cell window that is illuminated by the laser. This layer will reduce the resonant laser light penetrating the SEOP cell at any pressure. As demonstrated by Wagshul and Chupp [51] its effect is particularly detrimental at low pressures when the resonant absorption cross section of the rubidium is very high, leading to an almost complete absorption of the resonant laser light. The situation can be alleviated by detuning the laser to (slight) off-resonant illumination (not attempted in this work) and by the usage of very high laser power densities [51]. This effect was not investigated in this work.
Furthermore, the sudden drop in P NG with decreasing SEOP pressure may be caused by a dramatic increase in rubidium relaxation due to the combination of increased diffusion and wall relaxation [33,51]. The contribution of diffusion modes on the Rb relaxation in pure nitrogen becomes dominant and increases dramatically at pressures below 50 kPa of N 2 [51], i.e. at a pressure slightly above p max tot in the current work. This effect was also not further investigated in this work.

Recompression of Low Pressure hp Noble Gases, Equivalent Flow Rates and Storage
This work demonstrated that SEOP with mixtures containing high noble gas concentrations can produce high spin polarization. This concept may be used as a pathway to hp noble gas MRI without the need for cryogenic separation. However, the drawback of this technique is that the hp noble gases need to be recompressed after SEOP. As shown previously by Imai et al. [36], diaphragm pumps can be utilized for low pressure 129 Xe SEOP without significant depolarization. In the current work, recompression was found to maintain about 80% of the 129 Xe polarization and approximately 60% of 83 Kr polarization thus reducing P max app from 4.4% to approximately 2.6%. Further development is needed to make recompression of larger volumes routinely available. The SEOP cell used in this work has approximately 75 cm 3 volume and the 129 Xe SEOP is complete every 6 min. Assuming 80% gas transfer, SEOP with the 50% xenon mixture at 22 kPa (see Table 2) leads to 1.8 cm 3 /min hp gas (at 298 K delivery temperature) with 12.4% apparent spin polarization, P app . Similarly, SEOP with 25% krypton at 40 kPa results to an equivalent flow rate of 2 cm 3 /min hp gas with 2.6% apparent spin polarization.
The polarization and rates above have been obtained with a single 23.3 W laser (incident beam power at SEOP cell entry) and scaling of the volume should be possible by increasing laser power and SEOP cell volume. In any case the usage of multiple cells and lasers would increase the volume of hp gas per time unit. Furthermore, temporary storage of hp 129 Xe at ambient temperature has previously been successfully demonstrated by Saam and co-workers [70] as a viable alternative to cryogenic storage. Further studies are required to explore temporary storage of hp 83 Kr.

Conclusions
Cryogen free production of hp 83 Kr and hp 129 Xe for practical MRI applications is possible through stopped flow SEOP with high noble gas concentrations at low total gas pressures. Without cryogenic separation the apparent polarization (as defined in Eq. 6) was P app~1 5:5% for hp 129 Xe at a production rate of 1.8 cm 3 / min hp gas (volume at 298 K). Respectively, an apparent polarization of P app~4 :4% at a rate of 2 cm 3 /min was produced for hp 83 Kr. These results were obtained using 23.3 W of laser power (incident at the SEOP cell) and a laser linewidth of 0.25 nm. Recompression of the hp gases after SEOP is a necessary step with this technique and preliminary work resulted to P app~1 2:4% (for 129 Xe) and P app~2 :6% (for 83 Kr) after recompression.
Current theory (Eq. 2) appears to provide a reasonable qualitative description of the SEOP gas pressure dependence of the polarization although several simplifications were used in this work. Overall, the practical application of current theory would benefit if more studies and published data were available. For instance, little is known about the actual spin-rotation parameter for various gas mixtures. Further, an experimental procedure to measure the temperature distribution within the SEOP cell would be very useful. In this work, a corrected value for the rubidium density [Rb] was used for 83 Kr SEOP analysis (Eq. 8) that is 4 times higher than its predicted equilibrium value at the (externally) measured SEOP cell temperatures. A correction factor of 1.3 was used for 129 Xe SEOP analysis, although correction proved to be less important compared to 83 Kr SEOP. The rubidium density (and the pumping rate c op due to associated changes in laser penetration) also appeared to be dependent on the SEOP mixture, an effect attributed to different thermal conductivity of the various gas mixtures. Furthermore, the Rb D 1 absorption linewidth dependence upon the SEOP gas pressure at 373 K was taken into account for the hp 129 Xe data fitting (Eq. 9). The pressure dependence of the Rb D 1 transition appeared not to be relevant for 83 Kr SEOP because the D 1 linewidth at 433 K is much wider than that of the narrowed diode array laser. However, a non-linear Extrapolation using the obtained values of the fitting coefficients to pressure ranges outside the fitting range are shown by dotted lines. Results of this data analysis are listed in Table 6. doi:10.1371/journal.pone.0049927.g007 pressure broadening of the Rb D 1 linewidth was observed in all cases and this unexpected behavior warrants further study.
High SEOP temperature is needed for 83 Kr in order to increase the spin exchange rate c SE for 83 Kr and to decrease the 83 Kr relaxation rate C. The results from 83 Kr SEOP inversion recovery experiments suggest that surface relaxation is a strong contributor to C at SEOP below 200 kPa (see Appendix 2 in Supporting Information S1 for discussions). Therefore, higher 83 Kr spin polarization may be obtained through a reduction in surface to volume ratio using larger SEOP cells that reduce C and thus increase the ratio c SE = c SE zC ð Þ in Eq. 2. The technique would benefit from future development focusing on practical gas recompression units, in particular for hp 83 Kr, and on larger SEOP cell volumes to produce larger quantities of hp noble gas within a given time interval. Larger SEOP cells, that may also improve the polarization in 83 Kr SEOP, will require increased laser power. Further increased laser power density at narrow laser line widths may be particularly advantageous for SEOP with high noble gas concentrations, as demonstrated in this work. Laser line narrowing to approximately 0.25 nm provides a crucial increase in 129 Xe polarization compared to SEOP with a 2 nm laser and further narrowing would likely be helpful for 129 Xe SEOP at low pressures. Finally, the general concepts of cryogen free hp noble gas production are by no means restricted to SEOP with rubidium. SEOP with cesium vapor [59,71,72] has recently been shown to increase the 129 Xe polarization significantly compared to SEOP with rubidium [29]. The benefits of cesium vapor SEOP at low gas pressures, in particular with 83 Kr, are still unexplored. Figure S1.