Observing and preventing rubidium runaway in a direct-infusion xenon-spin hyperpolarizer optimized for high-resolution hyper-CEST (chemical exchange saturation transfer using hyperpolarized nuclei) NMR C. Witte, M. Kunth, F. Rossella, and L. Schröder Citation: The Journal of Chemical Physics 140, 084203 (2014); doi: 10.1063/1.4865944 View online: http://dx.doi.org/10.1063/1.4865944 View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/140/8?ver=pdfcov Published by the AIP Publishing Articles you may be interested in Quantitative chemical exchange saturation transfer with hyperpolarized nuclei (qHyper-CEST): Sensing xenonhost exchange dynamics and binding affinities by NMR J. Chem. Phys. 141, 194202 (2014); 10.1063/1.4901429 Analytical solution for the depolarization of hyperpolarized nuclei by chemical exchange saturation transfer between free and encapsulated xenon (HyperCEST) J. Chem. Phys. 136, 144106 (2012); 10.1063/1.3701178 Measurement of laser heating in spin exchange optical pumping by NMR diffusion sensitization gradients J. Appl. Phys. 107, 094904 (2010); 10.1063/1.3371249 Liquid hyperpolarized Xe 129 produced by phase exchange in a convection cell Appl. Phys. Lett. 85, 2429 (2004); 10.1063/1.1793350 Non-thermal nuclear magnetic resonance quantum computing using hyperpolarized xenon Appl. Phys. Lett. 79, 2480 (2001); 10.1063/1.1409279
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 152.14.136.77 On: Fri, 13 Mar 2015 08:58:35
THE JOURNAL OF CHEMICAL PHYSICS 140, 084203 (2014)
Observing and preventing rubidium runaway in a direct-infusion xenon-spin hyperpolarizer optimized for high-resolution hyper-CEST (chemical exchange saturation transfer using hyperpolarized nuclei) NMR C. Witte, M. Kunth, F. Rossella, and L. Schrödera) ERC Project BiosensorImaging, Leibniz-Institut für Molekulare Pharmakologie, Berlin, Germany
(Received 15 October 2013; accepted 4 February 2014; published online 25 February 2014) Xenon is well known to undergo host-guest interactions with proteins and synthetic molecules. As xenon can also be hyperpolarized by spin exchange optical pumping, allowing the investigation of highly dilute systems, it makes an ideal nuclear magnetic resonance probe for such host molecules. The utility of xenon as a probe can be further improved using Chemical Exchange Saturation Transfer using hyperpolarized nuclei (Hyper-CEST), but for highly accurate experiments requires a polarizer and xenon infusion system optimized for such measurements. We present the design of a hyperpolarizer and xenon infusion system specifically designed to meet the requirements of Hyper-CEST measurements. One key element of this design is preventing rubidium runaway, a chain reaction induced by laser heating that prevents efficient utilization of high photon densities. Using thermocouples positioned along the pumping cell we identify the sources of heating and conditions for rubidium runaway to occur. We then demonstrate the effectiveness of actively cooling the optical cell to prevent rubidium runaway in a compact setup. This results in a 2–3-fold higher polarization than without cooling, allowing us to achieve a polarization of 25% at continuous flow rates of 9 ml/min of 129 Xe. The simplicity of this design also allows it to be retrofitted to many existing polarizers. Combined with a direction infusion system that reduces shot-to-shot noise down to 0.56% we have captured Hyper-CEST spectra in unprecedented detail, allowing us to completely resolve peaks separated by just 1.62 ppm. Due to its high polarization and excellent stability, our design allows the comparison of underlying theories of host-guest systems with experiment at low concentrations, something extremely difficult with previous polarizers. © 2014 AIP Publishing LLC. [http://dx.doi.org/10.1063/1.4865944] I. INTRODUCTION
As xenon forms host-guest complexes it has been utilized to probe hydrophobic cavities, binding pockets, and channellike structures in proteins, using both nuclear magnetic resonance (NMR)1–3 and x-ray crystallography.4 For NMR experiments, this is often used in conjunction with spin hyperpolarization, which can increase the NMR signal from nuclei by several orders of magnitude. This facilitates the investigation of highly dilute spin systems and, as a result, the technique of hyperpolarization has been widely used in the design of NMR and MRI tracers and biosensors.5 Specifically, hyperpolarized (HP) xenon-129 has been proposed as a highly suitable NMR/MRI probe.6, 7 Its NMR signal is extremely sensitive to its molecular environment8 and it can be functionalized through host-guest interactions with synthetic molecules, such as cryptophanes6 or cucurbit-urils.9 The detection of Xe bound to host structures can be improved using Hyper-CEST7 (Chemical Exchange Saturation Transfer using hyperpolarized nuclei). This technique utilizes the reversible binding of Xe to its host molecule to effectively amplify its signal. Furthermore, as the chemical shift information is preserved, this indirect detection technique retains Xe nuclei’s in-
a) Electronic mail: [email protected]
0021-9606/2014/140(8)/084203/9/$30.00
herent high specificity alongside the significant improvements in sensitivity.10 HP Xe is produced via a technique known as spin exchange optical pumping11 (SEOP). This process transfers the easily available angular momentum of laser generated photons to a noble gas via an intermediary alkali metal. Previous experiments have observed a linear increase in Xe polarization with laser power at low cell temperatures, while for higher temperatures this trend is broken. Increasing the laser power beyond a certain amount can actually lead to a decrease in Xe polarization.12 Early publications13, 14 even suggested to abstain from high laser powers due to related unwanted drops in Xe polarization, especially when fast polarization build-up is desired. The observed reduction in Xe polarization with increasing laser power at high temperatures is believed to be due to a process known as rubidium runaway (first described by Appelt et al.14 though the term was coined later). As nitrogen is used to quench optical emissions from the rubidium, most of the energy from the absorbed photons is converted to heat. 150–300 mbar of N2 can quench ≈99% of the excited states of Rb15, 16 (which can be populated beyond the 9D energy level) and each D1 transition contributes 1.5 eV to the internal energy of the N2 . This additional heating by the laser can generate dense, highly absorptive rubidium clouds at the window where the laser beam enters the optical pumping cell. In a runaway chain reaction, the increased
140, 084203-1
© 2014 AIP Publishing LLC
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 152.14.136.77 On: Fri, 13 Mar 2015 08:58:35
084203-2
Witte et al.
absorption locally increases the temperature resulting in inhomogeneous cell illumination and, consequently, poor Xe polarization. Thus, while increasing laser power should result in greater Xe polarizations, efficiently utilizing high photon densities and achieving a stable Xe magnetization remains a challenge. As we explain below, rubidium runaway is exacerbated by optimizations we have made to the polarizer for direct infusion Hyper-CEST measurements. This makes preventing this effect of particular importance for such applications. In this paper, first we motivate and present the design of our hyperpolarizer and SEOP monitoring system. We then utilize the N2 gas-moderated heating effect to observe the onset of Rb runaway, identifying two major sources of heat production, namely, the Rb condensate layer and the injection of the N2 quench gas. We observe highly inhomogenous Rb densities and localized Rb runaway with our system and then validate the effectiveness of dissipating excess heat on the temperature conditions inside the cell. By feeding this information back into the system, we can maintain a stable temperature gradient across the cell, prevent Rb runaway, and improve the Xe polarization17 in a compact SEOP setup. Under these conditions we optimally use the high photon density to achieve a 2–3-fold higher Xe polarization when compared with polarization levels measured without active cell cooling. Combining this with improvements in techniques for the delivery of HP Xe into solution we reduce the shot-to-shot noise to less than 0.56%, enabling highly accurate Hyper-CEST measurements of host-guest systems. II. POLARIZER DESIGN AND NMR SETUP
While most polarizers are designed to maximize the volume and hyperpolarization of Xe produced, when performing Hyper-CEST experiments additional concerns are of similar or greater importance. In particular, as Hyper-CEST is an indirect detection technique, it requires at least two measurements with identical starting conditions. This means that being able to reliably deliver the same amount of Xe with the same polarization can be just as important as the absolute level of polarization achieved. While running the polarizer in batch mode facilitates higher Xe polarization (polarizations of 50%–90% for 129 Xe have been reported18–21 ) the relaxation of the reservoir of Xe becomes an issue for repeated measurements over longer periods of time. This necessitates running the polarizer in continuous flow mode, which achieves more consistent levels of polarization at the cost of Xe polarization. But even using continuous flow with direct infusion, where HP Xe travels directly from the polarizer into a high field magnet (without freezing) and is bubbled into solution, multiplicative noise from sources such as variations in the Xe polarization (due to fluctuations in gas flow or temperature of the pumping cell) or due to remaining bubbles in the sample have been reported to contribute as much as 20% uncertainty to the detected signal of indirect detection methods.22 Eliminating these sources of noise while maintaining a high polarization is a central part of our polarizer and direct infusion design. In our polarizer, a laser diode array with integrated volume holographic grating (QPC, Sylmar, CA) provides 150 W
J. Chem. Phys. 140, 084203 (2014)
continuous output at 794.8 nm in a bandwidth of < 1 nm (two peaks separated by ∼0.4 nm, see Sec. S7 in the supplementary material49 ). The laser diode array is cooled using a recirculated liquid that passes through a chiller module. The laser wavelength can be tuned by altering the set point of the chiller. A series of near-IR optical elements are used to circularly polarize the light and homogenize the beam profile to avoid intensity modulation due to individual diode bars as described by Young et al.23 Briefly this consists of a fiber, fiber collimator, aperture, polarizing beam splitter cube, zero order quarter wave plate, and a final beam expander to ensure that the entire optical cell is illuminated. The light rejected by the beam splitter cube (approximately 30% of the total laser output) is sent into a beam dump. The approximately cylindrical optical pumping cell, with a total length of 200 mm and outer diameter of 44 mm, has an inner volume of 214 cm3 and resides in an oven (see Fig. 1). Three magnetic field coils enclose the oven and generate a magnetic field of 26 G to break the degeneracy of the Rb electron energy levels and allow selective optical pumping.24 A silicone heating pad, attached to the outside of the cell, sits below and heats a rubidium droplet inside the cell. This is monitored by a thermocouple and fed back into a controller to adjust the temperature of the heating pad. A gas mixture, containing either 2% Xe (2% Xe/10% N2 /88% He) or 5% Xe (5% Xe/10% N2 /85% He) flows counter to the direction of propagation of the laser light, exiting the optical cell close to the entrance of the laser light. This ensures the photon density is highest where the Xe exits the optical cell. Such antiparallel flow conditions have been described as favorable over parallel flow.25 The gases are pre-mixed to allow easy switching between two storage tanks. Mass flow controllers and pressure sensors continuously monitor the condition of
FIG. 1. Schematic of the optical pumping cell. Circularly polarized 794.8 nm laser light illuminates the cell from the left. A silicone heating strip is attached to the pumping cell below a droplet of rubidium. Heating the droplet generates a vapor of rubidium. The transmitted light passes through a 100 μm diameter pin hole and absorption disk (optical density 5) and is reflected into an optical fiber. A near-infrared spectrometer monitors the transmission of laser light though the cell. The cell is contained within an oven that is divided into two regions, hot and cold. Room temperature air is blown into the cold side of the cell to maintain the front of the optical cell at a reduced temperature. Three thermocouples (near the front, center, and rear of the cell, with regards to the laser entrance window) monitor the temperatures throughout the cell.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 152.14.136.77 On: Fri, 13 Mar 2015 08:58:35
084203-3
Witte et al.
the polarizer. Additionally, three thermocouples monitor the temperature in different areas of the pumping cell, as can be seen in Fig. 1. The laser light that is transmitted through the cell is monitored by a near-IR spectrometer (HR4000 Ocean Optics, Dunedin, FL, USA). The cell is kept at an absolute pressure of approximately 4.7 bars to pressure broaden the rubidium absorption line allowing efficient optical pumping. Due to the reduced residence time of the Xe in the pumping cell under continuous flow conditions (typically less than 120 s), it is preferable to run at higher temperatures (≈150 ◦ C) than would be optimal for batch mode production. This increases the rubidium density and hence the spin exchange rate between rubidium and Xe.14, 17, 26 Under these conditions the Xe polarization buildup rate is increased,20 as characterized, e.g., by the time constant for Rb-Xe spin exchange being only ≈20 s in a high pressure system at 150 ◦ C.13 Additionally, the polarization of the rubidium should be maintained as high as possible, which under high rubidium density requires increased laser power. These two requirements, increased temperatures and laser power, combine to make Rb runaway much more likely. It has been reported by Nikolaou et al.19 that under certain conditions higher laser power (32 W vs. 27 W) is of no benefit at all and no temperature can be found where it produces better Xe NMR signal compared to lower laser power (Fig. 4 by Nikolau et al.19 ). Hence, to benefit from higher laser power, working with a more or less uniform cell temperature is not an option. If the chain reaction is not suppressed, theoretical predictions for the benefit of high temperatures become invalid.27 One method of preventing rubidium runaway is to presaturate the gas mixture with rubidium before it enters the optical pumping region. With no further source of rubidium in the area illuminated by the laser, rubidium runaway is less likely to occur.28 However, such a system is relative bulky and has no real means to monitor the potential onset of the runaway effect. Previous work17 has used some similar cooling but did not investigate the correlations between temperature in different cell areas and the details of adjustable inhomogeneous cell temperatures. As depicted in Fig. 1 we divide the oven into a hot and a cold region. The hot side of the oven is insulated with fiberglass. The temperature of the front of the cell is monitored using a thermocouple. This is used to control the flow of cold (room temperature) air being pumped into the cold side of the oven. Using this we can maintain the front of the cell at a reduced temperature to the rear of the cell and prevent Rb runaway. Rather than monitoring the Xe polarization using in situ low-field NMR17, 20 we observe the Xe polarization as it arrives in the NMR spectrometer. The gas continuously flows from the polarizer and through a high pressure 10 mm NMR tube. The NMR tube resides in a dual 129 Xe/1 H 10 mm coil in a Bruker wide bore AV 400 spectrometer. This setup introduces a slight delay (the travel time from the polarizer to the NMR spectrometer is less than 1 min) between changes in the condition of the pumping cell and recording of the Xe polarization but otherwise allows us to observe the polarizer running as intended for direct infusion, without altering the conditions in the pumping cell.
J. Chem. Phys. 140, 084203 (2014)
III. OBSERVING AND PREVENTING RUBIDIUM RUNAWAY
Walter et al.15 stated that the flow of energy is poorly understood in SEOP and investigated it on an atomic/molecular scale by N2 Raman spectroscopy. Though thermocouples on the glass surface are inappropriate to study internal temperatures in all detail as by Walter et al.15 and Newton et al.,39 we demonstrate they are a simple measure to reveal temperature variation on the macroscopic scale that contains a surprising amount of information to analyze the overall SEOP performance of a cell. We first identify different contributions to the heating of the optical cell, observing Rb runaway occurring in the center and rear of the cell. We then identify conditions for Rb runaway to occur at the front of the cell. Next, we investigate how dissipating the heat can improve the performance of the polarizer.
A. Identifying heating sources
To illustrate the significant role of N2 in the heating process, we first carefully monitored temperature developments in the three cell sections (front, center, and rear) during ramping up the polarizer with a pure argon atmosphere in the cell. This suppresses heat conversion initiated by the N2 quench gas which was added later. Whereas some studies17 report laser light absorption only for cell temperatures > 100 ◦ C, we found that especially for “aged” cells the IR light is absorbed relatively early during ramping up and causes substantial heating of the cell. This is illustrated by the data in Fig. 2. Switching on the laser (P ≈ 130 W) causes immediate heating of the rear part of the cell with the center and front temperature following in a less pronounced way. It is noteworthy that the rear thermocouple registers almost 100 ◦ C even before the silicone heating pad is turned on. Due to the pronounced difference to the other two temperature curves, we assign this heating to the interaction of a film of condensed Rb on the rear glass window with IR photons of direct incidence; the initial heating induced by the laser burns off the alkali metal layer coating this part of the cell leading to more heat production. Such Rb films have also been described by Fink and Brunner.25 However, the optical density (OD) at this point before activation of the heater is still relatively low as shown by laser spectrum 1 (LS1) in Fig. 2(b). This spectrum serves as a reference for further OD tests (note: the conditions of a used cell make it impossible to register a reference spectrum for a completely cold and transparent cell. Since the laser needs ∼10 min to stabilize, Rb burn-off is already happening and some Rb vapor is present when the first stable laser spectrum can be taken). Activating the silicone heater at approximately t = 17 min with a set temperature of 195 ◦ C causes the temperatures in the center and rear part to rise faster (as can be seen by the derivatives of the temperature curves in Sec. S1 in the supplementary material49 ). In the next step, we replaced the Ar by injecting the 2% Xe mix at t ≈ 20:30 min with the gas flow set to 0.7 SLM for the remainder of the experiment. The thermocouple next to the gas inlet (rear sensor) shows an immediate strong response, indicating the conversion of photon
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 152.14.136.77 On: Fri, 13 Mar 2015 08:58:35
Witte et al. 250
14000
intensity count
12000
LS1 LS2 LS3 LS4
15 20 time t [min]
16:30 24:30 30:30 30:35
10000 8000 6000 4000 2000 0 794.0 794.5 795.0 795.5 wavelength λ [nm]
25
35
0.4
0
10
20
30
40
50
60
70
80
0.2
0 90
time t [min]
FIG. 3. Relative integrated Xe signal and temperatures at the front, center, and rear of the optical pumping cell as we first induce and then recover from rubidium runaway. After the system reaches steady state with the laser operating at 100 W, at time 0 min the laser power is increased to 150 W. Two significant time points are indicated in the plot: at 69 min the cooling system is enabled with a set point of 120 ◦ C and at 80 min the cooling set point is increased to 140 ◦ C.
100 80 60 40 20 0
100
0
30
(c)
rel. absorption [%]
(b)
10
0.6
50
0 5
150
Cooling setpoint 140°C
rear center front
0.8
Cooling setpoint 120°C
50
temperature T [°C]
100
0
1 rear center front Xe signal
200
pumping field on
150
cell heater on
temperature T [°C]
200
Xe signal [a.u.]
laser on, pure Ar atmosphere
(a)
J. Chem. Phys. 140, 084203 (2014)
injection of (He/Xe/N2) mix
084203-4
LS2 24:30 LS3 30:30 LS4 30:35
794.0 794.5 795.0 795.5 wavelength λ [nm]
FIG. 2. (a) Monitoring temperatures throughout the cell during ramp up of the polarizer initially under an argon atmosphere. Significant time points as indicated on the plot are: at t ≈ 6:30 min the laser is turned on, at t ≈ 17:00 min the cell heater is turned on, at t ≈ 20:30 min the injection of the Xe gas mix begins, and at t ≈ 30:30 min the pumping field is turned on. (b) Laser spectra taken at time points: LS1 t ≈ 16:30 min, LS2 t ≈ 24:30 min, LS3 t ≈ 30:30 min, and LS4 t ≈ 30:35. (c) Absorption of laser spectrum LS2, LS3, and LS4 relative to LS1 as a function of wavelength.
energy into heat by the quench gas. The thermocouples register a propagating heating effect initiated by N2 , starting at the rear sensor and eventually reaching the front sensor, this can be more clearly seen in the derivatives of the temperature plots shown in Sec. S1 in the supplementary material.49 At this point, the pumping field Bpump is still turned off to allow for high photon absorption due to the degenerate Rb energy levels. Although this is an unrealistic scenario for the normal operation mode, it allows easy identification of the laser-induced heating effects due to higher photon absorption. It is also useful to bring the cell faster to target temperature. As the dense Rb cloud resides at the center, the rear sensor observes a decrease in temperature. Several correlations between changes in the three temperature curves can be more clearly seen in the derivatives of the temperatures in Sec. S1 in the supplementary material49 and illustrate further evolution of the effects on a macroscopic scale. Approximately 4 min after starting the Xe mix flow the rear already experiences poor illumination as shown by LS2 (t ≈ 24:30) which demonstrates that up to 95% of the photons are absorbed (Fig. 2(c)). Two minutes later (t ≈ 26:30 min) this situation is aggravated as the rear cools down further and the center shows another burst of temperature increase before reaching
values of T > 200 ◦ C. Pronounced signs of Rb runaway in the center now cause poor illumination of the rear part of the cell, demonstrated by LS3 with even higher absorption in the cell. In order to damp the runaway effect, Bpump is turned on at t ≈ 30:30 min to allow only for selected transitions within the Rb electron energy levels. This causes an immediate change in the absorption profile calculated from LS4 with 65% spin polarized xenon-129 for NMR spectroscopy and imaging,” J. Magn. Reson. 159(2), 175–182 (2002). 33 H. Imai, J. Fukutomi, A. Kimura, and H. Fujiwara, “Effect of reduced pressure on the polarization of 129 Xe in the production of hyperpolarized 129 Xe gas: Development of a simple continuous flow mode hyperpolarizing system working at pressures as low as 0.15 atm,” Concepts Magn. Reson. 33B(3), 192–200 (2008). 34 G. Schrank, Z. Ma, A. Schoeck, and B. Saam, “Characterization of a lowpressure high-capacity 129 Xe flow-through polarizer,” Phys. Rev. A 80(6), 063424 (2009). 35 S. Klippel, J. Jayapaul, M. Kunth, F. Rossella, M. Schnurr, C. Witte, C. Freund, and L. Schröder, “Cell tracking with caged xenon: Using crypto-
J. Chem. Phys. 140, 084203 (2014) phanes as MRI reporters upon cellular internalization,” Angew. Chem., Int. Ed. 126(2), 503–506 (2014). 36 D. Baumer, E. Brunner, P. Blümler, P. P. Zänker, and H. W. Spiess, “NMR spectroscopy of laser-polarized 129 Xe under continuous flow: A method to study aqueous solutions of biomolecules,” Angew. Chem., Int. Ed. 45(43), 7282–7284 (2006). 37 N. Amor, P. Zänker, P. Blümler, F. Meise, L. Schreiber, A. Scholz, J. Schmiedeskamp, H. Spiess, and K. Münnemann, “Magnetic resonance imaging of dissolved hyperpolarized 129 Xe using a membrane-based continuous flow system,” J. Magn. Reson. 201(1), 93–99 (2009). 38 C. Hilty, T. J. Lowery, D. E. Wemmer, and A. Pines, “Spectrally resolved magnetic resonance imaging of a xenon biosensor,” Angew. Chem., Int. Ed. 45(1), 70–73 (2006). 39 H. Newton, L. L. Walkup, N. Whiting, L. West, J. Carriere, F. Havermeyer, L. Ho, P. Morris, B. M. Goodson, and M. J. Barlow, “Comparative study of in situ N2 rotational Raman spectroscopy methods for probing energy thermalization processes during spin-exchange optical pumping,” Appl. Phys. B (published online). 40 A. Fink, D. Baumer, and E. Brunner, “Production of hyperpolarized xenon in a static pump cell: Numerical simulations and experiments,” Phys. Rev. A 72(5), 053411 (2005). 41 L. Schröder, “Xenon for NMR biosensing: Inert but alert,” Phys. Med. 29(1), 3–16 (2013). 42 L. Schröder, T. Meldrum, M. Smith, T. J. Lowery, D. E. Wemmer, and A. Pines, “Temperature response of 129 Xe depolarization transfer and its application for ultrasensitive NMR detection,” Phys. Rev. Lett. 100, 257603 (2008). 43 L. Schröder, L. Chavez, T. Meldrum, M. Smith, T. J. Lowery, D. E. Wemmer, and A. Pines, “Temperature-controlled molecular depolarization gates in nuclear magnetic resonance,” Angew. Chem., Int. Ed. 47(23), 4316–4320 (2008). 44 F. Schilling, L. Schröder, K. K. Palaniappan, S. Zapf, D. E. Wemmer, and A. Pines, “MRI thermometry based on encapsulated hyperpolarized xenon,” ChemPhysChem 11(16), 3529–3533 (2010). 45 M. Zaiss, M. Schnurr, and P. Bachert, “Analytical solution for the depolarization of hyperpolarized nuclei by chemical exchange saturation transfer between free and encapsulated xenon (HyperCEST),” J. Chem. Phys. 136(14), 144106 (2012). 46 M. Zaiss and P. Bachert, “Exchange-dependent relaxation in the rotating frame for slow and intermediate exchange – modeling off-resonant spinlock and chemical exchange saturation transfer,” NMR Biomed. 26(5), 507–518 (2013). 47 J. Sloniec, M. Schnurr, C. Witte, U. Resch-Genger, L. Schröder, and A. Hennig, “Biomembrane interactions of functionalized cryptophane-A: Combined fluorescence and 129 Xe NMR studies of a bimodal contrast agent,” Chem. - Eur. J. 19(9), 3110–3118 (2013). 48 M. Schnurr, C. Witte, and L. Schröder, “Functionalized 129 Xe as a potential biosensor for membrane fluidity,” Phys. Chem. Chem. Phys. 15, 14178– 14181 (2013). 49 See supplementary material at http://dx.doi.org/10.1063/1.4865944 for additional figures and data.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 152.14.136.77 On: Fri, 13 Mar 2015 08:58:35
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 152.14.136.77 On: Fri, 13 Mar 2015 08:58:35
THE JOURNAL OF CHEMICAL PHYSICS 140, 084203 (2014)
Observing and preventing rubidium runaway in a direct-infusion xenon-spin hyperpolarizer optimized for high-resolution hyper-CEST (chemical exchange saturation transfer using hyperpolarized nuclei) NMR C. Witte, M. Kunth, F. Rossella, and L. Schrödera) ERC Project BiosensorImaging, Leibniz-Institut für Molekulare Pharmakologie, Berlin, Germany
(Received 15 October 2013; accepted 4 February 2014; published online 25 February 2014) Xenon is well known to undergo host-guest interactions with proteins and synthetic molecules. As xenon can also be hyperpolarized by spin exchange optical pumping, allowing the investigation of highly dilute systems, it makes an ideal nuclear magnetic resonance probe for such host molecules. The utility of xenon as a probe can be further improved using Chemical Exchange Saturation Transfer using hyperpolarized nuclei (Hyper-CEST), but for highly accurate experiments requires a polarizer and xenon infusion system optimized for such measurements. We present the design of a hyperpolarizer and xenon infusion system specifically designed to meet the requirements of Hyper-CEST measurements. One key element of this design is preventing rubidium runaway, a chain reaction induced by laser heating that prevents efficient utilization of high photon densities. Using thermocouples positioned along the pumping cell we identify the sources of heating and conditions for rubidium runaway to occur. We then demonstrate the effectiveness of actively cooling the optical cell to prevent rubidium runaway in a compact setup. This results in a 2–3-fold higher polarization than without cooling, allowing us to achieve a polarization of 25% at continuous flow rates of 9 ml/min of 129 Xe. The simplicity of this design also allows it to be retrofitted to many existing polarizers. Combined with a direction infusion system that reduces shot-to-shot noise down to 0.56% we have captured Hyper-CEST spectra in unprecedented detail, allowing us to completely resolve peaks separated by just 1.62 ppm. Due to its high polarization and excellent stability, our design allows the comparison of underlying theories of host-guest systems with experiment at low concentrations, something extremely difficult with previous polarizers. © 2014 AIP Publishing LLC. [http://dx.doi.org/10.1063/1.4865944] I. INTRODUCTION
As xenon forms host-guest complexes it has been utilized to probe hydrophobic cavities, binding pockets, and channellike structures in proteins, using both nuclear magnetic resonance (NMR)1–3 and x-ray crystallography.4 For NMR experiments, this is often used in conjunction with spin hyperpolarization, which can increase the NMR signal from nuclei by several orders of magnitude. This facilitates the investigation of highly dilute spin systems and, as a result, the technique of hyperpolarization has been widely used in the design of NMR and MRI tracers and biosensors.5 Specifically, hyperpolarized (HP) xenon-129 has been proposed as a highly suitable NMR/MRI probe.6, 7 Its NMR signal is extremely sensitive to its molecular environment8 and it can be functionalized through host-guest interactions with synthetic molecules, such as cryptophanes6 or cucurbit-urils.9 The detection of Xe bound to host structures can be improved using Hyper-CEST7 (Chemical Exchange Saturation Transfer using hyperpolarized nuclei). This technique utilizes the reversible binding of Xe to its host molecule to effectively amplify its signal. Furthermore, as the chemical shift information is preserved, this indirect detection technique retains Xe nuclei’s in-
a) Electronic mail: [email protected]
0021-9606/2014/140(8)/084203/9/$30.00
herent high specificity alongside the significant improvements in sensitivity.10 HP Xe is produced via a technique known as spin exchange optical pumping11 (SEOP). This process transfers the easily available angular momentum of laser generated photons to a noble gas via an intermediary alkali metal. Previous experiments have observed a linear increase in Xe polarization with laser power at low cell temperatures, while for higher temperatures this trend is broken. Increasing the laser power beyond a certain amount can actually lead to a decrease in Xe polarization.12 Early publications13, 14 even suggested to abstain from high laser powers due to related unwanted drops in Xe polarization, especially when fast polarization build-up is desired. The observed reduction in Xe polarization with increasing laser power at high temperatures is believed to be due to a process known as rubidium runaway (first described by Appelt et al.14 though the term was coined later). As nitrogen is used to quench optical emissions from the rubidium, most of the energy from the absorbed photons is converted to heat. 150–300 mbar of N2 can quench ≈99% of the excited states of Rb15, 16 (which can be populated beyond the 9D energy level) and each D1 transition contributes 1.5 eV to the internal energy of the N2 . This additional heating by the laser can generate dense, highly absorptive rubidium clouds at the window where the laser beam enters the optical pumping cell. In a runaway chain reaction, the increased
140, 084203-1
© 2014 AIP Publishing LLC
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 152.14.136.77 On: Fri, 13 Mar 2015 08:58:35
084203-2
Witte et al.
absorption locally increases the temperature resulting in inhomogeneous cell illumination and, consequently, poor Xe polarization. Thus, while increasing laser power should result in greater Xe polarizations, efficiently utilizing high photon densities and achieving a stable Xe magnetization remains a challenge. As we explain below, rubidium runaway is exacerbated by optimizations we have made to the polarizer for direct infusion Hyper-CEST measurements. This makes preventing this effect of particular importance for such applications. In this paper, first we motivate and present the design of our hyperpolarizer and SEOP monitoring system. We then utilize the N2 gas-moderated heating effect to observe the onset of Rb runaway, identifying two major sources of heat production, namely, the Rb condensate layer and the injection of the N2 quench gas. We observe highly inhomogenous Rb densities and localized Rb runaway with our system and then validate the effectiveness of dissipating excess heat on the temperature conditions inside the cell. By feeding this information back into the system, we can maintain a stable temperature gradient across the cell, prevent Rb runaway, and improve the Xe polarization17 in a compact SEOP setup. Under these conditions we optimally use the high photon density to achieve a 2–3-fold higher Xe polarization when compared with polarization levels measured without active cell cooling. Combining this with improvements in techniques for the delivery of HP Xe into solution we reduce the shot-to-shot noise to less than 0.56%, enabling highly accurate Hyper-CEST measurements of host-guest systems. II. POLARIZER DESIGN AND NMR SETUP
While most polarizers are designed to maximize the volume and hyperpolarization of Xe produced, when performing Hyper-CEST experiments additional concerns are of similar or greater importance. In particular, as Hyper-CEST is an indirect detection technique, it requires at least two measurements with identical starting conditions. This means that being able to reliably deliver the same amount of Xe with the same polarization can be just as important as the absolute level of polarization achieved. While running the polarizer in batch mode facilitates higher Xe polarization (polarizations of 50%–90% for 129 Xe have been reported18–21 ) the relaxation of the reservoir of Xe becomes an issue for repeated measurements over longer periods of time. This necessitates running the polarizer in continuous flow mode, which achieves more consistent levels of polarization at the cost of Xe polarization. But even using continuous flow with direct infusion, where HP Xe travels directly from the polarizer into a high field magnet (without freezing) and is bubbled into solution, multiplicative noise from sources such as variations in the Xe polarization (due to fluctuations in gas flow or temperature of the pumping cell) or due to remaining bubbles in the sample have been reported to contribute as much as 20% uncertainty to the detected signal of indirect detection methods.22 Eliminating these sources of noise while maintaining a high polarization is a central part of our polarizer and direct infusion design. In our polarizer, a laser diode array with integrated volume holographic grating (QPC, Sylmar, CA) provides 150 W
J. Chem. Phys. 140, 084203 (2014)
continuous output at 794.8 nm in a bandwidth of < 1 nm (two peaks separated by ∼0.4 nm, see Sec. S7 in the supplementary material49 ). The laser diode array is cooled using a recirculated liquid that passes through a chiller module. The laser wavelength can be tuned by altering the set point of the chiller. A series of near-IR optical elements are used to circularly polarize the light and homogenize the beam profile to avoid intensity modulation due to individual diode bars as described by Young et al.23 Briefly this consists of a fiber, fiber collimator, aperture, polarizing beam splitter cube, zero order quarter wave plate, and a final beam expander to ensure that the entire optical cell is illuminated. The light rejected by the beam splitter cube (approximately 30% of the total laser output) is sent into a beam dump. The approximately cylindrical optical pumping cell, with a total length of 200 mm and outer diameter of 44 mm, has an inner volume of 214 cm3 and resides in an oven (see Fig. 1). Three magnetic field coils enclose the oven and generate a magnetic field of 26 G to break the degeneracy of the Rb electron energy levels and allow selective optical pumping.24 A silicone heating pad, attached to the outside of the cell, sits below and heats a rubidium droplet inside the cell. This is monitored by a thermocouple and fed back into a controller to adjust the temperature of the heating pad. A gas mixture, containing either 2% Xe (2% Xe/10% N2 /88% He) or 5% Xe (5% Xe/10% N2 /85% He) flows counter to the direction of propagation of the laser light, exiting the optical cell close to the entrance of the laser light. This ensures the photon density is highest where the Xe exits the optical cell. Such antiparallel flow conditions have been described as favorable over parallel flow.25 The gases are pre-mixed to allow easy switching between two storage tanks. Mass flow controllers and pressure sensors continuously monitor the condition of
FIG. 1. Schematic of the optical pumping cell. Circularly polarized 794.8 nm laser light illuminates the cell from the left. A silicone heating strip is attached to the pumping cell below a droplet of rubidium. Heating the droplet generates a vapor of rubidium. The transmitted light passes through a 100 μm diameter pin hole and absorption disk (optical density 5) and is reflected into an optical fiber. A near-infrared spectrometer monitors the transmission of laser light though the cell. The cell is contained within an oven that is divided into two regions, hot and cold. Room temperature air is blown into the cold side of the cell to maintain the front of the optical cell at a reduced temperature. Three thermocouples (near the front, center, and rear of the cell, with regards to the laser entrance window) monitor the temperatures throughout the cell.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 152.14.136.77 On: Fri, 13 Mar 2015 08:58:35
084203-3
Witte et al.
the polarizer. Additionally, three thermocouples monitor the temperature in different areas of the pumping cell, as can be seen in Fig. 1. The laser light that is transmitted through the cell is monitored by a near-IR spectrometer (HR4000 Ocean Optics, Dunedin, FL, USA). The cell is kept at an absolute pressure of approximately 4.7 bars to pressure broaden the rubidium absorption line allowing efficient optical pumping. Due to the reduced residence time of the Xe in the pumping cell under continuous flow conditions (typically less than 120 s), it is preferable to run at higher temperatures (≈150 ◦ C) than would be optimal for batch mode production. This increases the rubidium density and hence the spin exchange rate between rubidium and Xe.14, 17, 26 Under these conditions the Xe polarization buildup rate is increased,20 as characterized, e.g., by the time constant for Rb-Xe spin exchange being only ≈20 s in a high pressure system at 150 ◦ C.13 Additionally, the polarization of the rubidium should be maintained as high as possible, which under high rubidium density requires increased laser power. These two requirements, increased temperatures and laser power, combine to make Rb runaway much more likely. It has been reported by Nikolaou et al.19 that under certain conditions higher laser power (32 W vs. 27 W) is of no benefit at all and no temperature can be found where it produces better Xe NMR signal compared to lower laser power (Fig. 4 by Nikolau et al.19 ). Hence, to benefit from higher laser power, working with a more or less uniform cell temperature is not an option. If the chain reaction is not suppressed, theoretical predictions for the benefit of high temperatures become invalid.27 One method of preventing rubidium runaway is to presaturate the gas mixture with rubidium before it enters the optical pumping region. With no further source of rubidium in the area illuminated by the laser, rubidium runaway is less likely to occur.28 However, such a system is relative bulky and has no real means to monitor the potential onset of the runaway effect. Previous work17 has used some similar cooling but did not investigate the correlations between temperature in different cell areas and the details of adjustable inhomogeneous cell temperatures. As depicted in Fig. 1 we divide the oven into a hot and a cold region. The hot side of the oven is insulated with fiberglass. The temperature of the front of the cell is monitored using a thermocouple. This is used to control the flow of cold (room temperature) air being pumped into the cold side of the oven. Using this we can maintain the front of the cell at a reduced temperature to the rear of the cell and prevent Rb runaway. Rather than monitoring the Xe polarization using in situ low-field NMR17, 20 we observe the Xe polarization as it arrives in the NMR spectrometer. The gas continuously flows from the polarizer and through a high pressure 10 mm NMR tube. The NMR tube resides in a dual 129 Xe/1 H 10 mm coil in a Bruker wide bore AV 400 spectrometer. This setup introduces a slight delay (the travel time from the polarizer to the NMR spectrometer is less than 1 min) between changes in the condition of the pumping cell and recording of the Xe polarization but otherwise allows us to observe the polarizer running as intended for direct infusion, without altering the conditions in the pumping cell.
J. Chem. Phys. 140, 084203 (2014)
III. OBSERVING AND PREVENTING RUBIDIUM RUNAWAY
Walter et al.15 stated that the flow of energy is poorly understood in SEOP and investigated it on an atomic/molecular scale by N2 Raman spectroscopy. Though thermocouples on the glass surface are inappropriate to study internal temperatures in all detail as by Walter et al.15 and Newton et al.,39 we demonstrate they are a simple measure to reveal temperature variation on the macroscopic scale that contains a surprising amount of information to analyze the overall SEOP performance of a cell. We first identify different contributions to the heating of the optical cell, observing Rb runaway occurring in the center and rear of the cell. We then identify conditions for Rb runaway to occur at the front of the cell. Next, we investigate how dissipating the heat can improve the performance of the polarizer.
A. Identifying heating sources
To illustrate the significant role of N2 in the heating process, we first carefully monitored temperature developments in the three cell sections (front, center, and rear) during ramping up the polarizer with a pure argon atmosphere in the cell. This suppresses heat conversion initiated by the N2 quench gas which was added later. Whereas some studies17 report laser light absorption only for cell temperatures > 100 ◦ C, we found that especially for “aged” cells the IR light is absorbed relatively early during ramping up and causes substantial heating of the cell. This is illustrated by the data in Fig. 2. Switching on the laser (P ≈ 130 W) causes immediate heating of the rear part of the cell with the center and front temperature following in a less pronounced way. It is noteworthy that the rear thermocouple registers almost 100 ◦ C even before the silicone heating pad is turned on. Due to the pronounced difference to the other two temperature curves, we assign this heating to the interaction of a film of condensed Rb on the rear glass window with IR photons of direct incidence; the initial heating induced by the laser burns off the alkali metal layer coating this part of the cell leading to more heat production. Such Rb films have also been described by Fink and Brunner.25 However, the optical density (OD) at this point before activation of the heater is still relatively low as shown by laser spectrum 1 (LS1) in Fig. 2(b). This spectrum serves as a reference for further OD tests (note: the conditions of a used cell make it impossible to register a reference spectrum for a completely cold and transparent cell. Since the laser needs ∼10 min to stabilize, Rb burn-off is already happening and some Rb vapor is present when the first stable laser spectrum can be taken). Activating the silicone heater at approximately t = 17 min with a set temperature of 195 ◦ C causes the temperatures in the center and rear part to rise faster (as can be seen by the derivatives of the temperature curves in Sec. S1 in the supplementary material49 ). In the next step, we replaced the Ar by injecting the 2% Xe mix at t ≈ 20:30 min with the gas flow set to 0.7 SLM for the remainder of the experiment. The thermocouple next to the gas inlet (rear sensor) shows an immediate strong response, indicating the conversion of photon
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 152.14.136.77 On: Fri, 13 Mar 2015 08:58:35
Witte et al. 250
14000
intensity count
12000
LS1 LS2 LS3 LS4
15 20 time t [min]
16:30 24:30 30:30 30:35
10000 8000 6000 4000 2000 0 794.0 794.5 795.0 795.5 wavelength λ [nm]
25
35
0.4
0
10
20
30
40
50
60
70
80
0.2
0 90
time t [min]
FIG. 3. Relative integrated Xe signal and temperatures at the front, center, and rear of the optical pumping cell as we first induce and then recover from rubidium runaway. After the system reaches steady state with the laser operating at 100 W, at time 0 min the laser power is increased to 150 W. Two significant time points are indicated in the plot: at 69 min the cooling system is enabled with a set point of 120 ◦ C and at 80 min the cooling set point is increased to 140 ◦ C.
100 80 60 40 20 0
100
0
30
(c)
rel. absorption [%]
(b)
10
0.6
50
0 5
150
Cooling setpoint 140°C
rear center front
0.8
Cooling setpoint 120°C
50
temperature T [°C]
100
0
1 rear center front Xe signal
200
pumping field on
150
cell heater on
temperature T [°C]
200
Xe signal [a.u.]
laser on, pure Ar atmosphere
(a)
J. Chem. Phys. 140, 084203 (2014)
injection of (He/Xe/N2) mix
084203-4
LS2 24:30 LS3 30:30 LS4 30:35
794.0 794.5 795.0 795.5 wavelength λ [nm]
FIG. 2. (a) Monitoring temperatures throughout the cell during ramp up of the polarizer initially under an argon atmosphere. Significant time points as indicated on the plot are: at t ≈ 6:30 min the laser is turned on, at t ≈ 17:00 min the cell heater is turned on, at t ≈ 20:30 min the injection of the Xe gas mix begins, and at t ≈ 30:30 min the pumping field is turned on. (b) Laser spectra taken at time points: LS1 t ≈ 16:30 min, LS2 t ≈ 24:30 min, LS3 t ≈ 30:30 min, and LS4 t ≈ 30:35. (c) Absorption of laser spectrum LS2, LS3, and LS4 relative to LS1 as a function of wavelength.
energy into heat by the quench gas. The thermocouples register a propagating heating effect initiated by N2 , starting at the rear sensor and eventually reaching the front sensor, this can be more clearly seen in the derivatives of the temperature plots shown in Sec. S1 in the supplementary material.49 At this point, the pumping field Bpump is still turned off to allow for high photon absorption due to the degenerate Rb energy levels. Although this is an unrealistic scenario for the normal operation mode, it allows easy identification of the laser-induced heating effects due to higher photon absorption. It is also useful to bring the cell faster to target temperature. As the dense Rb cloud resides at the center, the rear sensor observes a decrease in temperature. Several correlations between changes in the three temperature curves can be more clearly seen in the derivatives of the temperatures in Sec. S1 in the supplementary material49 and illustrate further evolution of the effects on a macroscopic scale. Approximately 4 min after starting the Xe mix flow the rear already experiences poor illumination as shown by LS2 (t ≈ 24:30) which demonstrates that up to 95% of the photons are absorbed (Fig. 2(c)). Two minutes later (t ≈ 26:30 min) this situation is aggravated as the rear cools down further and the center shows another burst of temperature increase before reaching
values of T > 200 ◦ C. Pronounced signs of Rb runaway in the center now cause poor illumination of the rear part of the cell, demonstrated by LS3 with even higher absorption in the cell. In order to damp the runaway effect, Bpump is turned on at t ≈ 30:30 min to allow only for selected transitions within the Rb electron energy levels. This causes an immediate change in the absorption profile calculated from LS4 with 65% spin polarized xenon-129 for NMR spectroscopy and imaging,” J. Magn. Reson. 159(2), 175–182 (2002). 33 H. Imai, J. Fukutomi, A. Kimura, and H. Fujiwara, “Effect of reduced pressure on the polarization of 129 Xe in the production of hyperpolarized 129 Xe gas: Development of a simple continuous flow mode hyperpolarizing system working at pressures as low as 0.15 atm,” Concepts Magn. Reson. 33B(3), 192–200 (2008). 34 G. Schrank, Z. Ma, A. Schoeck, and B. Saam, “Characterization of a lowpressure high-capacity 129 Xe flow-through polarizer,” Phys. Rev. A 80(6), 063424 (2009). 35 S. Klippel, J. Jayapaul, M. Kunth, F. Rossella, M. Schnurr, C. Witte, C. Freund, and L. Schröder, “Cell tracking with caged xenon: Using crypto-
J. Chem. Phys. 140, 084203 (2014) phanes as MRI reporters upon cellular internalization,” Angew. Chem., Int. Ed. 126(2), 503–506 (2014). 36 D. Baumer, E. Brunner, P. Blümler, P. P. Zänker, and H. W. Spiess, “NMR spectroscopy of laser-polarized 129 Xe under continuous flow: A method to study aqueous solutions of biomolecules,” Angew. Chem., Int. Ed. 45(43), 7282–7284 (2006). 37 N. Amor, P. Zänker, P. Blümler, F. Meise, L. Schreiber, A. Scholz, J. Schmiedeskamp, H. Spiess, and K. Münnemann, “Magnetic resonance imaging of dissolved hyperpolarized 129 Xe using a membrane-based continuous flow system,” J. Magn. Reson. 201(1), 93–99 (2009). 38 C. Hilty, T. J. Lowery, D. E. Wemmer, and A. Pines, “Spectrally resolved magnetic resonance imaging of a xenon biosensor,” Angew. Chem., Int. Ed. 45(1), 70–73 (2006). 39 H. Newton, L. L. Walkup, N. Whiting, L. West, J. Carriere, F. Havermeyer, L. Ho, P. Morris, B. M. Goodson, and M. J. Barlow, “Comparative study of in situ N2 rotational Raman spectroscopy methods for probing energy thermalization processes during spin-exchange optical pumping,” Appl. Phys. B (published online). 40 A. Fink, D. Baumer, and E. Brunner, “Production of hyperpolarized xenon in a static pump cell: Numerical simulations and experiments,” Phys. Rev. A 72(5), 053411 (2005). 41 L. Schröder, “Xenon for NMR biosensing: Inert but alert,” Phys. Med. 29(1), 3–16 (2013). 42 L. Schröder, T. Meldrum, M. Smith, T. J. Lowery, D. E. Wemmer, and A. Pines, “Temperature response of 129 Xe depolarization transfer and its application for ultrasensitive NMR detection,” Phys. Rev. Lett. 100, 257603 (2008). 43 L. Schröder, L. Chavez, T. Meldrum, M. Smith, T. J. Lowery, D. E. Wemmer, and A. Pines, “Temperature-controlled molecular depolarization gates in nuclear magnetic resonance,” Angew. Chem., Int. Ed. 47(23), 4316–4320 (2008). 44 F. Schilling, L. Schröder, K. K. Palaniappan, S. Zapf, D. E. Wemmer, and A. Pines, “MRI thermometry based on encapsulated hyperpolarized xenon,” ChemPhysChem 11(16), 3529–3533 (2010). 45 M. Zaiss, M. Schnurr, and P. Bachert, “Analytical solution for the depolarization of hyperpolarized nuclei by chemical exchange saturation transfer between free and encapsulated xenon (HyperCEST),” J. Chem. Phys. 136(14), 144106 (2012). 46 M. Zaiss and P. Bachert, “Exchange-dependent relaxation in the rotating frame for slow and intermediate exchange – modeling off-resonant spinlock and chemical exchange saturation transfer,” NMR Biomed. 26(5), 507–518 (2013). 47 J. Sloniec, M. Schnurr, C. Witte, U. Resch-Genger, L. Schröder, and A. Hennig, “Biomembrane interactions of functionalized cryptophane-A: Combined fluorescence and 129 Xe NMR studies of a bimodal contrast agent,” Chem. - Eur. J. 19(9), 3110–3118 (2013). 48 M. Schnurr, C. Witte, and L. Schröder, “Functionalized 129 Xe as a potential biosensor for membrane fluidity,” Phys. Chem. Chem. Phys. 15, 14178– 14181 (2013). 49 See supplementary material at http://dx.doi.org/10.1063/1.4865944 for additional figures and data.
This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP: 152.14.136.77 On: Fri, 13 Mar 2015 08:58:35
