8352
Langmuir 2007, 23, 8352-8356
Tracking Single Lipase Molecules on a Trimyristin Substrate Surface Using Quantum Dots Andreas W. Sonesson,†,‡ Ulla M. Elofsson,† Thomas H. Callisen,§ and Hjalmar Brismar*,‡ YKI, Institute for Surface Chemistry, and Department of Cell Physics, Royal Institute of Technology, Stockholm, Sweden, and NoVozymes A/S, BagsVaerd, Denmark ReceiVed March 30, 2007. In Final Form: May 23, 2007 The mobility of single lipase molecules has been analyzed using single molecule tracking on a trimyristin substrate surface. This was achieved by conjugating lipases to quantum dots and imaging on spin-coated trimyristin surfaces by means of confocal laser scanning microscopy. Image series of single lipase molecules were collected, and the diffusion coefficient was quantified by analyzing the mean square displacement of the calculated trajectories. During no-flow conditions, the lipase diffusion coefficient was (8.0 ( 5.0) × 10-10 cm2/s. The trajectories had a “bead on a string” appearance, with the lipase molecule restricted in certain regions of the surface and then migrating to another region where the restricted diffusion continued. This gave rise to clusters in the trajectories. When a flow was applied to the system, the total distance and average step length between the clusters increased, but the restricted diffusion in the cluster regions was unaffected. This can be explained by the lipase operating in two different modes on the surface. In the cluster regions, the lipase is likely oriented with the active site toward the surface and hydrolyzes the substrate. Between these regions, a diffusion process is proposed where the lipase is in contact with the surface but affected by the external flow.
Introduction Glyceride lipases are enzymes involved in lipid metabolism and function at lipid-water interfaces.1 The turnover of lipases increases by several orders of magnitude when the lipases are adsorbed at an interface, which in some lipase molecules involves a conformational change where an R-helix surface structure moves and exposes the hydrophobic active site region.2 A critical feature of the lipase catalysis is the mobility at the interface for individual lipase molecules. For secreted phospholipase A2, two modes of interactions between the enzyme and the surface have been used to describe the surface dynamics: hopping and scooting.3 If the lipase binds possessively and does not leave the surface between catalytic cycles, the enzyme is said to act in the so-called scooting mode. If an exchange between the solution and surface occurs, i.e., if the lipase leaves the surface between catalytic cycles, the dynamics is referred to as the hopping mode. Enzyme diffusion on substrate surfaces has received very little attention in the literature. Studies that have been reported are, e.g., the enzyme-substrate systems of cellulase on cellulose,4 subtilisin protease on a monolayer of BSA,5 collagenase on the peptide FALGPA,6 gelatinase on gelatin,7 β-amylase on a starch gel,8 and lipase on trimyristin.9 Fluorescence recovery after photobleaching (FRAP) was used in those studies as a method * To whom correspondence should be addressed. E-mail: hjalmar@ cellphysics.kth.se. † YKI, Institute for Surface Chemistry. ‡ Royal Institute of Technology. § Novozymes A/S. (1) Schmid, R. D.; Verger, R. Angew. Chem., Int. Ed. 1998, 37, 1608-1633. (2) Brzozowski, A. M.; Savage, H.; Verma, C. S.; Turkenburg, J. P.; Lawson, D. M.; Svendsen, A.; Patkar, S. Biochemistry 2000, 39, 15071-15082. (3) Berg, O. G.; Gelb, M. H.; Tsai, M. D.; Jain, M. K. Chem. ReV. 2001, 101, 2613-2653. (4) Jervis, E. J.; Haynes, C. A.; Kilbourn, D. G. J. Biol. Chem. 1997, 272, 24016-24023. (5) Roy, S.; Thomas, J. M.; Holes, E. A.; Kellis, J. T.; Poulose, A. J.; Robertson, C. R.; Gast, A. P. Anal. Chem. 2005, 77, 8148-8150. (6) Gaspers, P. G.; Robertson, C. R.; Gast, A. P. Langmuir 1994, 10, 26992704. (7) Collier, I. E.; Saffarian, S.; Marmer, B. L.; Elson, E. L.; Goldberg, G. Biophys. J. 2001, 81, 2370-2377.
to quantify the diffusion coefficient. FRAP measures the rate at which a photobleached area becomes repopulated with a fluorescent specimen and thus gives an ensemble average diffusion coefficient for the studied system.10,11 However, since FRAP might be insensitive to more complex behaviors such as confined diffusion,12 single particle tracking (SPT) has been used to get more detailed information on molecular motions.13,14 A main advantage with this method is that it does not need a model of the bleaching process and the distribution of particles to describe the diffusion, but enables directly the tracking of the space and time coordinates for single particles. SPT has mainly been used to study receptor mobility in cell membranes12,15,16 but also lipid diffusion in bilayers.17 To study single molecules for a sufficient time, proteins can be labeled with quantum dots (QDs), which are nanometer-sized inorganic semiconductors.18,19 They are stable light emitters whose emission wavelength can be tuned by the particle size and chemical composition.19,20 QDs typically consist of a CdSe or CdTe core capped with a ZnS shell, which can be functionalized to increase solubility and biocompatiblity.21 Compared to conventional fluorescent dyes, QDs are superior in terms of (8) Henis, Y. I.; Yaron, T.; Lamed, R.; Rishpon, J.; Sahar, E. O ¨ .; KatchalskiKatzir, E. Biopolymers 1988, 27, 123-138. (9) Sonesson, A. W.; Brismar, H.; Callisen, T. H.; Elofsson, U. M. Langmuir 2007, 23, 2706-2713. (10) Reits, E. A. J.; Neefjes, J. J. Nat. Cell Biol. 2001, 3, E145-E147. (11) Lippincott-Schwartz, J.; Altan-Bonnet, N.; Patterson, G. H. Nat. Cell Biol. 2003, S7-14. (12) Bates, I. R.; He´bert, B.; Luo, Y.; Liao, J.; Bachir, A. I.; Kolin, D. L.; Wiseman, P. W.; Hanrahan, J. W. Biophys. J. 2006, 91, 1046-1058. (13) Choquet, D.; Triller, A. Nat. ReV. Neurosci. 2003, 4, 251-265. (14) Saxton, M. J. Biophys. J. 1997, 72, 1744-1753. (15) Dahan, M.; Le´vi, S.; Luccardini, C.; Rostaing, P.; Riveau, B.; Triller, A. Science 2003, 302, 442-445. (16) Goulian, M.; Simon, S. M. Biophys. J. 2000, 79, 2188-2198. (17) Kiessling, V.; Crane, J. M.; Tamm, L. K. Biophys. J. 2006, 91, 33133326. (18) Giepmans, B. N. G.; Adams, S. R.; Ellisman, M. H.; Tsien, R. Y. Science 2006, 312, 217-224. (19) Chan, W. C. W.; Nie, S. Science 1998, 281, 2016-2018. (20) Alivisatos, P. Nat. Biotechnol. 2004, 22, 47-52. (21) Chan, W. C. W.; Maxwell, D. J.; Gao, X.; Bailey, R. E.; Han, M.; Nie, S. Curr. Opin. Biotechnol. 2002, 13, 40-46.
10.1021/la700918r CCC: $37.00 © 2007 American Chemical Society Published on Web 07/06/2007
Single Lipase Mobility on Trimyristin
brightness and stability against photobleaching, and fluoresce at sharp and discrete wavelengths. This makes it possible to continuously excite and track individual QDs for several minutes, whereas single organic fluorophores or autofluorescent proteins might photobleach within a few seconds.15,16,20 The aim of this work was to image and track single QDconjugated lipase molecules on a substrate triglyceride surface. Lipase surface diffusion on trimyristin was earlier performed with FRAP,9 but SPT of single molecules makes it possible to analyze the modes of lipase surface dynamics in more detail. Single enzyme activity studies were performed earlier, both when the enzyme was trapped in narrow capillaries22,23 or nanoscopic vials24 and when the enzyme was present at interfaces.25-27 However, there are to our knowledge no earlier reports on single enzyme tracking on surfaces. The lipase studied was from the fungus Thermomyces lanuginosus (TLL), which is an enzyme with pronounced interfacial activation,28 and as substrates, spincoated surfaces of trimyristin were used. Materials and Methods Materials. Carboxyl 2-MP CdSe/ZnS EviTags (Adirondack Green) quantum dots (QDs) were purchased from Evident Technologies (New York). A lipase variant from T. lanuginosus was provided by Novozymes A/S (Bagsvaerd, Denmark). Trimyristin (C45H8606, Mw ) 723.18, 97%), toluene (g99, 5%), and N-[3(dimethylamino)propyl]-N′-ethylcarbodiimide (EDC; 97%) were from Fluka AG (Buchs, Switzerland). The buffers used were glycine, pH 9.0 (10 mM NaCl, 0.05 mM EDTA, 50 mM glycine, and 1 mM NaN3), and PBS, pH 7.0. All water was of Milli-Q grade. Preparation of Trimyristin Films. Poly(vinyl chloride) (PVC) surfaces of 13 × 18 mm were treated with detergent solution, Milli-Q water, and ethanol and finally blown dry with N2. Cleaned PVC surfaces were spin-coated with trimyristin using a PWM32 photoresist spinner (Headway Research, Inc., Garland, TX). A 50 µL drop of 15% (w/w) trimyristin in toluene was placed in the center of the surface. A speed of 4000 rpm for 4 min was used to rotate the surface, with an acceleration of 100 rpm/s. This led to a thin trimyristin film on the PVC surface, with a contact angle with water of around 121°. This procedure was also earlier found to give smooth triglyceride surfaces.9,29 Conjugation of QDs to Lipase. The QDs used had a PEG (polyethylene glycol) surface coating modified to react with amine terminal groups (primarily Lys) of proteins. The hydrodynamic diameter was about 25 nm, and the recommended excitation of the QDs was 505 nm. A drop of ∼20 nM QD-lipase sample was pipetted onto the trimyristin surface, which was then mounted in a flow cell for the microscope. This led to a surface density of less than one QD-lipase molecule per 50 × 50 µm. A 0.1 mm thick silicon rubber limited the flow cell volume to 10 µL (18 × 5 × 0.1 mm). The flow cell was rinsed with buffer before imaging. A scan speed of 1 image/s (in a few cases 1 image/3 s) was used to detect a series of maximum 300 images or was stopped when the studied QD-lipase drifted out of focus or left the imaged area. Imaging was performed at maximum 15 min after adsorption to the trimyristin surface. Data Analysis. As a molecule moves on the surface, it tends to be found farther and farther away from its starting point, as described for a random walk process. The spreading could be calculated from the captured image time series as the mean square displacement (MSD) of each molecule. QD-lipase trajectories were calculated using the Volocity 4.0.1 software from Improvision Inc. (www.improvision.com). MSD functions can be calculated by averaging over all pairs of points with a given time lag with the following relation: MSDn∆t )
1
N-n
∑[(x
N - n i)1
i+n
- xi)2 + (yi+n - yi)2]
(1)
where xi and yi are the particle coordinates on frame i, N is the total number of steps in the trajectory, ∆t is the time between frames, and n∆t is the time interval over which the MSD is calculated. The MSD function was calculated for five time steps from 100 to 300 pairs of points, which has in simulations been proven to be enough to estimate the short-term diffusion coefficient with high accuracy.14 The diffusion coefficient D could then be estimated by fitting the first points in the MSD function versus time using the relation for a two-dimensional random walk: MSD ) 4Dt
(2)
The Volocity software also allowed quantification of the QD step length between the image frames.
Results QD-Lipase Mobility on Trimyristin with No Flow. Image series of single QD-lipase molecule migration on trimyristin were collected under no-flow conditions. The QD-lipase could easily be distinguished in the collected images. Three representative trajectories are displayed in Figure 1. The single molecule lipase motion could be described as a motion restricted in one part of the surface, with the lipase then being transported to another part where the restricted motion continued (Figure 1). The calculated MSD functions for the whole trajectories are shown next to the three trajectories in Figure 1. The MSD is plotted as a function of time and shown for the first five time points. Linear functions could easily be fitted to these functions, and thus, the short-term diffusion coefficients could be calculated. For comparison, the MSD of the trajectory in Figure 1A calculated for the first 50 first time points is shown in Figure 4A. It was seen that the MSD deviated from a linear function at longer times and that the cutoff time of five time points was necessary to estimate the diffusion coefficient. The calculated diffusion coefficients varied by 1 order of magnitude, from D ≈ 1.5 × 10-10 cm2/s to D ≈ 15 × 10-10 cm2/s, calculated from 14 different trajectories. The mean diffusivity was D ) (8.0 ( 5.0) × 10-10 cm2/s. The average step length could also be calculated from the data in the trajectories (Figure 3A). The average step length in the cluster regions was denoted dc, and the average step length when the lipase is transported between regions was denoted dt. It was found that dc always was shorter than dt, although the standard deviation was quite large. QDs that had not been conjugated to lipases adsorbed less compared to QD-lipases
8354 Langmuir, Vol. 23, No. 16, 2007
Figure 1. Trajectories of quantum dot-lipase on trimyristin under no-flow conditions plotted for three different lipases. The mean square displacement as a function of the first five time points as well as the calculated diffusion coefficient are plotted alongside each trajectory.
and were difficult to even spot on the surface. Objects that could be found were trapped, i.e., showed no mobility at all. QD-Lipase Mobility on Trimyristin with Flow. Image series of QD-lipase mobility on trimyristin were collected according to the standard protocol, but under a continuous flow of 2 µL/ min. This led to a linear velocity of 60 µm/s, calculated from the dimensions of the flow cell. Three representative trajectories are displayed in Figure 2, where the flow direction is from the bottom left corner to the upper right corner. QD-lipases seemed to move across the surface in the same direction as the flow but with intermittent “stops” where it underwent random movement. The linear velocity of the lipase on the surface was in the range of 1-2 µm/s, i.e., much lower than the applied flow. These random movements are seen in the zoomed in trajectories on the right in Figure 2. When the MSD function was calculated for the whole trajectories, it was found that MSD had a positive curvature, i.e., deviated from a linear dependence on time. Instead, the MSD was proportional to time × 1.9. An example of this is displayed in Figure 4B, where the MSD for the whole trajectory in Figure 2B is shown. However, when the MSD calculations were restricted to the regions where the lipase underwent random
Sonesson et al.
Figure 2. Trajectories of QD-lipase on trimyristin under a constant flow (direction indicated by the arrows), plotted for three different lipases. The parts of the trajectories where the QD-lipase seemed to undergo random movement are zoomed to the right, as well as calculated diffusion coefficients for these random parts of the trajectories.
motion (Figure 2), the MSD showed a linear dependence with time and D could be estimated in these regions. The mean diffusivity of this movement was (5.4 ( 3.7) × 10-10 cm2/s, calculated from 10 different trajectories. When the average step length in the different parts of the trajectory was analyzed, it was found that the steps were significantly shorter during the random movements compared to when the lipase seemed affected by the flow (Figure 3B). With applied flow, dc was similar to dc without flow (∼0.5 µm) but dt increased to values of several micrometers.
Discussion The aim of this work was to study single lipase molecule mobility on a triglyceride substrate surface. This was approached by the use of confocal microscopy and lipases conjugated to QDs. The size of the QD, ∼25 nm in diameter, by far exceeds the size of the lipase, which can be described as a spherically shaped molecule with an average diameter of 4.6 nm.2,30 However, at the single molecular scale, mass can be neglected as viscous
Single Lipase Mobility on Trimyristin
Langmuir, Vol. 23, No. 16, 2007 8355
Figure 3. Calculated average step length in different regions in the trajectories in Figures 1 and 2. The average step length in the cluster regions (within the circles) is denoted dc, and the average step length in the rest of the trajectory (transport between clusters) is denoted dt. (A) Trajectories with no flow (from Figure 1): left, dc ) 0.55 ( 0.25 µm, dt ) 0.65 ( 0.40 µm (step time 3 s); middle, dc ) 0.45 ( 0.20 µm, dt ) 0.70 ( 0.30 µm (step time 1 s); right, dc ) 0.60 ( 0.30 µm, dt ) 0.80 ( 0.40 µm (step time 1 s). (B) Trajectories with flow (from Figure 2): left, dc ) 0.55 ( 0.20 µm, dt ) 1.80 ( 1.00 µm (step time 1 s); middle, dc ) 0.30 ( 0.15 µm, dt ) 6.60 ( 3.20 µm (step time 1 s); right, dc ) 0.50 ( 0.20 µm, dt ) 1.00 ( 0.50 µm (step time 1 s).
forces dominate movement.13 According to the Stokes-Einstein equation of diffusion of a spherical particle, the diffusion coefficient is inversely proportional to the viscosity of the medium. For a QD of diameter 25 nm, the diffusion coefficient in water should be D ≈ 2 × 10-7 cm2/s, i.e., several orders of magnitude higher than for the QD-lipase surface diffusion found in this work. If the QD, attached to the lipase, remains mainly in solution, the viscosity the QD experiences is far lower than what the adsorbed lipase experiences. Thus, the lipase movement at the surface should be rate-limiting and not sensitive to what has been attached if the label is “dangling” free in solution without imposing a viscous drag. With these assumptions, the QDlipase complex used in this study should mimic an unlabeled lipase molecule in terms of surface diffusion. Single QD-lipase complexes could be tracked on the trimyristin surface. The surface density of adsorbed QD-lipase molecules was very low and ensured that the molecular motion was not affected by intermolecular interactions. Under no-flow conditions, the lipase mobility seemed to follow a “bead on a string” movement, as seen in the trajectories in Figure 1, with several cluster regions visible. The average step length calculated in the different regions implied that the lipase motion was different in the different regions of the trajectory, since the steps were shorter in the cluster regions (dc) compared to when the lipase migrates between clusters (dt) (Figure 3A). Diffusion coefficients could be quantified by calculating the MSDs of the particles. For a randomly diffusing particle in a two-dimensional system, the MSD should have a linear dependence on time,31 which was seen for lipases on trimyristin (Figure 1). However, at long times, (30) Hedin, E. M. K.; Hoyrup, P.; Patkar, S. A.; Vind, J.; Svendsen, A.; Fransson, L.; Hult, K. Biochemistry 2002, 41, 14185-14196. (31) Qian, H.; Sheetz, M. P.; Elson, E. L. Biophys. J. 1991, 60, 910-921.
the averaging in the MSD might give very scattered values, which has been analyzed, e.g., by Saxton.14 The deviation of a linear dependence of the MSD on time is also consistent with anomalous diffusion, e.g., local trapping effects in membranes.32 As an example, the MSD of the trajectory in Figure 1A calculated for the first 50 time points is diplayed in Figure 4A. This clearly showed that the MSD deviated from a linear function when the time steps increased. Hence, a cutoff time in the MSD is necessary to calculate the short-range diffusion coefficient. In this work five time points were chosen, which has been suggested in earlier work on diffusion in cell membranes.33 The diffusion coefficient of lipases on trimyristin under no-flow conditions when the whole trajectory was analyzed was estimated to be D ) (8.0 ( 5.0) × 10-10 cm2/s. To further analyze the mode of lipase migration on the substrate surface, mobility under flow conditions was measured (Figure 2). The flow rate and the dimensions of the flow cell assured a laminar flow over the surface. It was shown that the lipase mobility was affected by the flow and the lipase was pushed across the surface in the flow direction, but could at different durations of time move randomly on the surface, similar to the cluster regions when no flow was applied. The calculated MSD functions for the whole trajectories had a positive curvature (Figure 4B); i.e., the lipase seemed to diffuse at a higher rate the further it moved, which made it difficult to apply eq 2 to estimate the diffusion coefficient. This positive curvature of the MSD function has been shown theoretically to be characteristic for a diffusive system with a superimposed flow, i.e., a driven diffusive system.31 (32) Nicolau, D. V., Jr.; Hancock, J. F.; Burrage, K. Biophys. J. 2007, 92, 1975-1987. (33) Charrier, C.; Ehrensprenger, M.-V.; Dahan, M.; Le´vi, S.; Triller, A. J. Neurosci. 2006, 26, 8502-8511.
8356 Langmuir, Vol. 23, No. 16, 2007
Figure 4. (A) MSD of the QD-lipase trajectory in Figure 1A displayed for the first 50 time points. (B) MSD for the whole QDlipase trajectory under flow conditions in Figure 2B, calculated for the first five time points, showing a positive curvature.
However, the mean diffusion constant could be calculated for those periods of time when the lipase seemed unaffected by the flow and was in the same range as with no flow, D ) (5.4 ( 3.7) × 10-10 cm2/s. Moreover, when the average step length was analyzed, it was seen that dc was similar to that when no flow was applied but dt increased remarkably (Figure 3B). Our combined results suggest that the lipase operated in two modes on the substrate surface. In one mode, the lipase moved in a small region of the surface at a rate that was unaffected by external forces, since both D and dc were found to be similar with and without flow. As lipase adsorbs at a hydrophobic interface, a large hydrophobic patch on the lipase surface, containing the active site, is exposed as a part of the so-called interfacial activation.1,34 The cluster mode is likely associated with an orientation of the lipase molecule with the active site region facing the surface and thus enzymatic degradation of the substrate. This orientation has on hydrophobic model surfaces been suggested to be of low lateral mobility.35 The other mode, when lipase migrates between the cluster regions seen in Figure 1, was severely affected by the applied flow, with a remarkable increase in step length. However, no complete detachment of the lipase was likely during the transport mode, since the linear velocity of the lipase across the surface was much lower than the flow rate and no desorption into the bulk solution was visible. This might be explained by the possibility of some segments of the lipase always being in contact with the surface, e.g., a rotational diffusion process, or that the lipase is temporarily detached and then immediately recaptured in the adsorbed state. The latter would however be difficult to document due to, e.g., limitations (34) Verger, R. Trends Biotechnol. 1997, 15, 32-38. (35) Sonesson, A. W.; Callisen, T. H.; Brismar, H.; Elofsson, U. M. Langmuir 2005, 21, 11949-11956.
Sonesson et al.
in the time resolution of the microscope. The two-mode diffusion pattern of the lipase could have the functional purpose to minimize the effect of product inhibition. To that end, generation of negative charge at the interface (due to substrate conversion to free fatty acids) could be an additional mechanism that would induce a “jump” of the overall negatively charged enzyme. Hence, the lipase migration could be described as a diffusion-trap process with local restriction when the active site faces the surface, in analogy to the diffusion-trap processes that have been described for lateral diffusion of proteins in cellular membranes.36 Another possibility instead of the proposed model might be that the surface consisted of different domains, or that the surface roughness somehow constrained the motion of the lipase. The same surface has been examined previously with white profilometry to have a root-mean-square (rms) roughness parameter of 0.3 µm9, but when examined in the CLSM, the structure difference seemed to be on a much longer length scale (∼100 µm) compared to the distance between lipase cluster regions (∼10 µm). The roughness might be explained by some partial crystallization of the trimyristin on the surface. The TLL-trimyristin system, but with lipases labeled with an organic dye, was earlier analyzed with FRAP,9 which required a higher adsorbed amount of enzyme on the surface. The diffusion coefficient extrapolated to infinite surface dilution, which would represent the concentration studied with the QD-lipase, was estimated to be D ) 1.8 × 10-10 cm2/s for the same lipase variant. Thus, the single particle tracking experiments gave slightly higher values of the diffusion. An explanation for this is that the MSD in the trajectory measured the short-term microscopic diffusion, whereas FRAP measures the macroscopic long-term diffusion coefficient, which is a combination of microscopic mobility within a cluster and a macroscopic diffusion between clusters.
Conclusions This is the first reported single molecule study of enzyme lateral mobility on a substrate surface. It was shown that lipase molecules, conjugated to quantum dots, could be tracked on spin-coated surfaces of trimyristin. The calculated diffusion coefficient was D ) (8.0 ( 5.0) × 10-10 cm2/s under no-flow conditions. The lipase molecules seemed to operate in two different modes on the surface, which gave the trajectories a bead on a string appearance. The movement in the cluster regions of the trajectories was unaffected, in terms of diffusion rate and step length, by an applied flow, whereas the migration between cluster regions was affected by a large increase in the step length. This was thought to be due to different orientational states of the lipase, where the orientation with the active site toward the surface gave rise to the restricted diffusion in the clusters seen in the trajectories. Acknowledgment. Novozymes A/S and the Swedish Foundation for Strategic Research (SSF) are acknowledged for financial support of the project. Supporting Information Available: A QuickTime movie sequence of a QD on trimyristin, 20 frames/s (QD.qt), a QD-lipase on trimyristin, 20 frames/s (QD-lipase.qt), and a QD-lipase on trimyristin with an applied flow, 6 frames/s (QD-lipase+flow.qt) (note: due to small fluctuations in focus, the contrast has been improved in all video sequences to clarify the molecular movements). This material is available free of charge via the Internet at http://pubs.acs.org. LA700918R (36) Scott, L.; Zelenin, S.; Malmersjo¨, S.; Kowalewski, J. M.; ZettergrenMarkus, E.; Nairin, A. C.; Greengard, P.; Brismar, H.; Aperia, A. Proc. Natl. Acad. Sci. U.S.A. 2006, 103, 762-767.
