Super-resolution RESOLFT microscopy of lipid bilayers using a fluorophore-switch dyad†

Dyads consisting of a photochromic switch covalently linked to a fluorescent dye allow the emission from the dye to be controlled by reversible photoisomerization of the switch; one form of the switch quenches fluorescence by accepting energy from the dye. Here we investigate the use of dyads of this type for super-resolution imaging of lipid bilayers. Giant unilamellar vesicles stained with the dyads were imaged with about a two-fold resolution-enhancement compared with conventional confocal microscopy. This was achieved by exciting the fluorophore at 594 nm, using a switch activated by violet and red light (405/640 nm).

To a solution of the deprotected dye (11 mg, 13 µmol) and SO-NHS (6.5 mg, 11 µmol) in DMF (1.5 mL) was added DIPEA (7.1 mg, 9 µL, 55 µmol) from a stock solution in DMF (3.4 M) and the mixture was stirred at 20 °C overnight. The next day DIPEA (7.1 mg, 9 µL, 55 µmol) was added from the same stock solution (3.4 M) and the mixture was stirred for 6 h. In order to drive the reaction to completion, SO-NHS (8 mg, 13 µmol) was added and the solution was stirred at 20 °C overnight. The crude mixture was concentrated and the residue was dissolved in CH 2 Cl 2 . The organic layer was washed with water (3 x 5 mL) and the combined organic layers were concentrated. The crude product was purified by column chromatography (silica gel, 100% CH 2 Cl 2 to 85:15 CH 2 Cl 2 /CH 3 OH) affording dyad 2 as a purple solid (11 mg, 85%). R f = 0.32 (CH 2 Cl 2 /MeOH 4:1); 1

6'-Atto590-ethylenediamine-SO dyad (3)
Dye 3a (6.5 mg, 8.5 µmol) was dissolved in anhydrous dichloromethane (0.9 mL) and trifluoroacetic acid (0.1 mL) was added. The resulting purple solution was stirred at 20 °C under argon for 30 min, after which time the solvent was removed under high vacuum. Dichloromethane (1 mL) was added to the residue and the solvent removed under vacuum. This process was repeated 5 times. The resulting blue solid was used immediately without further purification assuming quantitative conversion; R f : 0.06 (SiO 2 , 90:10 CH 2 Cl 2 :CH 3 OH).
Crude 6'-Atto590-ethylenediamine conjugate (7.3 mg, 8.5 µmol) was dissolved in DMF (0.5 mL). The SO-NHS (6.0 mg, 10 µmol) was added as a solution in DMF containing DIPEA (30 µL DIPEA in 1 mL). The resulting blue solution was stirred at 20 °C under argon for 18 h after which time the solvent was removed under reduced pressure. The residue was dissolved in dichloromethane (10 mL) and washed with water (10 mL). The aqueous solution was extracted with dichloromethane (3 × 10 mL), and the organic layers were combined, washed with brine (10 mL), and concentrated. The crude product was subjected to multiple column chromatography purifications (SiO 2 , CH 2 Cl 2 to 85:15 CH 2 Cl 2 :CH 3 OH) to yield several fractions containing impure product. The product was further purified by semi-preparative HPLC (Method B) to yield dyad 3 as a blue solid (

6'-Atto590-piperazine-SO dyad (4)
Dye 4a (31 mg, 39 µmol) was dissolved in dichloromethane (3 mL) and trifluoroacetic acid (0.3 mL) was added. The resulting blue solution was stirred at 20 °C under argon for 45 min. The solvent was removed under high vacuum and the residue was dissolved in dichloromethane. This process was repeated 5 times, to yield the crude product as a blue solid, which was used immediately without further purification (assuming quantitative conversion); R f : 0. 10  Cl + 6'-Atto590-Boc-protected 4,4'-bipiperidine conjugate (5a) 6'-Atto590 (20 mg, 32 µmol) was dissolved in DMF (1.0 mL). To this solution was added HBTU (12 mg, 32 µmol) as a solution in DMF (1.0 mL) and DIPEA (50 µL, 290 µmol) and the solution was stirred for 5 min. N-Boc-4,4'-bipiperidine (8.6 mg, 32 µmol) was added as a solution in DMF (0.5 mL) and the reaction mixture was stirred at 20 °C under argon for 20 min after which the reaction mixture was quenched with water (3 mL) and stirred for a further 20 min. The solvent was removed under reduced pressure and the residue dissolved in dichloromethane (20 mL). The solution was washed with aqueous HCl (2%, 20 mL), and the aqueous layer extracted with dichloromethane (3 × 20 ml). The organic layers were combined, washed with brine (20 mL) and concentrated. The crude product was purified by column chromatography (SiO 2 , CH 2 Cl 2 to 88:12 CH 2 Cl 2 :MeOH) to yield dye 5a as a deep blue solid (14.7

6'-Atto590-4,4'-bipiperidine-SO dyad (5)
Dye 5a (10.0 mg, 11.4 µmol) was dissolved in anhydrous dichloromethane (1.8 mL) and trifluoroacetic acid (0.2 mL) was added. The resulting blue solution was stirred at 20 °C under argon for 1 h. The solvent was removed under high vacuum and the residue was dissolved in dichloromethane. This process was repeated 5 times, to yield the crude product as a blue solid, which was used immediately without further purification (assuming quantitative conversion); were subsequently added and the suspension was stirred at 120 °C for 18 h. The reaction mixture was cooled to 20 °C and poured into water (100 mL). The resulting suspension was filtered through a Celite pad. The filtrate was discarded and the solid was washed through the pad with chloroform. This solution was dried over sodium sulfate, filtered and concentrated to yield a dark green oil which was purified by silica gel column chromatography (neat CH 2 Cl 2 ), giving 8 as a clear colorless oil (

Calculation of Förster Distances
Equation S1 was used to calculate Förster radii, R 0 , for the dyads (see Table S2): where κ 2 is the dipole orientation factor (assumed to be 2/3 given free rotation between the donor and the acceptor), Φ D is the fluorescence quantum yield of the donor in the absence of the acceptor (in this case, dyes with linkers attached), n is the refractive index of the solvent (1.33 for methanol, 1.34 for PBS buffer), and J(λ) is the spectral overlap integral, which is given by Equation S2 .
where f D is the normalized donor emission spectrum, ε A (λ) is the acceptor molar extinction coefficient at wavelength λ. Efficiency of the FRET process, E, can be calculated from the Förster radius, R 0 , and the distance between the donor and acceptor, r, according to Equation S3. We have used distances measured from molecular mechanics structural optimizations of the dyads.

Photoinduced Electron Transfer as a Potential Quenching Mechanism
Experimental redox potential measurements Redox potentials of Cy3.5, Atto590 and the spironaphthoxazine switch SO ( Figure S13) were measured using square wave voltammetry. Electrochemical experiments were performed in acetonitrile at compound concentrations of 0.1-1.0 mM with tetrabutylammonium hexafluorophosphate as the electrolyte (0.1 M). Square wave voltammograms were collected with a modulation amplitude of 50 mV and a frequency of 2 Hz. Glassy carbon (Ø = 1 mm), platinum wire and Ag|AgNO 3 (10 mM) were used as working, counter and reference electrodes, respectively. The potentials were referenced at the end of each experiment by addition of ferrocene.

Figure S13
Structures of the dye derivatives and spironaphthoxazine tert-butyl ester used for electrochemistry measurements.

Calculated redox potentials
Density functional (DFT) calculations were carried out using the ORCA 4.1.1 program, 5 on simplified chemical structures from the dyads, using previously-published analysis to guide structure conformation ( Figure S14). 6 Tight optimization criteria were used for all geometries, and numerical frequency calculations used to confirm stationary points as minima, and to calculate thermal energy corrections. Resolution of identity was used to speed up the SCF process employing the RIJCOSX approximation. Grid6 and GridX6 were used in all calculations. Implicit solvation was introduced employing the SMD model. Redox potentials were calculated from free energies using Equation S4, where z is the number of transferred electrons (z = 1) and F is the Faraday constant, and referenced to the SCE electrode with an absolute ° of 4.429 V, corrected for the liquid junction potential in acetonitrile. 7 These values were further corrected to ferrocene (0.40 V vs. SCE) 8 for comparison with experimental values. Free energies can be calculated in the gas phase and in solvent, and the change in free energy for the redox process calculated according to Equation S5, as illustrated in the Born-Haber Cycle ( Figure  S15). 7 However, this process can be simplified, and the free energies in solvent of the two species directly calculated and compared, as for the example of a reduction in Equation S6. Calculations were performed at the PBE0 level of theory in combination with the def2-TZVPP basis set and the SMD solvation model (acetonitrile), using the optimized PBE/def2-SVP geometries.
Born-Haber cycle for calculating the reduction potential of a species, X. An analogous cycle applies for oxidation.

Results
A comparison of the calculated and experimental values for the redox potentials of each compound is shown in Table  S3. The calculated values consistently under-estimate the experimental redox potentials. We have applied these values to a thermodynamic analysis of PET. We cannot measure the oxidation and reduction potentials of the spironaphthoxazine in its open (merocyanine) form as it does not persist in solution.
The ∆ !"# of electron transfer for all possible scenarios was calculated using the Weller equation (Equation S7), 9 in which (D/D ! ) and ∆ D are the oxidation potential of the donor and singlet excitation energy of the fluorophore, respectively, and (A ! /A) is the reduction potential of the acceptor. The final term (e 2 /εr) accounts for the free energy gained by bringing the ions together in solution, however it is smaller than the errors in the redox potential measurements, so it can be ignored for this calculation. In this analysis, excited state energies of 2.09 and 2.07 eV for Atto590 and Cy3.5, respectively, were calculated as the midpoint between the absorption and emission maxima. ∆ !"# of the possible photoinduced electron transfer processes are shown in Table S4. As might be expected for the relatively electron rich photoswitch and electron poor dyes, electron transfer is more favorable when the switch acts as the donor in both dye combinations. The negative ∆ !"# values in these scenarios imply that some PET quenching of fluorescence is possible in the dyads, both when the spironaphthoxazine is closed and open. Such a process may account for the decrease in quantum yield observed when the dyes are conjugated to the switch (Table S1). In particular, the Atto590 dyads (3)(4)(5), which have shorter linkers than the cyanine dyads (1 and 2), show a larger reduction in fluorescence upon conjugation to the switch. This could be due to the highly distantdependent nature of electron transfer, which becomes very unfavorable as the donor-acceptor distance increases. Indeed, the fluorescence quantum yield variation between dyads 3-5, where the linkers have different flexibility, may be accounted for by changes in the electron transfer process from the closed spironaphthoxazine to the Atto590 fluorophore.
Although electron transfer is thermodynamically favorable from both the open and closed spironaphthoxazine to the fluorophores, the difference in ∆ !"# is probably not large enough to account for the magnitude of the observed switching in the dyads. It is likely that FRET is the main mechanism by which the open spironaphthoxazine quenches the fluorophores, but PET may contribute to the quenching.

Vesicle Preparation Preparation of GUVs for imaging
GUVs were freshly prepared on the same day as imaging, according to the following electroformation procedure: 10 The electroformation chamber was cleaned with ethanol and dried thoroughly under N 2 flow. A solution of DOPC (5 µL of a 1 mg mL -1 solution in CHCl 3 ) was spread evenly over two Pt electrodes and dried under N 2 flow. A solution of sucrose (370 µL, 300 mM aqueous) was placed in the electroformation chamber and the Pt electrodes were inserted into this solution. Using a function generator (frequency 10 Hz, amplitude 5.7 V peak-to-peak = 2 V RMS ) an electrical potential was applied to the sample for 1 h to form the GUVs. After 1 h, the frequency was reduced to 2 Hz and left for a further 30 min to detach the GUVs from the electrodes. For confocal and RESOLFT imaging, part the GUV solution (75 µL) was transferred into an Eppendorf tube and a solution of dyad (1 µL, 1 mM in DMSO) was added and left for 15 min. For FCS measurements, a solution of the dyad (0.5 µL, 1 µM in DMSO) was added to the vesicle solution (50 µL). The vesicle solution was then placed into PBS (250 µL) in the well of an 8-well ibidi plate for imaging. All transfers of GUVs were carried out using trimmed pipette tips.
Phosphate buffered saline was made by dissolution of tablets (ThermoFisher) in deionized water. The concentrations of the components are as follows: NaCl at 8.0 g L -1 , KCl at 0.2 g L -1 , Na 2 HPO 4 at 1.15 g L -1 , KH 2 PO 4 at 0.2 g L -1 .

Preparation of SLBs for Zeiss confocal imaging
PBS solution (100 µL) containing GUVs was taken from the 8-well plate and deposited on a plasma-cleaned glass slide which induces the vesicles to collapse and form supported lipid bilayers on the glass surface. A further 500 µL of PBS was added to the imaging chamber to hydrate the bilayers.

RESOLFT microscopy
RESOLFT microscopy was performed on a modified Abberior Instruments RESOLFT microscope (Abberior Instruments, Göttingen, Germany), equipped with excitation lasers at 594 nm (pulsed 80 ps at  For imaging on the Abberior RESOLFT microscope, glass ibidi µ-Slide 8-well plates were passivated with poly-L-lysine for 45 min, before washing with PBS three times, leaving 250 µL in the chamber. Then, 75 µL of the GUV solution was transferred into the wells. In order to acquire RESOLFT images, we used a pixel-by-pixel imaging sequence where each pixel is irradiated with a number of laser beams sequentially. Each pixel was irradiated as follows: firstly, a Gaussian-shaped 594 nm excitation spot acquired the confocal signal; secondly, a donut-shaped 405 nm pulse converted the spironaphthoxazine to its active open form; thirdly, the Gaussian-shaped 594 nm excitation spot acquired the RESOLFT signal by inducing fluorescence from the central (unquenched) region of the pixel. Each laser was applied sequentially to each pixel in the image, and raster scanning was used to build up the image. Each RESOLFT image is accompanied by a confocal image acquired simultaneously which acts as an internal control. Excitation (594 nm) powers were set such that the detector was not saturated. Because of the nature of super resolution imaging, it was necessary to have small pixel sizes (40 × 40 nm here) to allow sufficient sampling. The sample area irradiated by a laser beam is much bigger than the size of the pixels. In addition to the pixel at the center of the laser beam, surrounding pixels are also irradiated, switching neighboring molecules into their dark states. This manifests itself in subsequent pixels appearing much darker than they should in the bright images, because molecules are still in their dark states following irradiation of earlier pixels.
In our initial experiments we introduced a waiting time to allow the molecules in neighboring pixels to recover before moving to the next pixel. This allowed us to acquire our first images with sub-diffraction limit resolution (see Figure  S16). However, we found that several hundreds of milliseconds of waiting time were necessary for full recovery of the molecules, making the imaging very slow (33 minutes for a 2 × 2 µm field of view). As an alternative solution, we introduced a fourth laser pulse to photochemically accelerate the ring closing reaction of the open merocyanine to the closed spiro form. We tested lasers at three available wavelengths and found that 640 nm was the most suitable. This allowed significantly faster image acquisition. The pixel-by-pixel imaging sequence is shown in Figure 17. Before attempting super-resolution imaging using this setup, we explored the fluorescence switching behavior in pixel-by pixel acquisition by confocal imaging. Confocal imaging of the GUVs clearly showed that the dyad was incorporated into the lipid bilayer and was not present in significant concentration in the aqueous solution either inside or outside the vesicle. Switching off of the dyad was achieved using a Gaussian-shaped 405 nm beam rather than a donut-shaped beam. We varied the dwell times and laser powers to find the optimal conditions for switching d S27 (see S18-S21). We observed that optimal switching for dyads 1 and 2 was achieved using 400-500 µW of the 405 nm laser, but for dyads 3-5 lower 405 nm laser powers were required. For dyad 1, increasing the power beyond 500 µW did not seem to lead to poorer quenching, whereas for the other dyads, high laser powers led to a drop in quenching efficiency. For all dyads, higher 405 nm laser powers resulted in faster switching fatigue, as demonstrated later.
Having optimized the power of the switch-off laser, we used similar experiments to find optimal parameters for the 640 nm laser. Addition of a phase plate in the path of the 405 nm laser produced a donut-shaped beam with zero intensity at the center. With this in place, we observed resolution enhancement in our images (Figure 4 and Figures S24-28), with around a 2-fold improvement in FWHM for all dyads (see Table S5 and Figure S29). In some of our images, we observed that the resolution enhancement was good in the top rows of pixels of the image (rows acquired early in the raster scan) but deteriorated towards the bottom of the image. This is probably due to the diffusion of the dyads through the membrane. In the time that it takes to record the image, the molecules can diffuse through the membrane such that as we reach the bottom of the image, we are irradiating photoswitches that have already suffered fatigue.

Analysis of Microscopy Line Profiles
The RESOLFT microscopy experiments on DOPC GUVs stained with dyads 1-5 resulted in the intensity line profiles, perpendicular to the membrane at the equatorial plane of the vesicles, as shown in Figures 4 and S24-S28. Each experiment simultaneously generated two line profiles: one from the conventional confocal imaging mode and one from the RESOLFT super-resolution imaging mode.
The confocal line profiles are well described by a simple Gaussian curve (Equation S8), where I is the signal intensity (photon counts), I 0 is a baseline correction, I p is the peak intensity, x is the position along the line, x p is the center of the peak, and w con is the peak width (FWHM, Table S5).
The RESOLFT line profiles are better fitted by a Lorentzian model, 11,12 rather than a single Gaussian (Figs S22 and S23), giving the FWHM values (w SRM-L in Table S5). In Figures 4 and S24-S28, we have shown Gaussian fits for confocal profiles and Lorentzian fits for RESOLFT profiles, and we quantify the increased resolution by the ratio w con /w SRM-L (see also values in Table S5 and Figure S29).
The RESOLFT line profiles are also well described by a double Gaussian model 13 (Equation S9), where w con is fixed as the peak width from the corresponding confocal profile and w SRM-2G is the fitted width of the RESOLFT component (parameters summarized in Table S5). The w SRM-2G values extracted from these fits are smaller than the corresponding w SRM-L values obtained from Lorentzian fits, suggesting that the unsuppressed confocal contribution limits the resolution enhancement we can achieve. The observed enhancement in resolution (w con /w SRM-2G ) also increases (see Table S5).

Confocal microscopy
Confocal images were obtained using a Zeiss LSM780 inverted laser scanning confocal microscope equipped with various excitation sources: 405 nm (diode laser, CW, 815 µW at 100%) and 594 nm (HeNe laser, CW, 84 µW at 100%). Images were acquired using either a Plan Apochromat 63×/1.40 NA DIC M27 oil immersion objective or an LD C-Apochromat 40×/1.1 NA DIC water immersion objective. Unless otherwise stated, images are 512 × 512 pixels, using 80 nm pixels, giving an image size of 42.5 × 42.5 µm. The microscope was operated via a PC running Zen software. Image analysis was carried out in FIJI ImageJ.
For images recorded on the Zeiss LSM780 microscope, the GUV solution was placed in a pre-treated plastic ibiTreat µ-Slide 8-well plate (No. 1.5 coverslip).
All imaging carried out on the Zeiss LSM780 microscope used line-by-line laser switching sequence, rather than the laser sequence changing between pixels (see Figure S31). For example, in the 'bright image', only the 594 nm laser was used to excite the dyads. In the 'dark image', the top line of pixels was first irradiated with the 405 nm light to switch the molecules to their dark state, then this line was scanned again with the 594 nm laser to read out the dark image. This process was repeated for all 512 lines of the image.
It can be seen that in images taken on the Zeiss LSM780 (which has a linearly polarized excitation laser), the vesicles appeared brighter at the top and bottom due to better overlap of the transition dipole moment of the dye with the incident light. This observation confirmed that the dyes were aligned in the tails of the membrane lipids rather than being randomly associated with the surface. We synthesized a 'dummy dyad' 6 which does not contain a switch. We subjected dye 1a (containing Cy3.5) and dummy dyad 6 (containing Atto590) to the same 100 imaging cycles experiment as the dyads to see whether there was significant photobleaching of the fluorophore in the experiment ( Figure S40). With 0 µW of 405 nm laser power (i.e. only 594 nm excitation), there was no observable photobleaching of either the Atto590 or Cy3.5 dye. At 85 µW of 405 nm laser power there some photobleaching of Cy3.5, and the Atto590 dye exhibits some photobleaching at higher 405 nm laser power. These photobleaching effects are dwarfed by the loss of switching observed for the spironaphthoxazine unit in the dyads. S40 the sigmoidal shape of the fatigue. One possibility is that the dyad molecules are sufficiently close together that the spironaphthoxazine on one molecule can quench the fluorescent dye of neighboring molecules. Because of the excellent FRET overlap between the fluorescent dye and the active spironaphthoxazine quencher, FRET is efficient over long distances. In this instance, we would expect to see an initial period where we have deterioration of some switches, but neighboring molecules can still efficiently quench the emission, so there is no loss in overall quenching efficiency. Once a critical number of switches has been degraded, the quenching efficiency begins to fall. In a 1:1 mixture of dummy dyad 6 with dyad 3 which has a switch, where there is no quenching of neighboring molecules, one would expect the quenching efficiency to fall significantly (below 50%, as 6 has a higher fluorescence quantum yield than dyad 3). In fact, our experiments showed that the initial quenching efficiency for a 1:1 mixture was over 80% ( Figure S41), confirming that molecules are quenching each other. However, the fatigue resistance of this system was very poor as there is a smaller pool of switches available to quench fluorescence, so the critical number of degraded switches is reached much faster. Another possibility is that diffusion of non-degraded dyad molecules from outside the imaging plane can replenish the partially-degraded population in the imaging plane between the imaging cycles. For this to be true, the rate of diffusion would need to be slower than the time required to acquire a single image (otherwise no dark image would be observed) but fast enough to replenish the dyad over the course of the imaging cycles. This would result in an initial period where fresh dyad replaces degraded dyad in the imaging plane, meaning minimal loss in quenching efficiency. After a period of time, the supply of fresh dyad is exhausted, and the quenching efficiency falls. We tested this hypothesis using supported lipid bilayers (SLBs; see ESI Section 6 for preparation), in which all the dyad molecules in the membrane were irradiated in each cycle, leaving no 'spare' pool out of the imaging plane from which to replenish between imaging cycles. There is still the possibility for diffusion between pixels during the raster scanning, as with imaging GUVs. However, diffusion of molecules that have already been irradiated will lead to poorer observed quenching efficiency. We observed that the fatigue curves in SLBs were still sigmoidal, but that the initial period before the decay began appeared shorter for SLBs than for GUVs ( Figure S42-S43). These results confirm that replenishment by diffusion is occurring but that it is not the only process which contributes to the sigmoidal shape.

Modeling Sigmoidal Fatigue
We can use a simple kinetic model to describe the sigmoidal appearance of the quenching efficiency fatigue. In the following explanation, we assume that the rate of switch fatigue is constant.
In the simplest example, we can consider that molecules behave independently of each other. We start with a state (A) in which the switch is capable of quenching the dye so there is no emission. Our concentration of quenched species is 1. Our switches are degraded with rate constant k to a state (B) where the switch cannot quench the dye, so the dye becomes emissive. The concentration of quenching species follows 1st order kinetics, giving an exponential decay.
If our molecules interact in pairs, in which one switch can quench both the dyes in the pair, we have 3 possible states (Scheme S7): State A, where both switches are active so the system is fully quenched; State B, where one switch has been degraded but the other can still quench both dyes so the system is still dark; State C where both switches are degraded so the system becomes emissive.

Scheme S7
Illustrative representation of kinetic model used to model the shape of quenching efficiency fatigue curves for an aggregate of two dyad molecules.
Assuming that the total concentration is given as We can extend this principle to any number of interacting molecules, n. As you increase the number of interacting molecules, the initial period of high quenching efficiency gets longer ( Figure S45). Unfortunately, because of the competing diffusion replenishment processes, it is impossible to fit our experimental data using these simple models, but the models do demonstrate that cooperative quenching between molecules leads to sigmoidal fatigue behavior.

Fluorescence Correlation Spectroscopy
For details of sample preparation for FCS measurements see Supporting Information Section 6. FCS measurements were carried out on a Zeiss LSM880 inverted laser scanning confocal microscope equipped with a 561 nm (HeNe) excitation laser. Data were acquired using an LD C-Apochromat 40×/1.  Transit times determined from FCS measurements were used to calculate diffusion coefficients. Firstly, the size of the observation spot was first calibrated according to previously reported procedures. 15 A FWHM of 297 nm was measured for the observation spot, and diffusion coefficients, D, were calculated according to Equation S10, where τ D is the transit time measured by FCS:

HPLC Analysis
Reverse phase HPLC was performed at 298 K using an Agilent 1100 Series system comprising an autosampler (G1313A), a vacuum degassing unit (G1379A), a quaternary pump (G1311A), a column oven (G1316A), a diode array detector (G1315B), and a fraction collector (G1364C). The instrument was operated using ChemStation software. For analytical HPLC an Agilent Eclipse XDB-C18 column (4.6 × 150 mm, 5 μm particle size) was used with a flow rate of 1.0 mL/min. For semi-preparative HPLC an Agilent Eclipse XDB-C18 column (9.4 × 250 mm, 5 μm) was used with a flow rate of 4.18 mL/min.