Open Access
Add to BibSonomyAbstract
We present a fluorescence correlation spectroscopy (FCS) approach to obtain spectral cross-talk free auto- and cross-correlation functions for probes with highly overlapping emission spectra. Confocal microscopes with either a hyperspectral EM-CCD or six-channel PMT array spectral detection were used, followed by a photon filtering correlation approach that results in spectral unmixing. The method is highly sensitive and can distinguish between Atto488 and Oregon Green 488 signals so that auto-correlation curves can be fitted without the need for cross-talk correction. We also applied the approach to the membrane dye Laurdan whose emission is dependent on the lipid order within the bilayer. With fluorescence spectral correlation spectroscopy (FSCS), we could obtain spectral cross-talk free auto- and cross-correlation functions corresponding to Laurdan located in liquid ordered and liquid disordered phases.
© 2014 Optical Society of America
1. Introduction
Fluorescence correlation spectroscopy (FCS) has become a widely used technique in biophysical and biological research [1, 2]. The most commonly investigated biophysical parameters measured with FCS are diffusion coefficients, concentrations and the existence of interactions between mobile species, namely for fluorescently labeled proteins and lipids. The advantages of FCS are that it measures the dynamic behavior of molecules at nanomolar concentrations over a large range of time scales and its minimal invasiveness makes it compatible with live cell measurements. One of the challenges of FCS is to capture the dynamics of two or more diffusing species simultaneously. The benefit of simultaneous data acquisition is that the recorded parameters are obtained exactly under the same experimental conditions for all species. This is particularly important for cellular measurements as conditions in live cell can change rapidly. In addition, the simultaneous data acquisition enables cross-correlation analysis to reveal interactions between the species [3, 4].
The main prerequisite for simultaneous multicomponent FCS analysis is that the species under investigation differ in their fluorescence properties. There are four main characteristics that can be used for species identification – excitation spectrum, emission spectrum, excited state lifetime and polarization. The most common approach is dual-color fluorescence correlation spectroscopy [3] (DC-FCS) that uses fluorophores with distinct emission characteristics allowing the separation of emitted photons by different spectral regions. The basic configuration of DC-FCS contains one [5] or two [3] excitation wavelengths to match the excitation spectrum of the used fluorophores and two non-overlapping detection channels, each matching emission spectrum of one fluorophore. The emission spectra of the fluorophores cannot significantly overlap, otherwise DC-FCS suffers from intensity cross-talk resulting in an artificial positive cross-correlations, which need to be corrected for with appropriate data fitting procedures [6]. The main source of spectral cross-talk is the ‘leakage’ of the emission from the shorter wavelength fluorophore into the detection channel of the fluorophore with the longer wavelength. This bleed-through can be minimized by using pulsed interleave excitation (PIE) [7]. PIE requires differences in excitation spectra of the fluorophores so that one laser exclusively excites one fluorophore. The fast sub-microsecond alternation of lasers makes it possible to distinguish between red fluorescence excited by green laser and red fluorescence excited by red laser, and removes the green-red cross-talk from FCS analysis. The limitation of this approach is that fluorophores with highly overlapping excitation and emission spectra cannot be used. Further, the use of two laser lines for excitation creates the detection volumes that slightly differ in their size and position, which undermines the accuracy of the measurement.
The concept of DC-FCS has been extended to fifteen [8, 9] and recently up to 80 spectral channels [10]. The higher number of spectral channels allows analyzing more than two species simultaneously and can decrease the spectral cross-talk by optimizing the selection of spectral bands. So far, the only attempts to remove the existing spectral-cross talk artifacts from DC-FCS data were introduced at the data analysis step. Here, calibration measurements [6] and global analysis approach [9] were employed to fit FCS correlation curves generated from multiple spectral channels. However, none of these post-processing corrections can handle species with highly overlapping spectra.
The second modality for multicomponent FCS analysis is exploring differences in excited state lifetimes. Contrary to DC-FCS, fluorescence lifetime correlation spectroscopy [11–13] (FLCS) uses a single excitation and emission channel together with pulsed excitation, time-tagged time resolved (TTTR) detection and photon filtering to obtain ‘clean’ auto- and cross- correlation curves for each component. The interesting feature of FLCS is that the excited state decays, which are used to separate the species, are by default highly overlapping as they both start at the same time. In DC-FCS terminology they have an extremely high cross-talk, but still result in cross-talk free correlation curves for all species. The cross-talk free FLCS data is obtained by implementing mathematical filters that weight photons based on the known excited state decays of each species. The limiting factors for FLCS analysis is that the filtering decreases signal to noise ratio for the correlation functions and that it is sensitive to the non-ideal characteristics of single photon counting detectors [13]. In addition, FLCS requires specialized hardware and software that is not often found in biological laboratories.
In this paper we describe a new approach that we named fluorescence spectral correlation spectroscopy (FSCS). As in FLCS, the method uses a photon filtering approach, but instead of excited state lifetimes it is based on spectral data from at least six spectral channels. We demonstrate that clean cross-talk free auto- and cross-correlation curves can be obtained even for fluorophores with highly overlapping emission and excitation spectra such as Atto488 and Oregon Green 488. We evaluate two experimental systems: a confocal sample-scanning microscope with an EM-CCD camera for spectral detection and a conventional laser-scanning confocal microscope with more than five highly sensitive spectral detectors. We also applied the method to the lipid phase-sensitive dye Laurdan to distinguish between compositionally homogenous and heterogeneous suspensions of lipid vesicles.
2. Experimental setup
2.1 A confocal sample-scanning microscope with EM-CCD based spectral detection
The multi-purpose optical setup [14–16] was built on an inverted microscope IX71 (Olympus, Hamburg, Germany) and included a 488 nm line of Ar-ion laser as an excitation source, an acousto-optical tunable filter (AOTFnC-400.650, AAOptoelectronic, Orsay, France) for wavelength selection and intensity modulation, a single-mode polarization-maintaining optical fiber (LINOS Photonics, Goettingen, Germany) for spatial mode filtering and an air-spaced objective (UPLSAPO 4X, Olympus) for excitation beam re-collimation just before the laser beam enters the back port of the microscope body. The microscope body contained the standard confocal part consisting of a z488rdc dichroic mirror (Chroma, Rockingham, VT), a water immersion objective (UPLSAPO 60x, Olympus), 3D sample scanning stage (PIMars XYZ NanoPositioner, 200 x 200 x 200 μm, PI, Karlsruhe, Germany), exchangeable 50 μm diameter pinhole placed at the focal plane of the left camera port, a recollimation lens (AC254-250-A-ML, Thorlabs) and an optional HQ500LP long pass filter (Chroma, Rockingham, VT). The collimated fluorescence beam entered a dispersive equilateral prism (PS852, Thorlabs) under a minimum deviation angle. The resulting spectrum was focused with a 50 mm focal length achromatic lens (AC254-050-A-ML, Thorlabs) onto an EM-CCD camera (iXon DU-860D 128 x 128 pixels, Andor, Belfast, United Kingdom), cooled to −70ºC. The camera was connected to water-cooling circuit to minimize the vibrations from cooling fan. The projected image of the pinhole on the EM-CCD was made smaller by a factor of 5 to fit into a single pixel (24 μm x 24 μm). The line spectrum was aligned parallel to the readout register into a fifth raw from chip’s storage area. As 10% of the light was reaching rows four and six, 3x1 binning was used to capture all light without any significant effect on camera readout speed and noise. The rest of the light sensitive area was devoid of any light. The prism and the EM-CCD camera were placed on independent adjustable rotation stages with a common rotation axis. The prism rotation allowed for finding the minimum deviation angle configuration, the change in the EM-CCD angle allowed for horizontal adjustment of the spectrum position. The nonlinear nature of prism dispersion caused a change in the spectrum resolution from 2 nm/pixel at 500 nm to 6 nm/pixel at 700 nm. As the nonlinear nature of the spectrum has no effect on the FSCS results no spectral calibration was used. An arbitrary waveform generator card (PCIe6259, National Instruments, TX) was used to control AOTF and synchronize it with EM-CCD detection.
The setup was operated with a home-written software in LabVIEW (NI). Communication with EM-CCD camera occurred via a manufacturer provided SDK dll library. The EM gain was set to 1000 for single photon read-out with pixel readout clock at 10 MHz. Data were read out in the continuous-crop mode with full vertical binning of three rows at a speed of 62 500 lines per second. Using this setting, the whole camera image including the storage area was only shifted by three rows every 16 µs. The three rows reaching the readout register were vertically binned and read-out as a single line of up to 128 pixels. Reading out fewer pixels or horizontal binning of pixels did not improve the readout speed. The acquisition used a circular buffer in run-till-abort mode, which allowed for an arbitrary length of measurements, and stored raw data directly to a hard disk. The typical length of measurement was 160 s (10 million of lines).
The multiplicative process during the EM-CCD readout increases the signal-to-noise ratio (SNR) by a factor of √2 for high EM gains, which effectively leads to halving the quantum yield of the detector [17]. Since the experiments described here typically had low light levels of less than 0.1 electron/pixel/frame, we improved the SNR by applying a photon counting treatment to the data [18]. Raw analogue data (in counts) from the EM-CCD camera were converted to single photon time-tagged spectrally resolved (TTSR) data format by setting a threshold for a successful single photon detection (see also Fig. 2(A)). The threshold level was set as the pixel and line specific-median value plus six times the standard deviation of read-out noise [17]. The pixel and line specific-median was obtained in two steps. First, median values for each pixel along the line were obtained from the full length of the measurement and used as a pixel-varying correction. Second, the moving median of all pixels in each 1000 lines, with the pixel correction applied, was obtained as a line-varying correction. The procedure assumes that less than 20% of pixels contain a photon. The standard deviation of the read-out noise was determined from experimental data with a closed shutter. The obtained value of 5 counts corresponds to a threshold offset of 30 counts above the median. If the number of counts exceeded the threshold, a single TTSR event was created with the time tag corresponding to the line number and the spectral channel corresponding to the pixel number. When average photon fluxes are higher than 0.1 photo-electron/pixel/readout, the photon counting conversion suffers from saturation effect and standard analogue values corrected for varying background levels could be used instead, but was not necessary here [19].
2.2 A confocal laser-scanning microscope LSM780 with an array of PMTs for spectral detection
A commercial confocal laser-scanning microscope (CLSM), namely the LSM780 (Zeiss, Germany) equipped with an array of 32 GAsP detectors for spectral detection was also used for this work. The linear spectral resolution of 8.91 nm per detector covered the spectral range of 410.26 - 695.38 nm. The detectors were operated at single photon counting mode to improve the SNR. Point FCS measurements allow simultaneous readout of 6 channels as photon streams. Each channel i.e. each stream represents a routed output from an arbitrary number of consecutive spectral detectors. For 488 nm excitation experiments, the emission was acquired from 499 nm to 657 nm with splitting at 516 nm, 534 nm, 552 nm, 578 nm and 604 nm. For Laurdan measurements with 405 nm excitation the emission was acquired from 421 nm to 622 nm with splitting at 447 nm, 473 nm, 499 nm, 525 nm and 560 nm. The laser power at the back aperture of the water immersion of the 40x 1.2NA objective was 5 µW for both 405 nm and 488 nm excitations.
The LSM 780 was operated by the software ZEN in a multichannel FCS mode. Raw data from 180 s acquisitions were stored directly to a hard disk in single photon (.fcs) format. A single raw data file (*.fcs) was created for each channel. A routine in LabVIEW was established to merge the files and convert them into a single TTSR file. The original arrival time for each photon was encoded in the time tag and the file number corresponded to a spectral channel. The TTSR data format is identical to .pt3 data format used in time tagged time resolved data acquisition (TTTR, PicoQuant, Germany). The spectral channel information was saved as the time resolved information and all photons were considered to be detected in one detector. The data conversion enables the use of FLCS analysis software for direct evaluation of TTSR data.
2.3 Data analysis
FSCS is an analogues technique to FLCS. It uses emission spectra where FLCS uses different excited state fluorescent lifetime. The data treatment is identical for both FSCS and FLCS methods. We present here a detailed explanation of the FSCS principle devised from the FLCS principle reported previously by Humpolíčková et al. [20].
The above-described data conversions result in single photon counting data in time tagged spectrally resolved (TTSR) format. In this format, every detected photon is assigned a time that elapsed since the beginning of the measurement t and a discrete spectral channel j, which corresponds to the wavelength of the photon. It should be highlighted that FSCS is not limited to single photon-counting data and analogue data of each spectral channel and time increment could also be used. The conversion to TTSR, however, not only improves the SNR at the low photon rates obtained here but also significantly decreases the size of the data file and increases the speed of the correlation function calculation.
At every time t, the fluorescence intensity Ij(t) in each spectral channel j is a linear combination of area normalized fluorescence spectra pjk (Fig. 2(B)):
where k stands for a particular emitter with a specific spectrum, n is the number of different emitters with different spectra and wk(t) is the contribution of the kth fluorescent species to the total fluorescence signal at time t.Equation (1) is an over-determined set of linear equations, provided the number of spectral channels is higher than the number of different emitters. Assuming that the photon detection obeys a Poissonian distribution and applying singular value decomposition, the solution of the Eq. (1) can be written as follows:
where N is the number of spectral channels and fjk is a discrete filter function (Fig. 2(C)), which is constructed from the area normalized fluorescence emission spectra of the different fluorescence species and the total intensities of the compound signal in spectral channels. Explicitly, fjk is given by:where the matrix elements are:Finally, the intensity normalized correlation function of the kth species with lth species (for auto-correlation k = l, for cross-correlation k ≠ l) is calculated as:In practice, the only difference to standard single photon data based calculation of correlation functions is that each photon enters the calculation with a double precision weight given by the filter value fjk for the selected species k and photon’s spectral channel j, instead of default value equal to one. We used a custom-written fast multi-tau correlation algorithm that accepts double precision values as the starting photon weight (dll library compiled in C/C + + MVS 10).The final spectral cross-talk free, intensity-normalized auto-correlation curves were fitted, using a non-linear least square Levenberg-Marquardt algorithm (LabVIEW, NI), to a mathematical model of a single freely diffusing species within the 3D Gaussian detection volume that also accounts for the transition of the fluorophores to the triplet state:
where PN stands for average number of diffusing species in detection volume, τres is the mean residence time, ω0 and ωZ are radii of the 3D Gaussian approximated focal volume in lateral and axial direction respectively, and T and τ0 is the fraction of molecules undergoing the triplet transition and the triplet constant, respectively. The diffusion coefficient D was obtained as D = ω02/ (4*τres).3. Sample preparation
3.1 Materials
1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC), 1,2-dioleoyl-sn-glycero-3-phospho-L-serine (sodium salt) (DOPS), cholesterol and N-palmitoyl-D-erythro-sphingosylphosphorylcholine (SM) were purchased from Avanti Polar Lipids, Inc. (Alabaster, AL). 2-(4,4-Difluoro-5-methyl-4-bora-3a,4a-diaza-s-indacene-3-dodecanoyl)-1-hexadecanoyl-sn-glycero-3-phosphocholine (β-BODIPY® 500/510 C12-HPC), Oregon Green® 488 1,2-dihexadecanoyl-sn-glycero-3-phosphoethanolamine (Oregon Green® 488 DHPE), 3,3′-dihexadecyloxacarbocyanine perchlorate (DiOC16(3)) and Alexa Fluor® 488 NHS-ester were obtained from Life Technologies Corporation. Atto488 with free carboxy group was obtained from ATTO-TEC (Siegen, Germany). 4-(2-Hydroxyethyl)piperazine-1-ethanesulfonic acid (HEPES) and sodium chloride (NaCl) were purchased from Sigma (St. Louis, MO).
3.2 Vesicle preparation
Lipid mixtures of a final lipid concentration 1 mM (4:1 molar mixture of DOPC:DOPS labeled with Oregon Green® 488 DHPE at lipid to dye ratio 10 000:1; 4:1 molar mixture of DOPC:DOPS labeled with β-BODIPY® 500/510 C12-HPC at lipid to dye ratio 1 000:1; 4:1 molar mixture of DOPC:DOPS labeled with DiOC16(3) at lipid to dye ratio 1 000:1; and 100% DOPC, 50:35:15 molar mixture of DOPC:SM:Chol and 70:30 molar mixture of SM:Chol labelled with Laurdan at lipid to dye ratio 100:1) were prepared from lipid stock solutions in chloroform. The organic solvent was evaporated under a stream of nitrogen and the resulting thin lipid film was kept for additional 2 h under vacuum. The dried lipid film was subsequently hydrated with 10 mM HEPES buffer (150 mM NaCl, pH = 7.4) and the solution was extensively vortexed for at least 2 min until multilamellar vesicles were formed. Next, the cloudy suspension of Oregon Green® 488 DHPE labeled lipid mixture was sonicated with a probe sonicator for 20 min, yielding a suspension of small unilamellar vesicles (SUVs). All other lipid preparations were extruded through polycarbonate membrane with 100 nm or 200 nm (DiOC16(3) lipid mixture only) diameter pores (Mini-extruder, Avanti Polar Lipids, Inc.), yielding clear suspensions of large unilamellar vesicles (LUVs).
4. Principle of fluorescence spectral correlation spectroscopy (FSCS)
The general scheme of a microscope setup suitable for FSCS is depicted in Fig. 1(A). The main difference compared to traditional emission detection with dichroics is that the collected fluorescence here first passed a pinhole and was then dispersed using a prism or a grid and projected onto a linear array of single photon sensitive detectors. Each detected photon was recorded and converted in a time-tagged spectrally resolved (TTSR) format, as described above. TTSR data format keeps the information of the detection time of the photon from the start of the experiment and the identity of spectral channel the photon was detected in.
Fig. 1 Principle of fluorescence spectral correlation spectroscopy (FSCS) data acquisition and analysis. (A) FSCS setup is based on a standard confocal microscope with a single excitation wavelength. The emitted fluorescence is re-collimated after passing through a pinhole, dispersed via a prism or grid and focused onto a line array of single photon counting detectors (APDs, PMTs or EM-CCD). Each detected photon n is assigned a spectral detection channel λn and detection time tn relative to the start of the acquisition. Data is saved in time-tagged spectrally resolved (TTSR) format. (B) Emission spectra were reconstructed from TTSR data obtained for 100 nm diameter DOPC vesicles labelled with BODIPY FL-DOPE at ratio 1:1000 (red line), 200 nm diameter DOPC vesicles labelled with DiO at ratio 1:1000 (blue line), a 3:1 mixture of 100 nm BODIPY FL-DOPE labelled vesicles with 200 nm DiO labelled vesicles (green line) and the detector’s background signal (black line). The data was obtained with an EM-CCD camera in single photon counting mode. (C) Photon weighting filters were calculated from the spectral patterns shown in B according to Eq. (3). Note the sum of all filters (brown line) is equal to 1 for each spectral channel. (D) Spectral cross-talk free autocorrelation functions of each component (DiO – blue line, BODIPY FL-DOPE – red line and EM-CCD background – black line) and their cross-correlation functions (only DiO versus BODIPY FL-DOPE cross-correlation is shown – magenta line) were obtained from the TTSR data of the 3:1 mixture of 100 nm BODIPY FL-DOPE labelled vesicles with 200 nm DiO labelled vesicles (green line in B) after applying the photon weighting filters according to Eq. (4).
A requirement for the FSCS analysis is that spectral patterns of all components contributing to the signal of the sample are determined. Typically, these are two distinct spectral patterns from two fluorophores and a background signal of detectors (Fig. 1(B)). The easiest way of obtaining the spectral patterns is to acquire reference TTSR data for all contributors separately and subsequently build histograms from spectral channels of all photons in the TTSR data. In Fig. 1(B), data was obtained with an EM-CCD camera resulting in 128 spectral channels. Alternatively, the spectrum of the mixture may be able to be decomposed into its components; however, this approach is not only more complicated but also likely to be more error prone.
The next step in the FSCS analysis is the calculation of photon weighting filters (Fig. 1(C)). Spectral patterns of all components plus the spectrum of the sample are used to calculate the photon weighting filters for each component (Eq. (3)). The sum of all filters in each spectral channel should be equal to 1 (brown line in Fig. 1(C)) and each filter has to have negative values for some channels. If the sum of the filters in some spectral channel significantly deviates from 1, the spectrum of the sample is not a linear combination of the used spectral components (Eq. (1)) and such filters cannot be correctly used for FSCS analysis. The need for negative values stems from the orthonormality of the spectral patterns with the filters [11]. As spectral patterns are always positive, the filters have to contain negative values in order that the scalar multiplication of a pattern of one component with a filter of second component can yield zero.
The last step of FSCS analysis is the calculation of clean auto- and cross- correlation functions for all components (Fig. 1(D)). Photon weighting filters are directly used during the calculation of the correlation function (Eq. (4)). The photon weighting adjusts intensities at any given time to essentially obtain the fractional intensity of a given component (Eq. (2)). This ensures that the resulting ‘filtered’ correlation functions are free of any spectral cross-talk. In the example shown in Fig. 1, the spectra of the BODIPY (β-BODIPY® 500/510 C12-HPC) and DiO (DiOC16(3)) dyes and the spectrum of the detector background clearly overlap. Despite this immense spectral cross-talk, three clean autocorrelation functions were obtained from a mixture of BODIPY and DiO-labeled liposomes after the photon filter (Fig. 1(C)) was applied. The completely flat auto-correlation function for the background signal demonstrates the successful separation of the BODIPY and DiO emission signal. If the method did not work, the autocorrelation function for the background would correlate to autocorrelation curves of the vesicles. The cross-correlation function between BODIPY and DiO that were incorporated into two distinct vesicles that were subsequently mixed is probably not a cross-talk artifact but caused by the exchange of dyes between the vesicle population since DiO is soluble in water.
5. FSCS with an EM-CCD camera for spectral detection
In a typical FLCS experiment, the decay of the excited state is sampled into several hundreds of time channels [20] so that the number of time channels is much higher than the number of species to be resolved. In order to use FLCS mathematics for FSCS analysis of spectrally resolved data, the number of spectral channels must also be significantly higher than the number of species that one wishes to be resolved. Tests on a commercial setup (MT200, PicoQuant) have shown that dichroic mirrors with four APDs detectors were not sufficient to unmix two dyes with highly overlapping emission spectra freely diffusing in water with FSCS (data not shown). We attribute this failure to the suboptimal spectral splitting by the dichroic mirrors and to the varying levels of afterpulsing of the APD detectors. The suboptimal spectral splitting caused most of the photons to be detected in the second and third spectral channel, lowering the effective contribution of the remaining two channels to the FSCS analysis. The high and intensity-dependent afterpulsing of the detectors also act as another spectral components, increasing the total number of species to above the number of spectral channels, so that the FLCS mathematics could no longer be used.
We built a confocal microscope with an EM-CCD camera resulting in 128 spectral channels [21, 22]. Previously, EM-CCD cameras were reported to be a suitable detection system for intensity fluctuation analysis [19, 23, 24]. The electron multiplication register increases a single photo-induced electron signal above the noise of the analogue-to-digital convertor, allowing for single photon detection (Fig. 2(A)). The raw analogue EM-CCD signal is converted to TTSR data by applying a threshold. The threshold used in our experiments was equal to background level plus a six time the standard deviation of read out noise. If the intensity in a pixel was above the threshold level, a single TTSR photon event was registered, in which the time t from the start of the experiment is reflected in the line number and the spectral channel is equivalent to the x pixel coordinate.
Fig. 2 Fluorescence spectral correlation spectroscopy (FSCS) with an EM-CCD camera. (A) An EM-CCD camera (128x128 pixels) was used in continuous crop mode to obtain a 62.5 kHz line spectral readout of 128 pixels with single photon sensitivity (EM gain 1000). A typical single line readout is shown (black line). Intensities above the defined threshold (red line) were considered as single photon events. (B) Comparison of experimental spectra of Alexa488 in water (1 nM at 10 µW – red line, 1, 2, 4 and 8 nM at 40 µW – blue, cyan, magenta and dark yellow lines, respectively, 488 nm excitation) with the spectrum of EM-CCD background (black line. Measured with a closed shutter). The high read-out background (integral intensity 375 kHz from all 128 channels) makes this detection unsuitable for single molecule experiments (1 nM Alexa488 at 10 µW 488 nm excitation has an integral fluorescence intensity of 5.4 kHz after background subtraction – red line minus black line). The rise of background with increasing pixel number is indicative of readout-induced noise (C) Amplitude-normalized autocorrelation curves measured for 100 nm diameter DOPC vesicles labelled with BODIPY FL-DOPE at ratio 1:1000 (red line), 200 nm diameter DOPC vesicles labelled with DiO at ratio 1:1000 (blue line) and BODIPY FL-DOPE (red dots) and DiO (blue dots) autocorrelation curves obtained from a 3:1 molar mixture of 100 nm BODIPY FL-DOPE-labelled vesicles and 200 nm DiO-labelled vesicles, indicating the successful separation of the diffusion times for 100 nm and 200 nm vesicles despite the spectral overlap of BODIPY FL-DOPE and DiO.(D) Autocorrelation curves of DiO, obtained from a 3:1 molar mixture of 100 nm BODIPY FL-DOPE-labelled vesicles with 200 nm DiO labelled vesicles at various binning of spectral channels. Number of channels are 128 (black line,) 32 (red line), 16 (blue line), 8 (cyan line) and 5 (magenta line).
It should be pointed out that the first limiting factor of commercial EM-CCD cameras for FSCS is their readout speed. These cameras have only one readout register operated at maximal frequency of 10 MHz. The theoretical minimal time to read-out one full line of 128 pixels is 12.8 µs. Due to a slight overhead caused by the vertical image shifting, we reached a maximum speed of 16 µs per line (62.5 kHz) using a continuous-crop mode. Decreasing the number of pixels per line or horizontal binning of pixels did not improve the line speed for our camera. Although the time resolution of 16 µs was not sufficient to measure small freely diffusing molecules in water with a typical correlation times of tens of microseconds, it is sufficient to follow the diffusion of proteins in cells with a typical correlation times hundreds of microseconds or lateral diffusion of lipids in lipid membranes with a typical correlation time in millisecond range.
The second limiting factor we faced is the read-out-induced background noise of EM-CCD camera. Figure 2(B) shows typical single photon-counting spectral readout of EM-CCD camera with a closed shutter (black line) and single molecule acquisitions of Alexa488 in solution at various concentrations with the shutter open. The background signal reached an integral value of 375 kHz for the sum of all 128 channels. In addition, the background was not uniform for all channels with higher levels for higher pixel numbers. Based on this feature we assumed that the main component of the background is the read-out-induced noise – a random creation of electron while shifting the accumulated charges into the read-out register. The overall background level of 375 kHz was almost two orders of magnitude higher than fluorescence intensity (5.4 kHz after background subtraction) of 1 nM Alexa488 in water at an excitation power of 10 µW (red line in Fig. 2(B)). It was still possible to obtain autocorrelation curve from these data (data not shown) but the high background prevented single molecule FSCS analysis of mixture of two components with highly overlapping spectra. To test the linearity of EM-CCD response and the stability of the EM-CCD background, we varied the Alexa488 concentration and increased the excitation power to 40 µW (Fig. 2(B)). The detected intensity scaled linearly with the concentration. The spectral distribution over the 128 channels for different fluorescence signal intensities helped to identify artificial peaks in the spectra caused by scattered excitation light (first peak in pixel 4) or by hot pixels (three peaks at pixels 14, 48 and 72).
Since the high EM-CCD background levels did not allow for single molecule FSCS experiments, we tested the principles of FSCS on BODIPY (β-BODIPY® 500/510 C12-HPC)-labeled and DiO (DiOC16(3))-labelled lipid vesicles that were mixed after preparation (Fig. 1(B)). Here, each vesicle contains several tens of dye molecules, which increases the brightness of the vesicles well above the background noise of the EM-CCD camera. We used a reasonably low molar labeling ratio of 1:1000 and moderate excitation intensity (10 µW) so that the detected fluorescence intensity still fulfills the assumption of the single photon detection. The background noise was contributing approximately one third of the total signal in the FSCS experiments (Fig. 1(B), the green line). As outlined above, Fig. 1(D) shows a successful separation of the autocorrelation curves for all three components (including the background) and cross-correlation function between the two dyes.
To further benchmark the performance of the FSCS method, we compared the shape of auto-correlation curves of 100 nm and 200 nm LUVs obtained in separated measurements with the shape of auto-correlation curves obtained from the same LUVs that were mixture before the FSCS measurements and analysis (Fig. 2(C)). The autocorrelation curves for each LUV population agrees extremely well, returning similar correlation times of 5.6 ± 0.3 ms for 100 nm BODIPY-labeled LUVs and 10.5 ± 0.6 ms for 200 nm DiO-labelled LUVs.
Next, we asked what is the minimal number of spectral channels needed for a successful FSCS experiment. It is not trivial to theoretically estimate the required number of spectral channels as factors like detectors’ background and spectral splitting relative to the spectra of the used dyes are important parameters. Hence we used our experimental data and performed FSCS analysis of the BODIPY-labelled and DiO-labelled LUV’s mixture in conjunction with software–based spectral binning. The resulting autocorrelation functions for DiO component are shown in Fig. 2(D). The clean autocorrelation functions obtained after photon filtering were not significantly different in shape even though the data was reduced to 5 bins and only a small variation in the amplitude was observed. It should be noted that binning increased the noise in the autocorrelation curve, particularly for long correlation times. The data indicate that it is possible to perform FSCS experiments with only 5 spectral channels, but a high number of channels improves the signal-to-noise ratio of the correlation curves.
6. Fluorescence spectral correlation spectroscopy using a PMT array based spectral detection
With the above experiments with an EM-CCD spectral detection, we showed that five spectral channels were sufficient to resolve three spectral components from a mixture but that the read-out noise of the camera was too high for true single molecule FSCS measurements. This finding indicated that FSCS measurements could be performed with commercial confocal microscopes with multiple sensitive PMT detectors, for example the LSM780 systems (Carl Zeiss Microscopy) that has a spectral 32-channel GaAsP array detector. Here, we simultaneously read out six spectral channels that were freely selectable between 410 and 695 nm and have a spectral resolution of 8.9 nm per channel.
To test the single molecule FSCS performance with six channel spectral GaAsP detectors, we mixed Atto488 freely diffusing in water with LUVs that were weakly labeled with Oregon Green 488 (Fig. 3), assuming that Atto488 does not spontaneously associate with the LUVs. The spectra of the dyes are highly overlapping (Fig. 3(A)) and dual-color FCS approach completely fails to obtain clean correlation functions for both species from this mixture (data not shown). We calculated the photon weighting filters from the spectral patterns of all species and from the spectrum of the mixture (Fig. 3(B)). Contrary to EM-CCD based detection we did not need to include a spectral signature for noise, as its contribution to the overall signal was negligible. The sum of the photon weighting filters of the two individual dyes was equal to one for all spectral channels (brown line in Fig. 3(B)), confirming that the spectrum of the mixture is indeed a linear combination of the two dyes. Next, we calculated the correlation functions using the weighted intensities. The filtering splits the overall correlation function for the mixture (green line in Fig. 3(C)) into autocorrelation functions for both species and into their cross-correlation function (Fig. 3(C)). As expected, there is no cross-correlation between Atto488 freely diffusing in water and Oregon Green 488 embedded in LUVs. This again demonstrates that the cross-correlation is spectral cross-talk free. A comparison of auto-correlation curves measured for both fluorophores individually with the autocorrelation curves obtained by FSCS analysis from the mixture clearly demonstrates that the curves fully overlap, giving identical correlation times of 40 ± 2 µs for Atto488 and 580 ± 30 µs for Oregon Green 488 in LUVs (Fig. 3(D)). The only difference between the two data sets was the level of noise. Due to its statistical nature, the FSCS analysis unavoidably increases the noise level; the more the spectra of the species overlap, the noisier the recovered correlation curves after photon weighting will be.
Fig. 3 Fluorescence spectral correlation spectroscopy (FSCS) with spectral GaAsP detectors. (A) Emission spectra for free Atto488 in solution (red line), 100 nm DOPC vesicles labelled with Oregon Green 488-DOPE at ratio 1:10 000 (OG488, blue line) and a 1:1 volume mixture of 100 nm OG488-labelled vesicles with free Atto488 in solution (green line) were reconstructed from six spectral channel TTSR data acquired with an LSM780 confocal microscope. (B) Photon weighting filters were calculated for Atto488 (red line) and OG488 (blue line) according to Eq. (1). The sum of filters (brown line) adds up to 1. (C) Autocorrelation curve for a 1:1 volume mixture of 100 nm OG488-labelled vesicles with free Atto488 in solution without any filtering applied (green line) and autocorrelation curves of Atto488 (red line) and OG488 (blue line) and cross-correlation curve between Atto488 and OG488 (magenta line) from the same TTSR data after applying the photon weighting filters shown in (B). Note the lack of cross-correlation, indicating successful unmixing of the Atto488 and OG488 signals. (D) Auto-correlation curves for Atto488 in solution (red line), 100 nm OG488-labelled vesicles (blue line) and FSCS auto-correlation curves for Atto488 (red dots) and OG488 (blue dots) from a 1:1 volume mixture of 100 nm OG488-labelled vesicles with free Atto488 in solution. Both direct FCS measurement and FSCS measurement in a mixture returned diffusion times of 40 µs for free Atto488 in solution and 580 µs for OG488 in vesicles. The filtered auto-correlation curves are noisier due to the inherent uncertainty caused by the statistical photon weighting.
7. Fluorescence spectral correlation spectroscopy with environment sensitive probes
There are a number of probes that change their fluorescence emission properties due to changes in local polarity, viscosity, charge, potential, pH, concentrations of other molecules and others [25]. The microenvironment induced-spectral shift of the probes is often rather minor, and certainly does not lead to a complete separation of spectra for different environments. We next asked whether FSCS could be applied to environment sensitive dyes so that the heterogeneity within a system, reflected by changes in the emission profile, could be identified on the time scale and resolution of a given FSCS experiment. In a homogenous system, the resulting intensity normalized autocorrelation and cross-correlation functions should be identical and independent of the spectral patterns that were used to create the FSCS filters. In contrast, in a heterogeneous system, any blind spectral photon weighting can change the shape of the correlation functions as it will change relative contributions of the probe from different environments. If there is additional knowledge on these environments available, i.e. the spectral patterns of the probe in these environments, spectral cross-talk free auto- and cross-correlation curves for probe in each environment can be obtained by FSCS, potentially addressing dynamic partitioning of the probe into the different environments.
We tested the applicability of FSCS to environment sensitive probes on mixtures of LUVs of different lipid composition and stained with the lipid phase reporter, Laurdan. Lipid membranes can co-exist in different phases, such as liquid ordered (Lo) and liquid disordered (Ld) phases, leading to different membrane micro-polarities and micro-viscosities [26]. Laurdan is a lipophilic probe with a huge change of the dipole moment upon excitation, which contributes to its dipolar relaxation sensitivity [27, 28]. When incorporated into lipid membranes, Laurdan can sense the amount of water molecules bound to the acyl-carbonyl and the mobility of these bound water molecules [29]. The more hydrated and less mobile the Laurdan environment, the more red-shifted is the emission spectrum. This feature is frequently used to image and quantify membrane order in model and cell membranes [30]. The liquid disordered phase is characterized by more hydrated and mobile carbonyl region, as is the case for the liquid ordered phase, causing a red shift of Laurdan emission relative to the ordered phase. The emission profile of Laurdan in homogenous bilayers with a single lipid phase depends on the exact lipid composition so that there are no pre-determined spectra that correspond, for example to liquid ordered and liquid disordered membranes. Solvent relaxation in lipid membranes is a nanosecond process, similar to excited state lifetime of Laurdan [31]. As a result the emission spectrum of Laurdan red shifts with increasing time after excitation and cannot be decomposed into principle components [32]. As the shortest correlation time of our FSCS acquisition is 100 ns, the solvent relaxation-induced spectral change is below the resolution of the method. Thus, for the FSCS experiments only the steady state spectrum, which is the spectrum integrated over the excited state lifetime of Laurdan, is relevant. In other words, for FSCS purposes, Laurdan located in two different lipid phases is the equivalent of two fluorescent dyes with different emission spectra.
In a proof-of-the-principle experiment, we examined whether FSCS would return no cross-correlation for Laurdan-stained vesicles with segregated lipid phases and would return cross-correlation for vesicles with co-existing phases (although the spectra of Laurdan in the co-existing phases are not known). The model LUVs samples were designed to fulfill the key assumption of FSCS analysis, which is that the emission spectrum for the analyzed sample is a linear combination of the used spectral patterns, in our case corresponding to Laurdan in either phase, at any given time (Eq. (1)). A homogenous membrane was mimicked by a homogenous suspension of lipid vesicles with two lipid phases within each vesicle (biphasic vesicles) and membranes with co-existing phases were mimicked by a heterogeneous mixture of two vesicle populations that each only contains one lipid phase (mixture of monophasic vesicles). The lipid phases in the biphasic vesicles may not be identical to those in the monophasic vesicles. Please note that in these experiments, each LUV containing many Laurdan molecules is the smallest detectable diffusing unit. Hence in the homogenous suspension of biphasic LUVs, the two Laurdan emission signatures that correspond to either phase should cross-correlate since FSCS sees only the diffusion of whole vesicles and thus the two lipid phases enter and leave the detection volume at the same time. In contrast, no cross-correlation between the ordered and disordered Laurdan signature is expected for the mixture of monophasic vesicles.
Biphasic LUVs were prepared from 50:35:15 molar mixture of DOPC:SM:Chol, which is known to undergo phase segregation at room temperature [33]. Monophasic LUVs in the ordered phase were prepared from 70:30 molar mixture of SM:Chol and monophasic LUVs in the disordered state from pure DOPC. The acquired spectra (Fig. 4(A)) confirmed that Laurdan spectrum in the disordered phase is red-shifted relative to the ordered phase. Further, the spectra of both the biphasic vesicle suspension and the mixture of monophasic vesicles are linear combinations of two Laurdan spectral patterns since the sum of the weighting filters for each spectral channel is equal to one (Fig. 4(B)). Hence the assumption that the overall spectrum is a linear combination of the used spectral patterns seems to be fulfilled.
Fig. 4 Fluorescence spectral correlation spectroscopy (FSCS) analysis of large unilamellar vesicles (LUVs) suspensions stained with environmentally sensitive membrane dye Laurdan. Laurdan is a membrane dye whose emission spectrum depends on the lipid phase. (A) 100 nm vesicles in liquid disordered phase (DOPC, red line), liquid ordered phase (70:30 molar mixture of SM:Chol, blue line) and with coexisting liquid ordered/disordered phase (50:35:15 molar mixture of DOPC:SM:Chol, cyan line) were stained with Laurdan before extrusion at molar ratio of 1:100 dye:lipid. Laurdan emission spectra from the three different types of lipid vesicles, plus a 1:1 volume mixture (green line) of liquid disordered phase vesicles and liquid ordered phase vesicles are shown. Data were reconstructed from six spectral channel TTSR data acquired with an LSM780 confocal microscope (405 nm excitation). (B) Photon weighting filters were calculated for spectral patterns devised from Laurdan in disordered (red line) and Laurdan in ordered (blue line), both for suspension of LUVs with coexisting liquid ordered/disordered phase (full square) and for mixture of monophasic vesicles (hollow square) according to Eq. (1). The sum of filters (black lines) is 1. (C) FSCS analysis of a 1:1 volume mixture of liquid disordered phase vesicles and liquid ordered phase vesicles (green in A) returned auto-correlation curves for Laurdan in liquid disordered phase pattern (red line) and Laurdan in liquid disordered phase pattern (blue line). As anticipated, no cross-correlation between the phases (magenta line) was detected. (D) FSCS analysis of the Laurdan-labelled vesicles with coexisting liquid ordered and liquid disordered phases (cyan line in A) returned auto-correlation curves for Laurdan in liquid disordered phase pattern (red line), liquid ordered phase pattern (blue) and the cross-correlation curve between the two patterns (magenta line). The high amplitude of the cross-correlation curve indicates that the patterns fluctuate together, which is expected since both lipid phases are located together in each vesicle.
We first applied the FSCS analysis to the mixture of monophasic vesicles (Fig. 4(C)). As each vesicle contains only one phase and the vesicles move independently of each other, no positive cross-correlation was obtained. The slight anti-cross-correlation (i.e. cross-correlation < 1) can be explained by the detector’s dead time effect and/or by an excluded volume effect [13]. In case of the dead time effect, the presence of a strong signal from one species lowers the probability of detection of the signal from another species. In the case of the excluded volume effect, the presence of a large entity in the detection volume lowers the available volume for other entities and thus decreases the probability of their detection. Both effects lead to anti-correlation.
Next, we applied the FSCS analysis to the homogenous suspension of biphasic vesicles. As the spectral patterns of all vesicles are nearly identical, there are no spectral fluctuations and a significant cross-correlation between the phases is observed (Fig. 4(D)). The amplitudes of auto- and cross- correlations differed slightly since not every vesicle can be expected to contain exactly the same amount of both phases and so the biphasic vesicles are not a completely homogenous doubly labelled system. However, the difference in cross-correlation between a mixture of monophasic vesicles and biphasic vesicles is dramatic, demonstrating that FCSC indeed can be applied to environmentally sensitive probes. Further experiments are required to precisely map the nature of the lipid phases in membranes with co-existing phases and apply FSCS to cells.
8. Conclusion
We have presented a new FCS-based method that eliminates spectral cross-talk for probes with highly overlapping emission spectra. The crucial and limiting factor of this new method is the spectrally resolved detection. Firstly, the detection has to be fast enough to capture molecular diffusion in sample of interest. Second, it has to have a high detection efficiency and low background noise in order to perform single molecule FCS measurements. And finally, the number of spectral channels has to exceed the number of analyzed components plus one (that accounts for noise). Although more spectral channels is clearly better, we demonstrated here that the six spectral channels of a commercial confocal microscope with an GaAsP detector array are sufficient to separate two dyes with highly overlapping emission spectra (Atto488 and Oregon Green 488). Fluorophores with less overlapping emission profiles may also allow a further reduction of the number of emission channels. We further showed that although in principle possible, a hyper-spectral detection with an EM-CCD camera suffers from a too high background noise for single molecule FSCS experiments.
We have further demonstrated the possibility to study dynamic environmental heterogeneity and to resolve the dynamics of environmentally sensitive dye in different locations. Since spectral properties of environmentally sensitive probes never completely segregate in different microenvironments, standard dual-channel FCCS cannot be used to measure the dynamics in the different environment. However, FSCS could obtain spectral cross-talk free auto- and cross-correlation for the lipid phase reporter, Laurdan. Hence should be possible to independently measure the diffusion of the probes in liquid-order and liquid-disordered lipid phases even when the phases co-exist in the same membrane. This unique feature makes FSCS highly attractive and opens the door to studying how membrane order affects the dynamics in live cells.
Lastly, we would like to point out that the photon weighting approach used here is not limited to point measurements and can be successfully combined with spectrally resolved line-scanning FCS or with spectrally resolved RICS. These are particularly attractive options for live cell measurements to reduce photo-toxicity and probe the fluctuation of for example, membrane order over a large range of length scales.
Acknowledgments
A.B. acknowledges financial support from the Academy of Sciences of the Czech Republic (grant KJB400400904) and the University of New South Wales for the Vice-Chancellor Postdoctoral Fellowship. M.H. and P.K. acknowledge financial support from the Czech Science Foundation (P208/12/G016). M.H. acknowledges funding from the Academy of Sciences for the Praemium Academie. K.G. received funding from the Australian Research Council and National Health and Medical Research Council of Australia including Program Grant 1037320. We acknowledge the support provided by the Biomedical Imaging Facility at UNSW.
References and links
1. E. L. Elson, “Fluorescence correlation spectroscopy: past, present, future,” Biophys. J. 101(12), 2855–2870 (2011). [CrossRef] [PubMed]
2. Y. Tian, M. M. Martinez, and D. Pappas, “Fluorescence correlation spectroscopy: a review of biochemical and microfluidic applications,” Appl. Spectrosc. 65(4), 115A–124A (2011). [CrossRef] [PubMed]
3. P. Schwille, F. J. Meyer-Almes, and R. Rigler, “Dual-color fluorescence cross-correlation spectroscopy for multicomponent diffusional analysis in solution,” Biophys. J. 72(4), 1878–1886 (1997). [CrossRef] [PubMed]
4. H. Park, C. Pack, M. Kinjo, and B. K. Kaang, “In vivo quantitative analysis of PKA subunit interaction and cAMP level by dual color fluorescence cross correlation spectroscopy,” Mol. Cells 26(1), 87–92 (2008). [PubMed]
5. K. G. Heinze, A. Koltermann, and P. Schwille, “Simultaneous two-photon excitation of distinct labels for dual-color fluorescence crosscorrelation analysis,” Proc. Natl. Acad. Sci. U.S.A. 97(19), 10377–10382 (2000). [CrossRef] [PubMed]
6. K. Bacia, Z. Petrášek, and P. Schwille, “Correcting for spectral cross-talk in dual-color fluorescence cross-correlation spectroscopy,” ChemPhysChem 13(5), 1221–1231 (2012). [CrossRef] [PubMed]
7. B. K. Müller, E. Zaychikov, C. Bräuchle, and D. C. Lamb, “Pulsed interleaved excitation,” Biophys. J. 89(5), 3508–3522 (2005). [CrossRef] [PubMed]
8. M. Burkhardt, K. G. Heinze, and P. Schwille, “Four-color fluorescence correlation spectroscopy realized in a grating-based detection platform,” Opt. Lett. 30(17), 2266–2268 (2005). [CrossRef] [PubMed]
9. M. J. Previte, S. Pelet, K. H. Kim, C. Buehler, and P. T. So, “Spectrally resolved fluorescence correlation spectroscopy based on global analysis,” Anal. Chem. 80(9), 3277–3284 (2008). [CrossRef] [PubMed]
10. F. Bestvater, Z. Seghiri, M. S. Kang, N. Gröner, J. Y. Lee, K. B. Im, and M. Wachsmuth, “EMCCD-based spectrally resolved fluorescence correlation spectroscopy,” Opt. Express 18(23), 23818–23828 (2010). [CrossRef] [PubMed]
11. M. Böhmer, M. Wahl, H.-J. Rahn, R. Erdmann, and J. Enderlein, “Time-resolved fluorescence correlation spectroscopy,” Chem. Phys. Lett. 353(5-6), 439–445 (2002). [CrossRef]
12. P. Kapusta, M. Wahl, A. Benda, M. Hof, and J. Enderlein, “Fluorescence lifetime correlation spectroscopy,” J. Fluoresc. 17(1), 43–48 (2006). [CrossRef] [PubMed]
13. P. Kapusta, R. Macháň, A. Benda, and M. Hof, “Fluorescence lifetime correlation spectroscopy (FLCS): concepts, applications and outlook,” Int. J. Mol. Sci. 13(12), 12890–12910 (2012). [CrossRef] [PubMed]
14. J. Humpolíčková, A. Benda, R. Macháň, J. Enderlein, and M. Hof, “Dynamic saturation optical microscopy: employing dark-state formation kinetics for resolution enhancement,” Phys. Chem. Chem. Phys. 12(39), 12457–12465 (2010). [CrossRef] [PubMed]
15. J. Humpolícková, A. Benda, and J. Enderlein, “Optical saturation as a versatile tool to enhance resolution in confocal microscopy,” Biophys. J. 97(9), 2623–2629 (2009). [CrossRef] [PubMed]
16. L. Beranová, J. Humpolíčková, J. Sýkora, A. Benda, L. Cwiklik, P. Jurkiewicz, G. Gröbner, and M. Hof, “Effect of heavy water on phospholipid membranes: experimental confirmation of molecular dynamics simulations,” Phys. Chem. Chem. Phys. 14(42), 14516–14522 (2012). [CrossRef] [PubMed]
17. A. G. Basden, C. A. Haniff, and C. D. Mackay, “Photon counting strategies with low-light-level CCDs,” Mon. Not. R. Astron. Soc. 345(3), 985–991 (2003). [CrossRef]
18. O. Daigle, C. Carignan, and S. Blais-Ouellette, “Faint flux performance of an EMCCD,” Proc. SPIE 6276, 62761F (2006). [CrossRef]
19. J. R. Unruh and E. Gratton, “Analysis of molecular concentration and brightness from fluorescence fluctuation data with an electron multiplied CCD camera,” Biophys. J. 95(11), 5385–5398 (2008). [CrossRef] [PubMed]
20. J. Humpolícková, L. Beranová, M. Stĕpánek, A. Benda, K. Procházka, and M. Hof, “Fluorescence lifetime correlation spectroscopy reveals compaction mechanism of 10 and 49 kbp DNA and differences between polycation and cationic surfactant,” J. Phys. Chem. B 112(51), 16823–16829 (2008). [CrossRef] [PubMed]
21. A. Benda, “Spectral fluorescence correlation spectroscopy with EM-CCD camera,” presented at 15th International Workshop on “Single Molecule Spectroscopy and Ultrasensitive Analysis in the Life Sciences”, Berlin, Germany, 15–18 Sept. 2009.
22. A. Benda, “Spectral fluorescence correlation spectroscopy with EM-CCD camera,” presented at Methods and Applications in Fluorescence 11, Budapest, Hungary, 6–9 Sept. 2009.
23. B. Kannan, J. Y. Har, P. Liu, I. Maruyama, J. L. Ding, and T. Wohland, “Electron multiplying charge-coupled device camera based fluorescence correlation spectroscopy,” Anal. Chem. 78(10), 3444–3451 (2006). [CrossRef] [PubMed]
24. M. Burkhardt and P. Schwille, “Electron multiplying CCD based detection for spatially resolved fluorescence correlation spectroscopy,” Opt. Express 14(12), 5013–5020 (2006). [CrossRef] [PubMed]
25. A. S. Klymchenko and Y. Mely, “Fluorescent environment-sensitive dyes as reporters of biomolecular interactions,” Prog. Mol. Biol. Transl. Sci. 113, 35–58 (2013). [CrossRef] [PubMed]
26. K. Gaus, T. Zech, and T. Harder, “Visualizing membrane microdomains by Laurdan 2-photon microscopy,” Mol. Membr. Biol. 23(1), 41–48 (2006). [CrossRef] [PubMed]
27. T. Parasassi, F. Conti, and E. Gratton, “Fluorophores in a polar medium: time dependence of emission spectra detected by multifrequency phase and modulation fluorometry,” Cell. Mol. Biol. 32(1), 99–102 (1986). [PubMed]
28. L. A. Bagatolli, T. Parasassi, G. D. Fidelio, and E. Gratton, “A model for the interaction of 6-lauroyl-2-(N,N-dimethylamino)naphthalene with lipid environments: implications for spectral properties,” Photochem. Photobiol. 70(4), 557–564 (1999). [CrossRef] [PubMed]
29. P. Jurkiewicz, L. Cwiklik, P. Jungwirth, and M. Hof, “Lipid hydration and mobility: an interplay between fluorescence solvent relaxation experiments and molecular dynamics simulations,” Biochimie 94(1), 26–32 (2012). [CrossRef] [PubMed]
30. D. M. Owen, C. Rentero, A. Magenau, A. Abu-Siniyeh, and K. Gaus, “Quantitative imaging of membrane lipid order in cells and organisms,” Nat. Protoc. 7(1), 24–35 (2011). [CrossRef] [PubMed]
31. P. Jurkiewicz, J. Sýkora, A. Olzyńska, J. Humpolícková, and M. Hof, “Solvent relaxation in phospholipid bilayers: principles and recent applications,” J. Fluoresc. 15(6), 883–894 (2005). [CrossRef] [PubMed]
32. T. Parasassi, E. Gratton, H. Zajicek, M. Levi, and W. Yu, “Detecting membrane lipid microdomains by two-photon fluorescence microscopy,” IEEE Eng. Med. Biol. Mag. 18(5), 92–99 (1999). [CrossRef] [PubMed]
33. N. Kahya, D. Scherfeld, K. Bacia, B. Poolman, and P. Schwille, “Probing lipid mobility of raft-exhibiting model membranes by fluorescence correlation spectroscopy,” J. Biol. Chem. 278(30), 28109–28115 (2003). [CrossRef] [PubMed]
