Using spectral decomposition of the signals from laurdan-derived probes to evaluate the physical state of membranes in live cells

Chemicals and probes

Laurdan, M-laurdan and C-laurdan were all synthesised in-house as previously described (Mazeres et al., 2014). Aliquots of 2 mM stock solutions in DMSO were kept in long-term storage at -20°C; when needed for experiments, aliquots were kept at 4°C for, at most, six weeks. Various staining procedures were used for different experiments, as described later. Nile Red was obtained from Sigma; staining was performed at 10ng/ml for 30–60 min at 37°C in tissue culture medium without serum. LysoTracker Red DND-99 was obtained from Invitrogen; staining was performed at 20 ng/ml for 30–60 min at 37°C in tissue culture medium without serum. Methyl-β-cyclodextrin (MßCD) (cell culture tested) was obtained from Sigma.

Multilamellar large vesicles

Multilamellar large vesicle (MLV)s were prepared as follows. DPPC (1,2-dipalmitoyl-sn-glycero-3-phosphocholine), DOPC (1,2-dioleoyl-sn-glycero-3-phosphocholine) and cholesterol were purchased from Sigma–Aldrich, and prepared as stock solutions in chloroform. The appropriate amounts of these stock solutions to obtain a final concentration of either 100 μM DOPC (for Ld MLVs), or 60 μM DPPC and 40 μM cholesterol (for Lo MLVs) in a final volume of 3 ml were placed in glass tubes. The chloroform solvent was first evaporated under nitrogen flow and then under vacuum for 2 hours. Three ml of MOPS buffer (3-(N-morpholino)propane sulfonic acid, 10 mM, pH 7.3, NaCl 100 mM, EDTA 10 μM) was then added either at room temperature (for the DOPC MLVs) or at 50°C (for the MLVs containing 60% DPPC and 40% cholesterol). The tubes were then vortexed vigorously for 30 seconds to form MLVs. Stock solutions of the various probes were prepared in DMSO such that 1/1000 v/v of the probe solution was added to the aqueous suspension of MLVs. To prevent dye aggregation in water, the probes were always added while vortexing the tubes containing the MLVs. They were then incubated at room temperature for at least one hour before analysis in a QM4 Spectrofluorimeter (Photon Technology International), or observation by biphoton microscopy (see below).

Cells

Mammalian cells were all grown in DMEM with 10% fetal calf serum (FCS) at 37°C with 5% CO₂ and passaged by trypsinisation every three or four days. Hela, HCT116 and L(tk-) cells were obtained from the ATCC. The primary culture of human foreskin fibroblasts was a gift from Laure Gibot (IPBS, Toulouse, France). Drosophila melanogaster Kc cells were provided by Vanessa Gobert (CBD, Toulouse, France) and were grown at room temperature in Shields and Sang medium with 10% FCS. Dictyostelium discoideum AX2 cells, provided by François Letourneur (DIMNP, Montpellier, France), were grown in HL5 medium at room temperature.

Cell preparation and staining

All mammalian cells were grown for 2–4 days on glass coverslips in six-well plates, with 2 ml of medium in each well. On the day of the assay, the glass coverslips were placed inside stainless steel culture chambers (either from SKE Research Equipment (Milan) or from ThermoFisher), which had been pre-warmed in the same incubator as the cells for at least 30 minutes. One ml of the medium that the cells were growing in was then transferred from the well of the six-well plate to the chamber, and the chamber containing the coverslip with the cells on it was then returned to the incubator for at least 30 minutes. The medium was then rinsed once with 1ml of DMEM with neither serum nor phenol red, which had been pre-warmed to 37°C. The probes (in 1µl DMSO) were placed in an Eppendorf tube, and 1 ml of pre-warmed DMEM was quickly pipetted up and down twice in the tube before transferring onto the cells. The chamber was then returned to the tissue culture incubator for various times (at least 20 minutes for C-laurdan, and 45 minutes for M-laurdan and laurdan) before observation. For the time-courses, such as the ones shown in Figure 7, C-laurdan was added by pipetting 500 µl out of the chamber, pipetting this liquid up and down twice into an Eppendorf tube containing the adjusted amount of probe in DMSO, and returning the 500 µl to the observation chamber with pipetting up and down a few times.

Drosophila Kc cells were grown for 48 hours on coverslips coated with poly-L-lysine (0.1 mg/ml for 1 hour followed by three rinses in PBS), and stained with 800 nM C-laurdan in Shields and Sang serum-free medium. Dictyostelium AX2 cells, were incubated for ~two hours in HL5 medium at room temperature to allow them to adhere to untreated coverslips. They were then stained with 800 nM C-laurdan in Sörensen's buffer.

Since the probes fluoresce only in a lipid environment, the cells were imaged in the staining medium. (Note that if rinsing is necessary, it should be done with medium without serum, or the labelling will fade rapidly, especially in the case of C-laurdan. If cells need to be returned to serum-containing medium, for example for very long incubation times, one should consider using laurdan or M-laurdan.)

Spectral imaging by biphoton microscopy

Mammalian cells were maintained at 37°C, 5% CO₂ throughout the imaging procedure. All images were recorded on a LSM 710 NLO-Meta confocal laser-scanning microscope controlled by Zen software (2010B SP1, v 6.0.0.485), equipped with a 40x/1.2 water immersion objective, a gas-controlled thermostatic chamber and a spectral detection module (Zeiss, Germany), and coupled to a biphoton laser source (3 watts at 800 nm; Chameleon Vision II, Coherent, France). When tuned to 720 nm, the laser delivered 2.14 watts, and with the AOTF set to 4%, the average measured power delivered at the level of the objective was 6 mW.

Unless otherwise specified, the settings of the microscope were: biphoton tunable laser set at 720 nm; 2–4% power; pixel dwell time of 6.3 µsec; average on four measures; pinhole open (600) or sometimes closed to 150 (never more) to gain higher resolution of close-up pictures (zoom 4 and above). When the settings differed from these, the specific details are given in the legends of the corresponding figures. The images were all acquired as stacks in lambda-mode with spectral resolution steps of 9.8 nm. For acquisitions with the laurdan-derived probes, we recorded the 17 channels between 418 nm and 584 nm plus a channel for transmitted light. When Nile Red or LysoTracker Red were included in the experiment, we recorded the 29 channels between 418 nm and 700 nm, plus a channel for transmitted light. All the images shown in all the figures are representative of at least six images (of six different fields), acquired on at least two different days.

Image analysis

Although the Zen software we used for acquisition of images on the Zeiss microscope allows extensive and convenient image analysis, including spectral decomposition (called ‘linear unmixing’ in the Zen software), this commercial software is not widely available. We have therefore chosen to base all the image analyses reported in this paper on ImageJ software (current version 1.48V, Java 1.6.0_65) complemented with various plugins, which are all open code software and freely available.

The PoissonNMF plugin (https://github.com/neherlab/PoissonNMF) performs nonnegative matrix factorisation to allow spectral decomposition without knowing the initial spectra (Neher et al., 2009). Here, we used it to perform spectral decomposition analysis of reference spectra acquired separately on MLVs of defined composition. The ‘blind’ mode used on live cells labelled with laurdan-derived probes usually gave results very similar to those obtained by using reference spectra, but the spectral identification was sometimes less accurate, especially when the signal to noise ratios were low.

When using fixed spectra with the PoissonNMF plugin, we used ten iterations (as recommended by Richard Neher, personal communication). The other values were the default settings suggested by the program: subsamples, 2; segregation bias, 0; saturation threshold, 4000; background threshold, 50; background spectrum, minimal values, and the spectral values were specified manually to match those of the spectral analyser of the biphoton microscope. Note: for the PoissonNMF plugin to function correctly when using fixed spectra, it is essential that the number of channels specified matches exactly the number of channels in the stack.

The Fraction Mapper plugin (see Software and data availability) was developed specifically for this study. This plugin converts the two-image stack obtained through spectral decomposition into fractional intensities by dividing the unmixed images by the total intensity image. The summation of fractional intensities is always 1 and since we only unmix two components, the fractional intensity can be indexed with only one number. Next, the fractional intensity is mapped on to the total intensity image and it is colour-coded using either readily available or custom made look-up table (LUT)s. The saturation of the colours is scaled with the pixel value of the total intensity. The plugin allows the choice of various LUTs and also thresholding of the image. After running the plugin on the stack of unmixed images, the plugin asks for the desired LUT, it generates the colour-coded fractional intensity image and also a 2D histogram, where the frequency is colour-coded. The axes of the histogram plot are the intensity of the two unmixed images. The plugin allows the user to make regions of interest on the 2D plot and to project them back into the fractional colour-coded image, rather like the gating of flow cytometry plots.

Flow cytometry

For analysis by flow cytometry, Hela cells were harvested by trypsinisation, and a single-cell suspension was prepared in tissue culture medium (DMEM with 10% FCS). Cells were then washed twice in pre-warmed DMEM with neither serum nor phenol red before adding the probes (1/1000 v/v of the DMSO stocks) followed by incubation in a 37°C water bath placed near the flow cytometer.

For cholesterol depletion, a fresh 10X stock solution of MßCD was prepared at 50mM (66mg/ml of MßCD powder) in PBS on the day of each experiment. Cell suspensions were rinsed twice in DMEM without serum before incubating with 5mM MßCD for 30 minutes at 37°C. Cells were then rinsed twice with warm DMEM without serum before staining with the laurdan-family probes. (Note, treatment of the cells with MßCD after labelling with the laurdan-family probes results in very effective removal of the probes.)

At the end of the incubations, the cells were analysed on a Beckton Dickinson LSR II flow cytometer equipped with a 355 nm UV laser, and the following set of filters were used for the analysis of the emitted light: first channel: 505LP + 530/30 BP; second channel 450/50 BP. For each sample, 5000 cells gated on FSC/SSC were analysed.

Using spectral decomposition to study colocalisation of two stains in live cells shows that laurdan and M-laurdan accumulate in lipid droplets

In a previous study, we compared the photophysical properties of C-laurdan and two intermediates in its synthesis, M-laurdan and MoC-laurdan, to those of laurdan (Mazeres et al., 2014). We found that the probes distributed very differently inside live cells. Laurdan stained predominantly cytoplasmic ‘vesicles’, which we speculated might belong to the endolysosome pathway (Mazeres et al., 2014). Since then, we observed that these vesicular bodies were very apolar, since they were seen much better at 440 nm than at 490 nm (Supplementary File 1), which suggested that they may in fact be lipid droplets. To discriminate between these two possibilities, we wanted to perform double labelling with either LysoTracker Red (which labels lysosomes) or Nile Red (which labels lipid droplets) and one of each of the three laurdan-derived probes. We found, however, that these vesicles were also very mobile (Supplementary File 1), so we could not perform successive image acquisitions to investigate whether the laurdan-based probes co-localized with LysoTracker Red or with Nile Red.

To circumvent this problem, we used the lambda mode of acquisition on our microscope to collect the fluorescent light emitted by the two dyes at exactly the same time, and then ‘unmixed’ the signals of the two probes by spectral decomposition. Thus, the mobility of the vesicles in live cells no longer confounded the interpretation of localisation of the two dyes. To unmix the emission spectra, we first used cells labelled with only one probe to acquire the corresponding emission spectrum from 420 to 700 nm. We then used these emission spectra to perform spectral decomposition on images of cells stained with two probes (Figure 1). Decomposition revealed that the staining patterns obtained with LysoTracker Red did not coincide with those obtained with any of the three laurdan probes (Figure 2). By contrast, the staining patterns obtained with Nile Red were essentially superimposable on those obtained with all three laurdan-derived probes, and in particular with laurdan: there was an almost perfect coincidence of the staining patterns obtained with laurdan and Nile Red (Figure 3). We conclude from this that the cytoplasmic vesicles detected with laurdan (and to a certain extent with M-laurdan) do not belong to the endolysosome pathway, as we initially postulated, but are lipid droplets. These findings are supported by those of Owen and colleagues who noted in the troubleshooting table of their article (Owen et al., 2012a) that lipid droplets tend to sequester laurdan. As previously documented by time-resolved microscopy (Ghosh et al., 2013), the environment inside lipid droplets is relatively viscous and very apolar, but not necessarily very ordered. Our findings are, therefore, a useful reminder that the solvatochromism of the laurdan family probes reflects not the order but the polarity of their environment; in other words, the presence or absence of water molecules in the immediate environment of the probes.

Figure 1. Scheme of the double staining and spectral decomposition procedure.

To perform spectral decomposition on live cells labelled with two fluorescent dyes, we first acquired lambda stacks of cells labelled with either Nile Red or laurdan. The emission spectra of the dyes were then extracted from those stacks. Those emission spectra were then used to perform spectral decomposition on the lambda stacks acquired from cells stained with both dyes (bottom right panel).

Figure 2. Intracellular vesicles labelled by laurdan and M-laurdan do not stain with LysoTracker Red.

Hela cells were grown on glass coverslips for four days before double labelling with LysoTracker Red and either laurdan (top row), M-laurdan (middle row) or C-laurdan (bottom row), as described in the Methods. The live cells, at 37°C, were then imaged over 30 channels: 29 fluorescence channels from 420–700 nm, and transmitted light (left column). Spectral decomposition was then performed on the lambda stacks using the PoissonNMF plugin and the spectra acquired with cells labelled with only one probe as references. The second column shows the signals attributed to the laurdan-family probes artificially coloured in green. The third column shows the signals attributed to LysoTracker Red artificially coloured in red. The fourth column shows a merge of the two signals, and the fifth column shows 10x magnifications of the areas indicated by the white squares in the fourth column. Width of the squares, 10 µm.

Figure 3. Intracellular vesicles labelled by laurdan coincide perfectly with Nile Red labelling of lipid droplets.

Hela cells were grown on glass coverslips for four days before double labelling with Nile Red and either laurdan (top row), M-laurdan (middle row) or C-laurdan (bottom row), as described in the Methods. The live cells, at 37°C, were then imaged over 30 channels: 29 fluorescence channels from 420–700 nm, and transmitted light (left column). Spectral decomposition was then performed on the lambda stacks by using the PoissonNMF plugin and the spectra acquired with cells labelled with only one probe as references. The second column shows the signals attributed to the laurdan-family probes artificially coloured in green. The third column shows the signals attributed to Nile Red artificially coloured in red. The fourth column shows a merge of the two signals, and the fifth column shows 10x magnifications of the areas indicated by the white squares in the fourth column. Width of the squares, 10 µm.

Of note, in the clumps of 15–20 cells labelled with either laurdan or M-laurdan (Figure 2 and Figure 3), the labelling intensity of the cells inside the clumps appeared somewhat weaker than the labelling of cells at the periphery. This may be due to inaccessibility of the probes to the cells inside the clumps, but more likely reflects a difference in the lipid composition of the cells in the crowded environment inside the clumps when compared to the cells at the periphery, as suggested previously (Frechin et al., 2015; Gray et al., 2015).

Using spectral decomposition on signals from solvatochromic probes

To reconstitute colour pictures from the stacks of images obtained in lambda mode, we used a chromatic LUT designed to match the wavelengths of the corresponding channels (Figure 4). For each of the four solvatochromic probes, we noticed differences in the colours emitted by various cellular compartments. For the laurdan-based probes, as expected from the results of several previous studies (Golfetto et al., 2013; Niko et al., 2016; Owen & Gaus, 2010; Owen et al., 2011; Yu et al., 1996), the cytoplasm was slightly greener than the blue PMs and lipid droplets. With Nile Red, however, there was a striking difference between the staining of the lipid droplets, which were bright yellow, and the staining of the intracellular membranes, which appeared red–orange. This solvatochromism of Nile Red in lipid droplets, which is due to their very hydrophobic nature, was documented previously (Greenspan & Fowler, 1985).

Figure 4. Colour-coded representation of lambda stacks reveals different colours for different cellular compartments.

From left to right: Hela cells labelled with laurdan, M-laurdan, C-laurdan and Nile Red, respectively. Top row, transmitted light. Bottom row, recoloured emitted fluorescent light after acquisition by the multichannel spectral analyser. Colouring was achieved using the built-in ‘time-lapse colour coding’ plugin of ImageJ, after converting the 29 channel stacks into 29 frame stacks and a custom-made LUT called Rainbeau (Supplementary File 3). In designing this LUT, we attempted to match the wavelengths of the corresponding channels of the microscope’s spectral analyser.

These observations led us to wonder whether we could decompose the various spectral components of the different colours emitted by the laurdan-derived solvatochromatic probes by a similar approach to that which we used to unmix the signals from two probes above. Indeed, rather than a progressive shift in their emission maximum, the laurdan-derived probes have more of a bimodal response to changes in the dielectric constants of their environment, which is particularly well marked between the more water-exposed environment of the probes in lipid membranes in disordered phase and the more hydrophobic, or apolar, environment of the probes in more ordered membranes, whether Lo or gel phases (Bagatolli, 2013; Bagatolli et al., 2003; Parasassi & Gratton, 1995). Thus, these bimodal fluorescence signals should be particularly amenable to spectral decomposition, and this might allow us to evaluate the proportion of ordered and disordered states of cell membranes.

We first compared the emission spectra obtained from fluorescently labelled DPPC–cholesterol (Lo) or DOPC (Ld) MLVs in a regular spectrofluorimeter to those obtained with the biphoton microscope in lambda mode from 480 to 600 nm for the three laurdan-based probes, and from 520 to 700 nm for Nile Red (Figure 5). The emission curves extracted using ImageJ software on the MLV images acquired on the biphoton microscope were remarkably similar to those obtained from the same MLVs in cuvettes in the spectrofluorimeter. Thus, the spectral analyser of the biphoton microscope can be used to analyse emission spectra in two dimensions at the submicrometer scale.

Figure 5. Emission curves obtained with the biphoton microscope are very similar to those obtained on a standard spectrofluorimeter.

The four fluorescent lipid probes were inserted into either Lo or Ld MLVs and their emission curves were recorded in cuvettes on a spectrofluorimeter (thick lines), or imaged on the biphoton microscope in the same conditions as used for live cells. After defining appropriate ROIs in the lambda stacks, the emission curves of individual MLVs were extracted using the ImageJ software. The thin lines correspond to the average curves obtained from 6–10 separate MLVs (the error bars are smaller than the symbols). Lo, DPPC–chol MLVs (blue or orange); Ld, DOPC MLVs (green or red).

We then tried our decomposition approach on the most challenging task available to us, i.e. we performed spectral decomposition into four different channels on stacks of 29 images acquired after double staining with laurdan and Nile Red (Figure 6). By using the four emission curves obtained with each probe on either DPPC–cholesterol (Lo) or DOPC (Ld) MLVs to perform spectral decomposition with the PoissonNMF plugin, we obtained a very convincing separation of the signals, where laurdan was found to be in an apolar, water-poor environment in lipid droplets and at the PM, whilst being much more exposed to water in the intra-cellular membranes. For Nile Red, there was effectively no detectable staining of the PM, but a marked dichotomy between hydrophobic lipid droplets and more hydrophilic intra-cellular membranes.

Figure 6. Image stacks can be separated into four different colours.

The PoissonNMF plugin was used to perform spectral decomposition on the lambda stack obtained from live cells double labelled with laurdan and Nile Red (see Figure 3). The reference curves used (lower panel) were obtained from MLVs labelled either with laurdan (Lo, DPPC–chol, cyan curve; Ld, DOPC, orange curve) or with Nile Red (Lo, DPPC–chol, green curve; Ld, DOPC, red curve). The settings used were all as described in Methods. Top left (cyan), laurdan in an apolar (hydrophobic) environment; bottom left (orange), laurdan in a water-exposed environment; top right (green), Nile Red in an apolar (hydrophobic) environment; bottom right (red), Nile Red in a water-exposed environment.

From these results, we came to suspect that spectral decomposition could turn out to be a very powerful approach to analyse the cellular distribution of solvatochromic probes such as laurdan, and more importantly, the actual physical state of their environment in different cellular compartments.

Improving the reproducibility of the results obtained with C-laurdan

Most signalling events take place at the PM. Given its cellular distribution, C-laurdan appeared as the best suited probe to investigate such events. At the beginning of our study, however, we found that the intra-cellular distribution of C-laurdan, as well as the polar/apolar ratios in various cellular compartments, were very variable, not only from one experiment to another, but even between cells on the same coverslip.

The first factor that we identified as having a major influence on the heterogeneity of our results was cellular stress. This could result, for instance, from increased osmotic pressure due to water evaporation when we carried out the staining steps in small volumes, from sudden temperature changes if we used buffers that had not been pre-warmed to the same temperature as that of the cells, or simply from unstable, oscillating temperatures when we used chambers of insufficient thermal inertia.

Intra-sample homogeneity was much improved by the use of heavy stainless steel chambers in which we could perform the staining steps with large volumes (0.5 – 1 ml) of buffers pre-warmed to 37°C (see Methods).

Despite these improvements, we were still witnessing very significant variability in the staining patterns of the cells stained with C-laurdan. In particular, we noticed that these patterns tended to change over time, with the high prominence of the apolar component at the PM that we obtained at the very early time points (< 5 min) tending to fade, and even disappear almost completely over time (cyan colour in Figure 7 upper row). We found out, however, that this did not occur when we used lower concentrations of the probe (see lower panel of Figure 7).

Figure 7. High concentrations of C-laurdan can result in autoquenching at the plasma membrane.

Hela cells growing on coverslips were placed in observation chambers in 1 ml of serum-free medium and placed in the thermostatic chamber of the biphoton microscope. After focusing on a chosen group of cells, C-laurdan was added to a final concentration of either 4 µM (top panels) or 200 nM (bottom panels). Lambda stack images were then taken at the indicated times, with the laser power set at 2% for staining with 4 µM, and 4% for 200 nM C-laurdan. All other settings were as described in Methods and spectral decomposition was carried out as described in Methods. The pictures shown are dual-colour overlays. Cyan, C-laurdan in apolar environments; orange, C-laurdan in water-exposed environments.

Having observed this, we felt that it was important to document at what precise concentrations this phenomenon was occurring. In addition, we also wanted to optimise our staining protocols with all three probes for both the concentrations used and the incubation times. One of the major drawbacks of fluorescence microscopy, however, lies with the difficulty in generating quantitative measurements. Another difficulty of microscopy lies with the relatively small number of samples that can be analysed.

Using flow cytometry for quantitative analyses

To circumvent this problem, we turned to flow cytometry, which allows quantitative analysis of many cells, in numerous different samples to be performed. For each cell, emissions in the 400–500 and 500–560 nm windows were recorded simultaneously. We realised that the ratio of the signals recorded in the first channel over that in the second should vary with the proportion of the probe emitting fluorescence in apolar versus water-containing environments, similarly to the GP factor, which is commonly used for the analysis of laurdan-based studies. Given that the intensities of the signals are completely dependent of the instrument settings, however, the values obtained are purely relative, and can thus only be used to compare the samples of one experiment with one another.

As can be seen in Figure 8, when we incubated cells with increasing concentrations of C-laurdan for 60 min at 37°C, fluorescence intensities increased in both channels, but the 450/530 ratio decreased progressively, and very notably, for concentrations superior to 1.2 µM. When we looked at the evolution of the 450/530 ratio over time in cells labelled with 200 nM of C-laurdan, we found that this ratio was very stable over time, and, if anything, tended to go up slightly for incubation times superior to 30 minutes (Figure 8A, upper right panel). The progressive disappearance of staining of the PM seen by microscopy at high concentrations of C-laurdan fits with the reduction in the 450/530 ratio seen by flow cytometry with increasing concentrations of the probe. Thus taken together, our results strongly suggest that, in cells kept at 37°C, C-laurdan does not have an intrinsic affinity for an intra-cellular compartment in which it would get progressively trapped. A more likely explanation seems to be that some autoquenching of the probe occurs preferentially in the PM. This more marked tendency of C-laurdan to autoquench in a Lo environment was confirmed by performing staining of MLVs with increasing concentrations of C-laurdan and measuring fluorescence in a standard spectrofluorimeter. As can be seen in Figure 8B, the fluorescence intensities recorded plateaued above 10 µM of C-laurdan, and this was even more noticeable at 440 nm than at 490 nm, with the signal at 440 nm actually decreasing for DPPC–cholesterol MLVs stained with 20 µM of C-laurdan.

Figure 8. Using flow cytometry (FACS) to quantify staining intensities at 450 and 530 nm, as well as the ratio between the two.

(A) FACS was used to quantify the mean fluorescence intensities (MFIs) of Hela cells labelled with C-laurdan (top), M-laurdan (middle) or laurdan (bottom). Blue squares indicate MFI at 450 nm and green circles indicate MFI at 530 nm, as plotted on the y-axis on the left side of each plot. Orange triangles indicate the ratio of 450/530 MFIs as plotted on the orange y-axis on the right side of each plot. The plots in the left column show dose–response curves after labelling cells at 37°C for 60 minutes. The plots in the right column show time courses of cells labelled at 37°C with either 200 nM C-laurdan, 500 nM M-laurdan or 4 µM laurdan. The unfilled blue squares and orange triangles indicate measurements as for the filled symbols, but on cells treated with 5 mM MßCD for 30 min at 37°C before labelling. (B) The preferential autoquenching of C-laurdan in Lo environments, suggested by the drop in the 450/530 ratio in the FACS analysis of cells labelled with high concentrations of the probe, can also be seen in MLVs analysed on a standard spectrofluorimeter. Serial two-fold dilutions of C-laurdan in DMSO were added (1% vol), whilst vortexing, to aliquots of MLVs prepared either with DOPC (Ld, dotted lines) or with DPPC–chol (Lo, continuous lines). After 20 min incubation at 37°C, the emission spectra of each aliquot were recorded in quartz cuvettes in a standard spectrofluorimeter (excitation: 360 nm), and the fluorescence intensities at 440 nm (blue lines) and 490 nm (green lines) were plotted against the concentration of C-laurdan.

When similar experiments were performed with cells labelled either with laurdan or M-laurdan, we found evidence for some autoquenching occurring with M-laurdan at concentrations above 600 nM, but not with laurdan (Figure 8A, middle and bottom panels). It can be noted that the tendency to autoquench correlates with the efficiency in labelling cells. In the experiment shown, the autoquenching becomes noticeable for levels of fluorescence superior to 10.000 for both C-laurdan and M-laurdan. The reason why autoquenching is not seen for laurdan may thus be that the levels of staining with that probe never attain these values.

When it came to optimizing the times necessary for staining cells, C-laurdan turned out to be a much simpler probe to use. Indeed, for C-laurdan, levels of staining and the 450/530 ratio were found to be remarkably stable between 10 and 30 min, but this was not the case for the other two probes. First, levels of staining with laurdan and M-laurdan kept increasing over time all the way to 60 minutes. Additionally, for M-laurdan the 450/530 ratios decreased during the first 20 minutes, probably as a consequence of the probe's diffusion from the surface to the intracellular membranes.

In order to validate the use of cytometry to evaluate the polar/apolar ratio in cell membranes, we performed staining of cells that had been treated with methyl beta cyclodextrin (MßCD) to deplete them of cholesterol (dotted lines and hollow symbols in right hand side panels of Figure 8A). As expected, this resulted in very significant decreases of the 450/530 ratio for all three probes, with evolutions over time that paralleled those recorded on untreated cells.

Based on these results, we elected to use 200 nM of C-laurdan, 500 nM of M-laurdan or 2 µM of laurdan as the concentrations that give the best compromise between good levels of staining and minimal autoquenching, and to incubate the cells with the probes for at least 45 min at 37°C before observation for the latter two probes, and 20 minutes for C-laurdan. We also find that maintaining the cells in the staining medium for imaging gives the best results. Of note, if rinsing needs to be performed, it should be done with medium with no serum, lest the labelling will rapidly fade away, especially for C-laurdan. But in our experience this is really not necessary since the probes are only fluorescent when inserted in a lipid environment.

Optimising the presentation of the results of spectral decomposition

Having established the appropriate concentrations and incubation times for the three probes, we could then compare the staining patterns obtained on Hela cells after spectral decomposition. The results of the decomposition consist of two-channel stacks, which correspond, respectively, to the fraction of the probe in either apolar or more water-exposed environments. After numerous tests and surveys among users to find out what two-colour combination worked best for most people, we selected to use the cyan colour for the first channel, and orange for the second one. As can be seen on Figure 9, all three probes did result in very similar overall patterns, but with some noticeable differences: whilst laurdan tends to accumulate inside cells, and very noticeably inside lipid droplets, C-laurdan labels the PM more efficiently, and M-laurdan has an intermediate behaviour.

Figure 9. After spectral decomposition, the staining patterns obtained with laurdan, M-laurdan and C-laurdan are similar, but C-laurdan labels the PM better than the other two probes.

Live Hela cells grown on glass coverslips for two days were labelled with either 2uM laurdan (top row), 500 nM M-laurdan (middle row) or 200 nM C-laurdan (bottom row), as described in Methods. Lambda stacks were then acquired by biphoton microscopy of the live cells at 37°C, and spectral decomposition was carried out, all with the standard conditions specified in Methods. Left column, transmitted light; second column (cyan), decomposed fluorescent signal corresponding to the probes in an apolar (hydrophobic) environment; third column (orange), decomposed fluorescent signal corresponding to the probes in a polar (water-exposed) environment; fourth column, overlay of the signals shown in the second and third columns.

Whilst the cyan-orange two-colour representation is already quite informative, we found that it was not at all appropriate to evaluate the ratio of the two channels in the various cellular compartments. To this end, we generated the Fraction Mapper plugin for ImageJ, which proceeds in successive steps (Figure 10). Starting from a two-channel image stack, such as the ones obtained with the PoissonNMF plugin, Fraction Mapper first calculates the fraction of signal in the first channel for every pixel (i.e. value of pixel in first channel / sum of values for that pixel in first and second channel). Fraction Mapper then colour-codes this fraction according to a chosen LUT, and gives each pixel the intensity of the sum of the two components. For our purpose, we find that a LUT based on six discrete colours from blue to red gives the clearest results (Supplementary File 2). Compared to relying just on the fluorescence recorded in the 440 and 490 nm channels, as classically used when calculating GP and shown on the panels at the bottom of Figure 10, the spectral decomposition approach based on combining PoissonNMF and Fraction Mapper can clearly result in a very marked improvement of the resolution between different cellular compartments.

Figure 10. The Fraction Mapper plugin.

Starting from a stack of two channels, Fraction Mapper calculates the fraction of the intensity in the first channel for every pixel (ch1/ch1+ch2), which can be obtained as a fraction map. A 2D plot of the intensities in each of the channels can also be generated. The plugin then colour-codes the numbers corresponding to the fraction according to a chosen LUT, and gives the pixels the light intensity of the sum of the two channels. The LUT used was based on six discrete colours from blue to red (LUT ‘6 colours 2,3,3,3,3,2’ available for download as Supplementary File 2). Bottom row: Starting from the 440nm and 490 nm channels, classically used for calculating GP, the picture obtained is much less informative. Of note, GP values are closely related to fraction values we use in this study: the GP can be obtained simply by multiplying the fraction by two and subtracting 1. Owen et al. have produced and published a ‘GP Calculator’ plugin (Owen et al., 2011) that provides very similar results to those obtained with Fraction Mapper, albeit by a more indirect approach. The 10× magnifications correspond to the 52×52 pixels regions indicated by white squares in the adjacent image; each pixel is 100×100 nm and the whole pictures cover 52×52 microns.

With the representation obtained using Fraction Mapper, one can rapidly see that, in Hela cells stained with the three laurdan-derived probes (Figure 11), although the distribution of the three probes is quite different, the overall picture is basically the same: intra-cellular compartments harbour a majority of water-exposed probes (yellow, orange and red), corresponding most likely to membranes predominantly in Ld states, whilst the regions near the PM show a strong dominance of apolar environments, which presumably correspond mostly to Lo states.

Figure 11. Fraction Mapper gives a much clearer picture of the relative polarity of the lipid environment in various cellular compartments than two-colour overlay.

In the middle column, the Fraction Mapper plugin was used on the two-colour stacks from Figure 9 (shown in left column), which were obtained by spectral decomposition using PoissonNMF. The right column shows 10x magnifications of the 52×52 pixel regions indicated by white squares in the middle column. Each pixel is 100×100 nm and the whole pictures cover 52×52 microns. Top row, 2uM laurdan; middle row, 500 nM M-laurdan; bottom row, 200 nM C-laurdan.

Upon ticking the corresponding box, Fraction Mapper will produce a ‘Fraction Map’ image, which can then be used to quantify the actual numerical values of the fraction of the first channel in a defined area of the pictures, such as individual pixels or ROIs drawn on particular regions of the cells.

Fraction Mapper can also generate 2D plots based on the intensity of each pixel in the two channels of the stacks of images used as starting material. On those 2D plots, it is possible to draw regions of interest or gates and to then use the ‘Gate to Image’ plugin tool to generate an image based on the fraction map, but containing only the pixels falling within the gate. This function was implemented because we felt that, considering the very sharp regional differences seen between the PM and the intra-cellular membranes, we might be able to identify separate populations of pixels on this type of graph, much like what can be done for cells in flow cytometry. This has, however, proven not to be the case: as long as the cells are alive and healthy, and as long as there are no over-exposed areas in the field, we observe no obvious clouds of dots, but always a regular distribution of the pixels instead. This has been true on all the numerous dot plots we examined in the course of this study, acquired on a variety of cell lines.

Optimizing the settings for biphoton acquisition

The Fraction Mapper plugin provided us with the capacity to analyse rapidly the physical state of the cellular membranes we were imaging with much improved clarity and precision compared to the two-colour cyan-orange pictures obtained through spectral decomposition. In turn, this allowed us to adjust the settings we were using to ensure the reproducibility of our results; in other words, to make sure that we were not damaging the cells during the acquisition process. Indeed, one of the major pitfalls of using biphoton excitation lies with the very high intensities of light necessary for an efficient absorption of two photons simultaneously. This is of particular concern in live cells because this type of illumination can very easily induce cell damage, in particular at the level of membranes. For our study, which focuses on the physical state of biological membranes, the localized delivery of heat was also of particular concern to us because, as anybody who has ever left butter in direct sunshine, or tried to spread butter on toast when it comes straight out of the fridge knows only too well, the physical state of lipids is especially sensitive to changes in temperatures.

Ultimately, we found that the systematic sequential acquisition of two pictures, coupled with the Fraction Mapper representation, was a very effective way to adjust our settings to obtain good signals, whilst ensuring that we were not inducing damage, or even just modifications to the membranes of the cells we were studying (Figure 12). For example, we find that using laser powers over 6% can very rapidly induce membrane blebs, and we therefore never use powers over 4%. Although biphoton excitation already results in excitation that is limited to a depth of about 1 μm, this resolution along the z axis can be somewhat improved a little bit further by closing the pinhole normally used for confocal acquisition with one-photon excitation. Whilst we did find that closing the pinhole could marginally improve the resolution in the pictures, we also found that closing it past a value of 150 resulted in an excessive loss of photons, which required increasing the laser power above 4%. Other factors that we identified as possibly resulting in altered pictures between the first and the second passage are excessively long dwell times per pixel or excessive number of averages, if only because both can also result in very long acquisition times. In the end, the standard settings we have elected to use are as follows: laser power: 2–4% (corresponding to 3–6 mW average power delivered, as measured at the level of the sample), pixel dwell time: 6.3µsec, average on 4 measures, pinhole: 150. With such settings, the acquisition of a 512x512 picture takes 15 seconds, and as can be seen on Figure 12, the state of the cells remains unaltered during the second acquisition compared to the first passage, whilst the micro heterogeneities seen at the level of individual 100 nm pixels are themselves remarkably fluctuant.

Figure 12. Using sequential acquisitions to verify that the acquisition has not modified the membranes or damaged the cells.

Hela cells grown for two days on glass coverslips were stained with 200 nM C-laurdan and imaged twice sequentially using the standard conditions defined in Methods (power 4%, speed 6, 4 averages, pinhole 150). With these settings, the acquisition procedure takes 15 seconds. The second pictures were thus taken 15 seconds after the first ones. Each pixel is 100×100 nm, and each picture, taken at zoom 4, covers 52×52 microns.

Comparing the results of spectral decomposition in a variety of cell lines

Having established these optimized settings and conditions using Hela cells, we turned our attention to a panel of other cell lines, and found very similar patterns in all the mammalian cells we looked at (see Figure 13 for three examples).

Figure 13. C-laurdan stains various cell types in similar patterns.

Human HeLa cells (first column), HCT116 cells (second column), foreskin fibroblasts in primary culture (third column) and mouse L(tk-) cells (fourth column) were all grown on glass coverslips before staining with C-laurdan, imaging and analysing the images by the standard procedure, as described in Methods. Top row: overlay of the two channels obtained after spectral decomposition with PoissonNMF (cyan, apolar environment; orange, polar, water-exposed environment). Middle row: the two channel stacks were processed with the Fraction Mapper plugin using the six-colour 2,3,3,3,3,2 LUT shown on the right-hand side. Bottom row: 10x magnifications of the 52×52 pixel regions indicated by white squares in the middle row. Each pixel is 100×100 nm, and each picture, taken at zoom 4, covers 52×52 microns.

Considering the importance of temperature in regulating the state and order in lipid bilayers, we wondered what the situation would be in cells from ectotherms. To this end, we turned our attention to cells from Drosophila melanogaster, and to Dictyostelium discoideum, which both grow at room temperature. For both of these cell types, we found that we needed to use slightly higher levels of the C-laurdan probe to reach satisfactory levels of staining (but which did not lead to any detectable autoquenching of the probe). The fact that these organisms are less efficiently labelled by C-laurdan may be due to their lipid composition being different from that of mammalian cells, such as different sterols and lipid chains better adapted to life at more variable temperatures. In line with this hypothesis, the results shown in Figure 14 reveal that the C-laurdan probe, when inserted in the membrane compartments of these two very different organisms, tends to emit fluorescent light that is blue-shifted compared to when it is in mammalian cells: the cytoplasm appears dominantly yellow rather than red-orange, and the PMs show a high prominence of dark blue rather than cyan pixels. On the whole, C-laurdan thus seems to be slightly less exposed to water when it is inserted in the membranes of those cells than in the membrane of mammalian cells, which would suggest that lipid membranes of these ectotherm organisms tend to be a bit less permissive for water to penetrate deep into the lipid bilayers. At the cellular scale, however, the overall picture remains the same, with the environment of the PMs of the cells of these two ectotherms being much more apolar than their intra-cellular compartments.

Figure 14. Staining of Drosophila and Dictyostelium cells.

Left column: Drosophila melanogaster Kc cells were grown at room temperature for 48 hours on PLL-coated coverslips in six-well plates before being transferred to an observation chamber and stained with 400 nM C-laurdan in serum-free medium. Middle and right columns: Dictyostelium discoideum amoebae (AX2 strain) growing at room temperature were placed directly into an observation chamber with a glass coverslip bottom for two hours in axenic medium before staining with 800 nM C-laurdan in Sörensen's buffer (see Methods). The pictures in the right column were taken 15 seconds after those in the middle column; they show that the motility of these photoreactive amoebae, induced by the first acquisition, resulted in no noticeable modification of the overall state of their lipid membranes. The microscope settings and analyses were identical to those described previously for mammalian cells. All pictures in this figure were taken at zoom 6. Whole picture 35×35 µM. Pixel size, 69×69 nm