Quantitative mapping of membrane nanoenvironments through single-molecule imaging of solvatochromic probes

Environmentally-sensitive fluorophores report on their local biochemical or biophysical environments through changes in their emission. We combine the solvatochromic probe di-4-ANEPPDHQ with multi-channel SMLM and quantitative analysis of the resulting marked point patterns to map biophysical environments in the mammalian cell membrane. We show that plasma membrane properties can be mapped with nanoscale resolution, and that partitioning between ordered and disordered regions is observed on length scales below 300 nm.


Main Text
The mammalian cell plasma membrane is the site of processes essential for cellular functionhomeostasis, mechanotransduction and signalling. Membrane biophysical properties play an integral role in regulating these processes. Once such property is membrane lipid order. It is hypothesised the plasma membrane can partition into disordered and ordered phase regions (Figure 1a) 1 . Ordered regions are enriched in saturated lipids and sterols, and are enriched in proteins which have high affinity for that environment. Ordered regions also display tighter lipid packing, thereby limiting the penetration of polar water molecules into the bilayer core. This feature can be exploited by solvatochromic probes that exhibit distinct changes in their fluorescence emission behaviour depending on the polarity of their surrounding environment 2 . Laurdan and di-4-ANEPPDHQ have previously been employed in this context using confocal microscopy, however, ordered domains are thought to exist below the resolution limit of conventional microscopy 3,4 . Nile Red has previously been shown to be amenable to study by single-molecule localisation microscopy 5 . By recording the full emission spectrum for each localisation, changes in the emission wavelength due to differences in local environment can be observed 6,7 . Here, we demonstrate SMLM with the more commonly used dye, di-4-ANEPPDHQ 8 , to probe membrane lipid nano-environments in mammalian cell plasma membranes. di-4-ANEPPDHQ shows a large blue-shift in emission spectrum in ordered membrane regions when compared to disordered membrane regions (Figure 1a). Using a 2-channel imaging approach based on point accumulation in nanoscale topography (PAINT), we show that we can acquire the x-y coordinates of di-4-ANEPPDHQ insertion events into the cell plasma membrane. Furthermore, a ratiometric analysis of the two acquired spectral channels allows us to calculate a specific parameter for each point -the so-called Generalized Polarisation (GP) value -a measure of membrane order 8 ; the resulting data, therefore, takes the form of a marked point pattern. We develop an analysis method for marked point pattern data, here in the context of membrane order, which has wide applicability for other marked data types.
di-4-ANEPPDHQ-PAINT was performed on fixed cells using 15 nM di-4-ANEPPDHQ in an oxygen scavenging buffer system (Glucose Oxidase:Glucose) to improve its photostability and making it amenable for SMLM imaging. Binding events at the plasma membrane were recorded simultaneously in the short and long wavelength channels (Supplementary Movie 1), and the sub-diffraction-limited positions of the individual binding events were extracted. Binding events were then matched frame by frame between the two channels, and the GP value calculated for each of the paired localizations using the extracted photon number (Supplementary Figure 1). The average position of the localisations in each channel is taken, and the GP value is ascribed to that position creating an x, y, GP coordinate. GP values run theoretically from -1 (highly disordered) to +1 (highly ordered). Using this approach, we were able to generate data with localisation precisions better than 30 nm (Supplementary Figure 2).
Voronoi tessellations have been implemented previously for both clustering and colocalization analysis of SMLM data. This turns a list of coordinates into a 2D set of tessellated polygons 9,10 , with the coordinates being at the centre of the polygons such that every position within one polygon is closer to that point than any other (Supplementary Figure 3). We implemented Voronoi tessellation on our GP SMLM data while retaining the GP value for each tile.
To visualise any ordered or disordered domains in the plasma membrane, we combined neighbouring polygons that had similar GP values (in this case, ± 0.1) into one supertile. We combined polygons until no more could be merged, thus giving a complete map of membrane domains within the data (Figure 1b). As a measure of domain size, we calculate the number of raw Voronoi polygons that are merged into each supertile (Figure 1c) and calculate the overall area of these tiles, displayed here as a cumulative probability distribution (Figure 1d).
While the Voronoi tessellations and generation of supertiles are useful for visualisation, they do not provide a statistical basis for the establishment of whether high or low order domains exist in the membrane. To achieve this, we applied a second approach, which interrogates the spatial separation, or colocalization, of the high and low order points. Given that we have points marked by a continuous range of GP values, (Figure 2a), we are able to split our localisations into "ordered" (Figure 2b, left) and "disordered" (Figure 2b, right) groups using a GP threshold. By splitting the data according to the membrane order, we can use analyses, which have been applied to conventional two-colour SMLM colocalization 11 , to determine whether high and low order points segregate from each other, or whether they are well-mixed. In this case, the GP threshold for splitting was set at 0.2.
The localised versions of the Ripley's K-function derivative, L(r), and the equivalent for the cross-K-function, Lcross (r), have been used previously to determine if point patterns in conventional two-channel SMLM data colocalise at a chosen length scale 11 (Supplementary  Figure 4). L(r) and Lcross(r) were calculated for each point and plotted against each other, allowing the Pearson's correlation coefficient to be calculated (Supplementary Figure 4b and  4d). We also performed the analysis on the same SMLM data, but this time randomly segregating points, irrespective of their GP value, into the two groups, as a control condition (Supplementary Figure 4c and 4e). This process was then repeated for different values of the search radius, r, allowing the degree of colocalization to be quantified over different length scales.
For small search radii, the average calculated Pearson correlation coefficients were statistically significantly lower when segregating points by their GP values than those in the randomly assigned case (Figure 2c and d). However, for larger radii, there was no significant difference between the GP and random segregation showing that the low and high order points were well mixed at large radii. Thus, segregated ordered membrane regions exist only at a sub-diffraction spatial scale.
In summary, we have demonstrated that marked point pattern data can be extracted from the solvatochromic probe di-4-ANEPPDHQ using a 2-channel PAINT SMLM acquisition. When combined with statistical cluster and colocalization analysis, this can serve as a basis for probing and mapping the nanoscale spatial organisation of ordered lipid domains in the cell membrane. Our data suggest that large scale segregation of domains may not be commonplace, but that ordered domains can be observed segregated from disordered areas at the nanoscale (~ 50-250 nm). This shows the potential for further membrane mapping approaches and lays the groundwork for analysis of other   , 500 nm (right). b) Points were segregated into two channels using the GP threshold of 0.2, corresponding to ordered (green) and disordered (red) channels. Scale bar 500 nm. c-d) L(r) vs L cross (r) analysis of the ordered and disordered channels respectively and compared to a random assignment of points into two channels. (ns: P > 0.05, *: P ≤ 0.05).

SUPPLEMENTARY FIGURES
Supplementary Figure 1: Processing and extraction of GP from SMLM data. di-4-ANEPPDHQ-PAINT binding events are recorded in long (> 650 nm) and short (< 650 nm) wavelength channels. The positions of the binding events and the number of photons per channel are determined by SMLM fitting. The photon numbers are then used to calculate the GP value for the event. Finally, the average position of the binding event is taken, and the GP value is ascribed to that point. Scale bar 500 nm.