Describing the distribution of proteins on the cell surface
How can we quantify the nanoscale distribution of proteins on the cell surface? How can we tell if they are randomly distributed or not? The answer lies in a new data analysis method for single-molecule super-resolution microscopy.
Fundamentally, fluorescence imaging is an attempt to capture and describe the spatial distribution of fluorescent molecules. In a fluorescence image for example, the pixel intensities are assumed to be proportional to the number of fluorophores in that part of the sample. Modern microscopy development can be seen as an attempt to better re-capitulate the ground-truth distribution of fluorescent molecules. One radical advance to do this is to attempt to record the x,y positions of fluorophores in a list, rather than resorting to a grey-scale pixel array (i.e. a picture). This is what super-resolution imaging by photoactivated localisation microscopy (PALM) does. Despite the data now consisting of x,y positions, each recorded with high precision, PALM data is still only a corrupted form of the underlying ground-truth molecule coordinates. There are many artefacts built into the data, but a key form of corruption is “multiple blinking” – that the same fluorophore can appear multiple times in the recorded dataset, owing to its complex photophysical behaviour.
The problem with correcting for multiple blinking in PALM data is that there are two things we don't know. If we knew the blinking dynamics of the fluorophores (problem 1); how molecules randomly blink over time to give multiple localisations, we could cluster the localisations into groups corresponding to a molecule, to best fit those dynamics. Conversely, if we knew which localisations corresponded to which molecules (problem 2), we could look at each group and fit the blinking dynamics. Knowing neither the clustering nor the blinking dynamics poses a seemingly difficult if not impossible estimation problem. However, in recent work, a mathematical approach was devised to exploit observed spatio-temporal correlations in the localisations to obtain a model for the blinking dynamics, circumventing problem 1. When we realised this was possible, it was a comparatively simple matter of integrating this model within a full, Bayesian model for the data to solve problem 2. This work is available here: https://www.biorxiv.org/content/10.1101/2021.03.24.436128v2.full
Apart from removing multiple blinking artefacts, which is the principal goal, there is an interesting opportunity to further improve the PALM data. The grouping of multiple coordinates that have come from the same molecule allow a better (more precise) estimate of that molecule’s position. Our method therefore moves microscopy further towards its goal of recapitulating the true x,y positions of fluorescent molecules by removing a key artefact corrupting the data and supplying more precise coordinates. Importantly, the method has no user-inputs making it objective and is backwards-compatible with previously acquired data.
Why is this important? Well, many researchers want to quantitatively describe the distributions of molecules that they see in cells, for example whether some membrane protein is randomly distributed on the cell surface or not. The presence of multiple-blinking makes even random distributions of real molecules look clustered in PALM data and so it is important to remove this effect before saying anything about whether proteins are really randomly distributed. In our paper, we show how this question can now be answered. Next, we might like to know how many proteins there are in a complex. Multiple-blinking generates data sets that are over-populated whereas with correction, the numbers of fluorophores in a cluster can now be accurately counted.
The new ability to quantitatively describe the distribution of fluorophores in cells with nanometer precision and free from multiple-blinking effects will help researchers understand how signals travel along pathways of interacting proteins to control such processes as cell division, migration and the immune response. All up to date materials including code, running instructions and example data is available from our Github repository: https://github.com/Louis-Jensen/MBC-for-PALM.