Optimising acquisition speed in localisation microscopy

Fluorescence microscopy is a crucial tool for cell biologists because it allows them to label different proteins with fluorescent molecules (fluorophores) and observe them in live cells. This yields information which can help us to understand diseases, and find new drugs to treat them. Until recently fluorescence microscopy had a major flaw: it could not resolve any features below 200nm. While human cells are at least twenty times this size, there are many parts of a cell which are much smaller. Over the last ten years a number of methods have been developed which allow fluorescence microscopes to image structures below 200nm, and these methods are now becoming standard in fixed (dead) cells. A major challenge in microscopy development is how to apply these methods in live cells, in a way that is reproducible enough that it can be used in cell biology laboratories where there are no experts in the technique. In this proposal we attack this problem with two approaches.

Firstly, we will investigate the theoretical limits of localisation microscopy. Localisation microscopy works by taking many images of the sample. The behaviour of the fluorophores is controlled so that in each image only a few of the fluorophores are emitting light. Even though each fluorophore results in a blurred spot, we can find the position of the centre of the spot very accurately. The image of the sample is then built up by putting a point down at the position of all the fluorophores we identify across all the frames. At the moment, people think about localisation microscopy as being similar to other microscopy techniques; you illuminate with light and you get an image, with the quality of the image depending on how good your microscope is and how bright your light is. However, for a localisation image to achieve a high resolution, you have to find the position of lots of fluorophores. Less obviously, the number of frames it takes to get a certain number of fluorophores depends on the structure of your sample, since you cannot image two fluorophores which are too close together. This means that the maximum speed depends on the structure of your sample. We will carry out simulations to work out how the maximum speed depends on the structure, which will allow cell biologists to know in advance what speed can be achieved for a given sample.

Secondly, we will develop a method which can examine the raw data from an experiment and determine whether, if you analyse it, you will get an image which reflects the structure of the sample, or if you will get an image with features caused by fitting the positions of fluorophores inaccurately. Currently, it is very hard to work out if this has happened, particularly if you try to get data quickly, which is necessary for live cell experiments. It may be possible to perform a quick test by looking at how the number of fluorophores which is detected changes over time. However, we are likely to need a more sophisticated test. We will use the images from an experiment and create a simulated image where we add a single fluorophore at a known position. We can then run the data analysis and see if the new fluorophore is correctly detected. By moving the fluorophore round, and performing the test on different frames, we will determine if there are particular times or places in the images where the data analysis is not working well.

By taking these two approaches, we will give every cell biologist with a localisation microscopy system the tools they need to calculate the maximum speed at which they can image the structure they are interested in. This will bring live cell localisation microscopy out of specialist labs and into the reach of cell biologists. Fixed cell localisation microscopy has already shown us many new and unexpected structures in the cell; by extending this technique into live cells, we will be able to see how these structures change and evolve over time.

In this project we aim to develop software which will enable researchers to carry out localisation microscopy in live cells. Currently the field has many analysis packages, and a few for experimental optimisation. However, there is no help for optimising live cell experiments, where speed is of the essence. Furthermore, the issue of whether artefacts might be present is very challenging to assess and has previously only been approached by doing correlative experiments or acquiring low density data, neither of which is practical in live cells.

To create this software we will need to develop our mathematical understanding of information transmission in localisation microscopy. This will allow us to quantify what speed of imaging is possible with a given localisation microscopy technique and a given sample structure. By building this information into a software tool, we will allow other cell biologists to optimise their localisation microscopy imaging of live cells.

Obtaining a reconstructed super-resolution image fast requires increasing the density of fluorophores in the raw data. However, these high densities are precisely the conditions most likely to lead to artefacts in the final reconstructed image. Therefore, we will also develop tests to identify when artefacts are likely to be present. We will check whether the fluorophore localisations follow Poisson statistics, since if they do not it implies there are incorrect numbers of fluorophores being fitted. We will also use simulations to add extra fluorophores to the raw data. If the position of the extra fluorophore is found by the fitting algorithm it will indicate that the data density is not too high in that region.

So, combining these two ideas, we will create an ImageJ plugin which can predict the maximum appropriate speed for a localisation microscopy experiment and can also take raw data and identify if there will be artefacts in the reconstruction.

The initial impact is primarily expected to be seen in an increase in the ability to perform live cell localisation microscopy experiments among cell biologists. There are a number of biological systems that could benefit from live cell localisation microscopy imaging, including but not limited to: fascin (upregulated in all known human cancers, cannot be imaged in fixed cells), the immune synapse in T cells and B cells, and structures associated with adhesion and migration such as focal adhesions and podosomes. These are all systems of high biomedical importance and in the longer term the greatest impact of this project is likely to be enabling and accelerating new biomedical research. A number of our collaborators (e.g. Dr Maddy Parsons) have strong links with pharmaceutical companies and would be well placed to help us take these results into an industrial context.

We have links to a number of microscopy companies, and on this project we have partnered with Leica. Recently we acquired the body of a Leica GSD system, and have adapted it using our own lasers and control systems. Leica have expressed strong support for those developing open-source solutions for other researchers, and they have offered to host a visit for both myself and Patrick at their development headquarters in Mannheim so that we can share our results with them. Leica stand to benefit from our work because it would extend the experiments that could be carried out on their systems and give users greater confidence in their results. In turn, their users will benefit because we will be able to advise on changes to the hardware and software of the system which would optimise it for live cell localisation microscopy.

We will also carry out experiments at the Nikon Centre, and so will be able to directly advise those using Nikon microscopes, and those using the N-STORM system in general.

Dr Patrick Fox-Roberts, the post-doc on this project, has received training in both analysis of microscopy data and in carrying out experiments. Super-resolution is a rapidly growing area, with many jobs being created, and a shortage of people with in-depth training.