Raimund Ober
Quantitative aspects of single molecule microscopy image analysis: an overview


Dylan Owen

Describing the distribution and dynamics of signalling molecules using data from single-molecule microscopy

In T cells, like many cell types, signalling pathways are regulated, at least in part, by the spatio-temporal arrangement of the proteins that make them up. In particular, protein nanoclustering has been shown to digitise signal transduction, setting basal levels and defining thresholds. Since appropriate T cell activation is crucial for mounting an immune response while at the same time avoiding autoimmune disease, understanding the nanoclustering behaviour of proteins is vital. Single-molecule localisation microscopy (SMLM), a variety of super-resolution imaging, is a commonly used tool for acquiring such data. Here, I will present a variety of tools to statistically analyse the output from SMLM to provide quantitative descriptions of protein nanoscale organisation. These will include Bayesian,  model-based methods with extensions to 3D and live-cell SMLM data and machine-learning approaches applicable to large data sets. Methods will be demonstrated using a variety of T cell proteins but in particular PTPN22, mutations in which are present in the human population, alter its nanoscale organisation in T cells, produce cell-level phenotypes and predispose the organism to autoimmune diseases including arthritis and diabetes. I will also present emerging challenges to the field going forward.


Scott Ward

Testing complete spatial randomness of the type 6 secretion system in Pseudomonas aeruginosa

In this talk we shall discuss methodology to explore spatial properties of the type 6 secretion system (T6SS) of Pseudomonas aeruginosa (P. aeruginosa). We model the underlying stochastic mechanism, giving rise to the observed T6SS, as a point process on the cell membrane of P. aeruginosa, where we approximate the cell membrane by an ellipsoid. Current methodologies in spatial point pattern analysis have been developed for planar and spatial data and although this theory can be applied a wide variety of applications it is not directly applicable here. This is becuase point patterns lie on an ellipsoid rather than being planar or spatial and the geometry of the space must be accounted for. We will discuss methodology to test complete spatial randomness of point patterns observed on an ellipsoid. By using the invariance of Poisson processes under transformations between metric spaces (known as the Mapping Theorem) we can transform one from any ellipsoid to a Poisson process on the unit sphere and take advantage of its rotational symmetries to construct functional summary statistics. In particular, we shall focus on Ripley's K-function and its inhomogeneous counterpart. Based on these functional summary statistics we can then determine whether a pattern observed on an ellipsoid exhibits complete spatial randomness or not and we highlight features of these derived summary statistics through simulations.


Juliette Griffié

How much should we trust quantitative bioimaging?

Fluorescence microscopy is an incredible toolbox for the investigation of the cellular machinery over multiple length scales.  It presents unique advantages such as protein specificity, multi-color, the access to 2D, 3D and whole cell data, from both fixed or live sample, to name a few. This, however, comes at a cost: complex microscopy set ups, with many user definable parameters to set for the image acquisition, and no clear guidelines to do so. This lack of robust framework for the optimization of acquisition parameters often impacts negatively on the output data: image quality deterioration and limited reproducibility are common issues faced by the field. In the face of these limitations, how much should we trust the statistics extracted from these images, independently of the analysis tool used? We propose here a Bayesian based framework for the optimization of acquisition parameters allowing for user-free microscopy acquisition, while insuring the reproducibility of the generated images. We will focus on the application of this framework to SMLM as well as a novel, real time SMLM simulator that we have developed for its validation.


Sumeetpal Singh

Identification of multi-object dynamical systems: consistency and Fisher information

Learning the model parameters of a multi-object dynamical system from  partial and perturbed observations is a challenging task. Despite recent numerical advancements in learning these parameters, theoretical guarantees are extremely scarce. In this article we aim to help fill  this gap and study the identifiability of the model parameters and the consistency of the corresponding maximum likelihood estimate (MLE) under assumptions on the different components of the underlying multi-object  system. In order to understand the impact of the various sources of observation noise on the ability to learn the model parameters, we study the asymptotic variance of the MLE through the associated Fisher information matrix. For example, we show that specific aspects of the  multi-target tracking (MTT) problem such as detection failures and  unknown data association lead to a loss of information which is quantified in special cases of interest. To the best of the authors' knowledge, these are new theoretically backed insights on the subtleties of MTT parameter learning.


Mansoor Sheikh

Analysis of over-fitting in the regularized Cox model

The Cox proportional hazards model is ubiquitous in the regression analysis of time-to-event data. However, when the data dimension $p$ is comparable to the sample size $N$, maximum likelihood estimates for its regression parameters are known to be biased or break down entirely due to overfitting. This prompted the introduction of regularizers, leading to the so-called regularized Cox model. In this paper we use the replica method from statistical physics to investigate the relationship between the true and inferred regression parameters in penalised multivariate Cox regression with $L_2$ regularization, in the regime where both $p$ and $N$ are large but with $\zz=p/N \sim \mathcal{O}(1)$. We thereby generalize a recent study from maximum likelihood to maximum a posteriori inference. We also establish a relationship between the optimal regularization parameter and $\zeta$, allowing for straightforward overfitting corrections in time-to-event analysis.


Heba Sailem

Deriving phenotypic signatures of angiogenesis using advanced image analysis of vascular networks

Angiogenesis plays an important role in many diseases including cancer invasion, cardiovascular disease, and Alzheimer disease. Endothelial tube formation assay provides a powerful approach for studying perturbations effects on endothelial cells ability to form vascular networks in vitro. However, the analysis of resulting imaging datasets has been limited to a few phenotypic features such as the total branching length or the number of branches and nodes. Here we develop an image analysis method for detailed quantification of various aspects of network formation including network symmetry and hypothetical flow efficiency. Using this approach we identified six biologically relevant phenotypes based on a high content screen of 1280 drugs. Clustering analysis revealed a novel group of proangiogenic drugs that have a similar mechanism of action and we validate its association with Alzheimer disease progression. In summary, our work shows that detailed image analysis of complex phenotypes can be highly valuable for targeted drug discovery.


Georege Ashdown

Machine learning identification of antimalarials using high-throughput, high-resolution imaging of malaria parasite cell development

As microscopy data becomes increasingly complex and dataset size increases towards population level; fast, automated acquisition and analysis are required for interpretation. By applying semi-supervised machine learning analysis to high-resolution fluorescent images of malaria infected red blood cells we demonstrate an approach which can successfully classify the different life-cycle stages within asynchronous parasite cultures. This architecture successfully organises parasite images into their natural developmental order and upon drug treatments of known antimalarials, detects and segregates phenotypes. This tool can now be extended to analyse large-scale imaging datasets to drive the discovery of novel antimalarials and may elucidate mechanism of action with more sensitivity and much faster than existing methods.


Iain Styles

Quantitative in vivo molecular imaging

Pre-clinical in vivo studies are an essential part of the path to approval of new drugs, and imaging using bioluminescent or fluorescent labels is one of the main tools used to assess where drugs go and the extent to which they affect disease progression. How much can we trust these experiments though? Can we really infer quantitative drug effects from these image when the remitted luminescence as measured at the detector is influenced by multiple factors that are typically not controlled for in these experiments – geometry, orientation, subject optical properties. Motivated by both technical and ethical concerns, we argue that these experiments should not be considered to be quantitative in any way, and describe a program of work in which we systematically account for and correct the different sources of error and uncertainty to take steps towards truly quantitative in vivo imaging.


Susan Cox

Information in localisation microscopy


Jean-Christophe Olivo Marin
TBC


Florian Levet

A tale of tiles: gathering tessellation and point clouds for SMLM data analysis

Over the last decade, single-molecule localization microscopy (SMLM) has revolutionized cell biology, making it possible to monitor molecular organization and dynamics with spatial resolution of a few nanometers. By identifying the molecule coordinates instead of producing images, SMLM holds an important paradigm shift towards conventional fluorescence microscopy. Consequently, developing dedicated analyzing tools has become essential to properly quantify SMLM data.

Due to their intrinsic geometrical characteristics, tessellations have been widely used for analyzing problems from domains as diverse as remeshing, astronomy or geology. In this family of space-partitioning technique, the Voronoï diagram is of particular interest as it encapsulates its seeds inside polytopes. A few years ago, we have developed an open-source framework called SR-Tesseler [1] in which the localization coordinates were directly used to reconstruct a Voronoï diagram. By using the Voronoï polytopes’ properties, we defined a per-localization local density that enabled a robust multiscale segmentation of biological systems with diverse shapes and sizes, ranging from the whole cells to small molecular complexes. More recently, we have used Voronoï diagrams to tackle the colocalization analysis of SMLM data. To this end, we proposed to extend their intrinsic multiscale capabilities to add a new pair-density descriptor to the localizations. We then used this new metric to adapt the well-known Manders’ and Pearson’s coefficients to SMLM data, resulting in a normalized colocalization analysis of SMLM data. We validated our method on simulations as well as on experimental data.


Jean-Baptiste Masson

Single biomolecule at the age of Big Data: Probabilistic pipeline and unsupervised analysis of random walks

Twenty years after its inception, the field of Single Molecule (SM) biology undergoes a transition towards a data-generating science [1-3]. At the nanometer scale, the dynamics of individual biomolecules is inherently controlled by random processes, due to thermal noise and stochastic molecular interactions. By accessing the distribution of molecular properties, rather than simply their average value, the great advantage of SM measurements is thus to identify static and dynamic heterogeneities, and rare behaviours.

In recent years, these experimental limits have been progressively alleviated with the advent of new, game-changing methods. Thanks to photoactivatable probes (protein-based or synthetic dyes), millions of individual trajectories can now be recorded in live cells in a few minutes. PALM/STORM images can be reliably acquired over many hours (or even days), yielding up to hundreds of millions of individual localizations.

As SM experiments enter the age of « big data », the development of a proper and unifying statistical framework becomes more necessary than ever.  « big data » approaches certainly open up new research venues for our understanding of biological processes, as they enable the inference of molecular dynamics. Yet they also come with a price. Often, adding more data brings both more information and more variability and noise. Specific tools are required to handle the complex structure of results associated to large datasets and to account for the sources of experimental and systemic variability.

Here, we show a global probabilistic pipeline: TRamWAy [4-7] that automatically analyse single molecule experiments from images to random walk analysis. TRamWAy relies on deep neural network to deconvolve single molecule images, Belief propagation coupled to ghost graph summing to perform probabilistic assignments between images, and both supervised and unsupervised Bayesian analysis to extract information from random walks.

We demonstrate the approach on two datasets: Glycine receptors in synapses and GAG dynamics during the formation of the Virion in HIV-1 [4]. We demonstrate two ways of applying the probabilistic pipeline TRamWAy. In the first we use model-based learning with automated results extraction and statistics. In the second we show that unsupervised learning with structured inference allows full analysis without assigning a model to the biomolecules dynamics.


Lekha Patel

A hidden Markov model approach to characterizing the photoswitching behaviour of fluorophores

Super-resolution imaging via STORM (Stochastic Optical Reconstruction Microscopy) exploits the inherent stochasticity of a photo-switchable fluorophore to reconstruct molecular positions at high resolutions. During an experiment, multiple blinks from each molecule can lead to misleading representations of their true spatial locations, therefore placing great importance in proper identification and inference of the unknown photo-switching rates. In order to characterise a molecule's photo-switching behaviour, we model its true photon emission state as a continuous-time homogeneous Markov process with m+3 states. The first m+1 states 0, 0_1, ..., 0_m refer to the unknown number of dark states, 1 refers to the photon-emission (On) state and 2 refers to the photo-bleached (permanently dark) state. During a single frame, the integral of this process gives rise to a binary discrete time imaged process indicating whether or not a molecule is detected. We describe a hidden Markov model (HMM) relating the observed process with the hidden continuous time signal, whereby observations are dependent on both current and past hidden states thereby producing emissions dependent upon the unknown photo-switching rates. We conceive transmission matrices that capture all such dependencies and which are needed to derive the log-likelihood function of observations. This likelihood can be numerically maximised to produce rate estimates and the Bayesian Information Criterion (BIC) for each model considered. We show through simulation studies the effectiveness of our procedure in rate estimation and the power of BIC in selecting the correct model given a range of different proposals.

Edward Avezov

Single particle tracking reveals reveals nanofluidic properties of the Endoplasmic Reticulum

The Endoplasmic Reticulum (ER), a network of membranous sheets and pipes, supports functions encompassing biogenesis of secretory proteins and delivery of functional solutes throughout the cell periphery. Molecular mobility through the ER network enables these functionalities. The diffusion-driven molecular motion (traditionally presumed by default), alone is not sufficient to explain the kinetics of luminal transport across supramicron distances. Understanding the ER structure-function relationship is critical in light of mutations in ER morphology regulating proteins that give rise to neurodegenerative disorders.

Applying super-resolution microscopy and stochastic analysis of single particle trajectories of ER luminal proteins revealed that the topological organization of the ER correlates with distinct trafficking modes of its luminal content: with a dominant diffusive component in tubular junctions and a fast flow component in tubules. Particle trajectory orientations resolved over time revealed an alternating current of the ER contents, whilst live ER fast structured illumination microscopy analysis identified energy-dependent tubule contraction events at specific points as a plausible mechanism for generating active ER luminal flow. The discovery of active flow in the ER has implications for timely ER content distribution throughout the cell, particularly important for cells with expensive ER-containing projections e.g. neurons, sanctioning efforts to understand the ER transport through mathematical modeling and biophysical analysis.


Olaf Ronneberger

Modelling ambiguities and uncertainty in biomedical image analysis with deep neural networks


Dominic Waithe

Advanced processing and characterisation of scanning fluorescence correlation spectroscopy aquired through conventional confocal microscopy

Scanning Fluorescence Correlation Spectroscopy (scanning FCS) is a variant of conventional point FCS that allows molecular diffusion at multiple locations to be measured simultaneously. It enables disclosure of potential spatial heterogeneity in molecular diffusion dynamics and also the acquisition of a large amount of FCS data at the same time, providing large statistical accuracy. In this talk we characterise the processing and analysis of these large-scale acquired sets of FCS data. On one hand we present FoCuS-scan, scanning FCS software that provides an end-to-end solution for processing and analysing scanning data acquired on commercial turnkey confocal systems. On the other hand, we provide a thorough characterisation of large-scale scanning FCS data over its intended time-scales and applications and propose a unique solution for the bias and variance observed when studying slowly diffusing species. Our work enables researchers to straightforwardly utilise scanning FCS as a powerful technique for measuring diffusion across a broad range of physiologically relevant length scales without specialised hardware or expensive software.


Falk Schneider

Statistical analysis of large sFCS data-sets discloses hindered diffusion dynamics

The plasma membrane of living cells is a profoundly bioactive structure and a platform for numerous interactions of proteins, solutes and lipids. Its functional organisation is highly spatiotemporally heterogeneous and heavily involved in regulating cellular function.

Here we use large scanning fluorescence correlation spectroscopy (sFCS) data-sets to differentiate free (Brownian) from hindered (non-Brownian) diffusion dynamics enabling a glance into the molecular membrane heterogeneity. Accurate determination of diffusion coefficients and diffusion behaviour can be performed by statistical analysis and fitting of transit time histograms. We make use of an inherent sampling bias in sFCS data and present a novel fitting approach including model selection criteria. Our biological findings line up with results previously obtained using super-resolution stimulated emission depletion (STED) nanoscopy combined with FCS. For instance, we observe free diffusion for phospholipids in model membranes and cell membranes whereas sphingolipids or GPI-anchored proteins undergo more complex diffusion behaviours in cellular plasma membranes.

Overall we are presenting a novel toolkit to investigate nano-scale molecular diffusion dynamics for shedding a new light on membrane organisation and heterogeneity.  Notably, our statistical analysis pipeline can be applied to data acquired on standard turn-key confocal microscopes using conventional fluorescent dyes or fluorescent proteins.


Ricardo Henriques

Democratising high-quality live-cell super-resolution microscopy enabled by open-source analytics in ImageJ

In this talk I will present high-performance open-source approaches we have recently developed to enable and enhance optical super-resolution microscopy in most modern microscopes, these are NanoJ-SRRF, NanoJ-SQUIRREL and NanoJ-Fluidics. SRRF (reads as surf) is a new super-resolution method capable of enabling live-cell nanoscopy with illumination intensities orders of magnitude lower than methods such as SMLM or STED. The capacity of SRRF for low-photoxicity, allows unprecedented imaging for long acquisition times at resolution equivalent or better than SIM.  For the second part of the talk, I will introduce SQUIRREL, an analytical approach that provides quantitative assessment of super-resolution image quality, capable of guiding researchers in optimising imaging parameters. By comparing diffraction-limited images and super-resolution equivalents of the same acquisition volume, this approach generates a quality score and quantitative map of super-resolution defects. To illustrate its broad applicability to super-resolution approaches, we demonstrate how we have used SQUIRREL to optimise several image acquisition and analysis pipelines. Finally, I will showcase a novel fluidics approach to automate complex sequences of treatment, labelling and imaging of live and fixed cells at the microscope. The NanoJ-Fluidics system is based on low-cost LEGO hardware controlled by ImageJ-based software and can be directly adapted to any microscope, providing easy-to-implement high-content, multimodal imaging with high reproducibility. We demonstrate its capacity to carry out complex sequences of experiments such as super-resolved live-to-fixed imaging to study actin dynamics; highly-multiplexed STORM and DNA-PAINT acquisitions of multiple targets; and event-driven fixation microscopy to study the role of adhesion contacts in mitosis.


David Gaboriau

Super resolution microscopy shows the organisation of the Escherichia Coli outer membrane

The early events of infection by Herpesvirus, such as genome uncoating, nuclear transport and the start of transcription from the viral genome, are not clearly understood.

We have developed new procedures to visualise single genomes and transcripts in human cells.

We used bioorthogonal chemistry to visualise viral genomes incorporating traceable precursors (Sekine et al., 2017, PLOS Pathogens) and single molecule RNA in-situ FISH to detect single transcripts.

By using these two techniques together, we were able to image the early stages of infection with widefield microscopy and deconvolution. Images were then processed with Icy, an open-source platform for bioimage analysis, with custom-made automated analysis sequences called protocols.

The multiple outputs generated by the protocols included number, geometry, intensity and distance measurements, as well as the position of each genome and transcript within the nucleus.

With these, we were able to characterise active transcription events by defining a transcription burst as a transcript of a certain size, intensity and distance from a genome.

We studied transcription of an immediate-early gene, ICP0, over time and followed the changes brought in by blocking viral genome replication and protein synthesis in the cell. We also used cells infected with a single virus, enabling us to follow transcription from one infecting genome.

Finally, by multiplexing the detection of transcripts, we followed transcription of two immediate-early genes, ICP0 and ICP4.

These results offer new perspectives on very early events of viral genome presentation after infection and transcription patterns from those genomes.


Jorge Bernardino de la Serna

Simultaneous spatiotemporal resolution of Lipid lateral packing and molecular diffusion


Sandip Kumar

Super resolution microscopy shows the organisation of the Escherichia Coli outer membrane

Antibiotic resistance in bacteria is on the rise for all classes of antibiotics. In the case of pathogenic Gram-negative bacteria, many are intrinsically resistant to some classes of antibiotics due to the outer-membrane (OM) acting as a diffusion barrier for drug molecules.  Bacteria maintains this barrier function during growth and division and this is mostly due the structure and organisation of the OM. The OM is asymmetric, where the outer surface has lipopolysaccharides (LPS) and outer membrane proteins (OMPs) exposed to the environment. While a great deal is known about the biogenesis of OM and how the LPS and OMPs are deposited on the surface of Gram-negative bacteria, surprisingly little is known about how these molecules are organised in a live bacterium. Here we fluorescently label LPS and OMPs and show their organisation in live E. coli. Fluorescence recovery (FRAP) and single particle tracking (SPT) shows that OM is immobile in live bacteria. Time lapse imaging of dividing cells shows that new LPS is deposited all along the OM surface while OMPs are preferentially incorporated in mid cell. Astigmatism based 3D localisation microscopy shows that LPS is uniformly distributed all along the OM and co-localises with OMPs.