Background

Fluorescence lifetime imaging microscopy (FLIM) is one of various advanced fluorescence imaging techniques, which allows for the measurement of how long a fluorophore stays in an excited energy state. Changes in fluorescence lifetime can reveal important associations between the physical structure and the chemical microenvironment of a molecule. On this basis, FLIM allows investigating molecular level cellular behaviours, including, for instance, the metabolic state of differentiating cells, or FRET of signaling events during cell division. FLIM can be used in two modes of operation: time-domain and frequency domain, and is compatible with multiple microscopy modalities. Emerging applications of FLIM continuously push forward innovation of FLIM-related hardware. However, although quantitative FLIM has the need for powerful downstream image analysis software, the development of tools for FLIM image analysis has lagged behind due to various hurdles. Currently unmet needs in the scientific community include: a) an open and extensible FLIM algorithm library that supports most FLIM formats and which can be modified and used by developers; b) a FLIM analysis tool that can use the library, and can be easily used by bench biologists; c) the integration of FLIM analysis with versatile microscopy analysis. While various tools have been developed in this direction, FLIM workflows have been largely segregated from the open source tool ImageJ/Fiji, due to the lack of the necessary FLIM workflow functionality. However, the recent development of the ImageJ Ops framework has laid a solid foundation for such a FLIM analysis toolbox. The ImageJ Ops framework is a suitable environment for developing modular, extensible image processing workflow components.

In their work, Gao et al (1) have built on previous efforts by their group and others, to develop an ImageJ-centric toolkit, FLIMJ, that addressed the needs discussed above. FLIMJ offers FLIM fitting routines with seamless integration with other ImageJ components, and the ability to be extended to create complex FLIM analysis workflows. Being built on ImageJ Ops, this also enables FLIMJ’s routines to be used with Jupyter notebooks and integrate naturally with science-friendly programming. A number of other ImageJ features could be utilized with FLIM data including scripting, 3D visualization, and feature tracking. Moreover,  multipleimaging file formats can be supported through the use of the SCIFIO (http://scif.io/) infrastructures to support a range of open and proprietary FLIM file formats. The authors propose this as the first FLIM analysis system that offers this range of flexibility and functionality, and explore FLIMJ in two analysis scenarios: lifetime-based image segmentation and image colocalization.

 

Key findings and developments

Overall development: methods

FLIMJ provides access to a variety of FLIM analysis techniques including standard nonlinear least-squares fitting (in the form of the Levenberg-Marquardt (LM) algorithm); advanced algorithms such as maximum likelihood, global, and Bayesian analysis optimised for FLIM; and simpler methods such as the rapid lifetime determination (RLD) by integration and frequency domain analysis via the method of phasors. FLIMJ allows the ability to account for an instrument response function that distorts the pure exponential decay. Moreover, it allows the addition of new methods, and can inherit the standard library code interface, as well as ensuring maximal compatibility with other parts of ImageJ, which allows powerful preprocessing and post-anaylsis for the FLIM workflow. The toolkit presented by Gao et al  comprises 3 major components: FLIMLib, FLIMJ Ops, and FLIMJ-UI.

1.FLIMLib

FLIMLib (https://flimlib.github.io/) is a cross-platform compatible library with contains C implementation of the algorithms. The library can be compiled to run as a native executable on Linux, Windows, or macOS. FLIMLib is equipped with a Java Native Interface (JNI) wrapper created by the SWIG framework (http://swig.org/), which offers efficient type conversion and data transfer between C and Java applications. More connectors can be added to make the library accessible to many high-level programmers using Python, MATAB, C++ and C# among others. The interaction could be through a graphical user interface such as with TRI2 or ImageJ, through a command line in scriptable form, or via a third-party framework such as MATLAB or R. Data analysis methods currently present in the open-source library for lifetime data include: rapid lifetime determination, Levenberg-Marquardt (LM) non-linear least squares fitting, maximum likelihood estimation, global analysis, phasor, and Bayesian inference.

2.FLIMJ Ops

FLIMJ Ops is a plugin built upon the ImageJ Ops framework that connects FLIMLib and the ImageJ ecosystem. With help from the ImageJ Ops framework, FLIMJ Ops provides a concise yet flexible programmatic interface that can be easily included in a scripting workflow.

3.FLIMJ UI

FLIMJ-UI is an ImageJ plugin created to facilitate integrating tools for visualizing results and allow fine-tuning of configurations. It is based on the SciJava command framework, and invokes FLIMJ Ops to carry out the computation. The plugin can be started through scripting, or may be launched by the user from Fiji during an image analysis workflow.

 The authors further provide two example demo notebooks with the FLIMJ Ops repository: 1) a groovy notebook running on the BeakerX kernel that accesses ImageJ Ops directly, 2) a python Jupyter notebook that accesses ImageJ ops through the PyImageJ interface to invoke FLIMJ Ops.

 

Overall development: validation

The authors demonstrate the FLIMJ workflow and validate the results using separate software and simulated data for two different cases:

1.Segmentation

This case consisted on linking fluorescence lifetime processing to other advanced image processing plugins within ImageJ, to accurately measure protein dimer formation in cancer tissue. The authors aimed for FRET efficiency calculation of segmented tumorous tissue. A challenging aspect of these measurements in tissue is the need to segment the tumour tissue from surrounding stroma and other normal tissue components. The authors created a pipeline that uses trainable Weka segmentation on the intensity image in parallel to FLIMJ to provide tumour segmented lifetime statistics.

2.Co-localization

This case consisted on measuring co-localization of NAD(P)H, and antibody distribution for microglia. This use case is based on the Fiji ROI colocalization plugin and links to fluorescence lifetime processing of autofluorescence images of microglia. The authors have previously explored the potential applicability of NAD(P)H FLIM in differentiating microglia functional state. The expectation in this case was that the hybrid method allowing lifetime estimation from raw decay data, and colocalization analysis, can be helpful in determining the effectiveness of FLIM based approaches in the identification of microglia. To properly evaluate microglia identification with endogenous NADH signal, colocalization analysis can be of great benefit to quantitatively analyze overlap.

For both cases, FLIMJ was shown to perform optimally with the central components of each workflow, as well as to have the potential to be integrated with more complex ones. Moreover, the authors validated the routines by comparing FLIMJ against commercially available FLIM analysis standards.

What I like about this preprint

I like all developments that are consistent with open science, and the democratization of science – making tools available to everyone, both in terms of cost accessibility, and expertise (i.e., that the development is equally accessible to developers and to bench scientists). Equally, I think this development will be of great use to the scientific community as it closes important gaps between hardware novelties and software needs.

References

1.Gao D et al, FLIMJ: an open-source ImageJ toolkit for fluorescence lifetime image data analysis, bioRxiv, 2020.

 

doi: https://doi.org/10.1242/prelights.25896

Read preprint (1 votes)