GSLab: Open-Source Platform for Advanced Phasor Analysis in Fluorescence Microscopy

Abstract

GSLab addresses the need for effective image analysis tools in fluorescence microscopy by providing an open-source platform that enhances traditional phasor analysis with advanced features. Key capabilities include machine learning-based clustering, real-time monitoring, and quantitative unmixing of fluorescent species. Designed for both commercial and custom systems, GSLab provides researchers with comprehensive lifetime and spectral phasor image analysis tools to tackle complex biological problems.

Main Text

Fluorescence lifetime microscopy (FLIM) is a powerful imaging technique that provides quantitative information about the local environment and molecular interactions of fluorophores within biological tissues or other materials. By measuring how long a fluorescent molecule remains in an excited state before emitting a photon, FLIM offers a new dimension for biological imaging, which, combined with spectral imaging, enables researchers to investigate complex cellular processes and interactions. The integration of advanced analytical methods, such as the phasor approach, enables more sophisticated image analysis and interpretation. As a result, this will further enhance the capabilities of FLIM and spectral microscopy.

The basis of phasor analysis lies in using the phasor transform to map every pixel of an image onto a two-dimensional space known as phasor space, based on the photon distribution within that pixel across the fluorescence lifetime or spectral dimensions^1,2. The position of each pixel in phasor space is determined by the shape of the photon distribution and is independent of the signal’s intensity. Analysis by means of the phasor representation does not require prior knowledge of the nature of the sample nor fitting of a model. In addition, utilization of the Fast Fourier Transform algorithm enables rapid computation. This analysis simplifies visual inspection and identification of distinct populations of pixels, which can subsequently be mapped back to the original fluorescence image (or set of images)³. Furthermore, the mathematical properties of the phasor transform enable researchers to understand the phenomena occurring in the sample by observing changes in the photon distribution represented in phasor space. A brief overview of the mathematics behind the phasor approach for analyzing fluorescence lifetime microscopy images is available in Online Methods.

Phasor analysis is widely used in the biophysics and bioimaging fields and has led to a vast number of quantification methods and applications. These include studying cellular metabolic states⁴, molecular interactions⁵, molecular dynamics⁶, drug delivery⁷, chromatin compaction⁸, sensing local polarity⁹, ion concentration¹⁰, pH¹¹, enhancing superresolution¹² and multiplexed imaging¹³.

For over a decade, the scientific community has primarily relied on Globals for Images SimFCS software developed by Prof. Enrico Gratton for phasor analysis¹⁴. However, maintenance and updates for this software were discontinued in 2021, prompting developers to create alternative phasor analysis tools. While several research laboratories have developed custom tools to address specific challenges^15–24, and commercial brands such as Leica, PicoQuant, Becker & Hickl, FLIM Labs, and ISS have incorporated phasor analysis into their software suites, there are few reports on open-source software solutions. Existing platforms, based on Phyton (PhasorPy¹⁵, FLUTE¹⁸, Phasor Identifier¹⁹, FLIMPA²⁵), Java (FLIMJ²⁰) or MATLAB (PAM¹⁷), have successfully incorporated traditional phasor analysis techniques, such as intensity thresholding, image filtering, color-mapping, and cursor analysis (manual clustering). Despite these developments, recent advances in machine learning-clustering for image segmentation²⁶ and unmixing of fluorescent species present in the same pixel via higher harmonics²⁷, are not yet available in open-source formats.

To address this gap, our team has developed GSLab, a pioneering, open-source software platform designed to provide researchers with a comprehensive set of tools for advanced lifetime and spectral phasor image analysis. GSLab not only incorporates traditional phasor analysis techniques, but also introduces unique and advanced capabilities, including automated machine learning-based clustering in phasor space for image segmentation, real-time monitoring of both image and phasor space, and quantitative unmixing of multiple fluorescent species from a single pixel, applicable to any imaging system, whether commercial or custom-built.

GSLab is developed in MATLAB. It offers robust computational and visualization capabilities, allowing researchers to enhance their phasor analysis workflows effectively. GSLab compiles these tools into a user-friendly program, featuring a graphical user interface that simplifies complex analyses. MATLAB’s extensive library of built-in functions facilitates rapid development, enabling users to tailor GSLab to their specific research needs. While a MATLAB license by Mathworks Inc. may be a barrier for some, the widespread availability in academia ensures that most researchers can benefit from the advanced functionalities of GSLab. We have uploaded GSLab in a public repository²⁸ under MIT license.

Figure 1 provides an overview of GSLab’ capabilities. It highlights its key features across three main areas: Input/Output, Basic Analysis, and Advance Analysis. The Input/Output section illustrates the software’s compatibility with various image formats for lifetime and spectral data, along with its export options for phasor plots and color-coded intensity images. The Basic Analysis section illustrates the commonly available tools such as cursor analysis, phasor filtering and image manipulation, enabling users to interact with and process the data efficiently. The current open-source solutions allow for the implementation of a subset of these analysis tools. The Advanced Analysis section emphasizes the software’s powerful functions for unmixing multiple fluorescence components, machine-learning based clustering, and real-time examination of the reciprocity between image and phasor space, demonstrating its robust analytical capabilities. A more detailed description of GSLab’ capabilities is available in the Supplementary Material.

Figure 1. GSLab capability overview.

1) Input/Output: Supports major image formats for lifetime and spectral data. Allows exporting of phasor plots and color-coded intensity images based on phasor coordinates, component unmixing, or clustering results, with various styling options for both phasor plots and images. 2) Basic Analysis: Offers cursor analysis, phasor filtering, image manipulation, and the ability to create or load masks. Users can measure pixel phasor coordinates and create color gradients on the phasor space, generate lifetime and spectral images based on the color gradients, and calibrate the data using reference files. 3) Advanced Analysis: Enables unmixing of multiple fluorescence components, machine learning-based clustering of phasor distributions, and real time inspection of the reciprocity between image and phasor space.

Figure 2 illustrates the advanced capabilities of phasor analysis for automated machine learning, clustering, and unmixing of fluorescent components. We demonstrate these capabilities by analyzing two sets of data. First, label-free FLIM images of a fresh human skin specimen (discarded tissue from surgery), acquired with a custom-built clinical multiphoton microscopy device²⁹ are used to demonstrate automated clustering and image segmentation (Figure 2 A–D). Second, time-resolved fluorescence images of labeled cell cultures, obtained using another custom-built FLIM-based multiphoton microscope for thick tissue imaging³⁰ are used to demonstrate unmixing of fluorescent components (Figure 2 E–H).

Figure 2. Advanced phasor analysis: Automated machine-learning clustering of phasor distributions and fluorescent component unmixing.

A) Intensity obtained through time-resolved two-photon excited fluorescence signal detection from a human skin specimen. B) Corresponding fluorescence lifetime phasor plot with gradient color distribution representing tau-phase values. The plot demonstrates automated machine learning clustering into four populations with ellipses outlining 88% of pixels within each Gaussian component. C) The intensity image from (A) shown as color-coded lifetime image, based on tau-phase values from the phasor plot. D) The intensity image from (A) displayed as color-coded image highlighting the four populations identified through automatic clustering. These populations correspond to known skin structures: keratin in epidermal keratinocytes (green), melanin in pigmented keratinocytes (red), elastin in the dermis (blue), and other structures characterized by a mixed fluorescence lifetime distribution, representing a linear combination of the other three clusters. E) Intensity images obtained through time-resolved two-photon excitation signal detection from five cell cultures: four of them stained independently with one of the nuclear dyes Ethidium Bromide, Acridine Orange, NucBlue, and Rose Bengal, and the fifth with a mixture of all four dyes. F) Corresponding fluorescence lifetime phasor plot with gradient color distribution representing tau-modulation values. The plot shows the fluorescence lifetime signatures for each dye and the signature of the mixed sample as an elongated distribution in the center. The theoretical lifetime of the four dyes is used to compute the coordinates for each of the four components, depicted by the vertices of a quadrilateral. Inset displays the same data in the 2nd harmonic required for the fluorescent component unmixing. G) Intensity images from (E) shown as color-coded lifetime images, based on tau-modulation values from the phasor plot. H) Results of fluorescent component unmixing: H1-H4) Unmixed photon fraction images for each component (rows) and each measurement (columns). Inset violin plots illustrate pixel value distribution for each image. H5) Merged images for each measurement using linear addition of panels (H1-H4).

Phasor analysis of the skin specimen’s FLIM image (Figure 2 B) shows automated machine-learning-based clustering into four populations, which are subsequently illustrated in the corresponding color-coded FLIM image revealing well-known skin structures. Phasor analysis of the fluorescence images of stained cell cultures highlights the ability to unmix four fluorescent components by clearly resolving the fluorescence lifetime signatures of the dyes and computing the pixel fractions of the components in the mixed sample. These examples demonstrate the powerful advanced capabilities of GSLab for resolving complex fluorescence signatures within a single sample.